doi
stringlengths 28
28
| title
stringlengths 19
311
| abstract
stringlengths 217
5.08k
| plain language summary
stringlengths 115
4.83k
| article
stringlengths 3.87k
161k
|
|---|---|---|---|---|
10.1371/journal.ppat.1002129 | The Impact of Recombination on dN/dS within Recently Emerged Bacterial Clones | The development of next-generation sequencing platforms is set to reveal an unprecedented level of detail on short-term molecular evolutionary processes in bacteria. Here we re-analyse genome-wide single nucleotide polymorphism (SNP) datasets for recently emerged clones of methicillin resistant Staphylococcus aureus (MRSA) and Clostridium difficile. We note a highly significant enrichment of synonymous SNPs in those genes which have been affected by recombination, i.e. those genes on mobile elements designated “non-core” (in the case of S. aureus), or those core genes which have been affected by homologous replacements (S. aureus and C. difficile). This observation suggests that the previously documented decrease in dN/dS over time in bacteria applies not only to genomes of differing levels of divergence overall, but also to horizontally acquired genes of differing levels of divergence within a single genome. We also consider the role of increased drift acting on recently emerged, highly specialised clones, and the impact of recombination on selection at linked sites. This work has implications for a wide range of genomic analyses.
| As bacteria diversify, many of the nucleotide changes that emerge will render the cell slightly less competitive, and these mutations will tend to be removed by natural selection. However, this purging process does not happen instantaneously, and this delay allows deleterious mutations to survive in the population long enough to be sampled. Genomes at the very initial stages of diversification therefore exhibit a relatively high proportion of slightly deleterious mutations and, as most of these are non-synonymous mutations, this is manifest as a high dN/dS ratio. However, the effective population size will also impact on this ratio, as will selection operating on neighbouring SNPs. Here we examine the distribution of synonymous and non-synonymous SNPs within recently emerged clones of two important nosocomial pathogens, methicillin resistant Staphylococcus aureus (MRSA) and Clostridium difficile. In both species, we note a much higher proportion of synonymous changes in those single nucleotide polymorphisms (SNPs) likely to have emerged through recombination compared to de novo mutations. We argue that this effect is explained by the very recent emergence of the mutational SNPs combined with a reduction in the efficiency of selection due to niche specialisation.
| The populations of many pathogenic bacterial species exist as a collection of discrete clonal complexes, many of which have emerged recently and exhibit specific resistance or virulence attributes. For example, molecular techniques such as multilocus sequence typing (MLST) [1] have identified a small number of widely disseminated methicillin resistant S. aureus (MRSA) clonal lineages [2], [3]. Of these, the MLST haplotype Sequence Type (ST) 239 is the most common globally [4], [5], [6], [7], [8], [9], [10]. This clone is multiple antibiotic resistant, shows increased virulence [11], [12], and is known to have emerged through a very large (635 kb) homologous replacement via an unknown mechanism [13], [14], [15]. Harris et al recently sequenced a global sample of 63 isolates of this clone using the Illumina Genome Analyser (IGA) platform [16]. Read mapping was carried out against a completely sequenced ST239 isolate (TW20) which caused an outbreak in a London hospital [11] [14], and this approach yielded powerful evidence concerning the global dissemination of this clone. Comparable studies using the Illumina platform have subsequently been carried out on single clones in other pathogenic species, including Streptococcus pneumoniae [17] and Clostridium difficile [18].
Robust phylogenetic and evolutionary analysis on many bacterial genomes, and S. aureus in particular, relies on drawing a clear distinction between the “core” and “non-core” genomes [19], [20], [21], [22], [23]. The S. aureus core genome is stable, with modest levels of sequence divergence (typically ∼1.5%), low rates of homologous recombination, and a high degree of synteny [24], [25], [26]. In contrast, the non-core genome is highly diverse and dynamic due to high rates of horizontal transfer of accessory elements, including a number of genomic islands, various “types” of the SCCmec chromosomal antibiotic resistance cassette, and prophages encoding various toxins and other virulence attributes [25], [27], [28]. Harris et al defined core SNPs conservatively as those affecting contiguous regions (>1 kb) that were universally present within all the ST239 isolates sequenced [16]. A very low rate of homoplasy confirmed that the vast majority of the core SNPs arose by recent de novo mutation, rather than recombination. Analysis of the core data also suggested that ST239 first emerged in the mid 1960s, shortly after the first administration of methicillin in 1959.
Intuitively, it might be expected that non-core genes will exhibit a higher proportion of non-synonymous change than core genes. As non-core genes are not ubiquitous and therefore - by definition - non-essential, they might be under weaker selective constraints. Additionally, non-core genes encoding virulence factors may be subject to positive selection from the immune response [29]. However, non-core genes are typically acquired by horizontal transfer, and tend to exhibit striking mosaic structure, implicating a long history of recombination [23], [30], [31], [32], [33]. It follows that the de novo (mutational) emergence of many SNPs in the non-core may pre-date the emergence of the clonal lineage in which they are observed, particularly so if the clone has emerged very recently and spread very rapidly, as is the case with ST239. The age of the SNPs is important because very recently emerged de novo mutations tend to contain a high proportion of non-synonymous changes, not as a result of positive selection but because purifying selection has had insufficient time to purge slightly deleterious mutations [34], [35], [36], [37], [38]. Whilst this effect might be pronounced for recently emerged SNPs in the core genes, many of the non-core SNPs will have already passed through a selective filter in the wider population prior to transfer and so should show a lower proportion of slightly deleterious (non-synonymous) changes.
An important caveat of this time-dependence model is that the relative enrichment of non-synonymous changes within mutational SNPs should only be apparent when comparisons are made between very closely related isolates belonging to the same recently emerged clone, and recombination should have little impact on dN/dS when more diverged strains of the species are considered. Furthermore, core genes which have experienced homologous replacement originating from an external lineage should, like non-core genes, also show an enrichment of synonymous change. This provides the means to test alternative explanations which assume differing selective pressures acting on core and non-core genes.
In addition to the age of the SNPs, it is also necessary to consider variation in the efficiency of selection, at both a genome-wide level and at linked sites. For example, the adoption of a highly specialised niche would decrease the effective population size leading to greater drift (higher dN/dS). This would be more pronounced in the core relative to the mobile non-core, as the latter has been acquired horizontally from the wider population. Local rates of recombination also impact on the degree to which selection removes linked neutral variation [39], [40]. Regardless of whether selection is positive (hitch hiking [41]) or negative (background selection [42]) a greater degree of linked neutral variation will be removed in lowly recombining regions. This might also account for differences in dN/dS between core (non-recombining) and non-core (recombining) components of the genome. Unlike the time dependence model, there is no obvious reason why these effects should be restricted to very short evolutionary time-scales (within single clones).
Here we revisit the genome-wide ST239 SNP data for S. aureus [16] and C. difficile [18] to examine the impact of recombination on the proportion of synonymous and non-syonymous SNPs. For S. aureus, we confirm a higher rate of recombination and a highly significant enrichment of synonymous changes in the non-core. Further, we note a striking time dependence of dN/dS in non-core genes, whereby the more diverged elements (older SNPs) correspond to lower dN/dS ratios. However, our results also point to a moderation of the efficiency of purifying selection (i.e. increased drift) in core genes of S. aureus ST239, possibly owing to ecological specialisation and a reduction in the effective population size. This difference between core and non-core is not evident when more diverged (interclonal) comparisons are made, which argues against a strong effect from selection at linked sites. We note a similar enrichment of synonymous SNPs in core genes which have experienced homologous replacement, both in S. aureus and in Clostridium difficile. This confirms that the effect is not limited to S. aureus, and controls for the possibility of differing selection pressures acting on the core and non-core.
Based on the core definition of Harris et al [16] (described in Methods), and including intergenic SNPs, we note a total of 4250 core SNPs (affecting 1492 core genes), and 2459 non-core SNPs (affecting 257 non-core genes) within 63 isolates composing the ST239 dataset. In total, there are 2562750 bp within the core region and 454915 bp within the non-core region, hence the overall SNP densities are 1.658×10−3 SNPs per site for the core, and 5.405×10−3 SNPs per site for the non-core. The higher SNP density in the non-core reflects the high proportion of mobile (“extra-chromosomal”) elements, including prophage, genomic islands and other MGEs (Figure 1) [14]. Note that the “extrachromosal” category in Figure 1 refers to the likely horizontal origin of these genes, and all SNPs are physically located on the chromosome. Short IS elements may also be assigned as non-core, as will regions that have encountered small deletions (Figure 1).
In order to check that the data is not overly biased by the presence of a large number of non-core genes in a small number of isolates, we checked how many non-core genes were present in only one strain, any two strains, three strains and so on. As many non-core genes are partially mapped, we took both the most inclusive definition of gene “presence” (at least one mapped read within the CDS), and the most exclusive definition (100% of the CDS is mapped) (supplementary Table S1). Plotting the cumulative proportions of non-core genes present in 1, 2 …63 isolates based on these two definitions revealed that essentially all non-core genes were present in at least 13 isolates, and >50% of non-core genes were present in at least 40/63 isolates (supplementary Figure S1).
Harris et al noted that recombination has been rare in the core, a view supported by the paucity of homoplasies [16]. In contrast, the frequent horizontal transfer, mosaic structure and modular nature of mobile elements in the non-core is consistent with frequent recombination [43]. To directly compare levels of recombination in the core and non-core, we constructed neighbour-net networks, as implemented in Splitstree 4 [44], based on core and non-core SNPs separately (Figure 2, A and B). Whereas no reticulation is apparent in the core genome (Figure 2A), confirming low rates of recombination, extensive reticulation is noted for the non-core, consistent with high rates of recombination (Figure 2B). This impression is borne out by the phi test which provided significant evidence of recombination for the non-core (P<0.001), but no significance evidence for recombination for the core (P>0.1).
Table 1 gives the total numbers and percentages of synonymous, non-synonymous and intergenic SNPs in the core and non-core. Whereas there are more than twice as many non-synonymous than synonymous SNPs in the core genome, for the non-core the reverse is true. A chi-squared test confirms that the core is highly significantly enriched for non-synonymous changes, relative to the non-core (χ2 = 719.325, p<0.00001). Given higher rates of recombination in the non-core, this observation provides a simple explanation as to why there is more reticulation when networks are reconstructed from all synonymous SNPs as compared to all non-synonymous SNPs (Figure 2, C and D). The non-core therefore differs from the core in three respects: i) a higher SNP density, ii) a higher rate of recombination and iii) a higher proportion of synonymous change.
We expanded this analysis by dividing the non-core SNPs into two approximately equal subsets; dispersed (<5 SNPs per 100 bp; n = 1469) and clustered (>5 SNPs per 100 bp; n = 986). We note over twice as many non-synonymous SNPs in the dispersed set (n = 476) compared to the clustered set (n = 207). In contrast, the numbers of synonymous SNPs are almost identical in the two subsets (694 and 695 respectively). This demonstrates that a higher proportion of synonymous SNPs are noted within regions of increased SNP density, even when only considering different non-core genes (χ2 = 72, p<0.005; Table 1; Supplementary Figure S2). We note that the non-core regions of high SNP density tend to correspond to the mobile elements (e.g. categorised as “extrachromosomal” in Figure 1). Non-core genes not included in this category are less SNP dense and show approximately equal numbers of non-synonymous (n = 148) and synonymous (n = 144) SNPs.
To complement the analyses described above we used BEAST to examine the degree of rate variation for the core and non-core SNPs. This approach provides an additional means to examine the relative strengths of selection on the core and non-core genome. In the absence of strong selection we would not expect significant rate variation between synonymous and non-synonymous SNPs. In contrast, strong purifying selection would decrease the rate of change for non-synonymous SNPs, resulting in site variation. Model selection was carried out as described in Methods, and the Akaike Information Criterion (AIC) and Bayesian Information Criterion gave essentially identical results. The best DNA substitution scheme for the core data set was TVMef, whereas the non-core selected the TVM scheme (these schemes are identical except the former assumes equal base frequencies). Far greater rate variation among sites was found in the non-core data than in the core, which is expected as non-core SNPs represent imports from multiple donor lineages. None of the robustly supported models required correction for rate variation among sites in the core genes, whereas all the robustly supported models for the non-core data required a correction for rate variation among sites. Furthermore estimates of the alpha parameter were far lower for the non-core data (∼3.7) than for the core data (approaching 100).
Whereas neither synonymous or non-synonymous SNPs (when considered separately) required correction for rate variation in the core data, both required this correction for the non-core, and in this case non-synonymous SNPs showed more rate variation (alpha parameter = 2.86) than synonymous SNPs (alpha parameter = 4.73). Figure 3A shows the relative rates of change for synonymous and non-synonymous SNPs in the core. Although the distributions substantially overlap, the mean rate is marginally faster for the synonymous SNPs and a larger variance is evident for the non-synonymous SNPs. This is consistent with a wider range of selective consequences of non-synonymous SNPs, and the initial purging of the most deleterious class. Figure 3B shows the relative rate distributions for the synonymous and non-synonymous SNPs in the non-core. In this case the distributions are significantly non-overlapping, consistent with the selective removal of a far higher proportion of slightly deleterious non-synonymous SNPs (the mean rate and confidence intervals are given in the figure legend).
The above analyses demonstrate that core SNPs are the least densely clustered, and show a much lower proportion of synonymous change than the non-core. As core SNPs are likely to have emerged more recently (on average) than non-core SNPs, this is consistent with the time dependency of dN/dS noted previously between genomes of differing levels of divergence [34]. However, it is not clear whether this time dependency is sufficient to explain the difference between the core and non-core or whether other factors are playing a role. For example, it is possible that a reduction in the genome-wide effective population size may have resulted in increased drift on core genes. Low rates of recombination also decrease the local population size by increasing background selection, and this could contribute to the relative paucity of neutral variation in core genes.
In order to disentangle these different effects we first controlled for time dependence by comparing core and non-core genes at similar levels of divergence. We calculated dN/dS for all 1953 pairwise comparisons of the 63 isolates, separately for the core and the non-core regions. For each pair we estimated divergence time by calculating the divergence at synonymous sites (again for the core and non-core separately). Figure 4, main panel, plots the average synonymous site divergence against the average dN/dS for 39 bins, each of 50 pairwise comparisons. This plot confirms the high dN/dS ratio in the core, relative to the non-core, and reveals that the maximum synonymous SNP density in the core is over an order of magnitude lower than that of the non-core. Figure 4, bottom left, rescales the plot in order to clarify the patterns in the core data and the most conserved 5 bins of the non-core, where levels of divergence overlap. We note that the most conserved bins for both the core and non-core correspond to a dN/dS ratio approaching parity, thus the most recent mutations have been subject to very little purifying selection, regardless of whether they emerged in core or non-core genes. However, whereas the subsequent decrease in dN/dS for the non-core is striking, and closely fits a power law (R2 = 0.96) (Figure 4, bottom right), the plot for core genome follows a very shallow trajectory. This strongly suggests that time dependence alone cannot explain the differences in dN/dS between the core and non-core genomes. A decrease in the efficiency of purifying selection owing to a reduction of the effective population size might explain the relatively slow purging of non-synonymous mutations in the core [34]. This weakened selection may have resulted from the rapid global spread of ST239, combined with specialised adaptation to the hospital environment. Such an ecological shift would disproportionately affect the core SNPs as these are more likely to have emerged de novo since the emergence of the clone. The high rates of recombination experienced by non-core genes might also act to buffer against increased drift acting elsewhere on the genome by maintaining a higher local effective population size.
It is well documented that in eukaryotic genomes low rates of recombination are associated with low levels of neutral variation [45], [46], [47], [48], [49], and that recombination facilitates protein adaptation [50]. An important mechanism underlying this is background selection, whereby the selective purging of deleterious mutations results in the loss of neutral variation at linked sites [42]. The emergence of an adaptive mutation has a similar effect through hitch hiking [41], and in both cases the size of the genomic region affected is determined by the local rate of recombination [40]. The effect will be stronger in lowly recombining regions of the genome where more neutral variation remains linked to the site under selection. The role of positive selection on core genes should also be considered [29], along with the possibility that recombination is mutagenic [51] (although it is unclear in this latter case how this can explain the strong enrichment of synonymous change in recombining genes).
When considering these alternative hypotheses, it is important to emphasize that the analyses described thus far considers isolates within a single clone, which probably emerged under 50 years ago [16] and thus corresponds to a tiny fraction of the diversity in the species as a whole. Our hypothesis of time dependence is broadly distinct from the alternative explanations listed above, in that it predicts that the high proportion of non-synonymous change in core genes should be substantially moderated over greater levels of divergence, such that the differences in dN/dS between core and non-core genes should diminish. Comparing a more representative sample of S. aureus genomes therefore provides a simple means to test our model of time dependence against these alternative hypotheses. To this end, we calculated the dN/dS ratio between core orthologs in TW20 (see Methods) and three other S. aureus genomes representing a range of divergence times: i) within the same clonal complex (TW20 vs USA300), ii) within different clonal complexes (TW20 vs MRSA252), and iii) highly divergent (TW20 vs MSHR1132). This latter genome corresponds to the unusually diverged S. aureus genotype ST75 which has been recorded in Northern Australia [52], Cambodia [53] and French Guiana [54].
For each of the three pairwise comparisons we calculated the mean dS and dN/dS, along with a standard error calculated by re-sampling the data as described in Methods. Note that we omitted those core genes corresponding to the large replacement region as described below. Figure 5 confirms that the dN/dS between core genes decreases with increasing synonymous site divergence. The mean dN/dS ratio for the intermediate comparison (TW20 vs MRSA252), which provides the most representative comparison for the species as whole, is below 0.1. We note this is lower than the average dN/dS observed for the non-core regions within the ST239 clone (0.18). This suggests that the difference between core and non-core genes has not only been diminished, but has reversed, in that non-core genes show a higher dN/dS than core genes when greater divergence times are considered. To confirm this we also calculated dN/dS separately for 57 orthologous non-core genes, identified on the basis of reciprocal Fasta best hits (as described previously [14]), between the most divergent comparison (TW20 vs MSHR1132) (Supplementary Table S2). The average dN/dS for these non-core genes is ∼0.1, which is over double the average of ∼0.04 noted in the core genes for the same comparison. Thus the relative inflation of neutral variation associated with non-core genes within the ST239 clone is only apparent over very short phylogenetic distances, and this argues against a strong role for background selection, positive selection on core genes or the mutagenic effects of recombination as alternative explanations.
The unique hybrid structure of the TW20 genome presents an opportunity to test whether the differences we observe between the core and non-core genes result from differing selection pressures on these two sets of genes. This large (635 kb) homologous replacement in the ST239 genome has affected many core and non-core genes, but for this analysis we only consider core genes where orthologs can be robustly identified in both parental genomes (see Methods). We compared patterns of dS and dN/dS within and outside the replacement region using the full genome sequences corresponding to the recipient clone (USA300; ST8) [55], the donor clone (MRSA252; ST36) [23] and the resultant hybrid (TW20; ST239) [14]. In order to gauge significance, we used a re-sampling procedure as described in Methods.
Overall patterns of divergence (dS) confirm that the replacement region within TW20 (hybrid) is much more similar to MRSA252 (the donor) than USA300 (the recipient), whilst the reverse is true for the rest of the genome (Supplementary Figure S3). When TW20 and USA300 are compared, the dN/dS ratios are much lower within the replacement (∼0.12) than the rest of the genome (∼0.33) (Figure 6). This is entirely expected, as the replacement represents an import from a diverged lineage so, similar to the non-core, should be relatively enriched for synonymous changes. The link between inflated neutral diversity and recombination therefore holds similarly for core genes which have been replaced by diverged homologous imports as it does for non core genes, and is therefore unlikely to be related to gene specific selection pressures. As a further check we confirmed that when the TW20 (hybrid) and MRSA252 (donor) genomes are compared the reverse is true; the dN/dS ratio within the replacement is higher (0.23) than the rest of the genome (0.08) (Figure 6).
The analysis above demonstrates that diverged homologous replacements may result in a relative local enrichment of synonymous change. In order to check whether a similar pattern could be observed in other species, we used the genome-wide SNP data for 25 isolates belonging to a single hypervirulent clone of Clostridium difficile presented by He et al [18]. These data are well suited to this analysis, as two of these isolates (bi11 and bi4) exhibit regions of high SNP density consistent with large-scale homologous recombination from outside of the clone. Following He et al. [18] we assign these blocks of high density SNPs as having arisen by homologous recombination. These blocks of recombination account for the vast majority (89.4%) of all the SNPs detected within all the 25 isolates of this clone. Whereas the average SNP density across all strains outside of these blocks was only 6×10−5 SNPs per site, all the blocks correspond to a SNP density at least an order of magnitude higher than this, and the average SNP density within the blocks was 1.4×10−3 SNPs per site. Because the recombination blocks within strains bi11 and bi4 correspond to such striking peaks of SNP density, we simply identified all the 1553 SNPs within these two strains likely to have arisen by recombination by visual inspection of the SNP alignment (Supplementary Figure S4). We then compared the number of synonymous, non-synonymous and intergenic SNPs within the blocks with the remaining 184 SNPs (Table 2). Consistent with expectation, a significant enrichment of synonymous SNPs was observed for those changes corresponding to the regions of densely clustered SNPs, and therefore assigned as having arisen by recombination (χ2 = 34, p<0.005).
Here we have exploited four complete S. aureus genome sequences, and revisited the genome-wide SNP datasets of S. aureus [16], and Clostridium difficile [18] to examine how recombination impacts on the level of neutral variation within recently emerged bacterial clones. A key starting point is the high level of non-synonymous change among very recently emerged mutations, and a commensurate decrease in dN/dS over divergence time [18], [34], [37], [38], [56], [57], [58], [59]. For S. aureus, which is a highly structured (clonal) population, this is evident as a preponderance of non-synonymous change within clones (dN/dS∼0.7) compared to between clones (dN/dS∼0.1) [23], [24]. Given this framework, it is clear that the importation of DNA into a bacterial clone from elsewhere is predicted to introduce a relative preponderance of synonymous change, and this is strongly supported by our data. Furthermore, we note that the proportion of synonymous change increases as the donor lineage is more divergent. This confirms that the relative enrichment of synonymous SNPs over time equally applies to genes acquired from disparate lineages within a single genome as it does to overall levels of divergence between pairs of genomes. The relationship of dN/dS to synonymous site divergence in the non-core fits closely to a power law (R2 = 0.96). Although more work is required to elucidate the properties of these plots for different datasets, we note that the decrease in dN/dS over greater scales of divergence (between clones; Figure 5) appears to show a broadly similar relationship.
Whilst it is clearly necessary to account for the time-dependence of dN/dS when comparing the strength of selection between genes, genomes or populations over very short time-scales, this model does not fully explain the difference we observe between the core and non-core genes in S. aureus. Even when controlling for this effect (i.e. comparing genes at the same level of divergence) the core genes show an enrichment of non-synonymous SNPs relative to the non-core. Rocha et al showed that differences in the trajectory of the decrease in dN/dS over time can be explained by changes in the effective population size [34]. This effect has been examined in detail by comparing trajectories of dN/dS over time in E. coli and highly specialised Shigella clones [36], and the impact of increased drift through bottlenecking has also been discussed in Mycobacterium bovis [60].
It is possible that the epidemiological characteristics of ST239 may reduce the effective population size, thus weakening the efficiency of purifying selection. Since its emergence in the mid 1960s this clone has disseminated globally, and it has been estimated that it currently causes >90% of all cases of hospital-acquired MRSA within regions which together account for >60% of the global population [7]. In contrast to methicillin sensitive S. aureus (MSSA), which are typically carried asymptomatically in the community [61], ST239 is very rarely noted outside of the hospital environment. Direct transmission between hospitals must therefore play a large role in the global dissemination of this clone, and this mode of dissemination will incur substantial bottlenecking if very limited variation is introduced into any given hospital. Commenting on the large number of impressively reinforced, yet extinct, species in the fossil record, Haldane remarked that “… in some cases the species literally sank under the weight of their own armaments” [62]. In the case of S. aureus ST239, which has already been replaced in Western Europe and is currently being replaced in other parts of the world [63], [64], the “armament” of multiple drug resistance may be costly in terms of cell function and resources, but also in terms of restricting the competitiveness of the clone to those health care settings where antibiotics are most aggressively deployed.
The argument above has two important implications. First, it raises the possibility that rapidly emerging clinically important clones which are exclusively maintained in health-care settings are inevitably self-limiting. This may help to account for the cycles of clonal expansion and replacement commonly noted within S. aureus and other pathogen populations [65]. It will be interesting to examine this further by comparing the trajectories of dN/dS over time in samples representing different ecological constraints and effective population sizes. Second, this analysis provides a novel approach for detecting increased drift through comparisons between core and non-core genes within a single genome. This contrasts with whole genome comparisons of different lineages as carried out previously for E. coli and Shigella [36]. The current study also illustrates the importance, and potential, of considering evolutionary processes within the context of the age and history of different genomic regions, rather than purely in terms of direct selective effects. Such a perspective underpins studies on the likely fate of acquired genes [66], [67], [68] or the effects of age on the functional and selective stability of proteins [69].
To what extent can selection at linked sites, leading to variation in local effective population size (as determined by the rate of recombination), explain our data? Although background selection [42] has been discussed extensively for eukaryotes, its role in bacterial evolution remains almost completely unknown. Touchon et al demonstrated a lower rate of recombination, a lower value of Tajima's D and a higher dN/dS ratio around the terminus of replication in E. coli [67]. These authors interpreted the co-occurrence of low rates of recombination and an enrichment of slightly deleterious change as evidence of background selection. Whereas we have argued that recombination introduces neutral variation into the genome from a diverged lineage, an advocate of background selection would argue that recombination has saved much of the neutral variation that has emerged de novo within the genome which would otherwise have been lost. As illustrated by Figure 4, main panel, this would need to be a very powerful force because the maximum level of neutral variation in the non-core of the ST239 data is at least an order of magnitude greater than that in the core.
We note in this context a broad distinction between recombination in eukaryotes, where the evolutionary consequences tend to be viewed in terms of the process itself (e.g. background selection, biased gene conversion or the Hill-Robertson effect), and recombination in bacteria, which (because it is less common, but can occur over large phylogenetic distances) is typically more simply viewed in terms of how the incoming SNPs (or genes) directly impact on the genome. We argue that our primary observation –the relative enrichment of synonymous SNPs within recombined regions at intra-clonal (but not inter-clonal) scales– is most parsimoniously explained by this latter perspective. Although the efficiency of purifying selection may have been compromised within the ST239 clone, our analysis confirms the decrease of dN/dS on the core genome over longer evolutionary time-scales which encompass the variation within the species. This means that the difference in the proportion of neutral variation between the non-recombining core and the recombining non-core disappears, or is even reversed, after the very initial (intra-clonal) stages of diversification. It is not obvious how to reconcile this observation with background selection, although we do concede the possibility that expansion of the “population” in this way may have unpredictable consequences concerning the relative effective population sizes of highly and lowly recombining regions. If positive selection played a major role in the emergence of high levels of non-synonymous changes in the core, or if the putative mutagenic effect of recombination enhanced neutral variation in the non-core, then we would expect these patterns to be maintained over greater levels of divergence.
The analysis of the large homologous replacement in TW20 suggests that it is not the process of recombination per se which acts to inflate neutral variation, as this effect is apparent only in those cases where the imported region (MRSA252-like) is more divergent than the two comparator genomes (USA300 and TW20) are to each other. If this is not the case, then recombination will have the opposite effect and remove neutral variation that has accumulated between the parental lineages (this applies when comparing TW20 with MSSA252). Finally, as this analysis is based on the core genes, the differences we observe in the SNP data are unlikely to be due to differing selective constraints acting on core and non-core regions.
One curious observation remains. For both the S. aureus and C. difficile data, the percentage of intergenic SNPs is lower when the changes are assigned as having been acquired horizontally from outside the clone (non-core or recombined; Tables 1 and 2). In the case of the non-core S. aureus data, this might be partly explained by difficulties in mapping very short and diverged intergenic sites in phage. However, this cannot explain the C. difficile data, where we note there is no significant difference between the numbers of intergenic and non-syonymous SNPs in the recombined and non-recombined datasets (in contrast, the difference between intergenic and synonymous SNPs is highly significant). Assuming that intergenic and synonymous SNPs are approximate selective equivalents, at least compared to non-synonymous SNPs, then this is the opposite trend to that expected. The reasons for the relative paucity of intergenic (compared to synonymous) SNPs within the non-core (S. aureus) or recombined (C. difficile) datasets may hint at selection on intergenic sites, and this observation clearly warrants further attention.
In conclusion, here we demonstrate that the dN/dS ratio varies according to the level of divergence and the past history of recombination between different genes, as well as due to population level effects associated with ecological specialisation. Whilst further work is required to elucidate the possible effects of selection at linked sites in natural bacterial populations, it is clear that variation imported into recently emerged bacterial clones from diverged lineages will contain a relatively high proportion of synonymous SNPs. This effect is not restricted to non-core genes, and the analysis on the C. difficile data demonstrates it extends to species other than S. aureus. As imported SNPs have been pre-filtered by purifying selection, so one might expect to see a general increase in the relative rate of recombination moving backwards in the tree as de novo mutations will tend to be purged more rapidly than recombination events. Recent studies have discussed the rapid rate of mutation at the very tips of the trees, which decreases over time as mutations are purged [35], [70], but it is not yet clear whether the rate of recombination shows a similar decrease. More practically, we propose that the enrichment of local synonymous change within bacterial clones might be used as an additional diagnostic to identify recombination events, thus facilitating detailed studies into their size and frequency or, through their subsequent removal, more robust phylogenetic analysis. Finally, although this work has focussed on the selective removal of slightly deleterious non-synonymous changes, future studies might also consider possible selective costs of SNPs at synonymous and intergenic sites [36], [71], [72], [73].
We used the core and non-core SNP datasets as previously defined by Harris et al [16]. The core genome was identified conservatively and objectively, as all sequences >1 kb that are present in all 63 isolates. Note that non-core regions present in the query strains, but absent in the reference sequence (TW20), are excluded. Rather than representing novel elements, the non-core SNP data in this study corresponds to the allelic variation of the accessory elements present in TW20 (e.g. polymorphisms present between closely related variants of the same non-core element). We computed the proportion of synonymous, non-synonymous, and intergenic SNPs for both core and non-core data sets using scripts written in Python.
We used four alignments in these analyses: i) all the synonymous and non-synonymous SNPs located in the core regions, ii) all the synonymous and non-synonymous present in the non-core regions, iii) all the synonymous SNPs irrespective of their presence in core or non-core regions, iv) all the non-synonymous SNPs regardless of their presence in core or non-core regions. For each alignment neighbour-nets were constructed using uncorrected p distances and were drawn using the Equal Angle method. Additionally, the Phi test for detecting recombination was conducted on all 4 alignments. All these analyses were carried out using SplitsTree 4 [44].
Statistical selection of models of nucleotide substitution was carried out using jModelTest [74] on an alignment of all SNPs for each data set, and on separate partitions (see below). Likelihood scores were computed for the different models. Corrections for unequal base frequencies and rate variation among sites were allowed, but we did not consider the correction for invariable sites as all SNPs are, by definition, variable. The Akaike Information Criterion (AIC) and Bayesian Information Criterion were used to perform model selection. The models used were: [TVMef SYM TVM GTR TVMef+G SYM+G TVM+G GTR+G TPM1 TIM1ef TPM1uf TIM1 K80 TrNef TPM1+G HKY TIM1ef+G TrN TPM1uf+G TIM1+G K80+G TrNef+G HKY+G TrN+G F81 JC F81+G JC+G].
We used the Bayesian methods implemented in BEAST [37] to estimate the rate of evolution for core and non-core SNPs. As the date of isolation was known for each single strain, we could calibrate the inferred phylogenies. In order to estimate relative rates, two data partitions (one consisting of the synonymous SNPs and other made up of non-synonymous SNPs) were used for the core and non-core data. These partitions were set up by editing the BEAST XML input files, as described in the BEAST manual (beast-mcmc.googlecode.com/files/BEAST14_Manual_6July2007.pdf). Since biological data sets will best fit a relaxed molecular model, which assumes independent rates on different branches, rather than a strict clock model, we used such an approach. The uncorrelated log-normal relaxed clock model was employed, using the GTR model with a gamma distribution for rate variation among sites. Substitution rate, rate heterogeneity, and base frequencies parameters were unlinked across partitions, otherwise default priors were used. For each analysis, one chain was run for 20,000,000 steps, and samples were taken every 2,000 steps. The first 2,000 steps were discarded as burn-in, and convergence was evaluated through TRACER, by examining the effective sample size (ESS) of the mean substitution rates and by examining trace plots of the likelihood scores.
The genomes used were those of S. aureus TW20 (hybrid), S. aureus MRSA252 (donor), and S. aureus USA300 (recipient). We identified 2207 orthologues present in all three genomes by a reciprocal Fasta analysis performed previously [14]. Of these, 36 were discarded as pseudogenes leaving 2171 genes. The co-ordinates of the large homologous replacement (with respect to the TW20 genome) were taken as position 1… 427725 and position 2848037 … 3043210 (note there is only one contiguous block, but this passes through the origin of replication so is located at either end of the linear sequence). All orthologous genes which fell completely within these boundaries were assigned as REC (357 genes), whilst those orthologues which fell outside were assigned as NON-REC (1755 genes). Note that many of the genes falling within these boundaries in TW20 belong to non-core elements (e.g. SCCmec, SCCmer, prophage φSa1, Tn552 and ICE6013), and as orthologues of many of these CDSs could not be identified in both MRSA252 and USA300 they were excluded. The exclusion of these non-core genes also allows us to test for gene specific effects between core and non-core genes. Individual alignments were made for each of the remaining genes as follows: first, using the protein sequences which are the translations of the genes we created protein alignments through MUSCLE [75] and, then, we used the program TRANALING, from The European Molecular Biology Open Software Suite [76], to generate alignments of the nucleic coding regions from the protein alignments (this was done in order to have DNA alignments in frame). To gauge the significance of differences in dS and dN/dS between the recombinant region and the rest of the genome, and to control for the different sizes of the datasets, we used a re-sampling procedure. We randomly sampled (with replacement) gene alignments until their concatenated length exceeded 378747 bp (the concatenated length of the 372 core genes of the recombinant region). This was repeated 200 times for the recombinant region and 200 times for the rest of the genome. For each replicate, pairwise dS and dN/dS values were computed from the concatenated alignments using the codeml program in PAML [77] to estimate synonymous substitution rates (dS), non-synonymous substitution rates (dN), and the ratio of the two (dN/dS). We specified “runmode = −2” in the control file to set pairwise calculations.
In addition to the 3 genomes used for the analysis of the large replacement, we included the genome of S. aureus MSHR1132 that is distantly related to the other 3 genomes. In this analysis we kept only those orthologues located in the non-recombinant region (NON-REC orthologues, n = 1755 genes, see above). Using these orthologues, we generated 200 replicates of concatenated alignments following the re-sampling procedure mentioned above. Pairwise dS and dN/dS values were then computed for each replicate through PAML as previously described.
|
10.1371/journal.pgen.1000527 | A Network of Conserved Damage Survival Pathways Revealed by a Genomic RNAi Screen | Damage initiates a pleiotropic cellular response aimed at cellular survival when appropriate. To identify genes required for damage survival, we used a cell-based RNAi screen against the Drosophila genome and the alkylating agent methyl methanesulphonate (MMS). Similar studies performed in other model organisms report that damage response may involve pleiotropic cellular processes other than the central DNA repair components, yet an intuitive systems level view of the cellular components required for damage survival, their interrelationship, and contextual importance has been lacking. Further, by comparing data from different model organisms, identification of conserved and presumably core survival components should be forthcoming. We identified 307 genes, representing 13 signaling, metabolic, or enzymatic pathways, affecting cellular survival of MMS–induced damage. As expected, the majority of these pathways are involved in DNA repair; however, several pathways with more diverse biological functions were also identified, including the TOR pathway, transcription, translation, proteasome, glutathione synthesis, ATP synthesis, and Notch signaling, and these were equally important in damage survival. Comparison with genomic screen data from Saccharomyces cerevisiae revealed no overlap enrichment of individual genes between the species, but a conservation of the pathways. To demonstrate the functional conservation of pathways, five were tested in Drosophila and mouse cells, with each pathway responding to alkylation damage in both species. Using the protein interactome, a significant level of connectivity was observed between Drosophila MMS survival proteins, suggesting a higher order relationship. This connectivity was dramatically improved by incorporating the components of the 13 identified pathways within the network. Grouping proteins into “pathway nodes” qualitatively improved the interactome organization, revealing a highly organized “MMS survival network.” We conclude that identification of pathways can facilitate comparative biology analysis when direct gene/orthologue comparisons fail. A biologically intuitive, highly interconnected MMS survival network was revealed after we incorporated pathway data in our interactome analysis.
| Cellular damage is known to elicit a pleiotropic response, but the relative importance of the constituent components in cell survival is poorly understood. To provide an unbiased identification of the proteins utilized in damage survival, we performed an RNAi survival screen in fly cells with methyl methanesulfonate (MMS). The genes identified are involved in 13 biologically diverse pathways. Comparison with analogous yeast data demonstrated a lack of conservation of the individual MMS survival genes but a conservation of pathways. We went on to demonstrate the MMS responsiveness for five representative pathways in both fly and mouse cells. We conclude that identification of pathways can facilitate comparative biology analysis when direct gene/orthologue comparisons fail. Incorporation of pathway data in interactome analysis also improved connectivity and, more importantly, revealed a biologically intuitive, highly inter-connected “MMS survival network.” This pathway conservation and inter-connectivity implies extensive interaction between pathways; for diseases such as cancer, such crosstalk may dictate disparate cellular responses not necessarily expected and confound treatments that are not tailored to the individual molecular context.
| Cellular damage is a normal component of life, with constant damage exposure from both endogenous and exogenous sources. Damage to DNA is considered to be the most biologically relevant lesion with the potential of mutagenic results, though most exogenous agents have the potential to damage many components of the cell. Responding appropriately to such insults, either mitigating cellular toxicity or initiating an appropriate cell death response, is critical, particularly in multi-cellular organisms. Inappropriate responses may facilitate deleterious effects, such as a destabilized genome and diseases such as cancer [1]. As such, DNA damage response (DDR) components are critical suppressors of deleterious effects of genotoxic agents by controlling cell cycle progression, DNA repair, and apoptosis [2]. For this reason, there have been many investigations using a variety of model organisms to identify components of DDR and subsequently to determine how they function and the consequences of their dysfunction. Recent reports suggest that DDR may involve pleiotropic cellular processes other than the central DDR components [3],[4], yet an intuitive systems level view of the cellular components required for damage survival, their interrelationship, and their contextual importance has been lacking. The most comprehensive attempt at understanding the interrelationship of damage response and survival components in yeast at a systems level has been mapping identified genes to a network by integrating the general biological processes as distinct modules [5]. It has been suggested that the inclusion of well-defined biological pathways in protein networks might provide a better understanding of biological interactions therein [6],[7].
In order to identify genes involved in damage response in an unbiased manner and to put them in a functional context, we used an RNAi library [8] to knock-down every predicted protein in the Drosophila melanogaster genome and assessed whether knock-down of individual proteins significantly altered cell viability following methyl methanesulfonate (MMS) exposure. MMS is a prototypical alkylating agent that attacks nucleophilic groups, such as those found in nucleic acid [9]. The resultant base methylation destabilizes the glycosidic bond, facilitating the production of the most biologically relevant cellular lesion of MMS, an abasic site in DNA. Other common sources of alkylation damage include endogenous S-adenoysyl methionine [10], the tobacco carcinogen N-nitorsoamine [11] and chemotherapeutics such as temozolomide [12], carmustine (BCNU) [9], and cyclophosphamide [13]. Considering the physiological relevance of alkylation damage, several recent studies using yeast as model organism investigated the global effects of alkylation damage induced by MMS. Viability [14], gene transcription [15] and protein expression [16] were measured in these studies, all of which provided insights into the diverse nature of biological responses to alkylation damage. However when transcriptional responses to various environmental stresses were compared between mammalian and yeast cells [17], Murray et al. reported clear distinctions and suggested that this might be the result of different selective pressures between multi-cellular organisms and single-cell organisms. In the attempt to understand diverse responses required for damage survival in a system that is evolutionarily closer to mammals than yeast, we performed a genome-wide, RNAi-based screen using cells derived from Drosophila melanogaster. Our experiments were based on “loss of function” and an assessment for cellular viability following exposure to MMS. Here we present results from this MMS survival screen, the pathways that were identified, and a comparative analysis of pathways to understand conservation of these pathways in yeast and mouse cells. Additionally, we present a novel approach of assembling a protein interaction network based on defined biological pathways, which facilitates network consolidation. By including biological pathways in our genomic data and protein network analysis, we were able to determine the commonality across species, which was obscured by direct orthologue comparison, and provide a simplified representation of the global survival responses that we identified.
An RNAi screen was designed to identify those genes that modulate cellular survival following exposure to a level of MMS-induced damage that resulted in only a limited amount of cell death. A linear decrease in cell viability was observed from 0.002% (w/v) of MMS to 0.008% (w/v) of MMS (Figure S1). A dose of 0.004% (w/v; 40 µg/mL) MMS was chosen, which resulted in a statistically significant decrease to approximately 65% viability, while allowing an additional, measurable decrease in cell survival. RNAi screens were performed using the Drosophila RNAi Screening Center (DRSC) version 1 library (about 23,000 D. melanogaster open reading frames) [8]. Kc167 Drosophila cells were exposed to three days of dsRNA to allow protein knock-down, followed by three days of growth in either the absence or presence of MMS exposure (Figure 1A).
To identify those genes required for cell viability following MMS treatment, viability results from the MMS treated RNAi screen were compared with that of the untreated (control) screen as previously described [18]. 1,398 different open reading frames were identified that affected MMS survival in a continuous distribution of cell survival values (see [18] and http://gccri.uthscsa.edu/ABPublished_Data.asp for original data), of which 996 had a unique assigned FlyBase gene number (FBgn; denotes known genes) (see Figure 1B and Table 1). Of these 996 genes, the top 537 were selected for validation analysis by a previously described, stringent validation method [18]. Whereas in [18] we validated normalization methods, here we independently validated individual genes using dsRNA targeting a different region of each gene. 202 protein knock-downs validated with a significant MMS viability effect, while 55 more had a notable trend effect without meeting our stringent statistical criteria (Table S1 and Table S2). An examination of gene ontology (GO) on the 202 validated MMS survival genes revealed a significant enrichment for genes involved in DNA metabolism, gene transcription, and cellular metabolism (Table 2). The overall variety of GO categories observed was quite broad, consistent with the findings reported for an analogous yeast screen [3] (data not shown). Surprisingly though, no significant enrichment between the gene orthologues for the two organisms was observed (G-test p-values of 0.29 and 0.057 or Fisher Exact test p-values of 0.37 and 0.08, yeast and fly, respectively). To further test this, we examined the MMS sensitivity following knock-down of 183 fly genes that were orthologues of to 118 yeast MMS survival genes [3], but only found conservation of MMS survival phenotype with 20 (Table S3). This apparent lack of gene enrichment between species has also been reported when comparing transcriptional profiles between mammalian and yeast cells in response to a variety of stresses [17]. Overall, these results suggest that there may be conservation of the biological processes that respond to MMS rather than the individual genes.
Assuming that biological function is a more informative measure of damage response than a requirement of individual genes, we therefore endeavoured to identify those MMS survival proteins within known signaling, metabolic, and enzymatic pathways. Using both a priori knowledge and KegArray, we identified 13 pathways that together included 41 MMS survival proteins (Figures 1, 2, 3 and Figures S2, S3, S4, S5), among them, five DNA repair pathways (Base Excision Repair; Nucleotide Excision Repair; Mismatch Repair; Homologous Recombination Repair; and RECQ). Many of these pathways have a statistically significant number of MMS survival proteins (Table S5), and this is without accounting for pathway “branching” or “subdivision.” Considering the likely role of these pathways in MMS survival, we used an RNAi assay method, which is both more sensitive and stringent than the original screen [18], to determine whether other members of the 13 identified pathway also affected MMS survival (Table S1). By individually interrogating the 346 pathway members of the 13 pathways that were not identified in the original screen, an additional 105 MMS survival genes were discovered. This observation significantly enriched the number of MMS survival genes in each of the 13 pathways (Figures 1, 2, 3 and Figures S2, S3, S4, S5) and provides additional support to the hypothesis that each of the identified pathways are involved in MMS survival. Though this result highlights the possibility of false negative screen data, compared to our previous false negative rate of 23.6% from randomly selected data, we observe a significant enrichment for false negatives within pertinent pathways ((χ2 p = 3.9E-5). In total we identified 146 MMS survival genes in 13 “MMS survival pathways,” encompassing 25% to 86% of all non-essential genes within each pathway (Table S4).
Using the genes we mapped to the MMS survival pathways, we then compared pathways identified from our Drosophila MMS screen with the analogous screen performed in yeast [14] using genes they found to be responsive to MMS. As previously noted, we did not observe a significant overlap between the two screens when comparing MMS survival gene orthologues, however, with this pathway comparison, we mapped orthologues of yeast MMS survival genes to 10 of the 13 Drosophila MMS survival pathways (Figures 2 and 3; Figures S2, S3, S4, S5; Table S6). The three Drosophila MMS survival pathways without yeast MMS survival genes either are not conserved in yeast or have almost all of the pathway components are essential for viability in yeast, thus making them refractory to MMS viability analysis (Table S5). The gene enrichment with each pathway between species is highly supportive of a conservation of processes involved in MMS survival. A notable absentee in the Drosophila screen, which was observed in yeast, is 3-methyladenine-DNA-glycosylase (MAG1/AlkA), the principal protection against MMS-induced DNA damage in yeast, but no direct orthologues exist in animals. One of the two glycosylase found in Drosophila that act at the same step in BER, namely Thd1, was found to be required for survival after MMS treatment. Furthermore, several of the pathways observed in both species, such as proteasome [19], the TOR pathway [20], and DNA repair pathways [21], have been shown to be functionally responsive to MMS in yeast. Altogether, these results suggest a conservation of pathway function, if not individual genes, in response to MMS between species.
To demonstrate that the identified Drosophila pathways are functioning in the expected manner in response to MMS, five were selected for further examination – base excision repair (BER), DDR, glutathione metabolism, proteolysis by proteasomal degradation (proteasome), and the TOR pathway. Two of these pathways, BER and DDR, are expected to play a role in MMS survival [9], while the others were selected for their apparent breadth of function and their non-canonical role in damage survival. For each pathway, an appropriate functional assay was chosen, and when possible, an appropriate upstream MMS survival protein within that pathway was knocked-down (Figure S6A) to demonstrate modulation of the MMS induced response.
First, we tested the two expected MMS survival pathways BER and DDR. MMS-induced DNA damage results in the production of apurinic/apyrimidic (AP) sites, a DNA damage typically repaired by BER [9]. We therefore quantified the amount of AP sites per microgram of genomic DNA following MMS treatment and observed a statistically significant increase compared to control (Figure 4A; p≤7.6E-8). Knock-down of the BER component XRCC1 resulted in an increased amount of AP sites in and of itself (p≤3E-3), but following MMS treatment, the amount of AP sites in the absence of XRCC1 was increased further (Figure 4A; compared with untreated XRCC1 knock-down, p≤3.3E-8; compared to MMS treated luciferase (Luc) control, p≤3.7E-3); MMS therefore produces the expected form of DNA damage to which BER appropriately responds. p53, a central component of the DDR pathway, is regulated at expression, protein stabilization, and posttranslational modification levels. As expected, p53 gene expression increased in response to MMS exposure (Figure 4B; p≤1.6E-3).
We then examined three MMS survival pathways that are not part of the canonical DDR and DNA repair process: glutathione metabolism, proteasome, and the TOR pathway. An increased activity of the glutathione synthesis pathway following MMS exposure was demonstrated by measuring total glutathione content per milligram of protein lysate (Figure 4C; p≤3.4E-3), as well as by examining the expression of the rate-limiting enzyme for glutathione synthesis, GCLc [22] (p≤8.8E-4). Therefore, as expected, knock-down of this same protein, GCLc, significantly reduced the total amount of glutathione present compared with control (p≤3.4E-3), but its knock-down also prevented the cells from increasing the level of glutathione in response to MMS exposure (Figure 4C). Since glutathione synthesis is considered to be an oxidative stress response, we demonstrated that MMS resulted in a dose-dependent increase in the level of 8-oxoguanine, a DNA lesion normally associated with oxidative damage (Figure S7 and Text S1). Thus it appears that MMS results in not only alkylation damage but also damage from oxidative stress.
For the analysis of the proteasome degradation pathway, we measured proteasome activity and found that it increased following MMS treatment (Figure 4D; p≤4.2E-2). Protein knock-down of the proteasome components Rpn2 or Pros26.4 significantly reduced proteasome activity compared to cells without knock-down that were either unexposed (Rpn2: p≤1.7E-2; Pros26.4: p≤3.7E-3) or exposed to MMS (Rpn2: p≤1.4E-4; Pros26.4: p≤7.0E-5). Following Rpn2 knock-down, cells were unable to mount a detectable increase in proteasome activity following MMS exposure, although we were unable to demonstrate this for Pros26.4 knock-down (p≤1.7E-3). Overall, it appears that MMS exposure results in increased proteasome activity.
Finally, to investigate TOR pathway activity, S6K phosphorylation status was monitored. TOR is a kinase that phosphorylates S6K to promote growth through ribosome biogenesis and is negatively regulated by the tumor suppressor protein Tsc1 [23]. MMS exposure resulted in a dose-dependent decrease in S6K phosphorylation, suggesting an inhibition of TOR activity (Figure 4E). This MMS-induced decrease in S6K phosphorylation was dependent on Tsc1 (Figure 4E). MMS exposure therefore elicited a down-regulation of the growth promoting TOR pathway, suggesting that this pathway is also a coordinated component of DDR similar to observations in yeast [20]. This supports a previously published observation by Matsuoka et al. [4], who showed that some components of the mammalian TOR pathway are phosphorylated following ionizing radiation exposure. In conclusion, the results of these functional assays validate the identification of the “MMS survival pathways” by RNAi screening and that the functionality of these pathways, or lack thereof, affects MMS survival.
Having thus identified 13 Drosophila MMS survival pathways, we went on to investigate their functional conservation in mammals. Using primary mouse embryonic fibroblasts (MEFs), we examined the same five pathways that were modulated following MMS exposure in Drosophila. In general we observed comparable results in MEFs to Drosophila (compare Figures 4 and 5). MMS increased the amount of AP sites in MEFs (Figure 5A); in the absence of XRCC1, however, we observed a significant increase in AP sites following MMS exposure (Figure 5A; p≤3.7E-4). For DDR in MEFs, we assessed the phosphorylation status of Chk1 (a kinase that, once phosphorylated, phosphorylates p53) [24], total p53 protein, and the phosphorylation status of p53 (Figure 5B), not to examine p53 expression levels, but to obtain a result equivalent to that of Drosophila result: that DDR is activated by MMS exposure. Also, similar to the observation in Drosophila cells, MEFs had increased proteasome activity following MMS treatment (Figure 5D; p≤2.0E-4); knock-down of either proteasome component, PSMC1 or PSMD1 (orthologues to Rpn2 and Pros26.4, respectively), resulted in decreased proteasome activity following MMS exposure (PSMC1: p≤8.1E-5; PSMD1: p≤4.1E-4). These results are comparable with Drosophila: it is clear that proteasome activity is responsive to MMS in MEFs. It should also be noted that, though we were able to demonstrate the same dose-dependent decrease in S6K phosphorylation following MMS exposure in MEFs as seen in Drosophila cells (Figure 5E) and that knock-down of TSC1 disrupted this MMS-induced effect, the disruption was not as evident as observed with Drosophila, probably due to inefficient knock-down of the TSC1 protein in MEFs (data not shown). Having observed the functionality of these pathways in response to MMS and the ability of the five siRNA to modulate the MMS response of their respective pathway, we also examined the effect of each knock-down on MMS survival in MEFs. Three of the protein knock-downs, GCLC, PSMC1 and PSMD1, affected MMS survival in MEFs (data not shown). Overall, it would appear that Drosophila could be used to accurately predict MMS survival pathways in mouse.
Given that MMS causes alkylation damage, it would be reasonable to suppose the identified biological pathways are generally involved in the response to other alkylating agents. Temozolomide is an alkylating agent used clinically in the treatment of cancer [12],[25]. It has already been demonstrated that BER and TOR are necessary for Temozolomide survival [12],[25]. Temozolomide causes DNA damage by increasing the number of AP sites, and similar to our MMS results, inhibition of BER increases sensitivity to temozolomide [12]. It has also been reported that rapamycin inhibition of the TOR pathway increased cellular sensitivity to temozolomide [25]. To further demonstrate the utilization of MMS survival pathways in response to temozolomide, we examined the DDR pathway, glutathione levels, and proteasome activity in MEFs and observed the same responses as observed following MMS exposure (Figure S8). Further, using HEK293 cells, we observed that both MMS and temozolomide exposure repressed Notch reporter activity (Figure S9 and Text S1). Absence of functional Notch protein has been shown to repress the activity of a downstream transcriptional activator (RBP-Jκ) [26], therefore this result demonstrates a functional Notch pathway response in alkylation damage exposure. Taken together, these data suggest a functional conservation of the MMS survival pathways responses with other similarly acting agents.
Of the 307 identified MMS survival genes (202 validated screen hits plus 105 from pathway analyses), we were able to assign 146 to 13 pathways (Figures 1, 2, 3 and Figures S2, S3, S4, S5). Because 161 MMS survival protein remain unassigned, we re-examined the inter-relationship between the identified MMS survival proteins using the currently available Drosophila protein:protein interactome (PPI) map [27]–[29]. To measure the connectivity among the MMS survival genes, we took the largest connected components of the PPI network and discarded the other, smaller components. The largest connected component contained 7364 of the 7504 proteins in the original PPI network. Taking the 202 original validated MMS screen hits as an unbiased starting point, we determined the number of proteins that were directly connected to one another, compared to a random set of the same number of proteins from the PPI, and observed a significant enrichment (Figure 6A and Table S7; p≤2.1E-10). Similar results were obtained for other assessments of network connectivity (Table S7). Because the MMS screen hits had more connections on average than a set of random genes, which might have biased the above analysis, we also compared the connectivity of the MMS screen hits in the real PPI network with the connectivity of the same set of proteins in a randomly rewired PPI network. Degrees were preserved in the random rewiring process. As shown in Table S7, the MMS screen hits have statistically significantly more direct interactions than would be expected in a randomly rewired network (p = 0.01). On the other hand, the randomized network had smaller average distance and higher global efficiency than the real network, which could be attributed to the well-known property of real-world networks: they usually have slightly longer average distances (and correspondingly, lower global coefficients) than their degree-preserving randomly rewired counterparts [30].
Because we were unable to assess the effect of essential proteins (proteins whose knock-down resulted in cell death regardless of treatment) on MMS survival, we repeated the connectivity analysis while including those essential proteins that were connected to two or more MMS survival proteins, since these are the ones most likely to have a functional role in MMS survival. This increased the size and significantly improved the connectivity within the resultant network (Figure 6B and Table S7; p≤1.3E-26). After pathway analysis, we identified an additional 105 MMS pathway hits, which were validated. In order to expand the network to include the proteins in these relevant pathways, these hits were then incorporated and a larger, and equally well-connected network was observed (Figure 6C and Table S7; p≤4.6E-26). Considering the apparent importance of pathways over the individual genes, we then included all components of the 13 identified MMS survival pathways and observed that the connectivity of the resultant enlarged network was significantly improved yet again (Figure 6D and Table S7; p≤4.6E-52). Similar to the analysis of the 202 validated MMS screen hits, we also compared the connectivity of the subnetworks containing the essential genes or pathway hits with that of a randomly rewired network consisting of the same nodes. Although at a lower statistical significance, the same general trend was observed (Table S7). This more inclusive network allows a view of all interactions both within and connecting to a pathway, even if the pathway components themselves are not critical to survival after MMS. To qualitatively visualize this result, each protein known to be in a pathway was grouped together and assigned to a “pathway node” (a single node within the interactome that retains the interactions of its constitutive proteins to proteins that are external to that pathway). This resulted in a highly connected interactome, or “MMS survival network” (Figure 6E), that now encompassed 179 of the 233 MMS survival proteins that are present in the PPI network (77%). Of the remaining 54 orphan proteins, 47 are only one protein removed from the MMS survival network, six are two proteins away, and only one is not connected at all. At each step in this analysis, we observed an improved network, either enlarged or better connected, not only when comparing the entire set of proteins in that network to a random set of proteins in the PPI, but also when randomizing the additional proteins added at each step (data not shown). Overall, our observations with the MMS survival network suggest that despite the general interconnectivity within protein interactomes, a pathway analysis is highly relevant because it may improve interactome connectivity and it simplifies a systems biology overview.
Studies performed using yeast as a model organism to predict network response to MMS reported an astounding involvement of diverse biological pathways [31]. Considering that the genes that respond to environmental stress differ between mammalian and yeast cells [17], we presumed that damage response might be different or more complex in higher eukaryotes, especially considering the presence of paralogues and thus increased genetic redundancy. We therefore performed a genome-wide, RNAi based screen with Drosophila cells to investigate which genes are essential for survival following damage exposure with MMS. We were able to identify and validate 307 MMS survival genes, the majority of which had not previously been associated with alkylation damage survival. Of these genes, 146 were components of 13 different MMS survival pathways. With the five pathways examined in detail, we observed that four were functionally conserved in yeast and all five conserved in mouse with regard to their utilization following MMS treatment (Figure 5). In yeast, experimental validation of response to MMS by proteasome [19], the TOR pathway [20], and DNA repair pathways [21] was previously reported. Similarly, glutathione response to MMS was observed in mammalian cells [32], and our observation of an increase in GCLc expression provides an underlying mechanism for this phenomenon. Our demonstration of a dose-dependent increase in 8-oxoguanine after MMS exposure indicates that MMS also results in oxidative stress damage, as previous studies suggested [33]. Additionally, several recent studies have demonstrated a role for the proteasome in regulating several DNA repair pathways (reviewed in [34]), supporting our observation of increased proteasome activity in response to MMS. Thus, our screen and pathway identification have revealed a conserved set of MMS survival pathways.
Our Drosophila based study has provided novel insights to the global cellular response to alkylation damage by identifying biological pathways whose functions are required for survival after this damage. The only other analogous genome-wide, loss-of-function screen for MMS survival genes was performed in yeast [14]. That study highlighted the general biological processes required for MMS survival based on gene ontology and integrated the identified proteins into a disorganized network [14]. Considering that pathways, whether signaling, metabolic, or enzymatic, have long been identifiable entities, it is logical to consider them as units within a network. Thus our experiments focused on identifying pathways required upon exposure to damage and validating the biological responsiveness of pathways following this damage exposure. Our results confirmed that these biological pathways are indeed functional in yeast, Drosophila, and mouse cells and therefore functional contribution of these biological pathways are pertinent to damage response in a network representation.
In addition to the functional conservation of the survival pathway in response to alkylation damage, these same biological pathways appear to have roles in response to other types of damages. It is interesting to compare our results with an elegant study by Matsuoka et al. [4], which identified proteins that are phosphorylated following ionizing radiation in human cells. Their study identified proteins that are components of nine of our 13 MMS survival pathways, including four of the DNA repair pathways, DDR, mTOR, proteasome, basal transcription, and ribosome. These results suggest that different types of damage, not just alkylation damage, may utilize different components of a DNA damage survival network in a functionally conserved manner and reemphasize the functional conservation of pathways, if not the individual genes, between species.
Our emphasis to reorganize the MMS survival network based on pathways is to facilitate the observation of biologically relevant interactions. Often protein:protein interaction networks may appear chaotic, but may be interrogated for simple sub-networks associated with protein(s) and pathways of interest [21]. However, when working with pleiotropic responses that encompass so many different biological processes, such as DDR, a chaotic network representation appears non-intuitive (Figure 6D). Thus, the integration of pathways within the conceptual framework of systems biology networking is logical. An additional advantage of including pathways is highlighted by our demonstration of MMS survival protein enrichment by detailed examination of pathways. Even with this detailed analysis and convincing evidence that the pathways were indeed functioning as expected, we were not able to assign every protein within each pathway as an “MMS survival protein.” There are many possible explanations for this, but nonetheless, considering the pathways as a whole provide a framework within the network that highlights novel interactions, cross-talk, and identified proteins not mapped to a canonical pathway, but present within the network, would unlikely be observed in a “chaotic network view” and encourages their investigation. This approach is similar to computational clustering of networks based on signaling pathways using interactome datasets [35], but our approach includes both identified proteins and pathway components. Our representation is simplified, using a single node to represent the entire pathway rather than a complex display of interactions for every component in a pathway [36]. This simplified pathway inclusive representation reveals a highly organized network, consistent with the requirement of each pathway for cellular survival (Figure 6E), and provides an effective strategy to integrate modular components into the network [37] and thereby inferring biological properties [38].
The interconnectivity between the survival pathways (Figure 5E) suggests potential pathway cross-talk. If such cross-talk exists, it would be highly pertinent to cancer therapy. Recent studies have demonstrated the utility of a global level analysis, allowing identification of altered pathway function in complex diseases such as the Notch pathway in pancreatic cancer [39] and similarly the DDR pathway in breast and colorectal cancers [40]. Considering our identification of Notch, TOR, DDR, and the proteasome as “survival pathways,” all of which are currently being explored as targets for cancer therapy [41],[42], our identified survival network would suggest the possibility of combining pathway-specific pharmacological agents in cancer therapy. Some of the pathway connections and potential cross-talk represented in our survival network (Figure 6E) have already been observed. For example, protein phosphatase, PP2A, a downstream component of the TOR pathway [23], interacts with the DDR component to regulate phosphorylation of ATM and ATR [43] and vice versa [4]; DDR interacts with BER via CHK2 and XRCC1 [44]; Notch interacts with DDR via Mastermind and p53 [45]; the glutathione pathway interacts with the nucleotide excision repair pathway (NER) [46]; and the proteasome interacts with various DNA repair components [34],[47]. Together, it would appear that our model of an integrated network of conserved damage survival pathways is both valid and biologically relevant.
In conclusion, we have identified a network of pathways that have a functional role in damage response by affecting viability; we also demonstrated the functional conservation between species of the MMS survival pathways. By considering the protein interactions between the MMS survival proteins and by incorporating the MMS survival pathways, a highly interconnected damage survival network is observed that encompasses at least 58% of the identified MMS survival proteins directly. This interconnectivity suggests a strong functional interrelationship between constitutive components of the survival network and the possibility of pathway cross-talk and coordination at a level greater than just their instigation by the DDR pathway. Although these MMS survival pathways have already been implicated in MMS damage response, we have identified these seemingly disparate pathways in a single screen, and with the network analysis, this lends to direct connection of these pathways in response to damage. The pleiotropic effects of alkylation damage therefore require a wide-variety of functioning pathways in order for the cell to survive.
Kc167 cells were grown in Schneider medium (Invitrogen, Carlsbad, CA) supplemented with 10% heat inactivated fetal bovine serum (65°C, 10 minutes), penicillin and streptomycin at 22°C in a humidified chamber.
Primary MEFs were obtained by harvesting 14.5 day C57BL/6 embryos as previously described [48]. Briefly, fetal liver and head were removed and the remainder of the embryo mechanically disaggregated in plating medium. A suspension of single-cells was plated out in Dulbecco's Modified Eagle's Medium (DMEM) supplemented with 10% fetal calf serum, 2 mM glutamine, 100 U/mL penicillin, and 100 µg/mL streptomycin. MEFs were grown for two passages before freezing aliquots. Aliquots were taken and expanded as needed for each experiment.
Methods were adapted from [49], with 1.2×104 Kc167 cells suspended in 10 µL of serum-free Schneider medium (Invitrogen) added to each well of sixty-three 384-well plates. Each well contains about 0.25 µg of a double-stranded (ds) RNA, with 22,915 individual dsRNA represented in the whole library (Version 1 dsRNA). Cells were incubated for 1 h at 22°C, allowed phagocytic uptake of the dsRNA, then 20 µL of serum (15% heat-inactivated fetal bovine serum) containing medium were added and the plates incubated for a further 72 h. Medium was exchanged for fresh serum containing medium, with or without 0.004% (w/v) of MMS (Sigma-Aldrich, St. Louis, MO). Following 72 h additional incubation, medium was removed and the number of viable cells assessed using pre-diluted Celltiter-Glo per manufacturers instructions (Promega, Madison, WI). Screens were performed in duplicate, with the replicate screen initiated on a different day.
dsRNA were produced similarly to previously described [8],[18]. Briefly, dsRNA was synthesized using cDNA prepared by gene specific amplification of reverse transcribed cDNA. For cDNA preparation, RNA was harvested from adult D. melanogaster using Trizol reagent. For gene specific amplification, primers designed by the DRSC to target genes of interest were custom synthesized with overhanging T7 promoter sequences (Invitrogen). DRSC validation primer sequences were obtained from the DRSC for targeting genes of interest, and oligonucleotides were purchased (Invitrogen). First round PCR amplifications were performed using gene specific primers, and the amplified products were gel purified, using an agarose gel purification kit (Qiagen, Valencia, CA) or a 96-well gel isolation kit (Invitrogen). The purified PCR products were then used for second round amplification also using the gene specific primers; these amplified products were purified using Millipore PCR purification 96-well plate system (Millipore, Billercia, MA). The purified products were then used for dsRNA synthesis using T7 Ribomax express system (Promega) and purified using the PCR purification plate system (Millipore). The quality of purified dsRNA was verified by agarose gel electrophoresis and, following spectrophotometric quantification, stored at −20°C.
Validation of MMS survival genes was performed as previously described [18], with the exception that DRSC “validation” dsRNA (targeting a gene mRNA transcript at a different location to the library amplicon), designed by the DRSC but produced in-house, were preferentially used in this study. Briefly, dsRNA were validated in quadruplicate, with one set of plates treated with MMS and one set used as an untreated control. Plates with or without MMS with the same dsRNA were then compared.
Genes required after MMS treatment and genes whose knock-down conferred resistance to MMS were validated by validation amplicons, if available. When validation amplicons were not available, the amplicon used in the original library was used for validation. In some instances, both validation and library amplicons were used in validation experiments. In these cases, the knock-down result from the validation amplicon was preferred over that of the library amplicon, if a difference was evident. If the gene of interest was annotated to a pathway that had already been determined to be involved in MMS response, the gene was considered as a hit regardless of which version amplicon validated.
Gene ontology was retrieved from FlyBase (http://flybase.org/). Enrichment analysis was conducted using FuncAssociate (http://llama.med.harvard.edu/cgi/func/funcassociate/) for those ontologies over-represented with a p-value of less than 0.05. Those ontology categories that overlapped were consolidated.
Protein lysates were prepared using radioimmunoprecipitation assay (RIPA) buffer containing 5% sodium deoxycholate, 0.1% SDS, 0.1% Igepal in PBS with a cocktail of protease inhibitors (1 mM PMSF, 1 mM Sodium Orthovandate and 30 uL/mL Aprotinin (Sigma)). Protein concentration was determined using Bradford Protein Assay Reagent (Biorad, Hercules, CA). Equal amounts of protein were resolved using 10% SDS polyacrylamide gel and transferred onto a nitrocellulose membrane (Hybond-ECL, GE Healthcare Lifesciences, Piscataway, NJ). Using appropriate dilutions of primary and secondary antibodies, immunodetection of the protein was performed using the ECL plus system (GE Healthcare Lifesciences). Primary antibodies include P-Chk1 (T68) (Abcam) and total CHK1 (Abcam), P-p70S6K (T389) (Cell Signaling Technology) and total p70S6K (Cell Signaling Technology), and P-p53 (S15) (Cell Signaling Technology) and total p53 (Cell Signaling Technology).
Apurinic or apyrmidinic sites (AP) that were generated following DNA damage were measured using a colorimetric assay kit for DNA damage quantification (Oxford Biomedical Science, Oxford, MI), following manufacturer's protocol. Briefly, DNA was isolated from the experimental cells using a Genomic DNA isolation kit (Oxford Biomedical Science). Equal amounts of DNA were then labelled using a biotinylated aldehyde reactive probe. Labelling was followed by purification and colorimetric quantification using streptavidin-horse radish peroxidase (HRP) conjugate and HRP-dependent substrate supplied with the kit. The aldehyde reactive probe labelled DNA standard, supplied by the kit, was used to determine the number of AP sites per 100 kilobase pairs of DNA in the experimental samples.
Intracellular glutathione levels were measured using a colorimetric assay kit for glutathione (Oxford Biomedical Science), following manufacturer's protocol. Briefly, cells from RNAi experiments were resuspended in ice-cold PBS and homogenized using a sonicator; the lysate was cleared by centrifugation and the protein concentration of the supernatant was determined using Bradford's reagent (Biorad). For estimation of glutathione, equal amounts of protein lysate were de-proteinized using metaphosphoric acid (MPA) with the final concentration of MPA adjusted to 5%. The precipitated proteins were cleared by centrifugation at 4000 g for 10 min at 4°C, and the supernatant was used for the assay. Reduced glutathione (Sigma) was used as a standard, and samples were arrayed in a 96-well plate. A two-step reaction was conducted, thioesterification of intracellular thiols using 4-choloro-1-methyl-7-trifluoromethyl-quinolinium methylsulfate followed by alkaline conversion of glutathione-thioester to chromophoric thione, followed by detection of total glutathione by absorbtion at 400 nm.
dsRNA was used to target a gene of interest; this dsRNA was the same used to validate the genes of interest by knock-down. The level of RNAi mediated silencing of gene expression was monitored by quantitative real time RT–PCR using QuantiTect SYBR Green RT-PCR kit (Qiagen GmbH) and an ABI 7500 Real Time PCR System (Applied Biosystems, Foster City, CA). For these experiments, 18.61×104 Kc167 cells were dispensed into a 24-well tissue culture plate containing 6.2 µg of dsRNA per well. Following 1 h incubation in a serum-free condition to allow the uptake of dsRNA, serum was replenished to a final concentration of 10%. On Day 3, RNA was isolated using RNeasy Mini kit (Qiagen) and quantified using an ND-1000 Spectrophotometer (Nanodrop, Wilmington, DE). For the PCR amplifications, distinct primers that were not encompassed within the dsRNA used to target the gene were used. For gene expression analysis following MMS exposure, RNA was isolated on Day 4. For experiments with Kc167 cells, CG6905, the expression of which remained unaltered following MMS exposure, was used as an endogenous control. For experiments with MEFs, the isolated RNA was reverse transcribed using an ImProm-II Reverse Transcription System (Promega) and the cDNA was used with TaqMan Gene Expression Assay kit for GCLc, purchased from (Applied Biosystems, CA) and TaqMan Universal Master Mix (Applied Biosystems). Mouse β-Actin TaqMan Gene Expression Assay kit was used as an endogenous control. The level of gene expression was determined using ▵▵Ct method [50].
Experiments were performed in 384-well plates using the same layout and timing as described for the RNAi validation experiments, [18], except the proteasome activity was measured on Day 4 of the experiment, 24 h after MMS exposure. For proteasome measurement, a Proteasome-Glo assay kit was used (Promega), as per instructions. To normalize the proteasome activity with cell density, a parallel experiment was performed in a separate 384-well plate, and the cell density was estimated using CellTiter-Glo (Promega) as described above. Proteasome activity was then normalized to the relative number of cells present.
Orthologues of Drosophila genes of interest in human, mouse, and yeast, were obtained from Ensembl49 (http://ensembl.org/).
SMARTPool siRNAs were purchased from Dharmacon RNA technologies (Lafayette, CO). For siRNA transfection into MEFs, cells were harvested by trypsinization, washed in a serum free medium, and cell density was adjusted to 1×106 cells in 100 µL of MEF-2 Nucleofector Solution (Amaxa, Gaithersburg, MD), containing 0.5 µg of siRNA, followed by transfection by electroporation using Nucleofector II Device (Amaxa). The transfected cells were seeded in either a 96-well or a 384-well tissue culture plate, as required by the experiment. For non-specific control, scrambled, non-targeting siRNA was used. The fluorescent probe siGLO red (Dharmacon) was used to monitor the efficiency of siRNA uptake, and the efficiency of protein knock-down was determined by western blot analysis (Figure S3B). Additional validation for SMARTPool siRNA experiments were performed using a minimum of four independent duplex siRNA for each gene (Figure S10).
Pathway analysis was conducted first by a priori identification of protein/pathway relationships and KegArray, yielding from which we included DNA damage response, glutathione metabolism, the TOR pathway, proteasome, and DNA repair pathways. For a systematic analysis on the MMS hits, KegArray (with default settings) was used (http://genome.jp/download/), yielding Notch signaling, ATPase, basal transcription, ribosome, proteasome and glutathione metabolism. These pathways were examined for the number of hits identified in the pathway versus the number of total pathway components. Pathways that we included from prior knowledge only were not top pathways retrieved by KegArray due to their relatively small size or completeness in the KEGG database in fly.
Protein interactome data were obtained from IntAct [27] (http://www.ebi.ac.uk/intact/), the Database of Interacting Proteins [28] (DIP, http://dip.doe-mbi.ucla.edu/) and the Biomolecular Interaction Network Database [29] (BIND, http://bond.unleashedinformatics.com/). Data from the three databases were combined into a single interactome, using the CG number of each gene as the identifier of the protein. Interactomes were visualized using Cytoscape 2.6 (http://www.cytoscape.org/). Pathway nodes were created external to Cytoscape by renaming all nodes representing protein in the pathways with the name of the pathway. If a protein was found in multiple pathways, it is represented in all relevant pathway nodes. Interactions between pathways were trimmed if the only interaction between them also existed within both pathways. For instance, if a protein:protein interaction occurs between two members of the NER pathway, and these two proteins also exist in the BER pathway, then the interaction is more likely to be pathway specific and not cross-talk between pathways and was therefore removed.
Four measurements of connectivity were made: (1) Counting the number of pairs of MMS survival proteins that directly interact in the interaction network, (2) computing the average geodesic distance (i.e., the number of edges in a shortest path) between each pair of MMS survival proteins in the interaction network [51], (3) the global efficiency of the network [51], and (4) the clustering coefficient [51]. The number of direct interactions provides an intuitive measure of the connectivity of a subnetwork, while the average distance measures the global connectivity of the sub-network. Global efficiency provides a similar measure to average distance, but allows for disconnected components. The global efficiency was measured by , where n is the number of vertices in the network and dij is the geodesic distance between vertices vi and vj. The clustering coefficient is a measure of local connectivity of the network. For each vertex vi, let gi be the subnetwork that consists of direct neighbours of vi (excluding vi itself) and the edges between them. The clustering coefficient of vi was measured by the total number of edges in gi divided by the maximum number of edges that could possibly exist in gi. The clustering coefficient of a network is given by the average of the clustering coefficient of each vertex, with a high clustering coefficient indicating a distinction between a real network from a random one. To assess the statistical significance of each measurement, the same number of proteins or proteins pairs, as appropriate, were randomly sampled from the PPI or PPI subnetwork. This sampling was repeated 1,000 times to estimate a p-value that the difference could be expected by chance.
To determine if knock-down of genes resulted in increased sensitivity to MMS, raw data obtained from the viability assays was normalized and statistically analyzed as described previously [18]. A T-test was performed between normalized quadruplicates to determine the significant difference between treated and untreated wells. Percent control survival with MMS treatment was then estimated for each experimental gene knock-down as described previously [18], and a second T-test was performed on the percent of viability as compared with luciferase control within each plate as described before [18]. From these analyses, significant hits were selected as death hits if there was no greater than 55% viability in treated wells as compared to untreated if a p-value of less than 0.05 resulted from at least one of the T-tests. For those genes with an essential phenotype of less than 40% viability of non-targeting dsRNA, more stringent requirements were made on the viability effect after MMS treatment, such that 30–40% viability when untreated needed 15% viability after MMS treatment, 20–30% needed 10% after treatment, 10–20% needed 5% after treatment, and 0–10% was too dead to determine if MMS had an effect. Genes were considered “trend death hits” if they exhibited less than 55% viability after treatment but did not have a p-value with significance or if they exhibited 55–65% viability after MMS treatment with a significant p-value. Knock-down of genes resulting in resistance to MMS were confirmed as those with greater than 85% viability after treatment as compared to untreated wells, with one of the two p-values of less than 0. 0001 and viability after no treatment to be at least 60% that of non-targeting knock-down.
To determine if a gene was essential for viability, a T-test was performed on untreated wells comparing values for non-targeting dsRNA against luciferase and dsRNA targeting the gene. If the targeting dsRNA resulted in less than 70% viability of luciferase with a p-value of less than 0.00001, the gene was deemed essential.
Comparisons between predicted and observed numbers of MMS survival genes between yeast and Drosophila studies were done by a standard G-test [52]. The G-test is equivalent to a contingency Chi-square test but allows for classes with zero events.
To determine protein enrichment within a pathway where there is an unbalanced distribution of data between the numbers of proteins within a pathway compared to the total number of genes within a genome, a Fisher's Exact test was employed [53].
For connectivity measurements, except in the case of clustering coefficient, p-values were estimated using a Z-test, given that the connectivity measurement for the random subnetworks approximately follows a normal distribution (Table S6). The p-value for the clustering coefficient measurement was estimated by simply counting the frequency that the clustering coefficient of a randomly sampled sub-network was at least as high as that of the real sub-network.
|
10.1371/journal.pgen.1003439 | Mouse Oocyte Methylomes at Base Resolution Reveal Genome-Wide Accumulation of Non-CpG Methylation and Role of DNA Methyltransferases | DNA methylation is an epigenetic modification that plays a crucial role in normal mammalian development, retrotransposon silencing, and cellular reprogramming. Although methylation mainly occurs on the cytosine in a CG site, non-CG methylation is prevalent in pluripotent stem cells, brain, and oocytes. We previously identified non-CG methylation in several CG-rich regions in mouse germinal vesicle oocytes (GVOs), but the overall distribution of non-CG methylation and the enzymes responsible for this modification are unknown. Using amplification-free whole-genome bisulfite sequencing, which can be used with minute amounts of DNA, we constructed the base-resolution methylome maps of GVOs, non-growing oocytes (NGOs), and mutant GVOs lacking the DNA methyltransferase Dnmt1, Dnmt3a, Dnmt3b, or Dnmt3L. We found that nearly two-thirds of all methylcytosines occur in a non-CG context in GVOs. The distribution of non-CG methylation closely resembled that of CG methylation throughout the genome and showed clear enrichment in gene bodies. Compared to NGOs, GVOs were over four times more methylated at non-CG sites, indicating that non-CG methylation accumulates during oocyte growth. Lack of Dnmt3a or Dnmt3L resulted in a global reduction in both CG and non-CG methylation, showing that non-CG methylation depends on the Dnmt3a-Dnmt3L complex. Dnmt3b was dispensable. Of note, lack of Dnmt1 resulted in a slight decrease in CG methylation, suggesting that this maintenance enzyme plays a role in non-dividing oocytes. Dnmt1 may act on CG sites that remain hemimethylated in the de novo methylation process. Our results provide a basis for understanding the mechanisms and significance of non-CG methylation in mammalian oocytes.
| Methylation of cytosine bases in DNA is an epigenetic modification crucial for normal development, retrotransposon silencing, and cellular reprogramming. In mammals, the vast majority of 5-methylcytosine occurs at CG dinucleotides, and thus most studies to date have focused on this dinucleotide. However, recent studies have shown that 5-methylcytosine is abundant at non-CG (CA, CT, and CC) sites in certain tissues and certain cell types in human and mouse. We previously identified non-CG methylation in CG-rich sequences, including the imprint control regions in mouse germinal vesicle oocytes, but its global distribution and the enzymes responsible are unknown. Using advanced high-throughput sequencing technology applicable to minute amounts of DNA, we obtained high-resolution methylation maps of newborn non-growing oocytes, adult germinal vesicle oocytes, and mutant germinal vesicle oocytes lacking any of the four DNA methyltransferase family proteins. Our results revealed that non-CG methylation accumulates genome-wide in close proximity to highly methylated CG sites during the oocyte growth stage. We also found that the de novo DNA methyltransferase proteins Dnmt3a and Dnmt3L are responsible for non-CG methylation in oocytes. Unexpectedly, we found that the maintenance methyltransferase Dnmt1 has a role in de novo CG methylation. Our study provides a basis for understanding the mechanisms and significance of non-CG methylation in mammalian oocytes.
| DNA methylation is a well-characterized epigenetic modification crucial for normal mammalian development, retrotransposon silencing, and cellular reprogramming [1], [2]. In mammals, a high proportion of 5-methylcytosines (mCs) occurs at CG dinucleotides, and thus studies on DNA methylation so far have focused on this dinucleotide. However, recent advances in the high -throughput DNA sequencing technology changed the scene [3], [4]. “Methylome” analyses using whole-genome bisulfite sequencing (WGBS) showed that mC occurs at non-CG sites, in addition to CG sites, in human and mouse embryonic stem (ES) cells, human induced pluripotent stem (iPS) cells, and mouse brain [5]–[9]. Moreover, we previously reported the prevalence of non-CG methylation in mouse germinal vesicle oocytes (GVOs), notably at maternally methylated imprint control regions (ICRs) and some CG-rich island regions (CGIs) [10], [11].
Methylation of cytosine bases in CG dinucleotides is catalyzed by enzymes called DNA methyltransferases (Dnmts). Among these enzymes, Dnmt1 is the maintenance methyltransferase that copies the pre-existing methylation patterns upon DNA replication, while Dnmt3a and Dnmt3b are the de novo methyltransferases that create new methylation patterns. Another member of the family, Dnmt3L, lacks enzymatic activity, but enhances the activity of Dnmt3a and Dnmt3b [12], [13]. It is unknown which Dnmt is responsible for non-CG methylation in oocytes. Using the reduced representation bisulfite sequencing (RRBS) method, which focuses mainly on CG-rich regions, Smallwood et al. (2011) demonstrated that the de novo methylation at CG sites occurs in many CGIs during oocyte growth, and depends on Dnmt3a and Dnmt3L [14]. Using WGBS, we have shown that global CG methylation in GVOs appears to be Dnmt3L-dependent [11]. However, the genome-wide distribution pattern of non-CG methylation and the enzymes responsible for this methylation in oocytes remain unknown.
To answer these questions, we have used WGBS to construct the methylome maps in non-growing oocyte (NGOs), GVOs, and mutant GVOs lacking either Dnmt1, Dnmt3a, Dnmt3b, or Dnmt3L. We demonstrate that non-CG methylation occurs concomitantly with CG methylation during oocyte growth and that Dnmt3a and Dnmt3L are responsible for non-CG methylation. Our study also reveals a new function of Dnmt1 in de novo CG methylation.
To obtain methylome maps at single-base resolution from a limited number of oocytes, we used the post-bisulfite adaptor tagging (PBAT) method that requires only nanogram quantities of DNA for amplification-free WGBS [15]. The PBAT method was previously applied to four hundred GVOs and 19.3 million uniquely mapped reads were achieved [11]. To elucidate the developmental timing of non-CG methylation, we determined the methylome of newborn NGOs in addition to adult GVOs. Furthermore, to examine the role of the Dnmts in non-CG methylation, we determined the methylomes of Dnmt1-knockout GVOs (designated Dnmt1-KO), Dnmt3a-knockout GVOs (Dnmt3a-KO), Dnmt3b-knockout GVOs (Dnmt3b-KO), and Dnmt3L-knockout GVOs (Dnmt3L-KO). Using PBAT, we obtained 158–460 million uniquely mapped reads for the respective methylome (Table 1). The average read depths were 2.8×–8.3× per strand (Table 1). We were able to determine the methylation status of >77% of the genomic CG sites and >74% of the non-CG sites by at least one read (Table 1 and Figure S1).
To study non-CG methylation, a high rate of bisulfite conversion is essential because unconverted cytosines are counted as mCs. We therefore spiked each sample with unmethylated lambda phage DNA before the bisulfite reaction and calculated the conversion rate using this substrate. We confirmed that the conversion rate always exceeded 99% and, in most of the cases, 99.5% (Table 1). Judging from the data of the previous WGBS studies [5], [11], our conversion rates are sufficient to analyze non-CG methylation.
To assess the quality of our data further, we compared the data from this study with our previous locus-specific observations [10], [16]. Consistent with the previous data, the CG sites at the maternally methylated ICRs were hypermethylated (90–98%), whereas those at the paternally methylated ICRs were hypomethylated (1–5%) in GVOs (Table S1). In Dnmt3a-KO and Dnmt3L-KO samples, the maternally methylated ICRs were methylated at a rate of less than 20% (Table S1). In addition, CGIs in intragenic (gene body) regions were more methylated than those in intergenic or promoter regions, as previously reported [14] (Figure S2).
We first focused on the methylome of GVOs for more detailed studies. Here 278 million (redundant) mCs were obtained in all mapped reads, of which 65.5% occurred at non-CG sites (17.6% at CHG and 47.9% at CHH; H = A, T, or C; Figure 1A). The average methylation level was 37.9% at CG, 3.6% at CHG, and 3.1% at CHH (Figure 1B). Among the non-CG sites, CA sites were methylated most often (6.1%), whereas CT and CC sites were less frequently methylated (1.9% and 0.8%, respectively); this was also a feature in human ES cells and mouse brain [5], [6], [9] (Figure 1B). When we examined the sequences around the methylated CHG/CHH sites, TACAGC and TACACC were the most frequently methylated (21.1% and 30.7%, respectively; Figure 1C).
A chromosomal view of CG methylation and non-CG methylation in GVOs revealed large variations in methylation levels throughout entire chromosomes, and a concordance between CG and non-CG methylation at a 50-kilobase (kb) resolution (Figure 2A and Figure S3). Using sliding, non-overlapping windows of 10 kb along all chromosomes, we found a genome-wide correlation between CG methylation and non-CG methylation (Figure 2B). This is consistent with the previous reports that non-CG methylation is generally found in regions containing CG methylation in mouse ES cells [8], [17]. The concordance between CG methylation and non-CG methylation was also observed in CGIs (Figure S2). We then analyzed the effect of the level of methylation at a CG site on the levels of methylation at nearby non-CG sites. The non-CG methylation levels near the highly (>80%) and weakly (<20%) methylated CG sites were 9.3% and 0.5%, respectively (Figure 2C). Taken together, these results indicate that non-CG methylation is highly linked to CG methylation in GVOs. Interestingly, we found that the levels of non-CG methylation at positions −1, −2, and +3 of a CG site are low and those at positions −4 and −5 are high (Figure 2C). The extremely low level of methylation at position −1 is consistent with the low level of methylation of the first cytosine at CCG sites (Figure 1C), but we do not have plausible explanations for the other peaks or dips. We also determined the distribution of CG and non-CG methylation across the RefSeq genes and found that their methylation levels are lowest in the region immediately upstream of the transcription start site (TSS), and gradually increase towards the end of the last exon (Figure 2D). Thereafter, the levels of methylation drop markedly in the downstream region (Figure 2D). The actual level of non-CG methylation was 5.2% in the intragenic region (gene body) and 1.9% in the intergenic region, which is consistent with previous reports [5], [6].
CHG sequences show partial strand-symmetry and have cytosines on both strands. We determined the methylation levels at CAG/CTG sites on the respective strands. We found that, while 98% of highly methylated CG sites (methylation level ≥70%) are methylated on both strands (Figure 3A), 89% of highly methylated CAG/CTG sites (methylation level ≥40% on one of the strands) are methylated only on one strand (Figure 3A), as previously reported in human ES cells [5].
It is known that the maternally methylated ICRs and a thousand non-imprinted CGIs acquire CG methylation during oocyte growth [14], [18]. We therefore determined the methylome of newborn NGOs (before growth) and compared it with that of GVOs (after growth). The level of CG methylation was 2.3% in NGOs, which increased to 37.9% in GVOs (Figure 4A; Also see Figure 1B), supporting the previous finding on CGIs [14]. Similarly, the level of non-CG methylation increased from 0.61% (NGOs) to 3.2% (GVOs; Figure 4A). (Note that the level is actually not above the non-conversion rate in NGOs (Table 1), suggesting that NGOs may actually have no non-CG methylation.) Thus, CG methylation and non-CG methylation occur at the same time during oocyte growth. The increase in non-CG methylation during this stage occurred globally (Figure 4B and Figure S4).
Next, we investigated the methylation changes during oocyte growth at different genomic elements. Both intragenic and intergenic regions demonstrated low levels of CG and non-CG methylation in NGOs, but higher levels in GVOs (Figure 5A). Most repetitive elements showed similar developmental changes (Figure 5B). However, intracisternal A particle (IAP) elements retained relatively high levels of CG methylation (36%) in NGOs, which indicates that approximately 62% of CG methylation found in IAPs in GVOs already exist in NGOs. This is consistent with the previous findings that IAPs are substantially resistant to epigenetic reprogramming in primordial germ cells [19]–[22]. Interestingly, even IAPs did not retain non-CG methylation in NGOs (Figure 5B). The maternally methylated ICRs generally showed low levels of CG methylation in NGOs, but the Peg10 and Impact ICRs retained relatively high CG methylation (17–19%; Table S1). Both of these ICRs contain tandem repeats [23], [24], and these repeats had higher CG methylation levels (34% and 23%, respectively) than the rest of the ICRs in NGOs (Figure S5), although other ICRs such as Igf2r, Kcnq1ot1, and Snrpn also contain tandem repeats. Again, non-CG methylation was not high even at the Peg10 and Impact ICRs in NGOs (Table S1). In general, there was no correlation between CG methylation and non-CG methylation in NGOs (Figure S6), which differs from the finding in GVOs (Figure 2B). Taken together, these results indicate that non-CG methylation is virtually absent in NGOs and accumulates during oocyte growth.
We next compared the methylomes of wild-type and mutant GVOs lacking Dnmts. Dnmt3a-KO and Dnmt3L-KO demonstrated a global reduction in both CG methylation and non-CG methylation, whereas Dnmt3b-KO showed no significant change (Figure 4A and Figure S4). In Dnmt3a-KO and Dnmt3L-KO, intragenic regions, intergenic regions, and repetitive elements (except IAP) all showed very low levels of methylation (Figure 5). Dnmt3a and Dnmt3L form a tetramer complex (composed of two molecules of each) and thereby act as an effective de novo CG methyltransferase [25]. Our results revealed that, not only CG methylation, but also non-CG methylation depends on this complex. Since non-CG methylation occurs on one strand, even at symmetrical CHG sites (Figure 3A), no maintenance methylation activity seems to exist. Thus, non-CG methylation is solely attributed to de novo methylation by the Dnmt3a-Dnmt3L complex.
Next, we separately identified and compared regions that showed significantly lower CG methylation and non-CG methylation in NGOs, Dnmt3a-KO, and Dnmt3L-KO, respectively, compared to GVOs, using a non-overlapping sliding window of 10 kb. We found that 80% (CG) and 92% (non-CG) of the regions identified in NGOs overlapped with the regions identified in both Dnmt3a-KO and Dnmt3L-KO, confirming that these regions are targeted by the Dnmt3a-Dnmt3L complex for de novo CG and non-CG methylation during oocyte growth (Figure S7).
Dnmt1 preferentially methylates hemimethylated CG sites and thus copies the pre-existing methylation patterns upon DNA replication. Interestingly, Dnmt1-KO showed a slightly lower level of CG methylation compared to GVOs (Figure 4A). This was associated with an increase in hemimethylated CG sites because the proportion of CG sites methylated on only one strand increased (Figure 3A). Most of the CG sites that were highly methylated on one strand in Dnmt1-KO were highly methylated on both strands in GVOs (Figure 3B). The findings were validated by conventional bisulfite sequencing of two selected loci (Figure S8). Because these hemimethylated CG sites were unmethylated in NGOs (Figure S8), the increase in hemimethylation is attributed to incomplete de novo methylation during oocyte growth. These results suggest that Dnmt1 acts on CG sites that have remained hemimethylated during the process of de novo methylation. Unexpectedly, Dnmt1-KO showed a slight increase in non-CG methylation (Figure 4A). This could be attributed to a secondary effect caused by compensatory up-regulation of Dnmt3a [26] (Figure S9).
Non-CG methylation in mouse oocytes was first identified in a few genes [27], [28], and then in several maternally methylated ICRs and two CGIs [10]. More recently, we have reported the abundant presence of non-CG methylation in mouse GVOs (3.4–3.8% methylation levels), which was revealed by WGBS using the PBAT method [11]. However, the detail of the genomic distribution, developmental timing, and enzymatic basis of non-CG methylation has been lacking.
To further investigate these facets of non-CG methylation, we have constructed and compared the methylome maps of GVOs, NGOs, and GVOs defective for either of the Dnmts (Dnmt1-KO, Dnmt3a-KO, Dnmt3b-KO, and Dnmt3L-KO). We found that a surprisingly high proportion (66%) of mCs occurs at non-CG sites in GVOs. This proportion is higher than that reported for human ES cells or mouse brain (ca. 30%) [5], [9]. When we examined the overall distribution of non-CG methylation, we found that non-CG methylation occurs almost exclusively in regions rich in CG methylation. This appears to be consistent with the fact that the same enzyme complex is responsible for both de novo CG and non-CG methylation in GVOs (see later). Furthermore, a comparison of the methylomes of NGOs and GVOs revealed that non-CG methylation occurs during the oocyte growth stage, concomitant with de novo CG methylation. This stage corresponds to meiotic prophase I, a stage in which oocytes remain non-replicating for an extended period (up to a year in mice).
Although CHG sites are more frequently methylated than CHH sites in human ES cells [5], we did not observe this trend in GVOs, which was consistent with the finding in mouse brain [9]. We found that the preferred sequence for non-CG methylation in GVOs is CA, and TACA(G/C)C in particular. This was also true in human ES cells and mouse brain [9], [29], [30], and thus the motif may be the common site for non-CG methylation in many mammalian species. Interestingly, the preference for thymine at position −2 and cytosine at position +3 is reminiscent of the sequence preference of Dnmt3a [31]; furthermore, murine Dnmt3a has been shown to mediate non-CG methylation in ES cells [32]. Indeed, we found Dnmt3a and its regulatory protein Dnmt3L to be responsible for non-CG methylation in GVOs. This was consistent with the role of the Dnmt3a-Dnmt3L complex in de novo CG methylation during oocyte growth [11], [14], [16]. Dnmt3b was dispensable, perhaps because of the extremely low expression of this gene in oocytes [33]. Consistent with this, the Dnmt3b promoter (±2 kb of TSS) showed a high level of CG methylation (>93%), whereas other Dnmt promoters showed lower methylation (<29%; data not shown). This is in contrast to the recent reports that non-CG methylation depends on both Dnmt3a and Dnmt3b in human ES cells [30] and on Dnmt3a, Dnmt3b, and Dnmt3L in mouse ES cells [17].
Based on the crystallographic analysis of the Dnmt3a-Dnmt3L complex, a model was proposed in which the two active sites of the complex recognize and methylate two CG sites spaced 8–10 base pairs apart [25]. Although Dnmt3a and Dnmt3L act in the same complex, the CG and non-CG methylation levels in Dnmt3a-KO were higher than those in Dnmt3L-KO. This could be explained by the difference in knockout strategy, as conventional knockout GVOs (Dnmt3L-KO) may show a more severe phenotype than conditional knockout GVOs (Dnmt3a-KO), depending on the efficiency of the Cre-mediated deletion in the conditional knockout.
Interestingly, the level of CG methylation was slightly lower in Dnmt1-KO than GVOs, and this was correlated with an increase in hemimethylated sites. We found that CG sites located in intergenic regions or retrotransposons are more likely to be hemimethylated than those in gene bodies in Dnmt1-KO (data not shown). It seems that Dnmt1 methylates CG sites that have remained hemimethylated during the de novo methylation process in oocyte. Uhrf1 (Np95) recognizes hemimethylated CG sites and helps to recruit Dnmt1 in somatic cells [34], [35]. Since Uhrf1 is also expressed in GVOs [11], [36], Dnmt1 could utilize this protein to recognize hemimethylated CG sites in GVOs. Thus, Dnmt1 has a role, even in a non-replicating (and non-dividing) cell type, in establishing fully methylated CG sites. Unexpectedly, the level of non-CG methylation was slightly higher in Dnmt1-KO compared to GVOs. Because a loss of Dnmt1 can cause up-regulation of Dnmt3a [26] (Figure S9), this gain in non-CG methylation could be attributed to this compensation mechanism.
Based on the data from this and other studies, we speculate that there are two important prerequisites for the presence of non-CG methylation in mammalian cells. The first requirement is a high level of expression of Dnmt3a and/or Dnmt3b, whose target sequence specificity is not strict [37], [38]. All mammalian cells that have non-CG methylation meet this criterion. The second is that the cells should be non-replicating or slowly replicating. Since non-CG methylation occurs only on one strand, even at symmetric CHG sites, there appears to be no maintenance mechanism for non-CG methylation. Thus, non-CG methylation needs to be re-established de novo after each cell division. Among the cells that contain non-CG methylation, oocytes and neurons are post-replicative and do not divide. Human pluripotent cells replicate more slowly than mouse pluripotent cells [39], and thus have higher levels of non-CG methylation. Presumably, the level of non-CG methylation is determined by the balance between the activity of Dnmt3a/Dnmt3b and the rate of cell division, and this may be the reason why mouse pluripotent cells, which divide faster, have lower levels of non-CG methylation. Consistent with this, non-CG methylation in oocytes is lost in a replication dependent way during the cleavage stage [10], [40]. This is similar to what we recently observed in mouse testis, where non-CG methylation accumulates in non-dividing germ cells but becomes reduced after the resumption of mitosis [41].
Lastly, 5-hydroxymethylcytosine (hmC) has been identified as an important intermediate for passive, and potentially active, demethylation [42]. Since bisulfite sequencing cannot distinguish between mC and hmC, we could not determine whether hmC is present in oocytes. Fine mapping of hmC at CG and non-CG sites in oocytes will require development of new technologies because currently available methods are not adapted for limited amounts of DNA [43].
In summary, we have determined the methylome maps of mouse NGOs and GVOs and revealed the genome-wide distribution and developmental timing of non-CG methylation. Using GVOs lacking either of the Dnmt proteins, we identified Dnmt3a and Dnmt3L as the proteins responsible for non-CG methylation during oocyte growth. At this point in time, it is unclear whether non-CG methylation is a by-product of CG methylation or has any biological role. Our data will provide a basis for understanding the mechanism and role of non-CG methylation in mammals.
NGOs and GVOs were collected from 0–3-day old and over 8-week old C57BL/6 females (Clea, Japan), respectively. Mutant GVOs, designated Dnmt1-KO, Dnmt3a-KO, Dnmt3b-KO, and Dnmt3L-KO, were obtained from [Dnmt12lox/2lox, Zp3-Cre] females, [Dnmt3a2lox/2lox, Zp3-Cre] females, [Dnmt3b2lox/2lox, Zp3-Cre] females, and Dnmt3L-null homozygous females, respectively [16], [33], [44]. The reason for the use of conditional knockout mice for Dnmt1, Dnmt3a, and Dnmt3b was the lethality of the relevant conventional knockout mice [16], [45]–[47]. In contrast, the conventional knockout mice for Dnmt3L are viable [44], [48].
Libraries for WGBS were prepared using the PBAT method, as described previously [15]. Approximately one-thousand oocytes were spiked with 1 ng of unmethylated lambda phage DNA (Promega), placed in a lysis solution (0.1% SDS, 1 mg/mL proteinase K, 1 µg tRNA) and incubated for 60 min at 37°C, and then 15 min at 98°C, followed by bisulfite treatment using MethylCode Bisulfite Conversion Kit (Invitrogen). Bisulfite-treated DNA was double-stranded with Klenow fragment [3′→5′ exo(-); New England Biolabs], using BioPEA2N4 (5′-biotin-ACACTCTTTCCCTACACGACGCTCTTCCGATCTNNNN-3′) (first strand). The biotinylated strand was captured using Dynabeads M-280 Streptavidin (Invitrogen) and double-stranded with Klenow fragment (3′→5′ exo(-)), with PE-reverse-N4 (5′-CAAGCAGAAGACGGCATACGAGATNNNN-3′) (second strand). After removing the first strand, the second strand was used as a template for primer extension by Phusion Hot Start High-Fidelity DNA Polymerase (Finnzymes) with Primer-3 (5′-AATGATACGGCGACCACCGAGATCTACACTCTTTCCCTACACGACGCTCTTCCGATCT-3′). Concentrations of the PBAT libraries were measured by quantitative PCR (qPCR) using PE-forward and PE-reverse primers (Illumina) [15]. The PhiX v2 Control Kit (Illumina) was used as a standard for quantification.
All sequencing runs were single-ended and 100 nucleotides (nt) in length, and performed on the Illumina HiSeq 2000 platform. Based on the qPCR quantification, 3×108 copies of double-stranded DNA from the PBAT library were sequenced per lane on HiSeq 2000, as previously described [15]. Cluster generation and sequencing were performed in single-read mode using the TruSeq SR Cluster Kit v3-cBot-HS (Illumina) and the TruSeq SBS Kit v3-HS (Illumina) according to the manufacturer's protocols. Sequenced reads were processed using the standard Illumina base caller (v.1.8.2). We truncated raw reads to 92 nt to remove lower quality bases near the end of the reads and any remaining adapter sequences incorporated in the read. The resulting reads were aligned to the reference genome (mouse mm9) using Bismark alignment software v.0.6.3 [49] with a maximum of two mismatches, and only uniquely aligned reads were retained. We estimated bisulfite conversion rates using reads that uniquely aligned to the lambda phage genome. For strand-independent analysis of CG methylation, counts from the two cytosines in a CG and its reverse complement were combined. We subsequently evaluated only CG sites with at least 6× coverage and non-CG sites with at least 4× coverage, and discarded cytosines with more than 100× coverage. CG islands, RefSeq genes, and repeat sequences for the mm9 genome were downloaded from the UCSC Genome Browser [50]. The logo plot images were generated with WebLogo [51].
Sequence data in this study have been deposited in DDBJ/GenBank/EMBL under accession number DRA000570.
|
10.1371/journal.pntd.0006154 | MAIT cells are activated in acute Dengue virus infection and after in vitro Zika virus infection | Dengue virus (DENV) and Zika virus (ZIKV) are members of the Flaviviridae and are predominantly transmitted via mosquito bites. Both viruses are responsible for a growing number of infections in tropical and subtropical regions. DENV infection can cause lethargy with severe morbidity and dengue shock syndrome leading to death in some cases. ZIKV is now linked with Guillain-Barré syndrome and fetal malformations including microcephaly and developmental disorders (congenital Zika syndrome). The protective and pathogenic roles played by the immune response in these infections is unknown. Mucosal-associated invariant T (MAIT) cells are a population of innate T cells with potent anti-bacterial activity. MAIT cells have also been postulated to play a role in the immune response to viral infections. In this study, we evaluated MAIT cell frequency, phenotype, and function in samples from subjects with acute and convalescent DENV infection. We found that in acute DENV infection, MAIT cells had elevated co-expression of the activation markers CD38 and HLA-DR and had a poor IFNγ response following bacterial stimulation. Furthermore, we found that MAIT cells can produce IFNγ in response to in vitro infection with ZIKV. This MAIT cell response was independent of MR1, but dependent on IL-12 and IL-18. Our results suggest that MAIT cells may play an important role in the immune response to Flavivirus infections.
| Dengue virus (DENV) and Zika virus (ZIKV) are responsible for a growing number of infections in tropical and subtropical regions. DENV infection can cause dengue shock syndrome leading to death in some cases, while ZIKV is now linked with Guillain-Barré syndrome and congenital anomalies including microcephaly. The protective and pathogenic roles played by the immune response in these infection is unknown. Mucosal-associated invariant T (MAIT) cells are a population of innate T cells with potent anti-bacterial activity. MAIT cells have also been postulated to play a role in the immune response to viral infections. In this study, we found that MAIT cells are activated in acute DENV infection and in vitro following ZIKV infection. MAIT cell IFNγ response to ZIKV infection was TCR independent, but IL-12 and IL-18 dependent. IFNγ produced from MAIT cells could help limit viral replication. Further studies are needed to determine the protective or pathogenic role of MAIT cells in Flavivirus infections.
| Dengue virus (DENV) and Zika virus (ZIKV) are members of Flaviviridae and both are transmitted mostly via mosquito bites. It is estimated that around 400 million people are infected with DENV annually[1]. DENV infection symptoms range from mild disease, to dengue fever, dengue hemorrhagic fever, and dengue shock syndromes, which can be fatal in some cases. The mechanisms by which DENV infection causes severe illness are not completely understood. An extensive immune activation, characterized by a cytokine storm, has been described in DENV infection, and host factors are also likely to be involved[2]. Conventional antiviral CD8+ T cells are activated and expanded following DENV infection[3], and have been proposed to be protective by reducing the viral load[4].
Until recently, ZIKV had been understudied because the infection was thought to be associated only with a mild viral illness and of limited geographical distribution. In 2014, the virus suddenly expanded its range dramatically and appeared in the Americas, leading to the most widespread ZIKV outbreak in history. It is now estimated that over 2 billion people are living in regions suitable for ZIKV transmission[5]. ZIKV infection is now linked with cases of Guillain-Barré syndrome[6] and with a plethora of fetal malformations including microcephaly, now called congenital Zika syndrome, following transmission from an infected pregnant woman to her developing fetus[7]. The protective or pathogenic roles of T cells in ZIKV infection remains to be investigated.
Mucosal-associated invariant T (MAIT) cells are a population of innate T cells that represent 1–10% of T cells in the blood of healthy individuals[8]. They express a semi-invariant TCR using Vα7.2 coupled with Jα33 and a limited Vβ repertoire[9]. A small fraction of MAIT cells have been found to express Vα12 or Vα20[10]. Recent studies suggest that the TCR β-chain has some influence on TCR dependent activation of MAIT cells[11, 12]. MAIT cells can be identified by the expression of Vα7.2 in combination with CD161 or the IL-18 receptor[13]. They have been shown to recognize microbial vitamin B2 (riboflavin) metabolites presented by the MHC class I-like protein MR1[14]. This allows MAIT cells to respond to a range of bacteria, mycobacteria, and yeasts[15]. MAIT cells can also be activated in a TCR independent way by IL-12 and IL18[16], allowing them to respond to pathogens not producing riboflavin, such as viruses[17, 18]. In chronic HIV-1 and HTLV-1 infections, MAIT cells are reduced in number and display impaired functionality in response to bacterial stimulation[19–21]. A similar MAIT cell impairement has been described in patients with chronic infections due to a primary immunodeficiency[22].
In this study, we investigated MAIT cells response in Flavivirus infection. We report that MAIT cells are activated in acute DENV infection and have a poor response to in vitro bacterial stimulation. We also report that MAIT cells can produce IFNγ in response to in vitro ZIKV infection. This response was dependent on IL-12 and IL-18 and was impaired in HIV-1-infected individuals.
25 DENV-infected individuals from Sao Paulo, Brazil, were enrolled in the study (9 males and 16 females, age 17 to 87, Table 1). Patients were diagnosed with DENV infection by detection of DENV NS1 antigen and/or IgM-specific antibodies using a commercially available rapid test (Dengue Duo Test Bioeasy, Standard Diagnostic Inc. 575–34, Korea) or by detection of DENV RNA by real time PCR (RT-PCR). Absolute cell counts were determined using an automated hematology analyzer (Abbott Cell-Dyn 3700 Hematology Analyzer) at the Hematology Laboratory, Hematology Service, at the Faculty of Medicine, University of Sao Paulo. The study was approved by the University of Sao Paulo institutional review board (CAPPesq), and written informed consent was provided by all participants according to the Declaration of Helsinki. Buffy coats from healthy donors were obtained from the New York Blood Bank as approved by the George Washington University institutional review board. Samples from HIV-1-infected patients were obtained from the Jacobi Medical Center (NY, USA) and written informed consent was provided by all participants. This study was approved by Jacobi Medical Center and the George Washington University institutional review boards. All samples from all sites were anonymized. Minors were enrolled in the study, in which case legal guardians provided written informed consent according to the Declaration of Helsinki.
Peripheral blood mononuclear cells (PBMCs) were isolated by density-gradient sedimentation using Ficoll-Paque (Lymphoprep, Nycomed Pharma, Oslo, Norway). Isolated PBMCs were washed twice in Hank’s balanced salt solution (Gibco, Grand Island, NY), and cryopreserved in RPMI 1640 (Gibco), supplemented with 20% heat inactivated fetal bovine serum (FBS; Hyclone Laboratories, Logan UT), 50 U/ml of penicillin (Gibco), 50 μg/ml of streptomycin (Gibco), 10 mM glutamine (Gibco) and 7.5% dimethylsulphoxide (DMSO; Sigma, St Louis, MO). Cryopreserved cells from all subjects were stored in liquid nitrogen until used in the assays. For DENV-infected patients, samples were collected during the acute phase of infection (before defervescence) and 1 month after (convalescent phase). Plasma was collected by centrifugation and stored at -80°C until used in the assays.
Vero cells were obtained from the American Type Culture Collection (ATCC, Manassas, VA, USA) and maintained using Eagle’s Minimum Essential Medium (ATCC) supplemented with 10% fetal bovine serum (FBS) at 37°C with 5% CO2. ZIKV MR766 (ATCC) was added to Vero cells at a MOI of 0.1 and incubated for 4–6 days. The supernatant was centrifuged at 12 000g for 5 min, filtered (0.44 μm), aliquoted and stored at –80°C. The viral titer was determined using plaque assays on Vero cells as previously described[23]. Briefly, virus stocks were serially diluted and adsorbed to confluent monolayers. After 1 h, the inoculum was removed and cells were overlaid with semisolid medium containing 1% carboxymethyl cellulose (Sigma Aldrich, St-Louis, MO, USA). Cells were further incubated for 5 days, fixed in 4% formaldehyde (Sigma Aldrich), and stained with 1% crystal violet in 20% ethanol (Sigma Aldrich) for plaque visualization. Titers were expressed as plaque forming units (PFU) per milliliter. In some experiments, ZIKV was heat inactivated by a 60 minutes incubation at 56°C.
Cryopreserved specimens were thawed and washed, and counts and viability were assessed using the Countess Automated Cell Counter system (Invitrogen, Carlsbad, CA). Cells were washed and stained in Brilliant Violet Stain Buffer (BD Biosciences, San Jose, CA) at room temperature for 15 min in 96-well V-bottom plates in the dark. Samples were then washed and fixed using Cytofix/Cytoperm (BD Biosciences) before flow cytometry data acquisition. Intracellular staining was performed in Perm/Wash (BD Biosciences). mAbs used in flow cytometry: CD3 AF700, CD3 PerCP-Cy5.5 (both clone UCHT1), CD8 BV711 (clone RPA-T8), CD38 APC-H7 (clone HB7), CD127 FITC (clone HIL-7R-M2), CD161 BV421 (clone DX12), CCR6 BV786 (clone 11A9), HLA-DR APC (clone L243), IFNγ APC (clone B27), and PD-1 PE-Cy7 (clone EH12.1) were all from BD Biosciences, PLZF APC was from R&D Systems (Minneapolis, MN), EOMES FITC (clone WD1928) was from eBioscience and TCR Vα7.2 PE (clone 3C10) was from Biolegend (San Diego, CA, USA). Live/dead aqua fixable cell stain was from Life Technologies (Eugene, OR, USA). Data were acquired on a BD LSRFortessa instrument (BD Biosciences) and analyzed using FlowJo Version 9.8.5 software (TreeStar, Ashland, OR, USA).
MAIT cell function was determined in vitro using paraformaldehyde-fixed E. coli stimulation (one shot top10, Life Technology, multiplicity of exposure 10) in the presence of 1.25 μg/ml anti-CD28 mAb (clone L293, BD Biosciences)[24] or ZIKV at a MOI of 5 (without anti CD28 mAb). E. coli was fixed for 5 minutes in 1% paraformaldehyde. PBMCs were further cultured for 24 hours at 37°C/5% CO2 in RPMI medium supplemented with 10% fetal bovin serum. Monensin (Golgi Stop, BD Biosciences) was added during the last 6 hours of the stimulation. In some experiments blocking antibodies against MR-1 (5μg/ml, clone 26.5, Biolegend), IL-12p70 (10μg/ml, clone 24910, R&D systems), and IL-18 (10μg/ml, clone 125-2H, MBL International, Woburn, MA, USA) were added.
IL-7 (RayBiotech, Norcross, GA, USA) and sCD14 (R&D Systems) were measured in plasma by ELISA following manufacturer’s instruction.
All statistical analysis was performed using Graph Pad Prism version 6.0h for Mac OSX (GraphPad Software, La Jolla, CA). The changes between acute and convalescent phases and before/after ZIKV stimulation with or without blocking antibodies were analyzed with Wilcoxon matched-pairs signed rank test. Associations between groups were determined by Spearman's rank correlation. P-values ≤ 0.05 were considered statistically significant.
We enrolled 25 individuals with acute DENV infection, and we followed them during the convalescent phase (Table 1). We evaluated MAIT cell (defined as CD3+ CD161+ Vα7.2+, Fig 1A) frequency by flow cytometry and found no significant difference between acute and convalescent DENV infection (Fig 1B). However, MAIT cell counts were decreased in the acute phase (Fig 1C) due to significant overall lymphopenia amongst infected patients (S1 Fig). Next, we characterized the phenotype of MAIT cells in the acute and convalescent phases of DENV infection. MAIT cells had significantly increased co-expression of the activation markers CD38 and HLA-DR (S1 Fig and Fig 1D), of the IL-7 receptor CD127 (S1 Fig and Fig 1E), and of PD-1 (S1 Fig and Fig 1F) in the acute phase. We did not observe any difference in the expression of CCR6 by MAIT between the acute and convalescent phases (Fig 1G). In chronic viral infections MAIT cell activation is associated with their reduced frequency[19, 21]. Thus, we investigated if there was an association between the reduced MAIT cell count in the acute phase and their increased co-expression of CD38 and HLA-DR, and found a trend for an inverse association (p = 0.0779, S1 Fig). Next, we compared the results for MAIT cells during the convalescent phase to healthy controls from Brazil. We found that there was no difference in the co-expression of CD38 and HLA-DR between the convalescent and healthy controls individuals (S2 Fig). PD-1 remained elevated during the convalescent phase of DENV infection (S2 Fig) and CD127 was decreased compared to healthy controls (S2 Fig). Our results show that MAIT cells are activated and reduced in number in acute DENV infection.
Because the majority of MAIT cells are CD8+, we evaluated the response of conventional CD8 T cells in acute DENV infection. Conventional CD8 T cells had significantly elevated levels of co-expression of CD38 and HLA-DR in the acute phase and the levels of co-expression in the convalescent phase were similar to healthy controls (S3 Fig). PD-1 was also elevated on CD8 T cell in the acute phase of infection. However, PD-1 levels in the convalescent phase trend to remain elevated compared to healthy controls (S3 Fig). However, in contrast to MAIT cells, the levels of CD127 on conventional CD8 T cells were not different between the acute and the convalescent phase, or healthy controls (S3 Fig).
MAIT cells have been shown to have decreased expression of key transcription factors in chronic viral infections[21, 25, 26]. Therefore, we investigated if MAIT cells showed a similar decrease of Eomes and PLZF expression in acute DENV infection. We found that Eomes expression was reduced in convalescent DENV infection (Fig 2A and 2B) compared to the acute phase and healthy controls. However, we did not observe any difference in PLZF expression between acute and convalescent DENV infection or healthy controls (Fig 2A and 2C). Our results suggest that different transcription factor expression profiles are associated with acute and chronic viral infections respectively.
Increased pro-inflammatory cytokines levels in DENV infection have been associated with microbial translocation[27]. sCD14 is a marker of monocyte activation and is considered an indirect marker of microbial translocation[28]. Thus, we measured the levels of sCD14 in our cohort of DENV-infected subjects. Levels of sCD14 were significantly higher in the acute phase of infection than in the convalescent phase (Fig 3A). Levels of sCD14 remained higher in convalescent DENV compared to healthy controls. However, we did not find any significant associations between the levels of sCD14 in acute DENV infection and co-expression of CD38 and HLA-DR by MAIT cells or with MAIT cell numbers (S4 Fig). Because we found elevated expression of the IL-7 receptor by MAIT cells in acute DENV infection, we measured the levels of plasma IL-7 in acute and convalescent DENV infection but did not find any significant change (Fig 3B).
Next, to establish the functionality of MAIT cells, we investigated the in vitro response of MAIT cells from the acute and convalescent phases of DENV infection to in vitro stimulation with E. coli. There was no difference in IFNγ production by MAIT cells in the acute and convalescent phases of DENV infection in the absence of stimulation (S5 Fig). MAIT cells in the acute phase produced significantly less IFNγ after E. coli stimulation compared to the convalescent phase (Fig 4A and 4B). The MAIT cell IFNγ response in the convalescent phase was similar to the response of healthy controls (S5 Fig). Interestingly, we found that the levels of sCD14 in acute DENV infection were inversely associated with the MAIT cell IFNγ response (Fig 4C), possibly suggesting a role for monocyte activation in the poor MAIT cell response.
Finally, we used in vitro infection with ZIKV to study the mechanism of MAIT cell activation in a different Flavivirus infection. MAIT cells from healthy individuals consistently produced IFNγ in response to in vitro ZIKV infection (Fig 5A and 5B). In contrast to E. coli, the MAIT cell IFNγ response to ZIKV could not be blocked by a MR-1 blocking antibody (Fig 5C). The IFNγ response from MAIT cells to ZIKV was partially reduced by blocking antibodies against IL-12 and IL-18 and was completely blocked when they were used in combination (Fig 5C). We also investigated if viral replication was needed for the MAIT cell response to in vitro ZIKV infection. For this purpose, heat inactivated ZIKV was added to PBMCs and the MAIT cell IFNγ response was compared to the response obtained using replication competent ZIKV. We observed only a small reduction in IFNγ production by MAIT cells in response to ZIKV when using a heat inactivated virus (Fig 5C), suggesting that viral replication is not needed for production of IL-12 and IL-18 and subsequent MAIT cell activation.
Viral co-infections are common and understudied. In Brazil, many people living with HIV will be exposed to dengue or zika viruses. MAIT cells from HIV-1-infected individuals exhibit decreased functionality following stimulation with E. coli[19]. Thus, we evaluated the capacity of MAIT cells from HIV-1-infected subjects (Table 2) to produce IFNγ in response to ZIKV infection and found that in 5 out of 6 individuals there was no increase in IFNγ production in response to ZIKV infection (Fig 5D). We then directly stimulated PBMCs with IL-12 and IL-18 and found a similar IFNγ response by MAIT cells from HIV-1-infected and uninfected subjects (Fig 5E), suggesting that MAIT cells from HIV-1-infected individuals have a normal capacity to respond to cytokine stimulation.
We found that MAIT cells are activated in acute human DENV infection as well as following in vitro ZIKV infection. However, in contrast to a previous study[17], we did not find any significant changes in MAIT cell frequency between acute and convalescent DENV infection. Wilgenburg et al. focused on the study of CD8+ MAIT cells, while in this study, we also included CD8- MAIT cells. In addition, the difference in timing of sample collection might explain the differences between the two studies. We found that MAIT cell counts were decreased in parallel with the total lymphocyte count during the acute phase. We found that MAIT cells were activated during acute DENV infection, as had Wilgenburg and colleagues. IL-12 and IL-18 have been shown to trigger MAIT cell activation[16], and monocyte production of IL-18 is required for MAIT cell in vitro response to influenza A virus (IAV)[18]. The levels of IL-12 and IL-18 are elevated in DENV infection[17, 29–31] and could therefore be involved in MAIT cell activation, as shown here. We showed that MAIT cell IFNγ production following in vitro ZIKV infection also depended on IL-12 and IL-18. Immune activation in acute DENV infection has been associated with elevated levels of LPS and of markers of microbial translocation[27, 32]. This raises the possibility that MAIT cells could also be activated in a TCR-dependent way by microbial products during acute DENV. However, we did not find an association between the levels of sCD14, an indirect marker of microbial translocation, and MAIT cell activation in acute DENV. Rather, we found that sCD14 was inversely associated with the in vitro IFNγ response of MAIT cells to E. coli. This suggests that monocyte activation could result in poor antigen presentation to MAIT cells. An alternative explanation could be of a temporary monocyte tolerance to stimulation induced by LPS. This could contribute to the reduced MAIT cell response in the acute phase. Finally, the elevated expression of PD-1 on MAIT cells during acute DENV infection could also contribute to the reduced IFNγ production. Further studies are needed to confirm that both MAIT cells and monocytes are involved in this defect.
Chronic viral infections have been associated with a reduced expression of the transcription factors PLZF and Eomes by MAIT cells[21, 25, 26]. Interestingly, we found that DENV infection did not change the levels of PLZF expression in MAIT cells and their Eomes levels were reduced in convalescent compared to acute DENV and healthy controls. CD56+ MAIT cells have been shown to have a higher Eomes expression and a more robust response to IL-12 and IL-18 than CD56- MAIT cells [11]. Therefore, it is possible that the decrease in Eomes expression by MAIT cells in convalescent DENV infection is part of a feedback loop to decrease their response to cytokines. Another possibility is a decrease in the CD56+ subset of MAIT cells in blood following acute DENV. We have also observed a decreased expression of the IL-7 receptor (CD127) by MAIT cells during the convalescent phase. IL-7 has been shown to increase MAIT cell response[26, 33]. Thus, reduced levels of Eomes and CD127 could by a mechanism by which MAIT cells could prevent sustained activation following an acute infection.
Patients that recovered from IAV infection had higher circulating MAIT cells than those that succumbed[18] and IFNγ production by MAIT cells has been shown to limit HCV replication in vitro[17]. Thus, there is increasing evidence that MAIT cells could play a protective role in viral infections. DENV and ZIKV infections are associated with a range of clinical symptoms. More studies are needed to investigate if MAIT cell frequency, functionality or activation status have an impact on the clinical outcome of DENV and ZIKV infections. In this regard, MAIT cell production of IFNγ could be part of an innate immune response to induce an anti-viral state and compromise Flavivirus replication. Levels of serum IFNγ have been reported to be inversely associated with DENV load and symptoms[34]. One limitation of our study is that we focused only on peripheral MAIT cells. MAIT cells are present in the skin[35, 36] and skin resident MAIT cells may play a more important role in early innate defense following mosquito transmission of Flavivirus.
MAIT cells from HIV-1-infected individuals have been shown to have a lower production of cytokines in response to E. coli stimulation[19]. In this study, we show that the cytokine mediated MAIT cell response to in vitro viral infection is also impaired. However, MAIT cells from HIV-1-infected subjects had a normal response to direct cytokine stimulation, suggesting that poor IL-12 and IL-18 production in response to ZIKV infection could be responsible for the impaired MAIT cell response in these individuals. This suggests that HIV-1-infected individuals could have a poor innate immune response to ZIKV and be at a higher risk to develop complications following Flavivirus infection. Case reports of HIV-1-infected individuals with ZIKV infection have been reported[37, 38], including one case with congenital Zika syndrome[39]. Defective MAIT cell activation could be one factor contributing to the increase incidence of severe dengue in HIV-1-infected subjects[40]. More studies are needed to determine if MAIT cells contribute to protection or to immunopathology during Flavivirus infections.
Overall, our results show that MAIT cells are activated in response to DENV and ZIKV infections. This innate response was TCR-independent and defective in HIV-1-infected individuals. Further studies are necessary to determine the importance of MAIT cell responses in the clinical outcomes of Flavivirus infections.
|
10.1371/journal.pntd.0000447 | Strong Host-Feeding Preferences of the Vector Triatoma infestans Modified by Vector Density: Implications for the Epidemiology of Chagas Disease | Understanding the factors that affect the host-feeding preferences of triatomine bugs is crucial for estimating transmission risks and predicting the effects of control tactics targeting domestic animals. We tested whether Triatoma infestans bugs prefer to feed on dogs vs. chickens and on dogs vs. cats and whether vector density modified host choices and other vital rates under natural conditions.
Two host choice experiments were conducted in small caged huts with two rooms between which bugs could move freely. Matched pairs of dog–chicken (six) and dog–cat (three) were assigned randomly to two levels of vector abundance and exposed to starved bugs during three nights. Bloodmeals from 1,160 bugs were tested by a direct enzyme-linked immunosorbent assay.
Conditional logistic regression showed that dogs were highly preferred over chickens or cats and that vector density modified host-feeding choices. The relative risk of a bug being blood-engorged increased significantly when it fed only on dog rather than chicken or cat. Bugs achieved higher post-exposure weight at higher vector densities and successive occasions, more so if they fed on a dog rather than on a cat.
Our findings strongly refute the hypothesis that T. infestans prefers to blood-feed on chickens rather than dogs. An increase in dog or cat availability or accessibility will increase the rate of bug feeding on them and exert strong non-linear effects on R0. When combined with between-dog heterogeneities in exposure, infection, and infectiousness, the strong bug preference for dogs can be exploited to target dogs in general, and even the specific individuals that account for most of the risk, with topical lotions or insecticide-impregnated collars to turn them into baited lethal traps or use them as transmission or infestation sentinels based on their immune response to Trypanosoma cruzi or bug salivary antigens.
| Chagas disease is a complex zoonosis with more than 150 mammalian host species, nearly a dozen blood-sucking triatomine species as main vectors, and 9–11 million people infected with Trypanosoma cruzi (its causal agent) in the Americas. Triatoma infestans, a highly domesticated species and one of the main vectors, feeds more often on domestic animals than on humans in northern Argentina. The question of whether there are host-feeding preferences among dogs, cats, and chickens is crucial to estimating transmission risks and predicting the effects of control tactics targeting them. This article reports the first host choice experiments of triatomine bugs conducted in small huts under natural conditions. The results demonstrate that T. infestans consistently preferred dogs to chickens or cats, with host shifts occurring more frequently at higher vector densities. Combined with earlier findings showing that dogs have high infection rates, are highly infectious, and have high contact rates with humans and domestic bugs, our results reinforce the role of dogs as the key reservoirs of T. cruzi. The strong bug preference for dogs can be exploited to target dogs with topical lotions or insecticide-impregnated collars to turn them into baited lethal traps or use them as transmission or infestation sentinels.
| Host choice of hematophagous insects mainly depends on relative host abundance and proximity, host defensive behavior, the density of blood-sucking insects, and the spatial and temporal concurrence of hosts and insects [1],[2]. Examples of innate (genetically determined) host-feeding preferences are few, and convincing evidence with both experimental and field support is scarce [1],[3]. Fleas (Xenopsylla conformis) do not have an innate preference but can discriminate between juvenile and adult hosts, and derive a higher reproductive reward when feeding on juvenile hosts [4]. In Lutzomyia longipalpis sandflies, host size was the main determinant of host-feeding choices among a human, a dog and a chicken exposed simultaneously to laboratory-reared sandflies [5], and its feeding success on chickens was density-dependent [6]. For Triatoma infestans bugs [7] and Simulium damnosum blackflies [8], the proportion of insects biting humans was strongly density-dependent. For Glossina palpalis gambiensis tsetse flies, male flies preferred to feed on cattle rather on reptiles in a stable; the host species selected for the second bloodmeal depended on the host encountered for the first bloodmeal, the between-meal interval and the interaction between these two factors [9]. In mosquitoes, acquired feeding preferences are reflected in their tendency to return to the same villages, houses, host species and oviposition sites [10]. A non-homogeneous distribution of vector feeding contacts on the same host species leads to a basic reproduction number of the pathogen (R0) greater than or equal to that obtained under uniform host selection, a result that still holds when groups of mosquitoes and hosts are highly structured in patches [11],[12].
Triatomine bugs (Hemiptera: Reduviidae) are the vectors of Trypanosoma cruzi, the causal agent of Chagas disease. Triatoma infestans (Klug), the main vector of T. cruzi, is a highly domiciliated species that also occurs in peridomestic structures housing domestic animals [13]. Like most species of triatomine bugs, T. infestans shows eclectic host-feeding patterns [14],[15]. Host proximity has usually been considered more important than host preference for hungry bugs seeking to feed [14]. In laboratory-based host choice experiments of Triatoma sordida (a species typically associated with birds), first-instar nymphs significantly preferred birds to humans [16] whereas fifth-instar nymphs feeding success and bloodmeal size were significantly larger on guinea pigs than on pigeons [17]. Triatoma infestans preferred caged chickens to guinea pigs though not in all replicates [18]. In a simultaneous exposure of four caged vertebrate species to separate groups of fifth-instar nymphs of T. infestans, Triatoma dimidiata and Rhodnius prolixus, none displayed dominant host-feeding preferences among dogs, chickens and opossums but toads were only rarely fed upon [19]. These authors [19] concluded that T. infestans showed a slight preference for dogs in short daytime experiments and a slight one for chickens in overnight trials. No measure of variability in host-feeding choices between the 7–22 replicates for each triatomine species was reported and neither were statistical procedures described. Within the restricted experimental conditions used, the tested triatomine species do not appear to have a fixed or dominant preference for any of the study hosts, and the question whether there are host-feeding preferences between dogs and chickens is still unresolved.
In rural areas of the Argentine Chaco, domestic T. infestans blood-fed more frequently on dogs or chickens than on the human hosts or cats with which they shared sleeping quarters [7]. Seasonal host shifts were recorded. In spring-summer bug collections, the proportion of domestic bugs that fed on dogs increased significantly with increasing numbers of dogs and T. infestans in bedroom areas, and decreased as bug feeding frequency on chickens rose. Feedings on cats increased significantly with the number of cats and decreased with the number of dogs in bedroom areas. Dog-fed T. infestans had higher infection prevalence with T. cruzi than bugs feeding on other hosts, but many bugs within a given house fed on up to four different bloodmeal sources in summer [20]. Both domestic dogs and cats acted as a source of T. cruzi infection to other species, including humans, whereas chickens (not susceptible to T. cruzi) contributed strongly to bug population growth [21],[22]. Using molecular typing techniques of T. cruzi, we recently showed that dogs, cats and a large fraction of the T. infestans within a household shared the same parasite sublineage and therefore were connected epidemiologically [23].
Understanding the factors that affect the host-feeding selection patterns of triatomine bugs is crucial to estimating transmission risks and predicting the putative effects of introducing or removing domestic animal hosts or targeting them for control. The recent emergence of pyrethroid resistance in T. infestans in northern Argentina and Bolivia [24], combined with the low effectiveness of standard residual spraying of pyrethroid insecticides in peridomestic structures [25], gave strong impetus to the search for cost-effective, alternative treatments based on the application of powder, topical lotions or insecticide-impregnated collars to the domestic animals themselves [26]–[28]. Whether chickens or dogs would be the preferred targets is one of the questions that motivated the current study, in which we report the first binary discrete host choice experiments of triatomine bugs conducted in small mud-and-thatch huts under natural climatic conditions. We tested whether T. infestans displayed host-feeding preferences between dogs and chickens and between dogs and cats, all unrestrained, and whether the density of vectors per hut modified host-feeding success, blood-engorgement and other vital rates in replicated trials. Analysis of field blood-feeding patterns and laboratory experiments supported the hypothesis that T. infestans would prefer chickens to dogs and dogs to cats, though the evidence regarding chickens and dogs was inconclusive [15],[19],[29]. We also hypothesized that blood-feeding success, engorgement and post-exposure bug weight would be reduced in a density-dependent way [13],[30]. To infer the putative processes accounting for the observed discrepancies, we re-examined the reported host-feeding patterns of domestic T. infestans in the field in light of experimental host choices and the demographic and behavior patterns of domestic animal hosts.
The trials were carried out in the field station run by the Argentinean National Vector Control Program in Punilla, Province of Córdoba (31°14′S, 64°28′W) in summer (late January) and in early winter (June) 2006. Study location and experimental set-up were previously described [27],[31]. For the present study, six small experimental huts simulating typical mud-and-thatch houses (1.60×0.80×0.80 m with a 40 cm-wide entrance) were built and subdivided into two equally-sized rooms that shared an adobe-bricked wall with loose bricks; this arrangement allowed the bugs to hide and move freely between rooms. The lower third of the middle wall and all of the other walls were plastered on the inside with a 7∶1 mixture of soil and cement, and a cement carpet was added over the floors of beaten earth. A cage of plastic mosquito netting mounted on a metal frame was placed above each hut to prevent bugs from escaping. The six huts were arranged in two rows over a 50 m2 rectangle.
Seven mongrel male dogs (approximate age range, 4–7 years; mean weight, 10.8 kg; SD, 3.4; range, 7–15) were used in the trial. All dogs had been exposed to T. infestans and had worn deltamethrin-impregnated collars for a four-month period ending six months before the current experiment [27] but not thereafter. According to the collars' manufacturer, the residual effect of the insecticide should cease within six months or one month after removing the collars; since collar use started >10 months before the first trial, no residual effect was expected to occur at this time. Dogs were vaccinated and dewormed with mebendazole prior to the start of the trial; they were kept in kennels made of chicken wire and a roof and fed twice daily. Chickens (all females; approximate age range, 2–2.5 years; mean weight, 2.4 kg; SD, 0.3; range, 1.9–2.8) of Lohmann breed were identified with a color ribbon and kept separately from other animals in a similar pen. Cats were a female and two male adults (approximate age range, 2.5–4 years; mean weight, 2.7 kg; SD, 0.2; range, 2.5–2.9). Dogs, but not chickens or cats, had previously been exposed to T. infestans bites six months before the first trial [27]. Chickens and cats had not been treated with insecticides. During the trials each animal was stationed individually inside a specific experimental hut at sunset and then released every morning into its specific area within the compound. This study complied with guidelines on research and biological testing activities involving live vertebrate animals from the Institutional Animal Care and Use Committee (IACUC) at FCEN-UBA, which is based on the International Guiding Principles for Biomedical Research Involving Animals developed by the Council for International Organizations of Medical Sciences.
The T. infestans bugs used in these experiments were first or third generation from bugs collected in Córdoba, Santiago del Estero and San Luis (Argentina); they had been reared on chickens at the insectary (at 27°C, relative humidity 70%), fed to repletion on the fourth instar, and starved for 2.5 (first trial) to 3.5 months (second trial) prior exposure to the hosts. This long starvation period (>2 months after they molted to fifth instars) normally does not increase bug mortality, and was used to secure that the previous bloodmeals on chickens were completely digested at the time of the trials (i.e., no ‘false positive” bloodmeal). Before release, a 20% sample of triatomine bugs for each trial was weighed individually with an electronic balance (precision, 0.1 mg), and the volume and shape of the bugs' midgut was observed by transparency against a torch light to check their nutritional status semi-qualitatively based on the size of the bug abdomen and occurrence of blood remnants [27],[32]. This classification (by which bugs are scored as unfed, little fed, medium fed, and fully fed) was consistent between observers. All bugs were classified as unfed immediately before the trials.
The first trial was started on late January 2006 (summer) and included six matched dog-chicken pairs, each housed in a different hut. Each pair was randomly assigned to one of two levels of bug abundance (30–31 or 90–91 fifth-instar nymphs of T. infestans); the upper bug density level was chosen because it had revealed negative density-dependent effects on domestic bug host-feeding patterns whereas the lower one did not [7]. The trial was replicated on three successive nights in the absence of any artificial source of light. Each host species was housed in a different room. Hosts were rotated among huts and between rooms every night, so that each individual host was matched with a different individual of the other host species during each of the three nights, and each individual room housed alternate host species in successive nights. Before the hosts were stationed within the huts at 8 p.m., the bugs were placed in a box with holes on the central wall at mid-day, and recovered after dismantling the movable parts of each hut on the next morning at 8 a.m. On recovery, all insects were immediately brought to the insectary, counted, scored for degree of engorgement, kept for 2 days post-recovery and then weighed (to allow them to approach the body weight plateau after eliminating the surplus of water in the bloodmeal), put in a vial labeled with a unique identifier for each bug, and then frozen at −20°C until dissection and bloodmeal identification. A subsample of 20 bugs not exposed to the hosts (control bugs) was frozen at −20°C at the same time as the recovered bugs to check whether there was any residual chicken bloodmeal in them. Given the experimental setup, we exclude the possibility that the small proportions of lost bugs escaped from the caged huts, and assume that lost bugs were most likely eaten by hosts. The proportion of blood-fed bugs is defined as the number of fed bugs (including little-fed, medium-fed and fully-fed bugs) plus bugs in the unfed nutritional class that later were ELISA-reactive to the test host species, relative to the total number of bugs examined for nutritional status; “fed” is therefore a composite category adjusted for bloodmeal reactivity among unfed bugs. The proportion of engorged bugs is defined as the sum of bugs medium fed and fully fed relative to the total number of fed bugs. The second trial, conducted in June 2006 (late fall), included three pairs of dog-cat and used the same protocol as the first trial except that hosts were stationed in the huts at 6 p.m. One replicate could not be finished properly because the cat fled away at the outset; this replicate was excluded from all calculations and analysis.
Temperature and relative humidity inside the huts were measured using data loggers (Hobo H08, Onset) inserted into the thatched roofs of both rooms and on the outside wall of a hut in the first trial, and on each of the three huts in the second trial. In the dog-chicken trial, mean internal temperatures from 8 p.m. (sunset) to 8 a.m. over the three trial nights were 22, 24 and 21°C, respectively; the mean temperature difference between rooms within a hut (dog-to-chicken) in the stated period ranged from −0.2 to +0.7°C. In the dog-cat trial, mean (minimum, maximum) internal temperatures from 6 p.m. to 8 a.m. over the three huts in each trial night were 8.8 (5.4, 13.3), 7.6 (3.7, 12.2), and 10.0°C (7.8, 14.9), respectively. The mean temperature difference between rooms within a hut (dog-to-cat) averaged over the three huts for each trial night was −0.52 (SD, 0.46), +0.09 (SD, 0.27), and −0.41°C (SD, 0.53).
Standardization of the direct ELISA assay was based on previous procedures [33],[34] and the ELISA reagents' manufacturer manual (Kirkegaard & Perry Laboratories (KPL) Inc., Gaithersburg, MD).
The data collected were entered in an Access database. Feeding indices (FI) were calculated as the ratio of the number of bugs that fed on a given host species X to the number of bugs that fed on the matched host species Y (whether or not the bugs that fed on X fed on Y and vice versa). As only one host of each host species was present we did not need to correct for the number of hosts [35].
Four related measures of blood gain by the bugs were used: i) feeding success, a binary variable measuring the likelihood of blood-feeding on any one or on both host species inside the hut (i.e., overall feeding success: fed bugs relative to the number of bugs recovered alive or dead), or on a specific host species as determined by ELISA (i.e., host choice); ii) engorgement (a binary variable including medium-fed and fully-fed bugs: engorged bugs relative to the number of fed bugs); iii) nutritional status (a categorical variable with four levels), and iv) post-exposure bug weight (a continuous variable, measured two days after host exposure). Engorgement and post-exposure bug weight measure the amount of blood imbibed overall or on a given host species. Exact 95% confidence intervals (95% CI) for mean vital rates (i.e., binary variables) were based on the binomial distribution.
The effect size on several binary response variables was estimated by fitting random-effects logistic regression models clustered by hut to the data using the command xtlogit in Stata 9.1 [36]. The use of random-effects models addresses the fact that insects within a hut roughly share the same environment and other undetermined characteristics that may create dependencies between responses within the same cluster of observations. We tested for significant (P<0.05) effects of trial (dog-chicken trial = 1; dog-cat trial = 2), vector density (two levels) and occasion (three levels) on several vital rates: bug recovery (including both dead and alive bugs relative to the number of released bugs); bug loss and mortality (missing and dead bugs relative to the number of released bugs, respectively); overall feeding success and engorgement, as defined above. An interaction term between vector density and occasion was added to each main-effects model. Host-feeding choices were analyzed by conditional (fixed-effects) logistic regression using McFadden's choice model with the command clogit and robust standard errors. These analyses only included unmixed host choices (i.e., dog, other) in each trial because bugs with mixed or no bloodmeal could not be considered for this analysis. To examine whether host choices were modified by vector density levels, occasion and individual dog, interaction terms were added to each of the models. Alternatively, exact binomial tests were used to test for differences between host choices in each replicate relative to the null hypothesis of no selective host choice. Random-effects multiple linear regression with the command xtreg was used to test for significant effects on post-exposure bug weight of vector density, host blood source (unmixed), nutritional status and occasion. Interaction terms were added one by one to the model with main effects and retained in the final model if P<0.1. When the response variable was nutritional status, multinomial logit models were used.
Of 1,622 triatomine bugs released in both trials, 1,536 (94.7%) were recovered and examined for nutritional status (Table 1). Random-effects logistic regression showed that the overall loss rate of bugs was significantly higher in the dog-chicken trial (6.8%) conducted in summer than in the dog-cat trial (2.4%) run in late fall (OR = 0.32, 95% CI, 0.11–0.96, P = 0.042), but the reverse happened with the observed mortality rate (0.3% vs 3.3%, respectively; OR = 12.23, 95% CI, 3.56–41.98, P<0.001), with no significant occasion effects in both cases. Most of the dead bugs recovered were unfed (17 of 21) and had very low weight. Significantly more bugs blood-fed (98.7%) in the dog-chicken trial than in the dog-cat trial (71.4%; OR = 0.031, 95% CI, 0.016–0.058, P<0.001), but among the fed bugs, engorgement status did not differ between trials (46.8% vs 44.4%, respectively, OR = 0.96; 95% CI, 0.68–1.36, P>0.8). Vector density adjusted for occasion effects did not modify significantly any of the vital rates in both trials (Table 1).
In the dog-chicken trial, we observed that most of the bugs were recovered from the thatched roof of the dog's room, followed by the adobe bricks in the mid-wall; the fewer bugs recovered from the chicken's room were in the thatched roof. Of all the bugs with identified bloodmeals, 81.8% had feedings on dogs and 24.0% on chickens. The dog-to-chicken mean feeding index was 7.0 (95% CI, 3.7–10.3). The total mean percentage of insects that fed on dogs only (75.0%, 95% CI, 71.5–78.3%) was significantly higher than that on chickens only (18.0%, 95% CI, 15.1–21.1%) (Fig. 1A). Both feeding choices were highly significantly correlated (r = 0.88, P<0.001) at each hut (Fig. 2A). Only 5.8% (95% CI, 4.1–7.9%) of bugs fed on both hosts, and 1.2% (95% CI, 0.5–2.4%) on none. Conditional logistic regression showed that dogs were highly preferred to chickens (OR = 11.2; 95% CI, 6.2–20.1, P<0.001) and high vector density significantly reduced feedings on dogs (OR = 0.51; 95% CI, 0.32–0.82, P = 0.005), with significantly reduced dog choice at occasion 2 (OR = 0.37, 95% CI, 0.22–0.62, P<0.001). Feedings on dogs were homogeneous among individual dogs (P>0.1). When each replicate was taken separately, dogs were significantly preferred over chickens in 16 of 18 replicates (binomial test, P≤0.001 in 13 replicates and P<0.05 in 3 replicates; one trial was marginally significant, P = 0.06, and one not significant). Eighteen (47%) of the 38 mixed bloodmeals recorded were from a single dog. None of the 20 control bugs not exposed to the hosts were positive for chicken bloodmeal.
In the dog-cat trial, the apparent dispersion pattern of bugs on recovery was more mixed among days; only in one day were most of the bugs located in the dog's thatched roof. Of the bugs with identified bloodmeals, 69.0% had feedings on dogs and 31.6% on cats. The dog-to-cat mean feeding index was 4.8 (95% CI, 2.7–6.9). Significantly more bugs blood-fed on dogs only (48.5%, 95% CI, 44.1–52.8%) than on cats only (22.0%, 95% CI, 18.5–25.7%); both indices were highly correlated (r = 0.71, P<0.001) though the relation was even stronger in the dog-chicken trial (Figs. 1B and 2B). Only 0.4% (95% CI, 0.05–1.4%) of bugs fed on both hosts, and no feeding was detected in 29.2% (95% CI, 25.3–33.2%). Conditional logistic regression showed that dogs were significantly preferred to cats (OR = 7.8; 95% CI, 1.7–35.8, P<0.001) and vector density reduced significantly the likelihood of feeding on dogs (OR = 0.09; 95% CI, 0.01–0.66, P = 0.018), with significantly increased dog choice at occasion 2 (OR = 8.9, 95% CI, 1.6–48.7, P = 0.012). Heterogeneous feeding rates on individual dogs 3 (OR = 6.5, 95% CI, 2.6–11.1, P<0.001) and 4 (OR = 0.24, 95% CI, 0.07–0.84, P = 0.025) relative to dog 2 were recorded. Bugs significantly preferred the dog in five replicates (P≤0.001 in four replicates, P<0.02 in one) and the cat only in one replicate (P<0.001), whereas no significant differences were found in two replicates (P>0.2). Two of the cats frequently allowed the bugs to blood-feed on them though with large variations between nights. In the excluded replicate that had no cat, 93% of the released bugs were recovered and 79% of them were fed on dog only, with no feeding on cat detected.
The relationship between proportional host body weight and host-feeding preferences in both trials is shown in Fig. 3. Two different patterns were obtained. The proportion of bugs that fed on dogs and proportional dog body weight were unrelated in the dog-chicken trial, whereas a significant relationship was found in the dog-cat trial (OR = 1.21; 95% CI, 1.03–1.44, P = 0.022) where exclusion of an outlier value gave a stronger relationship (OR = 1.32; 95% CI, 1.20–1.45, P<0.001).
Table 2 shows the relation between bug nutritional status on recovery of live and dead bugs, post-exposure mean bug weight, and bloodmeal source in both trials. Before exposure to hosts, the distributions of bug weight in the dog-chicken trial (mean, 61.9 mg; 95% CI, 60.3–63.5) and in the dog-cat trial (mean, 59.3 mg; 95% CI, 56.6–62.0) were not significantly different (Anova, F = 2.80, df = 340, P = 0.095). Post-exposure mean bug weight in the dog-chicken trial (238.1 mg) was significantly higher than in the dog-cat trial (136.8 mg) (Anova, F = 432.3, df = 1,503 and 1, P<0.001), and steadily and significantly increased with nutritional status class in both trials (Anova, dog-chicken: R2 = 0.61, F = 515.6, df = 973 and 3, P<0.001; dog-cat: R2 = 0.59, F = 253.0, df = 524 and 3, P<0.001). In total, 1,160 bugs were tested by ELISA and bloodmeals from 1,023 bugs were identified. The percentage of bugs with dog bloodmeal only varied marginally from 74.8% to 81.8% among nutritional classes in the dog-chicken trial (χ2 = 6.60, df = 3, P = 0.086), but increased significantly from 35.0% to 62.5–76.0% in the dog-cat trial (χ2 = 16.1, df = 3, P<0.001). The fraction of bugs with mixed meals on dogs and chickens steadily increased with nutritional status up to 12.1% in the fully-fed bugs. More bugs in the unfed nutritional class were ELISA-reactive in the dog-chicken trial (68.7% of 16) than in the dog-cat trial (15.9% of 126) (Fisher's exact test, P<0.001). Most bugs classified as little-fed in the dog-cat trial and not reactive to dog or cat by ELISA had residual chicken bloodmeals taken in the insectary three months before.
The post-exposure engorged status and mean bug weight of T. infestans according to vector density and individual host blood source are shown in Table 3 and Fig. 4, respectively. In the dog-chicken trial, the percentage of blood-engorged bugs was higher if the bug fed on dog only (43.4–48.5%) rather than only on chicken (34.1–36.7%), whereas bugs with mixed bloodmeals were more frequently engorged (70.4–72.7%) than those with unmixed meals (Table 3). Relative to little-fed bugs (unfed bugs were rare), the relative risk ratio (RRR) of a bug being medium-fed (RRR = 1.62, 95% CI, 1.02–2.57, P = 0.039) or fully-fed (RRR = 2.1, 95% CI, 0.8–5.3, P = 0.14) was significantly higher if the bug had fed on a dog only, after adjusting for significant occasion effects (P<0.001) and non-significant (P>0.4) vector density effects (n = 600, χ2 = 44.3, P<0.001, AIC = 1059.2, df = 10). All two-way interaction terms were not significant. Post-exposure mean bug weight varied significantly (P<0.001) with nutritional status and its interaction with occasion but not with vector density or host blood source (P>0.6) (R2 = 0.606, n = 583, P<0.001) (Fig. 4A). In the dog-cat trial, the dog-fed bugs engorged significantly more than the cat-fed bugs at lower vector densities (64.6% vs 18.8%, respectively), but there were smaller differences at higher levels of infestation (44.7% vs 37.0%, respectively) (Table 3). When compared to unfed bugs, the relative risk ratio of a bug being little-fed (RRR = 4.7, 95% CI, 1.8–12.4, P = 0.002), medium-fed (RRR = 7.2, 95% CI, 2.6–19.5, P<0.001) or fully-fed (RRR = 3.8, 95% CI, 0.98–15.1, P = 0.053) increased significantly if the bug had a feeding on dog only, after adjusting for significant occasion effects (P<0.02) and marginal effects (P = 0.056) of vector density (n = 372, χ2 = 31.9, P<0.001, AIC = 745.3, df = 15). Post-exposure mean bug weight (log-transformed to normalize the distribution) was significantly modified by host blood source (P = 0.022), vector density (positively, P = 0.008), nutritional status (positively, P<0.001) and occasion (P<0.001) (Fig. 4B). Addition of interaction terms revealed significant effects (P<0.03) between occasion and vector density or nutritional status (n = 372, R2 = 0.75, P<0.001).
Results of this study strongly refute the hypothesis that T. infestans prefers to blood-feed on chickens rather than on dogs. T. infestans consistently preferred dogs and engorged significantly more on dogs than on chickens or cats. Although chickens were relatively more often selected than dogs in domestic bug collections during the hot season [7] and in overnight experiments [19], in our semi-natural experiments T. infestans consistently preferred dogs to chickens despite: i) the bugs having had prior feeding experience on chickens in the insectary; ii) chickens having higher average body temperature (41–42°C) than dogs or cats (38–39°C), and iii) T. infestans blood-feeding more easily on pigeons or chickens than on mice and probably on other mammals in general (depending on the exact biting site) because bird blood lacks platelets and some coagulation factors and therefore is easier to imbibe [37]. Strict quality control methods practically exclude the chance of ‘false negative’ bloodmeal results, and restricted unspecific results to <3% (only in the case of chicken bloodmeals).
Several non-mutually exclusive factors may contribute to explain the feeding preference of T. infestans for dogs rather than chickens or cats: i) larger body size and surface; ii) greater attractiveness mediated by dog-specific cues or odors, yet to be identified and demonstrated, and iii) less defensive reactions to bug bites. On average, the study dogs had 4.5 times (range, 3.7–5.4) more biomass than the study chickens and 4.0 times (range, 2.8–5.2) more than the cats, thereby producing more total heat, moisture and carbon dioxide – known bug attractants. Yet the quantitative preference for feeding on dogs only was directly and significantly related to relative dog biomass in the comparison with cats but not with chickens. A laboratory-based comparison of host-feeding choices between chickens and guinea pigs also showed no direct relationship with host biomass [18]. It appears that relative host biomass would play a lesser role in determining feeding choices of T. infestans between two animal host species than in the case of mosquitoes and humans [38], possibly because of differences in behavior between host species.
The observed host-preference for and higher degree of blood-engorgement on dogs may be explained by higher host attractiveness and less effective host defensive behavior, factors that have been found to be inversely associated in other insect vectors, as predicted by evolutionary theory [39]. Our study provides the first evidence showing that the dog-fed bugs achieved a significantly larger relative degree of engorgement than the chicken- or cat-fed bugs, and higher post-exposure bug weight than the cat-fed bugs. The apparent inconsistency between blood-engorgement status and bug weight at two days after exposure in the dog-chicken trial may be attributed to large differences between chicken and dog blood in water and DNA content that modify the rate of bug diuresis and blood digestion depending on the type of blood imbibed. In field studies, both dogs and chickens were taken to be more tolerant to domestic T. infestans than cats because (i) dog and chicken bloodmeals were more often unmixed than cat bloodmeals (i.e., repeated or continuous feedings on cats were much less likely than on other hosts) [15],[20],[21], and (ii) the likelihood of feeding on dogs was positively density-dependent after adjusting for the observed number of each host species at each household [7]. However, the same mixed host-selection pattern could have been caused by equally tolerant hosts having radically different exposure patterns (i.e., stable for dogs and chickens vs unstable for cats) as discussed below. Unlike the study dogs, rural dogs in some endemic areas in northern Argentina frequently were larger, undernourished (sometimes severely), had lower hematocrits and suffered from other diseases [40]. This combination of factors probably made these rural dogs less responsive and allowed bugs to engorge more easily. Our current results show that the bugs' apparent preference for dogs still holds when the dogs were well-fed and healthy. The direct association between having fed on a dog and bug engorgement level or post-exposure weight demonstrates that on average, dogs were more tolerant to bug bites than cats or chickens. The host choice experiment results are therefore consistent with host-feeding patterns in real-life settings.
Mixed bloodmeals on dogs and chickens obtained within the same night most likely stemmed from interrupted, incomplete feeds caused by host-defensive behavior, and cannot be attributed to previous meals at earlier instars in the insectary (as shown by control bugs). Although our experiment does not allow to distinguish the host on which the bugs with mixed meals engorged first, bugs probably switched from the more defensive to the more tolerant or preferred host (i.e., dogs) to satiate, as suggested by mosquito-bird studies [2],[41]. Overnight videotaping of a chicken exposed to the bugs in one of our huts showed strong anti-feeding activity expressed in stamping and other defensive reactions (M.C. Cecere, unpublished observations). These anti-bug behaviors were frequently reported by local villagers when infestations associated with brooding hens were large. Similarly, after a two-week exposure to two matched pairs of hosts, T. infestans bugs were more frequently found spatially associated with guinea pigs than with chickens [18]. In our experimental setup, both chickens and dogs were sources of bug mortality [27],[31] whose relative magnitude is unclear. In addition, strong host heterogeneity in response to bug activity reflected in that individual dogs differed largely in the likelihood with which they provided a bloodmeal relative to cats, not to chickens; their contribution to mixed bloodmeals relative to chickens, and background history of eating bugs. When chickens were fed upon, they apparently allowed as large a bloodmeal as dogs.
Both trials displayed significant vector density-dependent effects on individual host choice, blood-engorgement and/or post-exposure bug weight (not on overall feeding success or other vital rates measured at the hut aggregate level), though such effects were more complex than expected. In laboratory settings several triatomine bug species frequently showed negative density-dependent engorgement rates on non-anesthetized, unrestrained, small hosts including mice, hamsters, guinea pigs, small chickens and pigeons [13], [17], [30], [42]–[44]. No such experiment including dogs or cats has been reported. In our study, negative density-dependent dog choice implied that host shifts occurred more frequently at higher vector densities, as predicted by optimal foraging models [2]. However, both engorgement and post-exposure bug weight were not modified by vector density in the dog-chicken trial, whereas bugs in the dog-cat trial achieved higher post-exposure engorgement and weight at higher vector densities over occasions, more so if they had fed on a dog. This may be due to the development of increasing host tolerance to increasing bug bites over successive occasions, as observed before [42], or to a relative increase in mean temperature from 7.6 to 10°C between occasion 2 and 3. Although the effects of vector density on bloodmeal size were highly significant, they were not particularly intense (Fig. 4B). However, the tested vector densities (and inferred attack rates) in the experiments were rather low compared with intense domestic infestations sometimes reaching a few thousand domestic T. infestans bugs per house. Density-dependent effects need to be investigated at a wider range of densities for a full description of the functional relationship between vector density and host-feeding success or bloodmeal size.
Temperatures were nearly homogeneous within each trial but differed largely between trials. The lower bug feeding success and post-exposure weight in the dog-cat trial conducted in late fall is most likely explained by a sharp fall-off in temperature relative to the dog-chicken trial conducted in summer, rather than by the presence of less suitable hosts (i.e., cats). Low temperatures also increased overnight mortality rate; probably reduced bug mobility, exposure to the hosts and losses due to predation, and reduced diuresis regardless of bloodmeal size [45]. We chose to complete the experiments in the fall because T. infestans populations blood-fed and maintained a high nutritional status during the cold season in the same study setting ([31], unpublished data), unlike in other places [46],[47]. Given that the spontaneous locomotory activity of T. infestans was strongly reduced below 18°C in laboratory experiments [48], it is remarkable that most of the bugs succeeded in blood-feeding at hut internal temperatures averaging 7.6–10.0°C (range, 3.7–14.9°C) during the host exposure period, or at 10.8–12.4°C during the first two hours after sunset when most bugs engage in host-seeking activities. Both the success and intake rate of T. infestans from an artificial feeding apparatus increased linearly with blood temperature between 6 and 35°C, but this occurred after an initial stimulation with a heated steel plate at 35°C [49]. In our study, the relationship between temperature and bug activity may have been severely modified by the use of starved bugs and the size and confined features of the huts, implying increased motivation and suitable conditions for host-seeking activities. A reappraisal of the association between the feeding frequency of T. infestans and temperature under different bug physiological states and field conditions is warranted.
Our experiments quantified host-feeding preferences in replicated trials using mongrel animals similar to those owned by rural villagers (though our dogs were older), and were conducted at more natural conditions than laboratory-based trials. Unlike field-based host-feeding patterns, host choices were not confounded by nonrandom temperature variations between house structures or huts over time; unequal host availability or accessibility, or by stationing host species in fixed rooms within each hut. Therefore, the patterns revealed by the host choice experiment may be taken as the base case against which to compare the more complex field data for additional inferences on the effects of host availability or accessibility.
The observed preference for dogs rather than chickens contradicts field data analyzed by household and bug collection habitat [7]. The dog-to-chicken median feeding index (0.4; interquartile range Q1–Q3, 0.1–1.6) of domestic T. infestans in spring-summer tended to favor chickens and was highly variable among households, whereas in the host choice trial the mean FI consistently favored dogs (7.0) and was 12-fold larger. Similarly, the median dog-to-cat FI of domestic T. infestans across seasons favored dogs (1.7; Q1–Q3, 0.5–4.1; calculated from 24 dog- and cat-owning households with dog and cat bloodmeals from the Appendix in [7]), but in the host choice trial the mean FI was nearly three-fold larger (4.8). Therefore, on a per capita basis and assuming host availability at observed values, T. infestans blood-fed on dogs much more often than on cats in the experimental host choice trial than in the field by a factor of 2.8 (i.e., feedings on cats occurred relatively more often in the field), with large variability in both settings.
Demographic and behavior patterns of domestic animals may explain the observed discrepancies between field and experimental host choice data. In rural villages, most of the between-house variability in chicken feedings may be attributed to the seasonal brooding curve of fowls, which peaked in mid-spring and decreased during the hot summer months, and to heterogeneous fowl breeding practices among households [50]. Domestic T. infestans took advantage of the constant indoor location of brooding hens during extended periods (put indoors for protection against predators), and of their apparent less responsive behavior during brooding. With respect to dogs and cats, both populations greatly differed in abundance (3∶1) and were approximately stationary in size, with high annual turnover rates (>30%) and predominance of males (range, 67–85%) in northern Argentina [51]. Although 50%–68% of the dogs and cats were reported to sleep in domestic sites (albeit with undetermined frequency), they were not restrained or confined and many roamed freely within and around the village to forage for all or part of their food. Dogs were sometimes reported to free-range in small packs at night (probably in oestrus groups with several males), whereas 56% of the cats were reported to stray in the forest to hunt. These qualitative features appear to be common in resource-poor rural settings in northern Argentina where Chagas disease is hyperendemic, and where there is little demarcation of property lines. In other rural locations, adult male dogs and cats have a larger home range and longer duration of activity than females; dogs display a particularly marked crepuscular free-ranging behavior with two peaks of daily activity [52],[53]. These peaks of activity mostly coincide with the main host-seeking periods of T. infestans bugs, and would reduce host availability and the likelihood of dog-vector encounter at such periods. We infer that current domestic bug host-feeding patterns reflect limited, heterogeneous dog exposures, and there would be ample room for increased feedings on dogs with increased dog exposure.
The long-standing controversy on the role of cats in the domestic transmission of T. cruzi infection revolved around the notion that their well-known nocturnal activity pattern and apparent intolerance to bug bites would reduce the likelihood that domestic bugs blood-fed on cats [29]. However, several studies showed that domestic T. infestans and other triatomine species blood-fed on cats as frequently as or more than on dogs [7],[20], a pattern that also emerged in our host choice experiment. To the apparent inconsistencies between different pieces of evidence already discussed [21], we add geographic variations in pet ownership and keeping practices, and heterogeneities in host behavior and exposure between demographic subgroups of cats (by age, sex and reproductive status). In our rural study area, kittens and young pups were frequently kept indoors all day long. Cats avoided exposure to hot weather by resting indoor in the cooler, dark mud-and-thatch houses during the daytime. Under such conditions, residents of highly infested houses sometimes reported to be attacked by T. infestans nymphs when resting on the floor indoors at noon (unpubl. observations). Age- and sex-dependent host activity patterns modify host exposure to domestic vectors, which combined with individual host heterogeneities and other factors, explain the large variability in feeding patterns observed between households and studies.
Some aspects of our study design limit the interpretation of results. Lack of direct observations on host attractiveness and defensive behavior limits the interpretation of the observed host-feeding preferences in terms of underlying mechanisms. Although both trials had the same initial density of bugs per hut, different temperatures affected attack, feeding and other vital rates; therefore, we refrained from establishing quantitative comparisons between trials and focused on within-trial outcomes. Bloodmeal size was not recorded directly, but nutritional status and post-exposure weight were strongly positively correlated; both may be considered valid surrogate indices of bloodmeal size given that pre-exposure bug weight was very similar between trials. Post-exposure bug weight should be increased roughly by 50% to estimate the actual weight immediately after feeding because of the rapid loss of excess water during the first day or two post-feeding depending on temperature. This rough calculation and comparison with laboratory-based estimates of bloodmeal size suggests that most of the bugs blood-fed to repletion in the summer trial. Because only the study dogs had been bitten by T. infestans before the experiments, differences in their background experience with bugs might modify the bugs' preference for dogs. This effect may not be too serious, considering that naïve chickens have an innate, stable tendency toward tolerating or defending from triatomine bites even after a rest time between exposures [42]. The individual study dogs also expressed some idiosyncratic behaviors against bugs over time. Despite the fact that the study dogs had been exposed to T. infestans bites six months before the first trial (i.e., were presumably immunized against bug salivary antigens), most of the dog-fed bugs blood-engorged close to repletion. Therefore, it seems very unlikely that the dogs' immune response reduced substantively vector host-feeding success or engorgement. Immune reactions to bug saliva on the skin of previously sensitized chickens facilitated vector blood-feeding [54] but we have not observed such skin reactions in dogs. It is unlikely that previous dog-bug contacts may have influenced the various steps involved in the host-selection process prior to engorgement; whether the degree of bug engorgement was modified by prior exposure of dogs remains to be determined. Unlike chickens found indoors in rural villages, the study hens were not brooding and their behavior in response to bugs may be potentially different.
Increased host tolerance implies increased residence and feeding times on the host, which in turn will increase fitness by increasing the overall rate at which blood is obtained, eggs are produced, and survival per feeding attempt [55]. The nutritional quality of blood may differ substantively between host species of R. prolixus [56], with chicken blood having half the hematocrit than mammals and much lower hemoglobin or plasma protein than dogs [57]. Therefore, the aggregate fitness implications of host choices remain to be established. Of note, the host bloodmeal choice variable includes a survival component because it was measured on recovered, fed bugs.
Because bloodmeal size increases the probabilities of T. infestans emitting dejecta sooner [58], ingesting trypanosomes and becoming infected [21], it follows that preferred, tolerant hosts such as dogs will seriously increase transmission rates relative to other domestic hosts. By virtue of allowing larger bloodmeals, the likelihood of dogs being repeatedly contaminated with bug feces and eventually superinfected with various parasite strains would be increased. The large frequency of unmixed dog bloodmeals shown by T. infestans in some field locations further suggests that a strong, stable link between individual dogs or groups of dogs and groups of bugs occurs in some households, thereby increasing transmission of T. cruzi back and forth from dogs to bugs and creating a transient partial refuge for other host species (a zooprophylactic effect). In most households, however, the frequency of mixed bloodmeals on dogs is high during spring-summer, and because domestic host species and bugs are more connected the flux of parasites between them is enhanced. Selective host choice amplified by a greater feeding success on diseased or infected hosts will increase the basic reproduction number of T. cruzi (though with possibly depressed prevalence and incidence as the outbreak follows through) compared with the base case represented by homogeneous contact rates [11],[12],[59]. An increase in dog or cat availability or accessibility in domestic areas will increase the rate of bug feeding on them which in turn will exert non-linear effects on R0 through the squared biting rate term. When the proportion of insects feeding on a given host species (i.e., humans) varies with the relative abundance of non-human (i.e., dogs, cats, chickens) and human hosts and with the ratio of vectors to hosts, as our studies have shown, the relationship between R0 and host blood indices is predicted to be strongly non-linear [2]. This implies that different tactics that seek to reduce vector abundance will exert very different impacts on parasite transmission depending on the exact relationship between R0 and the vector-to-host ratio. The empirical evidence further supports the prediction that removal of dogs from bedroom areas will strongly decrease domestic bug population size, transmission rates and human incidence of infection [22].
Heterogeneities in vector feeding rates and in host exposure and infection will tend to create ‘hot’ and ‘cold’ spots of transmission, which can be used to target more accurately and efficiently host species and individuals accounting for most of the risk. Application of pyrethroid-impregnated dog collars, causing reduced repellency but increased bug mortality for extended periods [27], are predicted to strongly reduce domestic bug population size and transmission rates. The various layers of heterogeneity involving dogs in rural endemic areas, including household aggregation of infection, infectiousness to bugs and exposure patterns [21],[51], can be used when designing control measures. For increased impact, collars or other similar tools should be preferentially applied to those dogs that are infected with T. cruzi and/or highly infectious to bugs and that are also closely associated with domestic sites (e.g., pups, females in reproductive state or restrained dogs). Such dogs can be turned into baited lethal traps, though a thorough cost-effectiveness assessment of such tactics is needed before large-scale field application. Other possible applications are to use dogs as baited sentinels of bug presence through the use of its immune response to salivary antigens for serologic surveillance during a bug elimination campaign [60] and as sentinels of parasite transmission [51].
|
10.1371/journal.ppat.1001333 | A New Model to Produce Infectious Hepatitis C Virus without the Replication Requirement | Numerous constraints significantly hamper the experimental study of hepatitis C virus (HCV). Robust replication in cell culture occurs with only a few strains, and is invariably accompanied by adaptive mutations that impair in vivo infectivity/replication. This problem complicates the production and study of authentic HCV, including the most prevalent and clinically important genotype 1 (subtypes 1a and 1b). Here we describe a novel cell culture approach to generate infectious HCV virions without the HCV replication requirement and the associated cell-adaptive mutations. The system is based on our finding that the intracellular environment generated by a West-Nile virus (WNV) subgenomic replicon rendered a mammalian cell line permissive for assembly and release of infectious HCV particles, wherein the HCV RNA with correct 5′ and 3′ termini was produced in the cytoplasm by a plasmid-driven dual bacteriophage RNA polymerase-based transcription/amplification system. The released particles preferentially contained the HCV-based RNA compared to the WNV subgenomic RNA. Several variations of this system are described with different HCV-based RNAs: (i) HCV bicistronic particles (HCVbp) containing RNA encoding the HCV structural genes upstream of a cell-adapted subgenomic replicon, (ii) HCV reporter particles (HCVrp) containing RNA encoding the bacteriophage SP6 RNA polymerase in place of HCV nonstructural genes, and (iii) HCV wild-type particles (HCVwt) containing unmodified RNA genomes of diverse genotypes (1a, strain H77; 1b, strain Con1; 2a, strain JFH-1). Infectivity was assessed based on the signals generated by the HCV RNA molecules introduced into the cytoplasm of target cells upon virus entry, i.e. HCV RNA replication and protein production for HCVbp in Huh-7.5 cells as well as for HCVwt in HepG2-CD81 cells and human liver slices, and SP6 RNA polymerase-driven firefly luciferase for HCVrp in target cells displaying candidate HCV surface receptors. HCV infectivity was inhibited by pre-incubation of the particles with anti-HCV antibodies and by a treatment of the target cells with leukocyte interferon plus ribavirin. The production of authentic infectious HCV particles of virtually any genotype without the adaptive mutations associated with in vitro HCV replication represents a new paradigm to decipher the requirements for HCV assembly, release, and entry, amenable to analyses of wild type and genetically modified viruses of the most clinically significant HCV genotypes.
| Two decades after its identification, hepatitis C virus (HCV) remains a leading cause of serious liver diseases worldwide. The poor in vitro propagation of patient isolates has impaired their study. Conversely, viral strains of the most prevalent (∼70% of total infections) and clinically problematic (∼45% cured with the standard of care) genotype 1 adapted for in vitro replication display mutations impairing yield and/or in vivo infectivity. We established a new cell culture model for producing infectious HCV in a cell line stably bearing a subgenomic replicon from West Nile virus (a flavivirus belonging to the same family as HCV) that circumvents the requirement for HCV RNA replication. To study viral infectivity in vitro, we devised several HCV genome-based constructs. This system produced wild type HCV particles of subtypes 1a, 1b, 2a and a 1b/2a chimera. All specifically infected permissive target cells, and HCV particles containing wild type genomes known to be infectious in vivo infected human liver slices ex vivo. The production of authentic HCV particles independent of HCV RNA replication represents a new paradigm to decipher requirements for HCV assembly, release, and entry, amenable to analyses of wild type and genetically modified viruses of the most clinically significant genotypes.
| HCV infects 2–3% of the world population. A majority of infected people fail to clear the virus and are at risk for developing serious liver complications (reviewed in [1]). HCV belongs to the genus Hepacivirus in the Flaviviridæ family, and at least six genotypes have been identified so far [2]. Greater than two thirds of HCV infections diagnosed worldwide are of subtypes 1a or 1b [2]. There is no approved vaccine and available treatments are much less effective against genotype 1 compared to other genotypes. The limited experimental availability of chimpanzees, the primary animal model for HCV [3], [4], and difficulties encountered in reproducing true infection in small animals have significantly limited the use of in vivo models to study the biology of this virus. The structure of the intact virion is unknown, and it is still unclear how the RNA genome [5] circulates in infected patients. In addition, although the natural target cells of HCV are primarily hepatocytes in the liver, in vitro most human hepatic cells poorly propagate HCV isolates from patients (e.g. [6]). In vitro studies were nevertheless marked by two breakthroughs allowing for the screening of new antiviral compounds. First, subgenomic replicons (i.e. without structural genes) of subtypes 1b [7], [8] and 1a [9] were established in selected subclones of the human hepatic Huh-7 cell line that are highly permissive for HCV replication, e.g. Huh-7.5 cells [10]. Subsequently, a full infectious cycle was reproduced in cell culture with JFH-1, a particular strain of genotype 2a [11], [12], or with a J6/JFH-1 chimera [13]; the released particles are referred to as HCVcc.
Although propagation of a few HCV strains in replication-permissive cell lines has been an important contribution to the field, it has long been recognized that these models are complicated by the particularly high error rate of the HCV RNA replicase [14]. Combined with the in vitro selective pressure, e.g. associated with the modifications acquired by the permissive cell lines [15], or viral recombination between genotypes [16]–[18], it inevitably results in the emergence of adaptive/escape variants [19]. However, cell culture-adapted HCV most often displays lack of infectivity or impaired fitness in vivo [20], [21]. Conversely, HCV genomes with a consensus sequence that are infectious in chimpanzees are not infectious in cell culture, e.g. in Huh-7.5 cells [9], [22]. This issue is especially perplexing with genotype 1 strains, for which the accumulation of cell-adaptive mutations that enhance its RNA replication results at best in low yields of HCVcc with impaired infectivity [19], [23]. Intergenotypic JFH-1 chimeras have been engineered to tentatively overcome such limitations [16]–[18] but have been shown to accumulate structural gene compensatory mutations [16]. As such mutations and their associated complications result from the viral RNA replication process, we reasoned that uncoupling the production of infectious HCV particles from HCV RNA replication would circumvent major limitations associated with existing in vitro systems requiring such coupling.
All known Flaviviridæ members replicate in the cytoplasm of their target cells and induce membrane rearrangements mostly deriving from the endoplasmic reticulum (ER) [24], [25]. Strongly connected to RNA replication [26], assembly of infectious flavivirus particles occurs within a distinct sub-compartment of rearranged membranes [27], [28]. It has been possible to produce flavivirus virions by providing their structural genes in trans. Thus, upon expression of WNV structural genes: core, pre-membrane (prM) and envelope (E), baby hamster kidney (BHK)-21 cells carrying a WNV subgenomic replicon encoding a reporter gene release infectious WNV reporter-particles (WNVrp) containing subgenomic replicon RNA [29], [30]. Although distantly related within the Flaviviridæ family, the Flavivirus and Hepacivirus genera display common features [31]. We therefore examined whether, as for WNV, infectious HCV particles could be formed when the structural proteins are encoded in trans. While we did not observe such trans-complementation in a HCV replication-permissive cell line, we made the surprising observation that non-hepatic mammalian cells previously used to study the biology of Flaviviridæ (including HCV) and bearing a flavivirus subgenomic replicon can produce infectious HCV of diverse genotypes from genomic RNA produced by a plasmid-based system involving cytoplasmic transcription by bacteriophage T7 RNA polymerase. The lack of involvement of the HCV RNA replication machinery avoids the occurrence of cell-adaptive mutations in the HCV genomes.
In initial analyses of the possible effects of flavivirus replicons on HCV virus particle production from proteins provided in trans, we observed that release of HCV structural proteins (expressed from a cytomegalovirus immediate early promoter and harvested by ultracentrifugation) was dramatically enhanced in BHK-21 cells carrying a lineage II WNV subgenomic replicon [32] (referred to as BHK-WNV cells in this study) compared to parental cells; a less pronounced increase was observed in the cell lysate (Fig. 1A). This result suggests that, in the complete absence of HCV RNA replication, the WNV subgenomic replicon had generated a permissive environment for releasing HCV particles. Surprisingly, these effects were not observed in the seemingly more relevant Huh-7.5 human hepatocyte cell line, in which we found that the presence of an HCV subgenomic replicon inhibited rather than stimulated release of HCV structural proteins (both of genotype 1a) provided in trans (Fig. S1A). Moreover, we were unable to stably establish an HCV subgenomic replicon in BHK-21 cells. Based on these results, we considered the potential of the BHK-WNV cell system to produce infectious HCV particles if appropriate HCV-based RNA molecules were generated in the cytoplasm. Such a system might potentially enable virus production of the most prevalent but experimentally difficult genotype 1 strains.
To test this hypothesis, we devised a strategy for generating HCV-based RNA molecules in the cytoplasm of BHK-WNV cells (Fig. 1B). One component of this approach consisted of a dual-plasmid bacteriophage polymerase (p2B) system consisting of the genes for the DNA-dependent RNA polymerases from both bacteriophages T7 and SP6 (T7pol and SP6pol, respectively), each linked to their reciprocal promoter. The other component was a plasmid encoding the HCV genomic sequence of interest flanked at the 5′ end by the bacteriophage T7 promoter, and at the 3′ end by a hepatitis delta virus antisense ribozyme (HDVrbz; cf. Materials and Methods). We reasoned that co-transfection of these two components into BHK-WNV cells would result in cytoplasmic co-amplification of both bacteriophage polymerases; T7 Pol would then drive high level cytoplasmic production of uncapped HCV genomic RNA with correct 3′ termini (by HDV rbz self-cleavage) that would serve as template for translation of HCV proteins (driven by the HCV IRES), including the structural proteins core, E1 and E2. Assembly and release of particles composed of HCV structural proteins and containing the HCV-based RNA might then occur (Fig. 1C), and such particles might be infectious for appropriate target cells.
We first generated HCV bicistronic particles (HCVbp) using a plasmid encoding HCV 5′-UTR to NS2 sequence upstream of the encephalomyocarditis virus (EMCV) IRES of a Huh-7.5 cell-adapted HCV subgenomic replicon of subtype 1a [9], thereby yielding a bicistronic RNA (Fig. 2A, top) capable of replicating in Huh-7.5 cells. Co-transfection of BHK-WNV cells with this plasmid plus the p2B system resulted in the formation of large vesicles (not classical multi-vesicular bodies) filled with 50–60-nm particles in the vicinity of dilated rough ER protrusions and mitochondria, as observed by transmission electron microscopy (Fig. 2B, top panels). In contrast, BHK-WNV cells transfected with a control plasmid (HCV subgenomic replicon minus the HCV structural genes) displayed the extensive membrane rearrangements previously shown to be triggered by the WNV subgenomic replicon [25] (Fig. 2B, lower left panel), such as vesicle packets (site of WNV RNA replication) and convoluted membranes (site of WNV RNA translation and polyprotein processing); however the large vesicles containing particles were not observed (Fig. 2B, lower right panel). Immuno-gold electron microscopy analysis with anti-HCV E1 and anti-core antibodies revealed the presence of the corresponding HCV proteins within membrane rearrangements or large vesicles (Fig. S2) in BHK-WNV cells expressing the bicistronic HCV full length construct.
Fig. 3A shows quantitation of viral RNA (WNV and HCV) in BHK-WNV cells and the corresponding culture supernatants (SN) after their ultracentrifugation. As expected, the cells contained a large amount of WNV RNA generated by the WNV subgenomic replicon, independent of transfection with the HCVbp-encoding plasmid. In contrast, HCV RNA was observed only in cells expressing this plasmid, at levels comparable to the WNV RNA. Strikingly, this was accompanied by the appearance in the SN of a large amount of HCV RNA, which was highly enriched (approximately 100-fold) relative to the WNV RNA. Sucrose density gradient analysis of particulate material from the culture supernatant indicated that the HCV-based RNA migrated over a broad buoyant density range of 1.05 to 1.20 g/cm3 (Fig. 3B). The HCV E1 glycoprotein was detected across the gradient, as were the other structural proteins core and E2 (Fig. S3A, upper panels). These results suggest that the BHK-WNV cell system is capable of releasing particles composed of HCV structural proteins that are preferentially associated with the HCV-based RNA from which they were translated.
We determined that the harvested particles were not exosomes or cell debris, consistent with a requirement for maturation of HCV envelope proteins for particle release in our system (Text S1 and Fig S3A–C). We also excluded that the WNV RNA released upon transfection of the HCVbp plasmid in BHK-WNV cells (Fig. 3A) was associated with infectious particles. First, previous reports suggest a requirement for WNV core protein [26]. In addition, after the transfection of BHK-WNV cells with a plasmid encoding WNV structural proteins, the secreted particles (WNVrp) were infectious for Huh-7.5 cells (Fig. S3D), consistent with previous findings using other target cells [32]. However, incubation of Huh-7.5 cells with HCVbp did not yield any Renilla luciferase activity. Finally, BHK-WNV cells were treated with antiviral drugs for two weeks, which inhibited the WNV replicon (measured by the reduced expression of Renilla luciferase) but did not affect the release of HCV particles (Fig. S3E). It is therefore highly unlikely that HCV RNA replication is responsible for the production of HCV in this system (data on the mechanism will be presented elsewhere).
Several criteria were examined to test the infectivity of the HCVbp in Huh-7.5 cells. First, we used RT-qPCR for the 5′-UTR to test individual fractions from the sucrose density gradient in Fig. 3B for their ability to induce HCV RNA replication. As shown in Fig. 4A, the amounts of HCV RNA in target cells at day 3 post-infection were negligible for nearly all fractions, and increased substantially by day 4. As previously reported for HCVcc [23], [33], we observed that the infectivity of HCVbp was spread over a broad range of buoyant densities, and that it did not directly correlate with the detected amounts of viral RNA. The peak of infectivity generally ranged between 1.08–1.13 g/cm3 (Fig. 4A), which corresponded to a low peak of HCV RNA (Fig. 3B). Infectious titers of HCVbp in the supernatants of BHK-WNV cells were measured in Huh-7.5 cells. TCID50 were between 0.6×104 units/ml at day 3 and 2.5×105 units/ml at day 4 (cf. Text S1), consistent with data presented in Fig. 4A. Such viral titers are about one log lower than with the JFH-1 strain [11], [12] and genotype 2a chimera [13] after a two-day incubation in permissive cell lines, but at least 10-fold higher than with HCVcc obtained with a cell culture-adapted strain of genotype 1 [23].
The relatively low buoyant density of most infectious particles could relate to their association with lipids, since lipid droplets were detected in the vicinity of non-structural proteins in BHK-WNV cells expressing the HCVbp-4cys construct, encoding a tetracysteine tag within NS5A (Fig. S4A). As BHK-21 cells express functional LDL receptor [34], another non-exclusive possibility is that HCV particles interacted with lipoproteins from the culture medium. Incubating HCVbp with (up to 0.15 µg/ml) human VLDL, LDL or HDL in vitro prior to Huh-7.5 cells enhanced the amount of viral RNA accumulating in target cells up to 5-fold (not shown), which would be consistent with a specific interaction of lipoproteins with pre-assembled HCVbp, as previously reported for HCV-like particles (HCV-LPs) [35], lentiviral particles pseudo-typed with HCV envelope proteins (HCVpp) and HCVcc [36].
We also analyzed HCVbp-induced synthesis of HCV proteins and RNA in Huh-7.5 target cells by laser-scanning confocal microscopy. Based on staining with a polyclonal antibody against NS5A (Fig. 4B; specificity of antibody validated in Fig. S4C), NS5A-positive patches were detected in the cytoplasm of Huh-7.5 cells infected with HCVbp for two days (center and right panels), but not in uninfected cells (left panel). Albeit in close proximity with ERGIC53, these patches did not co-localize with this lectin that transports glycoproteins from the ER to the Golgi apparatus, suggesting that NS5A was not associated with a ‘classic’ membrane compartment. We also examined HCV RNA replication in Huh-7.5 cells incubated with HCVbp; after several hours, the cells were treated with actinomycin D to block RNA polymerase II-dependent nuclear transcription, then loaded with 5-bromo-UTP, a nucleotide analog that is incorporated into elongating RNA. Staining of HCVbp-infected cells with anti-bromo-uridine (BrU) and NS5A antibodies resulted in the detection of both signals in a cytoplasmic subcompartment of Huh-7.5 cells incubated with HCVbp (Fig. 4C, center panels). This staining pattern was very similar to that observed in positive control cells, i.e. Huh-7.5 cells bearing an HCV subgenomic replicon (Fig. 4C, right panels), but not observed in the uninfected negative control cells (Fig. 4C, left panels). This result presumably reflects the local incorporation of BrU into replicating HCV-based RNA, as has been shown for flaviviruses [37]. Consistent results were obtained with live cells infected with particles encoding a tetra-cysteine tag in NS5A (Fig. S4D).
Treatment of cells with viral inhibitors (interferon α or β plus ribavirin) prior to their inoculation with HCVbp inhibited the accumulation of viral RNA by ∼10-fold (not shown). The sensitivity of HCV replication to these agents [38] suggests that HCVbp-mediated increase in HCV RNA reflects the activity of the introduced subgenomic replicon. The pre-incubation of HCVbp with serum from an HCV-cured patient (without circulating HCV RNA by PCR) decreased the amount of intracellular RNA (Fig. S4B) detected by RT-qPCR, compared to that with normal/naive human serum, suggesting the existence of a specific interaction of HCVbp with the immune serum (presumably IgG) interfering with their infectivity.
The CD81 tetraspanin has been implicated as an important receptor for HCV entry [39]. Albeit of human hepatic origin, the HepG2 cell line lacks CD81 and is poorly permissive for HCV entry but can be rendered permissive by CD81 expression, as previously shown by infection with HCVpp [40] or HCVcc [41]. We found that stable transduction of these cells with a recombinant lentivirus encoding human CD81 resulted in its surface expression (Fig. S5A); it enhanced the NS5A signal triggered by the incubation of HepG2 cells with HCVbp (Fig. 5A; compare right and left panels).
For more quantitative analyses, we devised a variation of the system involving the production of HCV reporter particles (HCVrp). To this end, the fragment encoding NS3 up to the last third of HCV NS5B in the HCVbp construct was replaced with one encoding the ORF of bacteriophage SP6 RNA polymerase (SP6 Pol; Fig. 5B, top). After HCVrp entry into target cells, these cells were co-transfected with the p2B system plus a plasmid encoding EGFP fused with Firefly luciferase, linked to the T7 and SP6 promoters and an EMCV IRES (Fig. 5B, bottom), and were treated with actinomycin D to decrease the background reporter gene expression in the absence of incoming SP6 Pol-encoding RNA, which triggers reporter gene expression in a dose-dependent manner, independent of most post-entry processes. As predicted, parental BHK failed to release infectious HCVrp (Fig. S5B). Although EGFP expression was also observed, only luciferase activity is reported. We tested the dependence of HCVrp entry (cf. Text S1 and Table S1) on surface molecules previously implicated as essential entry receptors in target cells (Fig. 5C). Inhibition of the Firefly luciferase signal generated by HCVrp entry occurred when Huh-7.5 cells were pretreated with siRNA pools targeting several HCV candidate receptor molecules: SR-B1 [42], CD81 [39], ASGP-R subunits 1 and 2 [43], and to a lesser extent claudin-1 [44] (Fig. 5C, filled bars). The same siRNA treatments had little effect on entry of WNVrp (generated by transfecting BHK-WNV cells with a plasmid encoding WNV structural proteins), as measured by the Renilla luciferase activities encoded by the WNV subgenomic replicon packaged into WNVrp (Fig. 5C, open bars). SR-B1 siRNA was the most effective at blocking both the protein expression (Fig. S5C) and HCVrp entry (Fig. 5C). Consistently, pre-incubation of Huh-7.5 cells with antibodies against CD81 and SR-B1 significantly inhibited HCVrp entry signal (Fig. S5D). The interaction of HCV E2 hypervariable region 1 (HVR-1) interaction with SR-B1 is critical for infection [42] and in vivo infection has previously been neutralized by an antiserum against HVR-1 [45]. Preliminary data (reagent was made available in very limited quantity) shows that incubation of HCVrp with these anti-HVR-1 antibodies also inhibited their entry into Huh-7.5 cells (Fig. S5E).
We also tested the possibility of producing infectious particles based on the ability of the JFH-1 strain to infect Huh-7.5 cells [11]. Plasmids encoding the genomic RNA of JFH-1 [11] or a Con1-JFH1 (1b-2a) chimera [18] under a T7 promoter were transfected into BHK-WNV producer cells and HCV particles were harvested, then incubated with Huh-7.5 cells. Fig. 6A–B shows the detection of viral RNA in the target cells for both constructs. Starting at day 3, increasing RNA amounts were measured, whereas in cells treated with interferon plus ribavirin no such increase was detected (Fig. 6A–B).
As an additional variation of this HCV expression approach, we tested the possibility that BHK-WNV cells could produce authentic infectious HCV particles. We co-transfected BHK-WNV cells with the p2B system and a plasmid encoding a full-length genomic RNA with the consensus sequence of a strain of genotype 1a (H77, Fig. 7A), which has been shown to be infectious in chimpanzees [46]. The ‘wild type’ particles (HCVwt) released into the supernatants were harvested by ultracentrifugation and analyzed by sucrose density gradient centrifugation. In fractions with buoyant densities of 1.08–1.13 g/cm3, spherical particles of 50–60 nm in diameter were observed by negative staining electron microscopy; these were not observed with corresponding fractions from control BHK-WNV cells. Some of these particles were positive by immuno-gold electron microscopy, indicating their recognition by immunoglobulins from an HCV-cured patient (Fig. 7B).
As HCV isolates from patients are poorly infectious in Huh-7 cells [6], the infectivity of HCVwt was tested in liver slices from non-infected patients (negative for HCV, HBV and HIV). Like primary human hepatocytes [47], liver slices can be infected ex vivo with HCVcc (unpublished). The liver slices presumably better reflect the real situation than cell lines do, as both the architecture and cell type diversity of the liver are maintained in their original configuration. After incubation of liver slices with BHK-WNV cell-produced HCVwt (H77 strain) [46] or Huh-7.5 cell-produced HCVcc (JFH-1 strain) [11], specific staining by anti-HCV antibodies was analyzed by multifocal confocal microscopy; after a few days, the signal appeared within the slices at various locations of a few lobules, and increased up to 6–10 days. Fig. 7C (upper panels) shows data obtained at day 8; positive staining by both serum from HCV-infected and monoclonal antibodies against structural proteins was observed (middle panel) within two to five lobules of HCVwt-infected slices (area >1 cm2). The results were similar to those obtained after infection with HCVcc (right panel), and specificity was verified by the undetectable staining in uninfected liver slices (left panel). Infection was detected in clusters of cells within a few lobules, consistent with a recent report showing that HCV infection of the liver involves a limited number of hepatocytes [48]. Results varied in shape and intensity with liver donor, but the specificity of the detected infection signal was further confirmed by additional analyses with control antibodies (Fig. S6).
Similar results were obtained with HCVwt-4cys, encoding a tetracysteine tag within its non-structural gene NS5A, as previously validated in Huh-7.5 cells (cf. Text S1). Six to eight days after their infection with HCVwt-4cys, liver slices were incubated with a permeable biarsenical dye and observed with a two-photon confocal microscope. Specific staining was detected predominantly in a few periportal spaces, and also in mediolobular areas (Fig. 7C, lower middle and right panels) of HCVwt-4cys-incubated slices. In spite of a high background that reduces the sensitivity of detection with this technology, the appearance of small clusters of positive signals (generated in live cells) is consistent with the local synthesis of HCV non-structural proteins in human liver slices after their ex vivo infection with HCVwt-4cys produced in BHK-WNV cells.
As HCV isolates from patients are poorly replicating in Huh-7 cells [6], [9], [22] and access to naïve human liver slices of good quality is limited, we tested the possibility that HCVwt could infect HepG2-CD81 cells, which have been previously reported to support replication of patient isolates [6]. To some extent these cells support HCVbp replication (Fig. 5A). The incubation of HepG2-CD81 cells with HCVwt (produced in BHK-WNV cells) of subtypes 1a, 1b, and to a lesser extent 2a (or 1b/2a chimera; not shown), resulted in high readings starting at day 0 (Fig. 8A–C). Although the detected amounts of HCV RNA sharply decreased during the first 24–48 hr, which could relate to some non-productive binding/uptake, it raised again afterward; the later increase was abolished by a treatment with interferon and ribavirin added to the cells both prior to and after infection (cf. results in Huh-7.5 cells). Incubation of HepG2-CD81 cells with HCVwt of subtypes 1a resulted in more intracellular accumulation of HCV RNA than what was measured after their incubation with HCVbp of the same genotype (not shown); one possible interpretation is that Huh-7.5 cell-adaptive mutations were detrimental to HCVbp replication in HepG2-CD81 cells, similar to what has previously been reported in the liver, in vivo [20].
Producing large amounts of infectious HCV virions in cultured cells has been difficult, especially for the most prevalent and clinically problematic genotype 1, which in part relates to its poor ability to replicate in vitro and the subsequent appearance of cell culture-adaptive mutations interfering with its propagation and infectivity. Here, we produced HCV particles of genotype 1 containing a genome previously shown to be highly infectious in vivo [7], [46]. Their ability to infect human liver slices demonstrates the biological relevance of the particles produced in this in vitro system. Two major features underlie independence from HCV replication, which avoided adaptive mutations typically associated with HCV propagation in cell culture: first, the unique and robust strategy for producing HCV genomes in the cytoplasm independent of HCV replication, and second, the WNV subgenomic replicon that created an appropriate cellular environment for HCV RNA translation as well as particle assembly and release.
HCV particle formation likely took place within membrane rearrangements derived from those induced by the WNV subgenomic replicon, as suggested by immuno-gold electron microscopy results. We also observed that the release of HCV particles by BHK cells was enhanced by lineage I WNV [49] and serotype-2 dengue virus [50] subgenomic replicons, but not by one of Semliki Forest virus [51], an alphavirus belonging to the Togaviridæ family (Fig. S1B). This indicates that, beyond similarities in genomic organizations and sequences [31], the increased production of infectious HCV could result from common functional properties conserved amongst members of the Flaviviridæ family rather than strict sequence specificity of the proteins encoded by the flavivirus subgenomic replicons. In BHK-WNV cells, this possibility is further substantiated by the lack of correlation between HCV production and translation of the WNV subgenomic replicon, upon inhibition of the latter's activity. The replication of flaviviruses and HCV induce similar membrane rearrangements in the cytoplasm of infected cells [24], [25], and our data confirmed that flaviviruses also infect hepatocytes [28], [52]. In Huh-7.5 cells, cholesterol metabolism has been implicated in HCV replication [53] and lipid droplets in its assembly [54]. Likewise, WNV replication involves cholesterol metabolism [55] and, for dengue virus particle formation, the interaction of the viral core with lipid droplets [56]. As the mechanisms involved in the production of HCV by hepatocytes are still debated, these similar features perhaps underlie part of the BHK-WNV cell permissiveness for HCV particle formation.
The correlation between reversal of membrane rearrangements and loss of HCV particles production (not shown) suggests that these rearrangements, and perhaps related cellular changes (e.g. cholesterol metabolism and lipid droplet formation), are playing a major role in the permissiveness of BHK-WNV cells. However, the down or up regulation of other cellular factors could be involved as well. Thus, several intracellular mechanisms involved in innate immunity interfere with flavivirus propagation [e.g. 57]–[59], and knockdown of interferon stimulating mechanisms or signaling pathways enhance WNV [58], [59] and HCV [60] productions in cell culture; WNV [61] and HCV [62] proteins have been shown to directly target such pathways. Here we cannot exclude that such a mechanism took place prior to or upon expression of HCV genes. However, the introduction of a BHK cell-adapted WNV subgenomic replicon into naïve BHK-21 cells rendered them rapidly permissive for the production of WNV, whereas that of HCV appeared after many more passages (not shown). One possible interpretation is that co-evolution of WNV subgenomic replicon and BHK cells under antibiotic selection led to the regulation of additional cellular factors, probably involved in fine tuning WNV replication and/or translation, but absolutely required for the production of infectious HCV. This prompted us to identify such cellular factors in BHK-WNV cells and test their relevance with the JFH-1 strain/Huh-7.5 cells paradigm, the results of which will be presented elsewhere.
The entry assay with particles produced in BHK-WNV cells (HCVrp) requires only the delivery of the associated RNA molecule into the cytoplasm of the target cell where it can be translated at sufficient levels to trigger the dual bacteriophage RNA polymerase amplification system. Thus, the target cell needs to be permissive only for viral entry, and possibly a limited number of post-entry steps (e.g. RNA uncoating). Most importantly, the non-involvement of RNA replication for the signal readout allows assessment of the entry permissiveness of diverse cell types, independent of their ability to support HCV replication. This represents a significant advantage over the HCVcc system that also relies on viral spreading to amplify the read out signal, and the involvement of only HCV structural proteins and RNA clearly distinguishes this system from HCVpp, which is based on non-HCV protein and nucleic acid platform.
We had previously observed that both ASGP-R subunits were required for internalization of HCV materials into hepatocellular carcinoma as well as non-target cells [43]. Here we show that these subunits are involved in delivering HCV reporter RNA into HCV-permissive hepatic cells. It is not yet known whether the role of ASGP-R in HCV uptake relates to incomplete maturation of E1 and/or E2 carbohydrate residues, as previously observed [63], [64], or involves another mechanism [65], [66]. HCV has been reported to enter cultured cells via clathrin-coated pits [67]–[69], and ASGP-R internalization follows the same path [66], [70]. Yet, ASGP-R can be targeted to various intracellular compartments including ER [43], which leaves open the possibility that this receptor plays a role at an early as well as a late step of the HCV entry process and RNA delivery.
As inter-genotypic differences and cell-adaptive mutations could affect viral production in hepatic cells, the BHK-WNV paradigm provides an alternative model to produce wild type virus for in vivo or ex vivo studies without the concern that adaptive mutations develop. It could also present major advantages for deciphering mechanisms of viral translation, assembly, release and entry, including involvement of non-structural genes in viral production independent of their role in replication.
BHK-21 cells were grown in E-MEM supplemented with 10% fetal bovine serum (FBS; HyClone), GlutaMax-I (Invitrogen); BHK cells harboring WNV lineage II SG-replicon encoding Renilla luciferase, BHK WNIIrep-REN cells [32], herein simply called BHK-WNV cells, were propagated in D-MEM supplemented with 10% FBS, GlutaMax-I and 5 µg/ml blasticidin (Invitrogen). Huh-7.5 cells and Huh-7.5 cells harboring HCV SG-replicon of 1a genotype (H77) with mutations in NS3 and NS5A (Huh-7.5-SG 1a rep) were maintained as described [9], [10]. HepG2 cells were grown in E-MEM supplemented with 10% FBS, GlutaMax-I and non-essential amino acid mix. Cells were cultured in an incubator with a 95% air/5% CO2 atmosphere saturated in humidity.
A new system of plasmids (p2B) was designed to amplify the cytoplasmic transcription of plasmids in which the gene of interest is under the control of a DNA-dependent RNA polymerase (DdRp)'s cognate promoter; this system consists of a set of two plasmids generating T7 polymerase (T7 Pol): 1) pCR-T7p/SP6pol in which bacteriophage SP6 DdRp (SP6pol) gene was cloned into pCR2.1 plasmid (Invitrogen) in frame with the second ATG start codon of EMCV IRES under the control of T7 promoter; 2) pSL-SP6p/T7pol in which bacteriophage T7 DdRp (T7pol) gene was cloned into pSL1180 plasmid (Clontech) in frame with the second ATG start codon of EMCV IRES under the control of SP6 promoter. This p2B system was used for all T7 Pol promoter-driven HCV coding plasmids, in which a sequence coding for an HDV antigenomic ribozyme [71] was added at their C termini. p90 HCVconFLlongpU encoding the FL genome of infectious H77 strain [46], or, pH-Neo-SG(L+I) encoding a subgenomic replicon of the same strain with cell-culture adaptive mutations [9] were used as templates to construct all HCV coding plasmids of genotype 1a. HCVbp was produced from p684-SG(L+I)-HDV plasmid, in which the neomycin resistance gene of pH-Neo-SG(L+I)-HDV, i.e. pH-Neo-SG(L+I) encoding an hepatitis delta virus antisense ribozyme (HDV rbz) after the HCV 3′-end, was replaced with HCV 5′-UTR to NS2 coding sequence. An HDVrbz gene was introduced at the 3′-end of p90HCVconFLlongpU to create p90-T7p/H77FL-HDV plasmid that will produce HCVwt, i.e. virus particles containing the full-length, consensus sequence of H77 strain. HCVbp-4cys and HCVwt-4cys were obtained using modified p684-SG(L+I)-HDV and p90-T7p/H77FL-HDV plasmids, in which a tetracysteine tag-encoding sequence [72] had been inserted within the NS5A gene. HCVrp was produced from pCMV(-)T7p/HCV-SP6pol-HDV plasmid that encodes HCV 5′-UTR and structural genes followed by those of SP6pol (entry signal) gene in frame with EMCV IRES and a sequence encoding carboxy-terminus of HCV NS5B (kissing loops) [73] and 3′-UTR. To detect incoming-SP6pol RNA upon HCVrp entry into target cells, pT7-SP6p2/EGFPLuc reporter plasmid was made. This plasmid was derived from pEGFPLuc plasmid (Clontech) in which EMCV-IRES-EGFPLuc expression is under the control of both bacteriophage T7 Pol and SP6 Pol cognate promoters in tandem. This construct lacks eukaryotic promoter and therefore is responsive either to T7 Pol, SP6 Pol, or both; it was found responding to either incoming DdRp, be it in the form of protein or DdRp encoding RNA (not shown). Two additional constructs, pHCVp7 and pHCVcore-NS2 are pcDNA3.1(+)-based plasmids (Invitrogen), respectively encoding HCV 1a structural genes (core, E1, E2, p7) and HCV 1a structural genes plus NS2. pIRES1hyg-WNV [32] encodes WNV structural genes (core, prM and E). These three plasmids are under CMV early promoter (not shown). pJFH1 [11], pFK1-Con1 (9605Con1) [7] and pFK-JFH1Con1C-842 [18] are plasmids encoding from a T7 Pol promoter the genomic RNA of, respectively, the JFH-1 strain (genotype 2a), the Con1 strain (genotype 1b) and a Con1-JFH1 chimera (1b/2a). A DNA fragment encoding an HDV rbz was inserted at the 3′-end of the HCV RNA coding region of each plasmid.
Anti-E2 monoclonal antibodies (ALP98 and AP33) [74] and anti-E1 (A4) monoclonal antibody were used for Western blot analysis, and rabbit polyclonal antibody against HVR1 of E2 [45] for inhibition of HCVrp entry. Anti-NS5A rabbit polyclonal antibody (in-house) was used for confocal microscopy analysis. To produce rabbit antibody against NS5A of genotype 1a, 48-amino-acid peptide: NH2-AEEDEREVSVPAEILRKSRRFARALPVWARPDYNPPLVETWKKPDYEP-COOH, corresponding to position 2261–2308 of the H77 strain was synthesized by Peptide Synthesis and Analysis Laboratory (RTB/NIAID/NIH); a cysteine residue was introduced at the amino-terminus and the peptide was coupled to KLH. Two rabbits were immunized from which two sera were harvested; both IgGs were peptide affinity-purified. Sequence of the peptide is almost identical (but amino acids 22, 25, 43 and 46) to that of Con1 (genotype 1b). Monoclonal antibody against HCV core protein (clone C7-50; Thermo Scientific) was used to analyze Huh-7.5-produced JFH-1 (HCVcc) infection by confocal microscopy. Antibodies against HCV candidate receptors and cellular proteins are as follow: anti-CD81 mAb (JS-81, BD Biosciences); anti-SR-BI rabbit polyclonal antibody (Novus Biologicals); anti-ASGPR-1 mAb (clone 8D7, Santa Cruz Biotechnology); anti-claudin mAb (Invitrogen); anti-Hsp70 (BD Biosciences); anti-ERGIC-53 (Alexis Biochemicals) and anti-BrdU (Invitrogen). FIAsH- and ReAsH-EDT2 labeling reagents were obtained from Molecular Probes (Invitrogen). For flow cytometry and immunofluorescence (confocal microscopy) analysis, the secondary antibodies used were Alexa Fluor 488-, 594-, or 635-conjugated goat anti-mouse and anti-human antibodies, and Alexa Fluor 594-, 635-, or 680-conjugated goat anti-rabbit antibodies from Molecular Probes (Invitrogen).
One day before transfection, BHK-WNV cells were seeded at a density of 6×106 cells per 162-cm2 flask. Plasmids encoding HCV sequence under the control of bacteriophage T7 promoter (or CMV early promoter where specified) were transfected using Lipofectamine LTX and Plus reagent according to the manufacturer's protocol (Invitrogen). Culture medium after transfection was D-MEM supplemented with 10% FBS, 1% non-essential amino acid mix, GlutaMax-I, 25 mM Hepes; cells were incubated at 37°C. One or two days later, 2.5 to 3.7 g/L sodium bicarbonate was added (to prevent further acidification of the medium), and culture medium were harvested at day 3, centrifuged at 30,000× g for 30 min at 4°C to remove cell debris, then clarified supernatants were centrifuged at 100,000× g for 3 hrs at 4°C. Pellets were either resuspended in culture medium and filtered through 0.45 µm PVDF membrane (Millipore), or loaded on the top of a 20–60% sucrose gradient in phosphate-buffered saline solution (PBS; Quality Biologicals, MD), then centrifuged in a SW55Ti rotor (Beckman) at 100,000× g for 16 hrs at 4°C. Gradients were manually harvested from the top in 150 µl fractions. HCVcc (Huh-7.5-produced JFH-1) was obtained by electroporating IVT RNA into Huh-7.5 cells as described [11]. Virus stock was concentrated, aliquoted and stored at −80°C.
BHK-WNV cells (2.5×105) seeded in a 6-well plate were transfected with HCVbp-coding plasmid. Three days later, cells were fixed in 2% glutaraldehyde in 0.1 M sodium cacodylate for 1 hr at RT, then at 4°C, overnight. Cells were subsequently processed for TEM as described [75]. Pooled sucrose fractions containing HCVwt were diluted with PBS then pelleted in Beckman SW55Ti (100,000× g, for 2 hr) at 4°C. Pellets were resuspended in 4% paraformaldehyde in PBS and analyzed for negative staining EM. Serum from a cured HCV patient previously infected with genotype 1a was used to detect HCVwt in the immuno-EM analysis.
Virus-containing supernatant from BHK-WNV cells were clarified at 30,000× g in SW28 Beckman rotor for 30 min, filtered through 0.45 µm PVDF membranes then concentrated (60-fold) with 106 MWCO Vivaspin filters (Sartorius Stedim, Gottingen, Germany). Huh-7.5 cells (7×103) were seeded in a 8-well chamber coverglass (Lab-Tek II, Nalge Nunc) and incubated with HCVbp for 2 hr at 37°C. After virus inoculum removal, cells were grown for another 48 hr to analyze the expression of HCV NS5A protein. Briefly, cells were washed twice with ice-cold PBS and fixed with 4% paraformaldehyde and 0.15 M sodium cacodylate buffer, pH 7.4, for 20 min at room temperature, followed by washing (5 minutes, twice) with PBS containing 50 mM glycine. After washing with PBS, cells were permeabilized with 0.3% Triton X-100 in PBS for 15 minutes at room temperature, then incubated with blocking solution (10% FBS, 3% BSA, 0.3% Triton X-100 in PBS) for 30 min. Cells were then incubated with primary antibodies: rabbit anti-NS5A IgG and anti-ERGIC-53 mAb (in 1% BSA, 0.1% Triton X-100, in PBS) overnight at 4°C. The fluorescent secondary antibodies were Alexa Fluor 488-conjugated anti-mouse IgG antibody and Alexa Fluor 594- or 635-conjugated anti-rabbit IgG antibodies. Nuclei were labeled with DAPI with antifade (Chemicon, CA). To test the infectivity of HCVwt (1a, 1b and 2a) produced by BHK-WNV cells, HepG2-CD81 cells were seeded on 24-well collagen plates, and the following day, cells were incubated with particles in the presence or absence of IFN-α and ribavirin. Total RNA was harvested daily and intracellular HCV RNA was measured by RT-Taqman PCR.
Cells were infected with HCV particles containing a genome encoding a tetracysteine-tag (HCVbp-4cys or HCVwt-4cys): Huh-7.5 cells were infected with HCVbp-4cys for 3 days, then incubated with the cell-permeant FIAsH-EDT2 or ReAsH-EDT2 biarsenical dye according to the manufacturer's protocol (Molecular Probes, Invitrogen). Adding FIAsH (or ReAsH) dye onto live cells expressing TC-tagged proteins should result in a specific fluorescent signal where the tag is present. Samples were observed under a confocal microscope (SP5 X-WLL (white light laser) mono-photon confocal microscope (Leica, Heidelberg, Germany) using a 63× oil immersion objective NA 1.32. Images were deconvolved with Huygens Essential software (Version 5.3, Scientific Volume Imaging BV, Hilversum, The Netherlands). A similar procedure was used to stain cultured human liver slices infected with HCVwt-4cys.
Huh-7.5 cells (7×103) were seeded in 8-well chamber coverglass and one day later, were infected with HCVbp. At 48 hr post-infection, medium was replaced with D-MEM complete medium containing 2.5 µg/ml actinomycin D (Sigma) for 30 min and transfected with 5-bromo-uridine 5′-triphosphate (BrUTP; Sigma) using Lipofectamine 2000 (Invitrogen). Briefly, 1 µl of Lipofectamine 2000 was added to 10 mM BrUTP, both in 25 µl Opti-MEM I, and incubated for 20 min at room temperature. The BrUTP-Lipofectamine complex was added drop wise onto cells and further incubated for 6 hours. Cells were then fixed, permeabilized and incubated with Alexa Fluor 488 conjugated-anti-BrdU mAb. Confocal microscopy analysis was performed as above.
Total RNA from sucrose fractions was extracted with Trizol LS (Invitrogen) and RT-TaqMan PCR of HCV 5′-UTR RNA was performed with QuantiTect Probe PCR kit (Qiagen) using IVT RNA standard corresponds to the HCV 5′-UTR. HCV RNA was analyzed directly from infected cells harvested daily using lysis buffer of TaqMan Gene Expression Cells-to-CT kit (Ambion, Applied Biosystems, Invitrogen); RNA was subjected to a RT step followed by HCV TaqMan qPCR analysis performed with HCV specific primers, and HCV 5′-UTR/NH2-core in vitro transcripts as RT-PCR standards. For HCV and WNV RNA analysis from BHK-WNV cells: Total RNA was extracted from cells and pelleted supernatants with Trizol LS followed by RT using random hexamer and Superscript III at 50°C, for 1 hr. Renilla luciferase-specific primers as the target gene for WNV-SG rep RNA. See Text S1 for details.
The released particles were filtered, concentrated and serially diluted before incubated with Huh-7.5 cells for 3–4 days. NS5A-positive cells were analyzed by immunofluorescence and the number of positive cells was determined using Odyssey In-cell Western system (Li-Cor Biosciences, Lincoln, NE). See Text S1 for details.
The cDNA of human CD81 (hCD81) from Huh-7.5 cells were amplified by reverse transcription (RT)-PCR and cloned into pENTR 2B (Invitrogen) followed by recombination with pLenti6.2/V5-DEST (Invitrogen) to according to manufacturer's recommendation. See Text S1 for details.
Human liver slices were infected with HCVcc (JFH-1) produced in Huh-7.5 cells, HCVwt produced in BHK-WNV cells, or not infected. Six-to eight days after infection, co-immunostaining was performed with HCV serum or monoclonal antibodies, followed by DyLight 488 conjugated-anti-human IgG F(ab′)2 (Jackson ImmunoResearch Laboratories, West Grove, PA), or Alexa Fluor 546 conjugated-anti-mouse IgG goat antibody (Invitrogen). Liver slices were analyzed with a mono-photon multi-focal confocal microscope (Leica SP5 Resonant Scanner, Heidelberg, Germany) coupled to a high resolution CCD. For live-cell staining, human liver slices were infected with HCVwt-4cys (HCVwt encoding a tetracysteine tag) for 6 days, incubated with the cell-permeant TC-FIAsH dye and analyzed (over a thickness of 100–150 µm) as above, using a multi-photon mono-focal confocal microscope (Leica TCS SP5 Resonant Scanner, Heidelberg, Germany).
See Text S1 for details.
See Text S1 for details.
All human samples were obtained during routine medical care and in compliance with the standard Ethical Guidelines of the Institutional Review Board of Cochin Hospital (Paris) that approved the study.
|
10.1371/journal.pmed.1002652 | Risk of adverse outcomes following urinary tract infection in older people with renal impairment: Retrospective cohort study using linked health record data | Few studies have investigated the risk of adverse outcomes in older people with renal impairment presenting to primary care with a urinary tract infection (UTI). The aim of this study was to determine the risk of adverse outcomes in patients aged ≥65 years presenting to primary care with a UTI, by estimated glomerular filtration rate (eGFR) and empirical prescription of nitrofurantoin versus trimethoprim.
This was a retrospective cohort study using linked health record data from 795,484 patients from 393 general practices in England, who were aged ≥65 years between 2010 and 2016. Patients were entered into the cohort if they presented with a UTI and had a creatinine measurement in the 24 months prior to presentation. We calculated an eGFR to estimate risk of adverse outcomes by renal function, and propensity-score matched patients with eGFRs <60 mL/minute/1.73 m2 to estimate risk of adverse outcomes between those prescribed trimethoprim and nitrofurantoin. Outcomes were 14-day risk of reconsultation for urinary symptoms and same-day antibiotic prescription (proxy for treatment nonresponse), hospitalisation for UTI, sepsis, or acute kidney injury (AKI), and 28-day risk of death. Of 123,607 eligible patients with a UTI, we calculated an eGFR for 116,945 (95%). Median age was 76 (IQR, 70–83) years and 32,428 (28%) were male. Compared to an eGFR of >60 mL/minute/1.73 m2, patients with an eGFR of <60 mL/minute/1.73 m2 had greater odds of hospitalisation for UTI (adjusted odds ratios [ORs] ranged from 1.14 [95% confidence interval (CI) 1.01–1.28, p = 0.028], for eGFRs of 45–59, to 1.68 [95% CI 1.01–2.82, p < 0.001] for eGFRs <15) and AKI (adjusted ORs ranged from 1.57 [95% CI 1.29–1.91, p < 0.001], for eGFRs of 45–59, to 4.53 [95% CI 2.52–8.17, p < 0.001] for eGFRs <15). Compared to an eGFR of >60 mL/minute/1.73 m2, patients with an eGFR <45 had significantly greater odds of hospitalisation for sepsis, and those with an eGFR <30 had significantly greater odds of death. Compared to trimethoprim, nitrofurantoin prescribing was associated with lower odds of hospitalisation for AKI (ORs ranged from 0.62 [95% CI 0.40–0.94, p = 0.025], for eGFRs of 45–59, to 0.45 [95% CI 0.25–0.81, p = 0.008] for eGFRs <30). Nitrofurantoin was not associated with greater odds of any adverse outcome. Our study lacked data on urine microbiology and antibiotic-related adverse events. Despite our design, residual confounding may still have affected some of our findings.
Older patients with renal impairment presenting to primary care with a UTI had an increased risk of UTI-related hospitalisation and death, suggesting a need for interventions that reduce the risk of these adverse outcomes. Nitrofurantoin prescribing was not associated with an increased risk of adverse outcomes in patients with an eGFR <60 mL/minute/1.73 m2 and could be used more widely in this population.
| It is not known if older adults with impaired kidney function are at increased risk of hospitalisation or death following a urinary tract infection (UTI).
Nitrofurantoin is an antibiotic used to treat UTI but is not recommended in people with impaired kidney function. However, the evidence supporting this recommendation is limited.
This study used linked health record data from general practices and hospitals in England and estimated risk of hospitalisation and death for older adults with impaired kidney function presenting to primary care with a suspected UTI.
Older adults with impaired kidney function had greater risk of a UTI-related hospitalisation and death in the 14–28 days following a UTI.
Older adults with impaired kidney function who were treated with nitrofurantoin were not at greater risk of an adverse outcome and were less likely to experience a hospital admission for worsening kidney function.
There is a need for strategies that prevent UTIs and reduce the risk of UTI-related hospitalisations and death in older adults with impaired kidney function.
Nitrofurantoin was not associated with worse outcomes and could be used more widely in this population.
| The Kidney Disease Improving Global Outcomes working group defines degrees of renal impairment using the glomerular filtration rate (GFR) [1]. GFRs <60 mL/minute/1.73 m2 are split into four groups and reflect worsening renal function, from mild impairment to renal failure. Around 6% of adults in the United Kingdom have an estimated glomerular filtration rate (eGFR) of <60 mL/minute/1.73 m2 [2]. This increases with age to around 20% of those aged ≥65. There is increasing evidence of an association between renal impairment and infection [3–6]. Oxidative stress in renal impairment disrupts the function of inflammatory cytokines and may impair immune response during an infection [7]. In more severe renal impairment, uremic toxins impair the function of T-lymphocyte and antigen-presenting cells, which play important roles in cellular and humoral immunity [8]. Despite the high prevalence in older adults, and the association with infection, few studies have investigated outcomes following an infectious illness in older people with renal impairment.
A cohort study in UK primary care showed that around 20% of adults aged 65 and over had at least one urinary tract infection (UTI) over a median follow-up of 5 years [9]. Most adults presenting to primary care with symptoms and signs of a UTI receive empirical antibiotics at the same consultation, without knowledge of microbiological findings or antibiotic susceptibilities [10]. Nitrofurantoin and trimethoprim (alone or with sulfamethoxazole) are the two most commonly prescribed antibiotics for empirical treatment of UTIs and are recommended by clinical guidelines in the UK, United States, and Europe [11,12]. Nitrofurantoin use was initially limited to those with an eGFR ≥60 mL/minute/1.73 m2, because of concerns about poorer efficacy in patients with lower eGFRs. In 2014, a review of the evidence [13] and a retrospective cohort study [14] prompted the UK Medicines and Healthcare products Regulation Authority to lower the threshold for nitrofurantoin use to an eGFR ≥45 mL/minute/1.73 m2. However, outcomes following empirical nitrofurantoin prescribing in older adults with a UTI and an eGFR <60 mL/minute/1.73 m2 are yet to be fully evaluated. There are also concerns about trimethoprim use in older adults with renal impairment, with increasing evidence of an association with hyperkalaemia and sudden death, especially when prescribed to patients already taking angiotensin-converting enzyme inhibitors, angiotensin-II receptor blockers, or potassium-sparing diuretics [15–19].
We used data from anonymised linked health records to estimate the risk of adverse outcomes in older patients with renal impairment empirically treated for suspected UTI in primary care. We firstly compared outcomes by eGFR to understand whether severity of renal impairment was associated with risk of adverse outcome following a UTI. This would help identify which patients would most benefit from interventions that improved prevention and/or treatment of UTI. We also compared outcomes for older patients with an eGFR <60 mL/minute/1.73 m2 who were prescribed empirical nitrofurantoin versus empirical trimethoprim to inform prescribing decisions and explore if nitrofurantoin prescribing is safe in patients with renal impairment.
We used the Clinical Practice Research Datalink (CPRD), an electronic database of anonymised primary care records covering 11.3 million patients from 674 general practices across the UK [20]. Approximately 7% of the UK population are included and patients are broadly representative of the wider UK population in terms of age, sex, and ethnicity. The CPRD holds data on demographics, clinical encounters, diagnoses (coded using Read codes), drug prescriptions, laboratory tests, and referrals to specialists. Data are available once they have met a series of quality checks on completeness and reliability and the CPRD deems them to be of the standard required for research purposes. Linked hospital and death registration data are available for patients from approximately 50% of contributing English practices. Hospital diagnoses and causes of death are recorded using the International Classification of Diseases-10th Revision (ICD-10).
The CPRD Independent Scientific Advisory Committee approved the study protocol and analysis plan (protocol number 17_250, S1 Protocol). Further ethical approval was not required as the proposed research was within the remit of the CPRD’s broad National Research Ethics Service approval. We used the Reporting of Studies Conducted using Observational Routinely-collected Health Data (RECORD) statement and checklist (S1 RECORD Checklist) to guide study reporting [21].
This was a retrospective cohort study using linked health record data. Patients were eligible for inclusion if, between 1 January 2010 and 31 December 2016, their data were of the quality required by CPRD, they were ≥65 years old, and eligible for data linkage. Only patients registered with practices that had consented to data linkage would have linked hospital and death registry data. We excluded patients if they were temporary residents (i.e., they registered with the practice for an acute problem but this was not their normal ‘long-term’ practice and thus medical record data would be limited) or had periods during their registration with the practice for which CPRD was unable to collect data, potentially leading to incomplete exposure/event capture. We identified eligible patients with a Read code indicating an incident primary care presentation with a suspected UTI (codes available in S1 Appendix), a prescription code indicating same-day empirical prescribing of a relevant antibiotic, and a creatinine record in the preceding 24 months. We defined ‘incident’ as a presentation without a previous consultation with a UTI-related Read code, or trimethoprim or nitrofurantoin prescription in the preceding 90 days. We used the first incident episode during each patient’s follow-up period. We excluded UTI episodes with a hospital discharge in the preceding 14 days to exclude hospital-acquired infections.
We used the most recent serum creatinine value recorded in the 24 months preceding the incident UTI and data for patient age, gender, and ethnicity to calculate an eGFR as per the Modification of Diet in Renal Disease (MDRD) Study equation [22]. We categorised eGFRs as ≥60 mL/minute/1.73 m2, 45–59 mL/minute/1.73 m2, 30–44 mL/minute/1.73 m2, 15–29 mL/minute/1.73 m2, and <15 mL/minute/1.73 m2. These categories are similar to those used by the UK National Institute of Health and Care Excellence to categorise the stages of chronic kidney disease, except we combined eGFRs of 60–89 mL/minute/1.73 m2 with those ≥90 mL/minute/1.73 m2. This was because eGFRs ≥60 mL/minute/1.73 m2 without additional evidence of kidney damage are clinically regarded as normal, and previous research found no difference in infection incidence or outcome between these two groups [5]. We used those with an eGFR ≥60 mL/minute/1.73 m2 as the reference and compared rates of adverse outcomes against the four other eGFR categories. We used the recorded empirical antibiotic prescription as the exposure variable to compare risk of adverse outcomes between patients with eGFRs <60 mL/minute/1.73 m2 prescribed trimethoprim versus nitrofurantoin.
We assessed the impact of our stated exposures on the following adverse outcomes for patients empirically treated in primary care for an incident suspected UTI:
We also initially planned to include hospitalisation for pyelonephritis as an outcome. However, our exploratory work showed that pyelonephritis was rarely coded in hospital records (only 8 events in total) and thus was unlikely to be a reliable outcome for use with these data.
We used primary care demographic and clinical codes to describe baseline characteristics for patients by exposure status. To assess the impact of eGFR, we compared rates of each outcome in those with an eGFR ≥60 mL/minute/1.73 m2 to those in each category related to an eGFR <60 mL/minute/1.73 m2, and used logistic regression to estimate odds ratios (ORs) and 95% confidence intervals (CIs). We adjusted for potential confounders of the association between renal impairment and outcome, including age; Index of Multiple Deprivation score quintile; Charlson comorbidity score [23]; the presence or absence of a record indicating diabetes, dementia, coronary heart disease, stroke, cancer, and heart failure; and polypharmacy. We inferred the presence of polypharmacy if the patient’s record showed repeated monthly prescribing of ≥5 medications in the year prior to the incident UTI. We also adjusted for the choice and duration of antibiotic therapy used to treat the incident UTI.
To assess the impact of empirical trimethoprim versus nitrofurantoin prescribing, we used a range of demographic and clinical variables to match patients on their propensity to receive a trimethoprim prescription. These included the confounders listed above and presence or absence of a record indicating urinary incontinence or long-term catheterisation, and long-term prescribing of angiotensin-converting enzyme inhibitors, angiotensin-II receptor blockers, or potassium-sparing diuretics. We used nearest neighbour matching with no replacement and matched each patient with a nitrofurantoin prescription to three patients with trimethoprim prescriptions. We assessed balance in measured baseline covariates between matched groups by visually inspecting jitter plots and histograms of covariate distribution before and after matching, and by calculating standardised mean differences for covariates between groups. We regarded standardised mean differences of <0.1 as reflecting adequate balance [24,25]. We used mixed effects logistic regression to account for clustering by general practice and calculated ORs and 95% CIs for each outcome.
All statistical tests were two-sided, with p < 0.05 considered statistically significant but an effect size of 10% considered clinically significant. p-values were derived using two-tailed Wald tests. Analyses were conducted in R version 3.2.1.
From a cohort of 795,484 patients aged 65 and over, we identified 123,607 with an incident UTI empirically treated with a relevant antibiotic (Fig 1). Of these, 116,945 (95%) patients had a creatinine measurement recorded in the 24 months prior to the incident UTI. In this final cohort, 32,428 (28%) were male and the median age at the time of incident UTI was 76 years (IQR 70–83). Almost one third of creatinine measurements were in the 90 days prior to the incident UTI. Median duration between most recent creatinine and UTI was 169 days (IQR 65–285). Using the MDRD study equation, 76,112 (65.1%) of patients were assigned an eGFR ≥60, 26,970 (23.1%) an eGFR of 45–59, 10,854 (9.3%) an eGFR of 30–44, 2,667 (2.3%) an eGFR of 15–29, and 342 (0.3%) an eGFR of <15. Baseline characteristics showed that patients with lower eGFRs had a relatively greater number of comorbidities and comprised greater proportions of patients with polypharmacy (Table 1). Trimethoprim was the most commonly prescribed empirical antibiotic across all eGFR groups. Nitrofurantoin was the second most common except in patients with an eGFR <15 mL/minute/1.73 m2.
There were 7,203 reconsultations with urinary symptoms resulting in another antibiotic prescription within 14 days of the incident UTIs, equating to about 6% of the cohort. The odds of reconsulting and receiving another antibiotic prescription were no different between those with an eGFR ≥60 mL/minute/1.73 m2 and those with eGFRs <60mL/minute/1.73 m2 (Table 2).
There were 1,991 hospitalisations for UTI (1.7% of the cohort), 176 for sepsis (0.2% of the cohort), and 865 for AKI (0.7% of the cohort) within 14 days of the incident UTIs. Compared to those with an eGFR ≥60 mL/minute/1.73 m2, odds of hospitalisation for UTI increased in those with eGFRs of 45–59 (adjusted OR 1.14, 95% CI 1.01–1.28), 30–44 (adjusted OR 1.25, 95% CI 1.08–1.44), 15–29 (adjusted OR 1.76, 95% CI 1.43–2.16), and <15 (adjusted OR 1.68, 95% CI 1.01–2.82). Odds of hospitalisation for sepsis were no different in those with an eGFR of 45–59 but were significantly higher in those with eGFRs of 30–44 (adjusted OR 1.70, 95% CI 1.06–2.72), 15–29 (adjusted OR 2.72, 95% CI 1.50–4.94), and <15 (adjusted OR 4.24, 95% CI 1.48–11.23). The risk of hospitalisation for AKI increased in a graded manner relative to renal function, with adjusted ORs of 1.57 (95% CI 1.29–1.91), 3.21 (95% CI 2.61–3.94), 6.70 (95% CI 5.24–8.55), and 4.53 (95% CI 2.52–8.17) for eGFRs of 45–59, 30–44, 15–29, and <15 mL/minute/1.73 m2, respectively.
There were 1,162 deaths in the 28 days following the incident UTIs, equating to about 1% of the cohort. Compared to those with an eGFR ≥60 mL/minute/1.73 m2, the odds of death were no different in those with an eGFR ≥30 mL/minute/1.73 m2, 63% higher in those with an eGFR of 15–29 (adjusted OR 1.63, 95% CI 1.27–2.10), and over 2-fold higher in those with an eGFR <15 (adjusted OR 2.37, 95% CI 1.44–3.89).
Of the 40,833 patients with an eGFR <60 mL/minute/1.73 m2, 24,471 (60%) were prescribed trimethoprim and 7,484 (18%) were prescribed nitrofurantoin. We matched 20,948 patients with an eGFR of 45–60 (15,711 prescribed trimethoprim, 5,237 prescribed nitrofurantoin), 7,260 with an eGFR of 30–44 (5,445 prescribed trimethoprim, 1,815 prescribed nitrofurantoin), and 1,728 with an eGFR <30 (1,296 prescribed trimethoprim, 432 prescribed nitrofurantoin). Inspection of jitter plots and histograms suggested matching had improved balance of covariates across trimethoprim versus nitrofurantoin groups. Standardised mean differences were all less than 0.1 (Table 3).
Empirical nitrofurantoin prescribing was associated with lower odds of hospitalisation for AKI across all eGFR groups (eGFR 45–59: OR 0.62, 95% CI 0.40–0.94; eGFR 30–44: OR 0.47, 95% CI 0.30–0.73; eGFR <30: OR 0.45, 95% CI 0.25–0.81) (Table 4). Nitrofurantoin was also associated with lower odds of reconsultation and re-prescription in patients with eGFRs of 45–59 (OR 0.74, 95% CI 0.61–0.91) and lower odds of death in patients with eGFRs of 30–44 (OR 0.61, 95% CI 0.39–0.95). There were no other statistically significant differences between empirical trimethoprim versus nitrofurantoin prescribing. Importantly, we did not detect any increase in odds of adverse outcomes in patients prescribed nitrofurantoin.
We restricted our eGFR and outcomes analysis to patients with a creatinine measured in the 90 days prior to the incident UTI, to increase the likelihood that the calculated eGFR reflected their current renal function. Results were consistent with our main analysis and most of the statistically significant ORs increased in magnitude (S1 Table). We also combined the hospitalisation and death outcomes in our trimethoprim versus nitrofurantoin analysis to increase statistical power to detect these adverse outcomes (S2 Table). Findings were consistent with our main analysis.
Our results show that compared to an eGFR of >60 mL/minute/1.73 m2, older patients with an eGFR of <60 mL/minute/1.73 m2 who were empirically treated for suspected UTI in primary care had greater odds of hospitalisation for UTI and AKI, those with an eGFR <45 had greater odds of hospitalisation for sepsis, and those with an eGFR <30 had greater odds of death. The magnitude of each association generally increased relative to the severity of the renal impairment. We also showed that, compared to trimethoprim, nitrofurantoin was associated with reduced odds of hospitalisation for AKI across all eGFR groups and was not associated with an increased risk of any adverse event evaluated in our study.
Previous research focussed on the risk of infection-related hospitalisation in adults with renal impairment and showed a greater risk of hospitalisation for pneumonia, UTI, bacteraemia, and cellulitis in those with eGFRs <60 mL/minute/1.73 m2 [4–6]. Previous studies also showed a greater risk of death following an infection-related hospitalisation in patients with renal impairment [3–5] but provided little information on health service contact prior to the adverse outcome, thus limiting interpretation about possible opportunities to intervene. Our study shows an increased risk of infection-related hospitalisation and death in older adults with renal impairment, following infection-related presentation and treatment in primary care. We also show increased odds of AKI hospitalisation in those with lower initial eGFRs, previously only investigated in a small cohort (n = 790) of patients hospitalised with UTI, who likely had more severe infection on initial presentation [26]. To the best of our knowledge, this is the first study to investigate the impact of eGFR on odds of reconsultation and re-prescription following an infection-related illness. We found no difference in the odds of this outcome across the different eGFR groups, suggesting that the increased odds of UTI, sepsis, and death were less likely to be due to treatment nonresponse and more likely to be related to other patient or renal factors.
Trimethoprim (with or without sulfamethoxazole) prescribing is associated with an increased risk of hyperkalaemia, AKI, and death, compared to amoxicillin [16–19]. Amoxicillin accounts for only about 5% of prescribing for UTI in the UK [9] and thus is a less helpful comparator for clinical decision-making. Furthermore, these studies did not investigate associations by degree of renal impairment, providing little information to guide prescribing in this population. Two studies assessed trimethoprim and nitrofurantoin prescribing in patients with renal impairment. The first compared treatment failure rates in women with UTI prescribed nitrofurantoin according to renal function and found no difference across the eGFR groups [14]. This study lacked a comparator group prescribed an alternative antibiotic, which makes it difficult to interpret their findings. The second compared outcomes in older women with a median eGFR of 38 mL/minute/1.73 m2, prescribed either nitrofurantoin or trimethoprim, and found no difference in risk of treatment failure or UTI hospitalisation [27]. We compared nitrofurantoin with trimethoprim across three eGFR groups and found that nitrofurantoin was associated with lower odds of reconsultation and re-prescription in patients with eGFRs of 45–59. This could be explained by recent data showing that 34% of community-acquired Escherichia coli UTIs in England are resistant to trimethoprim, compared to only 2.7% resistant to nitrofurantoin [28]. We did not find statistically significant differences between reconsultation and re-prescription rates in people with eGFRs <45. This could be due to less statistical power, as nitrofurantoin use was less common in these patients because of the advice to use with care in patients with eGFRs of 30–44 and to avoid in eGFRs <30. It may also be due to the possibility that nitrofurantoin efficacy was reduced in those with lower eGFRs but was offset by the high rates of trimethoprim resistance and thus resulted in apparent similar rates of reconsultation and re-prescription. Our finding that nitrofurantoin was associated with a reduced risk of death in those with moderate renal impairment is consistent with previously reported estimates in studies that compared nitrofurantoin with amoxicillin in the general population [18,19]. We also found a previously unreported lower risk of AKI associated with nitrofurantoin use across all three eGFR groups of our cohort.
We used data from a general practice database that is broadly representative of the UK population, increasing the generalisability of our findings. This is the largest cohort study to investigate the impact of eGFR on infection-related outcomes, with a sample size >4 times larger than the previously largest study [4]. Cohort entry was dependent on presentation and empirical treatment of UTI in primary care, and thus reduced indication bias. We adjusted for the presence/absence of more comorbidities than previous studies, increasing the likelihood of an independent association between eGFR and adverse outcomes. This is the first study to investigate trimethoprim versus nitrofurantoin prescribing in renal impairment, using clinically relevant eGFR groups analogous to stages of chronic kidney disease and without excluding men. We also reduced indication bias by matching patients on their propensity to receive trimethoprim, and achieving adequate balance of covariates across the two groups.
Our study has important limitations. We attempted to capture patients presenting with UTI but had no microbiological data to support this. However, whilst a limitation, this is also more representative of clinical practice. We were unable to investigate pulmonary/hepatic toxicity related to nitrofurantoin use because of the lack of reliable codes, and differential use of these codes by clinicians. However, two systematic reviews have shown that these toxicities are rare with short-term use [29,30]. We relied on a creatinine measurement from the 24 months prior to the UTI to estimate an eGFR, but this may not fully represent patients’ current renal function. Finally, despite our design, differential coding, indication bias, and residual confounding may still have affected our findings.
Around 20% of adults aged ≥65 present to primary care at least once with a UTI [9], and around 20% have renal impairment [2] and thus are at greater risk of an adverse outcome. The initial primary care visit presents a potential opportunity to address this increased risk. However, recommended interventions such as stopping co-prescribed angiotensin-converting enzyme inhibitors or angiotensin-II receptor blockers have not been evaluated in the primary care setting [31]. Therefore, there is a need for research that evaluates primary care–based interventions that may prevent adverse outcomes, including AKI, in patients with renal impairment and community-acquired infection. The absolute risks of hospitalisation and death are low. Therefore, interventions may be best targeted at those at highest risk—i.e., those with more severe renal impairment.
Current guidelines and the British National Formulary limit nitrofurantoin use to those with an eGFR >45 mL/minute/1.73 m2, although short courses can be used with care in those with eGFRs >30 mL/minute/1.73 m2 [11]. We found no evidence to support this limitation and actually found nitrofurantoin to be associated with a reduced risk of AKI, compared to trimethoprim. Research suggesting poorer nitrofurantoin efficacy in patients with renal impairment assessed urinary nitrofurantoin excretion, not clinical outcomes, and was restricted to small samples [13]. Our findings, combined with increasing rates of bacterial resistance to trimethoprim and the importance of avoiding broader spectrum agents, support wider use of nitrofurantoin for older patients with low eGFRs.
Our findings show that older patients with renal impairment presenting to primary care with a UTI are at greater risk of adverse outcomes independent of other comorbidities and of prescribed empirical antibiotic treatment. Despite documented concerns, we found no increased risk of adverse outcomes in patients with an eGFR <60 mL/minute/1.73 m2 prescribed nitrofurantoin and support its wider use in selected patients with moderate-severe renal impairment.
Disclaimer: The views expressed in this publication are those of the authors and not necessarily those of the NIHR, NHS Wales, HCRW, or the Welsh Government.
|
10.1371/journal.pcbi.1005860 | Fragmentation modes and the evolution of life cycles | Reproduction is a defining feature of living systems. To reproduce, aggregates of biological units (e.g., multicellular organisms or colonial bacteria) must fragment into smaller parts. Fragmentation modes in nature range from binary fission in bacteria to collective-level fragmentation and the production of unicellular propagules in multicellular organisms. Despite this apparent ubiquity, the adaptive significance of fragmentation modes has received little attention. Here, we develop a model in which groups arise from the division of single cells that do not separate but stay together until the moment of group fragmentation. We allow for all possible fragmentation patterns and calculate the population growth rate of each associated life cycle. Fragmentation modes that maximise growth rate comprise a restrictive set of patterns that include production of unicellular propagules and division into two similar size groups. Life cycles marked by single-cell bottlenecks maximise population growth rate under a wide range of conditions. This surprising result offers a new evolutionary explanation for the widespread occurrence of this mode of reproduction. All in all, our model provides a framework for exploring the adaptive significance of fragmentation modes and their associated life cycles.
| Mode of reproduction is a defining trait of all organisms, including colonial bacteria and multicellular organisms. To produce offspring, aggregates must fragment by splitting into two or more groups. The particular way that a given group fragments defines the life cycle of the organism. For instance, insect colonies can reproduce by splitting or by producing individuals that found new colonies. Similarly, some colonial bacteria propagate by fission or by releasing single cells, while others split in highly sophisticated ways; in multicellular organisms reproduction typically proceeds via a single-cell bottleneck phase. The space of possibilities for fragmentation is so vast that an exhaustive analysis seems daunting. Focusing on fragmentation modes of a simple kind we parametrise all possible modes of group fragmentation and identify those modes leading to the fastest population growth rate. Two kinds of life cycle dominate: one involving division into two equal size groups, and the other involving production of a unicellular propagule. The prevalence of these life cycles in nature is consistent with our null model and suggests that benefits accruing from population growth rate alone may have shaped the evolution of fragmentation mode.
| A requirement for evolution—and a defining feature of life—is reproduction [1–4]. Perhaps the simplest mode of reproduction is binary fission in unicellular bacteria, whereby a single cell divides and produces two offspring cells. In more complex organisms, such as colonial bacteria, reproduction involves fragmentation of a group of cells into smaller groups. Bacterial species demonstrate a wide range of fragmentation modes, differing both in the size at which the parental group fragments and the number and sizes of offspring groups [5]. For example, in the bacterium Neisseria, a diplococcus, two daughter cells remain attached forming a two-celled group that separates into two groups of two cells only after a further round of cell division [6]. Staphylococcus aureus, another coccoid bacterium, divides in three planes at right angles to one another to produce grape-like clusters of about 20 cells from which single cells separate to form new clusters [7]. Magnetotactic prokaryotes form spherical clusters of about 20 cells, which divide by splitting into two equally sized clusters [8].
These are just a few examples of a large number of diverse fragmentation modes, but why should there be such a wide range of life cycles? Do fragmentation modes have adaptive significance or are they simply the unintended consequences of particular cellular processes underpinning cell division? If adaptive, what selective forces shape their evolution? Can different life cycles simply provide different opportunities to maximise population growth rate?
A starting point to answer these questions is to consider benefits and costs of group living in cell collectives. Benefits may arise for various reasons. Cells within groups may be better able to withstand environmental stress [9], escape predation [10, 11], or occupy new niches [12, 13]. Also, via density-dependent gene regulation, cells within groups may gain more of a limiting resource than they would if alone [14, 15]. On the other hand, cells within groups experience increased competition and must also contend with the build up of potentially toxic waste metabolites [16, 17]. Thus, it is reasonable to expect an optimal relationship between group size and fragmentation mode that is environment and organism dependent [18–21].
Here we formulate and study a matrix population model [22] that considers all possible modes of group fragmentation. By determining the relationship between life cycle and population growth rate, we show that there is, overall, a narrow class of optimal modes of fragmentation. When the process of fragmentation does not involve costs, optimal fragmentation modes are characterised by a deterministic schedule and binary splitting, whereby groups fragment into exactly two offspring groups. Contrastingly, when a cost is associated with fragmentation, it can be optimal for a group to fragment into multiple propagules.
Our results show that the range of life cycles observed in simple microbial populations are likely shaped by selection for intrinsic growth rate advantages inherent to different modes of group fragmentation. While we do not consider complex life cycles, our results may contribute to understanding the emergence of life cycles underpinning the evolution of multicellular life.
We consider a population in which a single type of cell (or unit or individual) can form groups (or complexes or aggregates) of increasing size by cells staying together after reproduction [18]. We assume that the size of any group is smaller than n, and denote groups of size i by Xi (see the list of used variables in Table 1). Groups die at rate di and cells within groups divide at rate bi; hence groups grow at rate ibi. The vectors of birth rates b = (b1, …, bn−1) and of death rates d = (d1, …, bn−1) make the costs and benefits associated to the size of the groups explicit, thus defining the “fitness landscape” of our model.
Groups produce new complexes by fragmenting (or splitting), i.e., by dividing into smaller groups. We further assume that fragmentation is triggered by growth of individual cells within a given group. Consider a group of size i growing into a group of size i + 1. Such a group can either stay together or fragment. If it fragments, it can do so in one of several ways. For example, a group of size 4 can give rise to the following five “fragmentation patterns”: 4 (the group does not split, but stays together), 3+1 (the group splits into two offspring groups: one of size 3, and one of size 1), 2+2 (the group splits into two groups of size 2), 2+1+1 (the group splits into one group of size 2 and two groups of size 1), and 1+1+1+1 (the group splits into four independent cells). Mathematically, such fragmentation patterns correspond to the five partitions of 4 (a partition of a positive integer i is a way of writing i as a sum of positive integers without regard to order; the summands are called parts [23]). We use the notation κ ⊢ ℓ to indicate that κ is a partition of ℓ, for example 2 + 2 ⊢ 4. The number of partitions of ℓ is given by ζℓ, e.g., there are ζ4 = 5 partitions of 4.
We consider an exhaustive set of fragmentation modes (or “fragmentation strategies”) implementing all possible ways groups of maximum size n can grow and fragment into smaller groups, including both pure and mixed modes (Fig 1). A pure fragmentation mode is characterised by a single partition κ ⊢ ℓ, i.e., groups of size i < ℓ grow up to size ℓ and then fragment according to partition κ ⊢ ℓ. The partition κ can then be used to refer to the associated pure strategy. The total number of pure fragmentation strategies is ∑ ℓ = 2 n ( ζ ℓ − 1 ), which grows quickly with n: There are 128 pure fragmentation modes for n = 10, but 1,295,920 for n = 50. A mixed fragmentation mode is given by a probability distribution over the set of pure fragmentation modes. The relationship between pure and mixed fragmentation modes is hence similar to the one between pure strategies and mixed strategies in evolutionary game theory [24]. One of our main results is that mixed fragmentation modes are always dominated by pure fragmentation modes. Hence, we focus our exposition on pure fragmentation modes, and leave the details of how to specify mixed fragmentation modes to the Supporting Information (S1 Text, Appendix A).
Together with the fitness landscape given by the vectors of birth rates b and death rates d, each fragmentation strategy specifies a set of biological reactions. Consider the pure mode κ ⊢ ℓ, whereby groups grow up to size ℓ and then split according to fragmentation pattern κ. A set of reactions
X i → d i 0 , i = 1 , … , ℓ - 1 (1)
models the death of groups; an additional set of reactions
X i → i b i X i + 1 , i = 1 , … , ℓ - 2 (2)
models the growth of groups (without splitting) up to size ℓ − 1. Finally, one reaction of the type
X ℓ - 1 → ( ℓ - 1 ) b ℓ - 1 ∑ i = 1 ℓ - 1 π i ( κ ) X i , (3)
models the growth of the group from size ℓ − 1 to size ℓ and its immediate fragmentation in a way described by fragmentation pattern κ ⊢ ℓ, where parts equal to i appear a number πi(κ) of times. For instance, for the pure fragmentation mode 2 + 1 + 1 ⊢ 4, Eq (3) becomes
X 3 → 3 b 3 X 2 + 2 X 1 ,
which stipulates that groups of size 3 grow to size 4 at rate 3b3 and split into one group of size 2 and two groups of size 1; here, π1(2 + 1 + 1) = 2, π2(2 + 1 + 1) = 1, π3(2 + 1 + 1) = 0.
The sets of reactions (1), (2) and (3) give rise to the system of differential equations
x ˙ 1 = − ( b 1 + d 1 ) x 1 + ( l − 1 ) b l − 1 π 1 ( κ ) x l − 1 , x ˙ i = ( i − 1 ) b i − 1 x i − 1 − ( i b i + d i ) x i + ( l − 1 ) b l − 1 π i ( κ ) x l − 1 , i = 2 , … , l − 1
where xi denotes the abundance of groups of size i. This is a linear system that can be represented in matrix form as
x ˙ = A x , (4)
where x = (x1, x2, …, xℓ−1) is the vector of abundances of the groups of different size and
A = ( − b 1 − d 1 0 ⋯ 0 ( l − 1 ) b l − 1 π 1 ( κ ) b 1 − 2 b 2 − d 2 0 ⋮ ( l − 1 ) b l − 1 π 2 ( κ ) 0 2 b 2 − 3 b 3 − d 3 0 ( l − 1 ) b l − 1 π 3 ( κ ) ⋮ ⋮ ⋱ ⋱ ⋮ 0 0 ⋯ ( l − 2 ) b l − 2 ( l − 1 ) b l − 1 ( π l − 1 ( κ ) − 1 ) − d l − 1 )
is the projection matrix determining the population dynamics.
For any fragmentation mode and any fitness landscape, the projection matrix A is “essentially non-negative” (or quasi-positive), i.e., all the elements outside the main diagonal are non-negative [25]. This implies that A has a real leading eigenvalue λ1 with associated non-negative left and right eigenvectors v and w. In the long term, the solution of Eq (4) converges to that of an exponentially growing population with a stable distribution, i.e.,
lim t → ∞ x ( t ) = e λ 1 t w .
The leading eigenvalue λ1 hence gives the total population growth rate in the long term, and its associated right eigenvector w = (w1, …, wm−1) gives the stable distribution of group sizes so that, in the long term, the fraction xi of complexes of size i in the population is proportional to wi.
For a given fitness landscape {b, d}, we can take the leading eigenvalue λ1(κ; b, d) as a measure of fitness of fragmentation mode κ, and consider the competition between two different fragmentation modes, κ1 and κ2. Indeed, under the assumption of no density limitation, the evolutionary dynamics are described by two uncoupled sets of differential equations of the form (4): one set for κ1 and one set for κ2. In the long term, κ1 is not outcompeted by κ2 if λ1(κ1; b, d) ≥ λ1(κ2; b, d); we then say that fragmentation mode κ1 dominates fragmentation mode κ2. We also say that strategy κi is optimal for given birth rates b and death rates d if it achieves the largest growth rate among all possible fragmentation modes.
Fitness landscapes capture the many advantages or disadvantages associated with group living. These advantages may come either in the form of additional resources available to groups depending on their size or as an improved protection from external hazards. For our numerical examples, we consider two classes of fitness landscape, each representing only one of these factors. In the first class, that we call “fecundity landscapes”, group size affects only the birth rates of cells (while we impose di = 0 for all i). In the second class, that we call “survival landscapes”, group size affects only death rates (and we impose bi = 1 for all i).
To fix ideas, consider all pure fragmentation modes with a maximum group size n = 3. These are 1+1 (“binary fission”, a partition of 2), 2+1 (“unicellular propagule”, a partition of 3), and 1+1+1 (“ternary fission” a partition of 3). The three associated projection matrices are given by
A 1 + 1 = ( b 1 - d 1 ) , A 2 + 1 = ( - b 1 - d 1 2 b 2 b 1 - d 2 ) , A 1 + 1 + 1 = ( - b 1 - d 1 6 b 2 b 1 - 2 b 2 - d 2 ) .
The three growth rates are
λ 1 1 + 1= b 1 - d 1 , (5a)
λ 1 2 + 1 = − ( b 1 + d 1 + d 2 ) + ( b 1 + d 1 − d 2 ) 2 + 8 b 1 b 2 2 , (5b)
λ 1 1 + 1 + 1 = - ( b 1 + 2 b 2 + d 1 + d 2 ) + b 1 2 + 2 b 1 ( 10 b 2 + d 1 - d 2 ) + ( 2 b 2 - d 1 + d 2 ) 2 2 . (5c)
In the particular case of a fecundity landscape given by b1 = 1 and b2 = 15/8 (and d1 = d2 = 0), these growth rates reduce to λ 1 1 + 1 = 1, λ 1 2 + 1 = 3 / 2 and λ 1 1 + 1 + 1 = 5 / 4, and we have λ 1 2 + 1 > λ 1 1 + 1 + 1 > λ 1 1 + 1. We then say that ternary and binary fission are dominated by the unicellular propagule strategy.
Although for simplicity we focus our exposition on pure fragmentation strategies, we also consider mixed fragmentation strategies, i.e., probabilistic strategies mixing between different pure modes. A natural question to ask is whether a mixed fragmentation mode can achieve a faster growth rate than a pure mode. We find that the answer is no. For any fitness landscape and any maximum group size n, mixed fragmentation modes are dominated by a pure fragmentation mode (S1 Text, Appendix B). Thus, the optimal fragmentation mode for any fitness landscape is pure.
As an example, consider fragmentation modes 1+1 and 2+1, and a mixed fragmentation mode mixing between these two so that with probability q splitting follows fragmentation pattern 2+1 and with probability 1 − q it follows fragmentation pattern 1+1. For any mixing probability q and any fitness landscape, the growth rate of the mixed fragmentation mode is given by
λ 1 q = b 1 ( 1 - 2 q ) - ( d 1 + d 2 ) + ( d 1 + d 2 - ( 1 - 2 q ) b 1 ) 2 + 4 b 1 ( 2 q b 2 + ( 1 - 2 q ) d 2 ) 2 ,
which can be shown to always lie between the growth rates of the pure fragmentation modes, i.e., either λ 1 1 + 1 ≤ λ 1 q ≤ λ 1 2 + 1 or λ 1 2 + 1 ≤ λ 1 q ≤ λ 1 1 + 1 holds and the mixed fragmentation mode is dominated (S1 Text, Appendix C).
To further illustrate our analytical findings, consider groups of maximum size n = 4 and a fecundity landscape given by b = (1, 2, 1.4). We randomly generated 107 mixed fragmentation modes by drawing the probabilities for growth without splitting from an uniform distribution and letting the probabilities of splitting according to a given fragmentation pattern be proportional to exponential random variables with rate parameter equal to one. We then calculated the growth rate of these mixed strategies together with the growth rate of the seven pure fragmentation modes available for n = 4, i.e., 1+1, 2+1, 1+1+1, 3+1, 2+2, 2+1+1, and 1+1+1+1 (Fig 2A). In line with our analysis, a pure fragmentation mode (namely 2+2, whereby groups grow up to size 4 and then immediately split into two bicellular groups) achieves a higher growth rate than the growth rate of any mixed fragmentation mode, and the highest growth rate overall.
Having shown that mixed fragmentation modes are dominated, we now ask which pure modes might be optimal. We find that, within the set of pure modes, “binary” fragmentation modes (whereby groups split into exactly two offspring groups) dominate “nonbinary” fragmentation modes (whereby groups split into more than two offspring groups). To illustrate this result, consider the simplest case of n = 3 and the three modes 1+1, 2+1, and 1+1+1, out of which 1+1 and 2+1 are binary, and 1+1+1 is nonbinary. Comparing their growth rates (as given in Eq (5), we find that λ 1 1 + 1 ≥ λ 1 1 + 1 + 1 holds if b1 − b2 ≥ d1 − d2 and that λ 1 2 + 1 ≥ λ 1 1 + 1 + 1 holds if b1 − b2 ≤ d1 − d2. Thus, for any fitness landscape, 1+1+1 is dominated by either 2+1 or by 1+1. More generally, we can show that for any nonbinary fragmentation mode, one can always find a binary fragmentation mode achieving a greater or equal growth rate under any maximum group size n and fitness landscape (S1 Text, Appendices D and E). Taken together, our analytical results imply that the set of optimal fragmentation modes is countable and, even for large n, relatively small. Consider the proportion of pure fragmentation modes that can be optimal, which is defined by the ratio between the number of binary fragmentation modes and the total number of pure fragmentation modes. While this ratio is relatively high for small n (e.g., 2/3 ≈ 0.67 for n = 3 or 4/7 ≈ 0.57 for n = 4), it decreases sharply with increasing n (e.g., 25/128 ≈ 0.20 for n = 10 and 625/1295920 ≈ 0.00048 for n = 50).
Fig 2B shows the growth rate of the seven pure modes for n = 4 for a fecundity landscape given by b = (1, b2, 1.4) as a function of b2. In line with our analysis, only binary fragmentation modes (1+1, 2+1, 2+2, and 3+1) can be optimal, while nonbinary fragmentation modes (1+1+1, 2+1+1, and 1+1+1+1) are dominated. Which particular binary mode is optimal depends on the particular value of the birth rate of groups of two cells. For small values (b2 ≲ 0.45), the fecundity of such groups is too low, and the optimal fragmentation mode is 1+1. For intermediary values (0.45 ≲ b2 ≲ 1.11), the reproduction efficiency of groups of three cells mitigates the inefficiency of cell pairs, and the mode 3+1 becomes optimal. For larger values (1.11 ≲ b2 ≲ 3.52), the optimal fragmentation mode is 2+2, where no single cells are produced. Finally, for very large values (b2 ≲ 3.52), the optimal fragmentation mode is 2+1; this ensures that one offspring group emerges at the most productive bicellular state.
More generally, which particular fragmentation mode within the class of binary splitting strategies is optimal depends on all birth rates and death rates characterising the fitness landscape. To further explore this issue, we identified the optimal fragmentation modes for general fecundity and survival landscapes for the simple case of n = 4 (Fig 3; S1 Text, Appendix F). Since we can set b1 = 1 and min(d) = 0 without loss of generality (S1 Text, Appendix D), we represent fitness landscapes as points in a two-dimensional parameter space with coordinates b2/b1 and b3/b1 for fecundity landscapes, and coordinates d2 − d1 and d3 − d1 for survival landscapes. The exact boundaries of the parameter regions where a given fragmentation mode is optimal are often nontrivial mathematical expressions. Nevertheless, we identify general patterns dictating which fragmentation mode will be optimal. Consider first the optimality map for fecundity landscapes (Fig 3A). A sufficient condition for the unicellular life cycle 1+1 to be optimal is that the birth rate of single cells is larger than the birth rate of pairs and triplets of cells (b1 > b2 and b1 > b3). In this case, there is no apparent reason why a fragmentation mode different than 1+1 would be optimal. Perhaps less trivially, 1+1 can also be optimal in cases where single cells are less fertile than groups of three cells, i.e., even if b1 < b3 holds. This requires the birth rate b2 to be so small that the fecundity benefits accrued when reaching the size of three cells are not enough to compensate for the unavoidable penalty of passing through the less prolific state of two cells. Turning now to fragmentation mode 2+1, a necessary condition for this mode to be optimal is that pairs of cells have the largest birth rate, i.e., that b2 > b1 and b2 > b3 holds. Similarly, mode 3+1 can only be optimal if b3 > b1 and b3 > b2, so that groups of three have the largest birth rate. In these two cases, the optimal fragmentation mode (either 2+1 or 3+1) keeps one of the two offspring groups at the most productive size. Finally, for fragmentation mode 2+2 to be optimal, it is necessary that single cells have the lowest birth rate, i.e., that b2 > b1 and b3 > b1 holds. In this case, the fragmentation mode ensures that the life cycle of the organism never goes through the least productive unicellular phase. Under survival landscapes, fitness increases as death rates decrease. Taking this qualitative difference into account, the map of optimal fragmentation modes under survival landscapes (Fig 3B) follows similar qualitative patterns as the one under fecundity landscapes.
So far we have assumed that fragmentation is costless. However, fragmentation processes can be costly to the parental group undergoing division. This is particularly apparent in cases where some cells need to die in order for fragmentation of the group to take place. Examples in simple multicellular forms include Volvox, where somatic cells constituting the outer layer of the group die upon releasing the offspring colonies and are not passed to the next generation [26], the breaking of filaments in colonial cyanobacteria [27], and the fragmentation of “snowflake-like” clusters of the yeast Saccharomyces cerevisiae [28]. Fragmentation costs may also be less apparent. For instance, fragmentation may cost resources that would otherwise be available for the growth of cells within a group.
To investigate the effect of fragmentation costs on the set of optimal fragmentation modes, we consider two cases: proportional costs and fixed costs. For proportional costs, we assume that π − 1 cells die in the process of a group fragmenting into π parts. This case captures the fragmentation process of filamentous bacteria, where filament breakage entails the death of cells connecting the newly formed fragments [27]. For fixed costs, we assume that exactly one cell is lost upon each fragmentation event. This scenario is loosely inspired by yeast colonies with a tree-like structure, where cells can be connected with many other cells, so the death of a single cell may release more than two offspring colonies [19, 28]. Mathematically, both cases imply that fragmentation patterns are described by partitions of a number smaller than the size of the parent group (S1 Text, Appendix G).
For both kinds of costly fragmentation, we can show that mixed fragmentation modes are still dominated by pure fragmentation modes (the proof given in S1 Text, Appendix B also holds in this case). Moreover, for proportional costs the optimal fragmentation mode is also characterised by binary fragmentation, as it is the case for costless fragmentation (S1 Text, Appendix H). This makes intuitive sense, as the addition of a penalty for splitting into many fragments should further reinforce the optimality of binary splitting (whereby only one cell per fragmentation event is lost). In contrast, we find that under fragmentation with fixed costs the optimal fragmentation mode can involve nonbinary fragmentation, i.e., division into more than two offspring groups. This result can be readily illustrated for the case of n = 4 where the nonbinary mode 1+1+1 is optimal for a wide range of fitness landscapes (Fig 4).
Another interesting feature of costly fragmentation (implemented via either proportional or fixed costs) is that fragmentation modes involving the emergence of large groups can be optimal even if being in a group does not grant any fecundity or survival advantage to cells. If fragmentation is costless, as we assumed before, fitness landscapes for which groups perform worse than unicells (that is, bi/b1 ≤ 1 for fecundity landscapes or di − d1 ≥ 0 for survival landscapes) lead to optimal fragmentation modes where splitting occurs at the minimum possible group size i = 2, so that no multicellular groups emerge in the population (cf. Fig 3). In contrast, under costly fragmentation some of these fitness landscapes allow for the evolutionary optimality of fragmentation modes according to which groups split at the maximum size n = 4 (2+1 under proportional costs, and 1+1+1 under fixed costs), and hence for life cycles where multicellular phases are persistent. This seems paradoxical until one realises that by staying together as long as possible groups delay as much as possible the inevitable cell loss associated to a fragmentation event. Thus, even if groups are less fecund or die at a higher rate than independent cells, staying together might be adaptive if splitting apart is too costly.
Next, we focus on fitness landscapes for which either the birth rate of cells increases with group size (fecundity landscapes where larger groups are always more productive) or the death rate of groups decreases with group size (survival landscapes where larger groups always live longer). In this case, and for a maximum group size n = 4, the set of optimal modes is given by 2+2 and 3+1 if there are no fragmentation costs (Fig 3), by 2+1 if fragmentation costs are proportional to the number of fragments (Fig 3A and 3B), and by 2+1 and 1+1+1 if fragmentation involves a fixed cost of one cell (Fig 4C and 4D).
To investigate larger maximum group sizes n in a simple but systematic way, we consider fecundity landscapes with birth rates given by bi = 1 + Mgi and survival landscapes with death rates given by di = M(1 − gi), where gi = [(i − 1)/(n − 2)]α [29] models the fecundity or survival benefits associated to group size i and M > 0 is the maximum benefit (Fig 5). The parameter α is the degree of complementarity between cells; it measures how important the addition of another cell to the group is in producing the maximum possible benefit M [30]. For low degrees of complementarity (α < 1), the sequence gi is strictly concave and each additional cell contributes less to the per capita benefit of group living [31] and groups of all sizes achieve the same functionality as α tends to zero. If α = 1, the sequence gi is linear, and each additional cell contributes equally to the fecundity or survival of the group. Finally, for high degrees of complementarity (α > 1), the sequence gi is strictly convex and each additional cell improves the performance of the group more than the previous cell did. In the limit of large α, the advantages of group living materialise only when complexes achieve the maximum size n − 1 [31].
We numerically calculate the optimal fragmentation modes for n = 20 (costless fragmentation) or n = 21 (costly fragmentation) and the fitness landscapes described above for parameter values taken from 0.01 ≤ α ≤ 100 and 0.02 ≤ M ≤ 50 (Figs 6 and 7). In line with our general analytical results, optimal fragmentation modes are always characterised by binary splitting when fragmentation is costless or when it involves proportional costs, while nonbinary splitting can be optimal only if fragmentation involves a fixed cost. We also find that, for each value of α and M, and for both costless and costly fragmentation, the optimal fragmentation mode is one where fragmentation occurs at the largest possible size. This is expected since the benefit sequence is increasing in group size and thus groups of maximum size perform better, either by achieving the largest birth rate per unit (fecundity landscapes) or the lowest death rate (survival landscapes). Which particular fragmentation mode maximizes the growth rate depends nontrivially on whether fragmentation is costless or costly (and in the latter case also on how such costs are implemented), on the kind of group size benefits (fecundity or survival), on the maximum possible benefit M, and on the degree of complementarity α. Despite this apparent complexity, some general patterns can be identified.
Let us focus on the case of fecundity landscapes and first fasten attention on the scenario of costless fragmentation (Fig 6A). A salient feature of this case is the prominence of two qualitatively different fragmentation modes: the “equal binary fragmentation” strategy 10+10 (whereby offspring groups have sizes as similar as possible) and the “unicellular propagule” strategy 19+1 (whereby offspring groups have sizes as different as possible). A sufficient condition for equal binary fragmentation to be optimal is that increase in size is characterised by diminishing returns. The intuition behind this result is that, if the degree of complementarity is small, then groups (complexes of size i ≥ 2) have similar performance, while independent cells (i = 1) are at a significant disadvantage. Therefore, the optimal strategy is to ensure that both offspring groups are as large as possible, and hence of the same size. However, equal binary fragmentation can be also optimal for synergistic interactions, provided that complementarity is not too high. In contrast, the unicellular propagule strategy is optimal only for relatively high degrees of complementarity. This is because when complementarity is high only the largest group can reap the benefits of group living; in this case, the optimal mode is to have at least one offspring of very large size. Compared to 19+1 and 10+10, other binary splitting strategies are optimal in smaller regions of the parameter space, and in all cases only for synergetic interactions between cells.
Consider now the effects of introducing fragmentation costs proportional to the number of fragments (Fig 6B). Here, the region where the unicellular propagule strategy is optimal shrinks to the corner of the parameter space where benefits of group living and degree of complementarity are maximum, while the region of optimality for equal binary fragmentation expands. This makes intuitive sense. With fragmentation costs, the largest offspring group resulting from fragmenting according to the unicellular propagule strategy is of size 19, and hence always on the brink of fragmentation (once it grows to size 21) and incurring one cell loss. When group benefits are not high and synergistic enough, the unicellular propagule strategy is dominated by fragmentation modes (in particular, equal binary fragmentation) having smaller offspring for which the costs of fragmentation are not so immediate.
Finally, if costs of fragmentation are not proportional but fixed (Fig 6C), then two classes of nonbinary splitting become optimal in regions of the parameter space where equal binary fragmentation was optimal under proportional costs: (i) “multiple fission” (1+1+…+1) which is in general favored for small maximum benefit and increasing returns, and (ii) “multiple groups” (modes 2+2+…+2, 3+3+3+3+3+3+2, 4+4+3+3+3+3, 4+4+4+4+4, 5+5+5+5, and 7+7+6) which are often optimal for diminishing returns.
Fig 7 show the results for survival landscapes. The main difference in this case is that the unicellular propagule strategy can be the optimal strategy even when group living is characterised by diminishing returns. In general, fecundity benefits make equal binary fragmentation optimal under more demographic scenarios, while survival benefits make the unicellular propagule strategy optimal under more demographic scenarios.
Reproduction is such a fundamental feature of living systems that the idea that the mode of reproduction may be shaped by natural selection is easily overlooked. Here, we analysed a matrix population model that captures the demographic dynamics of complexes that grow by staying together and reproduce by fragmentation. The costs and benefits associated with group size ultimately determine whether or not a single cell fragments into two separate daughter cells upon cell division, or whether those daughter cells remain in close proximity, with fragmentation occurring only after subsequent rounds of division. We allowed for a vast and complete space of fragmentation strategies, including pure modes (specifying all possible combinations of size at fragmentation and fragmentation pattern) and mixed modes (specifying all probability distributions over the set of pure modes), and identified those modes achieving a maximum growth rate for given fecundity and survival size-dependent rates. Our research questions and methodology thus resonates with previous studies in life history theory [32, 33]. In the language of this field, our fragmentation strategies specify both the size at first reproduction and clutch size, where the latter is subject to a very specific trade-off between the number and size of offspring mathematically given by integer partitions.
We found that for any fitness landscape, the optimal life cycle is always a deterministic fragmentation mode involving the regular schedule of group development and fragmentation. This makes intuitive sense given our assumption that the environment is constant. However, this result might not hold if the environment is variable so that the fitness landscape changes over time. In this case different pure fragmentation modes will be optimal at different times, and natural selection might favour life cycles that randomly express a subset of locally optimal fragmentation patterns. Indeed, the evolution of variable phenotypes in response to changing environmental conditions (also known as bet hedging [34, 35]) has been demonstrated in other life history traits, such as germination time in annual plants [36], and capsulation in bacteria [37]. The extent to which mixed fragmentation modes can evolve via a similar mechanism is beyond the scope of this paper, but it can be addressed in future work by applying existing theory on matrix population models in stochastic environments [22].
We found that when fragmentation is costless, only strategies involving binary splitting (i.e., fragmentation into exactly two parts) are optimal. This result holds for all possible fitness landscapes, and hence any specification of how fecundity or survival benefits might accrue to group living. In particular, the optimal fragmentation mode under monotonic fitness landscapes is generally one of two types: equal binary fragmentation, which involves fission into two equal size groups, and the unicellular propagule strategy, which involves the production of two groups, one comprised of a single cell. Equal fragmentation is favoured when there is a significant advantage associated with formation of even the smallest group, whereas production of a unicellular propagule is favoured when the benefits associated with group size are not evident until groups become large. This makes intuitive sense: when advantages arise when groups are small, it pays for offspring to be in groups (and not single cells). Conversely, when there is little gain until group size is large, it makes sense to maintain one group that reaps this advantage. Interestingly, two bacteria that form groups and are well studied from a clinical perspective, Neisseria gonorrhoeae and Staphylococcus aureus, both show evidence of the above binary splitting fragmentation modes: Neisseria gonorrhoeae divide into groups of two equal sizes [6], while Staphylococcus aureus divide into one large group plus a unicellular propagule [7]. This leads to questions concerning the nature of the fitness landscape occupied by these bacteria and the basis of any collective level benefit as assumed by our model.
Once cell loss upon fragmentation is incorporated as a factor in collective reproduction, a wider range of fragmentation patterns becomes optimal. When fragmentation costs are fixed to a given number of cells, optimal fragmentation modes include those where splitting involves the production of multiple offspring. Among these, a prominent fragmentation strategy is multiple fission, where a group breaks into multiple independent cells. Such a fragmentation mode is reminiscent of palintomy in the volvocine algae [38]. A key difference between our “multiple fission” and palintomy is that the former involves a group of cells growing up to a threshold size at which point fragmentation happens, while the latter involves a single reproductive cell growing to many times its initial size and then undergoing several rounds of division. However, reinterpreting birth rates of cells in groups as growth rates of unicells of different sizes allows us to use our analysis to determine conditions under which such a mode of fragmentation is more adaptive than, say, the more standard strategy of growing to twice the initial size and then dividing in two (which for arbitrary sizes of offspring groups is equivalent to our “equal binary fragmentation” mode). Our results suggest that palintomy is favored over binary fission (and any other fragmentation mode) under a wide range of demographic scenarios (Fig 6C).
Many multicellular organisms are characterised by a life cycle whereby adults develop from a single cell [39]. Passing through such a unicellular bottleneck is a requirement for sexual reproduction based on syngamy, but life cycles with unicellular stages are also common in asexual reproduction modes such as those used by multicellular algae and ciliates [40], and colonial bacteria such as S. aureus [7]. If multicellularity evolved because of the benefits associated to group living, why do so many asexual multicellular organisms begin their life cycles as solitary (and potentially vulnerable) cells? Explanatory hypotheses include the purge of deleterious mutations and the reduction of within-organism conflict [39, 41]. Our results make the case for an alternative (and perhaps more parsimonious) explanation: often, a life cycle featuring a unicellular bottleneck is the best way to guarantee that the “parent” group remains as large as possible to reap maximum fecundity and/or survival advantages of group living. Indeed, our theoretical results resonate with previous experimental work demonstrating that single-cell bottlenecks can be adaptive simply because they constitute the life history strategy that maximises reproductive success [42].
Previous theoretical work has explored questions related to the evolution of multicellularity using matrix population models similar to the one proposed in this paper. In a seminal contribution, Roze and Michod [43] explored the evolution of propagule size in the face of deleterious and selfish mutations. In their model, multicellular groups first grow to adult size and then reproduce by splitting into equal size groups, so that fragmentation mode strategies can be indexed by the size of the propagule. In our terminology, this refers to either “multiple fission” or “multiple groups”. An important finding of Roze and Michod [43] is that, even if large groups are advantageous, small propagules can be selected because they are more efficient at eliminating detrimental mutations. We did not study the effects of mutations, but allowed for general fitness landscapes and fragmentation modes, including cases of asymmetric binary division (e.g., the unicellular propagule strategy) neglected by Roze and Michod [43]. Our results indicate that modes of fragmentation involving single cells can lead to growth rate maximisation even when small propagule sizes divide less efficiently or die at a higher rate. In particular, we have shown that if fragmentation is costly, a strategy consisting of a multiple fragmentation mode with a propagule size of one (i.e., the small propagule strategy studied by Roze and Michod [43]) can be adaptive for reasons other than the elimination of deleterious mutations.
Closer to our work, Tarnita et al. [18] investigated the evolution of multicellular life cycles via two alternative routes: “staying together” (ST, whereby offspring cells remain attached to the parent) and “coming together” (CT, whereby cells of different origins aggregate in a group). In particular, they studied the conditions under which a multicellular strategy that produces groups via ST can outperform a solitary strategy whereby cells always separate after division. The way they modelled group formation and analyzed the resulting population dynamics (by means of biological reactions and matrix models) is closely related to our approach. Indeed, their solitary strategy is our binary mode 1+1, while their ST strategy corresponds to a particular kind of binary mixed fragmentation mode. However, the questions we ask are different. Tarnita et al. [18] were concerned with the conditions under which (multicellular) strategies that form groups can invade and replace (unicellular) strategies that remain solitary. Contrastingly, we aimed to understand the optimal fragmentation mode out of the vast space of fragmentation strategies comprising all possible deterministic and probabilistic pathways by which complexes can stay together and split apart. A key finding is that, for any fitness landscape and if the environment is constant, mixed fragmentation modes such as some of the ST strategies considered by Tarnita et al. [18] will be outperformed by at least one pure fragmentation mode.
More recently, Rashidi et al. [20] developed a conceptual framework to study the competition of life cycles that involved five different life cycles defined by fragmentation patterns of the form 1+1+…+1 and an associated genetic control. Their model, which explicitly considers growing cells of different size, showed that depending on the fitness landscape, each of their five life cycles could prevail. By extending the range of life cycles to encompass all possible fragmentation modes (albeit with less detailed attributes), we have shown that certain life cycles will be suboptimal for any given fitness landscape.
In line with many studies in life history theory [32, 33], we made the simplifying assumption that the phenotype consists of demographic traits (in our case, probabilities of fragmenting into given fragmentation patterns) linked by trade-offs which interact to determine fitness (growth rate). This allowed us to predict the optimal phenotype at equilibrium at the expense of leaving unspecified whether, due to genetic constraints, such an equilibrium will be possible in an actual biological system. The question that inevitably arises is whether, given a presumptive genotype-phenotype mapping, it is possible for evolution to fine tune life cycles with group-level properties (such as specific fragmentation patterns) so that optimal fragmentation modes will be obtained as the endpoint of an evolutionary process. While a complete answer requires a more sophisticated analysis, we see no conceptual obstruction preventing seemingly arbitrary fragmentation modes to evolve. Firstly, genotype-phenotype maps of existing organisms can be complex and offer opportunity for adaptation, involving important qualitative behavioral changes [44–46]. Secondly, small genotypic changes can produce major phenotypic changes. For instance, Hammerschmidt et al. [3] observed the emergence of collective-level properties in a previously unicellular organism that was caused by a small number of mutations. Thirdly, even if a current set of genes cannot provide an appropriate template for given phenotypic traits, new genes can emerge de novo [47–51]. Finally, theoretical arguments suggest that genetic constraints can be effectively overcome in phenotypic evolution provided there is a rich variety of new mutant alleles [52]. We thus think that, both in the field and in the laboratory, multicellular organisms will be able to evolve a phenotype close to the optimal fragmentation mode in the (very) long run.
|
10.1371/journal.pgen.1007374 | Role of heterotrimeric Gα proteins in maize development and enhancement of agronomic traits | Plant shoot systems derive from the shoot apical meristems (SAMs), pools of stems cells that are regulated by a feedback between the WUSCHEL (WUS) homeobox protein and CLAVATA (CLV) peptides and receptors. The maize heterotrimeric G protein α subunit COMPACT PLANT2 (CT2) functions with CLV receptors to regulate meristem development. In addition to the sole canonical Gα CT2, maize also contains three eXtra Large GTP-binding proteins (XLGs), which have a domain with homology to Gα as well as additional domains. By either forcing CT2 to be constitutively active, or by depleting XLGs using CRISPR-Cas9, here we show that both CT2 and XLGs play important roles in maize meristem regulation, and their manipulation improved agronomic traits. For example, we show that expression of a constitutively active CT2 resulted in higher spikelet density and kernel row number, larger ear inflorescence meristems (IMs) and more upright leaves, all beneficial traits selected during maize improvement. Our findings suggest that both the canonical Gα, CT2 and the non-canonical XLGs play important roles in maize meristem regulation and further demonstrate that weak alleles of plant stem cell regulatory genes have the capacity to improve agronomic traits.
| Maize is one of the most important cereal crops worldwide. Optimizing its yields requires fine tuning of development. Therefore, it is critical to understand the developmental signaling mechanisms to provide basic knowledge to maximize productivity. The heterotrimeric G proteins transmit signals from cell surface receptor and have been shown to regulate many biological processes, including shoot development. Here we study the role of G protein α subunits in maize development by either making the only canonical Gα constitutively active or mutating all other non-canonical Gα subunits (XLGs). We demonstrate that CT2 and XLGs have both redundant and specialized functions in regulating shoot development. Importantly, we show that a constitutively active Gα functioned as a weak allele, which introduced multiple desirable agronomic traits, such as improved kernel row number and reduced leaf angle.
| The plant shoot system is derived from the SAMs, pools of stems cells that have the ability of self-renewal, while initiating new leaves and axillary meristems [1]. The CLV-WUS negative feedback loop has been identified as the key pathway to regulate SAM proliferation and differentiation in Arabidopsis, and is widely conserved in other species [2]. This pathway relies on the communication between a battery of receptors, peptides and transcription factors. WUS, a homeodomain transcription factor expressed in the organizing center, promotes stem cell fate [2], while CLV3, a small peptide ligand that is secreted from stem cells at the tip of the SAM, is perceived by leucine-rich repeat (LRR) receptor kinases, such as CLV1, and receptor-like protein CLV2, resulting in the repression of WUS transcription [3–5]. The CLV pathway is conserved in crops, for example maize CLV1 and CLV2 receptor orthologs THICK TASSEL DWARF1 (TD1) and FASCIATED EAR2 (FEA2) function in meristem regulation, and both td1 and fea2 mutants show enlarged meristems, or fasciated, phenotypes [6, 7]. However, the signaling players and mechanisms downstream of the CLV receptors are poorly understood.
A common class of proteins that signal directly downstream of cell surface receptors in mammalian systems is the heterotrimeric G proteins. These proteins, consisting Gα, Gβ, and Gγ subunits, are also key regulators in the transduction of extracellular signals in plants [8]. The classical model established in animals suggests that in the inactive state, the GDP-bound Gα associates with the Gβγ dimer. Ligand activation of an associated 7-transmembrane domain (7-TM) G-protein-coupled receptor (GPCR) induces the exchange of GDP for GTP on Gα, promoting dissociation of Gα from the receptor and Gβγ dimer. The activated Gα and Gβγ subunits then interact with downstream effectors to transduce signaling [9]. Emerging evidence suggests that instead of interacting with 7-TM GPCRs as in animals, the plant G proteins interact with single-TM receptors to regulate plant development and disease resistance [10–13]. Recent genetic screens in maize and Arabidopsis identified roles for heterotrimeric G protein α and β subunits in meristem regulation, by interacting with CLV related receptors [10, 12]. In maize, the Gα subunit COMPACT PLANT2 (CT2) interacts in vivo with the LRR receptor-like protein FEA2, to control shoot and inflorescence meristem development. ct2 mutants have enlarged SAMs, fasciated ears with enlarged ear inflorescence meristems and more rows of kernels [10]. In contrast, in Arabidopsis the Gβ subunit, AGB1, interacts with another CLV-related receptor RECEPTOR-LIKE PROTEIN KINASE2 (RPK2), to transmit the stem cell restricting signal, and agb1 mutants develop bigger SAMs [12].
In addition to interacting with a different class of receptors, the regulatory mechanism of Gα function in plants appears to be fundamentally different from that in animals, since plant Gα subunits spontaneously exchange GDP for GTP in vitro, without requiring GPCR activation [14, 15]. This novel mechanism of regulation involves a non-canonical Regulator of G-protein Signaling (RGS) protein in Arabidopsis, which contains a 7-TM domain coupled to an RGS domain [16], and promotes conversion of Gα-GTP back to Gα-GDP [16]. However, RGS homologs are missing from many grass species, including maize [15, 17–19]; therefore, the mechanism of plant G protein regulation, particularly the transition between the active and inactive states, remain largely unknown in these species. Expression of constitutively active Gα subunits that have lost GTPase activity, disrupting the balance between active and inactive Gα, results in distinct phenotypes, supporting the idea that Gα activity needs to be carefully controlled [16, 20, 21]. However, the implication of Gα constitutive activity on meristem development has not been addressed.
Plants also differ from animals in containing only a relatively small number of heterotrimeric G protein genes. Most plants have only one canonical Gα [15], however they also encode non-canonical Gα subunits, extra-large GTP binding proteins (XLGs), which contain a Gα domain at the C-terminus [22–28]. Arabidopsis has 3 XLGs, and they function either redundantly or independently, depending on the biological process [22–28]. Arabidopsis xlg1/2/3 triple mutants do not have an obvious shoot meristem phenotype, however knocking out the 3 XLGs in a Gα (gpa1) background leads to a significant increase in shoot meristem size [25], suggesting they function redundantly with the canonical Gα in meristem regulation; however, the importance of G protein signaling in diverse plant species remains obscure. Taking advantage of the strong developmental phenotypes of maize Gα mutant ct2, here we explore the roles of G proteins in maize development by either making Gα constitutively active or mutating all maize XLGs using multiplex CRISPR-Cas9. We demonstrate that CT2 and XLGs have both redundant and specialized functions in regulating meristem development, and importantly, manipulation of maize Gα subunits introduced desirable agronomic traits.
Our previous study showed that the maize heterotrimeric G protein α subunit CT2 plays an important role in shoot meristem regulation, by associating with a maize CLV receptor FEA2 [10]. However, the underlying signaling mechanism remains obscure, and the implication of Gα activity on meristem development has not been addressed. We took the opportunity of the strong maize phenotype to investigate the effect of forcing Gα to be constitutively active in vivo. We hypothesize that the GTPase activity and the GDP-GTP exchange cycle are required for full Gα function in transmitting the CLV signaling to regulate maize meristem development, and thus mutants that are defective in GTPase activity may act as a weak allele of ct2. Exchange of a single amino acid in mammalian, Arabidopsis, or rice Gα proteins is sufficient to block GTP hydrolysis, resulting in a constitutively active (GTPase-dead) form [16, 20, 29]. On this basis, we introduced an analogous point mutation, Q223L, in CT2, to generate a constitutively active protein, which we named CT2CONSTITUTIVELY ACTIVE (CT2CA). To ask if the Q223L mutation abolished GTPase activity, we performed in vitro GTP-binding and GTPase activity assays using fluorescent BODIPY-GTP, where an increase in fluorescence over time corresponds to GTP binding, and a subsequent decrease corresponds to GTP hydrolysis [30]. We first established that CT2 works as an authentic Gα protein, by testing GTP/GDP binding and hydrolysis specificity. CT2 rapidly bound then slowly hydrolyzed fluorescent GTP, with similar kinetics to other vascular plant Gα proteins (Fig 1A and S1A Fig) [15, 30], and the activity was efficiently competed by non-labeled GTP or GDP but not by ATP or ADP (S1C Fig). As expected, the CT2CA protein had similar GTP-binding, but lacked GTPase activity (Fig 1A and S1D Fig). We further asked if CT2CA interacted with Gβγ in a yeast-3-hybrid (Y3H) system. In contrast to CT2, we found that CT2CA did not interact with the Gβγ dimer, despite being expressed at a similar level as CT2 in the yeast cells (Fig 1B and S1B Fig). In summary, the Q223L point mutation abolished the GTPase activity of CT2, maintaining it in a constitutively active state that could no longer form a heterotrimeric complex with Gβγ.
To test if constitutive activation of CT2 impacted maize development, we introduced the Q223L point mutation into a native CT2 expression construct that also carried an in-frame fusion of mTFP1 at an internal position that maintains full protein function [10] (Fig 2A). After transformation into maize, CT2CA-mTFP1 was correctly localized in a thin line at the cell periphery that co-localized with a plasma membrane (PM) counterstain, FM4-64 (Fig 2B), and we confirmed this co-localization following plasmolysis (Fig 2B). We next backcrossed 6 independent transgenic events of CT2CA-mTFP1 into ct2 mutants in a B73 background. Our previous work established that a native CT2-YFP expression construct fully complements ct2 mutant phenotypes, and we found that both CT2CA-mTFP1 and CT2-YFP were expressed at a similar level as the endogenous CT2 [10] (S2A, S2B and S2D Fig). We first asked if CT2CA-mTFP1 was able to complement the vegetative growth defects of ct2 mutants, by measuring plant height and the first leaf length. CT2CA-mTFP1; ct2 plants were significantly taller than ct2 mutants, with longer leaves; however, they were significantly smaller than their normal, ct2 heterozygous siblings with or without the CT2CA-mTFP1 transgene, indicating that CT2CA-mTFP1 only partially rescued the vegetative growth defects of ct2 mutants (Fig 2C, 2D, 2F and 2G, similar results obtained with 6 independent transgenic events, S2C Fig). We also asked if CT2CA-mTFP1 could complement the enlarged meristem phenotypes of ct2 mutants. We again found partial complementation, indicating that CT2CA was only partially functional in meristem regulation (Fig 2E and 2H). Since CT2 is involved in the CLV-WUS pathway by interacting with FEA2 [10], we tested if CT2CA can still interact with FEA2 in an N. benthamiana transient expression system. The result showed that FEA2-Myc was pulled down by both CT2-YFP and CT2CA-YFP in the co-IP experiment (S3 Fig). Similarly, studies in human and insect cells showed that in some cases G protein subunits and receptors remain associated following receptor activation [31–33]. Further studies will be needed to elucidate the underlying mechanisms. In addition, to ask how CT2CA affected downstream signaling, we measured ZmWUS1 expression in inflorescence transition stage meristems by qRT-PCR. However, we found that ZmWUS1 expression was not significantly changed in ct2 mutants compared to wild type, nor in our constitutively active CT2CA-mTFP1 lines (S4 Fig), similar to other studies involving subtle changes in CLV pathway genes [34, 35] and reflecting the complex non-linear regulation of the CLV-WUS negative feedback loop. Collectively, our results suggest that ct2ca functioned as a weak allele of ct2, and that normal GTPase activity and the GDP-GTP exchange cycle is required for full Gα function in maize.
We next asked if CT2/Gα function in maize might be compensated by XLGs. We used phylogenetic analysis (Fig 3A) to compare the maize XLGs to Arabidopsis, and based on this named them ZmXLG1 (most similar to AtXLG1) and ZmXLG3a and b (most similar to AtXLG3). We first asked if the three ZmXLGs might function in a heterotrimeric G protein complex, by testing their interaction with a Gβγ dimer in Y3H system. All three were indeed able to interact, similar to CT2, suggesting that they function in maize heterotrimeric G protein complexes (Fig 3B).
To study the functions of ZmXLGs in maize development, we knocked out all three genes using a tandem guide RNA (gRNA) CRISPR-Cas9 construct. In one transgenic event, we recovered putative null alleles of all 3 genes, a 1-bp insertion allele for ZmXLG1, a 4-bp deletion allele for ZmXLG3a, and a 31-bp deletion allele for ZmXLG3b, each within the N terminal half of the protein coding region and before the Gα domain (Fig 4A). Inbreeding these plants produced offspring homozygous for all 3 loci, at the expected ratio. All Zmxlg triple mutant plants showed a striking developmental arrest phenotype, as they were lethal at the seedling stage (Fig 4B), much more severe than in Arabidopsis, where the triple mutants are smaller with reduced fertility, but can still complete the life cycle [27, 36]. To gain a deeper understanding into the lethal phenotype, we assayed for cell death using trypan blue staining. As shown in S5A Fig, the triple mutants had strong staining, suggesting they were undergoing cell death. We also measured the expression of two immune marker genes, PATHOGENESIS-RELATED PROTEIN 1 (PR1) and PR5, and found both were significantly up-regulated in the triple mutants, indicating that the lethality may be due to over-activation of immune system (S5B Fig). Rice Gβ mutants also display cell death and lethality [37, 38], indicating that the lethal phenotype of G protein mutants is not unique to maize. The reason for these differences between monocot and dicot G protein mutants remains elusive, but may be related to their dual role in immune signaling [13, 24, 39, 40]. Although the Zmxlg triple mutant plants stopped growing soon after germination, we could measure their shoot meristems, and found that they were normal in size and structure (S6A Fig).
As the Zmxlg1;3a;b triple mutants were lethal, we next analyzed the developmental phenotype of single or double mutants. Knocking out each single ZmXLG did not alter development; whereas knocking out any two ZmXLGs led to a modest but significant reduction in plant height, but did not affect SAM size (Fig 4C and 4D and S7 Fig), indicating that loss of any two ZmXLGs can be partially compensated by other XLGs or by CT2/Gα. Next, we asked if ZmXLGs function redundantly with the canonical maize Gα, CT2, by crossing the Zmxlg mutants into a ct2 mutant background. As expected, ct2 mutants were significantly shorter than wild-type siblings [10], and we found that mutation of any two ZmXLGs dramatically enhanced their dwarf phenotype (Fig 5A and 5B). In addition, mutation of any pair of ZmXLGs significantly increased SAM size in a ct2 mutant background (Fig 5C and 5D), indicating that ZmXLGs are partially redundant with CT2 in SAM regulation. In contrast, although both CT2 and ZmXLGs are expressed in the maize inflorescence, ZmXLG knockouts did not enhance the ct2 inflorescence fasciation phenotype (S6B, S6C and S8 Figs), suggesting that CT2 is the major Gα functioning in inflorescence meristem development. In summary, our results showed that XLGs are partially redundant with CT2 at some stages of development, but that all 3 XLGs redundantly function in early maize development, where they are essential for survival past the germination stage, and cannot be compensated by CT2.
Our previous results indicate that weak alleles of meristem regulatory genes, such as fea2 or fea3 can improve agronomic traits, such as increasing kernel row number (KRN), without the negative yield impacts associated with strong fasciation phenotypes [41, 42]. The results described above suggest that different Zmxlg mutant combinations reduce maize height, which is an important trait selected during breeding of many cereal crops [43, 44]. We also found that ct2ca functions as a weak allele of CT2, and therefore asked if its expression might affect agronomic traits. First, we measured tassel spikelet density, a trait associated with increased meristem size [10, 42], of CT2CA-mTFP1-expressing plants in a ct2 homozygous or heterozygous background. ct2 plants expressing CT2CA-mTFP1 had a significantly higher spikelet density compared with normal, ct2 heterozygous siblings with or without the CT2CA-mTFP1 transgene (Fig 6A and 6B). In addition, these plants did not develop stunted, fasciated ears as in ct2 mutants, but made ears of normal length with increased KRN compared with normal, ct2 heterozygous siblings with or without the CT2CA-mTFP1 transgene (Fig 6C and 6D). Since our previous results suggest that there is a positive correlation between the ear inflorescence meristem size and kernel row number [41], we next checked if this is also true for ct2 plants expressing CT2CA-mTFP1. Consistently, we found that they had significantly larger ear IMs compared with normal, ct2 heterozygous siblings with or without the CT2CA-mTFP1 transgene (Fig 6E and 6F), but were not fasciated.
Leaf angle is another important agronomic trait, because more upright leaves reduce shading and improve photosynthetic efficiency in modern high plant density production systems[45]. ct2 mutants have more erect leaves, however also have negative pleiotropic traits such as extreme dwarfing and very wide leaves [10, 46, 47]. Interestingly, we found that plants expressing constitutively active CT2 also had a more erect leaf angle compared with normal, ct2 heterozygous siblings with or without the CT2CA-mTFP1 transgene, without obvious negative pleiotropic phenotypes (Fig 6G and 6H). In summary, we found that ct2 plants expressing a constitutively active CT2/Gα develop phenotypes consistent with a weak allele of ct2. These finding suggest that the GTPase activity and the GDP-GTP exchange cycle is required for full CT2 function in vivo, but that expression of a constitutively-active version of CT2 can act as a partially functional (weak) allele that brings desirable agronomic traits.
Heterotrimeric G protein signaling in mammals and yeast transmits a plethora of developmental and physiological signals from GPCRs to downstream effectors [48, 49]. Mammals contain many Gα homologs, therefore the full significance of knocking out all Gα signaling has not been addressed. Plants contain a much smaller number, usually a single canonical Gα and ~3 related XLGs [9, 15, 22, 50]. XLGs are evolved from the canonical Gα, and share some redundant functions [25]. In some extreme examples such as moss Physcomitrella patens, the canonical Gα even has been lost during evolution, and its function has been completely replaced by the sole XLG [51]. XLGs have also gained independent functions during evolution, for example, Arabidopsis XLG2, but not the canonical Gα, interacts with the FLS2 receptor and mediates flg22-induced immune responses [13]. In Arabidopsis, knockouts of all 3 XLGs have no obvious effect on shoot meristem development, and the additional knockout of the canonical Gα leads to only modest effects on development, including a change in leaf shape and slightly larger shoot meristem [25]. These results suggest that the canonical Gα and XLGs work redundantly to regulate shoot development, and heterotrimeric G protein signaling plays a relatively modest role in plant development. In this report, we found that the maize XLGs work both redundantly and independently with the canonical Gα, CT2. Zmxlg mutations enhanced ct2 null phenotypes in plant height and meristem size during vegetative development, suggesting ZmXLGs function redundantly with CT2 in SAM regulation. However, knocking out all the 3 XLGs in maize leads to a striking early seedling growth arrest and lethality, independent of the presence of CT2, suggesting ZmXLGs are essential in maize early growth and development. In addition, knocking out ZmXLGs did not enhance ear fasciation, suggesting CT2 is the sole Gα functioning in inflorescence meristem development. Collectively, our results suggest that the maize XLGs and CT2 have overlapping functions at certain stage of development, however, both have evolved specialized functions. While we do not know the signaling pathways of the maize XLGs, it is likely that they interact with receptors involved in plant growth and development, analogous to the interaction between Gα and a CLV receptor [10].
The classic heterotrimeric G protein model established in the mammalian system suggests that Gα is usually in the inactive GDP-bound state, and is activated to switch to the active GTP-bound state by ligand binding to a 7-TM GPCR [9]. However, the plant G proteins, including those from grasses, are spontaneously active in vitro, and it is still under debate if plants have canonical 7-TM GPCRs [15]. Instead, several single TM receptors, such as CLV and innate immune receptors have been found to interact with G proteins [10–13]. Recent studies in Arabidopsis suggest that turning off plant Gα signaling is also an important step for its signal transduction [15, 52], indicating that the balance between the active and inactive Gα pool is important to fully exert its function. We found that native expression of CT2CA-mTFP1 in maize partially rescued ct2 mutant phenotypes. Sometimes partial transgene complementation of a mutant is due to improper transgene expression. However, our native CT2-YFP expression construct fully complemented ct2 mutant phenotypes, and CT2CA-mTFP1 was expressed at the same level as CT2-YFP and endogenous CT2 (S2D Fig), so we conclude that the partial complementation is indeed caused by the loss of GTPase activity. In yeast, a constitutively active Gα also similarly only partially complemented the growth defects of Gα null mutants [53], suggesting that the requirement for GTPase activity is universal. In addition to GTPase activity, GTP binding is also important for the function of Gα. For example, the T475N mutant of Arabidopsis XLG2, which lacks GTP binding activity, is not able to interact with a downstream effector RELATED TO VERNALIZATION1 (RTV1) [54]. Together, all of these studies suggest that Gα has to bind GTP and to cycle between the active and inactive state to fully exert its function. One explanation for the importance of the cycling is that the Gα controls meristem development through coordinating with the Gβγ dimer pool. Presumably, in both ct2 mutants and CT2CA background, more free Gβγ dimers are released. Arabidopsis Gβ regulates the meristem development via interacting with a CLV-like receptor RPK2 [12], while the maize Gα, CT2 interacts with another CLV receptor-like protein, FEA2 [10]. It is possible that Gα and Gβ function independently by coordinating with different receptors and downstream effectors at the cell surface, whereas signaling converges at some point. Although their downstream effectors remain largely unknown in plants, Gβ forms a complex with mitogen-activated protein kinases (MAPKs) [40], which may function in the CLV pathway [55]. Therefore, fine-tuning of the active and inactive states of G protein as well as the Gα and Gβγ pools may be important to maintain meristem development, and our study illustrates the complexity of G protein signaling in meristem regulation. Importantly, ct2ca functioned as a weak allele and introduced desirable agronomic phenotypes, similar to many weak alleles that underlie QTLs for crop traits [41, 42, 56]. Optimization of traits such as spikelet density, kernel row number, and leaf angle has been key to improvements in maize and other crops, both in improving yield per plant and planting density. Targeting specific regulators such as Gα by using CRISPR to generate weak alleles could enhance multiple yield related phenotypes to meet the food demands of the increasing global population.
Yeast codon-optimized ORFs of CT2 (GRMZM2G064732), CT2CA, ZmXLG1 (GRMZM2G127739), ZmXLG3a (GRMZM2G016858), and ZmXLG3b (GRMZM2G429113) were cloned between the EcoRI and XhoI restriction sites of MCS1 of pGADT7 (Clontech). ZmGB1 (GRMZM2G045314) was cloned between the EcoRI and BamHI restriction sites of MCS1. ZmRGG2 (GRMZM6G935329) was cloned between the NotI and BglII restriction sites of MCS2 of pBRIDGE (Clontech), respectively. The primer sequences are shown in the supplementary information. The yeast assay was performed in the AH109 yeast strain (Clontech). The double transformants were selected on SC -Trp -Leu (-LW) plates. The interaction was tested on the SC -Trp -Leu -His (-LWH) medium supplemented with 1 mM 3-Amino-1,2,4-triazole (3-AT) to suppress histidine synthesis. The HA-tag was detected using the monoclonal anti-HA antibody produced in mouse (Sigma, clone HA-7).
The guide RNAs were designed using the CRISPR-P website (http://cbi.hzau.edu.cn/crispr/) [57]. The multi-gRNA array was synthesized and cloned into pMGC1005 vector by the LR recombination reaction (Invitrogen) (S1 File) [58]. The construct was introduced into EHA101 and transformed into HiII background using Agrobacterium-mediated transformation by Iowa State University Plant Transformation Facility. The genomic regions spanning the gRNA target sites were amplified by PCR and sequenced. The T0 plants containing lesions in all three XLG genes were backcrossed with ct2 in the B73 background and self-crossed.
CT2CA-mTFP1 was constructed by amplification of genomic fragments and fusing with the mTFP1 gene in-frame at an internal position using the MultiSite Gateway Pro system (Invitrogen), as described [10]. All fragments were amplified using KOD Xtreme hot start polymerase (Millipore Sigma) and the Q223L point mutation was generated using PCR-based mutagenesis. The ORF of mTFP1 was inserted between the two amino-terminal α helices, αA and αB of CT2, as described [10]. All the entry clones were assembled in the pTF101 Gateway compatible binary vectors and introduced into the EHA101 Agrobacterium strain. The construct was transformed into HiII background using Agrobacterium-mediated transformation by the Iowa State University Plant Transformation Facility. The T0 plants were backcrossed twice with ct2 mutants in the B73 background. For genotyping, a 1.5 kb fragment of the CT2 gene was amplified and digested with AccI, as a single SNP causes a loss of the 5’ AccI site in the ct2-Ref allele. The transgene was amplified using one primer against the mTFP1 sequence and the other primer against the ct2 sequence. Primers are listed in the S1 Table.
For the SAM, ear IM, plant height, and leaf angle measurements, the plants were grown in the greenhouse with the light cycle 16/8 h light/dark and the temperature was maintained between (26–28°C). For the spikelet density and KRN measurement, the plants were grown at the Uplands Farm Agricultural Station at Cold Spring Harbor, New York between June to October.
For SAM measurements, maize seedlings were grown in the greenhouse for 15 days and then dissected and fixed in FAA (10% formalin, 5% acetic acid, and 45% ethanol). The fixed tissues were subsequently dehydrated with 70, 85, 95 and 100% ethanol for 30 min each and then immersed in an ethanol-methyl salicylate solution (1:1) for an additional 60 min. The tissue was then cleared in 100% methyl salicylate for 2 hours. The SAMs were imaged with a Leica DMRB microscope with a Leica MicroPublisher 5.9 RTV digital camera system. For IM measurements, ear primordia 2 mm in length were dissected. The pictures were taken using a Hitachi S-3500N scanning electron microscope or a Nikon SMZ1500 dissection microscope equipped with a camera. The SAMs and ear IMs were measured using Image J.
The 4-week old shoot apices were harvested for measuring ZmWUS1 (GRMZM2G047448) expression, and 1-wk old seedlings were used to measure CT2, CT2CA-mTFP1, and CT2-YFP as well as PR1 (GRMZM2G465226) and PR5 (GRMZM2G402631) expression. qRT-PCR was performed on a CFX96 Real-Time system (Bio-Rad). Total mRNA was extracted using the Direct-zol RNA extraction kit (Zymo Research). The cDNA was synthesized using the iScript Reverse Transcription Supermix (Bio-Rad) according to the manufacturer’s manual. The relative expression level of the targeted genes was normalized using ZmUBIQUITIN. Primers are listed in S1 Table.
The trypan blue staining was performed using 1-wk old wild-type and Zmxlg triple mutants, as described with slight modifications [59]. Briefly, the whole shoot was immersed in lactophenol containing 2.5 mg/mL trypan blue, and heat in a boiling water for 1 min. Then allowing the samples site at room temperature for overnight. The tissue was cleared in chloral hydrate solution (25 g of chloral hydrate in 10 ml of H2O) for 24 hours at room temperature.
The ORF of YFP was inserted between the two amino-terminal α helices, αA and αB of CT2 or CT2CA, as described [10]. The entry clones containing 2x35S promoter, CT2 or CT2CA-YFP, and Nos terminator were assembled in the pTF101 Gateway compatible binary vectors. The ORF of FEA2 was fused with the 6xMyc tag sequence and cloned into the pEARLEY301 vector [60]. All the binary vectors were introduced into the GV3101 Agrobacterium and infiltrated into 4-week-old N. benthamiana leaves with a P19 vector to suppress posttranscriptional silencing [61]. The protein extraction and membrane fraction enrichment were performed as described [10] with some modifications. Briefly, the leaves were harvested 3-day post infiltration and ground in liquid nitrogen to a fine powder then suspended in twice the volume of protein extraction buffer containing 150 mM NaCl, 50 mM Tris-HCl pH 7.5, 5% glycerol, and cOmplete, mini, EDTA-free protease inhibitor (Roche) and rotated in a cold room for 15 min. Then centrifuge at 4,000g for 10 min at 4°C, followed by filtration through Miracloth (Millipore Sigma), resulting a total protein extraction. The extract was then centrifuged at 100,000g for 1 h at 4°C to pellet the microsomal membrane fraction. The resulting pellet was re-suspended in 2 ml extraction buffer supplemented with 1% Triton X-100 with a glass homogenizer. Then the lysates were cleared by centrifugation at 100,000g for 1 h at 4°C to remove non-solubilized material. For co-immunoprecipitation experiments, solubilized microsomal membrane fractions were incubated with 30 μl magnetic beads coupled to monoclonal mouse anti-GFP antibody (μMACs, Milteny Biotec, 130-094-3252) for 30 min at 4°C. Flow-through columns were equilibrated using 250 μl membrane solubilization buffer before lysates were added. The MicroBead-bound target proteins were magnetically separated, and washed one time with 250 μl membrane solubilization buffer and three times with wash buffer 1 containing 150 mM NaCl, 50 mM Tris pH7.5, 0.1% SDS and 0.05% IGEPAL-CA-630 followed by one time with wash buffer 2 containing 20 mM Tris, pH 7.5, supplied by the company. Bound target proteins were eluted with 70 μl 1xSDS loading buffer. Following standard SDS-PAGE electrophoresis and blot transfer, FEA2-Myc protein was detected using an anti-Myc antibody generated from mouse (Millipore Sigma, 05–724) and a secondary HRP-coupled anti-mouse antibody (GE Healthcare Life Sciences, NA931). CT2 or CT2CA-YFP proteins were detected using an HRP-conjugated anti-GFP antibody (Miltenyi Biotech, 130-091-833).
BLAST search against the protein databases of Arabidopsis, maize, rice, and tomato using Arabidopsis GPA1 and maize CT2 was conducted in Phytozome (www.phytozome.com). The sequences were aligned with Clustal X [62] and the phylogenetic tree was constructed using the neighbor-joining model of MEGA7 [63]. One hundred bootstrap iterations were performed.
The coding sequence of CT2 was cloned into pPROEX-His vector between restriction sites EcoRI and XhoI. CT2CA was generated using PCR-based mutagenesis (Primers are shown in the S1 Table). Both constructs were transformed into Rosetta DE3 E.Coli cells for protein expression, as described by Urano et al. with modification [15]. The transformed cells were grown to an OD600 of 0.6 prior to induction by 0.5 mM 1-thio-β-D-galactopyranoside (IPTG) in LB medium for 18 hrs at 16°C. Cells were harvested by centrifugation and resuspended in 150 mM NaCl, 50 mM Tris, 10 mM imidazole, 5 mM β-mercaptoethanol (β-ME), 1 mM MgCl2, 10 μM GDP, and 10% glycerol, adjusted to a final pH of 7.5, and a cOmplete, mini, EDTA-free protease inhibitor tablet (Roche) was added. Cells were lysed by passage three times through a cell disruptor (Avestin) at greater than 15,000 psi, and the lysate was centrifuged at 29,000 g for 30 min to produce a clarified lysate. This lysate was loaded onto a cobalt-charged NTA resin column (GE Life Sciences), and washed with 500 mM NaCl, 50 mM Tris, 20 mM imidazole, 5 mM β-ME, 1 mM MgCl2, 10 μM GDP, and 5% glycerol, pH 7.5. Bound His-tagged protein was eluted with the same buffer including 300 mM imidazole and 10 mM MgCl2 prior to concentration and loading onto a Superdex-200 size-exclusion column (GE Life Sciences) equilibrated with 100 mM NaCl, 50 mM Tris, 10 mM MgCl2, 5 mM β-ME, 10 μM GDP, and 5% glycerol (final pH 7.5). Peak fractions were pooled and dialyzed to an appropriate buffer for later experiments.
The BODIPY-GTP assay was performed as described previously with slight modification [64]. Assays were performed at 25°C in a 200 μl reaction volume in the assay buffer (20 mM Tris-HCl, pH 8.0 and 10 mM MgCl2) with 25 μM purified protein and 50 nM BODIPY-GTP. For competition with non-labeled nucleotides, 25 μM of GTP, GDP, ATP or ADP was added to the assay buffer before starting the reaction. The fluorescence (excitation 485 nm, emission 528 nm) was recorded every 10 s for up to 40 min using a BioTek Synergy H4 fluorescence microplate reader.
For imaging of CT2CA-mTFP1, roots were counterstained with 1 mg/ml FM4-64 solution in water for 1min and washed with water. Images were taken with a Zeiss LSM 710 microscope, using 458 nm excitation and 488–515 nm emission for detection of mTFP1 and 514 nm excitation and 585–750 nm emission for detection of FM4-64. For plasmolysis, the tissues were treated with 20% sucrose for 30 min.
The significant differences between multiple groups were analyzed using ANOVA followed by the LSD test with Bonferroni correction in the R statistical programming language (www.R-project.org). All experiments were repeated at least twice and similar results were obtained. The result from one repetition is presented.
|
10.1371/journal.pgen.1004107 | Cell-Cycle Dependent Expression of a Translocation-Mediated Fusion Oncogene Mediates Checkpoint Adaptation in Rhabdomyosarcoma | Rhabdomyosarcoma is the most commonly occurring soft-tissue sarcoma in childhood. Most rhabdomyosarcoma falls into one of two biologically distinct subgroups represented by alveolar or embryonal histology. The alveolar subtype harbors a translocation-mediated PAX3:FOXO1A fusion gene and has an extremely poor prognosis. However, tumor cells have heterogeneous expression for the fusion gene. Using a conditional genetic mouse model as well as human tumor cell lines, we show that that Pax3:Foxo1a expression is enriched in G2 and triggers a transcriptional program conducive to checkpoint adaptation under stress conditions such as irradiation in vitro and in vivo. Pax3:Foxo1a also tolerizes tumor cells to clinically-established chemotherapy agents and emerging molecularly-targeted agents. Thus, the surprisingly dynamic regulation of the Pax3:Foxo1a locus is a paradigm that has important implications for the way in which oncogenes are modeled in cancer cells.
| Rare childhood cancers can be paradigms from which important new principles can be discerned. The childhood muscle cancer rhabdomyosarcoma is no exception, having been the focus of the original 1969 description by Drs. Li and Fraumeni of a syndrome now know to be commonly caused by underlying p53 tumor suppressor loss-of-function. In our studies using a conditional genetic mouse model of alveolar rhabdomyosarcoma in conjunction with human tumor cell lines, we have uncovered that the expression level of a translocation-mediated fusion gene, Pax3:Foxo1a, is dynamic and varies during the cell cycle. Our studies support that Pax3:Foxo1a facilitate the yeast-related process of checkpoint adaptation under stresses such as irradiation. The broader implication of our studies is that distal cis elements (promoter-influencing regions of DNA) may be critical to fully understanding the function of cancer-associated translocations.
| Rhabdomyosarcoma (RMS) is the most common childhood soft tissue sarcoma. Historically, RMS has been thought to arise from muscle because of the expression of myogenic markers. Most childhood RMS falls into one of two biologically distinct subgroups: alveolar (aRMS) or embryonal (eRMS). aRMS is the more aggressive variant with a survival rate of less than 20% when metastatic due to chemotherapy and radiation resistance [1]. aRMS is characterized by a frequent t(2;13) chromosomal translocation, which results in the PAX3:FOXO1A fusion gene, or less frequently by a t(1;13) mediated PAX7:FOXO1A fusion oncogene [1]. Clinically, the aggressive behavior of aRMS has been attributed to PAX3:FOXO1A transcriptional reprograming because fusion negative aRMS have a more favorable outcome similar to eRMS [2], [3], [4].
We previously developed a mouse model of aRMS employing a conditional knock-in approach that expresses Pax3:Foxo1a from the native Pax3 locus in fetal and postnatal myoblasts [5], [6], [7]. In this model, Pax3:Foxo1a was necessary but not sufficient for aRMS tumor initiation. Interestingly, cells expressing high levels of Pax3:Foxo1a were more prevalent in metastatic tumors [7]. The heterogeneity of Pax3:Foxo1a expression in primary and metastatic tumors, and enrichment in the latter, suggested that Pax3:Foxo1a might be selectively expressed in a subset of aRMS cells; alternatively, Pax3:Foxo1a expression might be temporally regulated. In the current study we present striking evidence that Pax3:Foxo1a is expressed in a dynamic manner and mediates a G2-specific program enabling checkpoint adaptation and refractoriness to therapy.
In our genetically-engineered conditional knock-in mouse model of aRMS, eYFP is expressed as a second cistron on the same mRNA as Pax3:Foxo1a (Figure 1A). We have observed heterogeneity of eYFP expression among tumor cells in situ (Figure 1B). To first examine Pax3:Foxo1a expression as a function of time, we flow sorted Pax3:Foxo1alow and Pax3:Foxo1ahigh cells using eYFP signal in two independent murine aRMS primary cultures (Figure 1C and 1D; Figure S1A and S1B). Comparison of Pax3:Foxo1a protein levels for sorted populations showed Pax3:Foxo1alow cells possessed much reduced levels of Pax3:Foxo1a protein (Figure 1E and Figure S1C). However, FACS analysis over time revealed that the eYFP signal of Pax3:Foxo1alow and Pax3:Foxo1ahigh tended towards the mean eYFP fluorescence intensity of unsorted tumor cells with time and/or cell divisions (Figure 1C and 1D; Figure S1A and S1B). Thus, Pax3:Foxo1ahigh cell could dynamically reduce expression of eYFP from the Pax3:Foxo1a locus, and Pax3:Foxo1alow cells could dynamically increase expression of eYFP from the Pax3:Foxo1a locus. We further confirmed that eYFP expression was indeed reflective of Pax3:Foxo1a expression in terms of protein half-life. Figure S1E and S1F shows levels of eYFP signal and Pax3:Foxo1a protein stability after translation inhibition by cycloheximide (CHX). Akin to the strong correlation between eYFP and Pax3:Foxo1a expression at the protein level (Figure 1 and Figure S1C), the protein half-lives of Pax3:Foxo1a and eYFP were roughly similar at 31.6 and 44.7 hours (Figure S1E and S1F), thereby affirming that eYFP is a reasonable surrogate for transcription of Pax3:Foxo1a from the Pax3 locus (we do however acknowledge that eYFP is a better marker of the start of Pax3:Foxo1a transcription than the end of Pax3:Foxo1a transcription or protein expression (i.e., since eYFP is expressed on the same mRNA as Pax3:Foxo1a, the beginning of fluorescence should coincide with the initial presence of the Pax3:Foxo1a transcript). Thereafter, eYFP is susceptible to photo-bleaching and possible proteasomal degradation sooner than the 44 hours observed under conditions of cyclohexamide treatment (Figure S1F)).
To investigate what conditions affect the dynamic alteration of Pax3:Foxo1a expression in aRMS cells, we compared eYFP fluorescence to cell cycle phase as determined by staining with the DNA dye Hoechst33342. Almost all Pax3:Foxo1alow cells existed in G0/G1 (2N) stage, while to our surprise Pax3:Foxo1ahigh cells were G2/M or hyperdiploid/multinuclear (≥4N) cells (Figure 1F and Figure S1D). We next performed time-lapse experiments of eYFP activity by confocal microscopy. Figure 1G shows in time-lapse images that eYFP activity during cell division is transiently but markedly increased, particularly in pre-mitotic cells. Interestingly, the level of eYFP in some multinuclear cells remained at a high level in cells that appeared to be unable to undergo telophase/cytokinesis (Movie S1).
We next performed QPCR of Pax3:Foxo1a and PAX3:FOXO1A using cell cycle specific sorted mouse and human aRMS cells, respectively. Both mouse and human aRMS cells showed significant differences in the mRNA expression of Pax3:Foxo1a and PAX3:FOXO1A in the transition from 2N (G1) to 3N (S phase) and 4N (G2/M) cells (Figure 2A and 2B) affirming cross-species relevance of the cell cycle dependent mRNA regulation of Pax3:Foxo1a expression.
To investigate the transcriptional basis of this Pax3:Foxo1a dynamic expression, we performed QPCR of Pax3 and Foxo1 using cell cycle specific sorted C2C12 mouse myoblast cells of the genotype Pax3(wt/wt) and mouse aRMS primary tumor cells of the genotype Pax3(wt/Pax3:Foxo1a). C2C12 myoblasts showed significant increases in Pax3 mRNA levels for 4N cells when compared with 2N cells (Figure 2C). Pax3 was not detectable in aRMS cells at the mRNA level (data not shown), which was also reflected in the absence of expression of Pax3 protein in aRMS cells by western blotting (Figure 2D). This result is consistent with our prior studies suggesting that Pax3:Foxo1a causes decreased expression of the wildtype Pax3 locus [5], [6]. By contrast, Foxo1 mRNA expression did not differ between 2N and 4N in either C2C12 myoblasts or aRMS tumor cells (Figure 2E). Thus, the cell cycle dependence of Pax3:Foxo1a may in some part be attributable to increased Pax3 promoter activity at G2/M versus G1 in C2C12 myoblasts, but Pax3:Foxo1a transcript level is so significantly increased over Pax3 in aRMS cells that other factors related to the chromosomal fusion are likely responsible, e.g. gain of a Foxo1a 3′ cis-enhancer, or loss of a Pax3 3′ cis-repressor repressor. From the design of the conditional knock-in allele [5], this element(s) can be inferred to exist in the 9.3 kB of the Foxo1a 3′ region containing exons 2 and 3 and untranslated region (6.5 kb), or exons 8–10 of Pax3. We also cannot exclude that stabilization of the Pax3:Foxo1a transcript may to some degree play a role, and this stabilization may or may not be related to the Foxo1a cis-elements on the chimeric mRNA.
Because Pdgfra [8] and Igf1r [9] are well known direct downstream targets of Pax3:Foxo1a, we determined whether these targets were expressed to any degree in 4N (G2/M) cells. We first sorted aRMS tumor cells for Pdgfra or Igf1r positivity versus negativity, then performed DNA content analysis. For both receptor tyrosine kinases (RTKs), the majority of cells with positive RTK surface expression were 2N (Figure 2F). However, nearly twice as many 4N cells are Igf1r (or Pdgfra) positive versus Igf1r (or Pdgfra) negative, suggesting these Pax3:Foxo1a targets may have a functional role late in the cell cycle, such as the Igf1r-mediated radioresistance seen for other forms of cancer [10].
To determine the role of Pax3:Foxo1a in G2, M or G2/M checkpoint, we examined markers of each cell cycle phase under non-stress or stress conditions. Immunocytochemistry is presented in Figure 3 is a for Pax3:Foxo1a (Pax3) with phospho-histone H3 (pHH3), a marker of mitosis, or CDC2-Y15 (pCDC2), a negative marker of entry into mitosis that is commonly expressed in G2 (CDC2-Y15 is phosphorylated by Wee1 kinase, which then negatively regulates Cdc2 kinase [11]; CDC2-Y15 is present starting in late G1 then also in S, and G2 phases, but absent in M [12]). In murine aRMS primary cultures U23674 and U42369, pHH3 positive metaphase cells did not express Pax3:Foxo1a protein and yet most pCDC2 positive cells expressed Pax3:Foxo1a very highly (Figure 3). These results suggest that Pax3:Foxo1a is expressed in the G2 cell cycle phase but not M phase. Human aRMS cell lines Rh3 and Rh41 showed identical results (Figure 3). Next, we sought to understand the function of Pax3:Foxo1a in G2. For this purpose we performed genome-wide expression analysis using cells sorted at specific stages of the cell cycle (2N vs. 4N) with or without Pax3:Foxo1a siRNA knockdown (Figure 4A). Because eYFP is expressed as a second cistron in the targeted Pax3:Foxo1a-ires-eYFP allele, we anticipated that siRNA for eYFP would knock down not only eYFP but also Pax3:Foxo1a. Western blotting of Pax3:Foxo1a and native Foxo1a protein 48 hours after siRNA transfection showed that eYFP siRNA efficiently and specifically knocked down Pax3:Foxo1a protein (Figure 4B). Protein expression of the Pax3:Foxo1a transcriptional target Pdgfra was also reduced (Figure 4B).
From genome-wide expression analysis of 2N vs. 4N sorted cells with or without Pax3:Foxo1a siRNA knockdown, we found several genes implicated in the process of G2/M checkpoint adaptation to be down-regulated in G2/M (4N cells) when Pax3:Foxo1a was knocked down (Figure 4C; Table S1 shows all data analyzed by ANOVA (<0.05) using the multiple comparison correction method of Benjamini and Hochberg). Checkpoint adaptation is the process by which unicellular organisms or cancer cells progress through a delayed cell cycle checkpoint (G2 or by analogy the mitotic spindle assembly checkpoint) in lieu of programmed cell death, but before DNA damage is completely repaired [13], [14], [15]. Factors implicated in checkpoint adaptation are similar to those involved in checkpoint recovery (after complete repair of DNA damage), but additionally require anti-apoptotic signals [14]. Select G2/M checkpoint adaptation genes implicated in this experiment, the DNA damage sensing/checkpoint progression factors Plk1, Cdc25b, H2afx and the cell survival factor Birc5 (Survivin), were validated for differential expression by QPCR (Figure 4D). Whether these genes are direct transcriptional targets of Pax3:Foxo1a was investigated by interrogating loci for reported nearby Pax3:Foxo1a binding sites [16]. Most potential regulatory sites were greater than 60 kB away (Table S2). While regulatory sequences can be hundreds of kBs away from the target gene, it remains possible that these genes may also be regulated indirectly by other Pax3:Foxo1a target genes or miRNAs.
As a test of checkpoint adaptation and the permissiveness of aRMS cells to transit from G2 to mitosis despite single- and double-stranded DNA damage, we irradiated tumor cells with or without Pax3:Foxo1a knockdown. Radiation resulted in a higher fraction of DNA breaks amongst mitotic cells (as represented by dual pHH3 positive, H2AX positive cells) under conditions of Pax3:Foxo1a expression than its knockdown (Figure 5A and Figure S2A), suggesting that Pax3:Foxo1a does facilitate G2 to M transition, consistent with checkpoint adaptation. Moreover, we performed cell cycle and Annexin V apoptosis detection assay after treatment with 10 Gy radiation for two independent eYFP shRNA knockdown clones compared to two other independent shRNA controls (as stated early, eYFP knockdown also achieves Pax3:Foxo1a knockdown) (Figure S2). Cell cycle analysis of the shRNA clones treated with radiation revealed increasing percentage of cells in cells having ≥4N DNA content after radiation for Pax3:Foxo1a knockdown cells compared to radiated controls (p<0.05)(Figure 5B). This result is consistent with a role of Pax3:Foxo1a in overcoming G2 arrest or M checkpoint arrest after radiation. Similarly, the Annexin V apoptosis detection assay showed a lower induction of apoptosis following radiation when Pax3:Foxo1a expression was preserved in shControl clones than shYFP cells (Figure 5C).
To test the acute role of Pax3:Foxo1a in tolerization to treatment-related DNA damage in vivo, we used eYFP siRNA to transiently knock down Pax3:Foxo1a in aRMS tumor cells treated with radiation versus non-irradiated controls that were then orthotopically injected into unirradiated host mice. Pax3:Foxo1a mediated a cell survival and tumor re-establishment advantage under the stress condition of irradiation, but not under homeostatic conditions (p = 0.02, Figure 6A and 6B).
To assess the extent to which the fusion gene mediates refractoriness to chemotherapy agents, we observed Pax3:Foxo1a to facilitate 2–4 fold refractoriness to clinical agents capable of causing double-stranded DNA breaks and mitotic arrest (vincristine, actinomycin-D, topotecan) more so than agent inducing single-strand breaks (mafosfamide, the active metabolite of cyclophosphamide) (Figure S3A–E). That a similar role of Pax3:Foxo1a may apply to targeted agents was previously suggested by enriched G2 expression of Pdgfra (Figure 2F) and then demonstrated by increased sensitivity to prototypic Pdgfr inhibitor, imatinib, after Pax3:Foxo1a knockdown (Figure S3F). Similarly, Pax3:Foxo1a knockdown sensitized tumor cells to siRNA inhibition of downstream signaling mediators of acquired imatinib resistance (Figure S3G) [17]. Thus, these in vitro and in vivo results are consistent with a function of Pax3:Foxo1a in mediating checkpoint adaptation and refractoriness to the established clinical therapies of radiation and chemotherapy, or more contemporary molecularly-targeted agents.
A key finding of this study is that Pax3:Foxo1a expression is dynamic and varies during the cell cycle. To our knowledge this is first report of a translocation-mediated chimeric transcription factor oncogene that is expressed in a cell cycle-specific manner – much less, one that is expressed specifically during G2. The master transcription factor MYOD is expressed strongly during G1 [18] but is inactivated by phosphorylation during mitosis, which results in deportation from the nucleus [19]. MYF5 is also expressed in a cell cycle-dependent manner, but neither MYOD nor MYF5 expression is increased during G2/M as observed in our study of Pax3:Foxo1a in aRMS. Our findings reveal that Pax3 expression in wildtype C2C12 myoblasts is dynamic and increased during G2/M, but that to account for the dramatic increase in Pax3:Foxo1a expression an additional enhancer effect of Foxo1a 3′ region DNA is likely to be present. This result opens the possibility that co-factors assembled at the Pax3 promoter or fusion gene specific cis-elements might be targeted to suppress Pax3:Foxo1a expression.
Cell cycle progression after DNA damage is regulated by checkpoint controls, which prevent continued transit through the cycle until the damage has been repaired, hence protecting the integrity of the genome. Arrest in G1 permits repair prior to replication, whereas arrest in G2 allows repair prior to mitotic chromosome segregation. The p53 tumor suppressor, which is mutated in roughly half of human aRMS, has been shown to be integral to both G1 and G2 damage checkpoint machinery, but some reports found p53 dispensable for the G2 checkpoint [13], [20].
Checkpoint adaptation is defined as the ability to divide and survive following a sustained checkpoint arrest despite the presence of unrepairable DNA breaks [14]. Cells undergoing checkpoint adaptation will frequently die in subsequent cell cycles if DNA damage goes unrepaired, yet, some cells may be able to survive and proliferate in an aneuploid state [14]. Furthermore, in unicellular eukaryotes and tumor cells, DNA repair can occur at G1 [21]. Here, we reveal that the G2/M adaptation genes (H2afx, Cdc25b and Plk1) were suppressed by Pax3:Foxo1a knockdown in G2 and M cell cycle phases and that fewer cells transited from G2 to M without initiating apoptosis under conditions of Pax3:Foxo1a knockdown in the context of radiation-induced stress. These results suggested that not only cell cycle dependent expression but also a clinically-relevant biology underlying Pax3:Foxo1a expression at the G2-M checkpoint, a critical cell cycle checkpoint following radiation or DNA double strand break inducing-chemotherapy.
That a myogenic cancer might utilize genomic instability, aneuploidy or multinucleation as a mechanism of cell survival or tumor cell evolution/progression may not be so unexpected, in retrospect. Normal myofibers are typically multi-nuclear by definition, and genetic conditions predisposing to mitotic disjunction such as Mosaic Variegated Aneuploidy (MVA) are strongly associated with the development of RMS [22]. Both aRMS and eRMS have also been documented to be hyperdiploid, tetraploid, polyploid or to even have mixed aneuploid populations [23], [24], [25]. At a cellular level, the heterogeneity of cells in rhabdomyosarcoma is notable for the subpopulation of multi-nucleated rhabdomyoblasts which appears with giant nuclei or as multi-nucleated giant cells, often with cross-striations – yet highly mitotic [26]. These rhabdomyoblasts might be compared to the multinucleated stemloid cells in fibrosarcoma, which have a tumor-repopulating ability [27]. Our recent study of aRMS and the PKC iota inhibitor, aurothiomalate, reveals that aRMS cells have a remarkable tolerance to polyploidy, which induces neither apoptosis or senescence [28]. This intrinsic capacity to tolerate aneuploidy as well as this report's observed Pax3:Foxo1a-mediated increase in checkpoint adaptation gene expression may be directly relevant to clinical care, given that decreased expression of these same factors (i.e., PLK1, CCCNB1, BIRC5, AURKB) have been reported to improve sensitivity to mitotic inhibitors [29]. Therefore, the interest generated from chemical screens identifying PLK1 as a potential therapeutic target in RMS [30] is likely warranted.
When considering the differences in treatment-related outcomes in RMS subtypes, the role of Pax3:Fox01a in checkpoint adaptation may be our most important clue yet as to how to improve outcome for fusion positive patients: while aRMS are certainly sensitive to standard chemotherapy and radiation, it is the survival of resistant clones which is the cause of disease progression and relapse – which occur to a greater extent in Pax:Foxo1a positive aRMS than fusion negative aRMS or eRMS [31], [32], and which we believe to be a result of Pax3:Foxo1a-mediated checkpoint adaptation. These effects on tumor cell sensitivity to radiation, chemotherapy and targeted therapeutics are likely to be cumulative and possibly critically important in defining the otherwise very narrow therapeutic window for fusion positive aRMS, for which the toxicity of chemotherapy and radiation is now dose-limiting [33].
Perhaps the most interesting aspect of this genetically-engineered conditional mouse model of a deadly but rare childhood cancer is that a labor-intensive knock-in approach to modeling the molecular pathophysiology of a fusion gene was beneficial. Successful transgenic tumor models have been generated by constitutive, ectopic expression of translocation-related fusion oncogenes for synovial sarcoma [34] as well as other “driver” oncogene related tumors [35]; similarly, retroviral transfection of oncogenes into hematopoietic cells has enabled this study of translocation-associated leukemia for many years [36], [37]. However, are these systems driven by non-native or partial-native promoters to be the definitive preclinical platforms for interrogating molecular physiology – or are distal native cis- and trans-regulation temporally critical? Every experimental system has its advantages and limitations, yet for cell and animal models where translocation-mediated fusion genes have yet to be modeled at the native promoter, we may have an entirely new spectrum of cancer genetics to explore.
All animal procedures were conducted in accordance with the Guidelines for the Care and Use of Laboratory Animals and were approved by the Institutional Animal Care and Use Committee (IACUC) at the University of Oregon Health & Science University (OHSU) or the Joslin Diabetes Center (Boston, MA). Every effort was made to minimize suffering.
The Myf6Cre,Pax3:Foxo1a,p53 conditional aRMS mouse model has been described previously [5], [6], [7], is described as caMOD Model 150064393, and is publically available through the NCI MMHCC Repository (MMHCC Strain Codes 01XBL B6; 129-Myf6<tm2(Cre)Mrc> and 01XBM B6; 129-Pax3<tm1Mrc>). SHO-PrkdcscidHrhr mice were purchased from Charles River Laboratories (Wilmington, MA) and bred/maintained at OHSU.
Mouse primary cell cultures (U23674, U42369, U57844) were established from tumor samples. Tumors were minced into small pieces and digested with collagenase (10 mg/ml) overnight at 37°C. The dissociated cells were then incubated in Dulbecco's modified eagle's media supplemented with 10% fetal bovine serum (FBS) and 1% penicillin-streptomycin in 5% CO2 at 37°C. C2C12 mouse myoblast cells were purchased from ATCC (Manassas, VA). Human aRMS cell lines were a gift from Peter Houghton (Rh3; Nationwide Children's Hospital, Columbus, OH) or Patrick Reynolds (Rh41; COG Cell Culture and Xenograft Repository). These cells lines were maintained in the same culture conditions as primary tumor cell cultures: DMEM supplemented with 10% Fetal Bovine Serum (FBS) and 1% Penicillin-Streptomycin. All primary cell cultures experiments using cells were carried out at passage 3–7.
For immunofluorescence staining of frozen sections, the polyclonal antibody for green fluorescent protein (1∶1000, AB16901, Chemicon) was used with DAPI counterstain.
siRNA transfections were carried out using Lipofectamine2000 (Invitrogen, Grand Island, NY) according to manufacturer's recommended protocol. siRNA's were diluted between 0.1 and 10 nM, and the final concentration of Lipofectamine2000 was 0.2%. siYFP Stealth RNAi siRNA Reporter Controls (cat. 12935-145; Invitrogen) were used as the eYFP siRNA to knockdown the Pax3:Foxo1a-ires-eYFP bi-cistronic mRNA, whereas Stealth RNAi siRNA Negative Control Med GC #3 (cat. 12935-113; Invitrogen) was used as the siRNA control (siCont).
To establish shRNA knockdown clones of primary tumor cell cultures, we used MISSION pLKO.1-puro eGFP shRNA Control Transduction Particles (cat. SHC005V; Sigma Aldrich) for Pax3:Foxo1a knockdown and MISSION pLKO.1-puro Non-Mammalian shRNA Control Transduction Particles (cat. SHC002V; Sigma Aldrich) as the control, respectively. shRNA transfections and clonal selection were carried out according to manufacturer's recommended procedures. Mouse RMS primary cell cultures were plated at 1.8×106 cells per 150 mm dish. After 24 h, hexadimethrine bromide was added (8 µg/ml, cat. H9268; Sigma Aldrich), followed by each particle solution (MOI 0.5). After another 24 h, media were removed and fresh media were added. The following day, puromycin was added (5 µg/ml, cat. P8833; Sigma Aldrich). Puromycin-resistant clones were selected cloning rings at day 14 (shControl) and day 17 (shYFP), with continuous puromycin selection at all times.
Cells were irradiated on a Trilogy linear accelerator (Varian, Palo Alto, CA) with a 10×10 cm AP field. Two centimeter of bolus material was placed on top of the 2 chamber slide or 6 cm dish and the target surface distance to the bolus was at 97 cm. Monitor units on the linear accelerator were then set to deliver 6 Gy or 10 Gy of dose to the cells.
Tumors were lysed in radioimmunoprecipitation assay (RIPA) buffer or NP40 buffer containing both protease and phosphatase inhibitor (Sigma). The lysates were homogenized and centrifuged at 8000 g for 10 minutes. The resulting supernatants were used for immunoblot analysis. Goat anti-FOXO1A antibody (cat. Sc-9808; Santa Cruz, Santa Cruz, CA), goat anti-GFP antibody (cat. 600-101-215, Rockland; Gilbertsville, PA) or rabbit anti-PDGFRa antibody (cat. #3164; Cell signaling Technology, Danvers, MA).
Cells were plated on 8-well CultureSlides (cat. 354118; BD Falcon, Franklin Lakes, NJ), fixed with 4% paraformaldehyde, permeabilized with 0.1% or 0.25% TritonX100, washed and incubated with mouse monoclonal anti-skeletal myosin (FAST) (cat. M4276; Sigma), rabbit anti-Ki67 (cat. RM-9106-F; Thermo Scientific, Waltham, MA), mouse anti-Pax3 (cat. MAB2457; R&D Systems), mouse anti-phospho Histone H3 (cat. #9706; Cell Signaling Technology), rabbit anti-phospho Histone H3 (cat. #3377; Cell Signaling Technology), mouse anti-phospho Histone H3 (cat. #9706; Cell Signaling Technology), rabbit anti-CDC2-Y15 (cat. #4539; Cell Signaling Technology) or rabbit anti-phospho H2AX antibody (cat. #9718; Cell Signaling Technology), overnight, rinsed with PBS, incubated with fluorescein isothiocyanate-conjugated anti-mouse and rabbit IgG (1∶200) for 1 h, and examined by confocal microscopy with a Zeiss LSM700 instrument. For immunocytochemistry experiments, at least 100 positive cells were scored per specimen.
Cells were suspended in Hank's balanced salt solution (HBSS) with 2% FBS and 2 mM EDTA. Antibody staining was performed for 20 minutes on ice. Prior to FACS sorting, cells were suspended in 1 µg/ml propidium iodide (Pi) and 10 µM calcein blue (Invitrogen) to identify viable cells (Pi−Ca+). Purity checks were performed to confirm that the sorted eYFP+ and eYFP- cell subsets had a purity of >98% using a eYFP expression threshold determined by the background fluorescence of eYFP- C2C12 cells. The following antibodies were used to evaluate receptor tyrosine kinase surface expression: APC-conjugated Pdgfrα antibody (#17-1401-81, eBiosciences) or anti-IGF1 Receptor antibody (cat. Ab32823; Abcam, Cambridge, MA; 1 in 25).
Mouse RMS primary cell cultures were trypsinized and incubated with Hoechst33342 (final concentration 15 µg/ml) and Reserpine (final concentration 5 µM). Cells were incubated in the dark for 30 min at 37°C, and analyzed and sorted by flow cytometry using an Influx FACS instrument (Becton Dickinson, Franklin Lakes, NJ). Cell cycle was determined with the FlowJo software (Tree Star, Inc., Ashland, OR).
Mouse primary cell cultures were stained with Annexin V and Propidium iodide using Annexin-V-FLUOS Staining Kit (cat. 11 858 777 001; Roche) following the protocol provided by the manufacturer. Briefly, 48 hour after irradiation, 106 mouse primary cell cultures were trypsinized, washed by PBS and resuspended in 100 µl of Annexin-VFLUOS labeling solution, incubated 10–15 min at 15–25°C, and analyzed by FACS Calibur.
U23674 cells were subfractionated by FACS sorting as described above. mRNA was isolated using RNeasy spin columns (Qiagen, Valencia, CA) and reverse transcribed using Superscript III First-Strand Synthesis System for RT-PCR (Invitrogen). QPCR was performed using an AV7900 PCR system (Applied Biosystem) with SYBR-green PCR reagents. Pax3:Foxo1a was detected using the following primer sequences: 5′-AGACAGCTTTGTGCCTCCAT-3′ and 5′-CTCTTGCCTCCCTCTGGATT-3′. Other primers are Taqman Gene Expression assay, H2afx (Mm00515990_s1), Cdc25b (Mm00499136_m1), Birc5 (Mm00599749_m1), Plk1 (Mm00440924_g1) and Gapdh (Mm99999915_g1) by Invitrogen. RT-PCR data were quantified using the standard curve method, and relative expression of Pax3:Foxo1a per sample was determined by normalization against the quantity of 18 s rRNA and Gapdh within each sample. For each sample, QPCR was performed in technical duplicates and results were averaged.
Mouse RMS primary cell cultures were plated at 1×103 cells of each cohort per well in a 96-well plate. After cell incubations, cytotoxic effects were assayed using CellTiter 96 AQueous One Solution Cell Proliferation Assay system (Promega, Madison, WI) and SpectraMax M5 luminometer (Molecular Devices, Sunnyvale, CA). IC50 and C.I. were determined with CalcuSyn software (BIOSOFT, United Kingdom).Drugs: Vincristine sulfate salt (cat. V8879; Sigma), Actinomycin-D (cat. A9415; Sigma), Mafosfamide (cat. sc-211761; Santa Cruz), Topotecan hydrochloride (cat. S1231; Selleck), Eribulin mesylate (NDC 62856-389-01; Eisai) or Imatinib Mesylate (cat. S1026; Selleck).
For these studies, individual siRNA were obtained from Dharmacon (Lafayette, CO), including the mouse siRNA library targeting the tyrosine kinome (siGENOME). These experiments are performed at 100 nM concentration and include non-specific pooled siRNA as a control purchased from Dharmacon. Transfection of siRNA was carried out using Lipofectamine 2000 in Opti-MEM Reduced Serum Media (Invitrogen). After cells were plated in 96-well plates in the presence of inhibitor or siRNA, and incubated for 96 hours, respectively, 20 µL CellTiter 96 AQueous One solution (MTS) was added to each well and absorbance values assessed by the BioTek Synergy 2 plate reader (BioTek, Winooski, VT).
Labeled target cRNA was prepared from 12 mouse total RNA samples (3 independent experiments×4 samples). Samples were amplified and labeled using the Ambion MessageAmp Premier RNA Amplification Kit following the manufacturer's protocol. Sample order was randomized. Each sample target was hybridized to an Illumina MouseRef 8 v 2 Expression BeadChip Array. Image processing and expression analysis were performed using Illumina BeadArray Reader and GenomeStudio (v. 2010.1) Gene Expression module (v. 1.6.0) software. Microarray data have been accessioned with the Gene Expression Omnibus (GEO) under series GSE41675. The following link has been created to allow review of record GSE41675 while it remains in in review/under private status: http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?token=xdajbqeisomcyhq&acc=GSE41675.
aRMS primary cultures (passage 5) were plated in 6 cm dishes. The next day cells were transfected with siYFP Stealth RNAi siRNA Reporter Controls or Stealth RNAi siRNA Negative Control Med GC #3. Two days later cells were irradiated on a Trilogy linear accelerator with a 10×10 cm AP field with two centimeter of bolus material was placed on top of the 6 cm dish. The target surface distance to the bolus was at 97 cm and monitor units on the linear accelerator were then set to deliver 10 Gy of dose to the cells. Subsequently, cells were trypsinized and 500,000 cells were injected into the gastrocnemius muscle of SHO mice that had been pre-injured 24 hours prior with 0.85 µg/mouse cardiotoxin intramuscularly. Tumor volumes (cm3) were measured 3-dimensionally with electronic calipers and calculated from formula (π/6)×length×width×height, assuming tumors to be spheroid. For statistical analysis of disease-free survival, a tumor volume threshold of 0.25 cc was applied. The log-rank test was used to contrast treatments. All analyses were performed using R 3.0.0 (The R Foundation for Statistical Computing, Vienna, Austria).
|
10.1371/journal.ppat.1006518 | Hijacking of the O-GlcNAcZYME complex by the HTLV-1 Tax oncoprotein facilitates viral transcription | The viral Tax oncoprotein plays a key role in both Human T-cell lymphotropic virus type 1 (HTLV-1)-replication and HTLV-1-associated pathologies, notably adult T-cell leukemia. Tax governs the transcription from the viral 5’LTR, enhancing thereby its own expression, via the recruitment of dimers of phosphorylated CREB to cAMP-response elements located within the U3 region (vCRE). In addition to phosphorylation, CREB is also the target of O-GlcNAcylation, another reversible post-translational modification involved in a wide range of diseases, including cancers. O-GlcNAcylation consists in the addition of O-linked-N-acetylglucosamine (O-GlcNAc) on Serine or Threonine residues, a process controlled by two enzymes: O-GlcNAc transferase (OGT), which transfers O-GlcNAc on proteins, and O-GlcNAcase (OGA), which removes it. In this study, we investigated the status of O-GlcNAcylation enzymes in HTLV-1-transformed T cells. We found that OGA mRNA and protein expression levels are increased in HTLV-1-transformed T cells as compared to control T cell lines while OGT expression is unchanged. However, higher OGA production coincides with a reduction in OGA specific activity, showing that HTLV-1-transformed T cells produce high level of a less active form of OGA. Introducing Tax into HEK-293T cells or Tax-negative HTLV-1-transformed TL-om1 T cells is sufficient to inhibit OGA activity and increase total O-GlcNAcylation, without any change in OGT activity. Furthermore, Tax interacts with the OGT/OGA complex and inhibits the activity of OGT-bound OGA. Pharmacological inhibition of OGA increases CREB O-GlcNAcylation as well as HTLV-1-LTR transactivation by Tax and CREB recruitment to the LTR. Moreover, overexpression of wild-type CREB but not a CREB protein mutated on a previously described O-GlcNAcylation site enhances Tax-mediated LTR transactivation. Finally, both OGT and OGA are recruited to the LTR. These findings reveal the interplay between Tax and the O-GlcNAcylation pathway and identify new key molecular actors involved in the assembly of the Tax-dependent transactivation complex.
| Human T-cell lymphotropic virus type 1 (HTLV-1) is the only human retrovirus associated to a cancer. Indeed, HTLV-1 is responsible for adult T-cell leukemia, an aggressive malignant proliferation of CD4+ T lymphocytes. The regulatory protein Tax governs HTLV-1 transcription from the 5’LTR, driving expression of all viral proteins, including itself, at the exception of the antisense product HBZ. Besides this critical role in HTLV-1 expression, Tax acts as an oncoprotein able to induce T-cell immortalization in vitro and tumor formation in mice. In this study, we report that Tax interacts with the O-GlcNAczyme OGT/OGA complex that catalyzes O-GlcNAcylation, a post-translational modification often deregulated in cancers. We found that Tax interacts with the OGT/OGA complex and inhibits the activity of OGA, increasing thereby cellular O-GlcNAcylation. Strikingly, we found that O-GlcNAcylation of CREB, the cellular transcription factor recruited by Tax on the viral promoter, is increased in a Tax-dependent manner. Moreover, increased CREB O-GlcNAcylation strongly enhances Tax-induced LTR transactivation as well as CREB binding to the viral promoter. Finally, both OGT and OGA are part of the transactivation complex. These findings shed new light on the mechanism of Tax-dependent LTR transactivation and may open the way to new molecular interventions targeting HTLV-1 expression.
| Human T-lymphotropic virus type 1 (HTLV-1) is the only retrovirus associated to a cancer in humans. HTLV-1 is indeed the etiologic agent of adult T-cell leukemia/lymphoma (ATLL), a very aggressive malignant proliferation of CD4+ T lymphocytes, which appears in 2–5% of infected individuals (reviewed in [1]). In addition, HTLV-1 is also associated with various inflammatory disorders, including HTLV-1-associated myelopathy/tropical spastic paraparesis (HAM/TSP) [2].
The oncogenic power of HTLV-1 is due in large part to the properties of the viral oncoprotein Tax. Tax is a powerful inducer of T-cell proliferation through its ability to activate a broad range of cellular promoters, promote cell cycle and inhibit apoptosis and repair machineries (reviewed in [3]). As a consequence, Tax has been shown to induce immortalization of primary T cells in vitro [4] as well as tumor formation in transgenic animals [5]. Tax is also critical for HTLV-1 gene expression by virtue of its capacity to transactivate the 5’ LTR that controls the transcription of all HTLV-1 structural, enzymatic and regulatory genes, including Tax itself, and auxiliary genes with the exception of the antisense product HBZ [6].
The transactivation of the 5’LTR depends on Tax interaction with the cellular transcription factor cAMP response element binding protein (CREB) that, together with Tax, binds to three conserved copies of a cyclic AMP-response element (CRE) located in the LTR U3 region (viral CRE/vCRE). CREB-mediated activation of cellular promoters has been shown to critically depend on CREB phosphorylation at Ser133 [7, 8]. It was initially proposed that CREB phosphorylation was dispensable in the context of Tax transactivation of the HTLV-1 promoter [9, 10]. However, further studies demonstrated on the one hand that the transactivation complex contains Ser133-phosphorylated CREB and on the other hand, that Tax is able to increase CREB phosphorylation [11–13]. The binding of Tax/CREB complexes to the vCRE then allows the recruitment of the CREB-Regulated Transcription Coactivator/Transducer Of Regulated CREB-Binding Protein (CRTC/TORC) [14], the CREB binding protein (CBP) [15] and CBP-associated factor (p/CAF) [16] general co-activators and ultimately, of components of the basal transcription machinery (reviewed in [17]).
O-GlcNAcylation is a reversible post-translational modification [18] that has been shown to regulate stability, sub-cellular localisation and/or activity of a large set of proteins, notably transcription factors or co-factors [19], including CREB [20–22]. O-GlcNAcylation consists in the addition of N-acetyl glucosamine (GlcNAc) on Serine and Threonine residues. Only a unique couple of enzymes controls O-GlcNAc level on proteins: OGT (O-GlcNAc transferase), which adds the GlcNAc motif on proteins, and OGA (O-GlcNAcase), which removes it [19]. OGT and OGA are known to be physically associated in a molecular complex (the O-GlcNAczyme complex), and this association was shown previously to be important for their regulatory activity on cell signaling and transcriptional processes [23]. Numerous studies have reported alterations in OGT, OGA and O-GlcNAc levels in solid tumors as well as hematopoietic cancers [24]. O-GlcNAcylation may promote tumor development through perturbation of signalling pathways and cell cycle regulators [24, 25]. In addition, major oncogenic factors were shown to be directly O-GlcNAcylated [24, 25]. Finally, O-GlcNAcylation has been recently recognized as a novel epigenetic mark (reviewed in [26]).
O-GlcNAcylation of CREB was initially described in rat brain [20]. Serine 40 of CREB was identified as a major O-GlcNAcylation site and found to function as a negative signal by preventing CREB association with CRTC/TORC [21]. CREB can be simultaneously O-GlcNAcylated at Ser40 and phosphorylated at Ser133 and indeed, CREB O-GlcNAcylation was shown to preferentially occur on the population of Ser133-phosphorylated CREB [21, 22].
In this study, we explore for the first time the status of O-GlcNAcylation in HTLV-1-transformed T cells. By using a combination of BRET, enzymatic and biochemical assays, we report that the HTLV-1 Tax protein binds to the O-GlcNAczyme complex, blocks the activity of OGA and increases total O-GlcNAcylation in both adherent cells and HTLV-1-transformed T cells. Moreover, we show on the one hand that Tax increases CREB O-GlcNAcylation and on the other hand that increasing O-GlcNAcylation through OGA inhibition enhances both Tax-induced LTR transactivation and CREB recruitment to the promoter. We also report that in contrast to wild-type CREB, the CREB S40A mutant fails to enhance Tax-mediated LTR transactivation. Finally, we show that both OGT and OGA are recruited to the HTLV-1 LTR. These findings identify new functional interacting partners of Tax and shed new light on the composition of the transactivation complex assembled by Tax on the HTLV-1 5’ LTR promoter.
To determine the status of O-GlcNAcylation in T cells upon HTLV-1-induced transformation, the levels of OGT and OGA were compared between T cells transformed or not by HTLV-1. Greater level of OGA mRNA was found in four HTLV-1-transformed T cell lines, compared to four non-HTLV-1 transformed T cells (Fig 1A), whereas OGT mRNA expression was not affected (Fig 1B). To determine whether increased OGA mRNA expression could be related to the activated phenotype of HTLV-1 transformed T cells, we evaluated the levels of OGA and OGT mRNA upon T-cell activation. In contrast to HTLV-1-induced transformation, activation of peripheral blood mononuclear cells with PHA and IL-2 strongly reduced the level of OGA mRNA, while increasing OGT mRNA expression (S1 Fig). Hence, HTLV-1-induced T-cell transformation and T-cell activation differentially modulate OGA and OGT mRNA expression. As shown in Fig 1C, western blot analysis confirmed increased OGA protein expression with no change in OGT protein expression in HTLV-1-transformed compared to non-HTLV-1 transformed T cells.
The enzymatic activity of OGA in each T cell line was then quantified. Cells were lysed and equal amounts of total proteins were used to measure either total OGA enzymatic activity or OGA protein level (Fig 1D). A statistically significant increase in total OGA activity (p = 0.0317) was found in HTLV-1-transformed T cells as compared to control transformed T cells (Fig 1D, left panel). Because OGA protein expression level was higher in HTLV-1-transformed T cells, OGA activities were normalized to the amount of OGA protein present in each assay, determined by quantification of the signal obtained by western-blotting using the same cell extracts (Fig 1D, middle panel). To validate this procedure, we verified that a linear relationship exists between OGA activity and the OGA signal obtained by western-blot (S2 Fig). When corrected for OGA expression levels, OGA specific activity was much lower (p = 0.0159) in HTLV-1-transformed T-cells than in control transformed T cells (Fig 1D, right panel).
These findings show that OGT and OGA expression levels are differentially affected by HTLV-1 transformation. They also show that OGA production is increased at both mRNA and protein levels in HTLV-1-transformed T cells but that the activity of the enzyme is impaired in these cells.
The HTLV-1 Tax protein is capable of interacting with and deregulating numerous cellular proteins and machineries [3]. We therefore evaluated the impact of Tax on OGA activity using a Tax-negative HTLV-1 transformed T cell line (TL-om1), which allowed us to study Tax activity in an HTLV-1-transformed T cell context. A Tax expressor plasmid was transfected into TL-om1 T cells and OGA activity was measured 24 hours post-transfection. We observed that Tax-expressing TL-om1 T cells exhibited lower OGA activity than TL-om1 T cells transfected with the control plasmid (Fig 2A). This reduction in OGA activity was not due to a change in OGA expression level (Fig 2A, insert).
In order to determine whether Tax-induced OGA inhibition was associated with a change in O-GlcNAcylation, we developed a BRET biosensor based on a previously described FRET O-GlcNAc biosensor (Fig 2B)[27]. This BRET O-GlcNAc biosensor is composed of Rluc8 fused to a lectin domain (GafD), a known OGT substrate peptide derived from casein kinase II, followed by the Venus variant of the yellow fluorescent protein. Upon O-GlcNAcylation, the casein kinase peptide binds to the lectin, resulting into a conformational change detected as an increased BRET signal (Fig 2B). We observed higher BRET signal in TL-om1 cells expressing Tax compared to control cells (Fig 2C, left panel and statistical analysis of Tax-induced delta BRET in middle panel). This result was confirmed by western-blotting with an anti-O-GlcNAc antibody, which showed increased O-GlcNAcylation of proteins in Tax-transfected cells (Fig 2C, right panel).
Since TL-om1 T cells still express the viral antisense product HBZ, we investigated the effect of Tax in an HTLV-1-independent context. In transfected HEK-293T cells, Tax expression also resulted in a marked reduction in OGA enzymatic activity, as compared to control cells (Fig 2D). Again, this effect was not due to a change in OGA expression level (Fig 2D, insert). In contrast to OGA, OGT activity was not affected by Tax expression (S3 Fig). Inhibition of OGA activity coincided with a significant increase in the BRET signal of the biosensor (Fig 2E left panel and statistical analysis of Tax-induced delta BRET in middle panel). An increase in O-GlcNAcylation level of HEK-293T cell proteins was also detected by western-blotting using the anti-O-GlcNAc antibody (Fig 2E, right panel).
These results suggest that Tax inhibits OGA activity independently of the HTLV-1 context, and that this inhibition results in increased cellular O-GlcNAcylation.
OGT and OGA have been previously shown to form a molecular complex, referred to as the O-GlcNAczyme, which plays an important role in their biological functions [23]. To determine whether Tax may alter O-GlcNAcylation by interacting with this complex, we first evaluated by BRET the interaction of Tax with either OGA or OGT. HEK-293T cells were transfected with a cDNA coding for a luciferase-tagged Tax (Rluc8-Tax) together with YPET-OGT, YFP-OGA, or YFP alone. Western blot analysis showed correct expression of each of these fusion proteins at their expected molecular weights (S4 Fig). A much higher BRET signal was observed with YPET-OGT or YFP-OGA than with YFP, indicating a specific interaction of Tax with the O-GlcNAc cycling enzymes (Fig 3A).
We then studied the effect of Tax expression on the formation of the OGT/OGA complex by BRET in HEK-293T cells co-transfected with OGT-Rluc and OGA-YFP constructs. As shown in Fig 3B, such complex could be readily detected as a BRET signal between OGT-Rluc and OGA-YFP. A higher BRET signal was found upon Tax expression, suggesting that Tax modulates OGT/OGA interaction (Fig 3B and statistical analysis of Tax-induced delta BRET in the insert). To further analyze the effect of Tax on OGT/OGA interaction, BRET saturation assays were performed (Fig 3C). This analysis permits to determine whether a change in BRET signal between two partners corresponds to an increased affinity between the two partners (reflected by decreased BRET50) [28] or, rather, a conformational change within the complex that modifies the relative orientation between the luciferase and the YFP, resulting in a higher efficiency of energy transfer, without change in BRET50 [29]. Analysis of the saturation curves using Prisme software indicated that Tax expression reduces the BRET50 (Fig 3C left panel and statistical analysis of BRET50 in the right panel), suggesting that Tax may regulate O-GlcNAcylation by increasing the affinity between OGA and OGT.
We next measured the enzymatic OGA activity in the OGT/OGA complex after immunoprecipitation of OGT. HEK-293T cells were co-transfected with OGT-Luc, OGA-YFP and either the Tax or control plasmid. OGA activity was measured on the immune complex and normalized to YFP fluorescence of the precipitated proteins. We found that Tax significantly reduced the activity of OGA co-immunoprecipitated with OGT (Fig 3D).
Taken together, these data support the notion that Tax regulates O-GlcNAcylation by modulating OGT/OGA interaction, resulting in inhibition of OGA activity in the O-GlcNAczyme complex.
We next studied the impact of increasing O-GlcNAcylation by using the specific OGA inhibitor Thiamet G on the activity of the HTLV-1 LTR in Tax expressing cells. C8166 T cells transfected with the HTLV-1-U3R-Firefly Luciferase construct (U3R-Luc) and the pRL-TK normalisation plasmid were cultured for 2 days with or without Thiamet G. As shown in Fig 4A, increased protein O-GlcNAcylation induced by Thiamet G (right panel) was associated with a significant increased activity of the U3R-Luc reporter construct compared to untreated cells (left panel), while comparable amount of Tax was produced in each condition (right panel). Importantly, similar results were obtained in HEK-293T cells transfected with the Tax plasmid (Fig 4B).
Hence, enhancing O-GlcNAcylation by pharmacological inhibition of OGA, to mimic the effect of Tax on OGA activity, significantly increases Tax-mediated LTR transactivation.
Tax activates the viral LTR via the recruitment of CREB, which has been shown previously to be modified by O-GlcNAcylation [20–22]. This raises the hypothesis that the higher level of LTR transactivation upon OGA inhibition was linked to higher O-GlcNAcylation of CREB. To investigate this point, the impact of Tax on CREB O-GlcNAcylation was studied using capture on wheat germ agarose (WGA), as previously described [30]. HEK-293T cells were transfected or not with the Tax plasmid and were also treated or not with Thiamet G. Two-days after transfection, cells were lysed and same amounts of total proteins were incubated with WGA. O-GlcNAcylated proteins captured on WGA were analyzed by western blot using an anti-O-GlcNAc antibody. As expected, Thiamet G treatment dramatically increased the amount of WGA-bound O-GlcNAcylated proteins (Fig 4C left panel, compare lanes 1 and 3 and quantification on middle panel). Expression of Tax in HEK-293T cells also increased the binding of O-GlcNAcylated protein on WGA (Fig 4C, left panel, compare lanes 1 and 2), albeit at a much lower level than in cells treated with 10 μM Thiamet G. In agreement with this observation, we found that the inhibitory effect of Tax on OGA enzymatic activity in HEK-293T cells corresponds to the inhibitory effect of a much lower concentration of Thiamet G (0.01 μM, S5 Fig). Adding N-acetylglucosamine during incubation of cell lysates with WGA almost completely abolished the anti-O-GlcNAc signal, showing the specificity of the enrichment method (S6 Fig).
Reprobing the membrane with the anti-CREB antibody indicated a massive increase in binding of CREB to WGA upon Thiamet G treatment, confirming CREB as a target of O-GlcNAcylation (Fig 4C, left panel, compare lanes 1 and 3 and quantification on right panel). Expression of Tax also significantly increased CREB retention on WGA, as demonstrated by the higher WGA/lysate ratio for CREB (Fig 4C left panel, compare lanes 1 and 2 and quantification on right panel) suggesting that Tax may induce CREB O-GlcNAcylation. In contrast, Tax was detected in the lysates but not among WGA-bound proteins, neither in absence or presence of Thiamet G. This suggests that Tax does not induce its own O-GlcNAcylation and is unlikely to be an O-GlcNAcylation target, as it is not retained on WGA even in conditions where a major general increase in protein O-GlcNAcylation is induced by pharmacological inhibition of OGA (S7 Fig).
We also evaluated in TL-om1 cells the effect of Tax on CREB retention on WGA (Fig 4D). Same amounts of total proteins were incubated with WGA. Immunodetection were performed on WGA-bound proteins (WGA) or total proteins (Lysates) with either the anti-O-GlcNAc or anti-CREB antibody. First, we observed that the level of total O-GlcNAcylated proteins retained on WGA was higher in lysates from Tax-transfected than in control cells (Fig 4D, left panel and quantification on middle panel). Importantly, the amount of CREB retained on WGA was also higher in Tax-transfected TL-om1 cells than in control cells (Fig 4D, left panel and quantification on right panel).
These data strongly suggest that Tax expression is sufficient to enhance CREB O-GlcNAcylation both in T cells and adherent HEK-293T cells.
Serine 40 was previously described as a CREB O-GlcNAcylation site [21]. To determine whether Tax induces CREB O-GlcNAcylation on this particular residue, we used YFP-tagged wild-type (wt) and S40A mutant versions of CREB. HEK-293T cells were transfected with Tax and either wt or S40A YFP-tagged CREB and lysed 48 hours post-transfection. After normalization for equivalent amount of YFP-CREB fluorescence, cell lysates were incubated with WGA beads. Western-blotting with the anti-CREB antibody indicated that Tax expression significantly increased the amount of YFP-CREB retained on WGA (Fig 5A). However, residual binding of mutated CREB on WGA suggests that either other O-GlcNAcylation sites still exists on S40A mutant, or that part of this binding occurs through O-GlcNAcylation of some CREB partner.
As a complementary approach, Tax-mediated O-GlcNAcylation of Serine 40 of CREB was analyzed by immunoprecipitation (Fig 5B). Cell lysates from HEK-293T cells transfected or not with Tax and either wt or S40A YFP-tagged CREB were normalized for YFP fluorescence and then immunoprecipitated with an anti-GFP antibody. Western-blotting using the anti-O-GlcNAc antibody revealed that mutation of S40 totally abolished Tax-induced O-GlcNAcylation of CREB, indicating that Serine 40 is indeed the main glycosylation site regulated by Tax. These findings also confirm that WGA binding of CREB mainly depends on O-GlcNAcylation of CREB itself.
We then investigated the effect of expressing either wt or S40A YFP-CREB on Tax-induced LTR transactivation. As expected, transfection of wt YFP-CREB into HEK-293T cells significantly enhanced Tax-mediated transactivation (Fig 5C, left panel). CREB S40A was produced at higher level than wt CREB (Fig 5C, right panel), as previously reported [21]. However, despite this higher expression level, significantly less transactivation was found in cells expressing the S40A mutant than in those producing wt CREB (Fig 5C, left panel).
These findings provide direct evidence that CREB O-GlcNAcylation, especially at Serine 40, is involved in Tax-mediated LTR activation.
Since CREB activity on the HTLV-1 LTR is linked to its recruitment to the vCRE regions, we directly analyzed the impact of increasing O-GlcNAcylation on protein recruitment to the vCRE LTR sequences by chromatin immunoprecipitation (ChIP) experiments. As CREB phosphorylated at Serine 133 was shown to be preferentially recruited to the vCRE, ChIP experiments were performed using an anti-phospho CREB (Ser 133) and primers specific for the distal U3 vCRE sequence. Thiamet G treatment of C8166 T cells dramatically increased the amount of amplified vCRE products as compared to untreated cells (Fig 6A).
Whether the O-GlcNAczyme complex was also recruited to the vCRE region was finally investigated by ChIP in C8166 T cells. Both anti-OGT and anti-OGA ChIP allowed the amplification of vCRE-specific products to levels significantly higher than the control IgG (Fig 6B). Moreover, very low amplification signals were detected with primers targeting alpha-satellite (alpha-sat) regions, showing the specificity of the anti-OGT and anti-OGA ChIPs. Importantly, similar results were obtained with another HTLV-1-transformed T cell line (MT2, Fig 6C) as well as with HTLV-1-immortalized T cells (CIB, Fig 6D).
Hence, CREB recruitment to the LTR is facilitated by O-GlcNAcylation and the OGT/OGA O-GlcNAczyme complex is recruited to the vCRE sequences of the HTLV-1 LTR.
The HTLV-1 Tax oncoprotein is critical for both HTLV-1 expression and HTLV-1-mediated T-cell immortalization. Therefore, the characterization of activators or co-factors responsible for the transactivation of the HTLV-1 5’ LTR is an important issue. In this study, we provide the first demonstration that a novel molecular actor, the O-GlcNAczyme complex, interacts with Tax and is recruited to the LTR as a positive co-factor in both HTLV-1 immortalized and transformed T cells.
We first documented that HTLV-1-transformed T cells express higher level of OGA than control transformed T cells but that this coincides with a dramatic reduction in the specific activity of OGA. Furthermore, expressing only Tax was sufficient to inhibit OGA activity (Fig 2A and 2D) and to increase O-GlcNAcylation of a BRET-based biosensor (Fig 2C and 2E) in both Tax-negative HTLV-1-transformed TL-om1 T cells and HTLV-1-negative HEK-293T cells. This suggests that the ability of Tax to inhibit OGA and increase O-GlcNAcylation is independent of HBZ, as this inhibition is observed in HEK-293T cells which do not express any HTLV-1 protein. However, a potential blocking effect of HBZ on Tax-induced inhibition of OGA activity cannot be ruled-out and should be investigated in future studies. The inhibitory effect of Tax appeared to be specific for OGA activity, as OGT enzymatic activity was not affected by Tax transfection (S3 Fig).
Increased OGA expression associated with impaired OGA activity is not unprecedented. Indeed, previous studies reported that pharmacological inhibition of OGA also leads to OGA accumulation, presumably as the result of a regulatory feedback mechanism compensating for loss of enzymatic activity [31, 32]. We propose therefore that the increased OGA expression found in HTLV-1 transformed T cells is an adaptive response, operating through a yet unknown mechanism, to counteract the inhibition of OGA activity by Tax.
Previous studies have indicated that OGA and OGT associate into a molecular assembly denominated O-GlcNAczyme [23]. Using BRET experiments, we found on the one hand that Tax interacts with both OGT and OGA (Fig 3A), and on the other hand, that Tax expression significantly increases the affinity between OGT and OGA (Fig 3C). This was associated with a significant reduction of OGA enzymatic activity in the complex (Fig 3D). Further experiments are needed to unravel the mechanism by which Tax regulates OGA activity within the O-GlcNAczyme complex, resulting in increased protein O-GlcNAcylation.
Our data also provided evidence that an important consequence of Tax-induced OGA inhibition is the higher O-GlcNAcylation of CREB. Indeed, we showed that expressing Tax in either HEK-293T cells or TL-om1 T cells significantly increased the amount of WGA-bound CREB (Fig 4C and 4D). Strikingly, higher WGA-bound CREB was also found upon pharmacological inhibition of OGA by Thiamet G, and Tax expression did not further increase CREB binding to WGA in Thiamet G-treated cells (Fig 4C). This confirms that Tax effect on O-GlcNAcylation is mediated by inhibition of OGA, as it is not anymore detectable in cells in which OGA activity is maximally inhibited by the drug. O-GlcNAcylation of CREB was confirmed in experiments showing that CREB binding to WGA was markedly reduced by mutation of Serine 40, a previously identified O-GlcNAcylation site on CREB (Fig 5A). Importantly, the effect of Tax on O-GlcNAcylation of CREB was directly shown by immunoprecipitation of CREB followed by western-blotting using anti-O-GlcNAc antibody (Fig 5B). Moreover, these experiments demonstrated that Tax induced O-GlcNAcylation of wt but not of S40A CREB (Fig 5B).
Regarding the functional impact of CREB O-GlcNAcylation, we found that treating cells with the selective OGA inhibitor Thiamet G increased Tax-mediated LTR transactivation both in C8166 T cells and adherent HEK-293T cells (Fig 4A and 4B). Moreover, Thiamet G treatment strongly enhanced the recruitment of Ser133-phosphorylated CREB to the vCRE region of the LTR (Fig 6A). This finding suggests that CREB O-GlcNAcylation and phosphorylation are not mutually exclusive, in agreement with a previous report [21]. Hence, our results indicate that OGA inhibition upon Tax expression or Thiamet G treatment increases CREB O-GlcNAcylation and thereby its activity on the LTR. In agreement with this hypothesis, we showed that Tax-mediated transactivation was enhanced upon expression of wt CREB but not of O-GlcNAcylation-defective S40A CREB mutant (Fig 5C), directly linking CREB O-GlcNAcylation, notably at Serine 40, and Tax-induced LTR activation. Interestingly, O-GlcNAcylation of CREB at Serine 40 was previously shown to block CREB transcriptional activity in neuronal cells by preventing CREB association with CRTC [21]. In contrast, we report here that CREB S40A is impaired for Tax-mediated transactivation, indicative of an activating role of this O-GlcNAcylation site in our model. This suggests that interaction with CRTC required for CREB function in neuronal cells is not a key determinant in the case of Tax-mediated LTR activation. CRTC/TORC was described as a coactivator of Tax-mediated LTR transactivation [14, 33]. However, Siu and collaborators showed that silencing all three CRTC/TORC family members only partially reduced Tax-mediated LTR activation [14], indicating that Tax can still activate the LTR without CRTC/TORC. Moreover, ATF4/CREB2, which does not need CRTC as coactivator [14], is able to activate the LTR in presence of Tax [34, 35]. Hence, depending of the promoter context, CREB O-GlcNAcylation at Serine 40 may mediate either activating or repressive functions. Interestingly, such opposite effect of O-GlcNAcylation has previously been reported for other transcription factors, notably RelA and Sp1 [19, 36–40].
OGT is now considered as an epigenetic regulator by virtue of its capacity to add O-GlcNAcylation on epifactors and histones (reviewed in [26, 41]). Consequently, OGT binding to promoters has been described [42]. We show here that not only OGT but also OGA are recruited to the vCRE region of the HTLV-1 LTR, suggesting the presence of the O-GlcNAczyme complex at the promoter and its involvement in HTLV-1 gene regulation. Our data therefore support a model in which, via the deposition of the O-GlcNAczyme complex onto the vCRE region, Tax facilitates the O-GlcNAcylation of CREB and possibly other transcription factors and co-factors while concomitantly modulating local chromatin architecture. This ultimately increases promoter activation, as documented here by the positive effect of OGA inhibition on the transactivation by Tax of the HTLV-1-U3R-Luc reporter construct. Importantly, recruitment of OGT and OGA to the 5’LTR was found in both HTLV-1-immortalized primary T cell and HTLV-1-transformed T cell lines. This provides the notion that the O-GlcNAczyme complex may play a key role in both HTLV-1-replication in vivo and HTLV-1-induced pathologies.
Only few studies have investigated the impact of O-GlcNAcylation on virus transcription. It has been shown that enhancing O-GlcNAcylation inhibits the expression of human immunodeficiency virus type 1 or herpes virus simplex [43, 44]. In these cases, the effect was linked to the modification of transcriptional regulators, Sp1 and HCF-1 respectively, involved in virus expression. Our data showing that O-GlcNAcylation increases the transactivation of the HTLV-1 LTR provide therefore the first example of a positive impact of the O-GlcNAcylation machinery on viral transcription via the recruitment of the O-GlcNAczyme complex to the viral promoter.
HEK-293T cells (American Type Culture Collection CRL-3216) were grown in Dulbecco’s modified Eagle’s medium supplemented with 10% fetal calf serum (Dutcher, S1402851810) and with 2 mM glutamine, 1 mM pyruvate and antibiotics (Invitrogen) and were transfected using the Fugene 6 reagent (Invitrogen). The non-infected CD4+ T-cell lines Jurkat (kindly provided by Dr. Schwartz, Institut Pasteur, Paris, France), Molt4 (American Type Culture Collection CRL-1582), CEM (American Type Culture Collection CRL-1992) and HUT-78 (American Type Culture Collection TIB-161) and the HTLV-1-transformed CD4+ T-cell lines C8166 and MT2 (NIH AIDS Research and Reference Reagent Program, USA) and TL-om1 (kindly provided by Dr. Harhaj, Johns Hopkins School of Medicine, Baltimore, USA) were grown in RPMI 1640 medium containing 25mM glucose and supplemented as above but with the addition of 20 mM HEPES and 5mL of 100X non-essential aminoacid solution (Invitrogen). TL-om1 T cells were transfected by nucleofection using the cell line nucleofector kit V (Lonza, France) and the program O-017. The HTLV-1-immortalized CIB T cells described in [45] were generated from peripheral blood mononuclear cells of a TSP/HAM patient. These cells were grown in supplemented RPMI medium in the presence of 50U/ml of IL-2 (Roche, France).
The pSG5M empty vector, pSG5M-Tax and pRL-TK plasmids have been described elsewhere [46]. The U3R-Luc construct described in [47] was kindly provided by Dr. A. Kress (Erlangen, Germany). The YFP-CREB wild-type plasmid was kindly provided by Prof. Montminy (La Jolla, USA). YFP-CREB S40A was generated by PCR-mediated mutagenesis using the following primers: Fw: TGCCACATTAGCCCAGGTAgCCATGCCAGCAGCTCATG and Rev: CATGAGCTGCTGGCATGGcTACCTGGGCTAATGTGGCA and the presence of the mutation was verified by sequencing. The pcDNA3Rluc8 plasmid was a kind gift of Prof. Gambhir [48]. The pcDNA3.1 Rluc8-Tax plasmid was generated following PCR amplification of the Tax sequence from the pSG5M-Tax vector using primers creating NheI restriction sites at both extremities of Tax cDNA (forward: GGCGCTAGCCACCATGGCCCACTTCCCAGGG; reverse: GCCGCTAGCTCCGA-CTTCTGTTTCTCGGAAATG). The PCR product was then inserted into the pcDNA3.1 RLuc8 after NheI digestion. YFP-OGA has been described previously [30]. Rluc8-OGT was generated by inserting OGT coding sequence [49] into the pcDNA3.1 Rluc8 vector after HinDIII/ BamHI digestion. YPet-OGT was obtained by insertion of cDNA OGT sequence into YPet-pcDNA3 vector after digestion with EcoRV-Apa1.
To monitor O-GlcNAcylation in living cells, we developed a BRET-biosensor based on the previously described FRET OS2-O-GlcNAc biosensor [27] by replacing the CFP by an Rluc8 sequence. The BRET biosensor is composed of Rluc8 fused to the fimbrial adhesin lectin domain GafD, a known OGT substrate peptide derived from casein kinase II placed between two flexible linkers (GGSGG) followed by a variant of the yellow fluorescent protein Venus (Fig 2B).
Tax was detected using sera from HTLV-1 infected individuals (kindly provided by Dr Gessain, Institut Pasteur, Paris, France) or the anti-Tax monoclonal antibody (mab) 168-A51 (NIH AIDS Research and Reference Reagent Program, USA). The following primary antibodies were used: anti-GFP recognizing GFP as well as the YFP and YPET variants (Roche Applied Science), anti-OGT (Sigma, DM-17 06264), anti-OGA (Santa Cruz, sc135093 or Sigma, SAB4200311), anti-O-GlcNAc (Abcam, RL2), anti-CREB (Millipore, CS 203204), anti-phospho CREB ser133 (Millipore, CS 204400), anti-actin (Santa Cruz, sc1616), anti-tubulin (GeneTex, GT114) and GAPDH (Santa Cruz, sc32233). HRP-conjugated anti-human, anti-mouse and anti-rabbit IgG (Promega) were used as secondary antibodies. Thiamet G (Sigma, SML 0244) was used at 10μM concentration.
C8166 T cells (2x106/12 well in duplicates) were cotransfected by nucleofection with 700 ng of the U3R-Luc reporter plasmid and 200 ng of the Renilla reporter plasmid pRL-TK. 293T cells seeded in duplicates in 24-well (3x104/well) were co-transfected with 500 ng of the U3R-Luc plasmid and 50 ng of pRL-TK, and with 500 ng of the control or the Tax plasmids with or without 200 ng of the YFP-CREB constructs. Luciferase activity was determined using the Dual Luciferase Assay System (Promega) and values were normalized with Renilla activity.
Cells were lysed in lysis buffer (50 mM Tris-HCl pH8, 1% NP40, 0.5% deoxycholate, 0.1% SDS and 150 mM NaCl) supplemented with protease and phosphatase inhibitors (Roche). Immunoprecipitations were carried out as follow: cell lysates were incubated overnight with primary antibodies at 4°C, and antibody complexes were captured on protein G-sepharose beads (GE Healthcare) 1h at 4°C. Sepharose beads were then washed 5 times in washing buffer (120 mM NaCl, 20mM Tris-HCl pH8, 0.2 mM NaF, 0.2 mM EGTA, 0.2% deoxycholate, 0.5% NP40) before elution in Laemmli buffer. O-GlcNAcylated proteins were precipitated on 40 μL of WGL-agarose (WGA) beads (Vector Laboratories, Paris, France) for 2h at 4°C. WGA beads were then washed 5 times in washing buffer and captured proteins then eluted in Laemmli buffer as described in [50]. In some experiments, N-acetylglucosamine (500 mM) was added during incubation with WGA beads as a control for non-specific binding of protein to WGA. Immunoprecipitated, WGA-precipitated proteins, and total cell lysates were separated by SDS-PAGE, transferred to membranes and blotted with specific antibodies.
Total RNAs were prepared with the Nucleospin RNAII kit (Macherey Nagel, France) and 1μg of RNA was reverse transcribed using the Maxima first strand cDNA synthesis kit (Thermo Scientific, France), according to the manufacturer’s procedure. Real-time-PCR was performed in the Lightcycler 2.0 (Roche, France) on 10 ng of reverse transcribed RNA using the following primers: OGT (forward: GCCCTGGGTCGCTTGGAAGA, reverse: TGC CAC AGC TCT GTCAAAAA), OGA (forward: TCTGCGGTGTGGTGGAAGGA, reverse: TGGGGTTAGAAAAAGTGATA) and the housekeeping gene HPRT (forward: 5’TGACACTGGCAAAACAATGCA3’, reverse: 5’GGTCCTTTTCACCAGCAAGCT3’) for normalization. PCR was conducted using the Sybr Green method with the following conditions: a first step of denaturation at 95°C for 8 min, followed by 40 cycles of denaturation (95°C for 10 sec), annealing (60°C for 10 sec), and extension (72°C for 8 sec) and a final step of melting curve (95°C for 5 sec, then 65°C for 15 sec. and finally 95°C for 10 sec).
Before the experiment, 107 C8166, MT2 or CIB cells were crosslinked using first 0,08% Disuccinimidylglutarate (SantaCruz Biotechnologies) during 30 min at room temperature and 1% Formaldehyde (Electron Microscopy Sciences) for 10 minutes at room temperature. Chromatin was then sheared using a Bioruptor Pico sonicator to obtain fragments of around 300 bp. Ten μg of chromatin were used for each condition. ChIP experiments were performed using the ChIP-IT high sensitivity kit from active motif. Primer pairs that specifically amplify the distal vCRE (position 201–275: Forward 5’ATCATAAGCTCAGACCTCCGGGAA3’, reverse 5’CCTGAGGACGGCTTGACAAACAT3’) were used for PCR.
HEK-293T cells were transfected in 12 well plates as described previously [51] using 300 ng of each cDNA construct, unless otherwise stated in the figure legend. One day after transfection, cells were transferred into 96-well microplate, and BRET measurements were carried out on the following day. TL-om1 T cells were transfected by nucleofection in 12 well plates. On the following day, cells were distributed into 96 well microplate and BRET measurements were performed.
For BRET measurements, cells were pre-incubated for 5 min in PBS in the presence of 5 μM coelenterazine. Light-emission acquisition at 485 nm and 530 nm was then started, and signal acquisition was performed every minute during 20–30 min using TECAN Infinite F200 Pro apparatus. BRET signal was expressed in milliBRET units (mBU). The BRET unit has been defined previously as the ratio 530 nm/485 nm obtained when the two partners are present, corrected by the ratio 530 nm/485 nm obtained under the same experimental conditions, when only the partner fused to Renilla luciferase is present in the assay [52]. Each measurement corresponded to the signal emitted by the whole population of cells present in a well (i.e., approximatively 4x104 HEK 293 cells or 106 TL-om1 T cells).
OGA activity was measured using 4-methylumbellifery-N-acetylβ-D-glucosamine (MU-GlcNAc, Sigma), which is converted into fluorescent 4-methylumbelliferon upon hydrolysis by OGA and other hexosaminidase [53]. 4-methylumbelliferon fluorescence was measured at 448 nm after excitation at 362 nm after 30 min and 60 min incubation at 37°C, to ensure that the determination was performed during the linear phase of the reaction. To determine the concentration of 4-methylumbelliferon, a standard curve was performed in each experiment using commercial 4-methylumbelliferon (Sigma). To specifically determine OGA activity versus other glycosydases, all reactions were performed in absence or presence of the highly specific OGA inhibitor Thiamet G. The difference of the fluorescent signal obtained in absence and presence of Thiamet G reflected the amount of 4-methylumbelliferon produced by OGA.
To measure OGT activity, OGT was immunoprecipitated using an anti-OGT antibody (Sigma-Aldrich) for 2h at 4°C. Precipitation was performed by incubating 50μL equilibrated protein G-sepharose beads (GE Healthcare) for 30 min at 4°C. After 3 washes, the precipitated proteins were submitted to an additional wash in OGT assay buffer containing 50 mM Tris-HCl and 12.5 mM MgCl2, pH7.5 and 1μM Thiamet G. OGT assay was then performed on protein-G sepharose bound OGT using the bioluminescent UDP-GloTM glycosyltransferase assay (Promega) exactly as described in the manufacturer instructions [54].
The use of peripheral blood mononuclear cells from patient CIB was approved by the French Comité Consultatif de Protection des Personnes dans la Recherche Biomédicale (CCPPRB) and the patient provided a written informed consent.
|
10.1371/journal.ppat.1007441 | Quantitative RNAseq analysis of Ugandan KS tumors reveals KSHV gene expression dominated by transcription from the LTd downstream latency promoter | KSHV is endemic in Uganda and the HIV epidemic has dramatically increased the incidence of Kaposi sarcoma (KS). To investigate the role of KSHV in the development of KS, we obtained KS biopsies from ART-naïve, HIV-positive individuals in Uganda and analyzed the tumors using RNAseq to globally characterize the KSHV transcriptome. Phylogenetic analysis of ORF75 sequences from 23 tumors revealed 6 distinct genetic clusters with KSHV strains exhibiting M, N or P alleles. RNA reads mapping to specific unique coding sequence (UCDS) features were quantitated using a gene feature file previously developed to globally analyze and quantitate KSHV transcription in infected endothelial cells. A pattern of high level expression was detected in the KSHV latency region that was common to all KS tumors. The clear majority of transcription was derived from the downstream latency transcript promoter P3(LTd) flanking ORF72, with little evidence of transcription from the P1(LTc) latency promoter, which is constitutive in KSHV-infected lymphomas and tissue-culture cells. RNAseq data provided evidence of alternate P3(LTd) transcript editing, splicing and termination resulting in multiple gene products, with 90% of the P3(LTd) transcripts spliced to release the intronic source of the microRNAs K1-9 and 11. The spliced transcripts encode a regulatory uORF upstream of Kaposin A with alterations in intervening repeat sequences yielding novel or deleted Kaposin B/C-like sequences. Hierarchical clustering and PCA analysis of KSHV transcripts revealed three clusters of tumors with different latent and lytic gene expression profiles. Paradoxically, tumors with a latent phenotype had high levels of total KSHV transcription, while tumors with a lytic phenotype had low levels of total KSHV transcription. Morphologically distinct KS tumors from the same individual showed similar KSHV gene expression profiles suggesting that the tumor microenvironment and host response play important roles in the activation level of KSHV within the infected tumor cells.
| Kaposi’s sarcoma (KS) is among the world’s most common AIDS-associated malignancies. The Kaposi sarcoma-associated herpesvirus (KSHV) was first identified in KS tumors and is now known to be the causative agent of all forms of KS, including classical, endemic, iatrogenic and HIV-associated. KSHV is endemic to sub-Saharan Africa with high infection rates in children and adults. Compounded with the high rate of HIV and AIDS in this area, pediatric and adult KS are some of the most common malignancies with the highest fatality rates. We used RNA deep sequencing to characterize KSHV expression in a large collection of KS biopsies from HIV-infected Ugandans. Using a novel approach to quantitate expression in complex genomes like KSHV, we found that RNA from a single KSHV promoter within the latency region constituted the majority of KSHV transcripts in the KS tumors. Alternate RNA processing produced different spliced and un-spliced transcripts with different coding potentials. Differential expression of other KSHV genes was detected which segregated the tumors into three different types depending on their expression of lytic or latency genes. Quantitative analysis of KSHV expression in KS tumors provides an important basis for future studies on the role of KSHV in the development of KS.
| Since its discovery in 1994, the Kaposi sarcoma-associated herpesvirus (KSHV), also known as human herpesvirus-8 (HHV-8), has been identified as the etiologic cause of all types of Kaposi sarcoma (KS), and is etiologically associated with primary effusion lymphoma (PEL) and multicentric Castleman Disease (MCD)[1]. The KSHV genome encodes more than 90 genes, including a core of genes highly conserved among the different herpesviruses [2]. In addition, a number of novel genes exhibiting sequence homology to cellular genes implicated in mitosis, cell cycle regulation and immunity have been identified [3].
In vitro, KSHV infects a variety of cell types including endothelial, epithelial, fibroblast and lymphocyte lineages [4] and establishes a latent infection in which only a subset of genes are detected, including LANA (ORF73)—the latency-associated nuclear antigen, vCYC (ORF72)–a cyclin D homolog, and vFLIP (ORF71)–a homolog of the Fas-associated protein with death domain-like interleukin 1beta-converting enzyme/caspase-8-inhibitory protein [5–7]. Early studies with cultured PEL cells determined that these genes were present on a tricistronic mRNA originating from the constitutive latency transcript (LTc) promoter upstream of ORF73 [6, 8]. Latent KSHV infections, first characterized in endothelial cells in vitro, resulted in the majority of cells expressing LANA as punctate dots in the nucleus, with a small population of cells (~1%) expressing ORF59, a DNA polymerase processivity factor [9–11]. In most cell types, the widespread lytic reactivation necessary for production of infectious virus was achieved only by using chemical inducers such as the phorbol ester TPA or the HDAC inhibitor sodium butyrate or by overexpression of exogenous recombinant ORF50, the replication transactivator (RTA) [4, 11, 12]. The complement of KSHV genes has been divided into functional groups based on their initial expression during establishment of viral latency and their response to artificial induction, with latency, immediate-early, early and late gene designations. While ORF50 RTA and ORF59 represent immediate-early and early genes, respectively, the genes encoding the major capsid protein (MCP; ORF25), and the virion envelope glycoproteins, gB(ORF8) and K8.1 are examples of KSHV late genes.
Early attempts to determine the expression profile of KSHV in KS tumors examined the RNA transcripts in KS lesions by Northern analysis. Two small RNA transcripts were detected, including T0.7 encoding the K12 Kaposin membrane-associated protein and T1.1, a polyadenylated nuclear RNA (PAN) [13]. Using in situ hybridization, the T0.7 RNA was detected in all KS tumor cells, while the T1.1 RNA was present in only 10% of the T0.7 positive cells [14]. In addition, RNA encoding the MCP late gene was detected in the same cells containing the T1.1 transcript. Once antibodies were available, the expression and localization of viral proteins was examined by immunohistochemical methods. The major latency-associated protein LANA was consistently detected in the nuclei of the vast majority of spindeloid tumor cells in the KS lesion [5, 15, 16]. In contrast, markers of lytic replication, including ORF50 RTA, ORF59, and the vIL-6 homolog K2 were detected very rarely (<1%) in the tumor cells, while no expression of late genes, including ORF26 and ORFK8.1 was observed [17–22]. Using an array of real-time PCR assays targeting the majority of known KSHV genes, the expression of mRNA transcripts from the latency locus, including ORFs 71 (vFLIP), 72 (vCYC) and 73 (LANA), was detected in 21 KS biopsies [23]. Subsequently, using the same PCR array, two types of transcriptional signatures were detected in a panel of KS tumors [24]. In half of these tumors, KSHV transcription was limited to the latency-associated genes. In the other half of the KS tumors, variable and incomplete expression of viral lytic mRNAs was observed.
We have utilized RNA deep sequencing (RNAseq) to globally examine the KSHV transcriptome in latently-infected tissue culture cells in vitro [25]. Due to the highly complex nature of the KSHV transcriptome, we developed a novel approach to more accurately quantitate specific viral transcripts using unique coding sequence (UCDS) features targeting non-overlapping regions of KSHV transcripts. We sequenced the RNA transcripts from in vitro infected cells to a great depth, with more than a million reads mapping to the KSHV genome. High levels of transcripts were observed across the complete KSHV genome in the absence of artificial induction with chemicals or recombinant proteins. This allowed us to develop a detailed map of KSHV transcription, which informed the development of the UCDS features in a new gene feature file (KSHV NC_009333 UCDS ver 020116.GFF) to globally analyze and quantitate KSHV gene expression [25].
Recently, RNAseq has been used to characterize the viral and cellular transcriptome in KS tumor and non-cancer biopsies of African epidemic HIV+ KS patients undergoing anti-retroviral therapy (ART) [26]. Lesions from four individuals were analyzed yielding 718–17,202 reads that mapped to known KSHV ORFs in the NC_009333 KSHV reference genome. High level expression of the latency region was reported but no obvious pattern was observed between the tumor biopsies. In the current study, we have used RNAseq to analyze and quantitate KSHV gene expression in a large collection of 41 KS tumor biopsies from HIV-infected individuals in Uganda who were naïve to ART. The RNAseq libraries were sequenced to an average depth of 100 million reads yielding up to 159,000 KSHV-mapped reads. Using the new gene feature file, we quantitated the RNA reads mapping to non-overlapping UCDS features and identified a set of transcripts from the latency region that was highly and consistently expressed in all the KS tumors.
Thirty Ugandan participants contributed 41 cutaneous KS tumors for RNA-Seq analysis, with 11 participants contributing 2 samples (S1 Table). The majority of participants were men (21/26; 80%) with a median age of 34 years (range, 23–56 years) (Table 1). At the time of KS diagnosis, all had advanced T1 tumor stage per AIDS Clinical Trials Group staging criteria, which includes those with extensive oral KS, visceral KS or tumor-associated edema [27]. All patients were naïve to antiretroviral treatment. Median CD4 T-cell count was 183 cells/mm3 (IQR, 58,331 cells/mm3) and median HIV plasma RNA was 5.5 log10 copies/mL (IQR, 5.1, 5.6 log10 copies/mL). Tumor samples represented a range of morphotypes, including 24 macular (58%), 13 nodular (32%), and 4 fungating (10%) lesions. Total nucleic acids were extracted from the 41 KS biopsies and cDNA libraries were prepared from poly-A-selected RNA and subjected to RNA deep-sequencing on the Illumina platform (S1 Table). 37 independent KS samples were sequenced for 50 bp from paired-end non-stranded libraries. Four additional KS samples were sequenced from stranded libraries to distinguish sense and anti-sense RNA transcripts. Total reads ranged from 81–124 million for the paired-end libraries and 35–52 million for the stranded libraries, which were analyzed at a lower depth of sequencing. RNA reads mapping to the human genome HG19 were subtracted from the libraries and the remaining reads were mapped onto the KSHV NCBI reference sequence NC_009333, strain GK18, with mapped KSHV reads ranging from 13 to 158,924, with a median of 10,232 (Fig 1A). Five of the KS biopsies had greater than 100,000 KSHV-mapped reads. While seven of the KS tumors had less than 1,000 total KSHV-mapped reads, the level of reads mapping to the human genome in these samples was comparable to the other KS samples (S1 Table). A comparison of KSHV mRNA expression in the different tumor morphotypes showed no significant differences in total KSHV-mapped reads (Fig 1B). The KSHV genome copy number per cell was determined for four of the KS samples (001_C, 006_B, 026_B, and 029_B). A comparison of the Ct values obtained from qPCR assays targeting KSHV and the cellular gene oncostatin M, essentially as described in [28], revealed similar KSHV genome copy numbers ranging from 0.5–0.9 KSHV genomes/cell (mean = 0.6). A comparison of the number of mapped KSHV reads/KSHV genome copies showed a variance of 8.6% across the four KS biopsies.
The RNA-seq reads mapping to the complete KSHV genome were visualized with the Integrated Genome Viewer (IGV), using a linear scale to provide an overview of the highly expressed regions of the genome. Representative data from KS tumors with 2,432 to 158,924 total KSHV-mapped reads are shown (Fig 2), while data from the seven KS tumors with total KSHV-mapped reads less than 1,000 are provided in S1 Fig The highest level of RNA reads mapped to the region of the T0.7 RNA transcript within the latency locus at the right end of the KSHV genome (Fig 2). T0.7 encodes ORF K12 Kaposin A, a small transmembrane protein implicated in cell adhesion and transformation [29, 30]. High levels of reads mapping to the T0.7/K12 region were consistently detected in all the KS tumor samples, regardless of the level of total KSHV-mapped reads in each sample. High levels of RNA reads also mapped to a small region of the KSHV genome located immediately to the right of ORF72, observable as a single vertical line of mapped reads (Fig 2; indicated with an asterisk in the bottom graphic). A lower but consistent level of reads mapped to the adjacent region containing ORFs 71 and 72. ORFs 71 and 72 encode vFLIP, which functions to promote cell survival and inhibit KSHV lytic replication, and vCYC, which regulates cell-cycle progression, respectively [31]. Downstream of the latency locus at the right end of the genome, a moderate level of reads mapped consistently to the region encoding ORF75, a large tegument protein essential for viral lytic replication [32], and K15, a multiple-pass membrane protein that modulates cellular signaling pathways associated with KSHV-induced angiogenesis [33–35](Fig 2). At the left end of the genome, a moderate level of reads mapped inconsistently to the multifunctional regulatory polyadenylated nuclear (PAN; T1.1) RNA transcript, with higher levels in the tumors with lower total KSHV-mapped reads shown in the upper part of Fig 2. One fungating tumor 023_B showed high levels of both PAN and total KSHV-mapped reads. Some tumor samples contained moderate levels of reads mapping to ORFK2, the viral interleukin-6 (vIL-6) homolog, or ORFK5, the ubiquitin ligase modulator of immune response (MIR2). One tumor, 008_B, showed high level expression of a sharply delineated region of the KSHV genome extending from ORFK3 to ORF19, with very high level of reads mapping to ORFK5 (Fig 2, bracketed). This transcript pattern was unique to this tumor and was not seen in the paired tumor from the same individual (008_C) or any other KS tumor. This sample is further described below.
To analyze the complete range of read depths across the KSHV genome, the mapped reads were also visualized using a log-based scale (Fig 3). This analysis revealed the presence of RNA reads mapping across the complete KSHV genome with concentrations in regions encoding specific genes associated with lytic replication (ORFs 6, 59, 60), gene regulation (ORFs K8, 57), virion structure (ORFs 11, 17.5, 27, 33, 38, K8.1, 65, 66) and immune modulation (ORFs 4, K2, K3, K4, K5, 45, K10) (Fig 3). Similar patterns of transcription were visually observed in the different KS tumors regardless of the total number of KSHV-mapped reads in each sample (see for example, tumors 11_D and 013_C, with 5,928 and 158,924 total KSHV-mapped reads, respectively). This similarity was also observed in the 7 KS tumors containing less than 1,000 total KSHV-mapped reads (S1 Fig), suggesting that overall the majority of KSHV-infected cells in the KS lesions expressed the same basic transcriptome pattern. Of note, occasional KS samples, such as 030_B, had minimal levels of reads mapping to ORFK15, even though high levels of reads mapped to the adjacent ORF75 (Fig 3, arrow). Different alleles of ORFK15 have large sequence variation extending into the ORF75 sequence [36, 37], which could have affected the ability of ORFK15 reads in the Ugandan tumors to map to the NC_009333 reference sequence.
The RNA-seq mapping studies were performed by aligning the RNA reads from the different Ugandan KS tumors to the GK18 NCBI Reference sequence (NC_009333) present in a patient with a case of the classic (HIV-negative) KS. The IGV view of the aligned reads from the Ugandan KS tumors revealed numerous mismatches indicating the presence of different KSHV strains in the KS tumor samples. To perform an initial phylogenetic analysis of these strains, we compared the RNA sequences of the large open reading frame encoding ORF75, which was highly expressed in nearly every KS tumor sample. Although most previous studies of KSHV phylogeny have used the widely divergent K1 and K15 sequences [36, 38], these genes, especially K1, were too minimally and inconsistently expressed in the KS tumors to obtain sufficient sequence for phylogenetic analysis. The complete coding sequences of ORF75 (3891 bp) were assembled from the 50 bp reads for 23 of the Ugandan KS samples. These sequences were aligned with published OR75 sequences from 16 unique KSHV genomes from Zambia [39] and ORF75 sequences in the NCBI database from KSHV strains in several KS biopsies and PEL cell lines obtained in Western countries. Maximum likelihood analysis revealed two major (A, C) and four minor ORF75 clusters (B, D-F) (Fig 4). A BLAST alignment with ORF75 partial sequences that have been previously typed [40] indicated that Cluster A corresponded to subtype [B], Clusters B and C corresponded to subtype A/[B], Cluster D corresponded to subtype R/A, Cluster E corresponded with subtype R/M and Cluster F corresponded with subtype N. The vast majority of Zambian sequences were distributed in the major clusters A and C. Previous studies have shown that ORF75 sequences are linked to the adjacent K15 subtypes, of which 3 distinct alleles P, M and N have been detected [36]. The majority of the ORF75 sequences, including those in Clusters A-D were linked to the K15 P-allele (Fig 4), which was present in the GK18 reference sequence. In contrast, the U030 and ZM123 sequences, like the BC1 sequence, were linked to the M-allele, while U012, ZM095 and ZM128 [39], were linked to the K15 N-allele. The sequence differences between the K15 P-allele of GK18 and the more distantly related M and N alleles resulted in essentially no RNA reads from the U012 and U030 tumor samples aligning with the K15 region of GK18 (see Fig 3). Although our phylogenetic analysis was based on a single gene, it is clear that the KSHV strains infecting the different Ugandan KS tumors show high variability, with strong similarity to KSHV strains identified previously in the Zambian KS samples [39]. Overall, the RNAseq analysis indicated the presence of a single KSHV strain in each tumor. In all cases, but subject 008 (discussed below), independent tumors from the same patient contained the same KSHV strain.
To accurately correlate the RNA reads to specific mRNA transcripts in the latency region, we previously developed a map of transcripts that had been identified in the literature [25]. Since these transcripts were initially characterized in different KSHV strains with different sizes, their positions were mapped onto the sequence of the NCBI reference sequence for KSHV, strain GK18 (NC_009333), and the sizes of the GK18 transcripts were predicted from the transcription start and polyadenylation (poly-A) termination sites (S2 Table). A map of the spliced and unspliced latency region transcripts is shown in Fig 5B, with the transcripts grouped according to their use of the poly-A termination sites at bp 117,553 (Group A), bp 122,342 (Group B), and bp 123,015 (Group C). Due to heterogeneity in the DR5, DR6 and DR7 repeat regions, the sizes of the transcripts from the GK18 strain vary from the published transcript sizes determined for other KSHV strains, such as BCBL-1, as indicated in the corresponding references (summarized in S2 Table).
To analyze the read depth and splicing events in the latency region, Sashimi plots were produced by IGV from the TopHat2 analysis of the RNA reads for the KS tumor samples. Sashimi plots graphically present the read depth across a selected region and show the location and quantity of split RNA reads that define the presence of a splicing event within an mRNA transcript. Sashimi plots of four representative KS tumor samples (Fig 5A) were aligned with the different spliced and unspliced transcripts generated from the latency region (ORFs 69-K14) (Fig 5B). This alignment clearly showed a high level of RNA reads mapping to the T0.7 RNA (herein designated as T0.7A) from the P5 latency promoter, which encodes Kaposin A. In addition, a high level of reads mapped to the adjacent DR5 and DR6 repeat regions, even though the TopHat analysis limited reads to a single map position. A number of spliced and unspliced transcripts that terminate at a common poly-A site (bp 117,553) downstream ORF K12 contain the repeat regions. These transcripts are derived from different latency transcript (LT) promoters and include the spliced T1.6A and T1.8A transcripts from the P1/LTc (constitutive) promoter, the spliced and unspliced T1.7A and T6.5A transcripts from the P3/LTd, (downstream) promoter, and the unspliced T1.5A transcript from the P4 promoter (Fig 5B).
Analysis of the Sashimi plots across the latency region revealed a high level of split reads mapping between the highly expressed DR6 repeat region and the highly expressed genomic region downstream of the P3 (LTd) promoter, herein designated as P3-exon1 (Fig 5A and 5B), which was indicated in Fig 2 with an asterisk. Examination of the corresponding sequences in the KSHV genome revealed classical splice donor (AG|gt) and acceptor (ttacgcccccttcgcag|G) sites at bp 123,843 and bp 119,047, respectively. These sites define the presence of a 4,796 bp intron (Fig 5B; labeled “a”), which is spliced from the pre-mRNA for transcripts T1.7A and T1.8A, and encompasses the sequences encoding ORFs 72 and 71, the right origin of replication and flanking long inverted repeat (LIR2) and the microRNAs miR K1-9 and 11. A high level of reads split across intron “a” were detected in the vast majority of KS tumors (Table 2). The depth of the split reads across intron “a” (for ex. see Fig 5A, ranging from 100 reads (030_C) to 1,307 reads (013_C); Table 2) was similar to the depth of the reads in the flanking exons suggesting that the majority of the reads mapping to the K12/DR5/DR6 region were derived from the T1.7A spliced transcript (Fig 5B, double asterisk). This conclusion was confirmed by RT-PCR amplification of the spliced T1.7A transcript from four KS biopsies using primers derived from the K12 region and the exon junction spanning the spliced “a” intron (S2 Fig). While there was evidence for the splicing of intron “b” in some tumors (see for example Fig 5A: 013_C- “b” = 32 reads; Table 2), compatible with the processing of transcript T1.8A from the upstream P1 (LTc) latency promoter, there was no corresponding accumulation of RNA reads mapping to the 5’ exon of this transcript (Figs 3 and 5A). Thus, the splicing data indicates that the high level of split reads across the K12/DR5/DR6 region and P3-Exon1 are derived from the spliced T1.7A transcript initiating from the P3 (LTd) promoter, which is highly expressed in all the KS tumors.
Quantitation of transcripts from complex genomes, such as KSHV, has been difficult due to the compact nature of the genome and the presence of numerous overlapping transcripts that are differentially expressed. We developed a novel approach to quantitate RNAseq reads mapping to specific genes and gene features in the highly complex KSHV genome using unique coding sequence (UCDS) features specific for all known KSHV ORFs and transcriptional regions [25]. Published information and transcript data from RNAseq analysis of primary latent KSHV infections of several endothelial cell types and long-term latent infections in PEL cells were used to identify transcription start and termination sites and globally map mRNA transcripts. The UCDS features were devised to be non-overlapping, separated by the length of a read (50 bp) so that algorithms, such as HTSEQ, using the “intersection_nonempty” setting could specifically identify and distinguish reads mapping to the different targeted gene features on both DNA strands. RNA read depth was determined by quantitating the reads aligning to the UCDS features in a simplified gene feature file (KSHV NC_009333 UCDS ver 020116.gff; S1 File) based on the KSHV reference genome NC_009333 [25]. The read count was normalized to the size of the feature and number of total KSHV-mapped reads in each tumor sample, yielding a relative transcript level, in transcripts per million KSHV reads (TPM), as described in Materials and Methods. Read data and normalization for each KS sample is provided in S3 Table This normalization allowed the expression of each KSHV gene to be compared as a proportion of the total number of KSHV-mapped reads in each sample. In this and subsequent quantitative analyses of read depths across the KSHV genome, the analyses were limited to 34 of the 41 KS tumor samples, which had sufficient levels of RNA reads mapping to the KSHV genome (>1000) for objective comparisons (see Fig 1). The median normalized level of transcripts targeted by our set of UCDS features across the entire KSHV genome for the 34 samples is provided in S4 Table. To more accurately reflect the protein coding potential for genes in regions with overlapping polycistronic transcripts, we determined the level of the primary transcripts for each ORF derived from its associated promoter, in which the ORF would be the first in the transcript to be translated through 5’ CAP-dependent initiation. The contribution of overlapping transcripts derived from distal promoters was subtracted, yielding an estimation of the primary transcript levels for each ORF (S4 Table), as discussed previously Bruce et al [25].
In the latency and flanking regions, UCDS features were identified targeting the coding sequences for the ORFs K12A (T0.7), 72, 71, 73, K14, 74, 75 and K15, as well as additional areas of interest, including the direct repeats DR5 and DR6, the region encoding microRNAs miR-K1-9, 11 (miR-region), and the short P3-Exon1 downstream of the ORF72 P3(LTd) promoter (K12Aa) (Fig 5C). Quantitation of the reads mapping to the latency region from the 34 KS tumor samples revealed very high transcript levels for the K12/T0.7 region (UCDS K12A) and flanking DR5 (UCDS DR5) and DR6 (UCDS DR6) repeat regions (median = 162,467, 310,389 and 65,018 TPM, respectively) (Fig 5D, boxed in red; S4 Table). Although the known transcripts in this region span both the DR5 and DR6 region, transcribed right to left (Fig 5B), significantly less reads mapped to the DR6 UCDS feature, which would detect the 5’ region of such transcripts. The DR6 repeat and upstream flanking regions are known to have significant sequence heterogeneity between different KSHV strains [41], suggesting that the relatively lower level of reads mapping to the DR6 UCDS feature could be due to mismatches between the reads from the Ugandan KSHV strains and the NC_009333 KSHV reference sequence used for the mapping. Very high levels of transcripts (median = 123,915 TPM) were detected using the K12Aa UCDS feature (Fig 5D, boxed in red), which targets the small P3-Exon1 downstream of the P3(LTd) promoter (Fig 5B and 5C). The similarity in the transcript levels detected across the K12A, DR5, DR6 and K12Aa UCDS features (Fig 5D, boxed in red) correlates with the presence of the common spliced transcript (T1.7A) derived from the P3(LTd) promoter, which encodes the Kaposin A/B/C complex (Fig 5A, double asterisk), as described above. This was confirmed by RT-PCR of RNA from 4 KS tumor biopsies (S2 Fig).
Moderate levels of transcripts (median = 15,794 and 14,430 TPM) were detected using UCDS features targeting ORF71 and ORF72, respectively (Fig 5D; S4 Table). Previous studies have determined that the major transcripts encoding ORFs 71 and 72 are either bicistronic (5’ ORF72/ORF71 3’) initiating from the P3(LTd) promoter or tricistronic (5’ ORF73, ORF72, ORF71 3’) initiating from either the P1(LTc) or P2(LTi) promoters (Fig 5B; S2 Table). The similarity in the read depths detected using the ORF71 and ORF72 UCDS features suggests that the majority of these reads are derived from the major unspliced bicistronic transcript (T1.7B) from the P3(LTd) promoter (Fig 5A, single asterisk). While these reads could also be derived from the common spliced bicistronic transcript (T1.8B) from the P1(LTc) promoter (see Fig 5B), only a small number of reads mapped to the 5’ exon of this transcript near the P1(LTc) promoter (see Fig 5A). Only a low level of transcripts (3,300 TPM) were detected with the miR UCDS feature, which targets transcripts containing the miR region flanking the LIR2 (Fig 5C), suggesting that the majority of the transcripts encoding ORF71 terminated at the poly-A site (bp 122,342) downstream of ORF71 (Fig 5B). Quantitation of the reads mapping to the non-repetitive regions of ORF73 using the ORF73A and ORF73B UCDS features revealed low levels of the tricistronic transcripts T5.2B, T5.4B, T5.5B and T5.7B encoding ORFs 73, 72 and 71 (Fig 5D; S4 Table). This supports the conclusion that the majority of the ORF71 and ORF72 reads map to the unspliced bicistronic T1.7B transcript from the P3(LTd) promoter and not from transcripts from the P1(LTc) promoter.
To further analyze transcription across the ORF72/71 locus, we determined the ratio of transcripts detected using the ORF72 and ORF71 UCDS features. The ORF71/ORF72 ratios in the different tumors ranged from 0.7 to 2.1, suggesting differential expression of the two ORFs that was not strictly compatible with the single bicistronic T1.7B transcript (S3 Fig). Seven tumors showed higher levels of transcripts encoding ORF72 (ORF71/72 ratios from .7 to 0.9) (S3A Fig) indicating the presence of a ORF72 monocistronic transcript T1.0C (Fig 5B), identified previously by Sarid et al [42]. Twenty-one tumors showed higher levels of transcripts encoding ORF71 (ORF71/72 ratios from 1.1 to 2.1) (S3A Fig), indicating the presence of a ORF71 monocistronic transcript T0.9B (Fig 5B), identified previously by Grundhoff and Ganem [43].
Kaposin A is part of a complex translational program that generates multiple novel proteins from the K12 locus. The Kaposin A sequence is downstream of the DR5 and DR6 repeat sequences, which encode variant translation products initiating at alternate CUG initiation codons that have been detected in cultured PEL cells [41, 44]. The CUG initiation codons occur in the sequence region upstream of the DR6 repeat region, which is highly variable in different KSHV strains [41]. Previous studies identified a non-spliced transcript from a promoter, herein referred to as P4, corresponding to T1.5A (see Fig 5B), whose major translation product in the KSHV strains present in BCBL-1 and JSC-1 cells was a CUG-initiated open reading frame encoding a highly repetitive protein sequence, with an “LAH” N-terminal sequence (Kaposin B)[44](Fig 6B). A minor translation product, Kaposin C, initiated from an alternate downstream CUG initiation codon with a “LQY” N-terminal sequence. Kaposin C contained a similar repetitive protein sequence as Kaposin B but was fused to Kaposin A [44] (Fig 6B). Functional studies have shown that the BCBL-1 Kaposin B can modulate mRNA turnover by stabilizing cytokine mRNAs [45]. However, the functions of Kaposins A and C are as yet unclear, as functions attributed to Kaposin A, such as transforming potential [30], have been further attributed to the microRNA miR-K10, which is embedded within the Kaposin A coding sequence [46].
Comparison of the K12 locus of the BCBL-1/JSC KSHV strain with the GK18 reference strain revealed a 2 bp deletion in the BCBL-1/JSC-1 sequence between the DR6 and DR5 repeat regions (Fig 6A and 6B). This insertion alters the open reading frames such that the “LAH” initiated open reading frame from the upstream CUG codon in GK18 is now fused with Kaposin A, suggesting that the major translational product of the GK18 K12 locus would be a novel protein, herein termed “Kaposin D” (Fig 6A). This indicates that the minor translation product of the GK18 K12 locus from the downstream CUG initiation codon would be a novel protein “Kaposin E”, with the N-terminal sequence “LQY”. Unlike Kaposin C in the BCBL-1 strain, Kaposin E would lack the downstream Kaposin A fusion. The putative DNA and encoded protein sequences in this region are provided in S4 Fig.
The RNA sequence data in our study revealed several nucleotide differences between the Ugandan KSHV strains and both the GK18 and BCBL-1/JSC-1 strains. A single nucleotide deletion was detected downstream of the P4 promoter before the CUG initiation codons in all the Ugandan KS samples (Fig 6C and S4 Fig). All Ugandan KSHV strains also had an additional nucleotide deletion between the DR5 repeat and the Kaposin A reading frame (“A” bp 118,228, NC_009333). The deletion of this base, which is also detected in all the published Zambian KSHV strains [39], changes the open reading frame through the repeat region that is contiguous with the Kaposin A sequence downstream. In a subset of Ugandan KSHV strains, from KS tumors 001, 005, 006, 015, 018, 020, 023, 026, 028, 029 and 034, this deletion results in neither open reading frame from the two CUG initiation codons creating a fusion with Kaposin A, as exemplified by the Zambian KSHV strain, ZM114 (Fig 6C). Thus, these KSHV strains would encode Kaposin B as the major CUG-initiated translation product and Kaposin E, as the minor-CUG initiated product, with Kaposin A as a downstream AUG-initiated product. The remaining Ugandan KSHV strains, from KS tumors 003, 004, 007, 008, 009, 010, 012, 013, 022, 024, 030, 037, 088, 099, and 101 had additional sequence differences that altered the possible protein products expressed from this locus. In these strains, the major CUG-initiated ORF, Kaposin B, was eliminated by a change in the putative “CUG” initiation codon to “CGC” (Fig 6D and S4 Fig). In addition, the minor CUG-initiated ORF, Kaposin C, was eliminated by a change in the codon flanking the “CUG” initiator from “CGA” to the stop codon “TGA”, thus immediately terminating translation. A similar situation was seen in the Zambian KSHV strain, ZM004 (Fig 6D), and was confirmed by PCR amplification and sequencing.
Small 5’ upstream ORFs (uORFs) are known to regulate expression of downstream ORFs by capturing or slowing the progression of ribosomes scanning the transcript for favorable initiation codons [47]. Previous studies have shown that a small 24 aa uORF is encoded upstream of ORF72 in the initial P3-exon1 of the T1.7B bicistronic ORF72/71 transcript from the P3(LTd) promoter (Fig 7A). This uORF attenuates the expression of the ORF72 10–20 fold in the T1.7B bicistronic transcript [43], and would be expected to attenuate ORF72 expression in the T1.0C monocistronic transcript and ORF71 expression in the T0.9B monocistronic transcript (Fig 7A). This uORF is also encoded in the initial 5’ exon of the highly expressed T1.7A spliced transcript from the same promoter and is positioned upstream of the ORFs encoding the Kaposin A/B/C complex (Fig 7A). Thus, this uORF would be expected to also attenuate translation of the Kaposin B/C complex proteins, especially since these proteins initiate from CUG codons.
The processing of the pre-mRNA for the T1.7A spliced latency transcript generates an RNA intron that is believed to be the source of the major KSHV microRNA species, miRs-K1-9 and 11 [48–50] (Figs 5B and 7A). Therefore, the high levels of the T1.7A spliced transcript could generate elevated levels of these microRNAs in the KS tumors. Since the other KSHV microRNAs, miR-K10 and miR-K12, are located within the retained exon in the T1.7A spliced transcript, such transcripts could also be processed to produce these microRNAs. Processed microRNAs are not detected in our RNAseq protocol, since they do not contain the 3’ poly-A region used for RNA purification. Thus, purification of small RNA species in the KS tumor samples would be needed to confirm the presence of KSHV microRNAs in the KS tumors.
RNA editing of bp 118,096 (NC_009333) (Fig 7A) has been shown to convert the transformation-associated miR-K10a containing an adenine in its seed sequence into the non-transforming miR-K10b containing an inosine [46]. We examined the RNA reads from the KS tumors for the presence of an edited base in the RNA reads. The fraction of ORFK12 transcripts containing an edited base ranged from 0 to 59% across the 34 KS tumors with a median of 13% (Fig 7C), suggesting that the transforming miR-K10a would be the major microRNA form produced in most of the KS tumors. Since miR-K10 is embedded within the mRNA transcripts encoding Kaposin A, RNA editing also affects the encoded Kaposin A protein sequence, changing a serine to glycine [51](Fig 7A and 7B). Our analysis revealed that only a fraction of the ORFK12 transcripts in the majority of KS tumors would encode the altered Kaposin A.
The spliced and non-spliced Group B latency transcripts transcribed from the P1(LTc) promoter and P2(LTi) promoter terminate at the poly-A site downstream of ORF71 (Fig 5B, S2 Table). Although previous in vitro studies indicated that the P1 and P2 transcripts were the major latency-associated transcripts in PEL cells, there was evidence for only minimal expression in the KS tumors (Figs 3 and 5A). Only two tumors had evidence for splicing of intron “e” within the major latency-associated spliced ORF73/72/71 tricistronic transcript (T5.2B), with only 3 reads detected (Table 2). No evidence was detected for splicing of intron “f” (T5.4B), which is similar to intron “e” but from an adjacent splice acceptor site (“f” acceptor: bp 127,626 [6] compared to “e” acceptor: bp 127,462 [8]). Transcripts initiating from the P1(LTc) promoter upstream of ORF73 LANA were quantitated using the ORF73A and ORF73B UCDS features (Fig 5C). These features flank the large DR7 repeat region within the ORF73 coding sequence, which was not used for quantitation due to its repetitive nature. Consistent low levels of LANA reads (median = 2,195 TPM) were detected in the KS tumors (Fig 5D; S4 Table), indicating low levels of the T5.2B, T5.4B, T5.5B, and T5.7B ORF73/72/71 tricistronic spliced and unspliced transcripts derived from the P1(LTc) or P2(LTi) promoter. The low level of transcription from the P1(LTc) promoter correlates with the very low levels of split reads detected for introns “e” and “f” in the Group B transcripts (Fig 5A and Table 2). RT-PCR analysis of RNA from four of the KS tumor samples with the highest levels of KSHV-mapped reads failed to detect any of the tricistronic transcripts from the P1(LTc) or P2(LTi) promoters, confirming the RNAseq data (S2 Fig).
Additional genes adjacent to the latency locus showed consistent and high-level expression in the KS tumors. These included ORF75 and K15 (Figs 2 and 3), which are expressed in the same orientation as the latency genes described above. While the vast majority of the KS samples showed comparable levels of reads mapping to both ORF75 and K15 (Fig 3), tumors from two individuals, 012 and 030, had high levels of reads mapping to ORF75 with no reads mapping to K15 (Fig 3, arrow). The different alleles of K15 have significant sequence variation extending into the ORF75 sequence. The phylogenetic analysis of the ORF75 sequences indicated that the KSHV strains in tumors 012 and 030 contain the K15 alleles N and M, respectively, thus reads from these strains do not map to the P-allele containing GK18 reference sequence. UCDS features were developed targeting the large ORF75 open reading frame (3,891 bp) (ORF75 UCDS) and the largest exon coding the C-terminal domain of K15 (464 bp) (K15a UCDS) from the heavily spliced K15 gene, which appears to be present in all K15 transcripts. Quantitation of the reads mapping to the ORF75 and K15a UCDS features revealed moderately high transcript levels across 32 KS samples (median = 21,490 and 15,866 TPM respectively) (Fig 5D; S4 Table). The KS samples from 012 and 030 were excluded, as their true read counts were not captured in this analysis. A single poly-A transcription termination signal has been identified for these transcripts downstream of ORF75 [52], suggesting that all of the transcripts detected by the K15a UCDS feature would be bicistronic, encoding K15 at the 5’ end and ORF75 at the 3’ end. Thirty of thirty-two tumor samples showed higher levels of ORF75 transcripts compared to K15 (Fig 5D; S4 Table; S3B Fig), indicating the presence of an ORF75 monocistronic transcript initiating upstream of the AUG initiator of ORF75. Quantitative analysis revealed that approximately two-thirds of the transcripts encoding ORF75 would be bicistronic, with K15 as the primary CAP-dependent translation product. The remaining transcripts would be monocistronic, with ORF75 as the primary translation product. It is not known whether ORF75 translation could occur through IRES-mediated initiation in the bicistronic transcripts, as was observed for ORF71 [43].
One KS tumor, 008_B, exhibited an unusual high level of reads mapping to the genomic region from ORFK3 to ORF19, as indicated above (Figs 2 and 3; bracketed in black) and in more detail (Fig 8A). The RNA reads at the borders of this region abruptly stopped in the middle of ORFK3 at one end and ORF19 at the other end, with no correspondence to possible RNA termination sites. High level expression of this region was not detected in KS tumor 008_C (Figs 2, 3 and 8A), which was isolated from a different location in the same patient, suggesting the presence of a genomic rearrangement in the KSHV strain in the 008_B tumor. By decreasing the alignment threshold in the initial Bowtie2 alignments, we identified reads containing partial sequences aligning to the ORFK3 and ORF19 regions flanking this highly expressed region. Sequence analysis of these reads revealed that a 14,813 bp ORFK3-ORF19 genomic region had been translocated from its original position at the left end of the genome (bp 19,168 to 33,980) to the right end of the genome within the long-inverted repeat (LIR2) (bp 119,504, numbering from the NC_009333 sequence) (Fig 8). A model genomic sequence of NC_009333 with this translocation was created and used to map the 008_B reads (as indicated in Fig 8). This analysis revealed high levels of spliced transcripts derived from the highly expressed latency promoter P3(LTd) upstream containing the 5’ P3-Exon1 (Fig 8D and 8E). Novel transcripts were detected in which the P3-exon1 was spliced to various acceptor sites in downstream genes within the translocated genome fragment. A spliced transcript encoding ORFK5 was expressed at the highest level. The majority of other genes contained in the translocated region were also highly expressed with non-spliced transcripts driven by their original promoters. Since high level expression of these genes was not observed in the paired 008_C tumor, it appears that the translocation of this genomic region to the highly expressed latency region was responsible for the increased gene expression.
At the left end of the KSHV genome, high levels of RNA reads mapped to the PAN (T1.1) mRNA transcript in some KS tumors (Figs 2 and 3; ex. KS tumors 011_D and 023_B). A UCDS feature targeting the PAN RNA was used to quantitate the RNA reads mapping to PAN. While transcript levels reached nearly 300,000 TPM in several KS samples, with a median of 37,413 TPM, significant variation was observed across the KS tumors (Fig 9A and 9D; S4 Table). The PAN transcript overlaps the longer T6.1 transcript encoding ORFK7 (Fig 9B). A UCDS feature was developed to target the ORFK7 transcript upstream of PAN (Fig 9C) to avoid overlap issues in which high levels of ORFK7 expression were observed in many previous studies that were actually due to PAN. While minimal expression of ORFK7 was detected, with a median of 267 TPM, a strong correlation was observed between the expression levels of ORFK7 and PAN in the KS tumors (R = 0.6401, P<0.0001). Using UCDS features, moderate levels of reads mapping to ORFK2 and ORFK5 were consistently detected in the vast majority of the KS tumors (median = 10,138 and 4,373 TPM, respectively) (Fig 9D; bracketed in red). Quantitative analysis of RNA reads mapping to the other KSHV genes in this region revealed 10 to 100-fold lower levels of transcripts with high variability across the 34 KS tumors (Fig 9D and S4 Table).
The transcript levels corresponding to each UCDS feature across the KSHV genome (described previously [25]) were compared for 34 of the KS tumor samples using a hierarchical clustering algorithm implemented in CIMminer [53]. This analysis was limited to the KS tumor samples with more than 1,000 total KSHV-mapped reads. The initial analysis showed expression of the genes in their order within the KSHV genome using the “equal width” binning method to map the TPM expression. This method divides the weightrange of data values into equal width intervals and each interval is mapped to one color for display in the clustered image map (Fig 10; left end genome-top, right end genome-bottom). Obvious high levels of transcripts were detected at the right end of the genome in the latency region, mapping to UCDS features for ORFK12 (UCDS-K12A), the adjacent direct repeat regions DR5 (UCDS-DR5) and DR6 (UCDS-DR6), the P3-Exon1 flanking ORF72 (UCDS-K12Aa), and the terminal ORF75 (UCDS-75) and ORFK15 (UCDS-K15a) (Fig 10). Elevated levels of PAN and ORFK2 transcripts were detected in some of the tumors at the left end of the genome. The hierarchical clustering analysis identified three clusters of KS tumors. Cluster I tumors displayed high level expression of transcripts containing the DR6 repeat region and minimal expression of PAN, while Cluster III tumors displayed high level expression of PAN and lower levels of expression of DR6 (Fig 10). Cluster II tumors were intermediate.
A second analysis was performed using the “quantile” binning method which divides the weightrange of expression data values into intervals each with approximately the same number of data points, spreading out the color differences in the image map. The KS tumors in Cluster I displayed low levels of gene expression throughout the KSHV genome outside of the latency region (Fig 11). In contrast, tumors in Cluster III showed moderate levels of lytic gene expression throughout the KSHV genome. Abnormally high expression of the genomic region ORFK3 to ORF19 was clearly observed in the KS tumor 008_B in this cluster. This is the region that was translocated downstream of the P3(LTd) latency promoter in this tumor. An obvious block of expressed genes was observed in the Cluster III tumors between the long-inverted repeat (LIR1) and PAN at the left end of the genome (Fig 11, top). The Cluster II tumors were intermediate with elevated gene expression across the majority of the KSHV genome. Both of the KS tumors from patient 030 showed unusually high expression of the genomic region from ORF39 through the right end of the genome. This patient was the only one to carry a KSHV strain with a K15 M-allele (Fig 4), as shown by the lack of reads mapping to the K15 UCDS.
The hierarchical clustering analysis with quantile binning showed an obvious pattern of transcription including expression of transcripts detected by UCDS features for ORFs K11, K2, K4, LIR1, K5, PAN, 38, 45, K8, K8.1, 57, 58, 59, and 65, in addition to latency transcripts K12A, DR5, DR6, miR, 71, 72 and K12Aa (Fig 11, labeled with red text). Other genes, such as ORFs 17.5, 27, 33 and 69 were expressed in a subset of tumors. Higher levels of reads mapped to genes that flanked poly-A transcription termination sites (Fig 11, indicated at right). Most of these genes are present in loci with bicistronic or polycistronic transcripts that terminate at the same site. Thus, the high level of reads mapping to the ORFs adjacent to the poly-A site, in most cases, is due to overlapping polycistronic transcripts that contain the sequences encoding the poly-A-flanking ORF, as we have shown previously [25]. Thus, the high levels of reads mapping to these poly-A-flanking ORFs would not correlate directly with their CAP-dependent protein expression.
The cutaneous KS tumor biopsies had been collected to represent different morphologies, sometimes from the same patient, including macular tumors with a flat appearance, nodular tumors with a distinct 3-dimensional structure and fungating tumors showing ulcerations and necrosis [54]. No correlations were observed between the morphological state and the level of total KSHV-mapped reads (Fig 1B), and the hierarchical clustering analysis revealed no obvious correlations between the morphological state and the patterns of gene expression (Fig 11). Cluster I was composed of 7 different KS tumors (5 macular; 1 nodular; 1 fungating) from 4 different individuals with a median read count of 38,605 (range 15,978–134,856). Cluster II was composed of 18 KS tumors (10 macular; 8 nodular) from 15 different individuals with a median read count of 25,538 (range 2,303–158,924). Cluster III was composed of 9 KS tumors (5 macular; 3 nodular; 1 fungating) from 8 different individuals with a median read count of 3,136 (range 2,548–146,773). Surprisingly, we observed a strong correlation between the gene expression patterns of KS tumor samples from the same patient, regardless of morphotype (Fig 11, bottom; red dot = macular, blue dot = nodular). Seven of 8 paired samples from the same individuals showed clustered gene expression profiles. The only paired sample that did not cluster was from individual 008. The 008_B sample in this pair (Fig 11, indicated with an asterisk at the bottom) exhibited the unusual high-level expression of the region between ORFK3 and ORF19, which was due to the translocation not present in the paired 008_C sample. Gene expression in the translocated region was mainly driven by the P3(LTd) promoter in the latency region, rather than by the normal gene specific promoters.
Principal component analysis (PCA) was used to reduce the complexity of the expression data from the 34 KS tumors. Principal components 1 (PC1) and 2 (PC2) captured 88.4% and 6.2% of the variation in the gene expression data, respectively. This analysis confirmed the grouping of gene expression detected by hierarchical clustering, where the tumor samples are indicated in blue text (Cluster I), black text (Cluster II) and orange text (Cluster III) (Fig 12A). As seen in the hierarchical clustering analysis, no similarities were detected between the patterns of gene expression and tumor morphotype in the PCA analysis (Fig 12A: morphotype of each tumor is color-coded). The PCA analysis revealed a gradient of KSHV gene expression from the more latent phenotype in Cluster I (upper right- hand quadrant) to the more lytic phenotype in Cluster III (lower left-hand quadrant). The similarity in the pattern of gene expression between the paired tumor samples from the same individual was confirmed, with the exception of sample 008_B (labeled with *), as discussed above (Fig 12A).
Since PAN expression is considered a marker of lytic activation, we compared PAN expression with the expression of the major T1.7A latency transcript. The RNA reads mapping to the PAN UCDS feature (size = 1.126 Kb) were compared to those mapping to the K12A UCDS feature (size = 0.183 Kb), which detects the Kaposin region in the major T1.7A latency transcript and overlapping transcripts. The reads were normalized to compare transcript levels using reads per kilobase (RPK), plotted as circular dots using the color code for morphotypes (Fig 12B). Because read data for these two features were present in all 41 of the KS tumor samples, we also compared the PAN/K12A ratio of the 7 KS tumor samples with less than 1,000 total KSHV mapped reads that were excluded in earlier expression analysis (plotted as triangles). Although the total number of KSHV-mapped reads was low in these samples, they showed an expression pattern that mirrored those seen in the lytic Cluster III (S1 Fig). The tumor sample 008_B with the altered genome was excluded. The three expression clusters showed statistically significant differences in the PAN/K12A ratio with median ratios of 0.046 (Cluster I; IQR = 0.009:0.084), 0.1545 (Cluster II; IQR = 0.0138:0.297) and 1.635 (Cluster III; IQR = 1.026:2.092) (Fig 12B), indicating that the PAN/K12A ratio could be used to distinguish the different tumor clusters. Notably, three of the four fungating samples showed high PAN/K12A ratios, compatible with a lytic phenotype.
The number of KSHV-mapped RNA reads present in each tumor sample was compared between tumors in the different expression clusters. KSHV-mapped reads were normalized to the total read depth of each tumor library, indicated as KSHV-mapped reads per 100 million library reads. The 7 KS tumors with less than 1,000 total KSHV-mapped reds were included in lytic Cluster III (triangles) due to the similarities in the PAN/K12A RPK ratios with the Cluster III tumors and the presence of lytic gene expression (S1 Fig). Both the latency Cluster I and intermediate Cluster II tumors showed high levels of KSHV-mapped reads in the tumor libraries (Cluster I median = 38,995, IQR = 18,393:139,027; Cluster II median = 25,059, IQR = 5,307:92,593). The lytic Cluster III tumors showed 23 and 10-fold lower levels of KSHV-mapped reads, respectively (Cluster III median = 1,694, IQR = 99:3,490) (Fig 12C). While the extremely high level of KSHV-mapped reads for the fungating tumor 023B (red dot) is shown in parentheses, it was excluded from the column analysis. The presence of high levels of KSHV-mapped reads in the KS tumors with latent phenotypes and the absence of KSHV-mapped reads in the KS tumors with lytic phenotypes is summarized graphically in Fig 12A.
A set of genes that were highly expressed across the set of 34 KS tumors was identified. Median transcript levels above 5,000 TPM were observed for 11 UCDS features targeting the latency region T1.7A (Kaposin A/B/C) spliced transcript and the ORF72/71 and ORF-K15/75 bicistronic transcripts, as well as the lytic region transcripts for PAN, K2 and K5 (Fig 13A). Median transcript levels of 1–5,000 TPM were observed for 14 other UCDS features targeting ORFs 11, OLAP, K4, 38, 45, 50, K8, K81, 57, 58, 59, 63, miR and 73 (Fig 13A). To identify biologically relevant KSHV gene regulation modules, the Pearson correlation was determined for each pair of highly expressed genes. The analysis was limited to the highly expressed genes as the read level across most of the other genes was low and variable. The 008_B tumor, which showed the unusual genomic translocation, was not included. Hierarchical clustering revealed groups of co-expressed genes that showed similar expression correlation profiles across the 33 KS tumor, indicated by cohesive purple squares along the diagonal (Fig 13B). These clusters of genes represent modules undergoing similar regulation of gene expression in the tumors. The orange areas showed clusters of genes that were negatively correlated. A strong correlated expression of the different transcripts derived from the latency P3(LTd) promoter was observed in Cluster A, including the spliced Kaposin A/B/C T1.7A transcript (detected with the K12A and K12Aa UCDS features), the unspliced bicistronic ORF71/72 T1.7B transcript (detected with the ORF 71 and 72 UCDS features), and the unspliced T6.5A transcript (detected with the miR UCDS feature (see Fig 5 for transcript details). The large Cluster C contains a number of co-regulated genes implicated in the lytic replication cycle, including genes involved in regulation (ORFs 45, 50, K8, 57, and PAN), immune modulation (ORFs K4, K5), replication (ORF59) and virion structure (ORFs 38, K8.1, 65). Other genes in Cluster C with unknown function include ORF58 and OLAP [55]. The expression of these genes is positively correlated with the expression of other genes in Cluster C, but negatively correlated with the expression of the other highly expressed genes. Cluster B contained a number of genes showing variable co-regulation, including K2 (vIL-6), ORF11, K15, ORF73 and ORF75. The expression of these genes negatively correlated with the expression of the P3 (LTd) transcripts in Cluster A and the lytic replication-associated transcripts in Cluster C.
Our study is the first to use RNAseq analysis to quantitatively evaluate KSHV transcription at a detailed individual gene level in in vivo KS tumors. Our analysis showed that the most highly and consistently expressed transcripts in the Ugandan KS tumors were derived from the downstream latency transcript promoter LTd flanking ORF72 [41, 49], herein designated as the third latency promoter—P3(LTd). Very high levels of reads mapped to each of the UCDS features for K12A(Kaposin), K12Aa(P3-Exon1) and the direct repeats DR5 and DR6 within the latency region of the genome. These features target different regions of a 1.7 Kb spliced transcript from the P3(LTd) promoter (herein referred to as T1.7A)[41, 49]. The read depths of the K12Aa (P3-Exon1) and K12A(Kaposin) features, which target the 5’ and 3’ ends of the T1.7A transcript, respectively, are quite similar, while the read depths of the DR5 and DR6 features were more variable, due to length of the UCDS feature, GC-content, sequence repeats in the DR5 region, sequence variability in the DR6 region and mismatches between RNA reads of the KS tumors and the KSHV GK18 reference sequence used for mapping. The presence of a single set of closely linked poly-A transcription termination signals in the region downstream of ORFK12 [52, 56], the strong correlation between the expression levels of the different regions of the T1.7A transcript in the different KS tumors, and the correspondingly high level of split reads mapping across the splice junction between upstream P3-exon1 splice donor and the downstream DR6 splice acceptor together support the conclusion that the P3(LTd) spliced T1.7A transcript is the most highly expressed transcript in the Ugandan KS tumors and was confirmed by RT-PCR analysis.
Early in situ studies of KS tumor lesions detected hybridization of a probe derived from the 0.7 Kb transcript encoding ORFK12 in every KS tumor examined [13, 14, 57–59]. Subsequent PCR and Northern analyses in cultured PEL cells indicated that transcripts hybridizing to the 0.7 Kb probe were almost always larger than 0.7 Kb, initiating from a promoter (P4) upstream of the GC-rich DR6 repeat region [44] (herein referred to as transcript T1.5A) or from the P3(LTd) promoter upstream of ORF72 (Transcript T1.7A) [41, 49]. The T0.7, T1.5A and T1.7A transcripts all terminate within a set of closely spaced poly-A termination signals (Group A). Our quantitative RNAseq and RT-PCR data provided strong evidence that the vast majority of transcripts hybridizing to the 0.7 Kb probe in KS tumors correspond to the spliced T1.7A transcript from the P3(LTd) promoter.
Previous In situ hybridization studies also detected strong signals with probes specific for the ORF71 and ORF72 latency genes in the majority of spindle cells in KS tumors [59, 60]. Subsequent analysis of latently-infected PEL cells identified three transcripts encoding ORFs 71 and 72, which are derived from the constitutive P1(LTc) promoter upstream of ORF73 and co-terminate at the poly-A site downstream of ORF71 (Group B transcripts)[6]. These transcripts include a full length 5.7 Kb unspliced tricistronic transcript (T5.7B in this study) encoding ORFs 73, 72 and 71, and three spliced transcripts, each containing the same short 5’ exon downstream of the P1(LTc) promoter that is alternatively spliced to either ORF73 producing large 5.2 and 5.4 Kb tricistronic ORF73/72/71 transcripts (T5.2B and T5.4B in this study) or ORF72 producing a 1.8 Kb bicistronic ORF72/71 transcript (T1.8B in this study). Although the tricistronic ORF73/72/71 transcripts have been detected in PEL cells in vitro using an ORF73 probe, in situ hybridization signals have not been observed in KS lesions in vivo, indicating the low abundance of tricistronic ORF73/72/71 transcripts encoding ORF73 LANA [61]. Our RNAseq analysis confirmed this observation as very few RNAseq reads from the KS tumors mapped to any of the tricistronic ORF73/72/71 transcripts and very few split RNA reads identified the splicing events downstream of the P1(LTc) promoter. Furthermore, no triscistronic transcripts were detected by RT-PCR analysis of the KS tumor RNAs. As the tricistronic ORF73/72/71 transcripts are the only transcripts known to encode the latency associated nuclear antigen ORF73/LANA, which is ubiquitously present in KSHV-infected KS spindle cells, the LANA protein appears to be more stable than its mRNA transcript. Thus, while the LTc promoter is constitutive in KSHV-infected PEL cells, its activity in the KS tumors appears to be quite low, regardless of morphotype or activation state.
We detected moderate and consistent levels of reads mapping to the ORF71 and ORF72 UCDS features in all the KS tumors, confirming the previous in situ hybridization studies. The RNA mapping and RT-PCR data indicate that the majority of these reads were derived from an unspliced bicistronic transcript encoding ORF72 at the 5’ end and ORF71 at the 3’ end (herein referred to as T1.7B), which has been previously characterized in cultured PEL cells [49, 50]. Like the spliced T1.7A transcript described above, the unspliced T1.7B transcript is transcribed from the P3(LTd) promoter but terminates at the poly-A site downstream of ORF71. A small upstream uORF in the initial 5’ region of this transcript has been shown to downregulate ORF72 expression, and a downstream IRES site facilitates expression of ORF71, resulting in coordinated expression of both ORF71 and ORF72 in PEL cells [43, 62]. As the unspliced T1.7B and spliced T1.7A transcripts are produced from the same P3(LTd) promoter, they both contain the same 5’ region encoding the small uORF upstream of ORF72, herein called P3exon1. A very high level of RNA reads mapped to the K12Aa UCDS feature that targets the P3exon1 region of both transcripts. The expression level of the bicistronic unspliced T1.7B transcript detected with the ORF71 and ORF72 UCDS features was only 10% of the expression level of the shared 5’ P3exon1, indicating that 90% of the P3(LTd) transcripts bypass the termination signal after ORF71 producing long initial pre-mRNA species, which terminate downstream of ORFK12 (the unspliced transcript is referred to herein as T6.5A). The majority of this mRNA is processed to remove the intron containing ORFs 72, 71 and microRNAs by splicing, generating the spliced T1.7A transcript described above. Read data mapping to the miR UCDS feature targeting the intronic region indicated that < 5% of the P3(LTd) transcripts correspond to the unspliced T6.5A precursor transcript.
A previous study suggested that the transcripts hybridizing to ORF71 and ORF72 probes in KS lesions were bicistronic spliced transcripts derived from the upstream P1 (LTc) promoter [6], as has been observed in PEL cells. TopHat2 analysis of RNA reads in our study detected little evidence for splicing of transcripts from the P1(LTc) promoter, and essentially no expression of the 5’ exon downstream of the P1(LTc) promoter. Thus, our RNA-seq data indicates that the transcripts encoding ORF72 and ORF71 in KS lesions are derived from the P3(LTd) promoter, not the P1(LTc) promoter. Note that the spliced T1.8B bicistronic ORF72/71 transcript from the P1(LTc) promoter and the unspliced T1.7B bicistronic ORF72/71 transcripts from the downstream P3(LTd) promoter are of similar size and could have been mistaken for each other by Northern analysis in previous studies. While the majority of transcripts detected using the ORF71 and ORF72 UCDS features appeared to be bicistronic, confirming previous data from PEL cells, our analysis provided evidence for expression in some tumors of monocistronic ORF71 (T0.9B) and ORF72 (T1.0C) transcripts from the P3(LTd) promoter that have been detected previously [42, 43].
Variable numbers of repeats and sequence heterogeneity in DR5/DR6 repeat region have been detected previously in other KSHV strains, which alter protein translation from the T1.7A mRNA [41]. Our RNAseq analysis revealed two nucleotide deletions in the overlapping T1.7A and T1.5 transcripts that are conserved in the Ugandan KSHV strains, both of which would alter the open reading frames in these transcripts from those described previously in BCBL-1 and other KSHV strains. In eleven Uganda KSHV strains typified by the Zambian KSHV strain ZM114, the putative protein products of the T1.5/T1.7A transcripts would be a BCBL-1 Kaposin B homolog and a novel protein, herein called Kaposin E, that would initiate with the same CUG codon as BCBL-1 Kaposin C but lack the fusion with Kaposin A. In fifteen other Uganda KSHV strains, typified by the Zambian KSHV strain ZM004, additional nucleotide changes in the T1.5A/T1.7A transcripts eliminate the CUG-initiated Kaposin B and C ORFs leaving Kaposin A as the only predicted translational product of this transcript. While other non-AUG translation initiation could be utilized in these KSHV strains, it is not clear what protein products would be translated. Due to the limited amount of RNA template in our biopsies and the repetitive nature of the sequence and the high GC content (in some regions >95%), we were unable to PCR amplify T1.7A transcripts from these or other KSHV strains to determine the exact nucleotide sequence. Since translation products from this mRNA could be the major KSHV-encoded proteins in the KS tumors, identification and functional characterization of the T1.7A-encoded proteins will be critical for understanding the role of KSHV in the KS tumors.
While the exact coding potential for the T1.7A transcripts in the Ugandan KSHV strains awaits functional analysis, it is clear that splicing of this transcript liberates a 4.8 Kb intron, which is the substrate for processing of the majority of the KSHV microRNAs [49]. Thus, a major outcome of the expression of the T1.7A transcript would be the production of this set of KSHV microRNAs [49]. Additional transcripts from the P1(LTc) promoter upstream of ORF73, including T1.8A and T1.6A, also generate the region encoding these microRNAs as an intron during splicing. However, we observed very low and variable expression of the P1(LTc) transcripts in the KS tumors, indicating that splicing of the T1.7A transcript from the P3(LTd) would generate the vast majority of processed microRNAs in the tumors. In addition to encoding the Kaposin A ORF downstream of the DR5/DR6 repetitive sequences, the T1.7A transcript is a precursor of the miR-K10a and miR-K12 microRNAs, which are embedded within the 3’ end of the transcript [46]. Thus, the RNAseq data indicates that a major functional role of P3(LTd) promoter in the establishment and progression of KS tumors would be to generate the precursor substrates for all the KSHV microRNAs. The P3(LTd) promoter has been shown to be constitutively active in PEL cells, and it is not clear whether its activity is increased by reactivation using RTA [41, 49, 50].
Unexpectedly, we identified one KS tumor in a sampling of paired tumors from individual 008, which contained a translocation of a 14 Kb region of the left end of the KSHV genome flanking the long-inverted repeat LIR1 into the LIR2 repeat within the latency region at the right end of the genome. This translocation is located downstream of the highly active P3(LTd) latency promoter, resulting in numerous novel transcripts from the P3(LTd) promoter containing the initial P3-Exon1 spliced to acceptor sites in the translocated K3, K5, and ORF17 genes. This translocation did not appear to inhibit the splicing of the highly expressed T1.7A transcript encoding the Kaposin A/B/C complex, even though the excised intron was 14 Kb longer. A spliced transcript containing the P3-Exon1 spliced to ORFK5 was the most highly expressed transcript in the 008_B tumor. Since the P3-Exon1 encodes the regulatory uORF, described above, the translation of protein products from the highly expressed novel spliced transcripts in this tumor could be downregulated. The majority of highly expressed genes in the translocated region were transcribed right to left in the same orientation as the transcripts from the P3(LTd) promoter from the same DNA strand. While elevated levels of reads also mapped to other genes in the translocated region, such as PAN, OLAP, and ORF18, these genes would have been transcribed from their proper promoters on the other strand in the opposite direction. As this tumor was not analyzed using a stranded RNA library, it was unclear whether the RNA reads mapped to PAN, OLAP and ORF18 transcripts from the opposite strand. No other KS tumor exhibited this translocation suggesting that the translocation was a unique event in the 008_B tumor. However, the existence of this translocation and the resulting altered transcription pattern indicates the plasticity of the KSHV genome and the strong role of the P3(LTd) promoter in the development of KS.
Using a PAN-specific UCDS gene feature, we detected expression of the PAN RNA in most of the 34 Ugandan KS tumors. Although the median expression level for PAN was high, it was 5-fold less than the expression level of the K12A (T0.7) region. The range of PAN expression levels was large, ranging from barely detectable in some tumors to extremely high in other tumors, with expression levels higher than K12A(T0.7) in 7/34 KS tumors. While previous in situ hybridization studies identified specific cells expressing PAN, our RNAseq analysis only detected the average expression level within the biopsied lesion. The expression of PAN correlated positively with the expression of a large number of lytic cycle genes, including K6 (vMIP-1), ORF59, K8 (bZIP), ORF25 (MCP), ORF26 (VP23), and ORF65 (SCP), which were previously detected by in situ hybridization in limited numbers of KS spindle cells. Only a weak correlation was detected between the expression of these lytic cycle genes and ORF50 (RTA). There was no correlation between the expression of these lytic genes and genes, including K2(vIL-6), ORF16(vBCL-2), K10 (vIRF-4), or K11 (vIRF-2), whose expression has also been observed infrequently in KS spindle cells. Surprisingly, the RNA-seq data showed a strong correlation between the expression of K2 (vIL-6) and the genes for the ORF11 tegument and the ORF4 complement control (KCP) proteins, indicating a common regulation of gene expression that was not shared with PAN or the other lytic genes. This data suggests that the expression of discrete subgroups of genes in the lytic pathway can be differentially regulated in the tumors.
A recent study by Tso et al. used RNAseq to analyze KS lesions from four HIV-infected individuals from Zambia and Tanzania that were undergoing anti-retroviral therapy [26]. These RNA samples were sequenced to a depth of 10–13 million total reads obtaining 718, 1,650, 3,441 and 17,202 reads (median = 2,545) mapping to the NC_009333 KSHV reference sequence. No obvious pattern of KSHV expression was reported across the four samples, presumably due to the limited sample size and variable KSHV read depths of the different biopsies. High levels of latency gene transcripts were reported, including ORF73 LANA, as well as an increased expression of viral immune modulation genes, including ORF-K2 (viral interleukin-6), ORF-K5 (modulator of immune recognition), ORF-K7 (viral inhibitor of apoptosis) and ORF75 (degradation of ND10 protein). In contrast, we analyzed RNA libraries from 41 KS biopsies at a 10-fold greater depth with a 5-fold greater depth of KSHV-mapped reads (median = 10,232; maximum = 158,924). The increased sample size and depth of sequencing allowed for clear patterns of KSHV expression to be detected in the panel of KS tumors from anti-retroviral therapy-naïve Ugandans in our study. We detected high levels of ORF75 and ORFK15 transcripts in all the KS biopsies, but only variable levels of ORFK2 and ORFK5 transcripts. We determined that 76% of the RNA reads mapping to ORF75 are derived from K15/ORF75 bicistronic transcripts, as K15 transcripts terminate downstream of ORF75 [52]. Since it is not known whether ORF75 protein is translated from the bicistronic transcripts through alternate initiation pathways, the biological relevance of the bicistronic transcripts to ORF75 function is unclear. While ORFK15 protein is abundantly expressed in KS tumor biopsies where it is believed contribute to the invasiveness and angiogenic properties of the tumor cells [34, 35], we have identified no reports of ORF75 tumor expression. Even though transcripts of both ORFK15 and ORF75 are highly expressed in the most latent KS tumors, both genes are required for viral lytic replication in vitro [32, 63], suggesting important pleiotrophic effects in the virus life cycle.
As we have indicated previously, mapping of RNAseq reads to the regions of the KSHV genome encoding entire ORFs, as performed in the Tso et al study, is problematic due to the compact size of the KSHV genome and overlapping nature of the KSHV transcripts [25]. While Tso et al detected high-level expression of ORF73 LANA by mapping reads to the entire ORF, we detected only minimal expression of ORF73 using UCDS features that avoided the large repetitive sequence within LANA. This raises the question whether the ORF73 expression in the Tso study could have been due to multiple mapping of reads to repetitive sequences. Our mapping protocol limited reads to a single mapping event to avoid this problem. As discussed above, a large T6.1 transcript derived from the positive strand of the KSHV genome encoding ORFK7 overlaps with the PAN RNA derived from the same DNA strand and ORFK5 transcripts from the opposite strand (see Fig 9B). Using the UCDS approach with non-overlapping PAN and ORFK7 UCDS features, we determined that the high-level expression of ORFK7 detected using the typical mapping to the NC_009333 ORFs is due to high levels of PAN transcripts, not ORFK7. With the UCDS approach, we detected high levels of PAN in some tumors with very low levels of ORFK7 transcripts in all tumors. Tso et al did not report PAN transcript levels as PAN is not an ORF in the NC_009333 accession record and therefore was not quantitated.
Using deep sequencing of RNA transcripts, our study quantitated the expression level of genes across the entire KSHV genome using UCDS features, providing a complete global analysis of the KSHV transcriptome in the KS lesions from HIV-infected Ugandans that were anti-retroviral therapy naive. Unsupervised hierarchical clustering and PCA analysis of gene expression revealed three distinct tumor clusters. Cluster I consisted of 7 tumors from four individuals with a mixture of morphotypes. These tumors showed a predominant latency phenotype with low levels of gene expression outside of the latency locus with 22 ORFK12 transcripts per PAN transcript. Cluster III consisted of seventeen tumors from fourteen individuals with a mixture of morphotypes, which showed a predominant lytic phenotype with gene expression detected across the entire KSHV genome. In contrast, these tumors expressed 1.6 PAN transcripts per ORFK12 transcript. Cluster II consisted of eighteen tumors from fifteen individuals with a mixture of morphotypes, which showed elevated gene expression across the right end of the genome and limited gene expression across the left end of the genome with 6.0 ORFK12 transcripts per PAN transcript. Surprisingly, the Cluster III KS tumors with the most lytic phenotypes contained the fewest number of RNA reads mapping to the KSHV genome, while the Cluster I tumors with the most latent phenotypes contained the highest number of KSHV-mapped RNA reads. This is in direct contrast to what was expected from previous studies on KSHV infection in vitro, where reactivation of latent virus was shown to initiate new lytic gene transcription but not alter constitutive latent transcription [64].
Of the 41 KS tumor samples sequenced, 7 had fewer than 1,000 KSHV-mapped reads, and 3 had fewer than 100 KSHV reads although all the tumors were advanced T1 tumor stage. We found that KS tumors with 15,000 to 150,000 total KSHV-mapped reads had similar viral loads of ~0.5 KSHV genomes/cell, suggesting that KSHV gene expression did not correlate with the level of KSHV infection in cells in the lesion. While the KS samples with less than 1,000 KSHV-mapped reads were excluded from hierarchical cluster analysis of the RNAseq data due to problematic normalization and clustering, we found that these tumors had a high PAN/K12 read ratio, consistent with a lytic phenotype. Thus, all the KS tumors except for one fungating sample (23B) showed a significant correlation between lytic gene expression across the entire KSHV genome and low levels of total KSHV transcription in the KS lesion. Interestingly, previous studies using double staining of KS lesions showed that vIL6- or vGPCR-positive KS spindle cells lacked punctate nuclear LANA spots indicating a decreased expression of LANA and possibly other latency proteins during lytic reactivation [21]. In a previous PCR profiling study of KSHV expression in KS tumors from Malawi, 10 of 61 tumors were excluded from analysis due to the lack of detectable mRNA for KSHV latent genes and three additional samples were excluded because they had low levels of KSHV latent mRNA [24]. Since these samples were excluded from analysis in this study, it is not known whether they displayed a lytic phenotype, as was seen for comparable tumor types using RNAseq analysis in our study.
It was previously reported that qPCR profiling of Malawian KS tumors detected abundant transcription within the viral latency locus including ORF73(LANA), ORF72(vCyc), ORF71(vFLIP), ORFK12(Kaposin) and microRNAs [24]. This study detected a poly-A site position effect in which the PCR signal for ORF71, which is proximal to the poly-A site, was higher than the distal ORF72 or ORF73. Since these signals were believed to originate from the same tricistronic mRNA transcript, it was concluded that poly-A purification of the mRNA templates had artifactually increased the transcript levels for the genes adjacent to the poly-A site. Using our RNAseq approach, we were able to characterize the latency region transcription in KS tumors in a more granular detail. We detected minimal numbers of RNA reads mapping to LANA on the tricistronic ORF73/72/71 latency transcript of the P1 promoter and showed that the vast majority of reads mapping to ORFs 71 and 72 were actually derived from the T1.7B bicistronic transcript of the P3/LTd promoter. These findings explain the results of the Hosseinipour study and indicate that the increased expression of ORFs 71 and 72 is not an artifact of poly-A purification.
An important consideration in our study was the use of the KSHV-GK18 reference sequence (NC_009333) to map the RNA reads from the different KSHV strains in the Ugandan KS tumors. We previously developed a detailed map of the KSHV transcriptome annotated to the NC_009333 reference sequence. We also developed a detailed gene feature file (GFF) for the NC_009333 sequence with UCDS features allowing quantitation of overlapping genes in the KSHV genome [25]. Since our phylogenetic analysis revealed the presence of multiple KSHV strains in the 34 Ugandan KS tumors, we decided to use the NC_009333 sequence as a common alignment target for consistency. We observed instances of sequence mismatches between KS tumor RNA reads and the aligned NC_009333 sequence throughout the KSHV genome, however, there were only a few obvious regions where the variations in read sequence affected transcript quantitation, as seen for K15, as described above. We observed some mapping issues within the K1 gene, which is known to exhibit high levels of variation in specific regions of the gene. However, there was sufficient homology between the reads and the GK18 sequence outside of the K1 variant regions to determine K1 expression. We also observed problems mapping reads to the DR5 and DR6 repeat regions, which have shown high levels of sequence variation. To compare sequences across problematic regions, we performed additional mapping studies using published KSHV strains with variant sequences as references.
In summary, quantitative RNAseq analysis using a unique set of UCDS gene features has provided an in-depth analysis of the KSHV transcriptome in 41 T1 stage KS lesions from 30 HIV-infected ART-naïve Ugandans, yielding a unique resource for subsequent analysis of specific transcripts by other approaches. Hierarchical clustering and PCA analysis of KSHV transcripts revealed three clusters of tumors displaying a gradient of KSHV gene expression ranging from minimal gene expression outside of the latency locus (latent expression) to wide-spread gene expression across the viral genome (lytic expression). Paradoxically, the tumors with the latent phenotype had high levels of total KSHV transcription while the tumors with the lytic phenotype had low levels of total KSHV transcription. Morphologically distinct KS tumors from the same individual showed similar KSHV gene expression profiles suggesting that the tumor microenvironment and host response played a determining role in the activation level of KSHV within the infected tumor cells.
HIV infected adults with KS were recruited from the Uganda Cancer Institute (UCI)/Hutchinson Center Cancer Alliance in Kampala, Uganda. Eligible participants had to be at least 18 years of age and have late stage (T1) KS by AIDS Clinical Trials Group staging criteria and be naïve for antiretroviral therapy (ART). All KS samples were obtained with written informed consent.
Total nucleic acids were extracted using RLT buffer (Qiagen) and RNA was isolated using the RNeasy mini kit with DNAse treatment step. Total RNA integrity was checked using an Agilent 2200 TapeStation (Agilent Technologies, Inc., Santa Clara, CA) and quantified using a Trinean DropSense96 spectrophotometer (Caliper Life Sciences, Hopkinton, MA). Unstranded RNA-seq libraries were prepared from 300 ng of total RNA using the TruSeq RNA Sample Prep Kit v2 (Illumina, Inc., San Diego, CA, USA). Four KS tumors libraries were prepared using the TruSeq Stranded mRNA Library Kit (Illumina) from 100 ng of total RNA. Library size distributions were validated using an Agilent 2200 TapeStation (Agilent Technologies, Santa Clara, CA, USA). Additional library QC, pooling of indexed libraries, and cluster optimization was performed using Life Technologies’ Invitrogen Qubit 2.0 Fluorometer (Life Technologies-Invitrogen, Carlsbad, CA, USA). The unstranded RNA-seq libraries were pooled (5-plex) and the stranded libraries were pooled (4-plex) and each pool was clustered onto a flow cell lane. Sequencing was performed using an Illumina HiSeq 2500 in “High Output” mode with a paired-end, 50 base reads (PE50) sequencing strategy for the unstranded libraries and non-paired end for the stranded libraries. Image analysis and base calling was performed using Illumina’s Real Time Analysis v1.18 software, followed by ‘demultiplexing’ of indexed reads and generation of FASTQ files, using Illumina’s bcl2fastq Conversion Software v1.8.2.
RNA from four KS tumors was analyzed by RT-PCR to detect different spliced and unspliced transcripts from the latency region of KSHV (S2 Fig). Forward PCR primers from the 3’ end of the ORF73 sequence (P1/2F, bp124121 5’ CCCTGCCATTAACCCAGCCAG 3’ bp124101) and the P3-exon1 sequence downstream of the P3 promoter (P3F, bp123950 5’ ACCCATCTACCTCAACTGAAC 3’ bp123930) were develop to use in conjunction with a reverse primer from the 3’ end of the ORF72 sequence (72R, bp123757 5’ CGATCCTCACATAGCGTGGGA 3’ bp123777) to amplify a 406 bp fragment of the tricistronic ORF73/72/71 (T5.2–5.7B) transcript and a 194 bp fragment of the bicistronic ORF72/71 (T1.7B) transcript, respectively. Reverse PCR primers from the 3’ end of the K12 sequence (KapR1, bp118037 5’ CGTTGCAACTCGTGTCCTGAATG 3’ bp118059) and the sequence spanning the exon junction of transcript T1.7A (P4R, bp119036 5’ TTATAGCGTTTC 3’ bp119047: bp123843 5’ CTGTAGAGCCTG 3’ bp123854), were developed to amplify 1,042 and 120 bp fragments of the T1.7A spliced transcript encoding the K12 Kaposin A/B/C region, respectively, using the P3F forward primer. A forward PCR primer from the sequence downstream of the P4 promoter flanking the DR6 repeat region (P4F, bp 118973 5’ TTGGATTTACACGTATCGAGG 3’ bp 118953) was developed to amplify a 186 bp fragment of K12 using the KapR1 reverse primer. PCR reaction conditions were developed using BCBL-1 DNA as template. cDNA was prepared from KS tumor RNA using Tetro reverse transcriptase (Bioline) and random decamers at 45 °C in the presence of RiboSafe (Bioline), and PCR amplification was performed using MyTaq HS polymerase (Bioline) and forward and reverse primers (5.0 μM), after denaturing at 95° 2.5 minutes with 45 cycles of 95° 15 sec/60° 30 sec/ 72° 60 sec). The PCR products were visualized by electrophoresis using ethidium bromide. Reactions without reverse transcriptase were performed to control for DNA contamination.
RNA reads mapping to the human genome (hg19) were removed using the Bowtie2 program (version 2.2.6) [65]. The remaining RNA reads were aligned to the KSHV reference sequence NC_009333 for the KSHV GK18 strain using TopHat2 (version 2.0.14) [66] in a local instance of Galaxy [67]. Mapped reads were visualized using the Integrated Genome Viewer (IGV; version 2.3.75) [68]. For quantitation purposes, the reads from paired-end libraries (non-stranded) were analyzed as unpaired (single-end data) to allow each read of the pair to map unambiguously to a single gene feature. The reads from both strands of the stranded libraries (non-paired end) were either concatenated and analyzed together (librarytype = unstranded) for comparison to the non-stranded paired end libraries or were analyzed separately (librarytype = FR) to show strand specificity. TopHat2 was used to detect splicing events ab initio. The default presets were used except that the maximum intron length was decreased to 10,000 and the maximum number of alignments allowed was decreased from 20 to 1, to avoid overcounting reads to repetitive regions. HTSeq (version 0.6)[69] was used to quantitate the reads mapping to the unique set of UCDS gene features within the novel revised gene feature file “KSHV NC_009333 UCDS ver 020116.GFF“(S1 File) [25]. The “intersection (non-empty)” setting in HTSeq was used to count all reads mapping completely or partially to a UCDS feature to maximize read count (featuretype = UCDS; IDattribute = gene). No reads were eliminated by ambiguity since the UCDS features were 50 bp apart, the length of a read. The read count was expressed as transcripts per million (TPM) by first normalizing the read count to reads per kilobase (RPK) by dividing the read counts by the length of the UCDS gene feature, in kilobases. The “per million” scaling factor was determined by summing all of the RPK values in a sample and dividing by 1,000,000. Each RPK value was then divided by the “per million” scaling factor to give TPM of mapped KSHV reads. Hierarchical clustering of TPM normalized expression levels was performed using the algorithm implemented in CIMminer [53]. Hierarchical clustering of the gene correlation matrix was performed by calculating the Pearson correlation between the normalized transcript levels (TPM) associated with each pair of UCDS gene features, using a script in R to create and output the correlation matrix. Shiny web applications were developed for R-based principal component analysis in Fig 12A (available at https://efg-ds.shinyapps.io/pcaApp/) and the boxplot analysis of gene expression levels in Fig 5C (available at https://efg-ds.shinyapps.io/boxplotApp/).
Phylogenetic analysis of the KSHV strains in the KS tumors was performed on the complete coding sequences of ORF75 (3,891 bp), which were assembled from the 50 bp RNA reads for 23 of the KS tumors with IGV, using maximum likelihood analysis. The sequences have been deposited in Genbank.
The biological samples in the study were obtained specifically for the study. All participants provided written informed consent. The protocol was approved by the institutional review boards at the Fred Hutchinson Cancer Research Center, the Makerere University School of Medicine, and the Uganda National Council for Science and Technology.
|
10.1371/journal.pmed.1002885 | Integrating preexposure prophylaxis delivery in routine family planning clinics: A feasibility programmatic evaluation in Kenya | Young women account for a disproportionate fraction of new HIV infections in Africa and are a priority population for HIV prevention, including implementation of preexposure prophylaxis (PrEP). The overarching goal of this project was to demonstrate the feasibility of integrating PrEP delivery within routine family planning (FP) clinics to serve as a platform to efficiently reach at-risk adolescent girls and young women (AGYW) for PrEP in HIV high-burden settings.
The PrEP Implementation in Young Women and Adolescents (PrIYA) program is a real-world implementation program to demonstrate integration of PrEP delivery for at-risk AGYW in FP clinics in Kisumu, Kenya. Between November 2017 and June 2018, women aged 15 to 45 from the general population seeking FP services at 8 public health clinics were universally screened for HIV behavioral risk factors and offered PrEP following national PrEP guidelines. We evaluated PrEP uptake and continuation, and robust Poisson regression methods were used to identify correlates of uptake and early continuation of PrEP, with age included as a one-knot linear spline. Overall, 1,271 HIV-uninfected women accessing routine FP clinics were screened for PrEP; the median age was 25 years (interquartile range [IQR]: 22–29), 627 (49%) were <24 years old, 1,026 (82%) were married, more than one-third (34%) had partners of unknown HIV status, and the vast majority (n = 1,200 [94%]) reported recent condom-less sex. Of 1,271 women screened, 278 (22%) initiated PrEP, and 114 (41%) returned for at least one refill visit after initiation. PrEP uptake was independently associated with reported male-partner HIV status (HIV-positive 94%, unknown 35%, HIV-negative 8%; p < 0.001) and marital status (28% unmarried versus married 21%; p = 0.04), and a higher proportion of women ≥24 years (26%; 191/740) initiated PrEP compared to 16% (87/531) of young women <24 years (p < 0.001). There was a moderate and statistically non-significant unadjusted increase in PrEP uptake among women using oral contraception pills (OCPs) compared to women using injectable or long-acting reversible contraception methods (OCP 28% versus injectable/implants/intrauterine devices [IUDs] 18%; p = 0.06). Among women with at least one post-PrEP initiation follow-up visit (n = 278), no HIV infection was documented during the project period. Overall, continuation of PrEP use at 1, 3, and 6 months post initiation was 41%, 24%, and 15%, respectively. The likelihood for early continuation of PrEP use (i.e., return for at least one PrEP refill within 45 days post initiation) was strongly associated with reported male-partner HIV status (HIV-positive 67%, -negative 39%, unknown 31%; overall effect p = 0.001), and a higher proportion of women ≥24 years old continued PrEP at 1 month compared with young women <24 years old (47% versus 29%; p = 0.002). For women ≥24 years old, the likelihood to continue PrEP use at 1 month post initiation increased by 3% for each additional year of a woman’s age (adjusted prevalence ratio [PR]: 1.03; 95% confidence interval [CI]: 1.01–1.05; p = 0.01). In contrast, for women <24 years old, the likelihood of continuing PrEP for each additional year of a woman’s age was high in magnitude (approximately 6%) but statistically non-significant (adjusted PR: 1.06; 95% CI: 0.97–1.16; p = 0.18). Frequently reported reasons for discontinuing PrEP were low perceived risk of HIV (25%), knowledge that partner was HIV negative (24%), experiencing side effects (20%), and pill burden (17%). Study limitations include lack of qualitative work to provide insights into women’s decision-making on PrEP uptake and continuation, the small number of measured covariates imposed by the program data, and a nonrandomized design limiting definitive ascertainment of the robustness of a PrEP-dedicated nurse-led implementation strategy.
In this real-world PrEP implementation program in Kenya, integration of universal screening and counseling for PrEP in FP clinics was feasible, making this platform a potential “one-stop” location for FP and PrEP. There was a high drop-off in PrEP continuation, but a subset of women continued PrEP use at least through 1 month, possibly indicating further reflection or decision-making on PrEP use. Greater efforts to support PrEP normalization and persistence for African women are needed to help women navigate their decisions about HIV prevention preferences as their reproductive goals and HIV vulnerability evolve.
| Adolescent girls and young women (AGYW) in Africa are disproportionately affected by HIV infections because of cultural, structural, biological, and behavior factors.
Preexposure prophylaxis (PrEP), as a highly potent and recommended discreet user-controlled HIV prevention strategy, has the potential to substantially reduce new HIV infections in African women if delivered with high coverage and if used with sufficient adherence.
However, data are limited on “real-world” implementation approaches to efficiently reach at-risk women who may benefit from PrEP.
Family planning (FP) clinics provide care to young women at risk for acquiring HIV and have in-built staffing, supply chain, and HIV testing access, which could contribute to more efficient PrEP implementation with less cost, but no study has evaluated this model in “real-world” settings.
We conducted a pilot open-label, “real-world” implementation program to evaluate the feasibility of integrating PrEP delivery into routine FP clinics to reach HIV at-risk young women.
General-population women accessing FP services were universally screened for HIV behavior risk factors and were counseled for PrEP by program-dedicated nurses embedded in 8 public health FP clinics in a high–HIV-prevalence region in Kenya.
We found that FP clinics can be an effective platform to efficiently reach HIV at-risk women who may benefit from PrEP.
PrEP screening was feasible, and 22% of the general population of women took PrEP home. There was a high drop-off in early PrEP continuation (41% PrEP continuation at 1 month) particularly among AGYW <24 years old, but a subset of women persisted on PrEP (25% at month 3 and 15% at month 6), with higher continuation among women with HIV-positive partners, possibly indicating further reflection or decision-making on PrEP use.
Women’s perceived risk for HIV, including having an HIV-positive partner, was an important driver of initiation and continuation on PrEP.
To our knowledge, this project provides the first demonstration of “real-world” delivery of PrEP for at-risk AGYW integrated into FP clinics in Africa, a priority population for HIV prevention.
These findings demonstrate that it may be feasible to integrate PrEP delivery in public health FP clinics, making this a platform potential “one-stop” location for FP and PrEP. With expanding PrEP awareness, uptake and continuation are likely to increase among women at risk.
Evidence from this work will inform next steps for wider delivery of PrEP and next-generation PrEP formulations (e.g., multipurpose HIV prevention and contraception technologies) in FP clinics, not only in Kenya but in other resource-limited settings globally.
| Young women in HIV high-burden settings are a priority population for HIV prevention because they account for a disproportionate fraction of new HIV infections [1]. Preexposure prophylaxis (PrEP) is a safe and highly potent intervention when taken daily and has the potential to substantially reduce new infections if delivered with sufficient coverage to populations with greatest HIV prevention needs [2–5]. As PrEP implementation gradually comes to scale in many HIV high-burden regions, care settings routinely accessed by young women could be leveraged as a platform for reaching this important at-risk group. Family planning (FP) clinics are a particularly attractive platform for integrating PrEP delivery because FP providers are uniquely positioned to counsel on PrEP as they already counsel women on sexual health services. Women also are routinely screened for sexual behavior and HIV risk factors in FP clinics, and PrEP screening could be integrated efficiently within this context. Importantly, PrEP can safely be used with commonly used hormonal contraceptives with no bilateral drug-drug interactions [6,7].
In September 2015, the World Health Organization (WHO) recommended PrEP as a prevention option for persons at high risk of HIV acquisition [8]. Subsequently, in July 2016, the Kenya Ministry of Health (MOH) released guidelines that recommended PrEP for all HIV-uninfected persons with substantial ongoing risk of HIV infection, including adolescent girls and young women (AGYW) as priority persons [9]. In Kenya and many other African settings, FP clinics already incorporate HIV prevention services such as HIV counseling and testing. However, limited data are available on implementation approaches on how to efficiently reach and counsel women for PrEP in these settings, particularly young women. Here, we report on uptake and early continuation of PrEP in a real-world implementation program integrated in routine FP clinic settings in Kenya.
The PrEP Implementation in Young Women and Adolescents (PrIYA) program was an implementation program to deliver PrEP to young women at substantial risk of HIV in Kisumu Kenya. The protocol (S1 File) and conduct of this project were fully compliant with the relevant Kenya MOH regulations and were approved by the Human Subjects Division of the University of Washington, the Kenyatta National Hospital Ethical Review Committee, and the Kisumu County administration and facility managers. Women provided verbal informed consent as is routinely done for standard of care services. This project is reported as per the Strengthening of the Reporting of Observational Studies in Epidemiology (STROBE) guideline (S1 Checklist).
The overall goal of the project was to demonstrate the feasibility of integrating PrEP delivery in public health maternal and child health (MCH) and FP clinics. PrIYA is part of the larger Determined, Resilient, Empowered, AIDS-free, Mentored, and Safe women (DREAMS) Innovation Challenge funded by the President's Emergency Plan For AIDS Relief (PEPFAR) managed by JSI Research & Training Institute. Following the release of the Kenyan national guidelines that recommended PrEP as part of standard of care HIV prevention [9], the Kenya MOH developed a national PrEP implementation framework and service provider toolkit in 2017 [10]. Kenya officially launched PrEP rollout nationally in May 2017. A preparatory phase for the PrIYA program commenced in July 2017, with full-scale implementation starting in November 2017. Between November 2017 and June 2018, in collaboration with the Kisumu County Government and the Kenya National AIDS and Sexually Transmitted Infection (STI) Control Programme (NASCOP), PrIYA operationalized PrEP counseling and delivery in 16 facilities (including 8 facilities with FP clinics as a delivery point for PrEP) in Kisumu County, Kenya. This region has an HIV prevalence of up to 28% among young women [11–13].
The current report details the operationalization of PrEP implementation integrated in routine FP clinics. The program targeted all HIV-uninfected women of reproductive age, 15 to 45 years old, seeking routine FP services in 8 high-volume public health FP clinics in Kisumu County, Kenya. Implementation clinics were selected based on high volume and geography in consultation with the Kisumu County health authorities: at pre-implementation assessment, the monthly volumes of women newly accessing FP services at these clinics were approximately <50, 50–100, and >100 at 1, 5, and 2 of the clinics, respectively. One clinic was classified as rural, 4 as semi-urban, and 2 as urban.
The primary implementation strategy was a PrEP-dedicated nurse-led delivery of counseling about HIV risk and provision of PrEP. Newly hired nurses were trained on HIV risk assessment, counseling, and PrEP provision using a 2-day case-based interactive Kenya MOH PrEP curriculum, and knowledge gain was assessed by pre- and post-test. Nurses only performed HIV risk counseling and provision of PrEP but did not participate in delivery of FP services. At nearly all of the 8 clinics, women first completed other services, including HIV testing, and were then referred to a PrEP-dedicated nurse. Specifically, women of reproductive age accessing FP services were universally counseled by a PrEP-program–dedicated nurse for HIV behavioral risk factors and willingness to consider PrEP for HIV prevention. Screening was conducted according to the Kenya PrEP national guidelines [9], guided by a Kenya MOH risk assessment screening tool (RAST) modified to include women’s self-assessed reasons for choosing or declining PrEP (S2 File); the tool was used only as a guide but not as a scoring tool for ruling in or out potential users. Behavioral factors defined by the Kenya PrEP guidelines to indicate a substantial ongoing risk of acquiring HIV include the following: (a) inconsistent or no condom use in the last 6 months; (b) having a high-risk sex partner(s) of unknown HIV status; (c) engaging in transactional sex; (d) history of ongoing intimate partner violence (IPV) and gender-based violence (GBV); (e) recent bacterial STIs, self-reported or etiologically diagnosed; (f) recurrent use of postexposure prophylaxis; (g) recurrent sex under the influence of alcohol and/or recreational drugs; (h) injection drug use with shared needles and/or syringes; and (i) having an HIV-positive partner [9]. Interested and medically eligible women were provided same-day PrEP initiation by the nurse.
Consistent with the programmatic nature of this work, visit schedules reflected approaches used in FP clinics to permit seamless integration in routine services in Kenya. Women initiated on PrEP were followed as per the Kenya national guidelines for PrEP, which include initiation, month 1, and then 3 monthly visits for clinical review. However, in order to cautiously manage PrEP commodities, most participants are dispensed monthly PrEP refills. PrEP commodities were supplied from the Kenya Medical Supply Authority.
For this implementation program, patient medical records were captured on standardized data collection tools including the MOH clinical encounter form for the clinical provision of PrEP for all populations in Kenya. Program data including women’s demographics, behavioral-risk characteristics, reported partner HIV status, PrEP uptake, self-reported adherence to PrEP, and adverse events were abstracted daily by program nurses. Continuation and adherence on PrEP was assessed by self-report and PrEP refill records at the clinic as well as through follow-up phone calls to ascertain PrEP continuation status and reasons for discontinuing PrEP. All data were entered into passward-protected tablets daily and uploaded to web-based encrytped Research Electronic Data Capture (REDCap) servers [14]. Internal quality control reports were run weekly to monitor program progress and discussed with clinics throughout program implementation. All relevant data underlying this manuscript are included in S1 Data.
The co-primary program implementation outcomes were the number of women screened for HIV behavioral risk factors (reach) and the number of women who initiated PrEP (uptake or adoption). Additional key outcomes were continuation on PrEP, behavioral risk profile of women, male-partner HIV status, reasons for declining PrEP, self-assessed adherence to PrEP, contraception method use, and correlates of PrEP initiation and early continuation. Early PrEP continuation was defined as return to clinic and PrEP refill within 45 days post initiation. We also summarized PrEP continuation rates at 3 and 6 months post initiation using available data from regularly scheduled follow-up visits.
Categorical variables were summarized as frequencies, and continuous measures were summarized as medians and ranges, as appropriate. Baseline demographic and behavioral factors characterized the HIV risk profiles of women screened for PrEP. Separately, we evaluated for the correlates of PrEP uptake and early PrEP continuation using Poisson regression methods with robust standard errors to generate prevalence ratios (PRs) and 95% confidence intervals (CIs) accounting for clinic clustering, an approach used when the occurrence of the outcome is high (>10%) [12,15]. For the correlates of PrEP uptake, age, marital status, reported male-partner HIV status, and contraception method were considered a priori to have substantial influence on uptake and were subsequently included in the multivariate analysis. However, because of sparse data involving the contraception methods variable (≤10% frequency of use for some contraception methods), contraception use was only evaluated at the unadjusted level. Age was fit as linear spline with a single knot at age 24, reflecting rate of change in outcome for each additional year of a woman’s age within each age group. To aid meaningful interpretation, we also present key outcomes stratified by age groups (i.e., ≥24 versus <24 years). A similar approach was used for the analysis of early PrEP continuation. Baseline HIV risk behavior covariates were evaluated for their independent effects on PrEP continuation if they had a p ≤ 0.2 in the unadjusted analysis. Self-reported adherence and reasons for discontinuing PrEP were presented overall and stratified by male-partner HIV status. Statistical analyses were conducted in Stata version 15 (Stata Corporation, College Station, TX).
Overall, we screened 1,271 HIV-uninfected women for behavioral risk factors and willingness to initiate PrEP among women accessing FP services in 8 clinics in Kisumu County, Kenya. The median age of women screened was 25 years (interquartile range [IQR]: 22–29), 8% (105/1,271) were <20 years old, and 49% (627/1,271) were <24 years old; 82% (1,026/1,271) were married, and more than a third (427/1,271) of the women did not know their male partners’ HIV status (Table 1). The vast majority of women (n = 1,200 [94%]) reported recent condom-less sex. Most women (1,121 [92%]) reported using some form of contraception at baseline; the most frequently used FP methods were injectable (56%), implants (31%), and oral contraception pill (OCP) (5%), with 3% using intrauterine devices (IUDs) and 2% condoms alone. Notably, 75% (45/60) of those who used OCPs were women ≥24 years old.
Of 1,271 HIV-uninfected women universally screened and counseled for PrEP, 278 (22%) initiated PrEP overall (Fig 1)—87 out of 531 (16%) among women <24 years old versus 191 out of 740 (26%) for women ≥24 years old (p < 0.001). Women who initiated PrEP had behavioral risk factors for HIV as defined by the Kenya national PrEP program guidelines; the most frequent HIV behavioral risk factors were recent condom-less sex (264/278; 95%), having a male partner of unknown HIV status (151/278; 54%), and having an HIV-positive male partner (61/278; 22%). History of IPV or being forced to have sex in the prior 6 months was reported by <10% of women, respectively.
Overall, age, reported male-partner HIV status, and marital status were baseline factors independently associated with PrEP uptake (Table 2). All but 4 of 65 (94%) women with HIV-positive partners initiated PrEP compared to 35% (151/427) of women with partners of unknown HIV status and 8% (65/722) for women with negative partners (p < 0.001). Notably, women with a positive male partner tended to be older compared to women with partners either negative or of unknown status (median age: 30 versus 24 years; p < 0.001). Similarly, a higher proportion of women ≥24 years old (26%; 191/740) initiated PrEP compared to 16% (87/531) among women <24 years old (p < 0.001). For women ≥24 years, the likelihood of initiating PrEP increased by about 3% for each additional year of a woman’s age (adjusted PR: 1.03; 95% CI: 1.0–1.05; p < 0.001). In contrast, for women in the group <24 years old, the likelihood of initiating PrEP for each additional year of age was approximately 6% but statistically non-significant (adjusted PR: 1.05; 95% CI: 0.99–1.12; p = 0.12). PrEP initiation was also independently higher among unmarried women compared to married women (28% versus 21% for married; p = 0.03). There was a moderate and statistically non-significant higher likelihood for PrEP uptake among women using OCP compared to women using injectable or long-acting reversible hormonal contraceptive (i.e., implants and IUD) methods (28% versus 18%; p = 0.06). Among women who declined PrEP but had at least one behavioral HIV risk factor as defined by national PrEP guidelines (n = 954), the most frequently reported reasons for not initiating PrEP were low perceived risk of acquiring HIV (43%), knowledge that their male partner was HIV negative (47%, among HIV-uninfected women), needing to consult male partner (21%), and pill burden (13%). Fear of IPV or side effects were reported by less than 5% of women who did not initiate PrEP. Women with partners of unknown HIV status who declined PrEP (n = 276) frequently reported needing to consult their partners (41%), low perceived risk of HIV (24%), and pill burden (21%) as reasons for declining PrEP. Of the 4 women with HIV-positive partners who chose not initiate PrEP, one had a partner who was virally suppressed, one needed to consult her partner, one had concerns about pill burden, and no reason was recorded for the fourth woman.
Overall, among all women who initiated PrEP (Table 3), 114 out of 278 (41%) returned to collect at least one PrEP refill within 45 days post initiation (Fig 2; continuation for key subgroups), 28 out of 278 (10%) returned to report discontinuing PrEP, and 136 out of 278 (49%) did not return for a PrEP refill.
Overall, among women with at least one post-PrEP initiation up visit (n = 278), no HIV infection was documented during the project period. Self-assessed adherence among women continuing PrEP at month 1 was reported as good (0–3 doses missed in a month) by 93% of the women, highest among women with HIV-positive male partners (partner HIV status: positive 100%, unknown 90%, negative 88%; p = 0.05). Overall, in adjusted analyses, early PrEP continuation (i.e. month 1) was strongly associated with reported male-partner HIV status with higher continuation among women with HIV-positive partners (HIV status: positive 67%, negative 38%, unknown 31%; p for overall effect = 0.001). Similarly, a higher proportion of women ≥24 years old continued PrEP 1 month post initiation compared with women <24 years old (47% versus 29%; p = 0.002). For women ≥24 years old, the likelihood of continuing PrEP at 1 month post initiation increased by approximately 3% for each additional year of a woman’s age (adjusted PR: 1.03; 95% CI: 1.01–1.05; p = 0.01). In contrast, for women <24 years old, the likelihood of continuing PrEP for each additional year of a woman’s age was high in magnitude (approximately 6%) but statistically non-significant (adjusted PR: 1.06; 95% CI: 0.97–1.16; p = 0.18). Overall, a pattern similar to 1-month continuation was observed for continuation with PrEP use at 3 and 6 months post initiation: 68 of 278 (24%) and 29 of 192 (15%), respectively. For covariates only assessed at the unadjusted level, women in polygamous marriage (n = 37; 62% versus 38% in monogamous married; p = 0.02) and those who engaged in transactional sex practices (67% versus 40% of those not engaged in transactional sex; p = 0.02) were more likely to continue PrEP, but these differences did not persist in the multivariable assessment. There were no discernable statistical differences in early PrEP continuation based on reported history of condom use, STI diagnoses, and IPV or GBV. For continuation at month 3 and 6, only knowledge of male-partner HIV status was statistically significantly associated with continuation of PrEP use in adjusted analyses. Among women who discontinued PrEP (n = 123), the most frequently reported reasons for discontinuation were low perceived risk of acquiring HIV (25%), finding out that their male partner was HIV negative (24%), experiencing side effects (20%), and pill burden (17%); few women reported discontinuing PrEP due to fear of IPV (7%).
In this real-world PrEP implementation program, integration of universal screening and counseling for PrEP in FP clinics resulted in PrEP uptake of 22% among HIV-uninfected women, overall, and 16% in AGYW from the general population. Women who initiated PrEP frequently had self-reported behavioral risks for HIV, and more than 40% continued PrEP use beyond 1 month, a continuation rate higher than recently reported from other programs targeting young women within the region [16,17].
FP clinics offer an opportunity for integration of a full complement of sexual and reproductive health services, including PrEP provision and management of STIs, particularly because risk behaviors for unintended pregnancy are similar to those for HIV and STIs and interest in prevention may also extend from pregnancy to HIV/STIs. A recent large clinical trial of contraceptive use and HIV acquisition (ECHO Study) emphasized that HIV risk is high for FP clinic attendees and called for integration of HIV prevention into FP settings [18]. To our knowledge, ours is the first evidence of real-world programmatic delivery of PrEP integrated in routine FP clinics in high–HIV-prevalence settings.
Awareness of PrEP and the individuals’ perceived risk for HIV are important drivers of PrEP initiation and continuation. We found that older women were more likely to perceive or self-assess to be at risk for HIV than AGYW. Women who reported an HIV-positive male partner or those who self-assessed to be at risk of acquiring HIV frequently initiated and continued PrEP. Notably, a substantial proportion of women had partners of unknown HIV status. Many of these women still felt they needed to consult their male partners before they could consider PrEP. Similar to contraception, women have diverse preferences for HIV prevention and need to be empowered to make informed decisions as they strive to achieve their sexual and reproductive health goals while mitigating risks for HIV acquisition. As PrEP comes to scaled implementation, in addition to efforts to create demand, equal priority should be placed on implementation strategies that support women to better evaluate and understand their own risk for HIV, especially AGYW. Such strategies may include distribution of HIV self-test kits to women to efficiently promote and reach male partners for HIV testing [19,20], investing in strategies to increase community PrEP awareness to normalize and minimize stigma for pill taking for HIV prevention in communities where women live, and accelerating delivery of proven HIV prevention options to satisfy the diverse preferences of women and their partners. As new PrEP technologies emerge, including different delivery options such as the dapivirine ring or combined FP and PrEP options, our work will provide informative and important first steps for building robust and integrated FP and HIV prevention systems, including PrEP provision to women at substantial HIV risk in this region.
As expected, a majority of women screened who subsequently initiated PrEP were either using injectable or long-acting reversible contraception methods (i.e., implant or IUD), and pill burden was a common reason for declining PrEP. We also found that younger women used OCP less frequently than older women and that those using OCPs were more likely to initiate PrEP than women using injectable or long-acting reversible contraception methods, possibly because they may have already navigated personal barriers for taking oral medications. Taken together, these findings are important for guiding new directions for PrEP formulation and delivery that respond to the needs of women for whom a daily pill may not be a viable prevention option.
In Africa, PrEP is being added to an already burdened health infrastructure. The ability to build sustainable PrEP programs necessitates making PrEP provision cost-effective and efficient. FP clinics are uniquely important platforms to efficiently reach at-risk women who may benefit from PrEP given that similar factors predispose women to unintended pregnancies and susceptibility for HIV acquisition. Although FP visits are also extremely busy, efficient systems for HIV prevention including PrEP provision can be built into existing routine services. Such implementation strategies may include less frequent PrEP visits and expanding the pool of providers who might be able to screen and provide PrEP beyond the few clinicians and nurses (e.g., training and empowering HIV testing counselors and community health workers or peer educators). These approaches have already been successfully implemented to expand access to injectable and implants contraceptive methods in FP clinics in many African countries using community health workers [21–24]. Similar approaches, commonly described as differentiated care services, are currently being promoted for stable virally suppressed HIV-infected persons in many HIV treatment programs in Africa [25,26].
This work has limitations. First, as for any implementation program, data were collected on standard MOH clinical encounter forms, and thus assessment for correlates of uptake and continuation are limited to covariates included on that standard tool. Second, we assumed that women agreeing to initiate and continue PrEP represented acceptability, but we did not explicitly conduct qualitative interviews with women for insight into young women’s choices, behaviors, beliefs, acceptability, experiences, and priorities as it relates to PrEP. However, an ongoing sister qualitative project that will include women and nurses who participated in this program will provide this contextual information. Third, in the absence of a clear comparator or randomized design coupled with the limited number of measured covariates imposed by the program data, the robust effectiveness of a PrEP-dedicated nurse-led implementation strategy could not be definitively ascertained. Despite these limitations, this work was executed with high rigor consistent with the implementation nature of the program and provides novel evidence for advancing HIV prevention for at-risk adolescents and young women.
In conclusion, integration of universal screening for HIV behavior risk factors and counseling for PrEP in routine FP clinics in Kenya was feasible and resulted in reasonable uptake and continuation in a general population of women including AGYW accessing FP services, which is comparable to continuation of PrEP use observed in general populations in other settings. The enthusiasm for PrEP and evidence demonstrated from this work will set the stage for next steps for full-scale PrEP delivery in FP clinics not only in Kenya but in other settings in Africa. Importantly, this work will lay the foundation for delivery of the next-generation women-controlled PrEP formulations to at-risk young women in this setting.
|
10.1371/journal.pgen.1000144 | Patterns of Positive Selection in Six Mammalian Genomes | Genome-wide scans for positively selected genes (PSGs) in mammals have provided insight into the dynamics of genome evolution, the genetic basis of differences between species, and the functions of individual genes. However, previous scans have been limited in power and accuracy owing to small numbers of available genomes. Here we present the most comprehensive examination of mammalian PSGs to date, using the six high-coverage genome assemblies now available for eutherian mammals. The increased phylogenetic depth of this dataset results in substantially improved statistical power, and permits several new lineage- and clade-specific tests to be applied. Of ∼16,500 human genes with high-confidence orthologs in at least two other species, 400 genes showed significant evidence of positive selection (FDR<0.05), according to a standard likelihood ratio test. An additional 144 genes showed evidence of positive selection on particular lineages or clades. As in previous studies, the identified PSGs were enriched for roles in defense/immunity, chemosensory perception, and reproduction, but enrichments were also evident for more specific functions, such as complement-mediated immunity and taste perception. Several pathways were strongly enriched for PSGs, suggesting possible co-evolution of interacting genes. A novel Bayesian analysis of the possible “selection histories” of each gene indicated that most PSGs have switched multiple times between positive selection and nonselection, suggesting that positive selection is often episodic. A detailed analysis of Affymetrix exon array data indicated that PSGs are expressed at significantly lower levels, and in a more tissue-specific manner, than non-PSGs. Genes that are specifically expressed in the spleen, testes, liver, and breast are significantly enriched for PSGs, but no evidence was found for an enrichment for PSGs among brain-specific genes. This study provides additional evidence for widespread positive selection in mammalian evolution and new genome-wide insights into the functional implications of positive selection.
| Populations evolve as mutations arise in individual organisms and, through hereditary transmission, gradually become “fixed” (shared by all individuals) in the population. Many mutations have essentially no effect on organismal fitness and can become fixed only by the stochastic process of neutral drift. However, some mutations produce a selective advantage that boosts their chances of reaching fixation. Genes in which new mutations tend to be beneficial, rather than neutral or deleterious, tend to evolve rapidly and are said to be under positive selection. Genes involved in immunity and defense are a well-known example; rapid evolution in these genes presumably occurs because new mutations help organisms to prevail in evolutionary “arms races” with pathogens. Many mammalian genes show evidence of positive selection, but open questions remain about the overall impact of positive selection in mammals. For example, which key differences between species can be attributed to positive selection? How have patterns of selection changed across the mammalian phylogeny? What are the effects of population size and gene expression patterns on positive selection? Here we attempt to shed light on these and other questions in a comprehensive study of ∼16,500 genes in six mammalian genomes.
| Positive darwinian selection is an important source of evolutionary innovation and a major force behind the divergence of species. The Neutralist-Selectionist debate of the past 30 years has gradually given way to a general consensus that both neutral drift and positive selection play major roles in evolutionary change. Interest has therefore shifted to questions of which genes positive selection has affected, how strong was the effect, when did it occur, and what were its functional consequences. Heightening interest in these questions is a growing appreciation that methods for detecting positive selection can also be valuable tools for gaining insight into gene function [1]. Consequently, a wide variety of methods for detecting positively selected genes (PSGs) have been introduced, including comparative or phylogenetic methods, which make use of patterns of substitutions between species, and population genetic methods, which primarily rely on patterns of intraspecies polymorphism [2],[3]. Using these techniques, strong evidence of positive selection has been found for various genes in various organisms, including many genes involved in sensory perception, immunity, host-pathogen interactions, and reproduction (reviewed in [1],[3]).
Phylogenetic and population genetic methods for detecting positive selection serve as complementary tools for functional and evolutionary analysis. These methods operate at different time scales, with phylogenetic methods being best suited for detecting selection that operates over relatively long periods in evolutionary history, and population genetic methods being best suited for detecting more recent selection. Population genetic methods can potentially detect selection operating at individual sites, through the effects of linkage with flanking alleles, while phylogenetic methods generally require multiple sites to have been affected in a sequence of interest. At the same time, decay of linkage disequilibrium at longer evolutionary time scales can allow phylogenetic methods to more accurately pinpoint the specific locations of functionally important substitutions. In some cases, phylogenetic methods also allow such substitutions to be mapped to particular branches of a phylogenetic tree, thereby providing useful insights about the evolutionary histories of the sequences in question.
With the availability of multiple complete genome sequences, it has become possible to apply phylogenetic methods for the detection of positive selection at a genome-wide scale. Within mammals, several genome-wide scans for positive selection on protein-coding genes have been conducted, using both phylogenetic [4],[5],[6],[7],[8],[9] and population genetic [10],[11],[12],[13],[14],[15] methods (reviewed in [16]). These analyses have identified many new genes showing strong evidence of positive selection and have revealed striking differences in the prevalence of positive selection on different lineages and among different classes of genes. For example, it has been reported that PSGs are enriched for roles in sensory perception, immunity and defense, tumor suppression, apoptosis, and spermatogenesis [4],[5]; that PSGs are associated with known Mendelian disorders [4]; that PSGs often coincide with segmental duplications [8]; and that more genes have undergone positive selection in chimpanzee evolution than in human evolution [9]. Genome-wide scans for PSGs have also helped to stimulate interest in detecting positive selection on noncoding sequences and on gene expression [17],[18],[19],[20].
Nevertheless, much remains to be learned about positive selection in mammalian genomes, even within protein-coding regions. The most comprehensive scans for PSGs so far [4],[5],[7],[8],[9] have been based on at most three genome sequences—typically the highly similar human, chimpanzee, and/or rhesus macaque genomes (>97% average identity in orthologous coding regions [8]). As a result, the power for detection of PSGs has been relatively weak [5],[8]. In addition, in several of these studies, at least one genome was of draft quality, which reduced the number of genes that could be examined and required additional care in avoiding false positive predictions.
Here we present a phylogenetic analysis of positive selection in the six eutherian mammalian genomes for which high-coverage, high-quality sequence assemblies are now available: the human [21], chimpanzee [6], macaque [8], mouse [22], rat [23], and dog [24] genomes. The phylogenetic depth of this data set helps considerably in addressing the problem of weak power. Indeed, these genomes have a near-optimal degree of divergence for PSG detection, being distant enough to produce a strong phylogenetic signal, yet close enough that gene structures are well conserved, alignment is fairly straightforward, and synonymous substitutions are not saturated (e.g., [25]). In addition, our data set for the first time allows positive selection of mammalian genes to be examined genome-wide on a nontrivial phylogeny, so that insight can be gained into the particular “selection histories” of individual genes—that is, the branches of the phylogeny on which they experienced positive selection. In our analysis, we employ models of codon substitution that account for variation of selective pressure over branches on the tree and across sites in a sequence, which can capture signatures of molecular adaptation that affect small numbers of sites [26],[27]. Using a series of likelihood ratio tests (LRTs) based on these models, we identify more than four hundred genes that show strong signatures of positive selection during mammalian evolution. Our detailed analysis of the functional roles, selection histories, and expression patterns of these genes follows.
Using the latest human, chimpanzee, macaque, mouse, rat, and dog genome assemblies, we identified 17,489 human genes with high-confidence orthologs in at least two of the remaining five species. These ortholog sets (human genes and non-human orthologs) were identified by an automatic pipeline that made use of syntenic whole-genome alignments, sequence quality scores, and other data (see Methods). Briefly, the pipeline began with 21,115 human genes drawn from the RefSeq [28], UCSC Known Genes [29], and VEGA [30] gene sets. These genes were mapped to the other genomes via syntenic pairwise alignments, then passed through a series of rigorous filters to ensure correct mapping, high sequence quality, and only minimal changes between species in gene structure. This approach exploits the fact that gene structures are generally well-conserved between mammalian species [22] and avoids any dependency on the non-human gene annotations, which—with the exception of mouse—are significantly less accurate and complete than those for human. Because low-quality sequence can produce a spurious signal for positive selection (e.g., [8]), all bases with low quality scores (Phred quality <20) were masked out for subsequent analyses. Masking (or truncation at the 5′ or 3′ end) was also used to exclude regions of genes in which minor differences in gene structure were apparent. Genes that showed signs of substantial disruptions to their exon-intron structures or open reading frames in one or more species (perhaps indicating pseudogenization) were masked out completely in those species. All masked bases were treated as missing data in the subsequent analysis of positive selection. This masking approach allowed the number of genes to be maximized while ensuring that the analyzed alignments were of high quality (Table 1).
For this study, we chose to avoid recently duplicated gene families and to focus on 1∶1 orthologs. This simplified the analysis, allowed for parameter sharing across genes (see Methods), and eliminated an important source of error by avoiding the need for a separate tree reconstruction for each gene family. (All ortholog sets were assumed to obey the species tree shown in Figure 1; because only an unrooted tree is needed, the topology is well accepted.) It was therefore necessary to discard any genes that showed evidence of recent duplication. This was accomplished in a pairwise fashion, by examining each human gene and orthologous non-human gene, and determining—based on BLAST matches to other genes and gene predictions in the same genome—whether either gene had a paralog that was more similar to it than the two orthologs were to each other (see Methods). Requiring that each human gene had a high-confidence 1∶1 ortholog in at least two other species reduced the total number of ortholog sets to 16,529. These sets contain a human gene and either five (42% of cases), four (28%), three (15%) or two (15%) non-human orthologs.
We performed a series of nine different LRTs to identify genes under positive selection on particular branches or clades of interest in the six-species phylogeny. In particular, we tested for selection on any branch of the tree (Figure 1A); on the branch leading to, and on any branch within, the primate clade (Figure 1B,C); on the branch leading to, and on any branch within, the rodent clade (Figure 1D,E); and on each of the four individual branches within the primate clade (Figure 1F–I). These LRTs were all based on widely used site or branch-site models of codon evolution [31],[26],[27] (see Methods). The test for all branches was applied to all 16,529 ortholog sets. For the branch- and clade-specific tests, ortholog sets were discarded if they did not contain adequate in-group or out-group data for the test in question, which somewhat reduced the number of tests (Text S1, Table S1).
The PSGs identified by each test ranged in number from only seven (the hominid branch) to 400 (the test for all branches; FDR<0.05 in all cases). As in previous studies, the numbers of genes identified by the tests for individual primate branches were small, primarily due to weak power caused by low levels of inter-species divergence. The inclusion of additional non-primate mammals does not appear to have improved the power of these tests substantially, but it does allow a distinction to be made between selection on the branches to the hominids and to macaque. The tests for selection on the branch to the primates and in the primate clade also yielded fairly small numbers of PSGs, but the tests for selection in, or on the branch to, the rodents identified somewhat (nearly three-fold) larger numbers. In general, even with the larger data set, our power to detect selection on individual lineages and clades is still fairly weak, and differences in numbers of identified PSGs almost certainly reflect differences in power more than differences in the prevalence of selection. Nevertheless, these LRTs together produced a fairly large set of high-confidence PSGs, permitting a more detailed and thorough functional analysis than has previously been possible in mammals (see below).
The identified PSGs are significantly enriched for a large number of functional categories, according to the Gene Ontology (GO) [32] and Protein Analysis Through Evolutionary Relationships (PANTHER) databases (Tables 2, S2, and S3). If these over-represented categories are clustered by the PSGs that are assigned to them, major groups corresponding to sensory perception, immunity, and defense emerge (Figure 2), in agreement with previous genome-wide scans [4],[5]. However, the increased power of our analysis allows biological processes and functions associated with positive selection to be identified at much finer resolution than in previous analyses, as discussed below. The increased power also seems to diminish the dependency of functional enrichments on the database or statistical methodology selected for the analysis. In particular, better agreement was observed between functional categories over-represented among the identified PSGs, as determined by Fisher's exact test (FET), and categories whose genes displayed significant shift toward smaller LRT P-values (whether or not they met the significance threshold for PSGs), as determined by the Mann-Whitney U (MWU) test (see Methods). Better agreement was also observed between analyses based on the GO and PANTHER databases (see Tables S2 and S3). The observed enrichments do not appear to be an artifact of differences between categories in gene length or alignment depth per gene (Text S1). In the discussion below, we focus on GO categories and nominal P -values based on the MWU test, as applied to P-values from the LRT for selection on any branch of the tree (except when otherwise indicated); full results are shown in Table 2 and Text S1.
The PSGs are enriched for a wide variety of functions related to immunity and defense. Several over-represented categories describe activation in response to external or environmental stresses, such as from bacteria (P = 4.2×10−8), viruses (P = 3.0×10−8), wounding (P = 3.2×10−8), and acute inflammation (P = 4.7×10−11). In some cases, different categories reflect the same or very similar sets of genes (e.g., “response to wounding” and “acute inflammatory response,” or “response to virus” and “response to bacterium”), while in others they reflect quite distinct gene sets (“response to wounding” and “response to virus”) (Figure 2). Genes involved in both innate (P = 1.9×10−9) and adaptive (P = 1.5×10−5) immunity are over-represented, with many PSGs contributing to both classes. The conventional division of adaptive immunity into humoral (P = 1.6×10−7) and cellular (P = 3.5×10−7) responses is reflected in the enriched GO categories. Various mechanisms of immune response are represented, including previously identified categories for natural killer cell (P = 1.6×10−8), B-cell (P = 4.8×10−7), and T-cell (P = 1.2×10−8) mediated immunity [5],[8], and new categories such as cytokine/chemokine-mediated (7.6×10−8) and complement-mediated immunity (P = 6.0×10−6; see Table S3).
Some of the enriched categories point to particular pathways with large numbers of PSGs. A striking example is the complement immunity system, a biochemical cascade responsible for the elimination of pathogens. This system consists of several small proteins found in the blood that cooperate to kill target cells by disrupting their plasma membranes. Of 28 genes associated with this pathway in KEGG [33], nine are identified as PSGs (FDR<0.05), and five others have nominal P<0.05 (Figure S1). Most of these PSGs are inhibitors (DAF, CFH, CFI) and receptors (C5AR1, CR2), but some are part of the membrane attack complex (C7, C9, C8A), which punctures cell membranes to initiate cell lysis. Many of these PSGs are known to interact with one another, suggesting possible co-evolution. Two of three biochemical pathways known to activate the complement system are also enriched for PSGs (the classical complement pathway [P = 6.1×10−7] and the alternative complement pathway [P = 1.5×10−6]), as is the coagulation cascade that interacts with the complement system (“blood clotting,” MWU P = 2.2×10−7; Table S3). Other pathways that contain multiple interacting PSGs include those for apoptosis, taste transduction, antigen processing and presentation, and cytokine- and chemokine-mediated signaling (e.g., Figures. S4, S5).
Several gene families of the immunoglobulin superfamily (“immunoglobulin mediated immune response,” P = 1.1×10−7) show particularly strong enrichments for PSGs. For example, five of the six SIGLEC genes included in our analysis are under positive selection (see [34]). A detailed examination of one immunoglobulin gene for which structural information was available—a cell-surface receptor for hepatitis A and other viruses called HAVCR1 (LRT P = 6.9×10−9)—revealed several sites under positive selection in its N-terminal V-like immunoglobulin (IgV) domain. Three of these sites correspond to regions of the protein believed to play critical roles in binding to viruses or in regulating the immune function of the gene (Figure 3). In addition to its role in viral defense, HAVCR1 is a key player in the hygiene hypothesis explaining the increase in allergies and asthma [35]. It also interacts with IgA (CD79A; P = 5.4×10−9), whose deficiency is associated with increased susceptibility to autoimmune and allergic diseases [36].
The hierarchical clustering of GO categories (Figure 2) reveals an unexpected similarity between the sets of PSGs involved in fertilization and cytolysis, and some similarity of both sets with immune-related PSGs. This association of immunity, fertilization, and cytolysis is driven by a group of genes that participate in sperm-egg interaction, but also have immune-related functions and destroy pathogens by cytolysis. Interestingly, PSGs with roles in both reproduction and immunity are often also related to cancer, and it has been hypothesized that most cancer genes under positive selection have been subject to antagonistic co-evolution, with lineage-specific variations in dynamics and strength [5],[37]. Several PSGs identified here are associated with both FAS/p53 apoptosis and cancer (Da Fonseca et al., in prep.), such as the protein p53, which also regulates maternal reproduction [38]; the cell adhesion gene ADAM2 (P = 2.9×10−6), which is integral to fertilization [39]; and the related genes ADAM15 (P = 5.4×10−4) and ADAM29 (P = 3.4×10−4), which are strong candidates for cancer evolution driven by sexual conflict. In addition, the testes development-related gene CCDC54 (P = 3.3×10−4) is currently a target of cancer immunotherapy research [40].
A smaller and somewhat less diverse group of enriched categories is associated with sensory perception. Among the most inclusive categories of this type are “sensory perception of chemical stimulus” (24 PSGs; P = 4.3×10−39) and “G-protein coupled receptor protein signaling pathway” (39 PSGs; P = 1.4×10−7). Previously, enrichments for such categories have been attributed primarily to olfactory receptors [4],[5]. Indeed, 15 PSGs are labeled as having “olfactory receptor activity” (P = 6.9×10−36). However, eight PSGs are involved in “sensory perception of taste,” including five taste receptors (P = 1.4×10−10). Interestingly, several of these are bitter taste receptors. The sense of bitter taste is critical in allowing organisms to avoid toxic and harmful substances, and extensive gene expansion of bitter taste receptors is known to have occurred during mammalian evolution [41], possibly driven by (or helping to drive) positive selection. Bitter taste receptors under positive selection include TAS2R1, TAS2R5, and a recently expanded cluster of genes at chr12p13 (TAS2R13, TAS2R14, TAS2R42, and TAS2R49). Another PSG, TAS1R2, is a receptor of sweet and umami taste, and the PSG RTP3 is a transmembrane protein that is involved in the transport of taste receptors and apparently influences their expression.
The PSGs in the “neurological processes” category (P = 7.5×10−7) are dominated by olfactory and taste receptors, but they also include other types of genes. For example, TMC2 (P = 1.1×10−4) is expressed in the inner ear and is important for balance and hearing [42]. The acid-sensing ion channel gene ACCN4 (P = 1.0×10−6) has been implicated in synaptic transmission, pain perception, and mechanoperception [43]. SLC6A5 (P = 3.0×10−4) is associated with hyperekplexia, a neurological disorder characterized by an excessive startle response [44]. The neuromedin receptor NMUR (P = 6.1×10−4) is involved in the mammalian circadian oscillator system [45],[46]. Finally, the neurotensin receptor NTSR1 (P = 8.1×10−4) mediates hypotension, hyperglycemia, hypothermia, antinociception, and regulation of intestinal motility and secretion [47].
Similarly, the PSGs associated with diet include but are not limited to taste and olfactory receptors. For example, MGAM (P = 2.4×10−8) is essential for the small intestinal digestion of starch, giving it a critical role in human metabolism, as starches of plant origin make up two-thirds of most human diets [48] (see also [49]). MAN2B1 (P = 1×10−6) is involved in the cleavage of the alpha form of mannose, a sugar monomer. Defects in this gene cause lysosomal alpha-mannosidosis, a lysosomal storage disease characterized by the accumulation of unbranched oligosaccharide chains [50]. TCN1 (P = 2.9×10−31) is a major constituent of secondary granules in neutrophils and facilitates the transport of vitamin B12 into cells, which is important for the normal functioning of the brain and nervous system, and for the formation of blood [51]. In addition, several PSGs participate in “steroid hormone metabolism” (P = 8.3×10−4) including genes that metabolize xenobiotics and drugs (e.g., SULT1C3, UGT2B7, and CYP2C8). Positive selection in these and other genes is likely to have been influenced by changes in food preferences during mammalian evolution.
Few functional enrichments were evident for the PSGs identified by the branch- and clade-specific LRTs, primarily because these sets were quite small in size. However, the more powerful LRTs, such as those for the primate and rodent clades (Figure 1C,E), did produce significantly lower P-values for genes of certain functional categories than for others. Interestingly, these categories were dramatically different for the primate- and rodent-clade LRTs, with nearly all of the primate categories relating to sensory perception, and nearly all of the rodent categories relating to immunity and defense (Table 4). Indeed, the PSGs identified by the primate-clade test include several taste and olfactory receptors, as well as receptors for the sensation of pain (e.g., MRGPRE, NPFF2) and color vision (e.g., OPN1SW), and receptors involved in immunity (e.g., CCR1). The PSGs identified by the rodent-clade test include few such genes, but they include many genes involved in responses to wounding, inflammation, and stress, as well as genes involved in complement activation and innate immunity. Thus, we find little evidence that genes directly involved in brain development and function have (as a group) been driven by positive selection in primates, but many genes that provide sensory information to the brain do appear to have experienced positive selection. These changes in sensory perception could conceivably have been brought on by, or could have contributed to, increased brain size and complexity in primates.
To gain further insight into the patterns of positive selection that have shaped present-day mammalian genes, we devised a model that allows for probabilistic inferences about the selection histories of individual genes. A selection history is defined as an assignment to each branch of the phylogeny of one of two evolutionary modes: positive selection (each site evolves with ω0<1, ω0 = 1, or ω2>1) or absence of positive selection (each site evolves with ω0<1 or ω0 = 1). The model allows a posterior distribution over selection histories to be inferred for each gene, and it allows for estimates of the number of genes under positive selection on individual branches and clades that consider uncertainty about selection histories. Unlike the branch- and clade-specific LRTs—which are simple one-sided hypothesis tests and are necessarily conservative about rejection of the null hypothesis—this model considers all candidate histories symmetrically, and allows for “soft” (probabilistic), rather than absolute, choices of history at each gene.
Briefly, the model is defined in terms of a simple switching process along the branches of the phylogeny. It has separate parameters for the rates of gain and loss of positive selection at several switch points on the tree, with two switch points per internal branch and one per external branch (see Figure 4A and Methods). The joint posterior distribution of these parameters and of all selection histories is inferred from the data by a Gibbs sampling algorithm (see Methods and Text S1). The inference procedure is computationally intensive, so it was applied only to the 544 genes identified by one or more LRTs as showing significant evidence of positive selection. Because in these cases the null model of no positive selection had already been rejected by a conservative test, the history without selection on any branch was excluded, leaving 29−1 = 511 possible histories for the nine-branch (unrooted) phylogeny. To reduce computational cost, the inference of selection histories was conditioned on the maximum likelihood estimates of the parameters of the codon models (see Methods).
The inferred rates of gain and loss are quite variable (Figure 4A and Figure S2), with posterior means ranging from about 0.01 to 0.53. These rates are sharply reduced for the external branches of the tree, probably in large part because of diminished power to detect changes in selective mode on these branches. The number of genes inferred to be under selection also varies by branch, but not as dramatically, with expected values ranging between 207.9 and 393.9 and many 95% credible intervals overlapping (Figure 4B). Despite differences at individual branches, gains and losses appear to be roughly in equilibrium overall, with 61% of genes estimated to have been under selection at the root, and between 38% and 62% (averaging 50%) under selection at the leaves. The slight tendency to lose selection over time could reflect an ascertainment bias for genes that experienced selection early in mammalian evolution, which will tend to display signatures of selection on multiple long branches of the tree and therefore will be more easily detectable by the LRTs. The branches with the most genes under selection (such as those leading to the rodent and primate ancestors, and to dog and macaque) are generally long (see Figure 1A), suggesting power may influence these estimates. Nevertheless, the unusually high rate of gain on the branch to the rodents, and the comparatively low rate of loss on that branch (both having fairly low posterior variance; Figure S2), suggest not just differences in power but a real tendency for a net gain of selection on this branch, perhaps due to larger population sizes in the rodents. Whether because of power or a genuine increase in selection, the rodent branch appears to play a major role in the identification of PSGs. An expected 72% of the 544 candidate PSGs are under selection on this branch.
The posterior distributions over histories suggest that few genes have experienced positive selection specific to individual branches or clades (Figure 4B). Instead, most genes appear to have switched between evolutionary modes multiple times. The estimated number of mode switches per gene (averaging across genes but considering the joint posterior distribution for all selection histories) is 1.6 (95% CI: 1.5–1.7), with 0.6 gains (0.5–0.7) and 1.0 losses (0.9–1.1). An expected 91% of PSGs have experienced at least one mode switch, and an expected 53% have experienced two or more switches. 54% of PSGs have 95% CIs excluding zero switches (i.e., with high confidence, these genes have switched modes at least once), and 10% have 95% CIs also excluding one switch (with high confidence, they have switched modes at least twice). Thus, this analysis suggests that positive selection tends to be gained and lost relatively frequently in mammalian genes. Episodic positive selection has been observed and analyzed in detail at individual loci (e.g., [52],[53]) but to our knowledge genome-wide evidence of this phenomenon in mammalian phylogenies has not previously been reported. Interestingly, our observations are qualitatively compatible with Gillespie's theoretical model of an episodic molecular clock [54], although our model differs from his in detail.
By pooling information across genes and allowing for uncertainty in selection histories, this method estimates much larger numbers of genes under positive selection on each branch of the tree than do the more conservative LRTs (Figure 1). For example, the expected number of genes under selection on the branch to the primates is 360.5 (95% CI 338–382), compared with 21 genes identified by the corresponding LRT, and the expected number under selection on the branch to the rodents is 393.9 (357–426), compared with 56 identified by the corresponding LRT. In this analysis, the estimated numbers of genes that have experienced positive selection on the various primate and rodent lineages are not dramatically different, suggesting that the sharp differences from the LRTs in large part reflect inequalities in power. They also suggest that the numbers of genes under selection in recent human and chimpanzee evolution are not as different as they appear from LRTs, which will identify only the most extreme cases [9]. Indeed, the 95% CIs for the human and chimpanzee estimates heavily overlap.
In addition to being useful in a bulk statistical analysis of all PSGs, the Bayesian framework can be used to identify the single most likely selection history for each gene. In some cases, these histories are consistent with known functional differences between species, and help to shed light on the evolutionary basis of these differences. For example, the sweet receptor TAS1R2 has been shown in knock-out experiments to be responsible for differences between species in preferences for sweet tastes [55]. (Humans can taste several natural and artificial sweeteners that mice cannot, such as monellin, thaumatin, aspartame, and neohesperidin dihydrochalcone.) This gene is predicted to have experienced selection on the primate clade and on the branches leading to the primate and rodent clades (posterior probability [PP] = 0.20), suggesting that positive selection on TAS1R2 in both primates and rodents could have contributed to differences in sweet taste preferences. Another example is the integral membrane glycoprotein GYPC, which plays an important role in regulating the mechanical stability of red blood cells. In humans, GYPC has been associated with malaria susceptibility, and predicted to have undergone recent positive selection [56]. However, we find evidence that GYPC has experienced positive selection on all branches of the primate clade (PP = 0.66), suggesting longer-term selective pressure that have also affected nonhuman primates. A third example is CGA, which encodes the alpha subunit of the four human glycoprotein hormones (chorionic gonadotropin, luteinizing hormone, follicle stimulating hormone, and thyroid stimulating hormone). This gene shows strong evidence of positive selection specific to the primate clade (PP = 0.82), consistent with the proposal that relatively recent adaptations in pregnancy and development have played a critical role in the evolution of the human endocrine system [57]. Interestingly, the closely related genes CGB1 and CGB2 (which encode two of the six beta subunits of chorionic gonadotropin) are thought to have originated by gene duplication in the common ancestor of humans and great apes [58], and these events could have contributed to positive selection on CGA. Finally, the complement components C7 and C8B, which encode proteases in the membrane attack complex, are predicted with high probability to be under selection in rodents only (C7: PP = 0.98 for selection in mouse; C8B: PP = 0.93 for selection in mouse and rat). Differences in complement proteases are thought to explain certain differences in the immune responses of humans and rodents [59].
We examined the human mRNA expression levels of PSGs non-PSGs using public data from the Affymetrix Human Exon 1.0 ST Array, which contains probes for nearly all of our genes and permits accurate estimation of expression levels [60]. Our most striking finding was that PSGs show reduced expression levels in all of the 11 available tissues (breast, cerebellum, heart, kidney, liver, muscle, pancreas, prostate, spleen, testes, and thyroid; see Methods). In particular, a significantly smaller fraction of PSGs than of non-PSGs produce a hybridization signal above the background level for the array (P<4×10−4 in all tissues for PSGs defined by the all-branch test, one-sided FET). Moreover, among genes expressed above background, expression levels are significantly lower for PSGs than for non-PSGs (P<7×10−5 in all tissues, one-sided MWU test; Figures 5A–C). PSGs also show significantly greater tissue bias than non-PSGs, as measured by the statistic τ [61] (Figure 5D) or by an alternative statistic here denoted γ [17] (Methods). The differences in expression level and tissue bias between the two sets of genes do not appear to be explained by differences in false negative or false positive rates in the detection of positive selection, and the differences in expression level do not appear to be a consequence of the differences in tissue bias (Text S1). In addition, the observed differences remain if the genes that belong to strongly enriched GO categories (Table 2) are excluded, indicating they cannot be attributed to particular classes of PSGs known to have tissue-specific expression patterns, such as those involved in immunity or spermatogenesis. That expression levels are reduced in all tissues further suggests the existence of a general relationship between expression patterns and the likelihood of positive selection.
Consistent with previous observations (e.g., [62]), we found a significant negative correlation of ω with expression level in all 11 tissues (Spearman's rank correlation coefficient ρ ranged from −0.25 to −0.43). In addition, we observed a positive correlation of ω with tissue bias, as measured by τ (ρ = 0.24) [63],[64]. (Similar correlations were observed when the log likelihood ratio in the test for positive selection on any branch—which increases with increasing evidence for selection—was used in place of ω.) Unlike in previous studies, however, we were able to examine these correlations separately for positively and non-positively selected genes, using the set of PSGs identified by the all-branches LRT. Interestingly, the correlations of ω with expression level τ are much stronger within the non-PSGs than within the PSGs, indicating that the observed correlations are primarily driven by negative rather than positive selection (see also [65]). Thus, while genes expressed at low levels and/or in a tissue-specific manner show an increased tendency to have experienced positive selection, the strength of positive selection does not appear to be strongly correlated with their expression patterns (see Discussion).
Of the 15,823 genes that were tested for positive selection and had detectable expression in at least one tissue, 1,509 showed a strong preference for one tissue and were designated as tissue specific (γt>0.25 for some tissue t and γt>0.25 for all t′ ≠ t; see Methods). Based on this designation, spleen- and testes-specific genes were strongly enriched for PSGs: 22 of 174 (12.6%) spleen-specific genes were PSGs, compared with only 2.2% of other genes (P = 8.7×10−11, one-sided FET); and 45 of 715 (6.3%) testes-specific genes were PSGs, compared with 2.1% of other genes (P = 8.2×10−10). There were also significant, but weaker, enrichments for PSGs among liver-specific (P = 9.1×10−3) and breast-specific (P = 1.0×10−2) genes. Not surprisingly, the spleen-specific PSGs generally appear to be immune-related, and many of the testes-specific PSGs are involved in spermatogenesis or sperm adhesion (they include ADAM2 and SPAM1; Table 3). The liver and breast specific genes are more heterogeneous. In contrast, only 2 of 254 (0.7%) cerebellum-specific genes were PSGs, compared with 2.3% of other genes (P = 0.066, one-sided FET). Only a few tissue-specific genes were identified by the clade tests, so it was not possible to compare the relationships between tissue-specific expression and positive selection in primates versus rodents. However, there were significant enrichments for primate PSGs among spleen-specific genes, and for rodent PSGs among testes-specific genes.
Despite our large data set, we found no indication of a correlation between expression in the primate brain and recent positive selection in protein-coding regions [66] (see [67],[68]). Indeed, we found some evidence to the contrary: PSGs identified by the primate-clade test show more sharply reduced expression levels (compared with non-PSGs) in the cerebellum than in any other tissue; cerebellum-specific genes are depleted, not enriched, for PSGs; and none of the primate PSGs show tissue-specific expression in the cerebellum. These findings, of course, do not rule out positive selection in individual genes of great importance in brain development, nor do they rule out positive selection on gene expression.
While positive selection was our primary focus, our data set also provides an opportunity to compare the average rates of protein evolution in various mammalian lineages. We estimated a separate nonsynonymous-synonymous rate ratio ω for each branch of the six-species phylogeny, pooling data from all ortholog sets (Figure 1A). Consistent with previous findings [6],[8], we observe that protein-coding genes, on average, have experienced moderately strong purifying selection (ω « 1) on all branches of the phylogeny, but that estimates of ω vary considerably within the mammals. These estimates are largest for the hominids (ω≈0.25), smallest for the non-primate mammals (0.12<ω≤0.14), and intermediate for non-hominid primates (0.17<ω<0.21). It is thought that increased estimates of ω in hominids primarily result from weakened purifying selection, owing to reduced effective population sizes [69],[5]. The intermediate values for non-hominid primates may also be influenced by population size.
To examine the relationship between ω and population size further, we made use of a theoretical relationship between ω and the scaled selection coefficient γ (see [70],[71]), which holds if nonsynonymous substitutions have equal (and small) selection coefficients, if synonymous substitutions are neutral, and if population sizes are sufficiently large (Methods). This relationship allows ratios of population sizes to be estimated from ratios of ω estimates, under the assumption of constant selection coefficients across species. Here we further assumed that the ancestral population sizes of humans and the chimpanzee subspecies Pan troglodytes versus (to which the sequenced animal belonged) were roughly the same (Nh = Nc) [5], and estimated the ratio of ωm in macaque to ωh in human/chimpanzee from our 10,980 human-chimpanzee-macaque ortholog trios. Our estimate of ωm / ωh = 0.732 implies an estimate for the ratio of the macaque to human ancestral population sizes of Nm / Nh = 1.41 [bootstrapping 95% CI (1.15, 1.64)]. In comparison, the ancestral macaque population size has been estimated at ∼73,000 [72] and ancestral human and chimpanzee population sizes at 40,000–70,000 [73],[74], which would imply a ratio of 1.04–1.82, in reasonable agreement with our estimate. We used the same theoretical relationship to devise a LRT indicating whether or not each gene deviated significantly from the assumed model with Nm / Nh = 1.41 (Methods). For the vast majority (96%) of the 10,980 genes examined, no significant deviation was observed, indicating that the differences in selection pressure in macaque and the hominids are generally well-explained by differences in population size.
To compare the power of our LRTs with the power of previous tests based on hominid or primate genomes, we simulated data sets under a range of parameter values and measured the fraction of cases in which positive selection was predicted (Figure 6). These experiments show that power increases substantially when the set of species under consideration is expanded from the two hominid species to the three primates then to all six mammals. With hominid species only, power is poor even when selection is quite strong (e.g., ∼20% with a constant ω = 2 and ∼40% with ω = 4), suggesting that a genome-wide scan will tend to identify only the most extreme cases of positive selection. If a rigorous correction for multiple testing is applied, a test based on hominids only has essentially no power, even for fairly long genes under strong selection (Figure S3; see also [5]). The situation is considerably improved by the addition of the macaque genome, but power remains poor when controlling for multiple testing unless genes are long and selection is strong. When all six mammals are considered, however, power increases substantially. With the full data set, power is reasonably good (≥70%) even when genes are short and selection is moderate in strength; it remains good when multiple comparisons are considered (Figure S3). The absolute estimates of power from these experiments depend on the simplifying assumptions used in the simulations (including the unrealistic assumption of constant ω among lineages and among sites), and they must be interpreted cautiously. However, estimates of relative power—which will be less sensitive to these simplifying assumptions—indicate a substantial improvement is achieved by the addition of the three non-primate mammals.
Since it first became possible to compare the sequences of complete mammalian genomes about five years ago, a number of genome-wide scans for positively selected genes (PSGs) have been conducted using phylogenetic methods [4],[5],[6],[7],[8],[9]. These studies have provided a valuable initial assessement of the genome-wide landscape of positive selection in mammals, but they have left many important questions unanswered. The analysis presented here, by incorporating non-primate mammalian genomes into a genome-wide scan for positive selection, represents a significant step forward. The larger, more divergent group of species improves power significantly, and the use of a nontrivial phylogeny provides insight into the particular patterns of positive selection that have helped to shape present-day genes. To our knowledge this is the largest and most detailed genome-wide analysis of positive selection to date, not only in mammals but in any group of organisms (although extensive analyses, similar in some respects, have been performed recently in Drosophila [75],[76]).
One finding of particular interest was that several whole pathways are especially rich in PSGs. Examples include the classical and alternative pathways for complement-mediated immunity and the FAS/p53 apoptotic pathway (Figures S1, S4 and S5). These findings suggest that positive selection may frequently act directly on whole protein complexes or pathways (see [77],[78]). Alternatively, adaptive changes in one protein may sometimes have a cascade effect, leading to changes in other genes that bring a system back into equilibrium. Whether or not all changes affecting a pathway are driven by positive selection, one might expect to see similarities in the selection histories of gene with closely related functions. Indeed, we have found that genes with similar selection histories on average have substantially greater similarity in their GO categories than do genes with more divergent histories (Figure S6). The observations that multiple interacting genes often show evidence of positive selection and that positive selection is frequently episodic may well be connected. For example, in some cases a transient external force could induce a burst of changes in multiple genes that participate in the same pathway, either separately or by triggering a cascade of interdependent events. Further unraveling the (co-)evolutionary histories of interacting PSGs promises to be a fertile area for future work. Care will be required to distinguish between true co-evolution and correlations that can be explained by dependencies on expression levels or other covariates of evolutionary rate [79].
Our finding that PSGs are expressed at lower levels and in a more tissue-specific manner than non-PSGs is consistent with a well-known negative correlation ω with expression level, and a positive correlation of ω with tissue bias (τ or γ). Various explanations have been proposed for the observed decrease in ω among genes expressed at high levels and/or expressed broadly across tissues, including selection for translational efficiency, selection against misfolding, or increased selection due to pleiotropy [62],[68],[65]. In any case, these genes do appear to experience a reduction in their evolutionary “flexibility” compared with genes expressed at low levels and/or nonuniformly across tissues. Our observation of decreased rates of positive selection among these genes—and increased rates among low-expression/high-tissue-bias genes—is consistent with this characterization. Interestingly, however, we observe that correlations of ω with expression level and τ hold strongly within non-PSGs, but are much less pronounced within PSGs. Thus, expression levels and patterns are strongly correlated with both the strength of negative selection and the likelihood of positive selection, but they are only weakly correlated with the strength of positive selection. It appears that genes may be more likely to come under positive selection if they are in a state of evolutionary flexibility brought on by reduced or tissue-specific expression, but once positive selection has taken hold their subsequent evolutionary course is not strongly dependent on their expression patterns.
As additional mammalian genomes become available, the statistical power to detect positive selection will improve. However, most forthcoming genomes are being sequenced at low coverage, and will inevitably exhibit increased levels of error in base calls, genome assemblies, ortholog identification (due to short contigs), and alignment—all of which can lead to spurious signals for positive selection. (The same errors tend to produce false negatives, rather than false positives, in the identification of conserved elements.) Thus, careful data quality controls will be needed to take advantage of these data. In addition, when considering the impact of additional sequences on statistical power, it is useful to distinguish between positive selection that acts continuously (or in recurrent episodes) over a long evolutionary period, and positive selection that acts transiently or in a lineage-specific manner. Deep phylogenetic sequencing should generally improve detection power for continuous or recurrent positive selection, but power for transient selection depends strongly on the sequenced species and the lineages of interest. For example, the genome sequences of a dozen non-primate mammals will likely have little effect on the power to detect human-specific selection, while the gorilla and neanderthal genomes could help considerably. There are fundamental limitations in the detection of weak, transient, or highly localized positive selection that will not be overcome by any amount of genome sequencing. Nevertheless, the availability of several new primate genomes, including those of the orangutan, marmoset, and gorilla, may significantly improve power for PSGs in primates.
Our ability to connect positive selection with function remains rudimentary, but gradual progress is being made. As additional sequence data becomes available, it will become possible to associate selection with individual residues of proteins with greater accuracy. At the same time, more data is becoming available on the specific functional roles of individual amino acids, for example, from structural or mutagenesis studies. As a result, it will increasingly become possible to find direct links between selection and function. Often these links will initially be tentative, as in our site-specific analysis of the HAVCR1 gene. Nevertheless, they provide a valuable starting point for experimental follow-up. At the same time, more can be done to incorporate non-sequence data—such as structural and expression data—into computational methods for detecting positive selection. Thus, improvements in both computational and experimental methods will be needed to establish deeper and more informative connections between evolutionary dynamics and molecular function.
The latest human (hg18), chimpanzee (panTro2), rhesus macaque (rheMac2), mouse (mm8), rat (rn4), and dog (canFam2) genome assemblies were obtained from the University of California, Santa Cruz (UCSC) Genome Browser. Human-referenced whole-genome alignments were constructed from syntenic pairwise alignments with human (the “syntenic nets”) using the UCSC/MULTIZ alignment pipeline [80],[81]. Low quality bases (Phred score <20) from the chimpanzee, macaque, rat, and dog genomes were converted to ‘N’s in these alignments.
A starting gene set was composed from of the human RefSeq [28], UCSC Known Genes [29], and VEGA [30] annotations (downloaded from UCSC Feb. 19, 2007). Transcripts that lacked annotated coding regions (CDSs), that had CDSs of <100 bp, or that had CDSs whose lengths were not multiples of three were discarded, leaving 88,879 nonredundant transcripts. These transcripts were grouped by same-stranded CDS overlap into 21,115 genes (transcript clusters). All transcripts were mapped from human to each of the other five mammalian species via the syntenic alignments, then subjected to a series of filters designed to minimize the impact of annotation errors, sequence quality, and changes in gene structure on subsequent analyses. Briefly, each human transcript was required (1) to map to the non-human genome via a single chain of sequence alignments including ≥80% of its CDS; (2) after mapping to a non-human species, to have ≤10% of its CDS in sequencing gaps or low quality sequence; (3) to have no frame-shift indels, unless they were compensated for within 15 bases; (4) to have no in-frame stop codons and to have all splice sites conserved. To allow for genes that are mostly conserved but whose start or stop codons have shifted, incomplete transcripts—with ∼10% of bases removed from the 5′ and 3′ ends of the CDS—were also considered. The final collection of ortholog sets was obtained by selecting, for each gene, the (complete or incomplete) transcript that successfully mapped to the largest number of non-human species. In the case of a tie, the transcript with the greatest total CDS length was selected. This procedure resulted in 17,489 genes with ≥2 non-human orthologs, averaging ∼5 species per gene (including human; see Table 1).
To establish 1∶1 orthology, each human gene and putative non-human ortholog was examined for evidence of an inparalog (a paralog arising from a recent duplication [82]) with respect to the other species. Specifically, if either gene had a BLASTN match within the same species (with ≥80% CDS alignment) that was more similar than the two orthologs were to each other, then that gene was considered recently duplicated and was excluded from the analyses of positive selection. The removal of a duplicated gene did not require an ortholog set to be discarded entirely, provided a human gene and ≥2 nonhuman orthologs still remained. A collection of genes and gene predictions from the UCSC Genome Browser were used in the identification of inparalogs. When comparing rodent vs. non-rodent and rodent vs. rodent distances, a simple correction for unequal rates of evolution was applied. Further details are given in Text S1.
The LRT for selection on any branch of the phylogeny is essentially Nielsen and Yang's [31] test of site models 2a versus 1a, and the lineage- and clade-specific LRTs are essentially instances of Yang and Nielsen's [26] test 2 (see also [83],[27]). However, to reduce the number of parameters estimated per gene, the complete set of 17,489 genes was divided into eight equally sized classes by G+C content in third codon positions. The branch lengths and the transition-transversion rate ratio κ were estimated for each class under the null model, and these estimates were subsequently held fixed, in a G+C dependent way, for the LRTs. Instead of a complete set of branch lengths, a single scale parameter μ was estimated per gene. Thus, only the parameters μ, ω0<1 and p0 for the null model and the additional parameters ω2>1 and p1 for the alternative model, were estimated per gene (see [31],[26]). This parameterization speeds up calculations substantially compared to estimating κ and a set of branch length per gene, while its sensitivity, specificity and power to detect positive selection are comparable (Text S1). We developed our own software for likelihood computation and parameter estimation to support this parameterization.
For the LRT for selection on any branch, P-values were computed empirically, based on simulation experiments. 10,000 alignments were simulated under the ‘nearly neutral model’ (allowing for a fraction p0 of sites to evolve with ω0<1 and a fraction 1−p0 to evolve with ω1 = 1) for each G+C class using evolver [84]. Alignment lengths and values of μ, ω0 and p0 were drawn from the empirical distribution defined by the real alignments (using estimates obtained under the null model), and the remaining parameters were fixed at global estimates for each G+C class. Log likelihood ratios (LLRs) were then computed exactly as for the real data. The nominal P -value for a LLR of r was defined as the fraction of all simulated alignments with LLR≥r, unless the number of such alignments was <10, in which case we assumed (an adequate approximation for small P-values, according to the simulation experiments). The method of Benjamini and Hochberg [85] was used to estimate the appropriate P-value threshold for a false discovery rate of <0.05. For the lineage- and clade-specific LRTs, P-values were computed assuming the null distribution was a 50∶50 mixture of a distribution and a point mass at zero (see [27] and discussion in Text S1).
Let X = (X1,…, XN) be the alignment data, with Xi denoting the alignment for the i th gene (1≤i≤N; here N = 544), and let Z = (Z1,…, ZN)be the set of selection histories, with Zi denoting the selection history for the i th gene (1≤Zi≤M; here M = 511). Recall that a selection history is defined as a pattern of presence or absence of positive selection on the branches of the unrooted phylogeny. Let Zib ∈ {0,1} indicate the selective mode (with 1 representing positive selection) for branch b ∈ {1,…,B} (here B = 9) under history Zi. The parameters of the switching model, denoted θ, are defined below. The model assumes independence of genes and independence of histories, and conditional independence of X and θ given Z. Thus, the complete data likelihood is given by:(1)
The probability of a history, P(Zi|θ), is a function of the set of switches in selective mode required to explain the history parsimoniously. For each history to be explained parsimoniously, switches must be allowed to occur early (near the ancestor) or late (near the descendant) on each internal branch, as well as (early) on each external branch (Figure 4A; see Text S1 for a justification of the model). Thus, there are twelve possible switch points, with three of them adjoining each of the four internal nodes of the tree. It is convenient to denote these points where is the set of internal nodes and represents the branches adjoining node n. Let Vnb ∈ {0,1} and Wnb ∈ {0,1} indicate the selective states before and after point Pnb, respectively. For a given history Zi, these variables are uniquely determined by parsimony according to a simple algorithm (see Text S1). The four possible values of (Vnb, Wnb) correspond to four possible scenarios at Pnb—gain of selection (0,1), loss of selection (1,0), absence of gain (0,0), or absence of loss (1,1). The probabilities of these scenarios (i.e., the conditional probability of each Wnb given Vnb) are defined by a parameter for gains (θnbG) and a parameter for losses (θnbL) at each point. In addition, the prior probability of selection at the root of the tree is given by a parameter θ0. (For this analysis, the most recent common ancestor of the primates and rodents is treated as the root of the tree; see Text S1.) The set of parameters can thus be described as . The prior probability of a history Zi is simply a product of the prior and the relevant switching probabilities:(2)where U0 represents the selective state at the root.
The switching model effectively defines a prior distribution over histories, which tends to favor simpler histories over more complex ones (typically θnbe<0.5). The prior probability for each element of θ is defined by a (conjugate) Beta distribution with parameters α and β (here, α = 1, β = 9). Because these elements are independent in the prior,(3)
The term P (Xi | Zi) in equation 1 is simply the likelihood at gene i of a branch-site codon model that assumes selection history Zi. A full Bayesian approach would integrate over the parameters of these codon models, but this would be computationally prohibitive. Instead, we make the Empirical Bayes simplification of conditioning the analysis on maximum likelihood estimates of the parameters of the codon models. The maximized log likelihoods Lij for all genes i and histories j are precomputed using existing software (in parallel, on a large computer cluster) and stored in an N×M matrix, which is then used in the inference of selection histories.
The variables Z and θ are unobserved, and the goal is to infer their joint posterior distribution,(4)
This inference was accomplished by a Gibbs sampling algorithm that alternates between sampling each Zi conditional on Xi and a previously sampled θ, and sampling each element of θ conditional on a previously sampled Z. It is straightforward to derive the required conditional distributions and to sample from them (Text S1). The Gibbs sampler converges rapidly and mixes well. Notice that, because the history without selection on any branch is excluded, all of the histories are described by codon models with the same number of parameters. Therefore, no penalty for parameter number is needed when comparing histories.
After an appropriate burn-in period, each iteration of the Gibbs sampler produces a sample (Z(t), θ(t)) from P(Z, θ|X). Estimated posterior expected values of interest were obtained by averaging these samples or functions of these samples, and Bayesian 95% confidence intervals were obtained by taking the 0.025 and 0.975 quantiles of the sampled values. For example, the posterior expected number of genes under selection on branch k (see Figure 4) was estimated as , where T is the number of samples and the function fk(Z) counts the number of genes under selection on branch k in a set of histories Z.
Each gene was assigned categories from the GO [32] and PANTHER [86] databases (downloaded on June 26, 2007), based on the Uniprot identifiers of associated transcripts. At least one GO category was identified for 14,137 (86%) genes, and at least one PANTHER category for 13,753 (83%) genes. To account for the hierarchical nature of these databases, each gene was also considered to belong to all parent categories of the ones to which it was directly assigned. For each category C and set of PSGs S, a 2×2 contingency table was constructed for the numbers of genes assigned or not assigned to C, and within and outsideS, then a (one-sided) P-value for independence of rows and columns was computed by Fisher's exact test. In addition, the distributions of LRT P -values among the genes assigned to C and not assigned to C were compared by a (one-sided) Mann-Whitney U (MWU) test. (Notice that S is not considered in this case.) Nominal P -values computed by the FET and MWU tests were corrected for multiple comparisons using the method of Holm [87].
The analysis of gene expression was based on the publicly available “Tissues+Mixtures” sample data set for the Affymetrix GeneChip Human Exon 1.0 ST Array (http://www.affymetrix.com/support/technical/sample_data/exon_array_data.affx). The RMA-based probeset summaries [88] and DABG (detected above background) -values were used. Each probeset was assigned genomic coordinates using the “Affy All Exon” track in the UCSC browser (hg17), then was associated with any human gene from our set having an exon on the same strand that completely contained the probeset. Nearly every gene (98%) had at least one probeset.
To calculate a P-value for each gene×tissue, the DABG P-values of all associated probesets (pooling the three replicates per probeset×tissue) were combined using Fisher's method [89]. A gene was considered to be significantly expressed above background if it had (nominal) P<0.001. Similarly, an estimated expression intensity for each gene×tissue was calculated by first taking the median over the three replicates of each RMA-based probeset summary, then taking the median of these values over all probesets associated with the gene. The analysis of expression intensities was restricted to genes significantly expressed above background so that genes expressed at or near the background level did not drive the results.
To measure tissue bias, we used: (1) the statistic τ [61], which represents the average difference in normalized expression intensity from that of the tissue of maximal expression, and (2) a statistic, here denoted γ, defined as γ = maxtγt, where γt is the squared cosine of the angle between the expression vector and the coordinate axis associated with t (see [17]). In defining genes as tissue specific for tissue t we required γt>0.25 and γt′<0.125 for all t′ ≠ t. Further details are given in Text S1.
Maximum likelihood estimates of ω for each branch were obtained using the codeml program in the PAML software package [84], with F3×4 codon frequencies, estimation of κ (fix_kappa = 0) and a single ω across sites per branch (model = 1, NSsites = 0). The tree topology shown in Figure 1 was assumed. The alignments for all genes were concatenated for this analysis.
Assuming all non-synonymous mutations at a given gene have the same selection coefficient and all synonymous mutations are neutral, population genetic theory says that ω should be given by [70],[90]:(5)where γ = 2Ns. Therefore, γ can be estimated as f−1(ω), where ω = f(γ) denotes the function above. (Values of γ can be obtained numerically; see Text S1.) Ratios of population sizes can therefore be estimated from ratios of ω estimates: .
The LRT to test whether differences in population size can explain the differences in ω in human and macaque was constructed as follows. The null model assumes ωh = ωc and ωm = 0.732ωh (see Results). The alternative model also assumes ωh = ωc but leaves ωm as a free parameter to be estimated from the data. Because the models are nested, a distribution is used for significance testing. This test was applied separately to each gene.
A website is available at http://compgen.bscb.cornell.edu/projects/mammal-psg/ with definitions of the candidate genes (accession numbers, genomic coordinates, and descriptions), multiple alignments of orthologous gene sets, GO and PANTHER category assignments, detailed results of the LRTs and the Bayesian analysis, and other resources. In addition, the candidate genes and predicted PSGs are displayed as a track in the UCSC Genome Browser (http://genome.ucsc.edu; assembly hg18).
|
10.1371/journal.pbio.1000334 | Sequestration of the Aβ Peptide Prevents Toxicity and Promotes Degradation In Vivo | Protein aggregation, arising from the failure of the cell to regulate the synthesis or degradation of aggregation-prone proteins, underlies many neurodegenerative disorders. However, the balance between the synthesis, clearance, and assembly of misfolded proteins into neurotoxic aggregates remains poorly understood. Here we study the effects of modulating this balance for the amyloid-beta (Aβ) peptide by using a small engineered binding protein (ZAβ3) that binds with nanomolar affinity to Aβ, completely sequestering the aggregation-prone regions of the peptide and preventing its aggregation. Co-expression of ZAβ3 in the brains of Drosophila melanogaster expressing either Aβ42 or the aggressive familial associated E22G variant of Aβ42 abolishes their neurotoxic effects. Biochemical analysis indicates that monomer Aβ binding results in degradation of the peptide in vivo. Complementary biophysical studies emphasize the dynamic nature of Aβ aggregation and reveal that ZAβ3 not only inhibits the initial association of Aβ monomers into oligomers or fibrils, but also dissociates pre-formed oligomeric aggregates and, although very slowly, amyloid fibrils. Toxic effects of peptide aggregation in vivo can therefore be eliminated by sequestration of hydrophobic regions in monomeric peptides, even when these are extremely aggregation prone. Our studies also underline how a combination of in vivo and in vitro experiments provide mechanistic insight with regard to the relationship between protein aggregation and clearance and show that engineered binding proteins may provide powerful tools with which to address the physiological and pathological consequences of protein aggregation.
| Alzheimer's disease is thought to be a result of neuronal damage caused by toxic aggregated forms of the Aβ peptide in the brain. There is no cure and existing treatments are ineffective in reversing or preventing disease progression. Here we describe a novel strategy that makes use of an engineered “Affibody” protein to study the disease and potentially combat its underlying causes. The Affibody occludes the aggregation-prone regions of Aβ peptides, preventing their aggregation into toxic forms, and it also acts to dissolve pre-formed Aβ aggregates. It is functional in vivo, as its co-expression with Aβ peptides in transgenic fruit flies prevents the neuronal damage and premature death that result from expression of Aβ peptides alone. Moreover, we show that the origin of this protection is the enhanced clearance of Aβ peptides from the brain. These findings open up new opportunities for using engineered binding proteins to probe the origins of Alzheimer's disease and potentially to develop a new class of therapeutic agents.
| Of the neurodegenerative disorders that have been linked to protein misfolding and aggregation [1], Alzheimer's disease (AD) is the most common [2],[3]. Transgenic animal models have shown that aggregation of the Alzheimer β-peptide (Aβ) causes memory impairment [4],[5] and cognitive deficits [6] similar to those seen in patients suffering from AD. Aβ aggregation precedes neuritic changes [7], and there is a quantitative correlation between the propensities of mutant forms of Aβ to aggregate and their neurotoxicity [8]. In vitro aggregation of Aβ proceeds from the initial association of monomers into oligomeric, but still soluble, assemblies that ultimately form highly structured and insoluble amyloid fibrils [1],[9],[10],[11]. Evidence suggests that the primary neurotoxic species are the soluble oligomeric aggregates [4],[5],[12],[13] and that a fundamental building block may be dimeric Aβ species [14]. However, despite this progress, the details of Aβ aggregation in vivo, the structure of toxic aggregates, the mechanism of toxicity, and in particular, the relationship between aggregate formation and peptide clearance are not known.
We set out to investigate a novel approach to study the dynamics of Aβ aggregation in vitro and neurotoxicity or degradation in vivo by using a conformation-specific Aβ binding protein, the ZAβ3 Affibody [15],[16]. Affibody molecules are engineered binding proteins, which are selected by phage display from libraries based on the three-helix Z domain [17],[18]. The ZAβ3 Affibody was selected [15] to bind specifically to Aβ monomers with nanomolar affinity (dissociation constant Kd≈17 nM) [16]. It forms a disulfide-linked dimer to which Aβ binds and folds by induced fit [19] into a hairpin conformation such that its two aggregation-prone hydrophobic faces become buried within a tunnel-like cavity in the ZAβ3 dimer [16],[19]. The specificity and well-characterized structural features of ZAβ3 binding to Aβ make it an ideal candidate for studying the effects of Aβ monomer binding in vivo. We find that the presence of the Affibody molecule, achieved by co-expression, can eliminate Aβ neurotoxicity in a fruit fly (Drosophila melanogaster) model of AD [20],[21], and we used biochemical and biophysical experiments to identify the molecular mechanism by which this process occurs.
We first generated Drosophila strains transgenic for ZAβ3. As ZAβ3 is most effective in binding Aβ when it is in its dimeric form, we also generated Drosophila in which two copies of ZAβ3 are connected head-to-tail—(ZAβ3)2—to enable the disulfide-linked dimer to form more readily. Drosophila transgenic for the wild-type Z domain were used as controls. These three Affibody fly lines were then each crossed with Drosophila transgenic for Aβ42, Aβ42e22g [22], or Aβ40, and the co-expression of both transgenes together in the brain or in the eye was initiated by crossing with appropriate driver flies [20],[21].
Expression of Aβ42e22g in the brain of Drosophila causes rapid neurodegeneration resulting in a drastic reduction in lifespan from 38 (±1.8) to 9 (±0.5) days, consistent with the findings of previous studies [8]. Co-expression of ZAβ3 with Aβ42e22g, however, increases the lifespan to 20 (±0.2) days. Strikingly, if the-head-to-tail dimer (ZAβ3)2 is co-expressed with Aβ42e22g, the toxic effects of the peptide are yet further reduced and the lifespan increases to 31 (±0.8) days, which is almost as long as in wild-type controls (Figure 1A, Table S1) and indicates that the neurotoxicity of Aβ has been almost entirely abolished. Co-expression of the Z domain, which has no affinity for Aβ, does not affect Aβ42e22g toxicity, demonstrating that the rescue of Aβ toxicity in vivo is specific to ZAβ3. Co-expression of ZAβ3 with wild-type Aβ42 also significantly prolongs the lifespan of these flies (from 28, ±0.4, to 32, ±0.7, days). Again, the (ZAβ3)2 head-to-tail-dimer is even more effective, completely eliminating the toxicity associated with Aβ42 (lifespan 40, ±1.2, days), whereas the Z domain control has no effect (Figure 1B). Expression of the less aggregation-prone Aβ40 has no effect on lifespan, and none of the Affibody molecules or the control significantly affected the lifespan of flies expressing Aβ40 or wild-type Drosophila (Figure 1C and 1D).
The ability of (ZAβ3)2 to abolish the toxic effects of Aβ42e22g was confirmed physiologically by its ability to abolish the abnormal eye morphology associated with Aβ42e22g expression in the photoreceptors in the fly (Figure 2).
To determine the mechanism by which ZAβ3 mediates suppression of Aβ toxicity, we assessed the levels of Aβ42 in the brains of flies co-expressing Aβ42e22g and either ZAβ3, (ZAβ3)2, or the Z domain by Western blotting. Fly brains were homogenized in 1% SDS, subjected to electrophoretic separation, and probed using an antibody against the N-terminus of Aβ, which detailed structural studies reveal remains exposed in the Aβ:ZAβ3 complex [16]. SDS soluble Aβ can clearly be detected in flies expressing Aβ42e22g, but it is absent in flies co-expressing ZAβ3 or (ZAβ3)2 (Figure 3A). The specificity of this effect is confirmed by the continued presence of the Aβ42e22g in flies that co-express the non-binding Z domain.
The ZAβ3:Aβ complex is stable in 1% SDS (B. Macao, unpublished), and Aβ remaining in complexes or in SDS insoluble aggregates in the fly brain might therefore not be detectable by Western blot. In order to address this possibility, fly brains expressing Aβ42e22g with or without (ZAβ3)2, ZAβ3 or the Z domain were homogenized in 5 M GdmCl, conditions known to dissociate both Aβ aggregates and Aβ:ZAβ3 complexes. The total level of Aβ42e22g in these extracts was then measured by a sensitive ELISA assay (Figure 3B). Flies expressing both (ZAβ3)2 and Aβ42e22g show a 97% (±3%) reduction in the concentration of Aβ42e22g compared to flies co-expressing Aβ42e22g and the inert Z domain (the most appropriate control for the non-specific effects of expressing a second transgene on the levels of Aβ). Decreased Aβ42e22g levels in the presence of different Affibody constructs correlate well with corresponding reduction in neurotoxicity measured by the survival assay (Figure 1).
The prevention of Aβ42e22g aggregation by ZAβ3 and (ZAβ3)2 is demonstrated by immunohistochemical detection of Aβ42e22g in whole mount brain preparations analyzed by confocal microscopy. Flies expressing Aβ42e22g under the control of the OK107-Gal4 driver, which drives expression in a subset of adult neurons, contain abundant deposits in the brain recognized by the anti-Aβ 6E10 antibody, whereas flies co-expressing Aβ42e22g and (ZAβ3)2 have almost no visible 6E10 immunoreactive deposits (Figure 3C). In good agreement with the results of the ELISA analysis, co-expression of ZAβ3 results in a significant reduction in the burden of aggregates but does not result in their complete removal, whereas co-expression of the Z domain gives levels of Aβ deposits similar to those present in flies expressing Aβ42e22g.
In order to determine whether the presence of Aβ42e22g had altered the levels of ZAβ3 or (ZAβ3)2 present in the fly brain, brain homogenates were analyzed using either anti-cMyc antibodies to detect ZAβ3 or anti-Affibody antibodies to detect (ZAβ3)2; both dimeric Affibody molecules can be observed as 12 kDa dimers under non-reducing conditions. The levels of these Affibody species are not detectably altered in flies co-expressing Aβ42e22g (Figure 3D) despite the marked reduction of the levels of soluble Aβ42e22g (Figure 3A). While this experiment suggests that Aβ clearance could be occurring without the corresponding clearance of its binding partner ZAβ3, the quantities seen by Western blot represent the equilibrium levels of these two proteins, and so would not detect any turnover in ZAβ3 that may also be occurring.
We have established that the reductions in the levels of Aβ42e22g peptide in the fly brain are not due to altered gene regulation in flies co-expressing Z, ZAβ3, or (ZAβ3)2, because the levels of Aβ42e22g transcription are not significantly reduced in any case (Figure 3E).
In summary, ZAβ3 causes a reduction in Aβ42e22g levels by actively promoting its clearance from the brain. The clearance does not involve any specific antibody-mediated process, since Drosophila lacks an adaptive immune system [23]. In order to determine at which stages of the Aβ aggregation process the ZAβ3 Affibody can intervene, we analyzed the effects of ZAβ3 on the dynamic interconversion of monomeric, oligomeric, and fibrillar Aβ species in vitro.
Sequestration of the hydrophobic regions of Aβ40 and Aβ42 (Figure 4A and Figure S1) allows ZAβ3 to inhibit amyloid fibril formation completely, even that of the extremely aggregation-prone Aβ42e22g variant, as judged by thioflavin T (ThT) fluorescence assays indicative of amyloid fibril formation (Figure 4B–D, Figure S2, and Figure S3). The addition of ZAβ3 to Aβ40 or Aβ42 aggregation reactions has the same effect on the aggregation kinetics as reducing the Aβ concentration by the equivalent amount (Figure 4C and Figure S3A), demonstrating that inhibition of fibril formation occurs by sequestration of monomeric Aβ. When a molar excess of ZAβ3 is added at different times during the aggregation process, it effectively inhibits all further aggregation (Figure 4B and Figure S3B), indicating that not only does ZAβ3 effectively block aggregation even after its initiation, but also that monomeric Aβ is accessible for binding throughout the process of fibril formation.
We noted, however, during the course of the experiments that the ThT fluorescence signal tends to fall after the addition of ZAβ3 at advanced stages of the fibril formation reaction, suggesting that ZAβ3 may also act to reverse the aggregation process (Figure 4D and Figure S3C). To determine the kinetics of fibril dissolution by ZAβ3 in vitro, we set up experiments in which Aβ40 monomers dissociating from pre-formed fibrils are captured by ZAβ3. We used 15N-labelled Aβ40 for these experiments so that monomeric Aβ40 in complex with ZAβ3 could be identified by solution nuclear magnetic resonance (NMR) spectroscopy at low micromolar concentrations. The large fibrillar aggregates of 15N-Aβ40 (Figure 4E) did not generate an observable NMR spectrum even after 24 h of data collection, as expected, due to slow molecular tumbling and no highly mobile residues. The addition of ZAβ3, however, generated resonances from ZAβ3-bound monomeric 15N-Aβ40, indicating a gradual dissolution of the fibrils (Figure 4F and Figure S4). Only a small fraction of the Aβ40, however, dissociates from the fibrils over the first three weeks; thereafter the dissolution process becomes very slow, even for fibrils fragmented by sonication (Figure 4G). Still, under these conditions the observed level of dissolution does not represent the equilibrium state, as the pre-formed Aβ40:ZAβ3 complex is stable in the presence of Aβ40 fibrils (Figure S5). Hence, even though binding of the ZAβ3 Affibody to monomeric Aβ40 can act to dissolve fibrils, the dissociation kinetics are too slow, at least in vitro, for dissolution to be achievable in practice under ambient conditions.
In order to determine the critical issue of whether or not ZAβ3 can prevent the formation of smaller Aβ aggregates (oligomers), we examined their formation in vitro by size exclusion chromatography (SEC) in the presence and absence of ZAβ3 (Figure 5A to 5D and Figure S6). Oligomeric species [24] appear within hours in solutions of Aβ42, prepared by dilution from alkaline conditions [25], where the monomeric species is initially dominant. The partitioning between monomeric and oligomeric Aβ then reaches an interim steady state after ∼10 h before the onset of the formation of amyloid fibrils (Figure 5A). By contrast, in the presence of the ZAβ3, oligomer formation is completely inhibited (Figure 5B), a result that can be attributed to the sequestration of Aβ42 within the complex formed with the Affibody.
Isolated Aβ42 oligomers contain elements of well-defined β-sheet structure as measured by circular dichroism (CD), but the β-sheet content is lower than in mature fibrils (Figure 5E). Their stability is also lower as isolated oligomers dissociate into monomers and convert into amyloid fibrils (Figure 5C). Addition of the ZAβ3 Affibody results in dissolution of the oligomers after a few days (Figure 5D, 5F, and 5G and Figure S7). This is because binding of monomeric Aβ acts to shift the dynamic monomer-oligomer equilibrium such that the oligomer population is reduced, and NMR (Figure 5H) and SEC analyses (Figure S6) consequently also reveal monomeric Aβ42 in complex with ZAβ3.
The presence of the ZAβ3 Affibody in vivo results in the effective inhibition of Aβ toxicity and the promotion of Aβ degradation. These effects can be attributed to the ability of the ZAβ3 Affibody to act in three distinct ways on the Aβ aggregation process. First, monomeric Aβ will be sequestered by ZAβ3, the result of which is that toxic Aβ aggregates will not be able to form in the brain. Second, if Aβ aggregation were to occur, it can be slowed, halted, and even reversed by the action of ZAβ3 on the dynamic Aβ monomer-aggregate equilibria. Furthermore, the presence of ZAβ3 not only prevents or reverses Aβ aggregate formation, it also promotes clearance from the brain. We envisage that this could occur either by intracellular lysosomal or proteasomal degradation, or alternatively by the secretion and uptake by phagocytic cells of the ZAβ3:Aβ complex.
The results furthermore demonstrate how engineered binding proteins, such as Affibody molecules, that target specific protein conformations can be used to gain important insights into the dynamics of the Aβ aggregation process and its toxic consequences both in vivo and in vitro.
Drosophila melanogaster transgenic for Aβ40, Aβ42, and Aβ42e22g have been described previously [20]. Drosophila transgenic for the Z domain, ZAβ3, and the (ZAβ3)2 head-to-tail dimer were created by standard p element mediated germ line transformation using pUAST (Brand and Perrimon) as the expression vector. Affibody cDNA was inserted into the multiple cloning site of pUAST using EcoR1 and Xho1, except for (ZAβ3)2, which was cloned between EcoR1 and Xba1 sites. Each transgene was preceded by the same secretion signal peptide (MASKVSILLLLTVHLLAAQTFAQ), derived from the Drosophila necrotic gene, in order to target its expression to the secretory pathway. Transgenes were injected into w1118 embryos.
Drosophila transgenic for Aβ40, Aβ42, and Aβ42e22g were each crossed with Drosophila transgenic for Z, ZAβ3, and (ZAβ3)2 to create stable double transgenic stocks. Expression of the transgenes was achieved using the UAS-Gal4 system. UAS-Tg flies were crossed with flies expressing Gal4 under the control of either a neuronal promoter (elavc155 or OK107) or eye specific promoter (gmr). All fly crosses were maintained on standard cornmeal/agar fly food in humidified incubators. Crosses to generate flies expressing Affibody molecules or Aβ were performed at 29°C.
Survival assays were performed as described previously [20]. Briefly, 100 flies of each genotype were collected, divided into tubes of 10 flies, and kept at 29°C. The number of live flies was counted every 2–3 days and recorded. Survival curves were calculated using the Kaplan-Meier method, and differences between genotypes were assessed using the log-rank test.
Transgenes were expressed in the eye by crossing with gmr-Gal4 flies. Crosses were performed at 29°C. Flies were collected on the day of eclosion and sputter coated using 20 nM of Au/Pd in a Polaron E5000. SEM images were collected using a Philips XL30 Microscope.
Fifty flies were snap frozen in liquid nitrogen and decapitated for each genotype. Fly heads were homogenized in PBS/1% SDS containing protease inhibitors (Complete, Roche Applied Science, UK). Homogenates were then centrifuged at 12,100 g for 1 min to remove insoluble material, and the supernatants were collected for analysis. Protein concentration in each supernatant was determined using the DC Protein Assay (Biorad). Equal quantities of protein for each genotype were loaded on to 4%–12% Bis/Tris SDS PAGE gels (Invitrogen) for detection of Affibody molecules and 4%–12% Tris/glycine SDS PAGE gels (Invitrogen) for detection of Aβ. Electrophoresis was performed under non-reducing conditions, and protein was transferred to nitrocellulose membranes for Western blotting. ZAβ3 was detected using a mouse monoclonal anti-c-Myc antibody (clone 9E10, Abcam), and (ZAβ3)2 was detected using a goat anti-Affibody antibody (Abcam). Aβ was detected using a mouse monoclonal anti-Aβ antibody directed against the N terminus of Aβ (6E10, Signet). All blots were developed using Supersignal West Femto Maximum Sensitivity ECL substrate (Pierce).
Heads from flies expressing Aβ42e22g with or without Affibody domains were subjected to mechanical homogenization in 5 M GdmCl, 50 mM Hepes, and 5 mM EDTA followed by 4 min of sonication in a water bath. Homogenates were centrifuged for 7 min at 12,100 g to pellet any GdmCl insoluble material. Supernatants were diluted in 50 mM Hepes and 5 mM EDTA with protease inhibitors to a final concentration of 1 M GdmCl. A sandwich ELISA was performed on the supernatants using biotinylated 6E10 (Signet) and a C terminal Aβx-42-specific antibody 21F12 (kind gift of D. Schenk, Elan). Protein levels were measured using a Sector Imager (Meso Scale Discovery) and normalized to a percentage of the level obtained for flies expressing Aβ42e22g alone.
Flies of all genotypes were crossed with OK107-Gal4 flies (Bloomington Stock No. 854) to drive expression in a subset of neurons that includes, but is not limited to, the mushroom bodies. For each genotype fly brains were dissected in PBS with 0.05% Triton X-100 and fixed in 4% paraformaldehyde for 1 h at room temperature. The brains were then washed three times in PBS/0.05% Triton X-100 and blocked in 5% w/v bovine serum albumin in PBS for 1 h at room temperature. Fly brains were incubated overnight in mouse anti-Aβ (6E10, Signet) diluted 1∶1000 in blocking buffer. After three further washes in PBS/0.05% Triton X-100, brains were then incubated in goat anti-mouse IgG Alexa 546 (Invitrogen) and counterstained with TOTO-3 (Invitrogen) to detect nuclei before mounting in Vectashield (Vectorlabs) anti-fade mounting medium.
Confocal serial scanning images were acquired at 2 or 4 µm intervals (for high magnification and low magnification images, respectively) using a Nikon Eclipse C1si on Nikon E90i upright stand (Nikon). The image stacks were projected using ImageJ (version 1.42k), and the resulting composite images were processed using Photoshop CS4 software (Adobe Systems).
Concentrations of mRNA were determined using quantitative real time PCR (RT-PCR). Twenty-five flies per genotype were collected and snap frozen in liquid N2. RNA was extracted from each group of 25 fly heads using TriZol followed by DNAse treatment to remove residual genomic DNA and reverse transcription to produce cDNA. Each sample was subjected to two separate quantitative PCR reactions to detect Aβ mRNA and the control gene Actin5c. Real time amplification of cDNA was monitored using SYBR Green fluorescence in a Bio-Rad iQ Cycler.
ZAβ3 was produced in Escherichia coli and purified as described elsewhere [16]. Aβ peptides were obtained from a commercial source (rpeptide, Bogart, GA, USA), synthesized in-house, or produced (with an N-terminal methionine) by recombinant co-expression of Aβ and ZAβ3 in E. coli [26]. Experiments were carried out in 20 mM sodium phosphate, 50 mM NaCl, except for the NMR experiments where NaCl was not included, and pH 7.2. 10 µM ThT was added prior to fluorescence measurements.
Fibril formation assays were carried out as described previously [16]. TEM images were obtained using a LEO 912 AB Omega microscope. CD spectra were recorded on a JASCO J-810 spectropolarimeter.
Fibrils were prepared from Aβ40 at a concentration of 100 µM with the same set-up and conditions as for the fibril formation assays, but in the absence of ThT. After 3 days of incubation at 37°C, fibrils were isolated by centrifugation at 16,000 g. To remove any residual soluble peptide, fibrils were washed by resuspension in buffer F [20 mM sodium phosphate, pH 7.2, 0.1% sodium azide, complete protease inhibitor (Roche; at the concentration recommended by the manufacturer)], followed by centrifugation. Fibrils were resuspended in buffer F supplemented with 10% D2O to a final concentration of 300 µM Aβ40 and investigated by 15N HSQC NMR with 24 h of data collection on a Varian Inova 900 MHz NMR spectrometer (equipped with a cryogenic probe) or on a Varian Inova 800 MHz spectrometer. The intensity of resonances originating from bound Aβ40 detected in the presence of 325 µM of unlabeled ZAβ3 was followed over time by recording a series of 24 h 15N HSQC NMR spectra. Five µM of 15N-ZAβ3 served as an internal concentration reference, assuming identical NMR-sensitivities of the intense resonances of the three C-terminal residues of bound Aβ40 and free ZAβ3. Sonication was achieved by placing the NMR tube with the fibril sample into a Misonix water bath sonicator for 2 min before acquisition of NMR data.
Oligomer formation was induced by adjusting the pH of alkaline (pH∼10.5) solutions of Aβ42 (concentration ≤100 µM) in 20 mM sodium phosphate and 50 mM sodium chloride to pH 7.2 (with 1 M HCl) [25]. The samples were incubated at 21°C and oligomer formation was monitored with SEC and ThT fluorescence. Fifty µl (for analytical runs) or 1 ml (for preparative oligomer isolation) aliquots were injected onto an ÄKTA Explorer system (GE Healthcare, Uppsala, Sweden) equipped with a Superdex 75 10/300 column, and the elution was monitored by UV absorbance at 220 nm. Preparative oligomer isolation was carried out 4–20 h after induction of oligomer formation and yielded oligomer solutions at 10–20 µM total Aβ42 concentration. The elution volumes of the ZAβ3:Aβ42 complex and free ZAβ3 were determined in separate runs of the isolated complex or free Affibody, respectively, and conformed to previous SEC studies [19]. The amounts of Aβ42 in the monomeric, oligomeric, or ZAβ3-bound fraction were determined from the elution peak areas obtained by integration using the Unicorn software provided with the chromatography system. The data were normalized by setting to unity the sum of the oligomer and monomer peak areas in the first SEC profiles (at t = 0.2 h for oligomer formation in Figure S6A, and at t = 0.5 h for oligomer dissolution in Figure S6C). The fraction of high molecular weight aggregates that did not enter the column bed was calculated as the difference between unity and the sum of the monomer and oligomer fractions. The fraction of ZAβ3-bound Aβ42 shown in Figure 5D was obtained by comparison of the integrated ZAβ3:Aβ42/free ZAβ3 peak area with those obtained in calibration runs of free ZAβ3 (set to 0) and ZAβ3:Aβ42 complex (set to 1) using the same protein concentrations as in the dissolution experiment. The fraction of Aβ42 bound to ZAβ3 was determined by 15N HSQC NMR employing an internal concentration standard.
|
10.1371/journal.pgen.1004747 | It's All in Your Mind: Determining Germ Cell Fate by Neuronal IRE-1 in C. elegans | The C. elegans germline is pluripotent and mitotic, similar to self-renewing mammalian tissues. Apoptosis is triggered as part of the normal oogenesis program, and is increased in response to various stresses. Here, we examined the effect of endoplasmic reticulum (ER) stress on apoptosis in the C. elegans germline. We demonstrate that pharmacological or genetic induction of ER stress enhances germline apoptosis. This process is mediated by the ER stress response sensor IRE-1, but is independent of its canonical downstream target XBP-1. We further demonstrate that ire-1-dependent apoptosis in the germline requires both CEP-1/p53 and the same canonical apoptotic genes as DNA damage-induced germline apoptosis. Strikingly, we find that activation of ire-1, specifically in the ASI neurons, but not in germ cells, is sufficient to induce apoptosis in the germline. This implies that ER stress related germline apoptosis can be determined at the organism level, and is a result of active IRE-1 signaling in neurons. Altogether, our findings uncover ire-1 as a novel cell non-autonomous regulator of germ cell apoptosis, linking ER homeostasis in sensory neurons and germ cell fate.
| Cells in the C. elegans germline undergo programmed cell death as part of the normal developmental program and in response to various stresses. Here, we discovered that more germ cells undergo programmed cell death under stress conditions associated with the accumulation of misfolded proteins in the endoplasmic reticulum, a cellular organelle responsible for protein folding and trafficking. Surprisingly, we found that germ cell death is a consequence of stress in neurons rather than in the germ cells themselves. This implies that germ cell death under ER stress conditions is regulated at the organismal level and implicates signaling between tissues.
| Apoptosis, also known as programed cell death (PCD), is a highly conserved fundamental cellular process that provides a self-elimination mechanism for the removal of unwanted cells. PCD is critical for organ development, tissue remodeling, cellular homeostasis and elimination of abnormal and damaged cells [1], [2]. The apoptotic machinery that actually executes cell death is intrinsic to all cells and can be activated in response to extracellular or intracellular cues. These are thought to be mediated by cell death receptors or by cytotoxic stress respectively [3].
In C. elegans, 131 somatic cells invariably undergo apoptosis during hermaphrodite development [4], [5]. In contrast, in the adult C. elegans, only germ cells undergo apoptotic cell death. These cell deaths can be either physiological or stress-induced [6], [7]. So far, stress-induced germ cell apoptosis has been associated with DNA damage, pathogens, oxidative stress, osmotic stress, heat shock and starvation [7]–[9]. These apoptotic events are restricted to germ cells at the pachytene stage which are located in the loop region of the gonad [6], where oogenesis transition normally occurs [10]. The physiological germ cell apoptosis pathway acts during oogenesis and is thought to act either as a part of a quality control process, preferentially removing unfit germ cells from the gonad, or as a resource re-allocation factor important for maintaining oocyte quality or simply as a gonad homeostatic pathway removing excess germ cells [6], [11], [12]. Both somatic and germ cell apoptosis rely on the highly conserved core apoptotic machinery comprised of the Caspase-3 homolog ced-3, the Apaf-1 homolog ced-4 and the anti-apoptotic Bcl-2 homolog ced-9 [6], [13]–[16].
All germ cell apoptosis, physiological and stress-induced, relies on the core apoptotic machinery [7]–[9]. However, different upstream genes activate the core apoptotic machinery in the germline in response to different stresses. For example, DNA damage-induced germ cell apoptosis involves the proteins EGL-1, CED-13, and the DNA damage response protein p53 homolog CEP-1 [9], [17]–[19]. In contrast, oxidative, osmotic, heat shock and starvation stresses induce germ cell apoptosis through a CEP-1 and EGL-1 independent pathway and rely on the MEK-1 and SEK-1 MAPKs instead [7].
The endoplasmic reticulum (ER) fulfills many essential cellular functions, including a role in the secretory pathway, in lipid metabolism and in calcium sequestration. Accordingly, ER homeostasis is essential for proper cellular function [20]. A specialized, conserved cellular stress response, called the unfolded protein response (UPR), is in charge of detecting ER stress and adjusting the capacity of the ER to restore ER homeostasis. In C. elegans, as in humans, three proteins located at the ER membrane sense ER stress and activate the UPR: the ribonuclease inositol-requiring protein-1 (IRE-1), the PERK kinase homolog PEK-1 and the activating transcription factor-6 (ATF-6) [21]. Of the three, IRE-1 is the major and most highly conserved ER stress sensor. In response to ER stress, IRE-1 activates the ER stress-related transcription factor XBP-1, which induces the transcription of genes that help restore ER homeostasis [22]–[24]. Accordingly, ire-1 and xbp-1 deficiencies perturb ER homeostasis [25], [26].
Although IRE-1 typically protects cells, upon excessive and prolonged ER stress, IRE-1 can also trigger cell death, usually in the form of apoptosis [27], [28]. For example, IRE-1 can lead to activation of the cell death machinery via JNK and caspase activation [29], [30] or by mediating decay of critical ER-localized mRNAs through the RIDD pathway, tipping the balance in favor of apoptosis [31]. These functions of IRE-1 are independent of XBP-1 [29]–[33].
Highly proliferating cells with a high protein and lipid biosynthetic load are thought to rely on ER function to a greater extent than other cells. This together with the general sensitivity of the germline to cellular stresses prompted us to investigate the effects of ER stress on germ cell fate. Strikingly, we discovered that ER stress does not simply kill the germ cells by not meeting their biosynthetic demands. Instead, we found that ER stress initiates a signaling cascade in neurons that regulates germ cell survival non-autonomously. Thus, our findings reveal that germ cell sensitivity to ER stress conditions can be regulated at an organismal level and can be uncoupled from germ cell stress.
To investigate whether ER stress induces apoptosis in the C. elegans germline, we first assessed the number of apoptotic corpses in the gonads of animals treated with tunicamycin, a chemical ER stress inducer which blocks N-linked glycosylation. Apoptotic corpses in the gonad were identified by staining with the vital dye SYTO12 and by their discrete cellularization within the germline syncytium. We found that tunicamycin treatment increased the number of apoptotic germ cells present in wild-type gonads by approximately 3 fold compared to control DMSO treatment from day-1 to day-3 of adulthood (P<0.001, Figure 1A–B).
If indeed the increased number of germline corpses in tunicamycin-treated animals is a consequence of ER stress, then additional manipulations that disrupt ER homeostasis should also increase germ cell apoptosis. tfg-1 encodes a protein that directly interacts with SEC-16 to control COPII subunit accumulation at ER exit sites and is required for the vesicular export of cargo from the ER [34]. We hypothesized that ER homeostasis would be disrupted in tfg-1-deficient animals. To examine the effect of tfg-1 deficiency on ER homeostasis, we assessed the effect of tfg-1 RNAi treatment on the levels of the ER stress response reporter Phsp-4::gfp [22]. tfg-1 RNAi efficacy was confirmed by the reduction in the animals' body size compared to control RNAi treated animals [35]. We found that treatment with tfg-1 RNAi specifically activated the ER stress response, as it increased the level of the ER stress response reporter without increasing the expression of oxidative stress response, heat shock response or mitochondrial stress response reporters (Figure S1).
In terms of germ cell apoptosis, we observed that tfg-1 RNAi consistently increased the number of apoptotic germ cells in the gonad by approximately 4 fold from day-1 to day-3 of adulthood compared to wild-type animals (P<0.001, Figure 2A,B). A similar 4 fold increase in germ cell apoptosis was observed by scoring germ cell engulfment by neighboring cells that expressed GFP-labeled CED-1, a transmembrane receptor that mediates cell corpse engulfment in C. elegans [36] (Figure 2C). tfg-1 RNAi treatment did not increase the number of SYTO12-labeled cells in the gonads of apoptosis-defective ced-3(n1286) mutants, confirming that the dye specifically labels apoptotic cells (Figure S2). Together, these results indicate that conditions that disrupt ER homeostasis, including tunicamycin treatment or blocking secretory traffic from the ER, increase apoptosis frequency in the gonad compared to non-stressed animals.
We next asked which apoptotic machinery is implicated in ER stress-induced germ cell apoptosis. To this end, we examined mutants deficient in core-apoptotic genes as well as mutants deficient in genes specifically implicated in germ cell apoptosis. This array of apoptosis-related mutants was treated with control or tfg-1 RNAi, and germ cell apoptosis was scored by SYTO12 labeling. As expected, we found that the core apoptosis machinery genes ced-3 and ced-4 [6], [13], [15], [16] were required for germ cell apoptosis in response to ER stress (Figure 3A). Importantly, the cep-1, egl-1 and ced-13 genes, previously implicated in DNA damage-induced apoptosis [9], [17]–[19], were also found to be completely essential for germ cell apoptosis in response to tfg-1 RNAi treatment (Figure 3A). Accordingly, the levels of a CEP-1::GFP translational fusion transgene driven by the cep-1 promoter were increased within the germ cells of tfg-1 RNAi-treated animals (P<0.001, Figure 3B). In contrast, pmk-1 and sek-1, previously implicated in oxidative stress-induced and pathogen-induced germ cell apoptosis respectively [7], [8], were dispensable for germ cell apoptosis in response to tfg-1 RNAi treatment (Figure 3C). Thus, the genetic analysis clearly implicated the apoptotic machinery that mediates DNA damage-induced germ cell apoptosis in ER stress-induced germ cell apoptosis as well.
The strong changes in CEP-1 levels observed in tfg-1 RNAi-treated animals suggest that ER stress controls CEP-1 activation within the germ cells. One possible explanation for the involvement of genes implicated in DNA damage-induced germ cell apoptosis is that ER stress indirectly damages DNA, which in turn leads to CEP-1 activation and DNA damage-induced germ cell apoptosis. However, whereas a previous study implicated the intestinal kri-1 gene in non-autonomous regulation of ionizing-radiation induced germ cell apoptosis [37], we found that tfg-1 RNAi treatment efficiently induced germ cell apoptosis in kri-1-deficient animals (Figure 3C). This genetic uncoupling between the requirements for ionizing-radiation induced germ cell apoptosis and ER stress-induced germ cell apoptosis suggest that ER stress does not simply induce DNA damage which in turn leads to germ cell apoptosis.
To further substantiate this conclusion, we examined directly whether the DNA damage response is activated in the germ cells of ER stressed- animals treated with tfg-1 RNAi. To this end, we followed the nuclear aggregation of HUS-1::GFP, which encodes a DNA damage checkpoint protein that relocalizes to distinct nuclear foci upon induction of DNA damage [38]. Whereas nuclear HUS-1::GFP aggregates were clearly observed in the germ cells of DNA-damaged rad-51 RNAi treated animals (P<0.001 compared to control RNAi), HUS-1::GFP aggregates were not detected in tfg-1 RNAi treated animals (P = 0.78 compared to control RNAi, Figure 3D). Thus, ER stress activates CEP-1/p53 to induce germ cell apoptosis without generating DNA damage.
We next asked whether any of the canonical ER stress sensing genes was implicated in ER stress-induced germ cell apoptosis. To this end, we examined mutants deficient in the ER stress-response sensor genes that comprise the UPR: ire-1, pek-1 or atf-6. This array of mutants was treated with DMSO or with tunicamycin and germ cell apoptosis was scored by SYTO12 labeling. We found that similarly to its effect in wild-type animals, tunicamycin treatment increased the number of germline corpses in atf-6 and pek-1-deficient animals by approximately 3 fold from day-1 to day-3 of adulthood (P<0.001, Figure 1A,C). In contrast, ER stress induced by tunicamycin treatment failed to increase germ cell apoptosis in ire-1 mutants (Figure 1A,B). Therefore, the insensitivity of the germline to tunicamycin is unique to ire-1-deficient animals and not seen in animals deficient in other UPR sensors.
ire-1 mutants are abnormal in terms of their gonad anatomy and their reproductive capacity: ire-1 mutants have approximately 2 fold less progeny and 2 fold less mitotic germ cells within their proliferative zones compared to ire-1(+) wild-type animals (P<0.001, Figures S3A,D). Thus, we wondered whether these abnormalities affected the ability of their germ cells to undergo apoptosis in general or whether they were specifically defective in their ability to undergo apoptosis in response to ER stress.
First we assessed germline apoptosis in ire-1(−) mutants under normal growth conditions, from day-0 (L4) to day-3 of adulthood. At all timepoints, we detected approximately half the amount of germline corpses in ire-1(−) gonads compared to ire-1(+) wild-type gonads, as assessed by SYTO12 labeling and by the CED-1::GFP engulfment marker (Figure 2A–C). The low levels of germ cell apoptosis persisted in ire-1 mutants treated with vps-18 RNAi (Figure 2D), a treatment that impairs germ cell corpse clearance [39]. However, normalization of the number of apoptotic corpses to the number of mitotic germ cells resulted in comparable levels of germ cell corpses in ire-1 mutants and in non-stressed wild-type animals (P = 0.12, Figure S3C). This indicates that in spite of their reproductive abnormalities, physiological germ cell apoptosis in ire-1(−) and ire-1(+) animals is comparable.
Next, we assessed stress-induced germline apoptosis in ire-1 mutants. We found that although manipulations that disrupt ER homeostasis fail to increase germ cell apoptosis in ire-1 mutants (Figure 1B,1D–E), DNA damage and oxidative stress conditions did increase germ cell apoptosis in ire-1 mutants (P<0.001 compared to non-stressed ire-1 mutants, Figure 1D–E). This resulted in a similar level of germline apoptosis as in stressed wild-type animals upon normalization of the number of apoptotic corpses to the number of mitotic germ cells (P = 0.44 for DNA damage and P = 0.42 for oxidative stress, Figure 1D–E). Thus, in spite of the reproductive abnormalities of ire-1 mutants, the germ cells of these mutants undergo stress-induced apoptosis similarly to wild-type animals, however not in response to ER stress. The inability of ire-1 mutants to increase germline apoptosis specifically in response to perturbations in ER homeostasis suggests that IRE-1 may be a critical mediator of ER stress-induced germ cell apoptosis.
The most established mode of action of IRE-1 under ER stress conditions is via the activation of the UPR-related transcription factor XBP-1 [22]–[24]. Therefore, if IRE-1 enabled ER stress-induced germ cell apoptosis via its downstream target xbp-1, then the number of germline corpses detected in xbp-1(−) mutants should remain low under ER stress conditions, similarly to ire-1(−) mutants.
In order to test this, we first examined germ cell apoptosis in xbp-1(tm2457) null mutants. Surprisingly, in contrast to ire-1(−) mutants, we consistently detected increased germ cell apoptosis in xbp-1(−) gonads compared to wild-type gonads under normal growth conditions. A 2.5 fold increase in the number of germline corpses in xbp-1(−) mutants was detected by SYTO12 labeling of gonads from day-1 to day-3 of adulthood compared to wild-type animals (P<0.001, Figure 2A–B). A similar observation was apparent by using the CED-1::GFP engulfment marker (Figure 2C). The 2.5 fold increase in the number of germ cell corpses was still apparent in engulfment defective vps-18 RNAi-treated animals (Figure 2D) and upon normalization to the number of mitotic germ cells located in the proliferative zone (Figure S3C). Thus, in contrast to ire-1 mutants and wild-type animals, xbp-1 mutants exhibit a high basal level of germ cell apoptosis.
We hypothesized that the increase in germline apoptosis in xbp-1 mutants may be due to perturbed ER homeostasis in these animals [25], [26], If so, then it should be mediated via ire-1, similarly to other ER stress conditions that induce germ cell apoptosis. Accordingly, we found that in an ire-1(−) background, the xbp-1 mutation did not increase germ cell apoptosis. This observation was consistent along different time points spanning from day-0 (L4) to day-3 of adulthood (Figure 2A–B). This also persisted upon normalization to the number of mitotic germ cells located in the proliferative zone (Figure S3C). Interestingly, the amount of mitotic germ cells of xbp-1; ire-1 double mutants was similar to that of xbp-1 single mutants (P = 0.072, Figure S3A), whose germ cells were responsive to ER stress-induced apoptosis (Figure S3B–C). This indicates that the reduced amount of mitotic cells in the gonad of ire-1 mutants can be uncoupled from the inability of their germ cells to undergo ER stress-associated apoptosis.
The finding that xbp-1 deficiency per se promotes ire-1-dependent ER stress-induced germ cell apoptosis suggests that xbp-1 is dispensable for increasing germ cell apoptosis in response to ER stress. Consistent with this, we found that tunicamycin treatment further increased germ cell apoptosis in xbp-1 mutants (P<0.001, Figure 1A–B). Altogether, these results lend further support to the notion that ire-1 is a critical signaling molecule in mediating ER stress-induced germline apoptosis, whereas its' downstream canonical target xbp-1 is not. Furthermore, since ER function is compromised both in ire-1 and in xbp-1 deficient mutants [25], [26], the differential ability to induce germ cell apoptosis in these mutants suggests that germ cell apoptosis may be the result of active IRE-1 signaling, rather than simply a consequence of ER dysfunction.
In mammalian cells, activation of IRE1 can cell-autonomously activate JNK via the adaptor protein TRAF. Consequently, IRE1-mediated activation of JNK initiates proapoptotic signaling, independently of XBP1 [29]. Thus, we examined whether the C. elegans homologs of TRAF and JNK proteins were required for ire-1/ER stress-induced apoptosis in C. elegans, which is also independent of xbp-1. To this end, trf-1 mutants or mutants deficient in all three C. elegans JNK homologs were treated with control or tfg-1 RNAi. We found that tfg-1 RNAi increased germ cell apoptosis independently of the trf-1 and the JNK-like genes (Figure 3E). Thus, since ER stress can effectively induce germ cell apoptosis in the absence of xbp-1, trf-1 and JNK homologs, the signaling mediated by IRE-1, in this case, must be executed by an alternative xbp-1-independent output of IRE-1.
Next, we examined whether ER stress triggers programmed cell death autonomously within the germ cells, or non-autonomously from the soma. To test this, we used tfg-1 RNAi to induce ER stress specifically in the germline or in the soma. To induce ER stress primarily in the germ cells, mutants in the rrf-1 gene, encoding an RNA-directed RNA polymerase (RdRP) homolog required for most somatic RNAi but not for germline RNAi [40], were treated with tfg-1 RNAi. No increase in the amount of germline corpses was observed as a result of tfg-1 RNAi treatment in rrf-1 mutants (P = 0.19, Figure 4A). To induce ER stress specifically in the soma, mutants in the ppw-1 gene, which is required for efficient RNAi in the germline [41], were treated with tfg-1 RNAi. This resulted in a 4.5 fold increase in the amount of apoptotic corpses in the gonads (P<0.001, Figure 4A). Thus, ER stress in the soma, rather than in the germ cells, is sufficient for the induction of germ cell apoptosis.
Does germ cell apoptosis occur upon disruption of ER homeostasis in the entire soma or does it occur in response to ER stress in a particular part of the soma? To answer this, ER stress was induced locally in specific somatic tissues. This was achieved by treating animals expressing functional RNAi machinery only in specific tissues with tfg-1 RNAi and assessing germ cell apoptosis in these animals. We found that tfg-1 RNAi treatment did not increase germ cell apoptosis in animals which respond to RNAi only in the intestine, in the muscle, in the hypodermis, in the uterine or in the distal tip cells (P>0.1 in each one of these strains, Figure 4B). In contrast, tfg-1 RNAi treatment increased germ cell apoptosis by approximately 7 fold in animals which respond to RNAi specifically in the neurons (P<0.001, Figure 4A).
Next, we examined whether ER stress-induced germline apoptosis is under pan-neuronal control or under the control of specific neurons. To this end, we introduced ER stress-inducing tfg-1 RNAi into animals expressing functional RNAi machinery specifically in the cholinergic, glutamatergic, GABAergic, dopaminergic or in a subset of sensory neurons. Importantly, we found that tfg-1 RNAi treatment increased germ cell apoptosis only in animals whose sensory neurons responded to RNAi (Figure 4C).
Among the sensory neurons whose exposure to ER stress increased germline apoptosis were the ASI neurons, which have been previously implicated in the regulation of germ cell proliferation and maturation [42]. Hence, we examined whether ER stress in the ASI sensory neurons alone is sufficient for the induction of germ cell apoptosis in the gonad. To this end, we first assessed germline apoptosis in daf-28(sa191) mutants, which produce a toxic insulin peptide that activates the UPR specifically in the ASI neurons [43]. We found that germ cell apoptosis in the gonads of daf-28(sa191) mutants was increased by approximately 4 fold compared to wild-type animals (P<0.001, Figure 4D). Importantly, germ cell apoptosis was not increased in a daf-28(tm2308) null strain, which is deficient in daf-28 and does not produce the toxic insulin peptide which induces ER stress (P = 0.15, Figure 4D). tfg-1 RNAi treatment of the two daf-28 mutant strains increased the number of germline corpses in daf-28(tm2308) null strain (P<0.001), but did not further increase germline apoptosis in the daf-28(sa191) strain (P = 0.09, Figure 4D). tfg-1 RNAi treatment did not alter ASI overall morphology as assessed by the expression pattern of a GFP reporter driven by an ASI-specific promoter (Figure S4A). Together, these findings suggest that expression of the toxic form of DAF-28 and tfg-1-deficiency increase germ cell apoptosis by similar means; most likely by causing ER stress and activating the UPR in the ASI neurons.
We have demonstrated that in the absence of the ER stress sensor ire-1, ER stress does not increase germline apoptosis. We further demonstrated that ER stress in the ASI sensory neurons is sufficient to induce germ cell apoptosis. Thus, we next examined whether it is also sufficient to express ire-1 in the soma, and specifically in the ASI neurons, to restore germ cell apoptosis in response to ER stress.
To this end, we restored ire-1 expression in the entire soma, pan-neuronally or specifically in the ASI/ASJ neurons of ire-1(−) mutants. This was achieved using multi-copy ire-1 transgenes under ire-1, rgef-1 and daf-28 promoters respectively. Since the expression of multi-copy transgenes is normally suppressed in germ cells [44], and due to the specificity of their promoters, these transgenes restore ire-1 expression within different parts of the soma but not in the germline. We found that expression of each of these ire-1 transgenes completely restored the increase in germline apoptosis in response to treatment with tfg-1 RNAi (P<0.001, compare white and black bars within each strain in Figure 5A). Similarly, we restored ire-1 expression in muscle cells and in the PVD and OLL neurons using multi-copy ire-1 transgenes under myo-3 and ser-2 promoters respectively. No increase in germline apoptosis in response to tfg-1 RNAi treatment was apparent in these two transgenic lines compared to control RNAi treatment (P>0.1, Figure 5A). The fact that not all ire-1 transgenes induced apoptosis supports the notion that ire-1-induced germline apoptosis is not the result of leaky expression of the transgenes in other tissues. Altogether, this implies that not all tissues and not all neurons are involved in the regulation of this process.
Next, we asked whether increasing IRE-1 levels to a greater extent may be sufficient for inducing germline apoptosis even in the absence of ER stress. To this end, we overexpressed ire-1 transgenes in various tissues or cells of ire-1(+) wild-type animals. This was achieved by using multi-copy ire-1 transgenes under ire-1, rgef-1, daf-28 and daf-7 promoters. This is consistent with the interpretation that some activation of IRE-1 is achieved merely by its over-expression, as has been previously observed in yeast and in mammalian cells [45], [46]. We found that this artificial activation of IRE-1 in the soma, pan-neuronally or specifically in the ASI/ASJ neurons of ire-1(+) animals was sufficient to induce high levels of germ cell apoptosis (P<0.001, compare white bars of transgenic animals to that of wild-type animals in Figure 5B). No increase in germ-cell apoptosis was observed upon overexpression of an ire-1 transgene in muscle cells or in the AIY neurons in ire-1(+) animals (P>0.1, Figure 5B). These findings support the claim that the rescuing activity of the ire-1 transgenes stems from their expression in specific neurons.
tfg-1 RNAi treatment of ire-1(+) animals over-expressing the ire-1 transgenes in the soma, in the neurons and specifically in the ASI/ASJ neurons did not further increase germline apoptosis (P>0.5, compare white and black bars within the strains, Figure 5B). This suggests that IRE-1 overexpression and tfg-1-deficiency increase germ cell apoptosis by similar means, i.e. by activating IRE-1. Taken together, our data demonstrate that activation of ire-1 specifically in the ASI neurons, either by ER stress in the ASI neurons or by IRE-1 overexpression, can non-autonomously regulate germ cell apoptosis. Furthermore, since over-expression of transgenic IRE-1 is sufficient for its artificial activation in a manner that is independent of ER stress, this further suggests that active IRE-1 signaling in the ASI neurons per se, rather than neuronal ER stress or ER dysfunction, is the cause of germ cell apoptosis.
Understanding the molecular events that regulate the life-death decision of cells is of fundamental importance in cell biology research, cell development, cancer biology and disease biology [47]. In this study, we gained new and fascinating insights into the complex coupling between ER stress in the nerve system and germ cell apoptosis.
We report for the first time that germ cells undergo apoptosis in response to ER stress. We find that activation of the ER stress response gene ire-1 is required and sufficient to induce germ cell apoptosis in response to several ER stress-inducing conditions. Strikingly, we find that germ cell fate is regulated non-autonomously by ER stress and/or through IRE-1 activation specifically in the ASI neurons. This implies that ER homeostasis and UPR signaling in the germ cells themselves is not a factor in determining their fate, ruling out the possibility that these apoptotic events are part of a quality control process that removes “stress-damaged” germ cells from the gonad [6], [11], [12]. Furthermore, this assigns a central neuroendocrine role for the ASI neuron pair in coupling between stress sensing and the onset of germ cell apoptosis. This is in addition to other central physiological processes in C. elegans, such as dauer formation [48], [49] and longevity [50], [51], that are also controlled by the sensory ASI neuron pair. Interestingly, another pair of sensory neurons, the ASJ neurons, has been previously implicated in the protection of germ cells from apoptosis under hypoxic conditions [52]. Thus, depending on the stress condition, different neurons can shift germ cell fate from survival to death or vice versa.
How might IRE-1 activation in the ASI neurons dictate germ cell survival or death? One possibility is that defects associated with ire-1 deficiency and/or ire-1 activation indirectly abrogate the communication between the neurons and the gonad. However, several lines of evidence undermine this hypothesis: (1) We find that ER stress-induced germ cell apoptosis proceeds normally in animals with a severely defective nervous system (Figure S4B,C). This implies that germline apoptosis does not result from a generic neuronal defect. (2) ire-1 deficiency is associated with germline abnormalities which include a significant reduction in the number of mitotic germ cells and in reduced progeny number. However, these gonad-related defects do not confer generic resistance to stress-induced apoptosis as the germ cells of ire-1 mutants do undergo apoptosis in response to a variety of stresses. Furthermore, a mutation in xbp-1, which improved the reproductive abnormalities of ire-1 mutants, did not restore responsiveness to ER stress induced germ cell apoptosis in ire-1; xbp-1 double mutants, thus uncoupling the two. (3) Whereas the comparison of germ cell apoptosis in ire-1 and wild-type animals may be confined by the basal discrepancy of their reproductive systems, this concern does not exist in the analysis of ire-1 overexpressing strains, whose gonad appears to be normal (P>0.1 for Pire-1::ire-1 and Pdaf-7::ire-1 compared to wild-type animals Figure S3A,D). Similarly, this concern does not exist in the intra-strain comparisons of germline apoptosis within the ire-1(−) strain under control and stress conditions.
If ire-1 misregulation in the ASI neurons does not indirectly abrogate the communication between the neurons and the gonad, how might it dictate germ cell survival or death? IRE-1 is a dual-activity enzyme, bearing both kinase and endoribonuclease activities and a propensity to self-aggregate at the ER membrane in response to ER stress. The most characterized mode of action of IRE-1 is the activation of its downstream transcription factor XBP-1 [53]. Significantly less characterized are XBP-1 independent targets of IRE-1, that include activation of the cell death machinery via JNK/TRAF signaling and degradation of ER-localized mRNAs that encode secreted and membrane proteins in a process called RIDD [29], [30], [32], [54]–[58]. Since we find that ER stress can effectively induce germ cell apoptosis in the absence of xbp-1, trf-1 and JNK homologs, the signaling mediated by IRE-1 in this case may be executed by the RIDD pathway or via a novel, yet undescribed, xbp-1-independent output of IRE-1.
We propose that activation of IRE-1 in the neurons (either as a result of ER stress or merely by its over-expression) actively regulates the production of a germ cell regulatory signal. In principle this may be a germ cell proapoptotic signal produced by the neurons upon IRE-1 activation. Alternatively, this may be a germ cell anti-apoptotic signal that is down-regulated by IRE-1 upon its activation. This ASI-regulated signal, whose identity and nature remain to be elucidated, propagates in the animal and affects the gonad where it acts upstream to the p53 homolog cep-1, activating the same apoptotic machinery in the germ cells as the DNA damage response, without inducing DNA damage in the germ cells (Figure 6). This indicates the existence of a new pathway that can activate CEP-1 independently of DNA damage upon activation of neuronal IRE-1.
Interestingly, in adult animals, exposure to ER stress or activation of IRE-1 in the soma induce apoptosis only in germ cells, as we did not detect any apoptotic corpses outside of the gonad of these animals. This is in contrast to the developing embryo, where exposure to ER stress can induce apoptosis in the soma [35], [59]. We propose that as the organism completes development, its ability to respond or execute programmed cell death upon exposure to ER stress is maintained in mitotic germ cells while being selectively abrogated in the post-mitotic soma, as has been demonstrated for their ability to execute apoptosis in response to DNA-damage [60]. This resistance of the soma is important in terms of survival of the animal as it prevents cell death of somatic tissues that lack stem cell pools and regenerative capacity, while allowing cell death of immortal germline cells at times of stress.
What could be the advantage in diluting the germ cell pool when neurons “feel” ER-stressed (i.e. when IRE-1 is activated naturally by ER stress or artificially by overexpression)? Recent studies demonstrate a tight inverse correlation between germ-cell proliferation and the maintenance of somatic proteostasis and longevity [61]–[65]. This inverse correlation is thought to be due to a limitation of resources shared by the germline and the soma and due to altered metabolic and cellular repair mechanisms in the soma that are enabled upon germ cell loss. Previous studies implicated the nervous system in systemic and hierarchical control of cellular stress responses elsewhere in the soma to maintain organismal homeostasis [66]–[72]. Our data further imply that neurons also have the ability to communicate with the germ cells to promote their death in response to stress in the ER. This, in turn, may orchestrate a proteostasis switch in the soma at the expense of a replenishable germ cell pool in times of stress. This adds a new layer of complexity to our understanding of how protein homeostasis is regulated and coordinated across tissues in multicellular organisms.
For single time-point experiments, the number of apoptotic germ cells was scored in day-2 animals stained with SYTO12 (Molecular Probes) as previously described [6]. For time course experiments, the number of SYTO12-labeled apoptotic corpses per gonad arm was scored in animals from day-0 (L4) to day-3 of adulthood. Where indicated, the average number of apoptotic corpses was normalized to the number of mitotic germ cells within the proliferative zone of the gonads, determined by section analysis of DAPI-stained gonads.
Day-1 adult animals were placed in 200 µl of M9 (control) or 10 mM paraquat (oxidative stress) for 1.5 h at 20°C. After the incubation period, 1 ml of M9 was added to dilute the paraquat. Animals were then transferred to eppendorfs with SYTO12 staining for 4.5 hrs. Animals were allowed to recover on plates for 40 min. Finally, the animals were mounted and observed under the microscope to determine cell corpse numbers.
Gonads of day-1 adults were dissected, fixed, and stained with DAPI as previously described [10].
Bacteria expressing dsRNA were cultured overnight in LB containing tetracycline and ampicillin. Bacteria were seeded on NGM plates containing IPTG and carbenicillin. RNAi clone identity was verified by sequencing. Eggs were placed on plates and synchronized from day-0 (L4).The efficacy of the tfg-1 RNAi was confirmed by the animals' reduced body size [35].
Animals were anaesthetized on 2% agarose pads containing 2 mM levamisol.
Images were taken with a CCD digital camera using a Nikon 90i fluorescence microscope. For each trial, exposure time was calibrated to minimize the number of saturated pixels and was kept constant through the experiment. The NIS element software was used to quantify mean fluorescence intensity as measured by intensity of each pixel in the selected area.
Error bars represent the standard error of the mean (SEM) of at least 3 independent experiments. P values were calculated using the unpaired Student's t test.
The following lines were used in this study: N2, CF2012: pek-1(ok275) X, CF2988: atf-6(ok551) X, CF2473: ire-1(ok799) II, CF3208: xbp-1(tm2457) III, SHK62: ire-1(ok799) II; xbp-1(tm2457) III, MD701: Plim-7 ced-1::gfp V, xbp-1(tm2482) III, CF2185: ced-3(n1289) IV, MT2547: ced-4(n1162) III, TJ1: cep-1(gk138) I, MT8735: egl-1(n1084n3082) V, FX536: ced-13(tm536) X, WS1433: hus-1(op241) I; unc-119(ed3) III; opIs34, CF2052: kri-1(ok1251) I, KU25: pmk-1(km25) IV, AU1: sek-1(ag1) X, CF3030: kgb-1(um3) kgb-2(gk361) jnk-1(gk7) IV, NS2937: trf-1(nr2014) III, CF2260: zcIs4[Phsp-4::gfp] V, CL2166: Pgst-4::gfp(dvls19) V, CF1553: muIs84 [(pAD76) Psod-3::gfp+rol-6], SJ4100: Phsp-6::gfp(zcIs13) V, CL2070: Phsp-16.2::gfp(dvls70) V, SHK57: xbp-1(tm2457) III; ced-3(n1286) IV, SHK189: zcIs4[Phsp-4::gfp] V; Pire-1::ire-1, NL2098: rrf-1(pk1417) I, NL2550: ppw-1(pk2505) I, SHK185: ire-1(ok799) II; Prgef-1::ire-1, SHK4: Pire-1::ire-1, SHK182: Prgef-1::ire-1, BB22 rde-4(ne299) III; adr-2(gv42) III, TG12: cep-1(lg12501) I; unc-119(ed4) III; gtIs1 [CEP-1::GFP+unc-119(+)], SHK8: Pmyo-3::ire-1, SHK14: Pmyo-3::ire-1; ire-1(ok799) II, Pser-2::ire-1; ire-1(ok799) II, SHK15: Pdaf-28::ire-1; ire-1(ok799) II, SHK6: Pdaf-28::ire-1, SHK237: Pttx-3::ire-1; Pdaf-7::gfp, SHK234: Pdaf-7::ire-1; Pdaf-7::gfp, daf-28(tm2308) V, CF2638: daf-28(sa191) V, VB1605: svls69[Pdaf-28::daf-28::gfp], SHK11: ire-1(ok799) II; svls69[Pdaf-28::daf-28::gfp], SHK27: ire-1(ok799) II; Pire-1::ire-1; svls69[Pdaf-28::daf-28::gfp], SHK60: unc-13(e51) I, SK7: unc-64(e246) III; unc-31(e928) IV and MT6308: eat-4 (ky5) III.
The following strains were used for tissue-specific RNAi experiments: TU3401: sid-1(pk3321) V; Punc-119::sid-1 (neuron only RNAi), VP303: rde-1(ne213) V; Pnhx-2::rde-1 (intestine only RNAi), WM118: rde-1(ne300) V; Pmyo-3::rde-1(muscle only RNAi), NR222: rde-1(ne219) V; Plin-26::rde-1 (hypodermis only RNAi), NK640: rrf-3(pk1426)II; rde-1(ne219) V; Pfos-1A::rde-1 (uterine only RNAi),JK4143: rde-1(ne219) V; Plag-2::rde-1::gfp (distal tip cell only RNAi).
The following strains were used for neuron-specific RNAi treatments as previously described [73]: XE1581: wpSi10 II [unc-17p::rde-1::SL2::sid-1+Cbr-unc-119(+)]; eri-1(mg366) IV; rde-1(ne219) V; lin-15B(n744) X - Cholinergic neuron-specific RNAi strain. XE1375: wpIs36 I [unc-47p::mCherry]; wpSi1 II [unc-47p::rde-1::SL2::sid-1+Cbr-unc-119(+)]; eri-1(mg366) IV; rde-1(ne219) V; lin-15B(n744) X - GABAergic neuron-specific RNAi strain. XE1582: wpSi11 II [eat-4p::rde-1::SL2::sid-1+Cbr-unc-119(+)] II.; eri-1(mg366) IV; rde-1(ne219) V; lin-15B(n744) X - Glutamatergic neuron-specific RNAi strain. XE1474: wpSi6 II [dat-1p::rde-1::SL2::sid-1+Cbr-unc-119(+)] II; eri-1(mg366) IV; rde-1(ne219) V; lin-15B(n744) X - Dopaminergic neuron-specific RNAi strain, SHK231: sid-1(pk3321) V; Pche-12::sid-1(+); rol-6(su1006).
Prgef-1::ire-1 - ire-1 cDNA was cloned under the 3.5 kb rgef-1 (F25B3.3) promoter and injected at 5 ng/µl with Pmyo-3::mCherry at 50 ng/µl.
Pire-1::ire-1 - ire-1 cDNA was cloned under the 4.5 kb ire-1 (C41C4.4) promoter in the L3691 vector and injected at 25 ng/µl with rol-6 at 100 ng/µl. Pttx-3::ire-1 - was created by cloning the ire-1 cDNA into a Pttx-3 vector [74] using KpnI/SphI.
Pdaf-7::gfp and Pdaf-7::ire-1 - daf-7 promoter fragment [49] was cloned into pPD95.75 (gift from A. Fire, Carnegie Institute) using SphI/XbaI to create daf-7p::gfp transcriptional fusion. The gfp fragment was then replaced by ire-1 cDNA using XmaI/AflII to make daf-7p::ire-1. daf-7p::ire-1 or ttx-3::ire-1 were injected at 10 ng/µl with daf-7p::gfp and pRF4 (rol-6) at 20 ng/µl each.
ser-2prom-3::ire-1 - was created by cloning ire-1 cDNA under ser-2prom-3 fragment [75] using XmaI/AflII. ser-2prom-3::ire-1 was injected at 10 ng/µl with ttx-3::mCherry at 40 ng/µl.
|
10.1371/journal.pntd.0002831 | Distribution and Clinical Manifestations of Cryptosporidium Species and Subtypes in HIV/AIDS Patients in Ethiopia | Cryptosporidiosis is an important cause for chronic diarrhea and death in HIV/AIDS patients. Among common Cryptosporidium species in humans, C. parvum is responsible for most zoonotic infections in industrialized nations. Nevertheless, the clinical significance of C. parvum and role of zoonotic transmission in cryptosporidiosis epidemiology in developing countries remain unclear.
In this cross-sectional study, 520 HIV/AIDS patients were examined for Cryptosporidium presence in stool samples using genotyping and subtyping techniques. Altogether, 140 (26.9%) patients were positive for Cryptosporidium spp. by PCR-RFLP analysis of the small subunit rRNA gene, belonging to C. parvum (92 patients), C. hominis (25 patients), C. viatorum (10 patients), C. felis (5 patients), C. meleagridis (3 patients), C. canis (2 patients), C. xiaoi (2 patients), and mixture of C. parvum and C. hominis (1 patient). Sequence analyses of the 60 kDa glycoprotein gene revealed a high genetic diversity within the 82 C. parvum and 19 C. hominis specimens subtyped, including C. parvum zoonotic subtype families IIa (71) and IId (5) and anthroponotic subtype families IIc (2), IIb (1), IIe (1) and If-like (2), and C. hominis subtype families Id (13), Ie (5), and Ib (1). Overall, Cryptosporidium infection was associated with the occurrence of diarrhea and vomiting. Diarrhea was attributable mostly to C. parvum subtype family IIa and C. hominis, whereas vomiting was largely attributable to C. hominis and rare Cryptosporidium species. Calf contact was identified as a significant risk factor for infection with Cryptosporidium spp., especially C. parvum subtype family IIa.
Results of the study indicate that C. parvum is a major cause of cryptosporidiosis in HIV-positive patients and zoonotic transmission is important in cryptosporidiosis epidemiology in Ethiopia. In addition, they confirm that different Cryptosporidium species and subtypes are linked to different clinical manifestations.
| The disease burden of Cryptosporidium parvum and role of zoonotic transmission in cryptosporidiosis epidemiology are poorly understood in developing countries. In this study, we examined the distribution and clinical manifestations of Cryptosporidium species and subtypes in HIV/AIDS patients in Addis Ababa, Ethiopia. Using molecular diagnostic tools, we detected Cryptosporidium infection in 26.9% of 520 HIV/AIDS patients studied. We have shown a very high diversity of Cryptosporidium species and subtypes in these patients, but unlike in other developing countries, C. parvum is overwhelmingly the dominant species in the study community, responsible for ∼65% Cryptosporidium infections. The common occurrence of C. parvum zoonotic subtype family IIa, combined with calf contact as a significant risk factor, suggest that zoonotic transmission is important in cryptosporidiosis epidemiology in HIV/AIDS patients in Ethiopia. We have also shown that different Cryptosporidium species and subtypes are linked to different clinical manifestations. Improved hygiene and avoidance of calf contact should be advocated to reduce cryptosporidiosis transmission in HIV/AIDS patients in the study setting.
| Cryptosporidium is an important protozoan parasite affecting HIV/AIDS patients, causing diarrhea, wasting syndrome, and reduced life quality [1]. Since specific therapy or vaccine for the control of this parasite is not yet available, preventing infections depends on avoiding exposure to the parasite and maintaining immune competence. In industrialized nations, access to highly active antiretroviral therapy (HAART) has significantly reduced the morbidity and mortality by cryptosporidiosis [2]–[4]. Nonetheless, cryptosporidiosis is still a major threat to AIDS patients who do not have access to HAART, especially in developing countries [5]–[8]. In industrialized nations, transmission of cryptosporidiosis via contaminated drinking and recreational water and contact with infected farm animals remains a major public health problem in both HIV-positive and immunocompetent persons [9]–[11].
The use of molecular epidemiologic tools has provided new insights into the diversity of Cryptosporidium species infecting humans and animals [12]. So far, 26 Cryptosporidium species have been described [13]–[17]. Most human cases are caused by C. hominis and C. parvum [12]. The latter also infects some other mammals, notably calves and lambs, and is responsible for most zoonotic infections in humans. Several other Cryptosporidium species are seen in humans at lower frequency, including C. meleagridis, C. felis, C. canis, C. ubiquitum, and C. cuniculus [12]. More recently, a new species, C. viatorum, has been described in 10 travelers returning to Great Britain from the Indian subcontinent [17]. This species appears to be a human-specific pathogen and has since been found in 2 Swedish travelers to Africa and Latin America [18]. In children and HIV-positive persons, differences in clinical manifestations have been observed among different Cryptosporidium species, especially between C. hominis and C. parvum, with the former more virulent than the latter [19]–[21]. In addition, infections with C. parvum were associated with chronic diarrhea and vomiting in HIV-positive persons more frequently than was C. hominis [21].
Sequence characterization of the 60-kDa glycoprotein (gp60) gene has been commonly used in subtyping C. hominis and C. parvum [12]. Differences have been observed in host specificity among C. parvum subtype families and in virulence among C. hominis subtype families. Thus, C. parvum subtype family IIa is commonly found in calves, IId is mostly found in lambs and goat kids, whereas IIc is mostly found in humans. Within C. hominis, studies in Peru have shown that subtype family Ib was more virulent than other subtype families in children, whereas subtype family Id was more virulent than subtype families Ia and Ie in HIV-positive patients [20], [21].
Cryptosporidiosis is endemic in Ethiopia; occurrence rates of 7.6% to 43.6% were reported in HIV/AIDS patients [22]–[26]. Some potential risk factors for cryptosporidiosis occurrence included contamination of drinking water, contact with calves, living in overcrowded households with many family members, and poor personal hygiene. Thus far, only one study has genetically characterized Cryptosporidium spp. from Ethiopia. In the study, 39 of the 41 specimens genotyped had C. parvum, one had C. hominis, and one had both species. All 12 C. parvum specimens subtyped by sequence analysis of the gp60 gene belonged to the subtype family IIa [22].
In the present study, we examined the occurrence of Cryptosporidium infection in HIV/AIDS patients in Ethiopia and characterized Cryptosporidium spp. at the species, subtype family, and subtype levels. We also examined the association between clinical manifestations and infections with specific Cryptosporidium species and subtype families. Data generated from the study have clearly shown a dominance of C. parvum in the study population, importance of zoonotic transmission in the epidemiology of cryptosporidiosis in Ethiopia, and differences in clinical manifestations among Cryptosporidium species and subtypes.
The research protocol was approved by the Ethical Clearance Committee of the Addis Ababa University. All study participants had given written informed consent before enrollment into the study. When the study participant was a child, written consent was obtained from his or her parent or guardian. Researchers at the Centers for Disease Control and Prevention (CDC) had no contact with patients and no access to personal identifiers. Laboratory work on the study specimens was covered under CDC IRB protocol No. 990115: “Use of residual human specimens for the determination of frequency of genotypes or sub-types of pathogenic parasites”.
This study was cross-sectional in nature. It was conducted between September 2009 and December 2011 in Addis Ababa, Ethiopia. A total of 520 HIV/AIDS patients were recruited from in-patients (hospitalized) and outpatients attending the Tikur Anbessa Hospital, Addis Ababa University, and patients referred to the study by attending physicians. The criteria for inclusion in the study were documented HIV infection, the ability to provide informed consent by the patient or the guardian, and willingness to provide one stool specimen. A structured questionnaire was used to collect CD4+ cell counts, demographic data, clinical symptoms, HAART history, antibiotics usage, and animal exposure history. Each participant was asked to provide a single fresh stool specimen.
Stool specimens were stored in 2.5% potassium dichromate at 4°C and shipped to the CDC laboratory in Atlanta for screening and molecular characterization of Cryptosporidium spp. by PCR. After washing the stool specimens twice with distilled water, genomic DNA was extracted from 0.5 ml of fecal materials using a FastDNA SPIN Kit for Soil (MP Biomedicals, Solon. OH) and eluted in 100 µl of reagent-grade water following the manufacturer-recommended procedures. DNA was stored at −80°C until analyzed by PCR.
Cryptosporidium oocysts present in the specimens were detected by nested PCR amplification of an approximate 830 bp fragment of the small subunit (SSU) rRNA gene as described previously [27]. Cryptosporidium species were determined by restriction fragment length polymorphism (RFLP) analysis of the secondary PCR products using endonucleases SspI and VspI [27]. PCR products and restriction fragments were subjected to electrophoresis in 1.5% and 2% agarose gels, respectively, and visualized after staining with GelRed (Biotium Inc., Hayward, CA). All secondary PCR products from species other than C. parvum were sequenced to confirm the identification. Specimens that contained C. parvum or C. hominis were further subtyped by DNA sequencing of the nested PCR product of the gp60 gene [28]. Each specimen was analyzed at least twice by PCR at each locus using 2 µl of the DNA extraction per PCR. As a positive control, C. baileyi DNA was used in SSU rRNA PCR and C. hominis DNA was used in gp60 PCR. A negative control using DNase-free water was also included in each PCR run.
PCR products were sequenced using the forward and reverse primers of the secondary PCR. An intermediary sequencing primer gp60-R3 [5′-GAG ATA TAT CTT GTT GCG-3′] was also used in the sequencing of gp60 PCR products. DNA sequencing was done using the ABI BigDye Terminator v. 3.1 Cycle Sequencing Kit (Applied Biosystems, Foster City, CA, USA) and an ABI3130 Genetic Analyzer (Applied Biosystems). Sequence accuracy was confirmed by sequencing of two PCR products from each positive specimen. Nucleotide sequences obtained were aligned with reference sequences using the ClustalX 1.81 package (http://www.clustal.org/) to identify Cryptosporidium species and C. parvum subtypes. Subtypes of C. parvum and C. hominis were named based on the established nomenclature system [29]. Unique sequences generated in this study were deposited in GenBank under accession numbers AB830575 to AB830590.
Data from the study were analyzed using the SPSS 20.0 for Windows software (IBM Corp, Armonk, NY) at three levels of parasite categorization: presence of Cryptosporidium, species of Cryptosporidium, and subtype families of C. parvum and C. hominis. Data from persons infected with low-frequency species were pooled based on their genetic similarities. Univariate and multivariate logistic regression modeling were used to analyze the association between Cryptosporidium infection and clinical symptoms or risk-factors while adjusting for potential confounders when the sample size was sufficient. For comparisons at the species or subtype family level, separate models and different subsets of the main dataset were run to examine the effect of each species or subtype family, with Cryptosporidium-negative as the referent. The Hosmer-Lemeshow test was used to assess the goodness-of-fit of each multivariate logistic regression model. The strength of the associations was estimated by odds ratios (OR) and 95% confidence intervals (CI). The association was considered statistically significant when the P value obtained was smaller than 0.05.
Among the 520 HIV/AIDS patients who participated in this study, 276 (53.1%) were males and 244 (46.9%) were females. The median age of the study participants was 41 years (range: 7 months to 86 years), and the mean CD4+ cell count was 278 cells/µL. Almost one-third (32.9%) of the study patients were bedridden and hospitalized. Of the 520 stool specimens examined in this study, 140 (26.9%) were positive for Cryptosporidium by the SSU rRNA-based PCR technique (Table 1). There were no significant differences in prevalence of Cryptosporidium infection among age groups and between the male and female gender (Table 1). No significant association between hospitalization and Cryptosporidium infection was found; percentages of Cryptosporidium infection in hospitalized and non-hospitalized patients were 35.7% and 37.4% (P = 0.83). In addition, no significant difference was observed in the occurrence of vomiting, diarrhea, age, HAART, antibiotic use, animal contact and CD4 between inpatients and outpatients. There was also no association between infections with Cryptosporidium spp. or any specific species and hospitalization rates (data not shown).
Species determination by RFLP was successful for 128/140 Cryptosporidium-positive specimens (Table 2). Cryptosporidium parvum (n = 92) and C. hominis (n = 25) were the species most frequently detected, followed by C. felis (n = 5), C. meleagridis (n = 3), C. canis (n = 2), and C. parvum and C. hominis co-infection (n = 1). All SSU rRNA PCR products from non-C. parvum specimens were sequenced, confirming the RFLP results. In addition, 12 specimens demonstrated a RFLP pattern that was similar to C. parvum, but with a slightly smaller upper SspI band. DNA sequences from two of the specimens were identical to a reference sequence of C. xiaoi (GenBank accession no. JQ413443) and those from 10 specimens were identical to a reference sequence (GenBank accession no. HM485434) of C. viatorum, a recently described species in humans [17].
Subtype family data were obtained from 82 (89.1%) of the 92 participants with C. parvum and showed the presence of subtype families IIa, IIb, IIc, IId, IIe, and a new subtype family genetically related to If in 71, 1, 2, 5, 1, and 2 persons, respectively. In contrast, 19 (76%) of the 25 participants with C. hominis were successfully subtyped and showed the presence of subtype families Ib, Id, and Ie in 1, 13, and 5 persons, respectively. Within C. parvum, 8 subtypes were from the subtype family IIa. This was followed by subtype families IId with 4, If-like with 2 subtypes, and IIb, IIc, and IIe with 1 subtype each. Thus, participants in the study were infected with 14 subtypes of C. parvum. The most frequently detected subtype was IIaA15G2R1, which was seen in 60 patients. Five subtypes were identified in C. hominis, including IdA20, IdA26, IdA24, IeA11G3T3, and IbA10G2 in 10, 2, 1, 5, and 1 patient, respectively (Table 2).
Cryptosporidium infection was significantly associated with diarrhea in univariate analysis (P<0.001), especially in patients with C. parvum and C. hominis (P<0.001 and P = 0.012 respectively). Vomiting was also more often seen in Cryptosporidium-positive than Cryptosporidium-negative patients (63.6% versus 51.8%; P = 0.018). Patients with C. hominis or rare species (C. meleagridis/C. felis/C. canis and C. xiaoi) were more likely to have vomiting compared with Cryptosporidium-negative patients (P = 0.057 and P = 0.027, respectively; Table 3).
Diarrhea was significantly associated with infections with C. parvum subtype family IIa compared with Cryptosporidium-negative patients (78.9%; P = 0.003). However, there were no significant differences in diarrhea occurrence between patients infected with C. hominis subtype families and Cryptosporidium-negative patients. Patients infected with different C. parvum and C. hominis subtype families also did not show significant differences in the occurrence of vomiting compared to Cryptosporidium-negative patients (P>0.05; Table 3).
Multivariable modeling was attempted for diarrhea and vomiting occurrence in order to adjust for age, gender, type of patients and clinical parameters (HAART and CD4+); however, it did not reveal major differences compared with the crude OR (Table 3).
After controlling for other potential risk factors, history of any contact with animals was associated with overall Cryptosporidium infections (OR = 1.6; P = 0.04), and with C. parvum (OR = 2.5; P = 0.002) and its subtype family IIa (OR = 2.1; P = 0.02) in particular. This association was mostly due to contact with calves (OR = 1.6 and P = 0.02 for overall Cryptosporidium infection; OR = 1.8 and P = 0.01 for C. parvum; and OR = 2.0 and P = 0.01 for C. parvum subtype family IIa; Table 4). However, there was no significant association between Cryptosporidium infection and age, gender, HAART history, CD4+ cell counts, antibiotics use, or type of patients (P = 0.56, 0.13, 0.59, 0.90, 0.84, and 0.83, respectively). No significant association was found between Cryptosporidium infection and CD4+ cell counts or HAART history at the species and subtype levels (Tables S1 and S2).
The present findings showed that (1) Ethiopian HIV/AIDS patients were infected with a diverse population of Cryptosporidium species, including the unusual species C. viatorum and C. xiaoi; (2) C. parvum was the most frequently detected species; and (3) Cryptosporidium species or subtype families were associated with different clinical manifestations. The dominance of C. parvum in this study is in agreement with the previous observation in a small study in Ethiopia [22]. It is, however, in sharp contrast with studies of human cryptosporidiosis in other developing countries where C. hominis dominates [12], [30], [31]. In Europe, the two species are almost evenly distributed, with C. parvum being more prevalent in some reports [32] and C. hominis in others [33]. Data from this study also indicate that the recently established species C. viatorum is more widely distributed than believed, and humans can be infected occasionally with the sheep and goat parasite C. xiaoi. As expected, the prevalence of cryptosporidiosis in HIV+ patients (26.9%) in this study is substantially higher than in the largely healthy persons (8.7%) in a previous study conducted in the same area [22].
Differences in geographical distribution of C. parvum and C. hominis are generally considered a reflection of differences in infection sources and transmission routes [12]. C. hominis is transmitted almost exclusively among humans, whereas C. parvum, especially its IIa subtype family, is more likely transmitted zoonotically. In industrialized nations, C. parvum infections have often been linked to contact with farm animals, and C. hominis infections to contact with children with diarrhea [34]. Both species have been associated with drinking-water outbreaks [33]. The predominance of C. hominis in most developing countries suggests that anthroponotic transmission is more important than zoonotic transmission in cryptosporidiosis epidemiology in developing countries in general [12]. In contrast, the dominance of C. parvum IIa subtype family in Ethiopia in HIV/AIDS patients suggests that unlike in other developing countries, Cryptosporidium infection in Ethiopia is mostly transmitted zoonotically. Previously, it was shown that that small numbers of C. parvum infections seen in humans in developing countries were mostly caused by the anthroponotic subtype family IIc [12].
Results of the risk factor analysis support the role of zoonotic transmission in cryptosporidiosis epidemiology in HIV/AIDS patients in Ethiopia. Even though the present study was performed in an urban area, more than 50% of patients had contact with animals, as households in the study area usually have farm animals living inside the residence. Thus, in this study, animal contact, especially with calves, was a significant risk factor for Cryptosporidium, especially C. parvum and its IIa subtype family. The distribution of C. parvum subtypes in this study reinforces the likely occurrence of zoonotic transmission, as the majority of C. parvum infections were caused by IIa subtypes (71/82 specimens subtyped), especially its IIaA15G2R1 subtype (60/82 specimens subtyped), which is a dominant C. parvum subtype in calves around the world [12]. In contrast, the anthroponotic C. parvum subtype family IIc was seen in only 2/82 of C. parvum infections.
In addition to IIa, C. parvum subtype family IId was also identified in five Ethiopian HIV/AIDS patients. Subtype family IId of C. parvum is generally considered a sheep and goat parasite [35], but has been found at high frequency in calves in China, Egypt, and Sweden [36]–[38]. In this study, only one patient infected with IId had contact with sheep in this study. Genotyping and subtyping studies of domestic animals from the study area and additional case-control-studies are needed to support the conclusion on the importance of zoonotic transmission in cryptosporidiosis epidemiology in Ethiopia.
In agreement with previous observations elsewhere [19]–[21], data from the study suggest that different Cryptosporidium species and subtypes are linked to different manifestations of cryptosporidiosis. As expected, cryptosporidiosis in our study was significantly associated with the occurrence of diarrhea. This association, however, was largely attributable to C. parvum and C. hominis; other species, including the newly described C. viatorum, were less pathogenic than these two species. Likewise, the significant association between Cryptosporidium infection and the occurrence of vomiting was also largely attributable to C. hominis and some less frequent species (C. meleagridis, C. canis, C. felis and C. xiaoi); C. parvum and C. viatorum were largely not associated with vomiting. Within C. parvum, the IIa subtype family also appeared to be more associated with the occurrence of diarrhea than other C. parvum subtype families. We did not observe any effect of CD4+ cell counts and HAART on the occurrence of cryptosporidiosis in HIV/AIDS patients in this study. This was probably largely the result of severe overall immunodeficiency in the study population, as reflected by the low mean CD4+ cell counts (278 cells/µL) and very high hospitalization rate (32.9%) and occurrence of cryptosporidiosis (26.9%).
In conclusion, Ethiopian HIV/AIDS patients with low CD4+ cell counts had an extremely high occurrence of Cryptosporidium infection, even when they were on HAART. Although the majority of cryptosporidiosis cases were caused by C. parvum, there was a high diversity of Cryptosporidium species, with a significant number of cases caused by the newly recognized C. viatorum. These Cryptosporidium spp. and C. parvum subtypes were linked to different clinical manifestations. Therefore, improved hygiene and avoidance of calf contact among this population should be advocated to reduce the occurrence of Cryptosporidium infections, especially those caused by C. parvum IIa subtypes of calves.
|
10.1371/journal.ppat.1004115 | Phosphorylation of KasB Regulates Virulence and Acid-Fastness in Mycobacterium tuberculosis | Mycobacterium tuberculosis bacilli display two signature features: acid-fast staining and the capacity to induce long-term latent infections in humans. However, the mechanisms governing these two important processes remain largely unknown. Ser/Thr phosphorylation has recently emerged as an important regulatory mechanism allowing mycobacteria to adapt their cell wall structure/composition in response to their environment. Herein, we evaluated whether phosphorylation of KasB, a crucial mycolic acid biosynthetic enzyme, could modulate acid-fast staining and virulence. Tandem mass spectrometry and site-directed mutagenesis revealed that phosphorylation of KasB occurred at Thr334 and Thr336 both in vitro and in mycobacteria. Isogenic strains of M. tuberculosis with either a deletion of the kasB gene or a kasB_T334D/T336D allele, mimicking constitutive phosphorylation of KasB, were constructed by specialized linkage transduction. Biochemical and structural analyses comparing these mutants to the parental strain revealed that both mutant strains had mycolic acids that were shortened by 4–6 carbon atoms and lacked trans-cyclopropanation. Together, these results suggested that in M. tuberculosis, phosphorylation profoundly decreases the condensing activity of KasB. Structural/modeling analyses reveal that Thr334 and Thr336 are located in the vicinity of the catalytic triad, which indicates that phosphorylation of these amino acids would result in loss of enzyme activity. Importantly, the kasB_T334D/T336D phosphomimetic and deletion alleles, in contrast to the kasB_T334A/T336A phosphoablative allele, completely lost acid-fast staining. Moreover, assessing the virulence of these strains indicated that the KasB phosphomimetic mutant was attenuated in both immunodeficient and immunocompetent mice following aerosol infection. This attenuation was characterized by the absence of lung pathology. Overall, these results highlight for the first time the role of Ser/Thr kinase-dependent KasB phosphorylation in regulating the later stages of mycolic acid elongation, with important consequences in terms of acid-fast staining and pathogenicity.
| Acid-fast staining has been used since 1882 as the hallmark diagnostic test for detecting Mycobacterium tuberculosis, the causative agent of tuberculosis. It has been attributed to the presence of a waxy cell envelope, and primarily to its key components, mycolic acids. Here, we report a new mechanism of regulation in which phosphorylation of KasB, involved in the completion of full-length mycolic acids, leads to shortened mycolic acids and loss of acid-fast staining. Moreover, a M. tuberculosis mutant strain mimicking constitutive phosphorylation of KasB is severely attenuated for growth in both immunocompetent and immunosuppressed mice and fails to cause mortality and pathophysiological symptoms. These results emphasize the critical role of kinase-dependent phosphorylation in the pathogenesis of M. tuberculosis by controlling the mycolic acid chain length. Our study demonstrates the importance of a regulatory mechanism governing acid-fastness and virulence of M. tuberculosis.
| Mycobacterium tuberculosis (Mtb) is an extraordinarily versatile pathogen that can exist in two distinct states in the host, leading to asymptomatic latent infection in which bacilli are present in a non-replicating dormant form, or to active tuberculosis (TB), characterized by actively replicating organisms. Establishment of these different (patho)physiological states requires mechanisms to sense a wide range of environmental signals and to coordinately regulate multiple metabolic and cellular processes. Many of the stimuli encountered by Mtb are transduced via transmembrane sensor kinases, allowing the pathogen to adapt to survive in hostile environments. In addition to the 12 classical two-component systems [1], Mtb contains 11 eukaryotic-like Ser/Thr protein kinases (STPK) [2], [3], suggesting that these two phospho-based signaling systems are of comparable importance in this microorganism. Knowledge of the substrates of each of the Mtb STPK is essential for understanding their function. Several kinase-substrates pairs have been identified and characterized during the last decade. In addition, a recent comprehensive understanding of in vivo phosphorylation event in Mtb was gained using a mass spectrometry-based approach to identify phosphorylation sites in Mtb proteins [4]. This provided insights into the range of functions regulated by Ser/Thr phosphorylation, underpinning the involvement of many STPK in regulating metabolic processes, transport of metabolites, cell division or virulence [5], [6], [7].
Recent studies focusing on mycolic acid biosynthesis regulation have shown that most essential enzymes forming the central core of type II fatty acid synthase (FASII) are phosphorylated by STPK [6] and that, at least in vitro, post-translational phosphorylation inhibits the activity of these enzymes. These include the β-ketoacyl ACP synthase KasA, the β-ketoacyl-ACP reductase MabA, the hydroxyacyl-ACP dehydratases HadAB and HadBC and the enoyl-ACP reductase InhA [8], [9], [10], [11], [12]. These studies culminated with the demonstration that InhA, also known as the primary target of the first-line anti-TB drug isoniazid (INH), is controlled via phosphorylation by STPK on Thr266 both in vitro and in vivo [10], [11]. The physiological relevance of Thr266 phosphorylation was demonstrated using inhA phosphoablative (T266A) or phosphomimetic (T266D/E) mutant strains. Not only was the enoylreductase activity severely impaired in the mimetic mutants in vitro, but introduction of inhA_T266D/E failed to complement an inhA-thermosensitive M. smegmatis strain, in agreement with mycolic acid inhibition, in a manner similar to that by isoniazid, and growth inhibition [10]. Altogether these results strongly suggest that Mtb may control in a very subtle manner its FASII system by regulating each step of the elongation cycle. Since phosphorylation of HadAB and HadBC enzymes was found to be increased during stationary growth phase, it was proposed that mycobacteria shut down meromycolic acid chain production under non-replicating conditions, a view which is supported by the fact that the mycolic acid biosynthesis is growth phase-dependent and is not functional during the stationary phase [13]. However, whether phosphorylation of FASII components may also directly participate in Mtb virulence, through the control of the meromycolic acid chain length has not been reported yet.
It was previously demonstrated that targeted deletion of kasB, one of two Mtb genes encoding distinct β-ketoacyl-ACP synthases, results in loss of acid-fast staining and synthesis of shorter mycolates [14]. Perhaps the most striking effect of kasB deletion was the ability of the mutant strain to persist in infected immunocompetent mice without causing disease or mortality. This indicated that KasB participates in the latest elongation steps by adding the last few carbon atoms to the growing acyl-ACP chains and plays a critical role in controlling Mtb physiopathology. From these data, it could be inferred that KasB activity may be tightly regulated in order to control the mycolic acid chain length during the infection process. We have previously reported that, like KasA, KasB was a substrate of STPK [8], suggesting that phosphorylation may represent a mechanism governing KasB activity and, as a consequence, mycolic acid chain length and Mtb virulence.
This prompted us to decipher an original mechanism linking post-translational phosphorylation of KasB with Mtb virulence in a mouse infection model. Herein, we demonstrate that phosphorylation of KasB on Thr334 and Thr336 dramatically alters the mycolic acid chain length and acid-fast staining. Importantly, a KasB phosphomimetic mutant of Mtb was found to be extremely attenuated in mice infection models. These results provide, for the first time, insights into the contribution and importance of FASII phosphorylation in vivo in the control of i) the clinically important feature of acid-fast staining in Mtb and ii) the physiopathology of TB.
Previous work demonstrated that KasB is a substrate for several Ser/Thr protein kinases with PknF being one of the most efficient kinase [8]. However, the role and contribution of KasB phosphorylation with respect to the Mtb physiology and pathogenicity remains unknown, mainly because of the lack of information regarding the identity of the phosphoacceptors. Therefore, recombinant wild-type KasB (unphosphorylated form) was expressed and purified from E. coli harboring pETPhos_kasB and used in an in vitro kinase assay in the presence of PknF and [γ-33P]ATP. The reaction mixture was then separated by SDS-PAGE and analyzed by autoradiography, revealing a specific band corresponding to the phosphorylated form of KasB (Fig. 1A), as reported previously [8]. To identify the number and nature of the phosphosites, the protein was phosphorylated in vitro with PknF and cold ATP and subjected to mass spectrometry analysis after tryptic and chymotryptic digestions, a method successfully used to elucidate the phosphoacceptors in a sequence-specific fashion for several other Mtb STPK substrates [7], [9], [10], [15], [16], [17]. Spectral identification and phosphorylation determination were achieved with the paragon algorithm from the 2.0 database-searching software (Applied Biosystems) using the phosphorylation emphasis criterion against a homemade database that included the sequences of KasB and derivatives. The sequence coverage was 92% with the non-covered sequence free of serine or threonine residues. Phosphorylation was detected only on a single peptide 315AIQLAGLAPGDIDHVNAHATGTQVGDLAEGR345. The MS/MS spectra unambiguously confirmed the presence of two phosphate groups on this peptide (data not shown). The 315–345 peptide possesses only two Thr residues representing the potential phosphoacceptors, Thr334 and Thr336. Definitive identification of the phosphosites was achieved by site-directed mutagenesis, by replacing Thr with Ala, preventing subsequent phosphorylation. The single (T334A, T336A) and double (T334A/T336A) mutants were expressed via the pETPhos, purified as His-tagged proteins and individually subjected to the kinase assay. As shown in Fig. 1A, decrease of the phosphorylation signal was only partially limited in the single mutants with respect to the wild-type protein. However, phosphorylation was abrogated in the double mutant as evidenced by the absence of a specific radioactive band, confirming that, in vitro, phosphorylation of KasB occurs at Thr334 and Thr336. Similar results were obtained when KasB_T334A/T336A was incubated in the presence of either PknA, PknB, PknD, PknH or PknL, indicating that these two residues are the phosphoacceptor for all six kinases (Figure S1 in Text S1).
In vivo phosphorylation of KasB was next investigated in recombinant M. bovis BCG by Western blotting using anti-phosphothreonine antibodies. Specificity of the antibodies was first assessed against KasB purified from either E. coli (pETPhos_kasB) or E. coli co-expressing PknF (pETDuet_kasB), thanks to a recently developed duet strategy [18]. Phosphorylated KasB derived from pETDuet_kasB was specifically revealed with anti-phosphothreonine antibodies, while the unphosphorylated isoform from pETPhos_kasB failed to react (Fig. 1B), confirming the specificity of Thr phosphorylation. To confirm the phosphorylation status of KasB in mycobacteria, and in order to exclude eventual interference/association between the endogenous and the recombinant His-tagged KasB monomers (KasB being a dimer, see below), a ΔkasB BCG mutant was transformed with pVV1::kasB_WT or pVV16::kasB_T334A/T336A in which the wild-type or phosphoablative kasB genes were placed under the control of the hsp60 promoter. BCG carrying either pVV16::kasB_WT or pVV16::kasB_T334A/T336A was harvested from exponential or stationary cultures prior to protein purification by affinity chromatography on Ni2+-containing agarose beads and Western blotting using phosphothreonine antibodies. Specific phosphorylation was only detected for the wild-type but not the T334A/T336A protein (Fig. 1B). The lack of reactivity with the double Ala mutant excludes the possibility of additional phosphorylation sites. Interestingly, phosphorylation of KasB was more pronounced during stationary phase than exponentially-growing bacteria, indicating that KasB phosphorylation is growth phase-dependent. Taken collectively, these data suggest that phosphorylation occurs at Thr334 and Thr336, in vitro and in vivo.
The crystal structure of KasB (438 residues, MW 46.4 kDa) in its apo-form has been determined to 2.4 Å resolution [19]. It consists of a dimer with each protomer adopting the typical thiolase fold decorated with specific structural features in the form of a cap (Fig. 1C). The structures of wild-type KasA (416 residues, MW 43.3 kDa), the other fatty acyl elongation β-ketoacyl synthase, and of the acyl enzyme mimic C171Q, both unliganded and with bound thiolactomycin (TLM), were also resolved to high resolution [20]. In line with their high sequence homology, KasA and KasB are structurally similar and superposition of the wild-type apo-dimers (PDB codes 2WGD and 2GP6, respectively) led to a root mean square deviation value of 1.1 Å for 814 aligned Cα atoms sharing 66% sequence identity. The active site, containing the Cys-His-His catalytic triad, is located in the core domain. As shown for KasA [20], TLM binds close to the active site in the malonyl-binding pocket and the hydrophobic acyl-binding channel of the substrate is connected to the malonyl-binding pocket and also directly accessible from the surface of the protein (Fig. 1C, left panel). Thr334 and Thr336 together with Ile235, Phe258, Val299, Ala300, Val338, Pro301, and Phe426 (most of these residues being strictly conserved in KasA) line a molecular tunnel that leads to the catalytic cysteine [19] (Fig. 1C, right panel). Thr334 is located close to the tunnel aperture whereas Thr336 directly faces the catalytic triad. Their side chains are aligned when looking from outside the tunnel with their OG atoms at a distance of 3.6 Å. Replacement of Thr334 and Thr336 by alanine would result in a slight broadening of one side of the tunnel. In contrast, Thr→Asp replacements that mimic constitutive phosphorylation [9], [10], [17], [21] are very likely to induce a profound perturbation in terms of steric hindrance and electrostatic potential. In addition, the carboxyl group of the Asp at position 334 could establish hydrogen bonds with the NE2 atoms of the two catalytic histidines. Thus, the perturbation brought by the Thr→Asp substitution might lead to severe impairment of the enzyme activity but not of the three-dimensional structure. Moreover, this was confirmed by analysis of the trypsinolysis kinetics of wild-type and mutated KasB proteins (Figure S2 in Text S1), and was consistent with the structural analysis indicating that introduction of Asp or Ala at position 334 and 336 does not seem to modify the folding of the protein as the proteolysis profiles of the different KasB derivatives were identical to the wild-type protein.
To study the effect of the two KasB phosphorylation sites in Mtb, the Thr334 and Thr336 amino acids were replaced either with phosphomimetic (aspartate) or phosphoablative (alanine) amino acid. Previous studies have shown that acidic residues such as aspartate qualitatively recapitulate the effect of phosphorylation with regard to functional activity [15], [17], [21], [22]. Specialized linkage transduction [23] was used to transfer single point mutant alleles, respectively kasB T334D/T336D and kasB T334A/T336A in Mtb CDC1551 (Table S1 in Text S1, Fig. 2A). These strains contained a sacB and hyg cassettes inserted between kasB and accD6. The introduction of the sacB and hyg cassettes was confirmed by Southern blot (Fig. 2B) and presence of the point mutation(s) was verified by sequencing kasB. An additional kasB deletion strain in Mtb CDC1551 was constructed using the same plasmid (pYUB1471) as the one used for the allelic exchanges.
Quantitative reverse transcription real-time PCR analyses were conducted to measure the kasB expression level in the different mutant strains. Expression levels were standardized using the sigA internal standard (Fig. 2C). As expected, no specific kasB mRNA was produced in the ΔkasB mutant. In contrast, kasB expression levels were found to be comparable in the parental strain, the double Ala and the double Asp mutants. Similar results were obtained when using either 16S rRNA or rrnAP1 as alternative internal control (data not shown). These results were confirmed by Western blot analysis using antibodies raised against KasA, which cross-react with KasB, and revealing comparable levels of KasB expression in the parental strain, the double Ala and the double Asp mutants (Fig. 2D). Moreover, since kasB belong to the fasII operon and is located upstream of accD6, we also checked whether the introduction of the T/A or T/D replacements may affect expression of the downstream accD6 gene. Immunoblotting using rabbit anti-AccD6 antibodies clearly revealed the presence of similar AccD6 levels in all strains (Fig. 2E). From these data it can be inferred that mutations in kasB did not exert a polar effect on AccD6 expression.
Together, these results indicate that introduction of the phosphoablative or phosphomimetic mutations does not affect kasB gene expression in Mtb, thus allowing analysis and comparison of the phenotypes associated to these mutations.
The first phenotype tested in the KasB mutants was acid-fastness using both the carbolfuchsin and auramine stainings (Fig. 3A). Like the parental strain, the Ala mutant remained acid-fast positive. In sharp contrast, the phosphomimetic T334D/T336D mutant strain behaved like the kasB deletion mutant, losing its ability to retain the primary stain following washing with the acid-alcohol decolorizer. The phosphomimetic and deletion strains regained acid-fast staining when a wild-type copy of kasB was introduced into these strains on a multicopy plasmid (Fig. 3B). This indicates that phosphorylation on Thr334/Thr336 abrogated the acid-fast property, presumably by negatively regulating the activity of KasB. The results support, for the first time, a control of acid-fastness by STPK-dependent signaling in Mtb.
The lack of acid-fastness prompted us to investigate the mycolic acid content in the various strains. Since kasB deletion results in a strain producing shorter mycolic acids [14], we compared the mycolic acid profiles of the double Ala and double Asp mutants to the parental and ΔkasB strains. Mycolic acids were extracted after saponification from the strains grown to stationary phase and analyzed by thin-layer chromatography (TLC) (Fig. 4A). When separated, the mycolates of the parental and phosphoablative strains presented similar mobility shifts, whereas those from the phosphomimetic and the ΔkasB strains were found to be slightly retarded. This lower mobility shift has been earlier reported to occur in M. marinum and Mtb kasB mutants and correlated to reduced carbon chain lengths [14], [24]. The normal mobility shift was restored in the corresponding complemented strains (Fig. 4A).
Individual mycolic acid species from all four strains were then purified from preparative TLC plates, re-extracted and analyzed by proton nuclear magnetic resonance (1H NMR) spectroscopy (Fig. 4B and Figure S3 in Text S1) and matrix-assisted laser desorption/ionization-time of flight (MALDI-TOF) mass spectrometry (Fig. 4C). 1H NMR analyses revealed that mycolates from the ΔkasB and KasB_T334D/T336D strains significantly differed from those of the parental strain. In particular, the relative quantification of signals at −0.33 and 0.55 ppm and signal at 0.45 ppm established that the ratio of cis/trans-cyclopropanation changed in oxygenated mycolic acids, as reported for the ΔkasB mutant [14]. For methoxy-mycolic acids, it decreased from 20% of trans-cyclopropanation in parental strain to 0% in the ΔkasB and KasB_T334D/T336D strains (Figure S3B in Text S1) and for keto-mycolic acids from 50% to 0% (Figure S3C in Text S1 and Fig. 4B). In contrast, mycolates from the phosphoablative mutant exhibited spectra similar to those of the parental strain (data not shown).
Next, the mycolic acid size distribution was assessed by MALDI-TOF MS, and similarly to the 1H NMR analysis, the double Ala mutant was found to produce mycolates identical to those from the parental strain. In contrast, the double Asp mutant, like ΔkasB, synthesized shorter α-, methoxy, and keto-mycolic acids (up to C80:2, up to C84:1, and up to C85:1) than the parental strain (up to C84:2, up to C89:1, and up to C89:1), and accumulated a trans-unsaturated precursor of both methoxy- and keto-mycolic acids, thus confirming the slightly reduced TLC mobility shift. The average size reduction of oxygenated mycolates was higher than that of α-mycolates because it resulted from both size reduction of aliphatic chain and loss of the -CH3 group of trans-cyclopropane groups (Fig. 4C).
NMR analysis indicates the lack of trans-cyclopropane rings in oxygenated mycolic acids in the phosphomimetic mutant, which accumulates the corresponding trans-double bonds precursors, as found in the ΔkasB strain and in agreement with previous work [14]. Trans-double bonds have been reported to be the substrates of the trans-cyclopropane synthase CmaA2 [25]. We reasoned that lack of trans-cyclopropanation in the phosphomimetic strain mutant may be due to the shortened oxygenated meromycolates that are poor substrates for CmaA2. Alternatively, alteration/lack of cmaA2 expression may occur in this particular strain. Indeed, multiple direct interactions have been reported to happen between various FASII enzymes in specialized multifunctional complexes [26], [27]. Consequently, phosphorylation of KasB may hinder/prevent association of KasB with other partners, including CmaA2. cmaA2 expression levels were therefore measured and found to be comparable in all four strains (Fig. 2C), indicating that neither phosphomimetic mutations within KasB nor deletion of kasB altered cmaA2 expression. Similar results were obtained using sigA, 16S rRNA or rrnAP1 as internal standards (data not shown). This implies that absence of trans-cyclopropane rings is likely due to the shortened mycolates that are poor substrate for CmaA2, rather than to a defect in cmaA2 expression.
Given the effect of KasB phosphorylation on the meromycolate chain length, we inquired whether shorter mycolic acids would alter drug susceptibility. The KasB strains were tested for growth in the presence of INH, ethionamide (ETH), and rifampicin (RIF) (Table 1). The KasB_T334D/T336D mutant was found to be more susceptible to INH, ETH, and RIF, suggesting that this strain exhibits a cell wall permeability defect.
To confirm the permeability defect of this mutant in logarithmic phase, we compared how the phosphomimetic, phosphoablative, deletion and parental strains incorporated two different types of dye: the cyanine dye 3,3′-diethyloxacarbocyanine iodide DiOC2, which penetrates all cells, and the SYTOX Red dead cell stain, which is excluded from cells with intact membranes [28], [29]. The highest amount of Sytox Red was found in the KasB phosphomimetic strain (Figure S4 in Text S1), suggesting that this strain had the most permeable membrane. This set of data support the view of a permeability defect in the KasB phosphomimetic strain.
Growth curves of Mtb bearing either the kasB_T334A/T336A or the kasB_T334D/T336D allele were also similar, indicating that the phosphomimetic or phosphoablative mutations did not impact in vitro growth (data not shown). To address the physiological relevance of KasB phosphorylation with respect to Mtb virulence and physiopathology, low-dose aerosol infection of mice were performed with the parental, double Ala, double Asp and ΔkasB isogenic Mtb strains. Previous work showed that whereas CDC1551 ΔkasB was severely attenuated and failed to cause active infection in immunocompetent mice, it caused mortality in immunodeficient SCID mice [14]. Therefore, in vivo growth of the different KasB mutants was tested in both SCID mice and immunocompetent C57BL/6 mice. Although all the strains grew in the lung of SCID mice, the double Asp was found to be much more attenuated than the other KasB variants. Indeed, the phosphomimetic mutant strain was strikingly less virulent relative to the other strains following aerogenic challenge, as assessed by the mean survival time of SCID mice infected with these strains (57.5, 75, 76 and 147 days post-infection for parental, KasB_T334A/T336A, ΔkasB and KasB_T334D/T336D, respectively). Mice infected with the double Asp mutant survived on average 90 days longer than the mice infected with the parental strain, while the mice infected with the double Ala or ΔkasB strain survived only 20 days longer than the mice infected with the control strain (Fig. 5A). In agreement with the survival data, quantification of tissue bacterial burden revealed a severe growth defect (3–4 Logs) of the KasB phosphomimetic strain (Fig. 5B). Manifestation of this hypovirulent phenotype is apparent as early as one week post-infection with the lung bacterial burden of mice infected with Mtb KasB_T334D/T336D about 30-fold lower than the parental strain-infected mice. The bacterial burden of the Mtb KasB_T334A/T336A strain was comparable to that of the parental strain. Consistent with previously published data [14], the ΔkasB strain exhibited a growth defect, albeit less pronounced than the Mtb KasB_T334D/T336D strain (Fig. 5B). Extensive granulomatous inflammation was visible in the lungs of SCID mice infected with the parental strain, and to a lesser extent in mice infected with the KasB_T334A/T336A and the ΔkasB strains but not in those infected with the KasB_T334D/T336D (Fig. 5C). Monitoring of colony-forming units (CFU) in spleen and liver at three and eight weeks post-infection revealed that the KasB_T334D/T336D mutant was also severely attenuated for growth in these organs (Fig. 5D). The restricted bacterial loads in both organs increase by three orders of magnitude after eight weeks of infection. In contrast, replication of the KasB_T334A/T336A strain was comparable to that of the parental strain in both organs, although a 1-log growth defect was observed in the liver eight weeks after infection, whereas replication of the ΔkasB mutant was also severely impeded in these two organs.
Next, immunocompetent C57BL/6 mice were infected via aerosol and assessed for both survival and bacterial replication. Monitoring of CFU in the lungs at different time points after infection indicated that both the parental and double Ala mutant grew for the first three weeks following a similar kinetic (Fig. 6A). Consistent with previous observations, the ΔkasB was severely attenuated for growth in mice [14]. In contrast to the ΔkasB, the KasB phosphomimetic mutant failed to replicate, even after the first three weeks of infection and was never found in the lungs of C57BL/6 in two independent aerosol infection experiments. Histological examination of stained lung sections from infected mice revealed multifocal, moderate infiltration after 21 days of infection with the parental strain (Fig. 6B) but not with the KasB_T334A/T336A, ΔkasB or KasB_T334D/T336D mutant strains, which may be attributed to the lower bacterial loads of these strains at this particular time point. Bacterial loads in the liver and spleen at different time points after infection indicated that the KasB phosphomimetic mutant was extremely attenuated, since CFU could not be obtained at either 3 or 8 weeks post-infection (Fig. 6C). Replication of the ΔkasB mutant was also severely impeded in these two organs.
Together, this suggests that KasB phosphorylation regulates the growth of Mtb in both immunocompromised and immunocompetent mice. In the presence of phosphorylated KasB, the tubercle bacillus fails to establish a chronic persistent infection, and exhibits a severely attenuated phenotype.
The difference in virulence between the ΔkasB and the phosphomimetic strains led us to investigate how these two strains infect and survive in macrophages, the primary cellular targets of Mtb. C57BL/6 bone marrow-derived macrophages (BMDM) were infected with the various Mtb KasB variants, lysed at 0, 1, 3 and 6 days post-infection, and the numbers of viable bacteria were counted (Fig. 7A). Both the parental and the KasB_T334A/T336A mutant grew similarly whereas the ΔkasB mutant, as expected, was attenuated for growth in macrophages, with day 6 CFU counts similar to day 0 numbers, versus an approximately one log increase for the parental strain. In contrast, the KasB phosphomimetic strain had no significant growth defect in macrophages but exhibited a marked impairment in macrophage uptake (Fig. 7A, day 0 time point). This effect was further investigated by measuring internalization of the various strains in BMDM. In three independent experiments, the uptake of the KasB phosphomimetic strain was significantly lower than the one of the parental strain by around 50% (Fig. 7B, left panel), a defect that was reversed when the phosphomimetic strain was complemented with a wild-type kasB on a multicopy plasmid (Fig. 7B, right panel). To confirm that these macrophages phenotypes were not specific to C57BL/6 BMDM, the growth of the KasB strains was also tested in the human acute monocytic leukemia cell line THP-1. In THP-1 macrophages, the phosphomimetic KasB mutant had also no growth failure and did present an uptake defect similar to the one observed in BMDM (Figure S5 in Text S1). Taken together, this set of data suggests that phosphorylation of KasB regulates the early interaction event between Mtb and macrophages.
The distinct in vivo phenotypes between the KasB phosphomimetic mutant and the ΔkasB mutant may reflect different physiological/metabolic status. To identify transcriptional differences that may be relevant to the differential phenotypes of these strains, a whole-genome transcriptional analysis was performed (Figure S6 in Text S1). Several genes were commonly up- or down-regulated two-fold or more in the KasB mutant strains. Most up-regulated genes were genes participating in cell wall, cell processes and lipid metabolism, whilst the majority of the down-regulated genes were associated with intermediary metabolism and respiration. Interestingly, a few genes were uniquely up-regulated in the KasB phosphomimetic strain such as the esterase/lipase lipF, proposed to play an important role in Mtb pathogenesis [30], and five genes involved in lipid or drug transport (MmpL4, MmpL5, MmpL10, MmpS4, and Rv1258c) [31], [32], as well as five other genes involved in lipid metabolism (Pks3, Pks16, PapA1, PapA3, and FadD21). The genes specifically down-regulated in the KasB phosphomimetic mutant encoded: NrdF1, an enzyme involved in DNA replication; two oxidoreductases (Rv3741c, Rv3742c) which are in an operon with a triacylglycerol synthase; two enzymes (Rv3084, Rv3085) part of the mymA operon which might be implicated in the modification, activation and transfer of fatty acids to the cell envelope [33]. Strikingly, most of the biosynthetic gene cluster (pks3/4, papA3, mmpL10, fadD21) required for the synthesis of polyacylated trehalose (PAT) [34] was specifically up-regulated in the KasB phosphomimetic mutant suggesting that PAT biosynthesis in this mutant might be altered.
Metabolic labeling with 14C-propionate and subsequent lipid analysis of parental, phosphoablative, phosphomimetic and deletion strains showed that the phosphomimetic KasB mutant produced more PAT than the three other strains (Fig. 8). Quantification of the TLC spots revealed a reproducible two-fold increase in PAT production in the KasB_334D/336D mutant compared to the three other strains and to its complemented strain, suggesting that phosphorylation of KasB positively affects PAT synthesis.
Recent studies have provided clear insights into the vast range of pathways regulated by STPK in Mtb. These include multiple metabolic processes, transport of cell wall components as well as cell division or virulence functions [5], [6], [7]. Several STPK, such as PknH or PknG, have been reported to play a crucial role in modulating Mtb virulence [6], [35], [36], [37], [38]. However, little is known regarding the specific substrates contributing to mycobacterial virulence and regulated by these kinases. Here, we report the critical role of STPK-dependent phosphorylation of KasB, which is directly linked to Mtb virulence. Through the design of a KasB phosphomimetic Mtb mutant, we demonstrate that, in vivo, the replacement of the two Thr by Asp residues was characterized by highly pronounced phenotypes corresponding to i) loss of acid-fastness, ii) production of shorter mycolic acids with defects in trans-cyclopropanation, iii) defect in macrophage invasion, iv) incapacity to grow and establish a persistent infection in both immune-compromised and immune-deficient mice, and v) absence of pathology in infected animals (Fig. 9). The long-term persistence of the KasB phosphomimetic strain without causing disease or mortality makes it an attractive model for studying latent Mtb infections and suggests that this attenuated strain may represent a valuable vaccine candidate against TB.
By analogy with a ΔkasB mutant, these phenotypes strongly support the view that phosphorylation has a detrimental effect on KasB activity. This is also emphasized by the modelling studies highlighting the strategic localization of the two phosphosites in the vicinity of the catalytic triad. That phosphorylation of these residues is likely to perturb the condensing activity of KasB contrasts with an earlier study proposing that phosphorylation of KasA and KasB reduces and increases their condensing activity in vitro, respectively [8]. The discrepancy between these two studies could be explained by several reasons. First, in the in vitro study, KasB activity was assessed using C16-AcpM, which represents a preferred substrate for KasA, but not for KasB [39] which catalyzes the last step in the elongation cycle [14], therefore acting on very long fatty acyl chains, such as C48-C52-AcpM substrate. Second, to conduct these studies, preparation of purified KasA or KasB from recombinant BCG were used, which contain large proportions of non-phosphorylated proteins, as usually within bacteria only a small proportion of proteins gets phosphorylated. A heterogeneous preparation containing a mixture of phosphorylated and non-phosphorylated isoforms may affect the overall activity of KasB. Finally, several studies have shown that most FASII components interact together [26], [27], [40] and a “mycolome” concept has recently emerged from work demonstrating that FASII system of Mtb is organized in specialized interconnected complexes composed of the condensing enzymes, dehydratase heterodimers and the methyltransferases [26], [27], [40]. In this model of interactome, three types of FASII specialized complexes are interconnected together: i) the initiation FASII is formed by a core consisting of the reductases, FabD and FabH, linking FASI and FASII together; ii) two elongation FASII complexes consisting of a core and either KasA (E1-FASII) or KasB (E2-FASII) which are capable of elongating acyl-AcpM to produce full-length meromycolyl-AcpM (Fig. 9); and iii) the termination FASII involving Pks13 which condenses the α branch with the meromycolic branch. One may therefore hypothesize that phosphorylation of KasB may also alter/disrupt heterotypic interactions with other FASII partners of E2-FASII which may in turn affect the activation of E2-FASII, leading to shortened mycolates. This would be missed in in vitro assays using single purified proteins.
To address the possibility that phosphorylation of KasB could also affect the localization of the protein, recombinant strains of M. smegmatis and M. bovis BCG expressing green fluorescent protein (GFP)-tagged KasB variants proteins were produced, comprising wild-type KasB, a phosphoablative (T334A/T336A) and a phosphomimetic (T334D/T336D) KasB isoform. We found that all KasB/eGFP fusion proteins localized in the mycobacterial cell wall, mostly on the bacterial poles (Figure S7 in Text S1). These findings strongly suggest that mycolic acid biosynthesis takes place during mycobacterial division at the bacterial poles and indicate that phosphorylation of KasB does not affect its localization in mycobacteria. Therefore, one can hypothesize that the primary consequence of KasB phosphorylation in vivo, is very likely to result in alteration of its enzymatic activity.
A recent study revealed that there were no detectable differences in the thickness of the cell envelope among the wild-type Mtb and ΔkasB mutant as determined by conventional transmission electron microscopy [41]. However, cryo-electron microscopy demonstrated that the region between the inner and outer membranes of the mutant strain, mainly composed of mycolic acids, showed a significant decrease in electron density as compared to the wild-type strain, suggesting that the altered mycolic acid pattern in the ΔkasB mutant may have affected the packing of the lipid-rich layer of the envelope [41]. One may hypothesize that the attenuated pathogenicity of the KasB phosphomimetic strain results in the increased permeability of its cell envelope due to the reduction in the number of tight bundles of mycolic acids, which facilitates the direct attack of effector molecules from the host cells. This altered mycolic acid packing and/or reduced chain length may also be responsible for the reduced capture of the dye, leading to loss of acid-fastness.
Previously, we have demonstrated that the mycolic acid cyclopropane synthase PcaA, which introduces cis-cyclopropane rings at the proximal position in α-mycolic acids, was phosphorylated by STPK [17]. As for KasB, phosphorylation of PcaA was associated with a significant decrease in the enzymatic activity. A PcaA phosphomimetic (T168D/T183D) strain, like the ΔpcaA mutant, exhibited reduced survival in human macrophages and was unable to prevent phagosome maturation, compared to the wild-type strain. This added new insights into the importance of mycolic acid cyclopropane rings in the phagosome maturation block and provided the first evidence of a Ser/Thr kinase-dependent mechanism for modulating mycolic acid composition and phagosome maturation block [17]. The present study extends these results by adding KasB in the growing list of virulence factors associated with mycolic acid metabolism, whose activities are directly regulated by phosphorylation. However, in contrast to PcaA, phosphorylation of KasB significantly altered colonization of macrophages and, unexpectedly, this phenotype was not shared by the ΔkasB strain. These differences were reflected in the different transcriptional profiles between the two strains. Among these transcriptional differences, specific up-regulation of the PAT biosynthetic gene cluster occurred in the phosphomimetic strain and correlated with increased production of PAT. Earlier studies demonstrated that PAT deficiency affects the surface global composition of the mycobacterial cell envelope, improving the efficiency with which Mtb binds to and enters phagocytic host cells [42], implying that PAT production affects early interaction between Mtb and macrophages. From these results, one can propose that the increased PAT production in the KasB phosphomimetic strain could be, at least partially, responsible for the decreased uptake by macrophages, which in turn may also explain the attenuation of the phosphomimetic mutant in mice. As shown in Figure S8 in Text S1, the phosphomimetic strain also produced PDIM, thus excluding the possibility that the attenuated phenotype of the phosphomimetic strain could be a consequence of the loss of PDIM. However, as suggested by the microarray data, multiple genes belonging to different classes such as adaptation, cell wall processes, intermediary metabolism, and respiration or regulatory proteins were found to be differentially regulated between the KasB phosphomimetic and knock-out strains. They may also, in addition to the PAT biosynthetic genes, participate in the phenotypic differences between the two strains.
Analysis of the phosphorylation status of KasB in BCG using anti-phosphothreonine antibodies indicated that phosphorylation was more prominent in stationary cultures than in replicating cultures. These finding are reminiscent of those reporting that phosphorylation of the FASII HadAB and HadBC complexes is growth phase-dependent and that phosphorylation occurred at higher levels in non-replicating bacteria [12], suggesting that phosphorylation is a mechanism by which mycobacteria might tightly control mycolic acid biosynthesis under non-replicating conditions. Mtb bacilli have two signature characteristics: acid-fast staining and the ability to cause long-term latent infections in humans. Acid-fast staining, such as Ziehl-Neelsen (ZN) staining, remains the cornerstone of diagnosis of TB, particularly in poor countries where the infection is highly prevalent [43]. Dormant bacilli have distinct structural alterations in the cell wall and are ZN-negative [44]. It is noteworthy that in a high percentage of patients exhibiting TB symptoms, analysis of patient's tissue samples may be positive for culture of mycobacteria and PCR analysis but negative for ZN staining. However, the reason(s) for the loss of acid-fastness during dormancy remain(s) unknown. A connection between loss of acid-fast staining and latent infection had been reported for a kasB deletion mutant [14] suggesting that regulation of KasB activity may cause these linked phenotypes. The present work provides the first evidence that phosphorylation of KasB correlates the loss of acid-fast staining and a loss of virulence allowing us to hypothesize that Mtb regulates these two related phenotypes through a signal transduction pathway. Further work will need to be done to elucidate these signals and determine which specific kinases regulate them in vivo. This knowledge might lead to better understanding of the molecular signals that trigger reactivation and TB disease. Although we show here that phosphorylation of KasB was more pronounced in stationary phase, additional studies are required to demonstrate whether this also happens in persistent mycobacteria and if it would result in the loss of acid-fast staining. This would open the way to improved methods for the diagnosis of latent TB infections.
All animal experiments and protocols described in the present study were reviewed and approved by the Animal Use and Care Committee of the Albert Einstein College of Medicine (Bronx, NY) complying with NIH guidelines under the Animal Study Protocol 20120114.
Strains used for cloning and expression (Table S2 in Text S1) of recombinant proteins were E. coli TOP10 (Invitrogen) and BL21(DE3)pLysS (Novagen) or BL21(DE3)Star (Novagen) grown in LB medium at 37°C. Media were supplemented with ampicillin (100 µg ml−1) or kanamycin (25 µg ml−1), as required. Mycobacterial strains used (Tables S1 and S2) were usually grown on Middlebrook 7H10 agar plates with OADC (oleic acid, albumine, dextrose, catalase) enrichment (Difco). Liquid cultures were obtained by growing mycobacteria either in Sauton's medium or in Middlebrook 7H9 (Difco) supplemented with 10% OADC enrichment, 0.2% (v/v) glycerol, and 0.05% (v/v) tyloxapol (Sigma) supplemented with either kanamycin (25 µg ml−1) or hygromycin (75 µg ml−1) when required. The multicopy expression plasmid pVV16 was reported earlier [45]. All plasmids used in this study are listed in Table S2 in Text S1 and the shuttle phasmid phAE159 was described previously [18], [46], [47].
The kasB gene was cloned into pCR-bluntII-TOPO using primers NtermKasB and CtermKasB (Table S3 in Text S1) to yield pCR-bluntII-TOPO::kasB. Site-directed mutagenesis was performed directly on pCR-bluntII-TOPO_kasB using inverse-PCR amplification with self-complementary primers (Table S3 in Text S1) carrying the desired mutation. The modified genes were subcloned in pETPhos using NdeI and NheI restriction sites, generating pETPhos_kasB_WT, pETPhos-kasB_T334A/T336A and pETPhos_kasB_T334D/T336D (Table S2 in Text S1). All constructs were verified by DNA sequencing. Recombinant KasB proteins were overexpressed in E. coli BL21(DE3)Star (Novagen) and purified as described previously [9]. Fractions containing pure KasB proteins were pooled, dialyzed when required and stored at −20°C until further use. The kasB gene was further subcloned from pETPhos_kasB_WT by NcoI/HindIII digest and ligated into pCDFDuet_1 vector already containing the PknF kinase domain [18], thus yielding pETDuet_kasB which allows co-expression of both PknF and KasB (Table S2 in Text S1).
A ΔkasB BCG mutant was transformed with either pVV16_kasB_WT [8] or pVV16_kasB_T334A/T336A which was derived from pVV16_kasB_WT by site-directed mutagenesis using inverse-PCR amplification with self-complementary primers carrying the desired mutation (Table S3 in Text S1). Transformants were selected on Middlebrook 7H10 supplemented with OADC enrichment, 50 µg ml−1 hygromycin and 25 µg ml−1 kanamycin and grown in Sauton's broth containing the same antibiotics. Exponential and stationary phase cultures were harvested, lyzed using a French Pressure Cell and purification of soluble KasB and KasB_T334A/T336A proteins was performed on Ni-NTA agarose beads as described earlier [8].
Mycobacteriophages used for transduction were prepared as described previously [47]. Briefly, kasB was PCR-amplified from the plasmids pETPhos_kasB carrying either the T334A/T336A or the T334D/T336D double mutations using the primers LL2 and LR1 (Table S3 in Text S1). accD6 was PCR-amplified using the primers RL and RR (Table S3 in Text S1). The kasB and accD6 PCR fragments were digested with AlwN1 and Van9I1, respectively and ligated with the 1.6 kb and 3.6 kb fragments of Van9I1-digested pYUB1471 (Table S2 in Text S1). The resulting plasmids were digested with PacI and ligated with the Pac1-digested shuttle phasmid phAE159. After packaging in vitro, the resulting phasmids were electroporated into M. smegmatis and the phages were amplified to obtain high-titer phage lysates.
Mtb (50 ml) was grown to log phase, washed twice with mycobacteriophage (MP) buffer (50 mM Tris, 150 mM NaCl, 10 mM MgCl2, 2 mM CaCl2; 50 ml) and resuspended in 5 ml of MP buffer. For each transduction experiment, 0.5 ml of cell suspension was mixed with 0.5 ml of high-titer phage lysate. The suspension was incubated at 37°C for 4 h without shaking, spun down, resuspended in 0.2 ml of 7H9 media (see above) and plated on 7H10 plates supplemented with hygromycin. Plates were incubated at 37°C for 4–5 weeks and transductants were picked and patched onto two hygromycin-containing plates. Colonies from one Hyg plate were used for DNA isolation using InstaGene Matrix (BioRad). PCR was performed using 5 µl of DNA for a 50 µl PCR reaction with the primers kasB_F and kasB_R (Table S3 in Text S1) and the resulting products were sequenced to check for the presence of the desired mutations. Southern blotting was done on genomic DNA isolated from Hyg-resistant transductants digested with BglII and probed with the kasB or kasA gene to confirm the allelic exchange and kasB deletion, respectively.
The kasB mutants were grown to log phase and 10 µl of culture were spread onto a glass slide. The slides were heated at 100°C for 2 min, dipped into 10% formalin for 30 min, dried and stained using the TB Fluorescent Stain Kit M (BD, Auramine staining) or the TB Stain Kit K (BD, Carbolfuchsin staining).
SCID mice and C57BL/6 mice (Jackson Laboratories) were infected via the aerosol route using a 2×106 CFU/ml mycobacterial suspension in PBS containing 0.05% tyloxapol and 0.04% antifoam. Five mice from each group were sacrificed at day 1, 7, 21 and 56 to determine the bacterial burden in the lung, spleen, and liver (one aerosol experiment was carried on for 119 days). Six mice per group were kept for survival experiments. All mice infected with Mtb were maintained under appropriate conditions in an animal biosafety level 3 laboratory.
C57BL/6 mice were used to obtain bone marrow-derived macrophages (BMDM). Isolated femurs were flushed with Dulbecco modified Eagle medium (DMEM; Gibco) supplemented with 10% heat-inactivated fetal bovine serum (FBS), 2 mM L-glutamine, and 1× non-essential amino acids (complete DMEM). The cells were cultured for 7 days in complete DMEM containing M-CSF (ebioscience) at 30–50 ng/ml, and then seeded into 24-well plates (∼4×105 cells/well) or 48-well plates (∼2×105 cells/well) as described previously [48]. The cells were allowed to adhere overnight prior to infection. The strains were grown as described above, washed and resuspended in DMEM supplemented with 10% FBS and diluted in this medium to achieve the appropriate titer. The bacteria were added to the wells at an approximate multiplicity of infection (MOI) of 1. Following 4 hrs incubation at 37°C to permit bacterial uptake, macrophage monolayers were washed twice with PBS to remove extracellular bacteria, following which wells were replenished with complete DMEM containing M-CSF at 10–50 ng/ml. At various times after infection, the medium in each well was removed to a tube containing sufficient SDS to give a final concentration of 0.025%; the cell monolayers were lysed with 0.025% SDS and combined with the medium. Lysates were diluted in PBS and plated onto Middlebrook 7H10 (see above) for determination of bacterial numbers.
Mtb strains were cultured to an OD600 of 0.5 in 7H9 broth supplemented with OADC, 0.2% glycerol and 2.5 ml 20% Tween 80. Cells were pelleted at 4500 rpm in 15 ml falcon tubes for 15 min, decanted, resuspended in 1 ml Trizol reagent (Invitrogen), and then transferred into new 2 ml screw cap microcentrifuge tubes containing 0.5 ml of zirconia-silica beads (diameter, 0.1 mm). Bacteria were disrupted in a Bead Beater (FastPrep Cell Disrupter, FP120) using two 45 second pulses at maximum speed of 6.5 m/sec, incubated on ice for 5 min, then 250 µl of chloroform was added, following by vigorous mixing for 15 s and then a 2–3 min room temperature incubation. Samples were then centrifuged at 12, 000 g for 5 min and the upper clear phase transferred carefully to a new tube and mixed with an equal volume of 70% ethanol, applied to an RNeasy mini column (QIAGEN) and processed according to the manufacturer's recommendations.
Five mRNA targets (kasB, cma2, sigA, 16S and rrnAP1) were reverse-transcribed, using a QuantiFast Multiplex RT-PCR kit (QIAGEN) according to the manufacturer's recommendations using a primer specific for each target gene. The conditions for Reverse Transcription PCR were 50°C for 50 min, 95°C 2 min. The amount of cDNA produced was quantified by real time PCR with the corresponding molecular beacon. All primer and molecular beacon sequences are listed in Table S4 in Text S1. The 10 µl PCR reaction mixture consisted of 1× PCR buffer, 250 µM dNTPs, 4 mM MgCl2, 0.5 µM each primer, 5 ng/µl molecular beacon and 0.03 U/µl Jumpstart Taq polymerase (Sigma-Aldrich). In order to normalize the individual reactions 6-carboxy-x-rhodamine (ROX) was always included as passive reference dye. PCRs were performed in 384-well microtiter plates in an ABI 7900 Prism (Applied Biosystems, Foster City, CA) according to the following parameters: initial denaturation at 95°C for 1 min, followed by 50 cycles of denaturation at 95°C for 30 seconds, annealing at 58°C for 30 seconds, and extension at 72°C for 15 seconds. PCR conditions were identical for all assays. The fluorescence was recorded during the annealing step of the assay. The quantity of specific target DNA was determined from the threshold cycle (CT) value with reference to a standard curve of genomic DNA. The copy numbers of target standards used ranged from 1 to 10E6 genomic copies per reaction (i.e. 10 fg to 10 ng DNA from CDC1551 strains). The lower limit of detection for each of the five assays was 10 fg which is equivalent to 1–5 copies of cDNA. RT reactions were performed in triplicate. The data for kasB and cmaA2 was normalized against 3 different control genes viz. sigA, rrnAP1 and 16S rRNA [49], [50], [51]. The mean of each triplicate was used in calculations.
Triplicate samples were prepared by harvesting growing cultures of Mtb CDC1551 KasB strains (OD600 nm≈1). Extraction of RNA, preparation of cDNA, and microarray analysis were performed as described previously [52]. The array data have been deposited in the Gene Expression Omnibus at NCBI with accession number GSE47640.
In vitro phosphorylation was performed as described [53] with 4 µg of KasB in 20 µl of buffer P (25 mMTris-HCl, pH 7.0; 1 mM DTT; 5 mM MgCl2; 1 mM EDTA) with 200 µCi ml−1 [γ-33P]ATP corresponding to 65 nM (PerkinElmer, 3000 Ci.mmol−1), and 0.2 to 1.0 µg of PknF kinase in order to obtain its optimal autophosphorylation activity for 30 min at 37°C. Cloning, expression and purification of the PknFGST-tagged kinase from Mtb were described previously [8].
Bacteria were disrupted by bead beating with 1-mm-diameter glass beads and the total protein concentration in each cell lysate was determined using a bicinchoninic acid (BCA) protein assay reagent kit (Pierce). Equal amounts of proteins (20 µg) were separated on 12% SDS-PAGE gels and were transferred to a nitrocellulose membrane. The membrane was saturated with 1% BSA in PBS/Tween 0.1% and either probed with rat anti-KasA antibodies (dilution1∶500) [54], rabbit anti-AccD6 antibodies (dilution, 1∶1,000) [55] or anti-phosphothreonine antibodies (dilution, 1∶1,000) (Cell Signaling). After washing, the membrane was incubated with either secondary antibodies labeled with IRDye infrared dyes (Odyssey Classic) or alkaline-conjugated anti-rabbit secondary antibodies (dilution, 1∶7,000) and revealed with BCIP/NBT, according to the manufacturer's instructions.
50–100 ml cultures of Mtb strains were grown to mid-log phase in Middlebrook 7H9 medium at 37°C. Cells were harvested and FAMEs and MAMEs were extracted as reported [56], subjected to one-dimensional thin layer chromatography (TLC) with silica gel plates (silica gel 60F254; Merck, Germany) and developed in hexane/ethyl acetate (19∶1, v/v; 3 runs). Lipids were revealed by spraying the plates with molybdophosphoric acid followed by charring. α-, methoxy- and keto-mycolic acids were then purified using preparative TLC plates and detection by spraying with ethanolic Rhodamine 6G to visualize the lipids under a UV lamp. Areas corresponding to the different mycolic acid subspecies were scrapped off the plates and extracted from the silica gel with diethyl ether. Samples were then resolved again on a second preparative TLC plate and re-extracted. Purity of the mycolates was then assessed on a standard TLC plate in hexane/ethyl acetate (19∶1, v/v; 3 runs) prior to NMR and MALDI-TOF analysis.
Mtb cultures (10 ml), grown in Middlebrook 7H9-OADC-glycerol-tyloxapol to an OD600 nm of 0.2, were treated with either 14C-acetate (10 µCi) or 14C-propionate (20 µCi) for 20 h or 2 days, respectively. Apolar lipids were extracted by mixing the cell pellets with methanol (2 ml), 0.3% aqueous NaCl solution (0.2 ml) and petroleum ether (1 ml) for 15 min. The suspensions were centrifuged and the upper petroleum ether phases were removed. The cell pellets were extracted a second time with petroleum ether (1 ml) for 15 min. The petroleum ether phases were combined, dried, and resuspended in dichloromethane (0.2 ml). The same amount of cpm was loaded onto a Silica gel 60 F254 250-µm aluminum plate (10×10 cm) and eluted [1st dimension∶petroleum ether/acetone 23∶2, ×3; 2nd dimension∶toluene/acetone 19∶1]. 14C-Radiolabeled species were detected by autoradiography after exposure at −80°C for 2 days on X-ray film.
For NMR analyses, MAMEs were dissolved into deuterated chloroform containing 0.01% of TMS and transferred into Shigemi tubes matched for D2O. Then 0.1 ml of deuterium oxide was added to avoid solvent evaporation during long acquisition. 1D proton NMR spectra were recorded at 300K on a 400 MHz Avance II Bruker spectrometer equipped with a 5 mm broad-band inverse probe.
Mass spectrometric analyses of MAMEs were performed on a Voyager Elite reflectron MALDI-TOF mass spectrometer (PerSeptiveBiosystems, Framingham, MA, USA), equipped with a 337 nm UV laser. Samples were solubilized in 1 µl chloroform/methanol (2∶1, v/v) and mixed on target with 1 µl of 2,5-dihydroxybenzoic acid matrix solution (10 mg/ml dissolved in chloroform/methanol 2∶1, v/v).
|
10.1371/journal.ppat.1002216 | Th2-polarised PrP-specific Transgenic T-cells Confer Partial Protection against Murine Scrapie | Several hurdles must be overcome in order to achieve efficient and safe immunotherapy against conformational neurodegenerative diseases. In prion diseases, the main difficulty is that the prion protein is tolerated as a self protein, which prevents powerful immune responses. Passive antibody therapy is effective only during early, asymptomatic disease, well before diagnosis is made. If efficient immunotherapy of prion diseases is to be achieved, it is crucial to understand precisely how immune tolerance against the prion protein can be overcome and which effector pathways may delay disease progression. To this end, we generated a transgenic mouse that expresses the ß-chain of a T cell receptor recognizing a PrP epitope presented by the class II major histocompatibility complex. The fact that the constraint is applied to only one TCR chain allows adaptation of the other chain according to the presence or absence of tolerogenic PrP. We first show that transgene-bearing T cells, pairing with rearranged α-chains conferring anti-PrP specificity, are systematically eliminated during ontogeny in PrP+ mice, suggesting that precursors with good functional avidity are rare in a normal individual. Second, we show that transgene-bearing T cells with anti-PrP specificity are not suppressed when transferred into PrP+ recipients and proliferate more extensively in a prion-infected host. Finally, such T cells provide protection through a cell-mediated pathway involving IL-4 production. These findings support the idea that cell-mediated immunity in neurodegenerative conditions may not be necessarily detrimental and may even contribute, when properly controlled, to the resolution of pathological processes.
| It is generally accepted that prion-specific antibodies can protect against mouse scrapie infection. However, passive antibody therapy is limited to the lymphoinvasion stage of the disease. Active immunization has been attempted but the results have been disappointing. There is therefore a need for developing analytical models that will allow a fine dissection of the immune mechanisms at play in prion diseases and help distinguish between protective effects mediated by B cells and antibodies, and the effect of T cells. The aim of our study was to thoroughly examine T cell tolerance to the prion protein and to evaluate whether a pure specific population of T cells adoptively transferred to a normal host could proliferate and confer protection against scrapie. We designed a transgenic mouse in which the majority of T lymphocytes recognize the prion protein. Our key findings are that prion-specific T cells remain functional when transferred to normal recipients, even more so when the host is infected with scrapie, and confer partial protection against the disease by slowing down prion replication, in complete absence of anti-prion antibodies. Anti-prion T cells may therefore be considered as a therapeutic tool in the future.
| Prion diseases, also termed transmissible spongiform encephalopathies (TSE), are fatal neurodegenerative disorders against which no treatment is available yet. The key pathogenic event is the conversion of the cellular prion protein (PrPc), a ubiquitous, host-encoded glycoprotein, into a misfolded protein, PrP scrapie (PrPSc) [1]. PrPSc forms oligomers that are self-propagating and cause neuronal damage. PrPSc is the presumed prion agent which is necessary and sufficient for disease transmission and gives strain-associated characteristics [2].
Many recent reports have shown that mice treated with antibodies (Abs) against PrP [3], [4], [5] or vaccinated against the protein [6], [7], [8] acquire resistance to scrapie peripheral infection. Encouraging as these findings may be, the results against TSE are not fully satisfactory yet and have precluded clinical trials. On one hand, passive Ab therapy is effective under restricted conditions, notably before neurological symptoms appear [4]. On the other hand, active vaccination is limited by the strong tolerogenicity of self PrP [9], [10]. With a few exceptions [6], vaccinated mice enjoyed only limited remission and ultimately succumbed to prion infection.
Focusing TSE immunotherapy on CD4+ T cells rather than on Ab-producing cells may overcome these difficulties [11]. Two strategies borrowed from cancer immunotherapy have already been probed for that purpose: dendritic cell (DC) vaccination [12] and adoptive CD4+ T cell therapy [13]. Improving T-cell-based immunotherapy, however, requires deeper insights into several issues. The way the anti-PrP repertoire is selected in the thymus and the chances that anti-PrP T cells with good functional avidity may overcome negative selection must be evaluated. One must also understand the way such lymphocytes, whether generated in the host or adoptively transferred, can be activated by prion-infected cells and neutralize prion progression. To address these issues, we have produced a transgenic (Tg) mouse expressing a single T cell receptor (TCR) β-chain from an anti-PrP TCR. The α-chains rearrange freely so that the anti-PrP repertoire adapts to the antigenic context. This model made it possible to follow repertoire development in PrP+ and PrP– mice through analysis of the α-chain rearrangements, and to produce highly enriched populations of T helper (Th) cells. The therapeutic efficiency of those T cells was assessed following adoptive transfer.
Among three founders, only one B6 male (Figure 1A) over-expressed the transgenic beta variable (BV) 12+ rearrangement in peripheral blood cells (29.8% versus 2.5% in controls, data not shown), and transmitted this phenotype to its progeny. Spleens and lymph nodes (LN) of Tg mice had normal size and normal total white cell content. Spleens of PrP+ and PrP– Tg mice contained 90±5×106 and 92±6×106 cells versus 96±8×106 cells for spleens of wild type (WT) littermates (n = 5, data not shown). Tg mice on both PrP+ and PrP– backgrounds displayed a minor reduction in the relative percentage of CD4+ T cells compared to CD8+ T and total B cells (Figure S1), suggesting that the expression of the TCR transgene had a slight impact on T cell ontogeny, as also reported in other TCR-Tg lines [14]. Secondary lymphoid organs from Tg mice showed fully developed germinal centers and the presence of cells with strong cell-surface PrPc expression (Figure S2A,B). The expression of PrPc was also similar in the brain between WT mice and Tg PrP+ mice, as assessed by immunohistochemistry and by FACS analysis (Figure S3A,B). Therefore, the transgenic insertion of the TCR β-chain in affects neither the architecture of secondary lymphoid organs nor PrPc expression.
As shown in Figure 1B, the BV12+ rearrangement was still dominant in the third generation of mice raised on PrP+ or PrP– backgrounds. The fact that endogenous rearrangements were not excluded and that only a third of total CD4+ T cells were BV12+ suggested a belated expression of the transgene due to the nature of the cassette. This inference was confirmed by flow cytometry analysis of thymocytes from Tg mice showing an increase in the percentage of transgenic BV12+ rearrangements only after they reach the single positive stage (Figure 1C). The expression of the β-chain transgene in the SP8 subset resulted from the deletion of the Cd8 gene silencer, which is present in the original plasmid, but which happens to be excised during the construction process [15].
Figure 1D shows the respective percentages of transgene-bearing CD4+ and CD8+ LN lymphocytes in several PrP+ and PrP– Tg mice. Percentages of BV12+ T cells were practically identical in PrP– and PrP+ mice, and the higher percentage of transgene expression among CD8+ LN T cells confirms the known preferential affinity of BV12+ rearrangements for MHC class I products during thymic selection [16].
As reported [9], [10], PrP+ WT mice do not normally respond to PrP. Because PrP+ Tg mice expressed a high percentage of TCR β-chain rearrangements, we decided to examine whether, in contrast to WT mice, these mice would be responsive. As Figure 2A shows, in vivo primed T cells from Tg mice proliferated weakly and to the same extent as T cells from WT littermates. Therefore, either pairings with alpha rearrangements conferring anti-PrP reactivity were deleted, or PrP-responders escaping selection were suppressed, notably by regulatory T (Treg) cells [17]. To test the latter possibility, Tg PrP+ mice were in vivo deprived of Treg cells by anti-CD25+ T cell Ab (clone PC61) before being primed with PrP158–187 4 days later. CD4+ T cells deprived of Treg lymphocytes proliferated more vigorously than untreated controls in response to antigenic challenge, but within the same range as non-Tg littermates similarly deprived of Treg cells (Figure 2B). Treg blockade therefore did not reveal the specific presence of PrP responders within the pool of transgene-bearing T cells.
The situation was remarkably different in PrP– Tg mice. In vivo primed CD4+ T cells from these mice proliferated almost five times as much upon in vitro challenge than non-Tg PrP– mice, which also respond to the peptide [18] (Figure 2C). T cells responded essentially to peptide PrP158–187 and only marginally to an I-Ab-restricted peptide of ovalbumin (OVA323–339) (Figure 2D). Conversely, priming CD4+ T cells from Tg mice with OVA323–339 generated a modest, but specific response against ovalbumin which suggests that a few non-PrP precursors are present in the total CD4+ repertoire of Tg mice (Figure 2D). PrP158–187-specific IFN-γ-producing CD4+ T cells evaluated by ELISPOT were also substantially increased in Tg versus non-Tg mice (Figure S4). No IL-4 response could be detected under the same conditions, suggesting an initial shift of anti-PrP T cells toward a Th1 profile. A characteristic of biased T cell repertoires is their capacity to respond to primary in vitro stimulation [19]. Naive CD4+ T cells from Tg mice did indeed proliferate in response to PrP158–187 in a dose-dependent manner, in contrast to naive non-Tg T cells (Figure 2E). Expansion of the anti-PrP repertoire in PrP– Tg mice could also be evidenced by the fact that the percentage of BV12+ T cells in LN underwent a substantial increase after peptide priming (Figure S5).
To gain insight into the mechanisms of repertoire selection on PrP+ and PrP– backgrounds, we analyzed the diversity of TCR α-chain rearrangements among sorted CD4+ BV12+ T cells by immunoscope. Figure S6 gives an overview of the 20 alpha variable chain (TRAV) families associated with the β-chain transgene in naive and primed Tg mice. Table 1 shows the actual percentages for each family. TRAV family usage did not differ between PrP+ or PrP– naive Tg mice. In particular, TRAV13, the variable segment constitutive of the α-chain of the original hybridoma that had donated the β-chain, was used to the same extent in the two Tg progenies. The pattern changed in primed BV12+ T cells. Whereas TRAV13 usage in PrP+ primed mice was practically unmodified, it represented almost 80% of all families in PrP– mice. Moreover, the length of the TRAV13 third complementarity determining region (CDR3) followed a Gaussian distribution except for the rearrangements in primed PrP– mice where the distribution presented a predominant peak of 12 residues (Figure 3A). This CDR3 was homogeneous enough to be sequenced directly from the PCR product. The sequence was identical to the α-chain CDR3 expressed in the original TCR hybridoma (Figure 3B).
Primed Ly5.2+ CD4+ T cells from Tg PrP– donors were labeled with carboxyfluorescein diacetate succinimidyl ester (CFSE) and injected into Ly5.1+ PrP+ and PrP– recipients. BV12+ lymphocytes among Ly5.2+ CD4+ gated T cells collected after a 3-day engraftment into PrP+ hosts had markedly proliferated, as shown by the presence of a resolved population in the upper left quadrant (Figure 4A). Far fewer non-BV12 CD4+ T cells were seen in the lower left quadrant. As expected, T cells injected into PrP– mice did not respond (Figure 4B). The “% divided” measuring the number of lymphocytes that had undergone mitosis and the average number of divisions per cell (division index) was much higher in the BV12+ than in the non-BV12 subset, indicating that responders were considerably more numerous and still reactive in the PrP+ environment (Figure 4C).
Next, we compared the proliferation of BV12+ CD4+ T cells in prion-infected versus healthy mice (Figure 4D). Of interest, the percentage of dividing T cells was strikingly higher in prion-infected mice. The same trend can be observed in Figure 4E, recapitulating data of more than 10 mice per group. Differences were statistically significant between infected and normal recipient mice (p<0.05 by ANOVA and Bonferroni's two-by-two comparisons). Infected antigen presenting cells (APCs) reactivated anti-PrP T cells more efficiently than non-infected APCs.
Having previously shown that PrP-sensitized polyclonal T cells attenuate scrapie evolution [13], we undertook to confirm those conclusions with transgene-bearing CD4+ T cells. Two ×105 CD4+ BV12+ primed T cells were transferred after sorting into CD3εo/o mice which had been infected 1 day before with 2×104 LD50 of the prion strain 139A. Recipients of BV12+ T cells were boosted with peptide PrP158–187 one day after transfer and once monthly thereafter or left non-boosted. Controls consisted of mice transferred with sorted BV12-negative CD4+ T cells further boosted and of non-transferred mice. Neurological symptoms appeared in two waves. Non-transferred mice and mice transferred with BV12-negative T cells became sick at 170 and 178.5 median day post-infection (dpi), respectively, whereas recipients of BV12+ T cells either boosted or not became sick at 217 and 197 median dpi, respectively (Figure 5A). One out of the 6 BV12+ T cell transferred and boosted mice remained free of symptoms (Figure 5A). Differences in kinetics were statistically significant among the four groups according to multivariate log rank test, as well as between the two control groups compared to the two experimental groups which had received BV12+ T cells, but not between boosted versus non-boosted recipients of transgene-bearing T cells nor between the two control groups which received BV12-negative T cells or no T cells. Overt disease duration from onset to terminal stage was significantly longer in the two experimental groups treated with BV12+ T cells than in the two control groups (42 median day versus 33 median day respectively) (Figure 5B). Brains of mice culled at terminal stage contained similar amounts of proteinase-K (PK) -resistant PrP, irrespective of the nature of the T cell transfer, the clinical onset or the overt disease duration (Figure S7A). This suggests that terminal stage occurs when a sufficient amount of PK-resistant PrP has accumulated in the brain. In contrast, PK-resistant PrP could not be detected in the spleen and brain of the mouse which had remained permanently free of neurological symptoms and was sacrificed at 350 dpi (Figure S7B).
To get an insight on where T cells exert their prionostatic effects, two types of experiments were performed. First, a few experimental and control mice were sacrificed at 90 dpi in order to compare the levels of PK-resistant PrP in their spleens by western blot. Second, we looked for the presence of T cells in the brain of the above mentioned mice sacrificed at 90dpi and of mice sacrificed at terminal stage. As shown in Figure 6, 4 out of 6 spleen samples of mice that had received BV12+ CD4+ T cells (either boosted or non-boosted), were almost completely free of pathological PrP, suggesting that anti-PrP T cells blocked the propagation of PK-resistant PrP in peripheral lymphoid tissues. At terminal stage, T cells were present in the brains of all mice which had received BV12+ T cells, whether boosted or non-boosted. At 90 dpi, one mouse transferred with BV12+ T cells and boosted already showed CNS infiltration. Fewer infiltrating T cells were detected, even at terminal stage in the brains of mice that had received BV12-negative T cells (Figure 7).
Blood samples were collected at 90 dpi to assess the presence of donor T cells in the CD3ε0/0 hosts (Figure S8). CD4+ T cells were clearly evidenced among the peripheral blood lymphocytes of transferred mice. The expansion of BV12+ CD4+ T cells was virtually identical in mice which had been boosted with peptide PrP158–187 or in non-challenged mice.
Serum samples were collected at 90 and 120 dpi to monitor for the presence of circulating Abs against native PrPc which presumably convey protection (Figure 8A). Abs were barely detectable in all serum samples, regardless of collection time and of treatment. Transferred T cells therefore had not initiated a productive T-B cooperation with host B lymphocytes. Next, we looked at antigen-specific T-cell proliferation and lymphokine secretion at 120 dpi. Engrafted BV12+ T cells could still proliferate in response to PrP158–187 (Figure 8B). A substantial proportion of T cells also released IL-4 after antigenic challenge, as evidenced by ELISPOT (Figure 8C), but at variance with freshly activated anti-PrP T cells, long-term engrafted lymphocytes were low producers of IFN-γ (Figure 8D). Contrasting with those low values, control T cells stimulated with concanavalin A displayed a frequency of IFN-γ secretors in the range of 200 spots per 1×105 cells (data not shown).
The mechanisms at work in prion immunotherapy are still poorly understood. Abs against PrP or PrP receptors have been considered to be the major protagonists so far, following the demonstration that they cured infected cell lines in vitro [20], [21] and retarded disease progression in vivo [3], [4], [5], [22]. By contrast, many authors have seen the involvement of T cells in neurodegenerative conditions as counterproductive, based on the fact that T cells may promote neuronal decay via microglia [23] or initiate autoimmune complications such as reported in Alzheimer's disease patients [24]. Recent evidence suggests, however, that in addition to providing help to B cells, CD4+ T cells may also make a contribution of their own against neurodegenerative conditions [25].
Our first objective was to find out whether anti-PrP T cells with good TCR functional avidity may escape central tolerance and be available for active vaccination or adoptive therapy. With the use of single β-chain Tg mice we were able to show how anti-PrP precursors emerge in a PrP+ or negative environment. A first conclusion is that the preselected PrP repertoire in the thymus is tightly controlled, leaving little chance for lymphocytes with good TCR avidity to escape selection and to settle in peripheral organs. Although the percentages of transgene-bearing CD4+ T cells in PrP+ Tg mice are as high as in PrP– mice, their reactivity to PrP peptide is as limited as that of T lymphocytes from non-Tg littermates. The blockade of CD25+ regulatory T cells prior to antigenic challenge increased proliferation and IFN-γ release to the same extent as it did for T cells from non-Tg mice. Therefore, Treg cells cannot presently account for the lack of responsiveness of PrP+ Tg mice even though a regulation of anti-PrP precursors by foxp3+ T cells has been described [7]. The situation is totally different in PrP– Tg mice in which CD4+ T cells displayed strong reactivity against the dominant PrP epitope. They proliferated more vigorously upon antigenic challenge than non-Tg littermates, producing more IFN-γ and responding in vitro even without in vivo priming. Although the TCR α/β repertoire against a given antigenic specificity is generally considered to be highly diverse, limited heterogeneity of β-chain rearrangements has been reported [26], [27], [28]. Limited heterogeneity is also reflected by the fact that the T cell repertoire of single TCR β-chain Tg mice is frequently skewed toward the specificity of the TCR that provided the rearranged chain [19]. But, as shown here, the β-chain is not sufficient by itself to impart TCR specificity since BV12+ T cells in PrP+ mice were unresponsive to PrP. As other studies have demonstrated, the α-chain makes an essential contribution to the final specificity and determines ultimately whether a T cell precursor responds or not to a defined epitope [28], [29], [30], [31]. Central selection in PrP+ mice probably eliminates the alpha pairings that confer anti-PrP responsiveness. This phenomenon is all or none because α pairings that would result in TCRs of intermediate avidity are visibly not spared either. Selection based on the choice of a TCR α-chain rearrangement is one of many strategies aimed at preventing autoreactivity [32]. A different TCR β-chain transgene originating from a TCR with lower functional avidity would possibly have allowed pairing with α-chains conferring responsiveness, as evidenced by the generation of activated T cells in normal WT mice challenged with immunogenic formulations of PrP.
Immunoscope analyses confirmed those conclusions. No bias in TRAV family usage was identified in naive CD4+ BV12+ T cells in a PrP+ or -negative context. This indicates a wide variety of α pairings among exiting T cells even though anti-PrP precursors are already abundant in the naive repertoire of PrP– Tg mice, as shown by primary in vitro responses. However, after antigen selection, the distribution of TRAV families differed radically in PrP+ and PrP– mice. TRAV13 usage in primed PrP+ Tg mice was similar to that of naive mice, whereas TRAV13 usage became predominant in primed T cells from PrP– Tg mice. Similar changes in α-chain diversity after antigenic priming were demonstrated in single-chain Tg mice challenged with lymphocytic choriomeningitis virus (LCMV) and in Tg non-obese diabetic (NOD) mice spontaneously developing autoreactive T cells against islets of Langerhans [29], [30]. Noteworthy, this dominating TRAV13 rearrangement showed a highly homogeneous CDR3 domain. The CDR3 nucleotide sequence was identical to that of the α-chain isolated in the original T cell hybridoma. From a practical point of view, this result shows that after antigen priming, the BV12+ T cell subset is composed of a quasi-monoclonal population of anti-PrP effector cells. A similar result was described in a single-chain Tg NOD mouse, in which CD8+ T cells had infiltrated the pancreas and had been activated locally [30].
Another valuable conclusion, from the perspective of adoptive T cell therapy, is that anti-PrP T cells maintain their reactivity in a PrP+ environment and are preferentially activated by prion-infected cells. CFSE-labeled CD4+ T cells from primed Tg mice showed vigorous proliferation after a 3-day engraftment in PrP+ mice or in PrP– mice immunized with peptide. As expected, they did not proliferate in non-challenged PrP– hosts, confirming that the response was antigen specific and not caused by homeostatic regulation [33]. Contrasting with the highly rigid selection at work in the thymus, the periphery appears to be more permissive, notably to primed T cells. As previously suggested [13], naive anti-PrP precursors may be more receptive to peripheral regulation. It is worth noting that, far from being inhibited, specific T cells proliferated more vigorously in scrapie-infected mice. The quantitative parameters indicated that the increased proliferation results from more BV12+ T cells being activated and undergoing division than from an increased number of divisions within the 3 days of engraftment. Several reasons could account for the observed heightened T cell proliferation. Prion-infected APCs might present a higher density of MHC class II molecules filled with PrP peptide (signal 1), thus activating T cells more efficiently. A non-mutually exclusive alternative could be that prion-infected APCs provide more co-stimulation (signal 2) to T cells than healthy APCs. Preliminary examination of CD80 and CD86 expression on DCs collected from infected mice did not favor this latter hypothesis (data not shown), but other co-stimulatory pathways deserve to be examined. Finally, because PrPc is mobilized at the immunological synapse [34], the possibility that transconformed PrPSc might enhance APC function cannot be totally ruled out, even though this pathway remains difficult to conceive given the structural properties of the conformer.
In accordance with previous observations [13], sensitized anti-PrP T cells passively administered into infected mice convey protection with no evidence of negative side effects. Presently, as few as 2×105 primed BV12+ T cells delay neurological onset and overt disease progression to the same extent as 50 times more polyclonal T cells [13]. The necessity of boosting Tg T cells seems less imperative than with polyclonal T cells, a fact which is consistent with the observation that PrP-specific T cells can be re-activated in vivo by prion infected-APCs or by APCs naturally processing endogenous PrPc, with no need to boost with exogenous peptide.
The lack of detectable anti-PrP Ab in T-cell transferred mice argues against the recruitment of resident B cells and the involvement of Abs in delaying disease progression. As previously reported [9], anti-PrP B cells differentiating in a PrP+ context, which is the situation in the recipient, are severely repressed and few precursors remain available in the periphery. In addition, by sorting donor cells for maximum purity, we made sure that donor B cells were not co-transferred. A T-cell-mediated pathway therefore seems to be the most likely explanation for the observed disease attenuation. T cells activated ex vivo after a 120-day engraftment still proliferate in the presence PrP peptide and produce substantial amounts of IL-4. We previously reported that a hallmark of polyclonal CD4+ T cells conveying protection was the production of both IFN-γ and IL-4 [13]. In the present experimental setting in which small numbers of highly enriched effectors were transferred, we can further suggest that protection is conferred by anti-PrP T cells that produce IL-4 rather than IFN-γ. Anti-PrP T cells, which are predominantly biased toward IFN-γ release shortly after priming, seem to evolve with time toward a Th2 oriented profile. In future experiments, the transfer of homogeneously polarized lymphocytes should confirm whether a Th2 profile preferentially conveys protection.
Where and how do T cells precisely halt disease progression? A reasonable hypothesis, also supported by the reduced content of PrPSc found in spleens of protected mice, is that PrP-sensitized T cells exert their prionostatic effects in the periphery, when prions propagate into secondary lymphoid tissues. The delayed clinical onset and protracted disease evolution would thus be a direct consequence of the slowed down lymphoinvasion, as already demonstrated in situations where the connection between follicular dendritic cells and nervous endings is perturbed [35] or when DCs or follicular dendritic cells are temporarily deleted [36], [37], [38]. Anti-PrP T cells would preferentially be activated at the sites of prion expansion such as in germinal centers, upon contact with infected DCs [39]. They would shift progressively toward a Th2 profile, and would mobilize agents of innate immunity such as alternatively-activated macrophages with a capacity to reduce inflammation and to degrade infectious oligomers in lymphoid follicles [40], [41], thereby retarding lymphoinvasion. But extraneural and neural invasion are not necessarily connected and the two processes could evolve independently [42], [43]. It is then conceivable that protective T cells exert their prionostatic effects in parallel, in peripheral lymphoid tissues and in the CNS. Both the extension of clinical disease duration and the detection of infiltrating BV12+ T cells in the brain of adoptively transferred mice speak in favor of a central action of T cells. Beneficial effects of IL-4-producing T cells have been described in various CNS pathological conditions as well as in physiological situations of memory acquisition [44], [45], [46]. Microglial cells or blood borne macrophages alternatively activated by Th2 T cells might retain their innate capacity to degrade pathological oligomers or fibrils [47] while protecting neurons from PrPSc-mediated toxicity. A recent study showing that prion propagation and neurotoxicity are dissociated processes [48] further supports the idea that the protective role of alternatively-activated effectors of innate immunity may not be confined to phagocytosis of amyloid aggregates only. Microglia activated by IL-4 induces oligodendrogenesis from adult stem cells. This effect is mediated, at least in part, by insulin-like growth factor-1 [49]. IL-4 can also counteract, in a dose-protective manner, the harmful effect of microglia activated by LPS. The mechanism by which IL-4 exerts its neuroprotective effects, involves the decrease of TNF-α and nitric oxide production [50]. The role of IL-4 in regulating CNS inflammation was also investigated in experimental autoimmune encephalomyelitis, a mouse model for multiple sclerosis. Ponomarev et al. showed that mice deficient in IL-4 had exacerbated neurological symptoms associated with a significant increase in infiltrating inflammatory cell number. CNS-resident microglial cells expressed in an IL-4 dependent manner the protein Ym1, a marker of alternatively activated macrophages [51].
In conclusion, our present results provide further support in favor of a positive role of cell-mediated immunity in TSE. T cells are not necessarily detrimental. They can be useful depending on their differentiation profile and, most likely too, on the stage of disease evolution. More generally, the success of immunotherapeutic strategies against neurodegenerative conditions will depend on our capacity to draw a clear line between useful and harmful processes and to set up beneficial synergies between the different arms of innate and acquired immunity.
All animal procedures were carried out in strict accordance with the French legislation (Rural Code articles L 214-1 to L 214-122 and associated penal consequences) and European Treaty ETS 123 (1986). They were approved by the Regional Ethic Committee “Charles Darwin” to which the animal facility is affiliated.
PrP– mice (Prnp0/0) [52] were propagated on a C57BL/6 (B6) background [13]. Ly5.1 and CD3ε0/0 mice were also on a B6 background.
A rearranged TCR β-chain (TRBV12-1-01*/TRBD1-01*/TRBJ1-4-02* (http://imgt.cines.fr)) was cloned from a T hybridoma generated by fusion of T cells from PrP– mice immunized with peptide PrP158–187 [18]. Genomic DNA was inserted into an expression cassette containing the Cd4 promoter [15]. The Cd8 silencer, originally present, was ultimately excised from the construct. The backbone vector was a PNNO3 plasmid, and the transgene was inserted at a SalI restriction site and excised with NotI. Purified DNA fragment was microinjected into WT B6 eggs (SEAT facility, Villejuif, France). Founders were identified by tail PCR, and the selected founder was mated with PrP+ and PrP– breeders to establish two independent lines. The transgene was kept heterozygous in both progenies.
Transgene expression was followed with anti-Vβ5.1/5.2 Ab (clone MR9-4) conjugated to FITC (BD Biosciences, Pont-de-Claix, France). Other reagents were CD4-PE, CD8-PerCP-Cy5.5, CD19-PE, and Vβ5.1/5.2-biotinylated revealed by streptavidin-APC, all from BD Biosciences. Samples were analyzed with BD CellQuest or FlowJo (TreeStar, Ashland, Oregon, USA).
Mice received 50 µg of peptide PrP158–187 in complete Freund's adjuvant. Spleens and LNs were collected 10 days later. CD4+ T cells were enriched above 90% purity using a negative isolation kit according to the manufacturer's recommendations (Dynal, Invitrogen, Oslo, Norway).
CD4+ T cells were assayed for proliferation in micro-culture plates containing mitomycin-C-treated PrP– spleen cells and various concentrations of PrP158–187 as previously described [18]. Cultures were pulsed overnight with tritiated thymidine after a variable number of days (indicated in figure legends). Lymphokine-secreting T cells were enumerated by ELISPOT assay according to standard procedures [34].
Enriched CD4+ T cells (1×107 cells) were labeled with 2.5 µM of CFSE (Molecular Probes, Invitrogen, Cergy-Pontoise, France) for 15 minutes in PBS at 37°C, washed twice with PBS-FCS 3%, and resuspended in PBS. Cells were injected i.v. at 2×106 per mouse. Specific proliferation was measured at day 3 on a LSR2 cytometer (BD). Cells were also stained with Vβ5-biotinylated-streptavidin−PerCP-Cy5.5, CD4-APC-Cy5.5 and CD45.2-APC. Results are expressed as “% divide” (% of cells which underwent divisions) or “division index” (the average number of divisions that a cell undergoes, including cells that did not divide).
The diversity of α rearrangements was analyzed by immunoscope [53]. TRBV12+ T cells were sorted on a cell sorter (FACS-ARIA BD). Total RNA was prepared using a micro kit (RNeasy, QIAGEN, Courtaboeuf, France), and cDNA was synthesized with SuperScript II Reverse Transcriptase (Invitrogen). The distribution of TCR-Vα germline genes clustered into 20 families (IMGT nomenclature) was obtained by PCR. The relative usage of each Vα family was calculated according to the formula:in which Ct(x) is the fluorescent threshold cycle number measured for a Vα (y) family.
For immunoscope profiles, products were subjected to run-off reactions for three cycles with a nested fluorescent primer specific for the constant region. The fluorescent products were separated and analyzed using an ABI-PRISM 3730 DNA analyzer. The size and intensity of each band were analyzed with “Immunoscope software” [53] further adapted to the capillary sequencer [54]. Fluorescence intensities were plotted in arbitrary units on the y axis, and CDR3 lengths (in amino acids) on the x axis.
Mice of both sexes, between 8 to 12 weeks of age, were infected i.p. with a brain homogenate of 139A prions containing 2×104 LD50. Mice were monitored twice a week for gait control on a set of parallel bars [39].
Spleens and brains collected at various time points post-infection were homogenized with the Ribolyser method and adjusted at 10% w/v in PBS plus a cocktail of anti-proteases. Precipitation with sodium phosphotungstic acid (Sigma) was performed as previously described [12]. Aliquots equivalent to 6 mg were PK-digested or left undigested. PK (proteinase K, Roche) was applied at 50 µg/ml, at 37°C for 30 min. Samples were run on 12.5% SDS-PAGE. PrP was revealed with Ab SAF-84 at 1/5000 (generously provided by Dr. J. Grassi, CEA). Signal was captured with a Fujifilm LAS3000 camera and quantified with a Fujifilm software program.
Abs against native membrane-bound PrPc were measured by indirect immunofluorescence [18]. Sera were assayed at three-fold consecutive dilutions. Results are shown as geometric mean fluorescence intensities (MFIs).
Hematoxylin eosin staining was performed on paraffin embedded sections at 5 µm, according to standard procedures. PrPc staining was performed on frozen sections of spleens and brains. Endogenous peroxydases were saturated with a solution of PBS-H2O2 at 0.3% final (Sigma, France) for 30 minutes. PrPc was labelled with a biotinilated SAF-83 Ab (gift of Dr. J. Grassi) at 5 µg/ml final concentration in tween-PBS for 1 hour at room temperature. Revelation was made with a horseradish-peroxidase streptavidin conjugate (GE Healthcare, Pittsburgh, USA) followed by incubation with a solution of diaminobenzidine (Diagnostics Biosystems, Pleasantville, USA).
Infiltrating T lymphocytes were identified on frozen crosswise brain sections as previously described [13]. Ten µm thick sections were air dried, fixed in acetone at 4°C and stained with an anti-CD3 rabbit monoclonal Ab (clone SP7) (Thermo Fisher Scientific, Fremont, USA) at 5 µg/ml. Revelation was achieved with a goat anti rabbit alexa fluor 488 (Invitrogen, Cergy-Pontoise, France).
Microphotographs were taken with an Olympus BX61 microscope equipped with an Olympus DP71 camera. Lens magnification is given in the legends to figures. Section areas were calculated with imagej software (http://rsbweb.nih.gov/ij).
Analyses were performed with GraphPad software (San Diego, CA, USA).
|
10.1371/journal.pgen.1007432 | A conserved role for Syntaxin-1 in pre- and post-commissural midline axonal guidance in fly, chick, and mouse | Axonal growth and guidance rely on correct growth cone responses to guidance cues. Unlike the signaling cascades that link axonal growth to cytoskeletal dynamics, little is known about the crosstalk mechanisms between guidance and membrane dynamics and turnover. Recent studies indicate that whereas axonal attraction requires exocytosis, chemorepulsion relies on endocytosis. Indeed, our own studies have shown that Netrin-1/Deleted in Colorectal Cancer (DCC) signaling triggers exocytosis through the SNARE Syntaxin-1 (STX1). However, limited in vivo evidence is available about the role of SNARE proteins in axonal guidance. To address this issue, here we systematically deleted SNARE genes in three species. We show that loss-of-function of STX1 results in pre- and post-commissural axonal guidance defects in the midline of fly, chick, and mouse embryos. Inactivation of VAMP2, Ti-VAMP, and SNAP25 led to additional abnormalities in axonal guidance. We also confirmed that STX1 loss-of-function results in reduced sensitivity of commissural axons to Slit-2 and Netrin-1. Finally, genetic interaction studies in Drosophila show that STX1 interacts with both the Netrin-1/DCC and Robo/Slit pathways. Our data provide evidence of an evolutionarily conserved role of STX1 and SNARE proteins in midline axonal guidance in vivo, by regulating both pre- and post-commissural guidance mechanisms.
| Syntaxin-1 is a core factor in tethering synaptic vesicles and mediating their fusion to the cell membrane at the synapse. Thus, Syntaxin-1 mediates neurotransmission in the adult nervous system. Here we show that this protein is also involved in axonal guidance in the CNS of vertebrates and invertebrates during the development of the nervous system: our systematic analysis of the phenotypes in the nervous system midline of fly, chick, and mouse embryos mutant for Syntaxin-1 unveils an evolutionarily conserved role for this protein in midline axonal guidance. Further, we also dissect the contribution of other proteins regulating neuronal exocytosis in axonal development. We propose that the coupling of the guidance molecule machinery to proteins that regulate exocytosis is a general mechanism linking chemotropism to axonal growth.
| Axonal growth and guidance are responsible for the correct formation of neural circuits. These processes rely on the tightly regulated response of the growth cone to both diffusible and membrane-bound guidance cues. In response to such cues, several intracellular signaling cascades are activated within the growth cone, leading to directional growth. For instance, the chemoattractant Netrin-1 binds to the receptor Deleted in Colorectal Cancer (DCC) at growth cones, activating several kinases and small GTPases, cyclic nucleotides, and calcium cascades, as well as cytoskeletal rearrangements [1–8].
In contrast, few reports have addressed the cross-talk mechanisms between axonal guidance and membrane dynamics and turnover, other than the fact that growth cones are filled by vesicles and express most SNARE (Soluble NSF Attachment Protein REceptor) and exocyst proteins [9–13]. A growing number of reports using in vitro approaches indicate that axon guidance mechanisms require the participation of SNARE-mediated exocytosis for chemoattraction and endocytosis for repulsion [14–18]. Thus, it has been demonstrated that the vSNARE (vesicular SNARE) VAMP2 is required for L1-mediated chemoattraction [19] and for Sema3A-induced chemorepulsion [17], that the vSNARE Ti-VAMP and the tSNARE (target SNARE) SNAP25 are necessary for neurite outgrowth [20–22], and that Syntaxin-1 (STX1) is required for Netrin-1-mediated attraction of axons and migrating neurons [15,16].
However, the participation of these proteins in neural circuit formation in vivo is still controversial. For instance, mice deficient for the SNAP25 and VAMP2 proteins show virtually no neural circuitry defects but do present a severe alteration of evoked synaptic activity [23–25]. Ti-VAMP-deficient mice display behavioral defects but no alterations in gross brain morphology [26]. STX1A knock-out (KO) mice show only mild cognitive defects and a normal brain structure [27] and axonal defects have not been described in STX1B KO [28]. In a previous study we showed that STX1A is required for the navigation of dorsal commissural in the chick spinal cord [16]. Syntaxin-1 loss-of-function in Drosophila and chick embryos results in motor axonal defects [29].
Drosophila melanogaster displays neural expression of a synaptobrevin (VAMP) gene, namely n-synaptobrevin (n-syb) [30] and a SNAP25 homolog, Snap25 [31]. Mutations in these components of the core SNARE complex give rise to neurotransmitter release phenotypes [32]. A single D. melanogaster STX1 homolog, Syntaxin1A (Syx1A), has been identified that shows strong homology to rodent STX1A [33–35]. Mutations in this gene are homozygous lethal, with severe alleles resulting in early embryonic death. In D. melanogaster, loss of Syx1A abolishes synaptic transmission [33,36], and other secretion phenotypes, such as soft cuticle and undigested yolk, have also been reported [33]. In addition, Syx1A is involved in cell membrane formation during cellularization [37] and in the condensation of the embryonic CNS [33]. In addition, Syx1A has been reported to affect the properties of neuronal membranes and influence membrane dynamics throughout development [38]. However, in D. melanogaster, in contrast to vertebrates, Syx1A has yet to be directly implicated in axonal guidance.
Here we systematically inactivated SNARE genes in three species, Drosophila melanogaster, chick, and mouse, to determine the role of SNARE proteins in CNS midline axonal guidance in vivo. We show the involvement of this protein complex, in particular STX1, in D. melanogaster and chicken axonal guidance, reporting aberrant phenotypes. Furthermore, we provide the first description of abnormal midline axonal defects in mouse embryos double mutant for STX1A and STX1B. Our results point to an evolutionarily conserved mechanism of SNARE complex proteins in midline axonal guidance.
To determine whether Syx1A affects axon guidance at the midline of D. melanogaster embryos, we started by analyzing the CNS phenotypes of Syx1A mutant embryos. For this purpose, we used zygotic mutants, since Syx1A has a maternal contribution and its function is required for proper cellularization [37].
In D. melanogaster, CNS axons of wild-type (wt) embryos are found in a stereotypic ladder-like arrangement. Within each neuromere, two commissures link the two halves of the nervous system. Individual neuromeres are connected by axons running in discrete fascicles in the lateral connectives. An antiserum that specifically recognizes D. melanogaster Syx1A revealed that, like Frazzled (Fra, the D. melanogaster DCC homolog) [39], Syx1A is expressed in developing axons in the embryo from stage 13. At later stages, this isoform is expressed at high levels on commissural and longitudinal axons in the developing CNS (Fig 1A). Embryos of the null Syx1AΔ229 genotype [33] displayed no detectable Syx1A protein in the ventral nerve cord (VNC) when stained with the same antibody (Fig 1B).
To analyze axonal midline phenotypes, we stained embryos with HRP, which marks all axons in the CNS, and anti-Fasciclin II antibodies (anti-FasII), which label lateral fascicles (Fig 1C). At stage 15 or later, FasII identifies three major axonal tracts, which are visible as parallel straight lines: the medial (FasII-m or 1st), intermediate (FasII-i or 2nd) and lateral (FasII-l or 3rd) fascicle. In wt conditions, these FasII-positive axons do not cross the midline. Examination of the commissural and longitudinal pathways labelled by HRP and anti-FasII in the VNC of Syx1A mutants revealed specific defects in diverse axonal pathways. From stage 14 onwards, by analyzing FasII-positive axons, we detected guidance defects in the VNC of 50% of the embryos (n = 38). These defects included aberrant midline crossing, as well as abnormal arrangement of ipsilateral fascicles (Fig 1D, 1E and 1F). Of these embryos, 18% displayed strong defects (Fig 1D and 1F) and 32% weak defects (Fig 1E and 1F, see Materials and Methods for quantification methods and definition of weak and strong phenotypes). FasII-positive fibers never crossed the midline in wt conditions. In contrast, Syx1A embryos displayed clear abnormal crossing of FasII-positive axons, thereby suggesting that longitudinal fibers aberrantly cross the midline in Syx1A mutants (Fig 1G). When HRP staining (a pan-axonal marker) was used, control embryos presented regularly spaced commissures (Fig 1C and 1H). However, embryos with the strong Syx1A phenotype showed collapsed commissures with no clear separation between anterior and posterior ones (Fig 1D and 1H). We analyzed and quantified commissural phenotypes in the VNC of Syx1A embryos displaying phenotypes (weak and strong phenotypes, 50% of embryos). Collapsed commissures (98% Fig 1H, arrows, and 1I) and “fuzzy” commissures were detected in many segments (36%, Fig 1H, asterisk, and 1I), as well as thinning of longitudinal fascicles between segments (57%, Fig 1H, arrowhead, and 1I).
On average, Syx1A mutant embryos displayed more than one defect in VNC axonal guidance per embryo (Fig 1G). Among these defects, the strongest were axonal midline crosses (Fig 1D, short arrow, 13%), commissural thinning (Fig 1H and 1I, 55%), and thinner longitudinal fascicles (Fig 1H and 1I, 32%), which were not detected in controls. The most penetrant phenotypes were fasciculation defects (Fig 1E, long arrow, 58%), but Syx1A mutant embryos also showed fascicle collapse (Fig 1D and 1E, arrowhead, 42%). Variability among these axonal phenotypes between individuals possibly reflects the differential contribution of the maternally deposited gene product. These results show that loss of Syx1A function induces axonal guidance defects in both commissural and longitudinal axons at embryonic stages of fly CNS development.
We next used the same approach to study whether genetic loss of additional components of the SNARE core complex (SNAP25, VAMP2 and Ti-VAMP/VAMP7) also alters the commissural and longitudinal pathways in D. melanogaster. To do so, we examined mutants for nSyb (human VAMP2 ortholog) [40] and Snap25 [41], and analyzed the D. melanogaster ortholog of Ti-VAMP/VAMP7 (Vamp7 source:Flybase). Overall, the nSyb, Snap-25, and Vamp7 phenotypes were weaker than the Syx1A one (Fig 2A–2D). When analyzing FasII positive axons for fasciculation defects, the strongest axonal guidance phenotypes were found in Snap25 mutants, in which 74% embryos displayed fascicle collapse, defasciculation, or both (Fig 2B and 2F, n = 35). 57% of nSyb mutants showed fascicle collapse, defasciculation, or both (Fig 2C and 2F, n = 41). In addition, a small percentage of nSyb embryos (4%, n = 41) showed axonal midline crosses of FasII positive axons (Fig 2C, arrow). Vamp7 mutants exhibited the weakest phenotypes (Fig 2D and 2E n = 31). We could detect stronger phenotypes when all axons were visualized with BP102 antibody. Again, the strongest phenotypes were detected in SNAP25 mutant embryos (Fig 2H and 2K) where 6.9% of embryonic segments showed fuzzy commissures and 27.5% showed thinning of longitudinal fibers (n = 80 segments). nSyb mutants displayed intermediate phenotypes, fuzzy commissures in only 3.1% of segments and thinning of longitudinal fibers in 10.6% (n = 80) (Fig 2I and 2K). In Vamp7 embryos we could only detect thinning of commissures (in 5% of embryonic segments, n = 80) (Fig 2J and 2K). These results indicate that Vamp7, nSyb, and Snap25 can also influence D. melanogaster axonal guidance at the midline, but to a lesser extent than Syx1A.
Next, we used in ovo electroporation of double-stranded RNAs derived from STX1A, SNAP25, VAMP2, and Ti-VAMP to study the role of these genes in dI1 commissural axon guidance in the chicken spinal cord. In untreated controls and in EGFP-expressing control embryos, most commissural axons followed a stereotypic trajectory (Fig 3A and 3B). The vast majority of the dI1 axons crossed the floor plate and turned rostrally along the contralateral floor-plate border. See Methods for details on the quantification method. In contrast, we found that the down-regulation of all SNARE-complex proteins (STX1A, SNAP25, VAMP2, and Ti-VAMP) generated defects in commissural axon guidance, as axons either failed to enter or to cross the floor plate, or failed to turn into the longitudinal axis along the contralateral floor-plate border (Fig 3C–3G). After silencing STX1A aberrant axon pathfinding was found at 36% of the DiI injection sites per embryo (Fig 3C and 3G). We found overall similar percentages of injection sites with aberrant phenotypes when SNAP25, VAMP2 or Ti-VAMP were down-regulated (Fig 3G). When electroporated at embryonic day 3 (E3/HH17-18) most axons reached the floor plate and entered the midline area in all groups (Fig 3H). A detailed analysis of the aberrant phenotypes indicated a significant increase in floor-plate stalling in embryos where STX1A, VAMP2, or Ti-VAMP was silenced (Fig 3I). Similarly, turning of post-crossing axons was affected in all experimental groups (Fig 3J).
In a separate series of experiments, we compared electroporation of dsSTX1A at HH13/14 (E2) with electroporation at HH17/18 (E3) (Fig 3K). These experiments revealed an effect of timing of gene silencing on the severity of the phenotype. Electroporation at E3, when axons are starting to grow in the dorsal spinal cord, resulted in normal axon guidance at 56.4% of all DiI injection sites per embryo. In contrast, electroporation of neural precursors at E2 resulted in normal axon navigation only at 23.8% of all DiI injection sites. Importantly, failure of some axons to enter the floor plate was observed in 3/8 embryos electroporated early but was never observed in the embryos electroporated late. These data support the notion that the silencing of SNARE proteins in the chicken spinal cord leads to various commissural axon guidance defects.
To study the involvement of STX1 in commissural axon guidance in mammals, we generated double KO mice for the two STX1 isoforms, STX1A and STX1B. Double KO embryos died just after birth and displayed strong motor abnormalities, thereby suggesting severe alterations in nervous system organization. To address axonal phenotypes in the midline, we examined the commissural pathway in the spinal cord in E12 embryos [42,43]. In wt embryos, commissural axons stained with TAG-1 antibodies were organized as a narrow axonal bundle extending from the dorsal spinal cord towards the floor plate without invading the motorneuron area (Fig 4A and 4B). In contrast, in STX1A/B (-/-) mice, TAG-1-positive commissural axons were still directed towards the ventrally located floor plate, although their organization differed from that of controls. Instead of forming a narrow bundle, axons in STX1A/B (-/-) mice were clearly defasciculated, invading the entire mantle zone. Individual fibers or bundles invading lateral motorneuron areas were frequently observed (Fig 4C–4F). However, axons appeared to reach the floor plate. To support these observations, we quantified the width of the commissural pathway at three dorso-ventral levels (Fig 4C–4G). At each one, we found a significant difference in the width of commissural axon bundles between wt and double STX1 KO mice.
To confirm these findings, we stained spinal cord sections with Robo3 antibodies, a marker of pre-crossing and post-crossing commissural fibers. In wt embryos, Robo3-stained fibers formed a tight fascicle directed towards the floor plate. Moreover, post-commissural fibers in the ventral spinal cord were heavily stained (S2A and S2B Fig). In contrast, commissural fibers in STX1A/B (-/-) embryos exhibited wider ipsilateral fascicles, often invading the lateral (motorneuron) domains in the ventral spinal cord. Many ipsilateral commissural axons in mutant embryos were tipped with growth cones indicating that they failed to reach the floor plate. Consistent with this observation, the bundles of post-crossing commissural axons (located in the ventral spinal cord) were markedly decreased in STX1A/B null-mutant embryos (S2C and S2D Fig). Taken together, our observations indicate that the lack of STX1 results in aberrant commissural axon growth on both ipsilateral and contralateral sides.
To confirm the above observations, we also examined commissural axon navigation in open-book preparations of E12 spinal cords (Fig 5). Comparable to our observation in the chicken spinal cord, we found aberrant navigation of dI1 axons at the floor plate. In the double STX1 KO embryos, almost no axons were found to cross the midline and to turn properly along the contralateral floor-plate border (Fig 5D and 5E). Axons mainly failed to reach the contralateral border of the floor plate. In embryos lacking STX1A but expressing STX1B from one or two wt alleles, axon guidance was still compromised, exhibiting intermediate phenotypes (Fig 5C, 5F and 5G). A quantitative analysis showed a pronounced decrease in normal axonal trajectories in mutant compared to wt embryos (Fig 5F). In wild-type mice we found normal axon trajectories at 75.6±6.3% of the DiI injection sites per embryo. In contrast, in mice lacking STX1A but expressing one or two wild-type allele(s) of STX1B normal trajectories were only seen at 10.2±5.4% or 21.7±12.3% of the DiI injection sites, respectively. In double KO mutants axon navigation was affected even more strongly, as normal trajectories were only seen at 1% of the injection sites (Fig 5F and 5G). A detailed analysis of the different guidance defects revealed a problem in floor-plate stalling in all mutants in addition to a failure to turn rostrally along the contralateral floor-plate border. Failure to enter the floor plate was only found in mutants but never in wild-type mice (Fig 5G). Embryos having at least one wt allele of either STX1A or STX1B exhibited weaker axonal defects than the double KO mice (Fig 5G).
To confirm the expression of SNARE proteins in spinal cord commissural axons we performed immunocytochemical analyses in neuronal cultures. Dissociated mouse commissural neurons were identified with two antibody markers: DCC and Robo3 antibodies. Cultures were co-immunostained for different SNARE proteins including STX1A, VAMP2, SNAP25 and Ti-VAMP. Confocal images revealed that commissural growth cones, identified by the expression of DCC and Robo3, co-expressed all the studied SNARE proteins (Fig 6). We also observed variable degrees of colocalization between DCC/Robo3 receptors and SNARE proteins. These results indicate that embryonic commissural axons express the SNARE proteins analyzed in the present study.
We next addressed whether STX1A is required for Slit-2 and Netrin-1 actions. Tissue explants of chick dorsal commissural neurons dissected from control embryos and grown on laminin are known to extend neurites readily in the absence of Slit-2 (Fig 7A). In the presence of Slit-2, neurite length was strongly reduced (Fig 7B). In contrast, neurite growth from explants taken from embryos electroporated with dsRNA derived from STX1A did not differ in the absence or presence of Slit-2 (Fig 7C, 7D and 7I), indicating that the absence of STX1A results in a markedly reduced Slit-2 response of chick commissural axons.
We have previously shown that the blockade of STX1 with Botulinum Toxin C1 reduces Netrin-1 induced chemoattraction in mouse spinal cord explants [16]. To confirm this finding, we co-cultured in collagen gels dorsal spinal cord explants from STX1 KO embryos with Netrin-1 expressing cell aggregates. In comparison with explants cultured with control cells (exhibiting radial growth, Fig 7E), wild-type explants confronted to aggregates of Netrin-1 expressing cells showed strong chemoattraction (Fig 7F and 7J). In contrast, Netrin-1 induced chemoattraction was absent in spinal cord explants derived from double STX1A/B KO embryos (Fig 7G, 7H and 7J).
Our previous studies have shown that exposition to Netrin-1 increases DCC surface expression in growth cones and that such increase is not diminished after blockade of STX1 [15,16]. Here we performed experiments to determine whether STX1 regulates Robo3 surface expression in the presence of Netrin-1. Mouse embryonic commissural neurons were cultured and stained for the differential immunolabeling of surface and intracellular Robo3 protein pools (S3 Fig). The data show that in control commissural growth cones, Robo3 trafficking and distribution is not substantially altered by Netrin-1 incubation. In contrast, STX1A/B knock-out growth cones incubated with Netrin-1 exhibited a decrease in surface Robo3 signals (S3D and S3E Fig). These findings suggest that the inactivation of STX1 alters Robo3 trafficking and/or surface expression, probably by increasing Robo3 internalization.
During our analysis of midline guidance defects in Syx1A mutant embryos, we detected that the arrangement of the longitudinal fascicles was shifted towards the CNS midline compared to controls (Fig 8A and 8B; n = 20). In order to clarify whether the FasII–positive axonal fascicles were closer to the midline in Syx1A mutants, we quantified these distances at embryonic stage 16 and compared them to wt, frazzled (fra) and robo2 mutants (Fig 8C–8E). Overall, Syx1A and robo2 mutations induce a shift of FasII-positive fibres (both FasII-m and FasII-i) towards the midline. Regarding both FasII-m and FasII-I, the opposite occurs in fra mutant embryos, thereby suggesting that Syx1A interferes with a midline repression pathway (Fig 8A–8E) [44].
In order to address whether Syx1A is involved in the Slit/Robo pathway, we generated double mutants for Syx1A and robo2 and examined the positioning of the longitudinal tracts in robo2;Syx1A embryos. As shown in Fig 8F–8J, the longitudinal fascicle distance to the midline decreased in the double mutant embryos (Fig 8I and 8P), where the strongest phenotypes showed a complete collapse of the tracts in the midline (Fig 8I, asterisk), as reported for slit (sli) mutants [45]. This result indicates that Syx1A and robo2 interact genetically. In molecular terms, our above results suggest that Syx1A is involved in axonal repulsion in the midline.
In vertebrates, it has been reported that Netrin1/DCC axonal guidance is coupled to exocytosis through STX1 [15]. Therefore, we examined whether differences in Syx1A levels could affect the Drosophila fra axonal guidance phenotypes. We found that the fra mutant phenotype was significantly aggravated by removing one copy of Syx1A (Fig 8K–8O), thereby suggesting that Syx1A levels severely interfere with the Netrin/Fra guidance pathway.
Here we used loss-of-function models in three species (fly, chick and mouse embryos) to examine the role of SNARE proteins in midline crossing of commissural axons. We report aberrant axonal phenotypes in the D. melanogaster midline and in the chicken spinal cord. Furthermore, we generated double mutant STX1A/B mice that show abnormal commissural phenotypes in the murine spinal cord. These findings indicate an evolutionarily conserved role of SNARE complex proteins in midline axonal guidance.
The guidance of commissural axons at the midline of the spinal cord is a complex process regulated by many molecules. For instance, the interaction between Netrin-1 and DCC attracts pre-crossing axons toward the floor plate [42,43]. Axon growth towards the floor plate is also altered in mice lacking VEGF/Flk1 signaling [46]. Midline crossing was shown to depend on attractive (Axonin-1/TAG1 with NrCAM; [47]) and repulsive interactions (Slit/Robo; [48]; Sema3B/NrCAM/PlexinA1/Neuropilin; [49] [50] [51,52]). The rostral turn of post-crossing commissural axons was shown to depend on morphogen gradients formed by Wnts [53,54] and Shh [55,56], although the latter is also involved in attraction of commissural axons toward the ventral midline in parallel to Netrin-1 [57,58]. In addition, an interaction between axonal Semaphorin6B and its ligand PlexinA2 [59] and interactions between the SynCAMs [60,61] are also involved in the turning of post-crossing commissural axons into the longitudinal axis.
Our results in D. melanogaster indicate that the SNARE complex is involved in midline axonal guidance. We show that Ti-Vamp/Vamp7, nSyb, and Snap25 influence axonal guidance at the midline, but to a lesser extent than Syx1A. In general, these phenotypes may be incomplete Drosophila KOs, because in this fly there is a maternal contribution for all these components. Additionally, it is known that there is a high level of redundancy in D. melanogaster. In this regard, it has been reported that Syb can replace nSyb and Snap24 can replace Snap25 [40,41]. Regarding the Syx1A phenotypes, we can conclude that Syx1A loss of function induces commissural and ipsilateral axonal phenotypes. Interestingly, some of the phenotypes observed in Syx1A mutants resemble the loss of function phenotypes of the Robo/Slit pathway [62,63]. We detected midline crosses of FasII-positive axons and a strong genetic interaction with robo2. These findings suggest that Syx1A is involved in repulsive midline guidance in Drosophila. In addition, most of the ipsilateral phenotypes detected are also common in frazzled (fra, the D. melanogaster homolog of DCC), and fra phenotypes are aggravated by decreased Syx1A levels. Furthermore, similar ipsilateral phenotypes to Syx1A are observed in netrin mutant ventral nerve cords [44], as well as in embryos mutant for Heparan Sulfate Proteoglycans (HSPGs) [64] and Hh [65], thereby suggesting that Syx1A may participate in various guidance pathways in D. melanogaster not just in axonal guidance but also in fasciculation/defasciculation events. Recent studies suggest that whereas axonal attraction requires exocytosis, chemorepulsion relies on endocytosis [14,18,66,67]. In addition, SNARE proteins have been reported to be involved in both exo- and endocytosis [68]. Our genetic interaction experiments support the notion that Syx1A participates in both Netrin/Fra attraction and Slit/Robo repulsion.
The data obtained in chicken reinforce the idea that the silencing of SNARE proteins induces an overall increase in various commissural guidance defects, with the silencing of STX1A leading to defects in all the guidance steps analyzed. Silencing STX1A at E2, on day before neurons start to extend axons rather than E3, just before they start to extend axons, also resulted in axons failing to enter the midline area. This phenotype resembled previously reported findings on the role of STX1A in DCC-mediated guidance of pre-crossing commissural axons towards the midline [16].
Based on these results, we conclude that SNARE proteins make a crucial contribution to the navigation of chick commissural axons at the floor plate. SNARE proteins are involved in both attractive [16] and repulsive (this study) decision-making steps in axon guidance, as STX1A, VAMP2 and Ti-VAMP silencing leads to a substantial increase in pre-crossing and post-crossing phenotypes. These phenotypes are in agreement with those reported in previous studies [15,16], confirming the involvement of STX1 in the regulation of chemoattractive guidance pathways for commissural axons.
The phenotypes seen in the chicken spinal cord resemble those observed after silencing Calsyntenin-1 and RabGDI in dI1 neurons [69]. In the absence of Calsyntenin-1 and RabGDI, Robo1 is not transported to the growth cone surface, resulting in axonal stalling in the floor plate. Furthermore, silencing Calsyntenin-1, but not RabGDI, prevents the expression of the Wnt receptor Frizzled-3 on the growth cone surface, leading to failure of post-crossing axons to turn rostrally in response to the Wnt gradient [54]. These findings are consistent with the analyses of commissural axon navigation at the midline in STX1 KO mice. In double mutants, the absence of STX1A and STX1B proteins prevented midline crossing.
In Netrin-1 [70,71] and DCC KO [44,70] mice, pre-commissural fibers display strong phenotypes, including defasciculation, aberrant trajectories, and a complete failure to cross the floor plate. Weaker phenotypes have been observed in mouse mutants for other genes involved in commissural axonal guidance, including Sonic Hedgehog and the VEGF/FlK1 pathways [46,57]. The phenotype described here in STX1A/B mutant mice is reminiscent of, that in Netrin-1 and DCC KO mice, with fewer axons reaching the floor plate, lateral invasion of motorneuron territories, and fewer post-crossing commissural axons, although the phenotype is much weaker and many fibers still reach the ipsilateral floor-plate border. A likely explanation could be that although STX1A/B might be required for correct sensing of Netrin1/DCC guidance [15,16], compensatory mechanisms may derive from the possible lack of STX1A/B requirement in other commissural attractive pathways (e.g., SHH or VEGF/Flk1). Our findings indicate that STX1A/B not only affects pre-crossing axons but strongly affects midline crossing and post-commissural axonal guidance. This conclusion is supported by our findings in Drosophila, which revealed a genetic interaction between Syx1A and the Robo pathway, as well as by the present in vitro experiments showing that STX1 is required for both Netrin-1 attraction and Slit-2 repulsion (Fig 7).
Our previous studies showed that the lack of STX1 did not affect DCC surface expression induced by Netrin-1 [15]. In the present study we show that the lack of STX1 results in a decreased surface expression of the Robo3 receptor in response to Netrin-1. It is therefore possible that the lack of STX1 may alter the surface expression of other Robo family members (eg Robo1) and Frizzled-3, the receptors used by post-crossing axons. This idea is supported by our STX1 loss-of-function data, showing that the lack of this SNARE protein abolished the responsiveness of post-crossing commissural axons to Slit, both in vitro (explant experiments) and in vivo (chick and mouse analyses). We thus propose that, in contrast to DCC expression, the effect of STX1 loss-of-function on post-crossing axons may be explained in part by preventing the surface expression of axon guidance receptors required for midline crossing and post-crossing navigation.
Robo3 is an atypical Robo receptor that has been proposed recently to potentiate DCC-mediated attraction to Netrin-1, but without binding Slits [72,73]. Thus, the observation that Netrin-1 in STX1-deficient growth cones results in decreased Robo3 surface expression, supports the notion of a possible contribution of Robo3 membrane downregulation to explain the reduced Netrin-1 chemoattraction found in STX1 loss-of-function models.
Many guidance molecules and their receptors involved in axonal guidance are conserved between vertebrates and invertebrates (e.g. Robo, Hh, and Netrins) [57,65,74]. Our systematic analysis of the phenotypes at the CNS midline of fly, chick, and mouse embryos mutant for STX1 unveils an evolutionarily conserved role for STX1 in midline axonal guidance. Overall, the ipsilateral phenotypes reported are consistent with the participation of STX1 in Netrin-1-dependent axonal guidance, as proposed previously [15,16]. In addition, here we describe post-commissural phenotypes that are reminiscent of those found in Robo, NrCAM and VEGF loss-of-function models [46,47,75,76], thereby suggesting that STX1 underlies various signaling pathways. Furthermore, the phenotypes described herein for other SNARE proteins point to the participation of SNAP25, VAMP2 and Ti-VAMP in midline axonal guidance, although the precise implication and relevance of individual SNARE proteins to specific axonal guidance signaling complexes remains to be determined. We propose that the coupling of the guidance receptor cell machinery to proteins that regulate exocytosis is a general and conserved mechanism linking chemotropic signaling to membrane trafficking [29].
The following stocks are described in FlyBase (http://flybase.org): Syx1AΔ229, SNAP-25 (Df(3L)1-16), nSybd02894, Ti-VAMP (P[CG1599G7738]), fra3 and robo2. All alleles used are genetic nulls. Wild-type control is yw. D. melanogaster stocks and crosses were kept under standard conditions at 25ºC.
D. melanogaster embryos were staged as described by Campos-Ortega and Hartenstein [77] and stained following standard protocols. For immunostaining, embryos were fixed in 4% formaldehyde for 20 min. We used antibodies that recognize FasII (mAb1D4, DSHB), βGal (Promega), Affinity-Purified Anti-HRP TRITC (Jackson immunoResearch), and mAbBP102 (DSHB). We used biotinylated HRP (Amersham) or non-biotinylated HRP (GE Healthcare), Alexa488, Alexa-555 and Alexa-647, Cy2, Cy3 and Cy5 conjugated secondary antibodies (Jackson ImmunoResearch). For HRP histochemistry, the signal was amplified using the Vectastain-ABC kit (Vector Laboratories) when required. In addition, the signal for the DAB reaction was intensified with NiCl2, except for double stainings, where it was omitted from one of the reactions. DIC photographs were taken using a Nikon Eclipse 80i microscope. Fluorescent images were obtained with a confocal microscope (Leica TCS-SPE-AOBS and TCS-SP2-AOBS systems) and processed using Fiji [78] and Adobe Photoshop. Images are maximum projections of confocal Z-sections.
Fertilized chicken eggs were obtained from a local supplier, and embryos were staged following Hamburger and Hamilton [79]. Electroporations were performed either at HH13-14 (E2) or at HH17-18 (E3). Images were acquired using a confocal spinning disk microscope (Olympus BX61).
Mouse embryos aged 11 or 12 days (E11 and E12 respectively) were used for the experiments. To obtain the tissue samples, pregnant female mice were sacrificed by cervical dislocation. A small portion of the tail was cut for further genotyping. Embryos were then fixed with 4% paraformaldehyde, and spinal cord sections were immunostained with mouse anti—TAG1 (mouse, Hybridoma Bank), followed by α-mouse IgM biotinylated (goat, Chemicon) and by Streptavidine Fluorescent FITC (490 nm, GE Healthcare). Sections were routinely stained with bisbenzimide. Confocal images were acquired using a Leica TCS SP5 microscope with 20x and 40x oil-immersion objectives. An average of 4–6 embryos per genotype and 10–15 slices per embryo were analyzed. For the quantification of TAG1 staining we used 4 KO and 3 WT embryos (32 sections from KO embryos and 8 from WT embryos). Data are presented as the means ± SEM. Statistical significance was determined using two-tailed Student’s t-test. Differences were considered significant at p<0.05. Embryonic mouse spinal cord sections were also stained with goat α-Robo3 antibodies (1:100, RD Systems) followed by incubation with α-goat Alexa-488 secondary antibodies.
Experiments with chicken embryos were approved by the Cantonal Veterinary Office Zurich. All the experiments using animals were performed in accordance with the European Community Council directive and the National Institute of Health guidelines for the care and use of laboratory animals. Experiments were also approved by the local ethical committees.
The percentage of axons displaying abnormal FasII phenotypes was quantified via anti-FasII stainings by an observer with no knowledge of the genotype. Defects were categorized as “midline crossing”, “defasciculation” or “fiber collapse”. Due to the variability of fasciculations and occasional fiber collapses in wt embryos, we allowed for up to 2 of these defects to be considered “normal” or “background” and therefore set our “zero” defects at this control level. From 2 to 5 defects, we considered that embryos had a weak phenotype, while with more than 6 defects they were considered to have a strong phenotype.
The percentage of axons displaying abnormal midline crossing phenotypes was quantified via anti-HRP or BP102 stainings by an observer with no knowledge of the genotype. Defects were categorized as “fuzzy commissures”, “collapsed commissures” or “thinning of longitudinals”.
Fascicle distances to the midline were quantified via anti-FasII stainings, where the distance between fascicles was measures. Distances were measured in arbitrary units, between media, intermediate and lateral fascicles and these values were normalized to the length of the axonal tracks in the VNC.
All data were analyzed statistically, SEMs were calculated and statistic significance assessed by Student’s t-test.
The analysis of commissural axon trajectories was performed as described previously [47,55]. In brief, fertilized eggs were windowed on the second or third day of incubation. Extra-embryonic membranes were removed to access the spinal cord in ovo. A plasmid encoding EGFP (20 ng/μl) and the dsRNA (250 ng/μl) derived from STX-1A, SNAP-25, VAMP2, or Ti-VAMP in PBS were injected into the central canal using glass capillaries. dsRNA was produced by in vitro transcription as described previously [80]. For the production of dsRNA derived from STX1A, SNAP25, VAMP2, or Ti-VAMP, we used ESTs obtained from Geneservices (now Source BioScience) [55]. For visualization and control of injection quality and quantity, 0.04% Trypan blue was added. For electroporation, we used platinum electrodes connected to a BTX square wave electroporator. Electrodes were positioned parallel to the longitudinal axis of the lumbosacral spinal cord of the chicken embryo, as detailed previously [55,81]. Five pulses of 26 V and 50 ms duration with a 1-s interpulse interval were applied. After electroporation, eggs were sealed with Scotch tape and incubated for another 2 or 3 days. The gene silencing specificity was verified by using two independent and non-overlapping sequences for the generation of the dsRNAs, when available. dsRNA sequences used in the present study were shown to downregulate the targeted proteins by 25%-66% [29]. Because in these conditions only about 60% of the cells are efficiently transfected, small decreases in the total amount of protein can still be indicative of efficient knock-down by dsRNA transfection.
For the analysis of commissural axon guidance, embryos were sacrificed between stages 25 and 26 [79]. The spinal cord was removed from the embryo, opened at the roof plate (“open-book” preparation), and fixed in 4% paraformaldehyde for 30–60 min.The trajectories of dI1 commissural axons at the lumbosacral level of the spinal cord were visualized by the application of the lipophilic dye Fast DiI (dissolved at 5 mg/ml in ethanol; Invitrogen) to the cell bodies. Care was taken to exclusively label the dorsal-most population of commissural neurons (dI1 neurons) to avoid confusion with more ventral populations that have distinct pathfinding behavior. Only DiI injections sites that were in the appropriate location in the dorsal-most part of the spinal cord and within the region expressing fluorescent protein were included in the analysis.
Quantification of the percentage of injections sites with axons displaying abnormal phenotypes was done by a person blind to the experimental condition. The injection sites were classified as ‘normal’ when the axons crossed the floor plate and turned rostrally along the contralateral floor-plate border. When at least 50% of the labeled fibers failed to reach the contralateral border of the floor plate, the DiI injection site was judged as ‘floor-plate stalling’, when at least 50% of the fibers reaching the floor-plate exit site failed to turn into the longitudinal axis, the DiI injection site was considered to exhibit ‘no turning’. Because stalling of all or most axons in the floor plate prevented the analysis of the turning phenotype, the quantification did not include a separate analysis of this phenotype between the different groups. In Fig 3G–3J, the average percentages of DiI injection sites per embryo with the respective aberrant phenotypes are shown.
Mutant STX1B mice were generated using a gene-trapping technique [82]. Mice (strain C57BL/6 from Charles River Laboratories) were cloned from an ES cell line (clone OST68841; Texas Institute for Genomic Medicine, TIGM). The ES cell clone contained an insertion of the Omnibank Vector VICTR24 in the first exon of the STX1B gene identified from the TIGM gene trap database and was microinjected into C57BL/6 host blastocysts to generate germline chimeras using standard procedures.
The retroviral OmniBank Vector VICTR24 (S1A Fig) contained a splice acceptor sequence (SA) followed by a 5’ selectable marker β-GEO, a functional fusion between the β-galactosidase and neomycin resistance genes, for identification of successful gene trap events followed by a polyadenylation signal (pA). Insertion of the retroviral vector into STX1B led to the splicing of the endogenous upstream exons into this cassette to produce a fusion transcript that was used to generate a sequence tag (OST) of the trapped gene by 3′ RACE [82]. More information on the gene trap strategies can be obtained from the TIGM website (http://www.tigm.org/). Chimeric mice were born three weeks later. Male chimeras where then mated with wt C57BL/6 to obtain germline transmission. We obtained four founders and used them to establish the colony. The derived F1 mice were screened by PCR. Genotyping of tail DNA was accomplished using PCR with forward and backward primers for the wt locus (5’-AAT CCG AAC AGA CTG AGA TAC ATT -3’; 5’-aGA GTT GGG CGG AAG GTA CAA GAG -3’) and two primers for the LTR mutant locus (5’-ATA AAC CCT CTT GCA GTT GCA TC-3’; 5’-AAA TGG CGT TAC TTA AGC TAG CTT GC-3’). A 330-bp band was amplified for homozygous wt mice, a 270-bp and 200-bp band for homozygous mutant mice, and the three bands for heterozygous mice (S1A and S1B Fig). Western blot analyses demonstrated absence of STX1B protein in STX1B mutants (S1C Fig).
STX1A mutant mice were a kind gift from Thomas Sudhof [83]. STX1A and STX1B mutant mice were mated to produce double heterozygous mutant mice. The double heterozygous mutant mice were then mated with each other, and the genotypes of their offspring were determined by PCR.
Open-book preparations of mouse spinal cords were essentially done as described above for chicken embryos. Embryos were collected and dissected at E12. Spinal cords were removed, opened at the roof plate and pinned down in a Sylgard dish in 4% PFA for 20–50 minutes. Commissural axons were traced with Fast DiI (5 mg/ml in ethanol) by incubating the open-book preparations in PBS for at least 3 days before mounting in PBS between two 24x24 mm coverslips sealed with vacuum grease.
Animal experimentation was conducted according to the European and National (Spanish) guidelines. The experimental protocol was approved by the local University Committee (CEEA-UB, Comitè Ètic d´Experimentació Animal de la Universitat de Barcelona) and by the Catalan Government (Generalitat de Catalunya, Departament de Territori I Sostenibilitat) with the approval number #9431.
Embryonic brains were lysed in hypotonic buffer (150 mM NaCl, 50 mM Tris pH 7.2, 2 mM EDTA, 1% Triton X-100, and protease inhibitors (Complete, Mini Protease Inhibitor Cocktail Tablets, Roche)).
SDS-sample buffer was added to the lysates, and the proteins were analyzed by SDS-PAGE and Western blot. Proteins were transferred onto nitrocellulose membranes, which were blocked with 5% non-fat dry milk in Tris-HCl buffered saline (TBS) containing 0.1% Tween 20, and incubated overnight at 4°C with mouse anti-STX1 (HPC-1 clone 1:500–1000, Sigma) and anti-actin (1:10000, Millipore) antibodies. After incubation with secondary antibodies, blots were developed following the ECL method (Amersham Pharmacia Biotech).
Spinal cord dorsal neurons were isolated from E11 mouse embryos. After dissection of the alar plate, tissue was treated with Trypsin 1x for 3 minutes at 37°C and with DNase and Fetal Bovine Serum at 37°C for 10 minutes. 50000 cells were seeded on 200 mm2 wells, previously treated with 0.5 mg/mL poly-D-lysin coverslips. Cells were maintained in vitro for 16–24 hours in Neurobasal medium with B27 1x, Glutamax 1x, Penicillin/Streptomycin 1x and Fetal Bovine Serum 10%. Neurons were fixed in 4% paraformaldehyde, and incubated with mouse anti-DCC (BD Pharmigen) and goat anti-Robo3 (RD Systems) antibodies (1:100). Mouse anti-STX1 (Sigma), mouse anti-VAMP2 (Synaptic Systems), mouse anti-SNAP25 (Covance); mouse anti-Ti-VAMP (Abcam) were incubated in blocking buffer to a final concentration of 1:100. Double immunodetection for DCC and SNARE proteins required incubation in blocking buffer containing anti-mouse IgG Fab specific antibodies. Alexa 488 anti-mouse and anti-goat secondary antibodies, and Alexa 568 anti-goat antibodies were incubated in blocking buffer (1:100). Neurons were imaged in a Leica TCS SP5 confocal microscope using a 63x oil-immersion objective.
To analyse Robo3 receptor surface expression, dorsal spinal cord neurons from wild-type and STX1A/B KO embryos were prepared as above. We followed the protocol described in [15]. Briefly, cultures were incubated with either 300 ng/ml of Netrin-1 (+Netrin-1) or with BSA (-Netrin-1) for 30 min at 37°C. After fixation, cultures were blocked with 10% of Normal Horse Serum (NHS) in PBS for 2 hours and incubated without detergent with primary goat anti-Robo3 (RD Systems, Robo3 1:100). Afterwards, cells were washed with PBS and incubated with secondary antibodies (Alexa Fluor 488 anti-goat, 1:50) in excess to block all the primary antibody epitopes. After several washes in PBS, cells were blocked again with 10% NGS and incubated with goat anti-Robo3 primary antibody, in the presence of 0,3% Triton X-100. Cells were then washed and incubated with Alexa Fluor 568 anti-goat (both at 1:50, Jackson). Cells were washed, stained with DAPI and mounted in Mowiol. Cells were imaged in a Leica TCS SP5 microscope using a 63x oil-immersion objective. Z stacks of 8–12 confocal planes were acquired and images were processed with Fiji software. 20 growth cones per condition were analyzed with GraphPad software to quantify surface and inside receptor expression. Statistics were calculated with a two-way ANOVA and a Tukey’s multiple comparison post-test, p<0.05.
Dorsal (alar plate) spinal cord explants were dissected from E11 mouse embryos. Tissue explants were co-cultured as described [16], along with cell aggregates of HEK293T cells stably transfected with a pCEP4-rNetrin-1c-myc construct or with control HEK293T cells. Explants and cell aggregates were embedded in a collagen matrix and maintained in vitro for 48h in Neurobasal supplemented media (Penecillin/Streptomycin 1x, Glutamax 1x and B27 1x). Cultures were fixed with 4% paraformaldehyde and immunolabeled with mouse anti-βIII-tubulin (1:1000, Covance) in blocking buffer. Images were acquired with an Eclipse E1000 microscope using a 10x objective. Axon elongation was quantified by calculating the area occupied by the axons in both the proximal and distal quadrants. Data was statistically analysed as above.
Explants of commissural neurons were dissected from untreated and experimental chicken embryos at HH25. Explants were cultured in serum-free DMEM medium with GlutaMax, 5 mg/ml Albumax, N3 and 1 mM sodium pyruvate. Eight-well Lab-Tek dishes (Nunc) were coated with polylysine (20 μg/ml) and laminin (10 μg/ml). After 24 h control medium or medium containing Slit-2 (100 ng/ml; R&D Systems) was added to the explants. After an additional 18 h, cultures were fixed in 4% PFA and the average neurite length of each explant was measured using the cellSens program (Olympus).
|
10.1371/journal.ppat.1002304 | Discovery of an Ebolavirus-Like Filovirus in Europe | Filoviruses, amongst the most lethal of primate pathogens, have only been reported as natural infections in sub-Saharan Africa and the Philippines. Infections of bats with the ebolaviruses and marburgviruses do not appear to be associated with disease. Here we report identification in dead insectivorous bats of a genetically distinct filovirus, provisionally named Lloviu virus, after the site of detection, Cueva del Lloviu, in Spain.
| A novel filovirus, provisionally named Lloviu virus (LLOV), was detected during the investigation of Miniopterus schreibersii die-offs in Cueva del Lloviu in southern Europe. LLOV is genetically distinct from other marburgviruses and ebolaviruses and is the first filovirus detected in Europe that was not imported from an endemic area in Africa. Filoviruses, amongst the most lethal of primate pathogens, have only been reported as natural infections in sub-Saharan Africa and the Philippines. Infections of bats with the ebolaviruses and marburgviruses do not appear to be associated with disease. Here we report identification of genetically distinct filovirus in dead insectivorous bats in caves in Spain.
| Filoviruses cause lethal hemorrhagic fever in humans and nonhuman primates. The family Filoviridae includes two genera: Marburgvirus, comprising various strains of the Lake Victoria marburgvirus (MARV); and Ebolavirus (EBOVs), comprising four species including Sudan ebolavirus (SEBOV), Zaire ebolavirus (ZEBOV), Ivory Coast ebolavirus (CIEBOV), and Reston ebolavirus (REBOV); and a tentative species Bundibugyo ebolavirus (BEBOV) [1]. MARV was discovered in 1967 in Marburg, Germany during an outbreak in laboratory staff exposed to tissues from monkeys imported from Uganda. ZEBOV was discovered in 1976 in Yambuku, Zaire during a 312-person outbreak associated with 90% mortality. With the exception of REBOV, that appears to be pathogenic in nonhuman primates but not in humans and is endemic in the Philippines, all known filoviruses are pathogenic in primates including humans and are endemic in Africa [2]. Bats are implicated as reservoirs and vectors for transmission of filoviruses in Africa [3]. ZEBOV sequences have been found in fruit bats (Hypsignathus monstrosus, Epomops franqueti and Myonycteris torquata) [4], [5]. MARV sequences have been found in fruit (Rousettus aegyptiacus) and insectivorous (Rhinolophus eloquens and Miniopterus inflatus) bats [6], [7]. Bats naturally or experimentally infected with ZEBOV or MARV are healthy and shed virus in their feces for up to 3 weeks [4], [5], [7].
In 2002, colonies of Schreiber's bats (Miniopterus schreibersii), sustained massive die-offs in caves in France, Spain and Portugal [8]. M. schreibersii, family Vespertilionidae, comprises at least four geographically discrete lineages distributed in Oceania, southern Europe, southern Africa, and southeast Asia [9]. Here we report the discovery of a novel ebolavirus-like filovirus in bats from Europe.
Bat carcasses from Cueva del Lloviu, Asturias, Spain (5° 32′ 8.1′ ´ N and 43° 30′ 5.6′ ´ W) were collected for anatomical, microbiological and toxicological analyses. Although no gross pathology was apparent, microscopy of internal organs revealed interstitial lung infiltrates comprised of lymphocytes and macrophages, and depletion of lymphocytes and lymphoid follicles in spleen (Figure 1). These findings were consistent with viral pneumonia; hence, nucleic acid from lung and spleen were analyzed by consensus polymerase chain reaction (PCR) for the presence of a broad range of viral agents including lyssa-, paramyxo-, henipa-, corona-, herpes- and filoviruses. Filovirus sequences were detected in extracts from lung, liver, rectal swabs or spleen of 5 animals. Pairwise distance analysis of the 186 nucleotide product showed highest similarity with ZEBOV (73.7%). A sensitive real time PCR assay established to quantitate viral burden confirmed the presence of filoviral sequences in the original 5 animals and from an additional 15 with similar pathology collected from the same cave (Table 1). A liver sample with the highest viral load by PCR (4.0×106 genome copies/gr) was selected for high-throughput sequencing yielding 225,758 reads that represented 12.1 kilobases of viral sequence. Gaps between fragments and genomic termini were completed by specific PCR and rapid amplification of cDNA ends (RACE) to obtain a nearly complete genome.
Reports of bat die-offs in additional caves prompted analysis of a second set of samples from caves in Cantabria, Spain, wherein many dead M. schreibersii were observed. Throat and rectal swabs, brain, lung and liver were collected from five dead M. schreibersii, and nine dead Myotis myotis. Whereas filovirus sequences were detected by real time PCR in all M. schreibersii samples, no filovirus sequences were found in the M. myotis (Table 1). Real time PCR analysis of throat swabs and stool samples from 1,295 healthy bats representing 29 different bat species (including 45 healthy M. schreibersii from Lloviu cave collected after the die-offs) collected in several geographic locations in Spain revealed no evidence of filovirus infection (Figure S1).
Sequencing of regions of the L and NP genes of the original Lloviu Cave bat samples resulted in nearly identical sequences to the prototype sequence; a similar lack of variation was observed within each lineage of MARV in fruit bat reservoirs in the Kitaka cave, Uganda, although in that instance two clearly differentiated lineages were observed [7].
Consistent with the genomic organization characteristic of filoviruses, Lloviu virus (LLOV), named for the cave in which it first was found, has a 19 kb negative sense, single stranded RNA genome that contains seven open reading frames (ORF) (GenBank Accession number JF828358). However, LLOV differs from other filoviruses in transcriptional features. Analysis of conserved transcriptional initiation and termination sites suggests that the seven LLOV ORFs are encoded by six mRNA transcripts, one of which is dicistronic and contains both the VP24 and the L ORF (Figure 2). Additionally, although the LLOV termination signal is identical to ebolaviruses, the LLOV initiation signal is unique (3′-CUUCUU(A/G)UAAUU-5′). Several attempts by RACE to obtain complete genomic sequence were unsuccessful. By analogy to other filoviruses we assume that up to 700 nt may be missing at the 5′ terminus of the genome. This assumption is based on the observation that all known negative-strand RNA viruses have complementary termini and that length of noncoding sequences at the termini of filoviruses do not exceed 700 nt.
LLOV sequence was analyzed for similarities to EBOVs and MARV. In EBOVs a C-terminal basic amino acid motif in VP35 mediates type I interferon antagonism by binding to double-stranded RNA and inhibiting RIG-I signaling. This domain is conserved in LLOV VP35 (Figure S2). In non-segmented, negative strand RNA viruses, matrix proteins are not only key structural components of the virions, but also play important roles in the maturation and cellular egress steps of the viral life cycles. Short amino-acid sequences, termed late-budding motifs or L domains, are crucial for these events. The matrix protein in EBOVs, encoded by VP40, has overlapping P(T/S)AP and PPXY late-budding motifs at the N-terminus [10], [11] and YXXL late-budding motifs in the C-terminus. MARV VP40s contains only PPXY motifs. LLOV contains only a PPXY motif in the N-terminal domain of the VP40; hence, in this aspect, it is more similar to MARV than to EBOVs. The filovirus GP2 has an immunosuppressive motif [12], [13] (Figure S3); this motif is highly conserved in LLOV. EBOV VP24 interacts with the KPNα proteins that mediate PY-STAT1 nuclear accumulation [14]. Two domains of VP24 are required for inhibition of IFN-β-induced gene expression and PY-STAT1 nuclear accumulation (region 36–45 and 142–146) [15]. LLOV VP24 ORF has significant homology to EBOV VP24s; however, interaction domains are not well conserved (Figure S4, shaded areas).
Phylogenetic analysis of conserved domain III of the RNA-dependent RNA polymerase demonstrates that LLOV belongs to the Filoviridae and may represent a complex of viruses related to all EBOVs (Figure 3A). Phylogenetic analysis of complete genome sequences (∼21,800 nucleotides) confirmed that LLOV is a distinct genetic lineage that originates after MARV (Figure 3B). Bayesian and ML phylogenetic analyses using 7 outgroup species supported these conclusions (Figure S5).
MARV and EBOV are proposed to have diverged 7,100–7,900 years ago [16]. The inclusion of LLOV and use of Bayesian methods suggests a most recent common ancestor for all filoviruses ∼155,500 years ago (95% HPD of 87,375–249,630 years) and divergence of EBOVs and LLOV approximately 68,400 years ago (95% HPD of 38,857–109,460 years).
MARV and EBOV genomes differ by more than 50% at the nucleotide level. MARV genomes also differ from EBOV genomes in that they have only one, rather than several instances of gene overlap [17]. Whereas the MARV gene four (GP) encodes only one protein, the spike glycoprotein GP1,2, the EBOV gene four encodes proteins (sGP, Δ-peptide, GP1,2, ssGP) via transcriptional polymerase stuttering that results in frame shifts and, in the case of sGP/Δ-peptide, proteolytic processing [18], [19], [20], [21]. MARV spike proteins are highly N- and O-glycosylated but lack sialic acids, whereas EBOV spike proteins may contain sialic acids. Based on these differences, MARV and EBOV are assigned to two different genera. LLOV differs at the nucleotide level from MARVs by 57.3–57.7% and from EBOVs by 51.8–52.6% (Figure S5). The LLOV contains four instances of gene overlap and is predicted to express six transcripts rather than the seven observed in EBOV and MARV. Like EBOV, LLOV gene four (GP) possesses three overlapping ORFs coding for sGP/Δ-peptide, GP1,2 and ssGP analogs while maintaining the proteolytic site that would generate the Δ-peptide. The product is predicted to be highly N- and O-glycosylated. Given these features, LLOV represents the prototype of a new genus, tentatively designated Cuevavirus [17].
Although the dynamics of epidemic filoviral diseases among humans, great apes, and other primates have been described in detail [22], [23], [24], the natural reservoirs, modes of transmission to hominids and pongids (gorillas, and chimpanzees), and temporal dynamics remain obscure. Life forms of diverse taxa have been suggested as potential reservoirs, including bats, rodents, arthropods, and plants [5], [25]. Several lines of evidence support a role for bats including virus replication at high levels in experimentally inoculated insectivorous bats [5]; asymptomatic infection of fruit and insectivorous bats with EBOV in central Africa [4]; asymptomatic infection of fruit and insectivorous bats with MARV [6], [26]; and a history consistent with human exposure to a fruit bat reservoir during a ZEBOV outbreak in the Democratic Republic of Congo (DRC) in 2007 [27].
The discovery of LLOV in M. schreibersii is consistent with filovirus tropism for bats. However, unlike MARV and EBOV, where asymptomatic circulation is posited to be consistent with evolution to avirulence in this long-term host-parasite relationships, several observations suggest that in the case of LLOV, filovirus infection may be pathogenic. LLOV was found in the affected bat population (M. schreibersii) but not in other healthy M. schreibersii or in bats of other species that cohabited the same caves (M. myotis). Furthermore, lung and spleen, tissues with evidence of immune cell infiltrates consistent with viral infection, contained LLOV RNA sequences (Table 1).
The sudden outbreak of bat die-offs in Spain that precipitated this study destroyed several bat colonies in less than 10 days [8]. As recently highlighted by the example of white nose syndrome, a lethal fungal skin infection that is associated with recent declines in North American bat populations [28], bats play critical ecological roles in insect control, plant pollination, and seed dissemination. Although we have not demonstrated a causal relationship between LLOV and mortality in M. schreibersii, the discovery of a novel filovirus in western Europe, and the wide geographical distribution of the associated insectivorous bat is a significant concern. While the virus was detected in the north of Spain, simultaneous bat die-offs also have been observed in Portugal and France [8]. Filoviruses had been posited to show a geographically related phylogeographic structure [29]. Viruses and subtypes from particular geographic area cluster together phylogenetically, suggesting a stable host-parasite relationship wherein viruses are maintained in permanent local-regional pools. Whereas EBOV is associated with humid afrotropics, MARV is focused in drier areas in eastern and south-central Africa [29]. In that analysis, CIEBOV and ZEBOV coincided ecologically, while MARV, a more distantly related filovirus, did not. M. schreibersii distribution does not overlap with the predicted or observed areas of ZEBOV or MARV activity. Thus, LLOV appears not to share the known ecological or geographical niches of other recognized filoviruses.
Recently, the discovery of integrated filovirus elements has led to the proposal that filoviruses have co-evolved with mammals over millions of years [30], [31]. Phylogenetic analyses of LLOV indicate a common ancestor of all filoviruses at least 150,000 years ago. The discovery of a novel filovirus in a distinct geographical niche suggests that the diversity and distribution of filoviruses should be studied further.
The study was made under projects SAF2006-12784-C02-02 and SAF2009-09172 approved by the General Research Program of the Spanish Government. Processed samples came from death bat carcasses collected from the floor of the caves. Sample collection was performed under special permit 14.03.443F (c.p. 1994-01680) from Principado de Asturias and regulation 32/1990 and 68/1995 from the “Dirección de Recursos Naturales y Protección Ambiental de la Consejería de Medio Ambiente del Principado de Asturias” and Royal Decree 439/1990. Sample collection in Cantabria was approved by the “Dirección General de Montes y Conservación de la Naturaleza” at the “Consejería de Agricultura y Ganadería y Pesca” under register E/07505.
Thirty-four bat carcasses (25 M. schreibersii; 9 M. myotis) were collected during the bat die-offs occurring in 2002. Throat and rectal swabs, spleen, brain, lung and liver were stored when available. Then, during the period 2004–2008, rectal and throat swabs were obtained from 1295 healthy bats representing 29 different species (including M. schreibersii from distinct geographic locations in Spain)(Figure S1).
Six M. schreibersii bats were sent to the Service of Pathology of the Veterinary Teaching Hospital of the Veterinary School of the Complutense University of Madrid. During the course of necropsies no macroscopic lesions were observed, and samples for microbiology were obtained. Likewise, samples from the most significant organs and tissues were fixed in 10% buffered formalin for histology, embedded in paraffin and stained with hematoxylin and eosin.
Amplification was carried out in a PCT-200 Peltier thermal cycler (MJ Research, Watertown, MA, USA) utilizing thin-walled reaction tubes (REAL, Durviz, Valencia, Spain). cDNA was obtained with SuperScript III RNase H Reverse transcriptase kit (Invitrogen SA, Spain/Portugal, Barcelona, Spain). A degenerate consensus PCR method for filovirus developed at the Instituto de Salud Carlos III, Madrid, was used for detection of the filovirus RNA-dependent RNA polymerase. Specific primers and protocols can be obtained from the authors on request. DNA bands of the correct size were purified using QIAquick Gel Extraction Kit (Qiagen) and sequenced using standard protocols (Applied Biosystems). After detection of filoviral sequences, primer-walking techniques utilizing degenerate primers on the L and NP gene were also used to obtain additional sequences of the genome (up to 2.5 kb).
Total RNA was extracted from the selected liver sample by using the Trizol procedure (Invitrogen, Carlsbad, CA, USA). Total RNA extracts were treated with DNase I (DNA-free, Ambion, Austin, TX, USA) and cDNA generated by using the Superscript II system (Invitrogen) for reverse transcription primed by random octamers that were linked to an arbitrary defined 17-mer primer sequence as previously described in detail [32]. The resulting cDNA was treated with RNase H and then randomly amplified by the polymerase chain reaction (PCR); applying a 9∶1 mixture of a primer corresponding to the defined 17-mer sequence and the random octamer-linked 17-mer primer, respectively. Products >70 base pairs (bp) were selected by column purification (MinElute, Qiagen, Hilden, Germany) and ligated to specific linkers for sequencing on the 454 Genome Sequencer FLX (454 Life Sciences, Branford, CT, USA) without fragmentation of the cDNA. Removal of primer sequences, redundancy filtering, and sequence assembly were performed with software programs accessible through the analysis applications at the CII Portal website (http://www.cii.columbia.edu). When traditional BLASTN, BLASTX and FASTX analysis failed to identify the origin of the sequence read, we applied FASD [33], a novel method based on the statistical distribution of oligonucleotide frequencies. The probability of a given segment belonging to a class of viruses is computed from their distribution of oligonucleotide frequencies in comparison with the calculated for other segments. A statistic measure was developed to assess the significance of the relation between segments. The p-value estimates the likelihood that an oligonucleotide distribution is derived from a different segment. Thus, highly related distributions present a high p-value.
After detection of several pieces of the genome of LLOV, specific PCR amplifications were performed to fill the gaps. Conventional PCRs were performed with HotStar polymerase (Qiagen) on PTC-200 thermocyclers (Bio-Rad, Hercules, CA, USA): an enzyme activation step of 5 min at 95°C was followed by 45 cycles of denaturation at 95°C for 1 min, annealing at 55°C for 1 min, and extension at 72°C for 1 to 3 min depending on the expected amplicon size. Amplification products were run on 1% agarose gels, purified (MinElute, Qiagen), and directly sequenced in both directions with ABI PRISM Big Dye Terminator 1.1 Cycle Sequencing kits on ABI PRISM 3700 DNA Analyzers (Perkin-Elmer Applied Biosystems, Foster City, CA).
Three alternative data sets were analyzed in the study. The Mononegavirales data set 1 (hereafter DS1) collected 609 cDNA-aligned characters from the conserved domain III of the L gene along 21 species of the order. The filovirus data set 2 (DS2) collected the complete genome (21,794 aligned nucleotides) of 48 viruses of the family. The mononegaviral data set 3 (DS3) collected 19 genomes of filoviruses, and 7 genomes of pneumoviruses and paramyxoviruses used as outgroups to root the tree. In this case a total of 8,547 aligned characters from the L gene were used. For DS1 and DS2 alignments the corresponding polymerase protein sequence data were used as references. All DS were aligned using Muscle v3.7 (http://www.drive5.com/muscle/).
To override distance saturation in the mononegaviruses DS1, the conservative Ka distance was estimated for a subset of 303 second codon positions. Neighbor-Joining (NJ) tree, and 1,000 bootstrap pseudo-replicates were used to evaluate the tree support. Distances estimation, bootstrap and tree reconstruction were performed with SeaView 4.0 [34].
Filoviruses in particular (DS2) and mononegaviruses in general (DS3) were analyzed using maximum-likelihood (ML), and Bayesian methods of phylogenetic reconstruction. In both cases GTR+Γ fitted the parameters of the evolutionary model with the best AIC support. MrBayes v3.1.2 [35] was run using 1,000,000 generations for the filoviruses DS2, and 500,000 generations for the mononegaviruses DS3. In both cases sampling was done every 1,000 generations. To summarize topologies and parameters we retained the last 300 and 200 samples on each data set (which were 600 and 400 samples for DS2 and DS3 considering the two parallel runs of MrBayes). Markov chain Monte Carlo (MCMC) convergence was assessed by checking the average standard deviation of split frequencies (below 0.01) during more than 10,000 generations. Maximum-likelihood (ML) phylogenies were computed in PhyML v3.0 (http://www.atgc-montpellier.fr/phyml/). Shimodaira-Hasegawa (SH) test, and 500 pseudo-replicates of bootstrap analyses were computed to measure the statistical support of ML trees in the two data sets. Bayesian and ML topologies agreed upon the definition of the main clades of the phylogeny. Tree representations were prepared with FigTree V1.3.1.
Using DS2, we also inferred a Maximum Clade Credibility (MCC) tree using the Bayesian Markov Chain Monte Carlo (MCMC) method available in the Beast package [36], thereby incorporating information on virus sampling time. This analysis utilized a strict molecular clock and a GTR+Γ model of nucleotide substitution for each codon position, although very similar results were obtained using other models. The analysis used a Bayesian skyline model as a coalescent prior. All chains were run until convergence for all parameters with 10% removed as burn-in.
Quantitative assays were established based upon virus specific sequences obtained from the high throughput sequencing for LLOV. A TaqMan real time PCR assay on the L gene was developed (primers available on request).
Genbank accession number JF828358 is available online through NCBI (http://www.ncbi.nlm.nih.gov/).
|
10.1371/journal.ppat.1007746 | Regulation of arginine transport by GCN2 eIF2 kinase is important for replication of the intracellular parasite Toxoplasma gondii | Toxoplasma gondii is a prevalent protozoan parasite that can infect any nucleated cell but cannot replicate outside of its host cell. Toxoplasma is auxotrophic for several nutrients including arginine, tryptophan, and purines, which it must acquire from its host cell. The demands of parasite replication rapidly deplete the host cell of these essential nutrients, yet Toxoplasma successfully manages to proliferate until it lyses the host cell. In eukaryotic cells, nutrient starvation can induce the integrated stress response (ISR) through phosphorylation of an essential translation factor eIF2. Phosphorylation of eIF2 lowers global protein synthesis coincident with preferential translation of gene transcripts involved in stress adaptation, such as that encoding the transcription factor ATF4 (CREB2), which activates genes that modulate amino acid metabolism and uptake. Here, we discovered that the ISR is induced in host cells infected with Toxoplasma. Our results show that as Toxoplasma depletes host cell arginine, the host cell phosphorylates eIF2 via protein kinase GCN2 (EIF2AK4), leading to induced ATF4. Increased ATF4 then enhances expression of the cationic amino acid transporter CAT1 (SLC7A1), resulting in increased uptake of arginine in Toxoplasma-infected cells. Deletion of host GCN2, or its downstream effectors ATF4 and CAT1, lowers arginine levels in the host, impairing proliferation of the parasite. Our findings establish that Toxoplasma usurps the host cell ISR to help secure nutrients that it needs for parasite replication.
| Parasites that live inside a host cell must develop strategies to ensure sufficient delivery of nutrients required for survival and replication. After invasion, Toxoplasma rapidly usurps the supply of its essential amino acid arginine from the host cell. Sensing low levels of arginine, the host cell initiates a nutrient starvation response designated the integrated stress response (ISR) that leads to enhanced expression of CAT1, a transporter that facilitates arginine uptake. Through activation of the host ISR and increased expression of this transporter, Toxoplasma secures a continued supply of arginine for its growth and reproduction. Inhibition of these pathways by therapeutic intervention could be a novel strategy to impair survival of the intracellular parasite.
| Toxoplasma gondii is an obligate intracellular protozoan parasite that can infect any nucleated cell. Toxoplasma resides and replicates within a non-fusogenic parasitophorous vacuole that functions to siphon nutrients from its host cell [1]. As an intracellular pathogen, Toxoplasma is auxotrophic for a range of nutrients, including tryptophan, arginine, polyamines, purines, and cholesterol, and relies on its host cell to supply them [2]. Parasites rendered incapable of salvaging these nutrients from host cells suffer reduced growth and virulence. For example, Toxoplasma lacking TgNPT1, a selective arginine transporter, show decreased survival [3]. A major unresolved question is how intracellular parasites, such as Toxoplasma, are able to ensure that a continued supply of essential nutrients is available as they rapidly replicate in host cells.
Phosphorylation of the α subunit of eukaryotic initiation factor-2 (eIF2α) is a well-characterized response to amino acid starvation. Mediated by the protein kinase GCN2 (EIF2AK4), phosphorylation of eIF2α (eIF2α-P) lowers translation initiation, which serves to conserve nutrients and energy [4]. Accompanying repression in global protein synthesis, eIF2α-P also enhances the translation of select mRNAs involved in stress adaptation. An example of a preferentially translated gene target is ATF4, which encodes a transcription factor that directs amino acid metabolism and transport, antioxidation, and cell survival [5]. In addition to GCN2, there are three other mammalian eIF2α kinases: PERK (EIF2AK3/PEK), PKR (EIF2AK2) and HRI (EIF2AK1), which are activated by endoplasmic reticulum (ER) stress, viral infection, and heme deprivation in reticulocyte cells, respectively [6]. Because eIF2α-P can induce ATF4 translation in response to different stresses, this pathway is referred to as the integrated stress response (ISR) [7].
This study addresses the mechanisms by which Toxoplasma ensures that its host cell continues to provide sufficient nutrients for parasite replication. We show that upon Toxoplasma infection, host cells become depleted for amino acids such as arginine, a nutrient stress that triggers the host ISR. Specifically, Toxoplasma infection prompts GCN2 phosphorylation of eIF2α in host cells, which leads to increased expression of ATF4. Enhanced levels of ATF4 triggered transcriptional expression of CAT1 (SLC7A1), which encodes a cationic transporter that facilitates arginine uptake by the host cell, thereby maintaining a ready supply for rapidly replicating parasites. Deletion of any component of the host GCN2/ATF4/CAT1 pathway lowers arginine levels in Toxoplasma-infected cells, dramatically reducing parasite replication.
We hypothesized that depletion of nutrients in Toxoplasma-infected cells would initiate the host ISR. To test this idea, we infected mouse embryonic fibroblast (MEF) cells with Toxoplasma and measured the level of eIF2α-P. Two hours after infection, MEF cells showed increased levels of eIF2α-P accompanied by a reduction in global protein synthesis (Fig 1A–1C and S1A and S1B Fig). Coincident with increased ATF4 protein (Fig 1A), infected host cells also showed increased ATF4 mRNA levels (Fig 1D), both hallmark features of the ISR [4]. Induction of eIF2α-P was also observed upon infection of HFF cells, HEK293T cells, and J774.1 macrophages, albeit HFF cells showed some differences in the timing of induction (S2A–S2E Fig). These findings indicate that the ISR can be activated in different types of host cells in response to Toxoplasma infection.
To determine if GCN2 activates the ISR during Toxoplasma infection, we infected MEF cells lacking GCN2 [8]. Following infection of GCN2-/- cells, there was a significant delay in the induction of host eIF2α-P, with appreciable levels detected only after 18 hours post-infection (hpi) (Fig 1E and 1F) that was accompanied by a delay in the induction of ATF4 mRNA levels (S3A Fig). These data show that GCN2 is a “first responder” eIF2α kinase during Toxoplasma infection of host cells, but other eIF2α kinase(s) can function later during the course of infection.
In the ISR, a primary eIF2α kinase is activated in response to a given stress, with one or more secondary eIF2α kinases being induced with extended cell perturbations [9]. To identify the host eIF2α kinase(s) that serve as secondary ISR responders during Toxoplasma infection, we infected MEF cells lacking PERK or PKR, or multiple eIF2α kinases [10]. While PERK-/- cells showed robust eIF2α-P early in infection along with a rise in ATF4 mRNA levels starting at 2 hpi (Fig 2A and 2B, S3B Fig), eIF2α-P was detected in the combined GCN2-/- PERK-/- cells only after 24 hpi, followed by an increase in ATF4 mRNA levels at 36 hpi (Fig 2C and 2D, S3C Fig). In cells lacking PKR, there was no detectable change in eIF2α-P or ATF4 mRNA levels during Toxoplasma infection until 36 hpi (Fig 2E and 2F, S3D Fig). MEF cells lacking GCN2, PERK, and PKR completely ablated induction of the host ISR during Toxoplasma infection, with no measureable eIF2α-P and minimal ATF4 mRNA even after 36 hpi (Fig 2G, S3E Fig). These findings suggest that PKR may perform a modest role in the induction of the ISR late during infection (after 24 hpi). The lack of host eIF2α-P in the triple knock out MEF cells also suggests that the eIF2α kinase HRI does not play a significant role throughout infection (Fig 2G). Our results suggest that host GCN2 is activated early in Toxoplasma infection, with induction of the secondary eIF2α kinase PERK (which is activated by ER stress) occurring later in the course of infection. Consistent with the idea that Toxoplasma infection produces ER stress in the host cell, we found that activation of IRE1, an ER-resident riboendonuclease that facilitates cytosolic splicing of XBP1 mRNA [11], occurs 12 hpi (Fig 2H). Furthermore, there were increased levels of cytosolic calcium in infected cells (Fig 2I, S4C and S4D Fig), a feature reported to occur upon disruption of the ER [12].
Our measurements of eIF2α-P in infected MEF cells suggest that Toxoplasma initially activates GCN2, followed by PERK at ~18 hpi. The inability of host cells to induce the ISR has a detrimental effect on Toxoplasma infection. Parasite replication was decreased nearly 50% in MEF cells lacking either GCN2 or PERK or both at 36 hpi (Fig 2J). Interestingly, at 48 hpi, parasites grew more slowly in MEF cells lacking both GCN2 and PERK compared to MEF cells lacking either GCN2 or PERK, suggesting that optimal parasite growth relies on both of these host eIF2 kinases (Fig 2J). Supplementation with arginine rescues parasite replication in MEF cells lacking GCN2 but not in MEFs lacking PERK (Fig 2J). These findings show that the host ISR is a significant contributor to robust Toxoplasma replication.
We next tested whether depletion of an essential amino acid, such as arginine, occurs during Toxoplasma infection, contributing to activation of host GCN2. Coincident with the time during infection when GCN2 is activated (2–12 hpi), host arginine levels were depleted; arginine was reduced by more than 40% within 2 hpi and remained low 12 hpi (Fig 3A). Of note, a partial restoration in arginine levels was observed 24 hpi (Fig 3A).
To further test the influence of arginine depletion during Toxoplasma infection on the activation of host GCN2, we supplemented the culture medium with additional amounts of arginine, tryptophan, or leucine (Toxoplasma is auxotrophic for arginine and tryptophan). The addition of 100-fold arginine to the medium significantly delayed and lowered levels of host eIF2α-P and ATF4 mRNA during infection (Fig 3B–3D); similar results were obtained with just a 10-fold supplementation of arginine to the DMEM medium (S5 Fig). By comparison, supplementation with leucine did not alleviate the ISR in infected host cells. The combined addition of arginine and tryptophan further lowered eIF2α-P during infection compared to arginine alone, suggesting that host tryptophan availability may be also affected during infection (Fig 3B–3D). Furthermore, supplementation with arginine can rescue parasite replication in MEF cells deleted for GCN2 (Fig 2J). These results bolster the model that activation of GCN2 during Toxoplasma infection occurs as a consequence of lowered availability of amino acids in the host cells, with arginine being a predominant nutrient required for Toxoplasma replication.
Another host protein kinase regulated by arginine depletion is mTORC1, which regulates many cellular processes including protein synthesis [13]. As amino acid starvation represses mTORC1, there is decreased phosphorylation of its substrate, S6 kinase (S6K). Upon Toxoplasma infection and the accompanying depletion of host cell arginine, we found that S6K phosphorylation was rapidly reduced in the host; supplementing the infected cells with arginine partially restored the phosphorylation of the mTORC1 substrate (Fig 3E and 3F).
We sought to further elucidate the mechanism by which the host ISR is co-opted to ensure that sufficient levels of arginine are available for Toxoplasma replication. Transcriptional and translational expression of the arginine transporter CAT1 (SLC7A1) is induced by amino acid depletion [14,15]. We found increased levels of CAT1 mRNA in infected host cells at 2 and 6 hpi, which were partially diminished by 12 hpi (Fig 4A). Furthermore, we detected elevated levels of CAT1 protein at 6 hpi that were sustained throughout the time course of Toxoplasma infection (Fig 4B and S6A Fig). By comparison, mRNA expression of the related cationic amino acid transporter genes, SLC7A2 and SLC7A3, did not change after infection, suggesting that these transporters do not play a major role during Toxoplasma infection (S7A and S7B Fig).
Deletion of GCN2 or its downstream effector ATF4 in MEF cells significantly lowered the induction of CAT1 mRNA upon infection with Toxoplasma, with no change in protein levels (Fig 4C, S6B and S6C Fig). By contrast, the absence of PERK, which does not respond directly to amino acid depletion and is activated later in infection, did not diminish the induction of CAT1 transcript at 6 hrs (Fig 4D, S6D Fig). Treatment of infected cells with ISRIB, a small molecule that blocks eIF2α-P induction of the ISR [16], lowered the induced expression of both ATF4 and CAT1 (Fig 4E and 4F, S6E Fig). These results indicate that GCN2-mediated phosphorylation of eIF2α, and the ensuing induction of ATF4, enhances CAT1 expression in response to Toxoplasma infection.
We next determined whether the enhanced expression of CAT1 in the host cell was a crucial determinant for Toxoplasma infection. Using CRISPR/Cas9, we knocked out the CAT1 gene in a population of MEF cells, leading to sharply lowered levels of CAT1 mRNA and protein (Fig 5A and S8A Fig). Confirming the specificity of the CAT1-targeted deletion, levels of SLC7A2 and SLC7A3 mRNAs remained similar to wild-type (WT) MEF cells during the course of Toxoplasma infection in the CAT1 knockout cells (S8B and S8C Fig).
Loss of CAT1 led to sharply reduced arginine levels in the MEF cells, which were exacerbated upon Toxoplasma infection (Fig 5B). Consequently, parasite replication was significantly compromised in CAT1-depleted host cells (Fig 5C, S6F, S6G and S8D Figs). Supplementing the CAT1-deficient host cells with additional arginine partially rescues parasite replication, suggesting that arginine uptake can take place at least in part by alternative transporters (Fig 6A and S8F Fig). These results support the model that increased expression of CAT1 by the ISR ensures that host cells can provide sufficient arginine for replication of Toxoplasma.
Toxoplasma encodes four different protein kinases that phosphorylate the parasite eIF2α (TgIF2α) and confer translational control [17]. Each of these Toxoplasma TgIF2α kinases serve in stress adaptation, with two nonessential GCN2-related variants designated TgIF2K-C and TgIF2K-D functioning during nutrient deprivation in the parasite [18,19]. We reasoned that if Toxoplasma infection of CAT1-depleted host cells led to arginine depletion in the host and subsequently the parasite, then the parasite GCN2-related protein kinases would be critical for Toxoplasma replication. In agreement with our earlier studies that TgIF2K-C and TgIF2K-D are not essential for parasite replication in HFFs grown in normal culture conditions [18,19], deletion of either of these GCN2-related protein kinases in Toxoplasma had no effect on parasite replication in wild-type MEF cells expressing CAT1 (Fig 5C). However, Toxoplasma lacking either TgIF2K-C or TgIF2K-D showed reduced replication in CAT1-depleted host cells compared to WT MEF cells (Fig 5C). These results indicate that eIF2α-P plays a pivotal role in nutrient sensing and adaptation in both parasite and host cells.
We next determined whether activation of the ISR and the ensuing enhancement of CAT1 alters arginine uptake by the host cell during Toxoplasma infection. We monitored the transport of [3H]-arginine from the medium into WT and CAT1-knockout MEF cells over a time course of Toxoplasma infection. Note that the radiolabelled arginine was applied to the cultured cells for 8 minutes to measure the efficiency of arginine transport at the indicated hpi. WT MEF cells showed modest arginine transport prior to infection, which increased >50-fold by 18 hours of infection (Fig 6A). By comparison, cells with diminished levels of CAT1 showed low arginine transport, even at later time points of infection. These findings support the critical role of CAT1 for arginine uptake in MEF cells infected with Toxoplasma.
Next, we addressed the contribution of the selective arginine transporter in Toxoplasma, TgNPT1, for salvaging arginine from the host cell [3]. We reasoned that if the parasites take up less arginine from the host cells, there would be diminished induction of the host ISR during the course of parasite infection. Consistent with this model, we found that MEF cells infected with Δnpt1 parasites for up to 12 hpi showed 50% lowered induction of eIF2α-P compared to those cells infected with WT parasites in RPMI medium (Δnpt1 parasites must be cultured in RPMI rather than DMEM, as RPMI has higher arginine concentration [3]) (Fig 6B). Furthermore, there was a delay and a reduced induction of ATF4 and CAT1 mRNA in the host cells infected with Δnpt1 parasites (Fig 6C and 6D). We also confirmed that lowered CAT1 expression in host cells infected with Δnpt1 parasites led to sharply lowered arginine transport into infected host cells (Fig 6E). Collectively, these findings indicate that the host ISR is activated by parasite-dependent depletion of host arginine, which culminates in the host cell enhancing CAT1-dependent transport of the amino acid. The interplay between the arginine transporters of the parasite (TgNPT1) and host cell (CAT1) is important for parasite replication. Whereas deletion of host CAT1 partially lowered parasite counts, the combined loss of host CAT1 and the parasite NPT1 sharply ablated parasite replication (Fig 6F).
Toxoplasma and other obligate intracellular parasites satisfy their resource needs by appropriating essential nutrients from their host cells. However, by doing so the parasites can quickly deplete available nutrients in the host cell, which would jeopardize parasite survival and replication. This study describes an intricate balance between Toxoplasma and host that ensures that a continual supply of nutrients is available for parasite replication. As illustrated in the model represented in Fig 7, Toxoplasma is auxotrophic for certain amino acids, and upon infection can rapidly deplete arginine levels in host cells via its arginine transporter TgNPT1 (Figs 3A and 6). We note that it was recently reported that Toxoplasma may also acquire arginine through the ingestion of host proteins [20]. Deprivation of amino acids can induce the host ISR, featuring GCN2-mediated phosphorylation of eIF2α, which enhances expression of ATF4 (Fig 1). ATF4 directly induces transcriptional expression of genes involved in the uptake and synthesis of amino acids, including the cationic amino transporter CAT1. Furthermore, CAT1 translation was reported to be enhanced by eIF2α-P during amino acid limitations [21]. The ensuing increase in CAT1 leads host cells to take up more arginine, unwittingly securing a constant stream of this critical amino acid for the parasites growing within (Fig 3). The finding that addition of both arginine and tryptophan further lowered eIF2α-P during Toxoplasma infection compared to arginine alone (Fig 3B–3D) suggests that regulatory interplay between the host ISR and parasite is applicable to other amino acids for which the parasite is an auxotroph.
Activation of a cascade of host ISR factors function to ensure that Toxoplasma is supplied with necessary nutrients required for replication. Loss of host GCN2, or its downstream targets ATF4 and CAT1 (Figs 2J and 5C and S6G and S8E Figs), sharply reduces Toxoplasma replication. It is noteworthy that while GCN2 is the first responder in the host ISR during parasite infection, a second eIF2α kinase PERK is activated later during Toxoplasma infection, suggesting that an ER stress is experienced by host cells as parasite numbers expand (Fig 2). The ER stress in host cells appears to involve calcium release from this organelle and may be a consequence of parasitophorous vacuole enlargement and/or its association with the host ER [22]. The parasitophorous vacuole is critical for parasite nutrient acquisition from the host cell, and our findings show that the GCN2/ATF4/CAT1 pathway in the host ISR facilitates a steady supply of amino acids to the parasite.
Analogous to the host ISR, Toxoplasma also senses nutrient depletion via its GCN2-related kinases TgIF2K-C and TgIF2K-D. Deletion of either of these TgIF2α kinases had no effect on parasite replication in WT MEFs cultured in DMEM; however, loss of either TgIF2K-C or TgIF2K-D significantly lowered parasite replication in MEF cells depleted for CAT1 (Fig 5C). Interestingly, TgIF2K-C appeared to be more critical than TgIF2K-D when cultivated in CAT1-depleted host cells. The regulatory mechanisms for these TgIF2α kinases have yet to be resolved, but we have previously shown that TgIF2K-C responds to amino acid deprivation experienced by intracellular parasites [19]. In contrast, TgIF2K-D appears to be necessary to maintain the fitness of extracellular parasites [18]. Parasite amino acid transporters, such as TgNPT1, are also crucial for salvaging nutrients from the host (Fig 6), and it would be of interest to determine whether the TgIF2Ks contribute to the expression and function of these transporters upon nutrient stresses. Considered together, our findings highlight the intricate balance between parasite and host, with each possessing complex nutrient responsive systems involving translational control that function together to optimize parasite survival. The complexity of these pathways provide for potential therapeutic intervention to subvert the ability of intracellular parasites such as Toxoplasma to thrive.
Our findings bolster a growing body of literature describing how pathogens manipulate host cell translation during their intracellular stages. Viruses, bacteria, and other intracellular pathogens have been shown to regulate translation through the ISR or mTOR signaling pathway using effector proteins or by creating a nutrient imbalance [23]. Our study is the first to demonstrate that apicomplexan parasites hijack components of host translational control to ensure nutrient acquisition by the parasite. In this case, it does not appear that a parasite effector protein is involved, but rather the host ISR is triggered in response to the parasite appropriating host nutrients.
Wild-type MEF (mouse embryonic fibroblast) and GCN2-/-, PERK-/-, GCN2-/-/PERK-/-, PKR-/-, and GCN2-/-/PERK-/-/PKR-/- counterparts [10], along with HFF (human foreskin fibroblast) cells and J774A.1 macrophages were maintained in Dulbecco’s modification of Eagle’s medium (DMEM) supplemented with 10% heat-inactivated fetal bovine serum (FBS) (Gibco/Invitrogen) and penicillin/streptomycin at 37°C with 5% CO2. The lack of host eIF2 kinase(s) had no detectable effect on ability of Toxoplasma to invade the mutant host cells (S4A Fig). The ATF4-/- MEF cells were cultured in DMEM that was supplemented with nonessential amino acids, and 55 μM β-mercaptoethanol (Sigma-Aldrich) as described [10]; the media was adjusted to the standard pH 7.2 and filtered before use. Prior to infection, host cells were split to a density of 2x105 cells/well and cultured overnight. Infection media consisted of DMEM supplemented with 1% FBS; mock infection showed that these conditions resulted in negligible eIF2α-P (S1F Fig). Type I RH strain Toxoplasma parasites were used at a multiplicity of infection (MOI) of 3:1. The WT (TATi/Δku80) and Δnpt1 parasites (a gift from Dr. Giel van Dooren, Australian National University) were cultured in RPMI medium as described [3]. RPMI medium contains additional nutrients, including arginine, which helps to overcome Toxoplasma growth defects associated with loss of NPT1 observed for infected cells cultured in DMEM [3]. Infected cells were harvested at the indicated times in RIPA buffer supplemented with cOmplete and EDTA-free Protease Inhibitor Cocktail (Roche) following of protein quantification by Bradford Reagent (Sigma-Aldrich) then the proteins were separated by SDS-PAGE.
To determine the levels of protein synthesis during Toxoplasma infection, MEF cells were infected with Toxoplasma and at the indicated hpi, the infected cells were then incubated with 10 μg/mL puromycin (Sigma-Aldrich) for 15 min. Infected cells were harvesting, and total protein lysates were analyzed by immunoblot analyses using the anti-puromycin antibody (EMD Millipore). Protein synthesis was quantified by densitometry using Image J and normalized by eIF2α. The Infected cells supplemented with arginine were incubated with puromycin as a control (S1 Fig).
Relative arginine levels were measured in host cells using a colorimetric arginase activity assay kit according to manufacturer’s instructions (Abcam ab180877). The infected cells were washed with PBS and scraped from the plates. Following centrifugation, the pellet was resuspended in assay buffer with 0.01% of Triton X-100 to disrupt the host cell membrane. A second centrifugation step was performed to remove host cell debris and parasites, with the supernatant containing the host cell cytosol fraction used to perform the assay shown in S4B Fig. Fidelity of the purification was assayed by immunoblot for parasite surface antigen SAG1 and host eIF2α (S4B Fig). Supernatant samples were diluted 1:10 and 40 μL was used for the assay. After 10 min of incubation with the arginase enzyme, absorbance was measured at 570 nm in kinetic mode for 30 min in a BioTek microtiter plate reader. The values were compared with an arginine standard curve.
Calcium was measured by the Calcium Assay kit (colorimetric, (Abcam ab102505). For this measurement, 2x106 cells were infected with Toxoplasma for up to 24 h and then harvested in PBS containing 0.04% of digitonin on ice. Following centrifugation, the resulting supernatant represented the cytosolic fraction (S4C Fig), which was diluted 1:50 in dH2O and adjusted to 50 μL/well following the manufacturer’s instructions. Absorbance was measured in a BioTek microplate reader at OD575 nm. For controls, uninfected cells were treated with 1 μM tunicamycin for 2 h, 1 μM thapsigargin for 2 h, or 1 μM A23187 for 5 min. Alternately, calcium levels in the uninfected and infected MEF cells were measured with cell permeant Fluo-4, AM (Thermo Fisher Scientific, F14201) according to the manufacturer’s instructions (S4D Fig).
Deletion of the CAT1 (SLC7A1) gene in a population of MEF cells was performed using CRISPR/Cas9, generating CAT1-KO cells. Briefly, four different sgRNAs (g1-ATGGGCTGCAAAAACCTGCTCGG, g2-CCAGGACTTACCGATGATGTAGG, g3-CACAAACGTGAAATACGGTGAGG, g4-CATCATGAGCGTGAGAGCGGCGG) were prepared using the EnGen sgRNA Synthesis Kit (New England BioLabs). The individual sgRNAs were associated with EnGen Cas9 NLS protein (New England BioLabs), which were then transfected into MEF cells using the 4D-Nucleofector System (Lonza) in combination with the P4 Primary Cell 4D-Nucleofector X Kit. As a negative control, EnGen sgRNA Control Oligo (CATCCTCGGCACCGTCACCC) was associated with Cas9 NLS and transfected into MEF cells. Targeted cell lines transfected with one of the sgRNAs, or a combination of all four, were validated by RT-qPCR using specific primers (S8A Fig) and by immunoblot using antibody that specifically recognizes CAT1 (Abcam ab37588).
A PCR-based assay was used to determine the number of parasites in host cells as previously described [24]. Briefly, host cells were infected with Toxoplasma at a MOI 3:1; at 2 hpi, the infection medium was replaced with fresh DMEM. At 30 hpi, genomic DNA was isolated and measured by quantitative PCR using primers to a parasite-specific gene region designated B1 [25]. We note that it is technically difficult to enumerate tachyzoites growing inside of MEF cells as they are not as easily visualized as in larger HFF cells. Therefore, to independently verify changes in parasite growth between WT and CAT1-KO MEF cells, equal numbers of parasites were allowed to infect the MEF host cells for 24 hours. Infected MEFs were then scraped and the material passed through a syringe; equal portions of the lysate preparations were then used to infect HFF cells. Five days post-infection, parasite viability was assessed using a standard plaque assay.
Equal amounts of protein lysates were separated by SDS-PAGE, and immunoblot analyses were carried out for three independent experiments using horseradish peroxidase–tagged secondary antibody. Primary antibodies used for immunoblots included total eIF2α (Cell Signaling Technology, #9722), eIF2α-P (Cell Signaling Technology, #9721), custom affinity-purified ATF4 antibody [26], CAT1 (Abcam, ab37588), GAPDH (Abcam, ab9485), puromycin (EDM Millipore, #17H1), and p70 S6 Kinase (49D7) rabbit mAb (Cell Signaling Technology #2708). Blots were incubated with Pierce ECL Western Blotting Substrate prior to imaging on FluorChem M- Multiplex fluorescence (Protein Simple).
2x105 MEF cells were plated in 6-well plates and allowed to adhere overnight. Cells were infected with tachyzoites for 2 h, washed in PBS, then cultured in DMEM for the indicated times. RNA was isolated from the infected cells using TRIzol LS Reagent (Invitrogen) and cDNA was generated using Omniscript (Qiagen). RT-qPCR was carried out using primers specific to the indicated gene transcript (Table 1), in combination with SYBR Green Real-Time PCR Master Mixes (Invitrogen) and StepOnePlus Real System. Relative levels of transcripts were calculated with the ΔΔCt method using genes encoding GAPDH and β-actin as internal controls. The relative levels of the target mRNAs from the mock-infected samples were adjusted to 1 and served as the basal control value. Each experiment was performed three times, each with three technical replicates.
MEF cells were infected with Toxoplasma for 24h and then were fixed with 4% paraformaldehyde for 20 minutes and blocked with PBS supplemented with 2% BSA. Cells were permeabilized in blocking buffer containing 0.1% Triton X-100 for 30 min then incubated with rabbit anti-CAT1 (Abcam) and mouse anti-SAG1 (Invitrogen) for 1 hour. Alternatively, the cells were incubated with anti-CAT1 without permeabilization, followed by incubation with anti-SAG1. Secondary goat anti-rabbit Alexa-fluor 488 and anti-mouse Alexa-594 (Invitrogen) was applied for 1 hour followed by Vectashield mounting media. DAPI was used as a co-stain to visualize host and parasite nuclei (Vector Labs). Images were acquired with Leica inverted DMI6000B microscope with 63x oil immersion objective.
Radiolabeled arginine uptake assays were based on previously published methods [27–29]. Briefly, MEF cells were infected with Toxoplasma for the designated time points. Infected cells were then incubated with 0.5 μCi [3H]-arginine (Perkin Elmer) in HEPES buffer with 5.6 mM D-glucose at pH 7.4, 24°C. After 8 min, uptake of radiolabelled arginine was thwarted by incubating the cells with 50 mM L-arginine. Arginine uptake was terminated by rapidly washing the cells with ice-cold HEPES buffer following lysis, with 1 ml of 0.5% SDS in 0.5 N NaOH. 700 μl of the lysate was mixed with 5.2 mL of scintillation buffer and read for 1 min in the Packard 1600TR Liquid Scintillation Counter. The remaining sample aliquot was used to determine protein concentration.
Quantitative data are presented as the mean and standard deviation from biological triplicates. Statistical significance was determined using One-way ANOVA with Tukey's post hoc test and multiple t-test in Prism (version 7) software (GraphPad Software, Inc., La Jolla, CA). The number of biological replicates (n) and p values are indicated in the legend of each figure. For immunoblot analyses, the reported images are representative of at least three independent experiments.
|
10.1371/journal.pntd.0000641 | Genotyping of Human Lice Suggests Multiple Emergences of Body Lice from Local Head Louse Populations | Genetic analyses of human lice have shown that the current taxonomic classification of head lice (Pediculus humanus capitis) and body lice (Pediculus humanus humanus) does not reflect their phylogenetic organization. Three phylotypes of head lice A, B and C exist but body lice have been observed only in phylotype A. Head and body lice have different behaviours and only the latter have been involved in outbreaks of infectious diseases including epidemic typhus, trench fever and louse borne recurrent fever. Recent studies suggest that body lice arose several times from head louse populations.
By introducing a new genotyping technique, sequencing variable intergenic spacers which were selected from louse genomic sequence, we were able to evaluate the genotypic distribution of 207 human lice. Sequence variation of two intergenic spacers, S2 and S5, discriminated the 207 lice into 148 genotypes and sequence variation of another two intergenic spacers, PM1 and PM2, discriminated 174 lice into 77 genotypes. Concatenation of the four intergenic spacers discriminated a panel of 97 lice into 96 genotypes. These intergenic spacer sequence types were relatively specific geographically, and enabled us to identify two clusters in France, one cluster in Central Africa (where a large body louse outbreak has been observed) and one cluster in Russia. Interestingly, head and body lice were not genetically differentiated.
We propose a hypothesis for the emergence of body lice, and suggest that humans with both low hygiene and head louse infestations provide an opportunity for head louse variants, able to ingest a larger blood meal (a required characteristic of body lice), to colonize clothing. If this hypothesis is ultimately supported, it would help to explain why poor human hygiene often coincides with outbreaks of body lice. Additionally, if head lice act as a reservoir for body lice, and that any social degradation in human populations may allow the formation of new populations of body lice, then head louse populations are potentially a greater threat to humans than previously assumed.
| While being phenotypically and physiologically different, human head and body lice are indistinguishable based on mitochondrial and nuclear genes. As protein-coding genes are too conserved to provide significant genetic diversity, we performed strain-typing of a large collection of human head and body lice using variable intergenic spacer sequences. Ninety-seven human lice were classified into ninety-six genotypes based on four intergenic spacer sequences. Genotypic and phylogenetic analyses using these sequences suggested that human head and body lice are still indistinguishable. We hypothesized that the phenotypic and physiological differences between human head and body lice are controlled by very limited mutations. Under conditions of poor hygiene, head lice can propagate very quickly. Some of them will colonize clothing, producing a body louse variant (genetic or phenetic), which can lead to an epidemic. Lice collected in Rwanda and Burundi, where outbreaks of louse-borne diseases have been recently reported, are grouped tightly into a cluster and those collected from homeless people in France were also grouped into a cluster with lice collected in French non-homeless people. Our strain-typing approach based on highly variable intergenic spacers may be helpful to elucidate louse evolution and to survey louse-borne diseases.
| Lice are extremely well-adapted ectoparasites that are usually host-specific [1]. Three recognized taxa of lice feed on humans: head lice (Pediculus humanus capitis), body lice (Pediculus humanus humanus), and pubic lice (Pthirius pubis), feed on humans. As one of the most intimate parasites of humans, lice have been widely used as a genetic model to infer host evolutionary history by providing genetic date independent of host data [1],[2]. Several nuclear and mitochondrial DNA sequences have previously been used in population genetic studies of human lice. Of these, the nuclear DNA sequences, EF-1α and 18S rDNA, discriminated lice into two subgroups, lice from Sub-Saharan Africa and lice worldwide[3]. In each subgroup, the head lice were genetically different from the body lice [3]. However, Leo et al. suggested that 18S rDNA phylogeny was concordant with the phylogenies from mitochondrial genes, but the EF-1α phylogeny was concordant neither with the mitochondrial phylogenies nor with the 18S rRNA phylogeny [4]. Furthermore, the mitochondrial DNA markers, partial COI and cytB classified the lice into three deeply divergent clades (Clades A, B, and C), and each having unique geographical distribution. Clade A includes both head and body lice and is worldwide in distribution. Clade B consists only of head lice from America, Australia and Europe, and Clade C consists only of lice from Ethiopia and Nepal [5].
More variable genetic markers, such as internal transcribed spacers (ITS) of ribosomal DNA and microsatellite DNA, were also used to deduce the louse phylogeny. However, the ITS that was chosen was not useful to study the populations structure of human lice because some of the lice had more than one copy of ITS2 in their genome [6]. A subsequent microsatellite DNA-based study has suggested that human head and body lice are genetically distinct [7], however recent studies contradict this hypothesis [5],[8]. Taken together, the population structure of human lice is complex and still unclear. The previously used genetic markers were mostly mitochondrial and nuclear genes that were too conserved to generate more information of genetic diversity of studied louse isolates. So far, no genetic marker has been found that can discriminate among individual human lice.
While being used as a suitable genetic model to study the evolutionary history of humans, lice have long been associated with infectious diseases. Of the three types of lice associated with humans, body lice can be a serious public health problem because they are known vectors of Rickettsia prowazekii, Bartonella quintana, and Borrelia recurrentis, which cause epidemic typhus, trench fever and relapsing fever in humans, respectively [9]. However, medical interest in louse-borne diseases had waned for more than 30 years until 1997, when an outbreak of infection of louse-transmissed R. prowazekii and B. quintana occurred among the displaced population of Burundi [10],[11].
Body lice have long been recognized as human parasites and although typically prevalent in rural communities in upland areas of countries close to the equator, high incidence of louse-borne infections are also increasingly found in the homeless in developed countries [9],[12],[13]. In contrast, head lice represent a major economic and social concern throughout developed nations, because head louse infestations are often associated with school-aged children.
Faster evolving molecular markers are needed in order to epidemiologically survey the vectors of these bacterial infections and to address more recent population-level questions, [8],[14]. Among these fast-evolving genetic markers, intergenic spacers are promising for individual discrimination of lice because they are under less evolutionary pressure, and are more variable than coding genes. These factors make intergenic spacers useful for understanding the population genetics of lice. Highly variable intergenic spacers are useful for strain-typing many bacteria, including louse-transmitted R. prowazekii and B. quintana [15],[16] as well as other pathogenic bacteria [17]. Additionally, intergenic spacer sequences for individual discrimination of lice, are now publicly available [14]. In this study, we used four highly variable intergenic spacers that were selected from the genomic sequence to study the genotypic distribution of a large collection of lice of worldwide origins.
Two hundred and eighty-four human lice collected from Russia, France, Portugal, Mexico, USA, UK, Morocco, Algeria, Peru, Thailand, Australia, Rwanda, and Burundi were included in this study. Lice were collected by experienced entomologists from patients who had only one type of infestation (head or body) and classified according to the site where they were found. Among them, only 97 lice were tested with four nuclear intergenic spacers, other 110 and 77 lice were tested with two intergenic spacers, respectively, due to limited DNA quantity. The strain information, including origin, the body part where they were removed (body or head), and the year when it was collected are given in Figure 1 and Figures S1 and S3. In addition, to estimate the utility of multi-spacer typing (MST) of louse populations, we also studied two body lice from our laboratory colony (Culpeper strain) per year, collected in 1998, 1999, 2000, 2003, 2004 and 2009. From 1998 to 2009, our louse colony went through 132 generations.
All lice were stored at −20°C until processed further. Before DNA isolation, each louse was rinsed twice in sterile water for 15 minutes and cut lengthwise in half. Then, total genomic DNA of each half louse was extracted using the QIAamp Tissue kit (QIAGEN, Hilden, Germany) as described by the manufacturer. The extracted genomic DNA was stored at −20°C until PCR amplification.
The nuclear intergenic spacers were randomly selected from the genomic sequence of Pediculus humanus humanus UDSA strain (http://phumanus.vectorbase.org/index.php) and were identified with flanking genes which exhibited >40% sequence identity with homologous genes in the Vectorbase [14],[18]. Primers used for amplification and sequencing of these intergenic spacers were chosen from the flanking genes using the Primer 3.0 software (http://frodo.wi.mit.edu/cgi-bin/primer3/primer3_www.cgi) and are listed in Table 1. All Primers used in this project were obtained from Eurogentec (Seraing, Belgium).
As the testing of all intergenic spacers on all louse samples was labor intensive, a panel of 16 lice from a wide range of origins was first used to test for the validity and the presence of polymorphisms for each of the intergenic spacers. Subsequently, the intergenic spacers with successful amplification and sequencing from the 16 tested louse samples, were finally used as markers in order to genotype the remaining strains.
PCR amplification of each intergenic spacer was carried out in a PTC-200 automated thermal cycler (MJ Research, Waltham, MA, USA). 1 µl of each DNA preparation was amplified in a 20 µl reaction mixture containing 10 pM of each primer, 2 mM of each nucleotide (dATP, dCTP, dGTP and dTTP), 4 µl of Phusion HF buffer, 0.2 µl of Phusion polymerase enzyme (Finnzymes, Espoo, Finland) and 12.4 µl of distilled water. The following conditions were used for the amplification: an initial 5 min of denaturation at 95°C, followed by 35 cycles of denaturation for 1 min at 94°C, an annealing time of 30 sec at 56°C, and an extension cycle for 1 min at 72°C. The amplification was completed by an extension period of 5 min at 72°C.
PCR products were purified, using the MultiScreen PCR filter plate (Millipore, Saint-Quentin en Yvelines, France), as recommended by the manufacturer. PCR products were then sequenced in both directions, with the same primers used for PCR amplification, using BigDye Terminator version 1.1 cycle sequencing ready reaction mix (Applied Biosystems, Foster City, CA). Sequencing products were resolved using an ABI 3100 automated sequencer (Applied Biosystems). Sterile water was used as a negative control in each assay.
In order to compare the discriminatory power of intergenic spacers with genes, as well as to compare their phylogenetic organizations, the mitochondrial gene, cytB (cytochrome b) was amplified and sequenced from those louse samples when there was DNA to perform the PCR experiments. The primers used for this experiment were CytbF1 (5′-GAGCGACTGTAATTACTAATC-3′) and CytbR1 (5′-CAA CAA AAT TAT CCG GGT CC-3′) [19].
Nucleotide sequences were obtained using Sequencher 4.8 (Gene codes Corp, Ann Arbor, MI, USA). The primers used to amplify intergenic spacers were selected based on flanking gene sequences. The sequences from the coding sequence fragments were not used in the analyses. For each intergenic spacer, and cytB, a genotype was defined as a sequence exhibiting a unique mutation. Each genotype was confirmed to be unique by BLASTn search in all the obtained sequences [20]. Multiple sequence alignments were carried out using the Clustal W software [21]. Phylogenetic analysis of the lice that were studied was obtained using the neighbor-joining and maximum parsimony methods within the MEGA 3.1 software with complete deletion [22] and using the maximum likelihood method in PhyML 3.0 with SH-like approximate likelihood-ratio test and HKY85 substitution model [23],[24]. For this purpose, sequences of the selected intergenic spacers were concatenated.
The discriminatory power (D) of each intergenic spacer, and cytB, was calculated with the Hunter and Gaston's formula [25].
DNA sequences obtained from the S2 and S5 spacers, the PM1 and PM2 spacers and the cytB gene were deposited in GenBank under accession numbers EU928781-EU928862, EU913096-EU913223, and GU323324-GU323334respectively.
Twenty-two nuclear intergenic spacers were initially selected from the genomic sequences and preliminary tested on 16 louse samples (Table 1). However, due to non-specific amplification, or low sequencing quality, 18 intergenic spacers were removed from this study. Finally, four intergenic spacers, hereafter termed S2, S5, PM1, and PM2, were used as typing markers in this study (Table 1).
Through amplification and sequencing, 165–185 bp of S2 and 156–189 bp of S5 were obtained from 207 louse samples and133–155 bp of PM1 and 323–328 bp of PM2 were obtained from 174 louse samples. Sequences from the different genotypes of the four intergenic spacers have been deposited in the EMBL/GenBank database with access numbers: EU928781-EU928862 for S2 and S5 and EU913096-EU913223 for PM1 and PM2.
Two hundred and seven lice were differentiated into 84 and 49 genotypes based on intergenic spacers S2 and S5, respectively. Concatenation of S2 and S5 sequences differentiated the 207 lice into 148 genotypes.
One hundred and seventy-four lice were differentiated into 25 and 62 genotypes based on intergenic spacers PM1 and PM2, respectively. Concatenation of PM1 and PM2 sequences differentiated the 174 lice into 77 genotypes.
Further concatenation of S2, S5, PM1, and PM2, discriminated a panel of 97 lice into 96 MST genotypes. Except for two lice collected from French homeless people which shared the MST genotype 89, the other 90 lice exhibited unique MST genotypes based on the concatenation of four intergenic spacers.
Sequences from each of the four intergenic spacers S2, S5, PM1, and PM2 were identical among the 12 body lice from our laboratory colony collected over 12 years. The genotypes obtained were: 8, 6, 18, and 39 for S2, S5, PM1 and PM2, respectively.
A partial cytB gene sequence was amplified and sequenced from 170 lice. A 316 bp fragment was obtained from each louse after sequence correction and assembling. The cytB sequences were used to classify the 170 lice into 11 genotypes. The body lice sampled from our laboratory colony over 132 generations exhibited identical cytB sequences (genotype 4).
The discriminatory power (D) of the intergenic spacers, PM1, PM2, S2, and S5 was respectively 0.6988, 0.8406, 0.9677, and 0.8913. The D value of cytB was 0.6445. The D value of concatenation of intergenic spacers varied from 0.9123 for concatenation of PM1 and PM2 to 0.9945 for concatenation of S2 and S5. A D value of 0.9998 was reached by combined use of the four intergenic spacers.
The dendrograms of studied lice inferred by the methods of neighbor-joining, maximum parsimony, and maximum likelihood exhibited similar phylogenetic organizations (Figures 1 – 3, Figures S1 – S6).
The 148 genotypes of intergenic spacers S2 and S5, were grouped into 3 clusters, C1, C2, and C3 (Figure S1). Each cluster included both head and body lice. In addition, genotypes 96 and 101 consisted of two body lice and two head lice, respectively (Figure S1). A subcluster (Burundi and Rwanda subcluster) in cluster C3 was comprised of 29 lice from Burundi and Rwanda and one louse from Russia. The other 24 lice collected in Rwanda and Burundi were grouped into cluster C2. The majority of French lice, including those collected from homeless people, were grouped into a sub-clade within cluster C1 (Figure S1).
The 77 nuclear intergenic spacer-genotypes, for PM1 and PM2, were grouped into 2 distinct clusters (Figure S3). Cluster C1 included 148 lice collected from Russia, Mexico, France, UK, USA, and Portugal as well as one louse from Rwanda (Figure S3). A subcluster in cluster C1 contained 31 lice from France and 2 lice from Portugal, which is hereafter referred to as the “French subcluster”. The 19 lice collected from French homeless individuals were tightly grouped with French lice in cluster C1 (Figure S3). Cluster C2 was comprised of 25 lice collected from Rwanda and Burundi. Genotypes 6, 29, and 32, were observed in both head and body lice.
Based on the concatenation of the four intergenic spacers PM1, PM2, S2, and S5, the 97 lice were discriminated into 96 MST genotypes and grouped into two clusters (Figures 1 and 2). Cluster C1 included 75 lice from Russia, France, Mexico, and Portugal, and Cluster C2, the Burundi and Rwanda cluster, contained 22 lice from Burundi and Rwanda (Figure 1). Lice collected from the French homeless individuals were tightly grouped with French lice into two subclusters within C1 (Figure 1).
The 11 genotypes of cytB from 170 lice were grouped into 2 clusters (Figure 3). Cluster C1 included 111 lice from France, Russia, UK, USA, Mexico, Portugal, Burundi, and Rwanda (Figure 3), and corresponded to Type A in a study by Light et al [5]. Cluster C2 included 59 lice from the UK, USA, and Mexico, and corresponded to the Type B reported by Light et al. [5]. Genotypes 4 and 6 comprised both head and body lice.
In this study, MST based on four highly variable intergenic spacers selected from the genomic sequence of a body louse, classified 97 lice into 96 MST genotypes. To date, MST appears to be the most sensitive discriminatory genotyping system of human lice, allowing for discrimination of individuals. In addition, MST helped us to address several important debates associated with human lice. One of the ongoing debates is whether head and body lice are separate species or two subspecies within Pediculus humanus [3],[4],[7],[8],[26]. To address this issue, most of the previous studies have used mitochrondrial or nuclear genes to evaluate and compare the genetic variability of human head and body lice. Studies based on the mitochondrial gene COI [26], the mitochondrial genes cytB and ND4 and nuclear genes EF-1α and RPII [27], the mitochondrial genes cytB and COI [28], or the nuclear gene 18S rDNA [4], supported the hypothesis that human head and body lice are conspecific. Using previously published sequence data, by reticulated networks, gene flow, population genetics, and phylogeny analysis, Light et al. [8] also observed that human head and body lice are conspecific.
However, a recent study performed by Leo et al. [7], in which microsatellites were used as genetic markers, concluded that human head and body lice are two distinct species. These studies made opposite conclusions by using different genetic markers. The low discriminatory power of previously used markers limited their ability to provide convincing evidence whether head and body lice are subspecies of one species or two distinct species.
Our genotypic and phylogenetic analyses using MST did not support the hypothesis that human head and body lice are separate species. For instance, genotype 32, which was a concatenation of the intergenic spacers PM1 and PM2, was comprised of 44 head lice from the USA and the UK as well as one body louse from Europe (Figure S3). Genotypes 6 and 29 were comprised of both head and body lice collected in France (Figure S3). Genotypes 96 and 101, a concatenation of the intergenic spacers S2 and S5, also were comprised of both head and body lice (Figure S1).
Phylogenetic organizations of head and body lice based on each of the four intergenic spacers, and on concatenation of both, support the hypothesis that head lice were grouped with body lice in the same clusters or subclusters (Figure 1, Figures S1 and S2). The changing tree topography observed among spacers may be related to differences in selection pressure that their flanking genes undergo. The genotypic distribution of 170 lice based on partial cytB gene sequences, and the phylogenetic organization of 11 cytB genotypes, also demonstrated that head and body lice shared the same cytB genotypes and were grouped in the same cluster (Figure 2), which further confirmed the hypothesis that human head and body lice are conspecific [3],[27],[28].
In our study, although only two clusters were observed based on partial cytB gene sequences, cluster C1 contained both head and body lice from worldwide origins, and cluster C2 included only head lice from America and Europe (Figure 3). This result did not contradict the previous observation [19] of three deeply divergent clades of human lice, as our study did not include lice from either Ethiopia or Nepal. However, the phylogenetic organization of cytB sequences was significantly simpler than those based on intergenic spacers. Three and two clusters were respectively obtained from the concatenations of the intergenic spacers S2 and S5 (Figure S1), and the intergenic spacers PM1 and PM2 (Figure S3). Additionally, two clusters were generated from the concatenation of the four intergenic spacers PM1, PM2, S2 and S5 (Figure 1, Figures S1 and S2). Each cluster was comprised of several subclusters, such as the French subcluster, including the majority of French lice, and the Rwanda/Burundi cluster, which also consisted of lice collected from sub-Saharan Africa (Figure 1, Figures S1 and S2). This discrepancy of phylogenetic organizations obtained from intergenic spacers and cytB sequences resulted, at least partially, from the high variability of intergenic spacers, which enabled individual discrimination of human lice. In addition, these differences may also be explained by the fact that the louse samples incorporated in each phylogenetic analysis were different due to limited DNA available for such experiments. Furthermore, louse genomic DNA may be highly recombined, which would in turn result in distinct phylogenetic organization from different markers [8]. Thus, collecting more louse samples with wide origins, especially lice from Ethiopia and Nepal, and subjecting them to MST analysis, would help to further clarify the relationship between head and body lice.
Human head and body lice are strict obligate human ectoparasites that differ in several aspects of their morphology, physiology and life histories. Head lice are mostly found on the head and attach their eggs to the base of hair shafts, whereas body lice reside in clothing and attach their eggs to clothing fiber, a life history strategy that probably arose when humans first began wearing clothes [27]. By comparison with body lice, head lice have been described as having shorter and broader antennae, shorter legs, more marked indentations between successive abdominal plates, and as being larger and more deeply pigmented [29],[30]. However, such morphological differences have been determined on a small number of lice and may not hold at the species level [29]. Body lice also take a larger blood meal, lay higher numbers of eggs and develop faster than head lice [29],[31],[32]. In addition, body lice are more resistant to environmental conditions, can stay alive for longer period of time outside the host, are able of transmitting infectious diseases, and are mostly found in adults whereas head lice are essentially found in children [29]. Despite various genetic differences [1]–[8], detailed above, head and body lice have been shown to be able to interbreed [30].
Lice are extremely well adapted ectoparasites, which are usually host-specific by co-speciation with their host [2],[28],[33]. Thus, lice have become a good genetic model for studying specific aspects of human evolution, including addressing when our human ancestors began to wear clothing. Very recently, a study based on sequence analysis of COI and cytB from human head and body lice suggested direct contact between modern and archaic humans [28]. More recently, Light et al. [34] verified this hypothesis by using both nuclear and mitochondrial genes [34]. However, these studies were based on conserved mitochondrial or nuclear genes, which provided limited genetic variability of studied lice. In our study, we also tested the use of highly variable intergenic spacers for strain-typing of human lice to explore human evolutionary history. Concatenation of these highly variable intergenic spacer sequences classified some lice from Rwanda and Burundi into a basal cluster or subcluster and grouped other lice collected in Rwanda and Burundi with lice from North Africa, Europe, USA, and Asia, which supports the hypothesis that human beings originated in Africa (Figure 1, Figures S1 and S2). Thus, highly variable intergenic spacer sequences could be used to study the evolution history of human lice and its host. It might be argued that, due to fast evolution and high polymorphism, intergenic spacers may not be able to fully reflect long-term dynamic changes of populations. However, we observed that MST was a valuable tool for tracing distinct louse populations, and was not biased by mutations that might arise within a single population over time, for at least 132 generations. Nevertheless, we recommend using a combination of coding genes and intergenic spacers because coding genes are conserved enough to highlight evolutionary relationships, and the intergenic spacers are variable enough to identify fine-scale genetic variability.
While lice may present a valuable model to study its host evolution, human head and body lice cause serious health and social problems. Head lice are common worldwide, infesting millions of school children every year and the resistance of Pediculus humanus capitis to insecticides is spreading [35]. Body lice are less prevalent parasites, associated mainly with those living in poor conditions, but are potentially more harmful because they are known vectors of at least three bacterial pathogens in humans: R. prowazekii, B. quintana, and B. recurrentis. There have been several outbreaks of louse-borne R. prowazekii infections in Burundi and Rwanda jails in 1997 and 2001, and sporadic R. prowazekii infections were also recently reported [9]. Epidemiological surveys of these louse-borne diseases are also very important for us to understand and potentially combat these diseases. In addition, recent evidence has been brought that head lice are potential vectors of B. quintana [36],[37], and their role in the epidemiology of epidemic typhus has been questioned [38]. Other studies have identified two endosymbiotic bacteria that have co-evolved in head and body lice [39]–[41]. However, whether these symbionts have any influence on louse behavior, development and/or competence as disease vectors is as yet mostly unknown.
Based on phylogenetic analysis of the four intergenic spacers, S2, S5, PM1, and PM2 as well as a concatenation of them, the head and body lice collected from Rwanda and Burundi tightly grouped together to form clusters as well as subclusters (Figure 1, Figures S1 and S2). In addition, the lice collected from homeless people in France grouped tightly with those collected in non-homeless French people, which suggested louse populations migrate between homeless people and non-homeless people in France and homeless people are known to be at high risk for louse-borne diseases [9],[11],[12].
MST may ultimately be a good tool for performing surveys associated with louse transmission and louse-borne diseases. In addition, our MST analysis demonstrated that head and body lice collected in Rwanda and Burundi in 1997, 2001, 2003, and 2008, were closely grouped (Figure 1, Figures S1 and S2). Thus, the outbreak of louse-borne R. prowazekii infection that happened in 1997 and 2001 opens up the possibility that lice in this region may still pose a risk for the transmission of R. prowazekii to humans. However, clear separation of African lice (collected from Rwanda and Burundi) from other lice was not recovered by intergenic spacers S2 and S5, likely due to recent recombination of nuclear DNA [8].
As mentioned above, head lice are different from body lice morphologically and physiologically. It is possible that these phenotypic differences are controlled by a single mutation or potentially a regulatory gene (or genes) governing, for example, the volume of ingested blood. This is the simplest explanation to understand the genetic data showing that lice have exactly the same origin. Under certain conditions of low hygiene, a head louse infestation can transform into a massive infestation (Figure 4). Certain head lice could colonize clothing (Figure 5), and produce a body louse variant by purifying selection or allotropism, which can in turn generate an epidemic of body lice (Figure 6). Several previous observational studies had also suggested that head lice could become body lice when raised in appropriate conditions [42]–[44]. If this scenario is true, the body louse reservoir is not autonomous and actually depends upon head lice. Previous work has shown that all body lice arose from mitochondrial Type A [5], which suggests that only that genotype has the ability to evolve into the body louse niche. This also makes it possible to understand the difficulties to eradicate body lice in a community, especially when the patients are surrounded by other individuals that are infested by head lice. In our clinical work in Marseilles, France, despite 10 years of attempts to minimize human louse populations, body lice continually reappear and may be due to the persistence of head louse populations [45],[46]. Recent work demonstrating the presence of B. quintana in head lice [36],[37] suggested that they might also transmit infectious diseases, which supports our hypothesis presented in Figure 6, giving them a greater opportunity to ingest circulating bacteria [29],[31], and that head lice are rarely collected and tested, even when present, in outbreaks of louse-borne infections, may explain why head lice have long been considered to be free from human pathogens.
In conclusion, by strain-typing of human head and body lice using both coding sequences and highly variable intergenic spacers, our data supports the hypothesis that human head and body lice belong to the same species. Based on genotypic and phylogenetic analyses, we also hypothesize that head lice may transform into body lice [47] and cause outbreaks of louse-borne diseases. However, more efforts on the genetic studies of head and body lice are necessary to link their genetic difference with morphological and physiological diversity. Whole genome sequencing of head lice and comparative genomics between head and body lice would be useful to address these questions. In addition, due to its high resolution and reasonable phylogenetic classification, MST based on highly variable intergenic spacer sequences may be helpful for the epidemiological survey of louse-borne diseases.
|
10.1371/journal.ppat.1004692 | The Epigenetic Regulator G9a Mediates Tolerance to RNA Virus Infection in Drosophila | Little is known about the tolerance mechanisms that reduce the negative effects of microbial infection on host fitness. Here, we demonstrate that the histone H3 lysine 9 methyltransferase G9a regulates tolerance to virus infection by shaping the response of the evolutionary conserved Jak-Stat pathway in Drosophila. G9a-deficient mutants are more sensitive to RNA virus infection and succumb faster to infection than wild-type controls, which was associated with strongly increased Jak-Stat dependent responses, but not with major differences in viral load. Genetic experiments indicate that hyperactivated Jak-Stat responses are associated with early lethality in virus-infected flies. Our results identify an essential epigenetic mechanism underlying tolerance to virus infection.
| Multicellular organisms deploy various strategies to fight microbial infections. Invading pathogens may be eradicated directly by antimicrobial effectors of the immune system. Another strategy consists of increasing the tolerance of the host to infection, for example, by limiting the adverse effects of the immune response. The molecular mechanisms underlying this novel concept remain largely uncharacterized. Here, we demonstrate that the epigenetic regulator G9a mediates tolerance to virus infection in Drosophila. We found that G9a-deficient flies succumb faster than control flies to infection with RNA viruses, but that the viral burden did not significantly differ. Unexpectedly, mutant flies express higher levels of genes that are regulated by the Jak-Stat signaling pathway, which in other studies was found to be important for antiviral defense. Exploiting the genetic toolbox in Drosophila, we demonstrate that Jak-Stat hyperactivation induces early mortality after virus infection. Precise control of immune pathways is essential to ensure efficient immunity, while preventing damage due to excessive immune responses. Our results indicate that G9a, an epigenetic modifier, dampens Jak-Stat responses to prevent immunopathology. Therefore, we propose epigenetic regulation of immunity as a new paradigm for disease tolerance.
| Efficient immunity against pathogens requires the coordinated activation and repression of genes within multiple signaling networks. Insufficient immune activation results in high microbial burden, severe pathogenesis, and high mortality from infection; overly strong immune responses may lead to tissue damage, immunopathology, and auto-inflammatory diseases. The inevitable tradeoff between immunity and immunopathology necessitates tightly regulated induction and resolution of immune responses. This is achieved by negative regulatory circuits within and among immune signaling cascades and by complex cellular and molecular programs that terminate inflammation [1,2].
It was recently proposed that host defense depends on a combination of resistance mechanisms, which lower or eliminate pathogen burden, and tolerance mechanisms [3,4]. Tolerance reduces the negative effects of an infection on host fitness, which could be either direct damage inflicted by the pathogen itself or adverse effects of the immune response on host tissues. Little is known about the molecular basis for tolerance, but it likely involves regulatory mechanisms that control the magnitude of the immune response [3,4].
The fruit fly Drosophila melanogaster is a powerful model to genetically and functionally dissect innate immunity. Past studies found that the evolutionarily conserved NF-κB pathways Toll and Immune Deficiency (Imd) mediate the humoral response against bacteria and fungi [5]. Defense against viruses, in contrast, requires the constitutively expressed RNA interference (RNAi) pathway [6]. In addition, the RNA viruses Drosophila C virus (DCV), Cricket paralysis virus (CrPV), and Drosophila X virus (DXV) activate the Janus Kinase-Signal transducers and activators of transcription (Jak-Stat) pathway that orchestrates a transcriptional response to fight the infection [7,8].
The evolutionarily conserved Jak-Stat pathway controls important developmental and homeostatic processes, including hematopoiesis and immunity [9,10]. Deficiencies in Jak-Stat pathway genes cause serious immune disorders and increase susceptibility to infections [11–13], whereas hyperactivated Jak-Stat responses are associated with autoimmune diseases and carcinogenesis in humans [13,14]. Also in insects, the Jak-Stat pathway needs to be tightly controlled. The Jak-Stat pathway is required for efficient antiviral immunity in fruit flies and Aedes aegypti mosquitoes, [15]. For example, loss-of-function fly mutants for the Jak kinase hopscotch (hop) support high levels of virus replication and show increased mortality rates upon infection with the RNA viruses DCV and CrPV [7,8]. Yet, hyperactivation of the Jak-Stat pathway in Drosophila can have serious consequences, such as the formation of lethal hematopoietic melanotic tumors in hop gain-of-function mutants [16].
Spatiotemporal regulation of immune responses occurs via a variety of mechanisms. At the transcriptional level, chromatin structure is a major determinant of gene expression. Histone-modifying enzymes deposit covalent modifications on specific amino acid residues of histone tails that alter the structure of chromatin and its accessibility to the transcriptional machinery. Histone H3 lysine 9 dimethylation (H3K9me2) is commonly regarded as a marker for heterochromatic genomic regions and transcriptional repression. Yet, G9a, one of the three H3K9 methyltransferases in Drosophila, mediates H3K9 dimethylation in vivo, but is associated with euchromatic regions [17]. Loss of G9a does not affect heterochromatin formation or global heterochromatic H3K9me2 levels in flies and mice [18,19], but G9a fly mutants show loss of H3K9 dimethylation at about 5% of the euchromatic genome [20]. Moreover, H3K9me2 can be associated with actively transcribed genes [21] and its presence does not globally correlate well with gene repression, unlike other repressive marks such as H3K27me2 and H3K27me3 [20,22]. These observations suggest that H3K9me2 is not solely associated with stably repressed genes, and that G9a might regulate defined sets of euchromatic genes.
A previous study revealed that G9a controls genes that are involved in processes that require tight and dynamic regulation and high transcriptional plasticity, including neuronal processes, stress responses, and immunity [20]. These observations prompted us to study the role of G9a in antiviral defense. Here, we report that G9a mutant flies are hypersensitive to RNA virus infection and that their inducible immune responses are highly dysregulated. We show that G9a and the Jak-Stat pathway epigenetically and genetically interact to modulate immune defense. Genetic hyperactivation of Jak-Stat signaling causes early lethality after viral infection, thus phenocopying loss of G9a. Together, our results uncover an epigenetic mechanism for tolerance that shapes Jak-Stat pathway activity in response to virus infection.
To investigate the role of G9a in Drosophila antiviral defense, we used the loss-of-function allele G9aDD2 and its wild-type genetic background control (hereafter referred to as G9a-/- and G9a+/+)[20]. Since the H3K9me2 mark is essential for the establishment of heterochromatin and proper gene regulation, we first assessed the overall fitness of G9a-deficient flies. G9a mutants were viable, fertile, and showed no obvious defects in development, confirming previous observations [20,23]. Moreover, the average life span of G9a-/- flies was slightly longer than that of wild-type controls (mean survival = 105.8 days and 87,8 days, respectively; P < 0.001) (Fig 1A).
We challenged wild-type and mutant flies with DCV, a positive-sense RNA virus from the Dicistroviridae family, by intra-thoraxic injection. G9a mutants were more sensitive to infection than their wild-type controls, with a mean survival of 3.6 and 6.9 days for G9a-/- and G9a+/+ female flies, respectively (P < 0.001; Fig 1B). Male G9a-deficient flies were also more sensitive to DCV infection than control flies, indicating that hypersensitivity to virus infection was not sex-dependent (S1A Fig).
To analyze whether G9a-/- flies were also more sensitive to other virus infections, we challenged flies with a panel of viruses with different genome organization and genetic makeup. Upon challenge with another Dicistrovirus, Cricket paralysis virus (CrPV), the mean survival of G9a mutants was 3.4 days, compared to 7.3 days for wild-type flies (P < 0.001; Fig 1C). Similarly, when infected with Flock House Virus (FHV), a positive-sense virus of the Nodaviridae family, G9a mutants succumbed faster than wild-type flies to infection (mean survival = 7.9 days and 13.1 days, respectively; P < 0.001; Fig 1D). Also upon challenge with the dsRNA virus Drosophila X Virus (DXV, member of the Birnaviridae), G9a mutant flies displayed higher lethality rates compared to wild-type controls (mean survival 7.6 days and 13.6 days, respectively, P < 0.001; Fig 1E). To analyze whether G9a mutants are also more sensitive to DNA virus infection, we challenged flies with Invertebrate iridescent virus 6 (IIV-6). As we observed before, IIV-6 infected wild-type flies survived for prolonged periods of time and mortality only became apparent in the later stages of the infection (>25 days post infection) [24]. In contrast to their hypersensitivity to RNA virus infection, survival rates of G9a mutants after IIV-6 infection were similar to wild-type levels (mean survival = 33.6 days and 34.4 days, respectively; P = 0.2; Fig 1F). Of note, mock infection with Tris buffer did not affect the survival rate of G9a mutants for up to 40 days (Fig 1F). Flies carrying another loss-of-function allele, G9aDD3, exhibited the same phenotype and succumbed more rapidly than their wild-type controls to DCV, but not to IIV-6 infection (S1B–S1D Fig). As G9a mutants displayed increased sensitivity against all RNA viruses tested, we used the model RNA virus DCV for follow-up studies.
To confirm the role of G9a in antiviral defense, we performed genetic rescue experiments by expression of a G9a transgene in the mutant background using the UAS/Gal4 system. We were unable to recover adult flies expressing the G9a transgene under control of the drivers actin-Gal4, daughterless-Gal4 and tubulin-Gal4, suggesting that ubiquitous overexpression of G9a is detrimental to fly development. The fat body, an organ that is involved in metabolism and immunity [5], is a major target organ of DCV [25]. We therefore used a fat body driver (C564-Gal4) to induce tissue-specific expression of the G9a transgene. Early lethality of infected G9a mutants (mean survival = 3.1 days) was rescued to control levels by fat body-specific G9a expression in the G9a-deficient background (mean survival = 5.6 days, compared to 6.3 days for G9a+/+; P = 0.482; Fig 1G). Survival of genetic control flies that only express the C564-Gal4 driver or the UAS-responder in the G9a-/- background remained significantly different from wild-type flies (mean survival = 4.1 and 3.8 days, respectively, P < 0.001 for both), indicating that the observed rescue was dependent on functional expression of the G9a transgene (Fig 1G). The rescue was tissue specific, since expression of G9a in other tissues, such as hemocytes (using the hemolectin-Gal4 driver), or glia (using the repo-Gal4 driver) did not rescue the phenotype of G9a mutants (S2A and S2B Fig). These experiments indicate that G9a is required specifically in the fat body during virus infection. Moreover, these experiments genetically segregate the role of G9a in antiviral defense from its function in other organs [20].
To analyze whether the reduced survival of G9a mutants is due to a defect in resistance or to reduced tolerance to infection, we analyzed viral load over time. No differences in infectious viral titers were observed between G9a-/- and G9a+/+ flies during the first 3 days post-infection (dpi) (Fig 2A). Since G9a was specifically required in the fat body during DCV infection (Fig 1G), we analyzed viral titers in dissected fat bodies of virus-challenged flies. Virus titers in G9a-/- flies were slightly higher than in wild-type flies, but no significant difference was observed at any time point (Fig 2B). To confirm these data, we measured viral RNA levels in whole flies and fat bodies by quantitative RT-PCR (RT-qPCR). Consistent with the results from the titration, we did not detect significant differences in DCV RNA levels in wild-type and mutant flies over three days post-infection. In the fat body, we observed a modest 3-fold increase in viral RNA at 1 dpi (P = 0.014), but not at the other time points (Fig 2C and 2D).
Together, our results demonstrate that G9a mutants are more sensitive to DCV infection, but that this is not associated with a major and generalized increase in viral titers. Moreover, the modest increase in viral load at 1 dpi in the fat body seems insufficient to explain the strongly reduced survival upon virus infection. We conclude that G9a mutant flies exhibit defects in tolerance to RNA virus infection.
RNA interference (RNAi) is a major antiviral pathway in Drosophila [6]. Given the hypersensitivity of G9a-/- flies to virus infection, we analyzed whether this pathway is functional in mutant flies. To this end, we first monitored RNAi activity using an in vivo sensor assay, in which the inhibitor of apoptosis thread (th) is silenced by expression of an RNAi-inducing hairpin RNA (thRNAi) [26,27]. Expression of thRNAi using the eye-specific driver (GMR-Gal4) leads to severe apoptosis in the developing eye. Consequently, adult thRNAi flies display a reduced eye size, roughening of the eye surface, and loss of pigmentation (S3A Fig). This phenotype is fully dependent on the RNAi pathway, since the phenotype is lost in mutants lacking the central catalytic component of the pathway, Argonaute 2 (AGO2) [26,27]. We expressed the thRNAi hairpin in the eye of G9a-/- and G9a+/+ flies and analyzed the phenotype. Both in G9a-/- and G9a+/+ flies, expression of thRNAi resulted in strong RNAi-induced eye phenotypes (S3A Fig). These results suggest that G9a-/- mutant flies have no major defect in RNAi.
To further evaluate the efficiency of the RNAi response of G9a mutants, we adapted a luciferase-based RNAi sensor assay that we routinely use in Drosophila S2 cells [27,28], to adult flies. Flies were subjected to in vivo transfection with firefly and Renilla luciferase reporter plasmids along with either firefly luciferase-specific dsRNA or control dsRNA, and three days later, efficiency of silencing was assessed in whole fly lysates. As controls, we included Ago2 null mutants and their wild-type controls (w1118). As expected, silencing was abolished in Ago2-/- flies, confirming that loss of FLuc expression was RNAi dependent (S3B Fig, left panel). Efficiency of silencing was similar in G9a-/- and G9a+/+ flies (S3B Fig, right panel), indicating the RNAi pathway is fully proficient in G9a mutant flies.
Since the RNAi pathway was fully functional in G9a-/- flies, we next analyzed whether inducible immune responses were intact in these flies. Virus infection of Drosophila activates the Jak-Stat pathway to induce expression of downstream genes, such as virus induced RNA-1 (vir-1) [7,8]. In addition, the NF-κB pathways Toll and IMD have been implicated in the response to virus infections in some studies [29–31]. We measured expression of vir-1, the stress-induced genes Turandot A and M (TotA and TotM), and the antimicrobial-like peptide Listericin as markers for Jak-Stat activation. To monitor activation of the Toll and IMD pathways, we measured expression of genes encoding the antimicrobial peptides Drosomycin (Drs), Metchnikowin (Mtk), Diptericin (Dpt). In addition, we measured expression of Vago, which is induced in DCV infection via an unknown signaling pathway [25].
We monitored expression of these genes by RT-qPCR at 24 hours after DCV infection (hpi) in whole flies (Fig 3A) and isolated fat bodies (Fig 3B). As observed before [7,8], DCV infection induced expression of the Jak-Stat dependent genes vir-1, TotA, and TotM, but not of NF-κB dependent Drs, Mtk, and Dpt genes. Strikingly, in G9a-/- flies we noted a much higher induction of Jak-Stat dependent genes than in wild-type flies, but no induction of NF-κB dependent genes (Fig 3A). In the fat body, even stronger overactivation of Jak-Stat dependent pathway genes was observed in G9a mutants (Fig 3B). However, basal expression levels of these genes did not differ between non-challenged G9a+/+ and G9a-/- flies (Fig 3C and 3D), suggesting that G9a is not required for steady-state repression of these genes, but that it mitigates their inducibility in response to viral infection.
We also monitored expression of the Jak-Stat dependent genes upon infection with 3 other RNA viruses: CrPV, DXV and FHV (Fig 3E–3G). As observed upon DCV infection, a strong upregulation of vir-1, TotA, and TotM was found in G9a mutants compared to wild-type flies. Upon infection with the DNA virus IIV-6, we detected only slight expression of these genes (1 to 4-fold, at 24 hpi and 7 dpi, when the replication plateau is reached), and expression levels were not significantly different between wild-type and G9a mutant flies (Fig 3H and 3I). We note that those viruses that induce higher Jak-Stat activation also induce higher mortality rates in G9a mutants (Fig 1B–1F). Our results are in line with a previous report showing that DCV, CrPV, DXV, and FHV, but not IIV-6, induce expression of the Jak-Stat dependent genes vir-1 or TotM [7]. In that study, DXV induces strong TotM expression, and DCV, CrPV and FHV induce mainly vir-1 expression, whereas under our experimental conditions, TotA and TotM are induced at higher levels than vir-1 for all viruses.
Jak-Stat deficient flies were reported to display higher viral load and increased mortality upon DCV and CrPV infection [7,8], suggesting that the Jak-Stat pathway controls expression of antiviral effectors. Our data suggest that robust induction of Jak-Stat dependent genes is not sufficient for efficient host defense, which is in line with previous observations [8]. Moreover, the G9a phenotype seems counter-intuitive, since the antiviral Jak-Stat pathway is strongly activated in G9a mutant flies, yet they are hypersensitive to virus infection.
To analyze the transcriptional response to viral infection at a genome-wide scale, we performed transcriptome analyses by next-generation sequencing (RNA-seq). We infected wild-type or G9a mutant flies with DCV, and collected whole flies or dissected fat bodies at 24 hpi (Fig 4A). At this time point, flies do not yet exhibit pathological symptoms, such as reduced locomotion and abdominal swelling.
We first determined the number of differentially expressed genes (≥ 2-fold) upon DCV infection in whole fly (Fig 4B) or fat body (Fig 4C) relative to mock-infected flies. We noted that only a limited number of genes were induced upon DCV infection in whole wild-type flies (n = 31), whereas many more genes were induced in the fat body (n = 129), possibly because the fat body is a major immune organ and a target organ for DCV [25]. In G9a mutants, significantly more genes were induced upon DCV infection than in wild-type flies, both in whole flies and in dissected fat bodies (n = 74, P < 0.0001 and n = 548, P < 0.0001, respectively, Pearson’s chi-squared test). We also observed a large number of genes that were downregulated upon DCV infection. These genes followed the same trends as the virus-induced genes, with greater number of genes affected in G9a mutants both in whole fly and fat body. These observations are in agreement with the results from Fig 3 and suggest that the transcriptional response to virus infection is dysregulated in G9a mutants.
Only a limited number of genes were induced by DCV in both wild-type and G9a mutant flies (13 and 28 genes in whole fly and fat body, respectively; Fig 4D and 4E). This core set of virus-induced genes consisted of genes involved in stress responses such as heat shock proteins (Hsp70 family, Hsp68) and the Jak-Stat dependent Turandot proteins (TotM, TotX, TotC), as well as other Jak-Stat dependent genes, Diedel [32] and Suppressor of Cytokine Signaling 36E (Socs36E) [8], and genes of unknown function (S4A and S4B Fig).
We focused our subsequent analyses on the genes that were differentially expressed (≥2-fold) in G9a mutants, based on the prediction that if G9a represses genes by depositing H3K9me2 marks, direct target genes are most likely de-repressed in G9a mutants. To analyze whether specific biological processes are dysregulated in G9a mutants, we analyzed Gene Ontology (GO) terms of genes that were expressed at least 2-fold higher in DCV-infected G9a mutant flies over infected wild-type flies. In the whole fly dataset, we observed significant enrichment for GO terms, such as “response to abiotic stimulus" and "response to stress" (within the ancestral GO term "response to stimulus") and "immune response" (ancestral GO term "immune system process") (Fig 4F and S4C Fig). GO term analysis on the fat body dataset identified several additional processes, including "reproduction" and “locomotion” (Fig 4G and S4D Fig). Using Pscan [33] to predict transcription factor binding sites in the promoter regions of the differentially expressed genes, we observed, in addition to the TATA-box binding motif, strong enrichment of Stat binding sites, and target sites of the JNK cascade transcription factor, AP-1 (Fig 4H and 4I). In accordance, we noted among the categories “response to stress” and “immune system processes” genes of the Jak-Stat and c-Jun N-terminal Kinase (JNK) signaling pathways, which included pathways components (dPIAS, Socs36E for Jak-Stat; Hemipterous, Gadd45, Jra, Kay for JNK) as well as some of their downstream targets (Socs36E, vir-1, CG13559, CG1572 for Jak-Stat; Puckered and Rab-30 for JNK) [34].
Our results indicate that the transcriptional response to infection is highly dysregulated in the absence of G9a and that Jak-Stat pathway components and downstream targets are among the genes that are derepressed in G9a mutants. We then asked whether these derepressed Jak-Stat genes are direct targets of G9a, or whether they are affected indirectly.
A previous study identified putative G9a target sites by comparing genome-wide H3K9me2 profiles obtained by chromatin immunoprecipation (ChIP) followed by next generation sequencing in wild-type and G9a mutant larvae [20]. Interestingly, these predicted targets are enriched for the GO term "Jak-Stat cascade" (P = 0.0011, 2.3-fold enrichment). We therefore selected Jak-Stat genes that fulfilled three criteria for further analysis: i) harboring a reported loss-of-methylation site in G9a mutants, ii) previously shown to be involved in defense responses, iii) being upregulated in the transcriptome sets of challenged G9a mutants. This set of five genes consisted of pathway components and regulators (domeless, dPIAS, Socs36E), as well as the downstream targets vir-1 and TotM [10,35–38]. Using RT-qPCR, we confirmed that all five predicted G9a target genes show over-induction in response to virus infection in G9a mutant fat bodies (domeless, dPIAS, Socs36E, Fig 5A; vir-1, TotM, Fig 3B). For none of these genes, a difference in basal expression was observed in the absence of viral infection, indicating that these genes are only derepressed upon viral infection in G9a mutants (Fig 5B and Fig 3D).
We next analyzed G9a-dependent targeting of these Jak-Stat genes by H3K9me2 ChIP followed by qPCR (ChIP-qPCR) in dissected fat bodies of wild-type and G9a mutants. We designed qPCR primers in the loss-of-methylation regions observed in ChIP-seq, as shown (domeless in Fig 5C; dPIAS, Socs36E, vir-1, TotM in S5A–S5J Fig). We found that Socs36E and domeless were significantly depleted of H3K9me2 in the fat body of G9a mutants at previously predicted G9a target sites [20] (Fig 5D). Not all G9a targets sites could be confirmed, possibly because ChIP-seq and ChIP-qPCR have been performed at different developmental stages and tissues (whole larvae versus adult fat body, respectively). Although we could not confirm direct targeting by G9a of dPIAS, vir-1 and TotM using ChIP-PCR, we did observe higher expression of these genes in infected G9a mutants. Upregulation of these genes could be a secondary effect resulting from the dysregulation of pathway components, such as domeless and Socs36E, in G9a mutants, rather than from direct epigenetic regulation by G9a. Taken together, these observations suggest that G9a epigenetically regulates a subset of Jak-Stat genes in the adult fat body to shape their transcriptional response to virus infection.
Our data suggest that G9a regulates Jak-Stat responses to prevent excessive expression of downstream target genes. We performed genetic epistasis tests to analyze the relationship between G9a and the Jak-Stat pathway. Epistasis is defined as a genetic interaction in which a mutation in one gene masks the phenotype of a mutation in another gene. We hypothesized that if G9a mediates viral tolerance through dampening Jak-Stat-induced transcription, inactivation of the Jak-Stat pathway would mask the hypersensitivity of G9a mutants to virus infection. Alternatively, if G9a confers tolerance to DCV infection in a Jak-Stat independent manner, simultaneous loss of G9a and Jak-Stat function would result in more dramatic hypersensitivity to virus infection.
To test our hypothesis, we combined the mutant G9a allele with a dominant negative allele of the Jak-Stat pathway receptor domeless (domeΔCyt) under control of a UAS enhancer [39]. We drove expression of domeΔCyt in the background of G9a mutants and wild-type controls using the ubiquitous actin-Gal4 driver and challenged flies with DCV. As expected [8], overexpression of domeΔCyt increased mortality rates in a G9a+/+ background. Remarkably, the difference in survival between G9a-/- and G9a+/+ flies was masked in the Jak-Stat impaired genetic background (Fig 6A). Moreover, mortality rates of double mutant flies (G9a-/- and Jak-Stat deficient) were similar to those of flies in which either G9a or Jak-Stat was inactivated. Therefore, our data suggests a genetic interaction between G9a and the Jak-Stat pathway receptor, domeless. Additionally, we found that domeΔCyt negated the over-induction of TotA and vir-1 in DCV-infected G9a mutants, demonstrating that G9a regulates these genes in a Jak-Stat dependent manner (Fig 6C).
To confirm these results with another Jak-Stat loss-of-function allele, we performed a second epistasis experiment using a fly strain overexpressing the negative regulator of the Jak-Stat pathway Socs36E under control of the UAS sequence [40]. Similar to the experiment with domeΔCyt, overexpression of Socs36E masked the hypersensitivity phenotype of G9a mutants to virus infection, suggesting a genetic interaction between G9a and Socs36E (Fig 6B). Again, as expected, Socs36E overexpression significantly reduced expression of TotA and vir-1 (Fig 6D). In both assays, mock infections were performed in parallel, confirming that differences in survival cannot be attributed to the injury caused by the injection itself (S6A and S6B Fig). As the DCV inoculum of 1,000 TCID50 induced high mortality rates in G9a mutants, as well as in Jak-Stat deficient flies, it remained possible that we may have missed higher mortality rates in flies carrying both mutations. Therefore, we repeated the epistasis experiments using a lower inoculum of 100 TCID50, and confirmed that combining Jak-Stat inactivation with G9a loss-of-function did not yield higher mortality rates than in single mutants (S6C and S6D Fig).
In both cases, inhibition of Jak-Stat signaling in wild-type flies masked the effect of a G9a null mutation upon viral challenge, indicating a genetic interaction between G9a and the Jak-Stat components. Taken together, these results suggest that G9a regulates viral tolerance through modulation of Jak-Stat pathway activity.
Our results suggest that G9a buffers Jak-Stat dependent responses to prevent excessive expression of Jak-Stat dependent genes. We hypothesize that hyperactivation of the Jak-Stat response induces immunopathology that causes increased mortality of G9a mutants upon virus infection. This hypothesis predicts that ectopic activation of the Jak-Stat pathway results in increased rates of mortality upon virus infection.
To test this prediction, we activated the Jak-Stat pathway in adult flies by ubiquitous expression of Unpaired (Upd), a ligand for the domeless receptor, and subsequently infected flies with DCV. Since the Jak-Stat pathway has important functions in development, we used the temperature sensitive Gal80ts allele [41] to induce ubiquitous Upd expression in adult flies by transferring them from 18–20°C (non-permissive temperature) to 29°C (permissive temperature) (Fig 7A). We confirmed by RT-qPCR that Upd as well as the Jak-Stat target gene TotA were strongly induced at 3 days after the shift to 29°C (Fig 7B). We next challenged Upd-overexpressing adult flies with virus. Strikingly, flies with a hyperactivated Jak-Stat pathway succumbed earlier to DCV infection (mean survival = 3.3 days) than genetic control flies expressing only the UAS-Upd transgene or the tubulin-Gal4, tubulin-Gal80ts drivers (mean survival = 5.5 days for both, P < 0.001) (Fig 7C). Moreover, irrespective of the genotype, mock infection did not induce mortality, excluding the possibility that incubation at 29°C is a stressor that triggers early lethality. We conclude that ectopic Jak-Stat activation phenocopies loss of G9a, indicating that immune hyperactivation may underlie the hypersensitivity of G9a mutant flies to DCV infection.
Disease tolerance was recently defined as a defense strategy that reduces the negative impact of infection on host fitness, without a concomitant reduction of pathogen burden [3,4]. The concept of tolerance (also termed resilience) provides an exciting, novel perspective on pathogen-host interactions in metazoans. A few examples of tolerance to bacterial or viral infections have been described in flies [42–48], but the mechanisms of tolerance remain largely unknown. In this study, we elucidate a novel epigenetics-based mechanism for tolerance. We provide evidence that the histone methyltransferase G9a contributes to tolerance by regulating the antiviral Jak-Stat signaling pathway.
G9a mutant flies are hypersensitive to RNA virus infection. Transcriptome analyses indicate that Jak-Stat pathway genes are highly upregulated upon DCV challenge in G9a mutants, whereas their basal levels prior to viral infection are normal. This phenotype, like others reported previously [49,50], seems paradoxical: the antiviral Jak-Stat pathway is strongly activated, yet G9a flies are hypersensitive to infection, showing that immune induction per se is not sufficient for efficient host defense. We propose that increased expression of Jak-Stat dependent genes causes immunopathology, eventually resulting in earlier mortality upon virus infection. In support of this hypothesis, we demonstrated that G9a limits the strength of the immune response through Jak-Stat and that ectopic hyperactivation of Jak-Stat signaling triggered early lethality after DCV infection, thus phenocopying the G9a phenotype. Therefore, we propose that epigenetic regulation by G9a dampens Jak-Stat signaling to avoid immune hyperactivation and subsequent mortality.
G9a seems to be required for tolerance to RNA viruses, but not to DNA viruses. G9a mutants induce higher expression of the Jak-Stat dependent genes vir-1, TotA, and TotM and show increased lethality rates upon infection with four RNA viruses (DCV, CrPV, FHV, and DXV). A previous study found that these viruses all induce either vir-1 or TotM to some extent, but that a resistance phenotype for Jak-Stat mutants (higher lethality rates in combination with increased viral load) was only observed after DCV and CrPV infection [7]. Thus, whereas Jak-Stat is only required for resistance to DCV and CrPV infection, our results suggest that all RNA viruses activate the Jak-Stat pathway and that precise epigenetic control of the pathway is required to prevent immunopathology.
Like in mammals, hyperactivation of immune pathways in Drosophila is detrimental for fitness and survival. For instance, overexpression of antimicrobial peptides or loss of negative regulators such as Caudal or the catalytic peptidoglycan receptor proteins (PGRP-LB and PGRP-SCs) triggers severe tissue pathology in the gut that are reminiscent of chronic inflammatory syndromes in mammals [49,51]. The mechanism by which Jak-Stat overactivation triggers lethality remains to be determined, but may involve expression of potentially toxic gene products that require tight regulation. Moreover, we cannot exclude that additional derepressed genes upon loss of G9a contribute to increased mortality of mutant flies. Alternatively, the G9a phenotype might be caused by defects in cell growth, differentiation, tissue homeostasis or apoptosis, which are also under control of the Jak-Stat pathway. We do note, however, that an external infectious stimulus, i.e. virus infection, was required to cause increased mortality upon genetic hyperactivation of the Jak-Stat pathway, and that G9a mutants appear to develop normally, thus excluding more generalized defects.
Our transcriptome analysis uncovered that, in addition to the Jak-Stat pathway, a multitude of pathways are activated by virus infection, many of which are of interest for follow-up studies. We observed a strong activation of the JNK pathway upon DCV infection. In accordance, predicted binding sites for the AP-1 complex, the transcriptional module of the JNK pathway, were highly enriched in promoters of genes upregulated upon DCV infection in G9a mutant fat bodies. Whether Stat and AP-1 associate upon virus infection to regulate immune genes cooperatively, as previously described in lipopolysaccharide stimulated Drosophila cells [52,53], is an interesting question for future investigation.
Our study makes an important contribution to understanding tolerance mechanisms beyond Drosophila. Two EHMT/G9a paralogs exist in mammals, EHMT1/GLP and EHMT2/G9a [20]. They form a heterodimeric complex, and loss of either protein results in nearly identical phenotypes [54]. We analyzed published microarray data of mice in which the G9a and GLP genes were inactivated in forebrain neurons and observed enrichment for the GO term "immune response", and over-representation of NF-κB binding sites in differentially regulated genes, suggesting that G9a also regulates immune signaling cascades in mammals (S7 Fig). Indeed, a previous study suggested that the G9a-dependent H3K9me2 mark is an epigenetic determinant of the interferon response in murine and human cells [55]. In that study, the abundance of H3K9me2 at the promoters of the Interferon-β (Ifnβ) gene and Interferon stimulated genes (ISG) correlates with expression levels of these genes in different cell types, but deficiency in G9a did not affect basal gene expression. Pharmacological inhibition or genetic ablation of G9a increased Ifnβ and ISG expression in mouse fibroblasts and rendered these cells resistant to viral infection.
Our results demonstrate that the role of G9a in controlling the responsiveness to immune challenge is evolutionarily conserved. Moreover, while the in vitro cell culture model suggested that loss of G9a would be beneficial to the antiviral response of the host [55], our data show that loss of G9a disrupts tolerance mechanisms at the organismal level, and is therefore detrimental for survival. This seems to better match the observations in humans. Heterozygous loss of EHMT1/GLP causes Kleefstra syndrome (OMIM number 610253). This rare disorder is characterized by developmental delay and severe intellectual disability. Interestingly, up to 60% of Kleefstra syndrome patients suffer from recurrent infections; yet, these patients do not suffer from primary immune deficiencies [56]. Whether defects in tolerance explain this aspect of the clinical presentation of Kleefstra syndrome remains an interesting hypothesis.
Flies were reared on standard cornmeal-agar media at 25°C on a light/dark cycle of 12h/12h. G9aDD2 mutants were generated previously by mobilization of the P-element KG01242 located in the 5’ UTR of the gene [20]. G9aDD2 has been used throughout the main text and is referred to as G9a-/-. A precise transposon excision line, referred to as G9a+/+, has been generated in the same genetic background and serves as a control in all experiments. An independent null allele, G9aDD3, has been generated by mobilization of the same P element and contains a deletion of 1850 bp that spans the translation start site [20] (S1B Fig). The following fly stocks and alleles have been described before: UAS-G9a (ref. [20]), C564-Gal4 fat body driver (ref. [57]), Hml-Gal4 hemocyte driver (ref. [58]), UAS-domeΔCYT (ref. [8,37]), UAS-Socs36E (ref. [59]), UAS-Upd (ref. [59]), tubulin-Gal4, tubulin-Gal80ts (ref. [60]), and Argonaute 2414 (ref. [61]). The driver lines armadillo-Gal4 and repo-Gal4 were obtained from the Bloomington Stock Center. In vivo RNAi experiments were performed by crossing GMR-Gal4, UAS-thRNAi/CyO male flies [26] with G9a+/+or G9a-/- virgins. The eye phenotype was monitored in two to four-day-old male F1 offspring lacking the CyO balancer. Upd was conditionally overexpressed by crossing tubulin-Gal4, tubulin-Gal80ts with UAS-Upd flies. Flies were reared at 20°C until hatching. Zero to three-day-old F1 offspring were then incubated at 29°C for 3 days prior to viral challenge, and cultured at 29°C throughout the remainder of the experiment.
Fly stocks were raised for two generations on standard fly flood containing 0.05 mg/ml tetracycline hypochloride (Sigma) to clear Wolbachia infection. Absence of Wolbachia was verified by PCR on DNA of whole flies using Wolbachia-specific primers, as described previously [24]. Persistent virus infections were cleared by bleaching embryos, and absence of DCV, DAV and Nora virus was verified by RT-PCR, as previously described [24].
Virus stocks were prepared as described [24]. Three to five-day-old flies were anesthetized with CO2 and injected with virus suspension using a Nanoject II injector (Drummond) in the thorax, between the mesopleura and the pteropleura. Virus suspensions in 10 mM Tris-HCl, pH 7.3 contained 1,000 median tissue culture infectious dose (TCID50) of DCV and CrPV; 14,000 TCID50 of IIV-6; 3,000 TCID50 of FHV and 2,000 TCID50 of DXV for all survival experiments. 10,000 TCID50 of DCV was used in experiments in which transcriptional responses were analyzed. Flies were cultured at 25°C and transferred to fresh food every 3 days. Survival was monitored daily; lethality at day 1 was attributed to the injection procedure and subtracted from the survival analysis. Unless noted otherwise, three pools of 10 to 15 flies were injected per condition with independent dilutions of virus stock. Fat body tissues were isolated by careful dissection of the abdominal carcasses of adult flies and removal of the gut and reproductive system. This procedure recovers cuticle-associated fat body with minor contamination by muscular and epidermal tissues [62].
Drosophila S2 cells (Invitrogen) were cultured at 25°C in Schneider’s Drosophila Media (Gibco) supplemented with 10% heat-inactivated Fetal Calf Serum (PAA), 50 U/mL Penicillin and 50 μg/mL Streptomycin (Gibco). DCV titers were determined by end-point dilution, as described previously [24]. Briefly, 2.104 cells were seeded in 96-well plates and ten-fold dilutions of fly homogenate were inoculated in quadruplicate. Cells were transferred to fresh medium at day 5, and cytopathic effect (CPE) was monitored until day 14. Viral titers were calculated according to the method of Reed and Muench [63].
RNA was isolated from flies using Isol-RNA lysis Agent (5-Prime), treated with DNase I (Ambion), and cDNA synthesis was performed on 1 μg RNA using TaqMan Reverse Transcription Reagents (Applied Biosystems) according to the manufacturer’s instructions. qPCR was performed on a LightCycler 480 using SYBR Green I Master Mix (Roche). The qPCR program was the following: 95°C for 5 min, and 45 cycles of 95°C for 5s, 60°C for 10s, 72°C for 20s. Expression of the gene of interest was normalized to transcript levels of the housekeeping gene Ribosomal Protein 49 (Rp49). The following primers were used for qPCR:
Rp49 forward, 5’- ATGACCATCCGCCCAGCATAC-3’;
Rp49 reverse, 5’-CTGCATGAGCAGGACCTCCA-3’;
Vago forward, 5’- CAGCCAAGCGATTCCTTATC-3’;
Vago reverse, 5’- CTCATACAGTGGGCAGCATC-3’;
vir-1 forward, 5’-ATTACTCCGAATTCGAAGCTTCC-3’;
vir-1 reverse, 5’- CGAATTCTTCACGCTCCTTC-3’;
Listericin forward, 5’-TTGCGGCCATTCTGGCCATG-3’,
Listericin reverse, 5’- TTTACGTCCCCAACTGGAAC-3’;
TotA forward, 5’- CCCTGAGGAACGGGAGAGTA-3’;
TotA reverse, 5’- CTTTCCAACGATCCTCGCCT-3’;
TotM forward, 5’- ACCGGAACATCGACAGCC-3’;
TotM reverse, 5’- CCAGAATCCGCCTTGTGC-3’;
Drosomycin forward 5’-GTACTTGTTCGCCCTCTTCG-3’;
Drosomycin reverse, 5’- ACAGGTCTCGTTGTCCCAGA-3’;
Metchnikowin forward 5’- TACATCAGTGCTGGCAGAGC-3’;
Metchnikowin reverse, 5’- AATAAATTGGACCCGGTCTTG-3’;
Diptericin forward, 5’- TGTGAATCTGCAGCCTGAAC-3’;
Diptericin reverse, 5’- GCTCAGATCGAATCCTTGCT-3’;
DCV forward, 5’- TTGCCATTGCACCACTAAAA -3’;
DCV reverse, 5’- AAAATTTCGTTTTAGCCCAGAA -3’;
Domeless forward, 5’- AGCTCTGATCCGGATTGTTG-3’;
Domeless reverse, 5’-ATCTCACCGCATTCACCAAG-3’;
dPIAS forward, 5’-AACTGCCCTGTATGCGACAA-3’;
dPIAS reverse, 5’-ACACCTCCTGGAAGTAGCCA-3’;
Socs36E forward, 5’-GTTGCTGCTCCCATTGAAAG-3’;
Socs36E reverse, 5’-GCAAAAGTCGGAGTGTGAGAG-3’;
RNAi competency of adult flies was analyzed using a reporter assay that was adapted from a previously published method in S2 cells [27,28]. In vivo plasmid transfection was based on a method described for Aedes aegypti mosquitoes [64,65]. Three to five-day-old female flies were injected in the abdomen with a 100 nl suspension containing a 1:1 mixture of Schneider’s Drosophila Media (Gibco) and Lipofectamine 2000 (Invitrogen) complexed with 80 ng pMT-GL3 (encoding firefly luciferase, FLuc), 50 ng pMT-Ren (encoding Renilla luciferase, RLuc) and 1 ng FLuc-specific or non-specific control dsRNA. After incubation for 3 days at 25°C, flies were homogenized with a Douncer in passive lysis buffer (Promega). Supernatant was collected after 10 min centrifugation at 16,000 × g and transferred to a new tube, followed by centrifugation for 5 min at 16,000 × g. 25μL of fly lysate was used to measure FLuc and RLuc activity using the Dual Luciferase assay reporter system (Promega). Ratios of FLuc/Rluc were calculated for each sample, and data are presented as fold silencing relative to the non-specific dsRNA control (GFP).
Eighty dissected fat bodies were homogenized in PBS with a douncer and crosslinked with 3.7% formaldehyde for 30 minutes at room temperature. The cross-linking reaction was quenched by addition of 1.25 mM glycine, and the samples were washed with 1 mL PBS and resuspended in a buffer containing 150 mM Tris-HCl (pH 7.5), 600 mM KCl, 150 mM NaCl, 10 mM EDTA, 1 mM EGTA, 1.5 mM spermine (Sigma) and 5 mM spermidine (Sigma). Tissue was further homogenized using a QiaShredder column, and cells were lysed by adding the same buffer supplemented with 2% Triton-X100. Nuclei were pelleted by centrifugation at 6,000 rpm for 10 min, and resuspended in 250 μL incubation buffer (0.75% SDS, 5% Triton-X100, 750 mM NaCl, 5mM EDTA, 2.5 mM EGTA, 50 mM Tris pH 8.0, 0.4% BSA, 1x protease inhibitor cocktail complete (Roche)). After nuclei purification, chromatin was sonicated at 4°C using a Bioruptor sonicator (Diagenode) for 30 minutes at high power with cycles of 30 seconds ON, and 30 s OFF. Anti-H3K9me2 (ab1220, Abcam), anti- H3 (ab1791, Abcam), anti-V5 (R960-20, Invitrogen) antibodies, and Prot A/G beads (Santa Cruz) were used to capture antibody-bound chromatin overnight at 4°C on a rotating wheel. Chromatin was eluted and de-crosslinked for 4 hours at 65°C in 416 μL elution buffer containing 0.2 M NaCl, 1% SDS and 0.1 M NaHCO3. DNA was then isolated using phenol/chloroform, precipitated overnight at -20°C with 1 mL 100% ethanol, 5 μg linear acrylamide, 0.1 M NaAc, pH 5.2. The pellet was washed with 70% ethanol and resuspended in water. Non-immunoprecipitated DNA was isolated in parallel from purified nuclei and used as an input control in qPCR.
qPCR was performed on a LightCycler 480 using SYBR Green I Master Mix (Roche) using the following qPCR program: 95°C for 10 min, and 40 cycles of 95°C for 15s, 60°C for 1 min. The percentage of immunoprecipitated DNA relative to the input was calculated after qPCR. Fold enrichment in H3K9me2 positive DNA was calculated by normalizing the percentage of input of the gene of interest to the euchromatic control gene previously shown to lack H3K9me2, moca [66]. We confirmed in our conditions that H3K9me2 marks are indeed nearly absent on moca in both wild-type flies and G9a mutants. Also, we show that histone H3 levels are identical between G9a mutant and wild-type flies, both on moca and domeless. Using an aspecific IgG isotypic control antibody, we verified very low aspecific background binding to chromatin (S5E–S5J Fig).
Primers for qPCR were designed in regions previously shown to be depleted of H3K9me2 in G9a mutants by ChIP-sequencing [20]. Sequences are as follows:
Socs36E forward, 5’-GAAATCCGATGTGCTGAAG-3’;
Socs36E reverse, 5’-ACATGGGGGTGTTTTACAGG-3’;
Domeless forward: 5’-CACGTGGATCCAAAATACCC-3’;
Domeless reverse, 5’-GATTGCGATTCCGAGAACTG-3’;
dPIAS forward, 5’- CACTGACTCAACCACGCTTC-3’;
dPIAS reverse, 5’-CCGTAAAAGGTGAACCGAAA-3’;
vir-1 forward, 5’- TTGTTCTGGGGCAGAGAAAG-3’;
vir-1 reverse, 5’- ATCGCTTCATGTCAGTGTCC-3’;
TotM forward, 5’-TTCGGGACGGTCACAGATAG-3’;
TotM reverse, 5’-TCTCGAAAAACCCCTGTAGC-3’;
Thirty whole flies or 100 fat bodies of three to five-day-old flies were collected at 24 hours after infection with 10,000 TCID50 of DCV. Samples were frozen on dry ice and stored at -80°C before RNA was isolated using Isol-RNA Lysis reagent as described above. The cDNA library was prepared with the Illumina TruSeq mRNA kit and single-end sequencing was performed on an Illumina HiSeq 2000 (Baseclear BV, Leiden, the Netherlands). RNAseq was performed on a single biological replicate, and should be considered an exploratory analysis.
The FastQ sequence reads were generated in the Illumina Casava pipeline version 1.8.0. Initial quality assessment was based on data passing the Illumina Chastity filter. Reads containing only adapters or PhiX control sequences were removed by filtering protocols developed by Baseclear BV. The second quality control on the remaining reads was performed with FastQC quality control tool 0.10.0. Reads were mapped to the reference genome (Drosophila melanogaster R5/dm3, released in April 2006, UCSC Bioinformatics) using TopHat version 1.4.0. Differential expression between two datasets was analyzed with the Genomatix analysis suite (using DESeq 1.0.6). Gene Ontology enrichment was analyzed using GoToolBox [67], with a hypergeometric test with Benjamini & Hochberg correction. Level 2 GO terms are shown in Fig 4, and level 3 GO terms in S4 Fig Fold enrichment is the ratio of the GO term frequency in the G9a datasets to the genome-wide GO term frequency. Promoter binding-sites for transcription factors were predicted with Pscan [33] on the 500-bp region upstream of the transcriptional start site using the TRANSFAC database. Venn diagrams were generated using Biovenn [68]. The RNA-Seq data are available at the NCBI Gene Expression Omnibus under series accession number GSE56013.
Kaplan-Meier analyses and log-rank tests, as implemented in SPSS Statistics (version 20, IBM), were used to evaluate whether differences in survival were statistically significant. For all other experiments, unpaired two-tailed Student’s t-tests and Pearson’s chi-squared test, as implemented in Graphpad Prism version 6, were used to determine statistical significance. P-values below 0.05 were considered statistically significant.
|
10.1371/journal.pgen.0030188 | Unexpected Novel Relational Links Uncovered by Extensive Developmental Profiling of Nuclear Receptor Expression | Nuclear receptors (NRs) are transcription factors that are implicated in several biological processes such as embryonic development, homeostasis, and metabolic diseases. To study the role of NRs in development, it is critically important to know when and where individual genes are expressed. Although systematic expression studies using reverse transcriptase PCR and/or DNA microarrays have been performed in classical model systems such as Drosophila and mouse, no systematic atlas describing NR involvement during embryonic development on a global scale has been assembled. Adopting a systems biology approach, we conducted a systematic analysis of the dynamic spatiotemporal expression of all NR genes as well as their main transcriptional coregulators during zebrafish development (101 genes) using whole-mount in situ hybridization. This extensive dataset establishes overlapping expression patterns among NRs and coregulators, indicating hierarchical transcriptional networks. This complete developmental profiling provides an unprecedented examination of expression of NRs during embryogenesis, uncovering their potential function during central nervous system and retina formation. Moreover, our study reveals that tissue specificity of hormone action is conferred more by the receptors than by their coregulators. Finally, further evolutionary analyses of this global resource led us to propose that neofunctionalization of duplicated genes occurs at the levels of both protein sequence and RNA expression patterns. Altogether, this expression database of NRs provides novel routes for leading investigation into the biological function of each individual NR as well as for the study of their combinatorial regulatory circuitry within the superfamily.
| NRs are key molecules controlling development, metabolism, and reproduction in metazoans. Since NRs are implicated in many human diseases such as cancer, metabolic syndrome, and hormone resistance, they are important pharmaceutical targets and are under intense scrutiny to better understand their biological functions. In the present study, we determined the expression patterns of all NR genes as well as their main transcriptional coregulators during zebrafish development. We used zebrafish because the transparency of its embryo allows us to perform whole-mount in situ hybridization from early development to late organogenesis. This complete developmental profiling offers an unprecedented view of NR expression during embryogenesis, uncovering their potential function during central nervous system and retina formation. We observed that in contrast to NR genes, only a few coregulators exhibit a restricted expression pattern, suggesting that tissue specificity of hormone action is conferred more by the receptors than by their coregulators. Lastly, by evolutionary analysis of expression pattern divergence of duplicated genes, we observed that neofunctionalization occurs at the levels of both protein sequence and mRNA expression patterns. Taken together, our data provide the starting point for functional analysis of an entire gene family during development and call for the study of the intersection between metabolism and development.
| Diverse processes such as reproduction, development, metabolism, and cancer are genetically regulated to a large extent by nuclear hormone receptors (NRs), a prominent transcription factor superfamily [1]. Several small lipophilic molecules, including steroids, thyroid hormones, and retinoids, function by binding members of this superfamily. In addition, a significant fraction of NRs (approximately 50% in human) are defined as orphan receptors since the identity of their ligand, if one exists, is unknown [2]. With a few exceptions, such as DAX and SHP in vertebrates, all NRs show a common structural organization with a highly conserved DNA-binding domain, and a less conserved ligand-binding domain. Regardless of their status as orphan or liganded receptors, they interact with hormone response elements in gene promoters or enhancers to regulate transcription [2]. NRs repress or activate the transcription of target genes through varied interactions with numerous transcriptional coregulators, which, together with other transcription factors, mediate chromatin modifications, leading to the repression or activation of target genes [3].
The conservation of several domains of NRs allows for relatively easy isolation of their sequences and permits efficient phylogenetic reconstruction of the superfamily [4,5]. This is why several global studies of the whole superfamily have been performed in terms of structural genomics [6–8]. Apart from having implications in evolutionary biology, these comparative approaches have provided an important source of information on the function of human NRs. For example, interspecific comparison of amino acid residues of the ligand-binding domain can help identifying key functional residues required for ligand recognition [9–11]. The number of NR genes present in complete genome sequences has been used as a tool to trace gene duplication and gene loss events that have shaped the structure of the superfamily [4]. Indeed, the number of NR genes varies considerably in metazoan genomes: in humans, 48 receptors were found, 49 in mouse, 21 in Drosophila, 17 in Ciona, 33 in sea urchin, and more than 270 in Caenorhabditis elegans [4,6,7,12,13]. In two species of pufferfish, Takifugu rubripes and Tetraodon nigroviridis, at least 71 NR genes were found, thus highlighting the impact of the ancestral fish-specific genome duplication that took place early in evolution of actinopterygian fish [14,15] (Figure 1).
In addition to this structural and evolutionary information, several resources are now available to provide functional information on NRs (e.g., NURSA, http://www.nursa.org/; NUREBASE, http://www.ens-lyon.fr/LBMC/laudet/nurebase/nurebase.html; and NucleaRDB, http://www.receptors.org/NR/). Several bioinformatic and experimental searches for hormone response elements have led to a better understanding of the transcriptional hierarchies controlled by NRs and their ligands [16–18]. Systematic analysis of NR interactions with themselves and with their coregulators allowed for precise elucidation of each receptor's interactome [19,20]. More recently, systematic expression studies using reverse transcriptase PCR (RT-PCR) and/or DNA microarrays have been performed in classical model systems such as Drosophila and mouse [21–24]. However, for studying the implications of NRs in development, it is critically important to know when and where individual genes are expressed. This is why we have established the complete spatiotemporal profiles of the expression of all NR genes during embryonic development using the zebrafish as a model system, because the optical transparancy of its embryo allows studies of gene expression with a cellular resolution using whole-mount in situ hybridization [25].
Other studies have been performed on NR expression during embryonic development in vertebrates, mainly in mouse, rat, chicken, and Xenopus [2]. However, most of them are partial and only describe expression by northern blot analysis or by in situ hybridization restricted to one organ or a few developmental stages for a limited number of genes. Moreover, for many NRs, expression during development was only studied regarding their roles in the adult, therefore introducing a bias in the interpretation of the data.
To carry out this large-scale project, we isolated all 70 NR genes in zebrafish plus 31 of their coactivators and corepressors. We analyzed the expression of these 101 genes from gastrula to early larval stages by whole-mount in situ hybridization. This allowed us to detect extensive correlation of expression between NR genes and their coregulators. Our results reinforce the notion that NRs are mainly expressed during organogenesis, with few of them expressed at early developmental stages. Our most unexpected finding is that the large majority of NR genes are expressed during central nervous system (CNS) and retina development, since classically, the primary role NRs was thought to be metabolism control in endodermal derivatives [2]. Finally, evolutionary analysis of the NR genes that were retained following the fish-specific genome duplication, shows that neofunctionalization of these genes occurred at the levels of both protein sequence and RNA expression patterns.
Taken together, our data extend and refresh our vision of NR involvement during vertebrate development, calling for a closer look at metabolic pathways and the control of homeostasis in developmental processes.
Using RT-PCR, we isolated probes corresponding to 70 NR genes from Danio rerio, all of which correspond to a distinct locus in the zebrafish genome, which is publicly available. The assignment of each sequence was done for each NR group by phylogenetic analysis (Figure S1). Figure 1 gives the complete list of the 70 NR genes that we found (see also Table S1). When we compared with the mammalian NR complement, we did not find orthologs of RARβ, LXRβ, or CAR using either RT-PCR or database searches. An ortholog of RARβ was found in the complete pufferfish genomes but was apparently lost in zebrafish. Thus far, neither LXRβ nor CAR has been described in any fish. Because it is always difficult to decide on the absence of a gene in a given genome, especially when the complete genome sequence is not published, we performed additional RT-PCR and PCR experiments on various tissues and/or DNA preparations with several primer pairs for these genes. We nevertheless cannot formally rule out that we artifactually missed a specific duplicate.
It is now clearly established that actinopterygians underwent a complete genome duplication [14,26]. Indeed, compared to mammals, 19 genes are duplicated in zebrafish (TRα, RARα, RARγ, PPARα, PPARβ, Rev-erbβ, Rev-erbγ, RORα, RORγ, VDR, RXRα, RXRβ, COUP-TFα, EAR2, one ERRβ or γ, ERβ, SF-1, GCNF,and SHP). Eighty percent of these duplications are shared with pufferfish. For clarity, we name these duplicates with capital letters after the gene name: PPARα-A and PPARα-B are thus the two duplicates of PPARα. Our phylogenetic analysis also reveals five NR paralogues that have no counterparts in mammals. These genes are Rev-erbγ, COUP-TFγ, ERRδ, FF1C, a member of the FTZ-F1 family, and HNF4β. They were also all found in the pufferfish genomes, while HNF4β is present in Xenopus laevis and in chicken.
Many different coactivators and corepressors of NRs have been described and these molecules exhibit highly variable functions, specificities, and modes of action [2,3,27]. Therefore, in contrast to NR genes, we did not attempt to perform an exhaustive analysis and decided to isolate only the most obvious ones. We have isolated representatives of the four main coregulator complexes (Figure S2), namely, the p160 complex (containing the three SRC/p160 factors, CBP/P300, Cited3, and CARM), the SMCC or Mediator complex (with TRAP220), the SWI/SNF complex (Baf53, Baf60 and BRG1), and the corepressor complex containing NCoR, SMRT, and histone deacetylases. Table S1 contains the list of the 31 probes that we have isolated, along with their Genbank accession numbers. As for NR genes, for each coregulator isolated, a tree was constructed to assign clear orthology and in some cases we noticed the presence of actinopterygian-specific duplicates (Figure S1). However, we cannot exclude that duplicates may exist for some coregulators for which only one copy was detected.
We have determined the spatiotemporal expression pattern of the 101 zebrafish genes by whole-mount in situ hybridization at seven different developmental stages that are classically studied [28]. Plates describing individual expression patterns are presented in Figure S3, and have been deposited in the ZFIN database (http://zfin.org) and will be available at the Nurebase Web site.
At a global scale, we can define three different types of expression profiles for NRs during zebrafish embryogenesis: (i) genes not expressed during embryogenesis and larval stages or expressed under the limit of detection of in toto in situ hybridization, (ii) genes expressed ubiquitously, and (iii) genes that exhibit a spatially restricted expression pattern. If we compare the expression profiles of NRs at each developmental stage, we observe that the number of spatially restricted NR expression profiles increases dramatically from gastrula to 48 h post-fertilization (hpf) (from 11% to 60%), whereas the number of ubiquitously expressed genes is almost constant (around 20%; see Figure 2A and Table S3). Therefore, it appears that the vast majority of NR genes are not expressed during early embryogenesis but rather late, i.e., during organogenesis. A similar observation was made in Ciona, where only 17% of NR genes are expressed early, whereas 48% were found expressed during later stages [29]. We did not notice any obvious correlation between the phylogenetic position of NR genes, their orphan versus liganded status, and the type of their expression patterns (restricted, ubiquitous, or not expressed).
We then analyzed in detail the expression pattern of NR genes that are spatially restricted during embryogenesis. Strikingly, we observed that many of them are expressed in the retina and in the CNS (e.g., spinal cord and/or brain), even if for each receptor, the expression is restricted to a part of these organs (Figure 2B). Figure 3 presents a selection of the expression patterns we detected in the brain, stressing the diversity of expression of NR genes in the CNS. At the mid-somitogenesis (MS) stage, more than 55% of spatially restricted NRs are expressed in the brain, and this proportion increases up to 71% at 48 hpf. The same picture holds for the retina (from 29% at MS stage up to 59% at 48 hpf). In addition, all genes expressed in the retina, except for TRβ, are also expressed in the brain and/or in the anterior spinal cord.
To test whether this high percentage of genes expressed in CNS and retina could be specific to NRs, we analyzed a set of 1,900 genes with spatially restricted expression patterns available in the ZFIN database. We found 40% to 54% of these genes expressed in the CNS between 24 and 48hpf, whereas for NR genes, this percentage rose to 71%. Eleven percent to 37% of genes were expressed in the retina, whereas 30% to 59% NR genes were expressed in the same organ (Figure 2C). Therefore, even if many genes are indeed expressed in CNS and retina, NR genes tend to be expressed more often than expected in these organs.
In contrast, some organs or tissues express very few NR genes in a restricted manner. This is the case for the lens, blood, somites, and heart, even if these organs express the NRs that show a ubiquitous expression pattern. Phenotypic analyses of mouse knockouts, as well as studies on the implication of NRs in human diseases, have suggested a major role for NRs in the control of homeostasis, and specifically in lipid metabolism, including cholesterol and steroid metabolism (see [2] for a review). These processes occur in organs such as liver, intestine, pancreas, and adipose tissue, all of which are endodermal derivatives, as well as in the adrenal gland, which is derived from neural crest cells. Looking at NR expression in these organs, we found, at various stages, VDR-B, EAR2-B, and FF1C expressed in the intestinal bulb, whereas FXRα, ERβ-A, and LRH1 were found in the liver. In addition, three genes, PPARβ-A, PXR and HNF4α, were detected in both organs. Therefore, we are confident that we did not miss expression of NR genes in endodermal tissues before 5 d of development. The case of PXR, which in mammals is restricted to endodermal derivatives, nicely illustrates this point. In zebrafish, we found this gene expressed at 24 hpf in the pituitary and at 36 hpf with a complex pattern in the telencephalon and diencephalon (see Figure S3). At 48 hpf, expression remains in the CNS but is also found at a relatively low level in intestine and liver. This demonstrates the power of whole-mount in situ screens in revealing heretofore unsuspected expression patterns. Recently, two analyses of genes of the NR2E, RAR, and RXR groups also revealed extensive expression in retina and CNS, globally supporting our findings [30,31].
Another well-known target of NRs in mammals is the sexual organs. Sex determination is a complex and late event in zebrafish and sexual organs are not yet differentiated at the stages examined by whole-mount in situ hybridization. We thus cannot discuss the eventual implication of NRs genes in differentiation of sexual organs in this species and this may account for the lack of expression of AR, PR, and ERα that we noticed in our study. However, at the studied stages, primordial germ cells are present in the embryo and migrate along the body axis, but we did not detect specific NR expression (e.g., GCNFs) in these cells.
Because we found frequent and complex restricted expression in the developing retina, we performed high-resolution analysis at 72 hpf, when the retina is already well differentiated. We then analyzed systematically the expression of the 25 NR genes expressed in the retina at this stage (Figures 4 and S4). At 72 hpf, the retina is divided into three main layers: the outer nuclear layer (ONL) containing cell bodies of photoreceptors, the inner nuclear layer (INL) which contains four classes of interneurons (amacrine, bipolar, horizontal and interplexiform) as well as Müller glia, and finally the ganglion cell layer (GCL), which contains ganglion cells.
By examining the retinal expression of these 25 genes, we observed a large diversity of patterns (Figure 4). TRβ, PNR, COUP-TFα-B are only expressed in the ONL (Figure 4B). RORβ, NURR1 and ERRγ are found only in the INL (Figure 4C), whereas no NRs are expressed only in the GCL. COUP-TFβ and EAR2-B are expressed in an asymmetric manner in the dorsal part of the INL and the ONL, respectively (Figure 4D). COUP-TFγ shows expression in the ventral part of these two layers (Figure 4E). TLL expression is not restricted to a specific layer of the retina, but is associated with cell proliferation (Figure 4F) [31]. Finally, the remaining 15 NRs are expressed in more than one layer and often ubiquitously. All these data highlight a diversity of NR gene expression in the retina suggesting that these genes may be implicated in a wide variety of processes.
The fact that the retina expresses a large proportion of the members of the NR superfamily has not been noticed in other vertebrates. This may be due to the fact that no global spatiotemporal expression pattern study of this superfamily has been performed with whole-mount in situ hybridization in mammals or Xenopus, or that there are specific differences between mammals and zebrafish concerning NR gene expression in the retina. We thus specifically verified if the genes that are expressed in the zebrafish retina are implicated in retinal development in other vertebrates. By an extensive survey of the literature, we found that among the ten genes that express in specific cell layers or cell types in the retina, four (TRβ, PNR, RORβ, and TLL) are known to be important for retina development in mammals. Indeed, retinal phenotypes in knockout mice and mutations in human diseases have been associated with these genes [33–35]. In addition, expression in the retina has been observed in other vertebrates for five more genes: NURR1 [36–38], ERRγ [39], COUP-TFγ, COUP-TFβ, and EAR2 [40–42]. Finally only one of these genes, COUP-TFα-B is expressed in zebrafish retina, while its mammalian counterpart is not [43]. We noticed that some genes ubiquitously expressed in the zebrafish retina (Rev-erb and ROR) have also been described as expressed in the mammalian retina [44,45]. Taken together, our data strongly suggest that some receptors have a conserved role in vertebrate retinal development and that the importance of this organ for the study of NR biological functions has been largely overlooked. This nicely illustrates the power of large expression screens, such as the one we performed here, in unraveling potential functions of NRs in specific organs.
In striking contrast with NR genes, most of the CoA/CoR we studied show ubiquitous expression (CBP-A, CBP-B, P300-A, P300-B, BRG1, PCAF, NCoA6, Baf 53, SRC1, SRC2, NCoA4, Baf 60, N-CoR, Alien, Sin3A, HDAC1, HDAC3, and TIF1α) or do not display embryonic expression that could be detected by whole-mount in situ hybridization (TRAP220, MYST-HAT2, TRIP13, and ARA54) (see Figure S3). In fact, only 30% of the coregulators (SRC3, RIP140-A, RIP140-B, PGC1, TRIP7, TIF1γ, Cited3, CARM1, SMRT, and HDAC4) show a spatially restricted expression pattern, suggesting that tissue specificity of hormone action is conferred more by the receptors than by their coregulators. Apart from TIF1γ, which is expressed in ventral hematopoietic mesoderm [46], all other spatially restricted coregulator genes are expressed in the CNS, stressing again the importance of NR signaling in this organ.
Among the ten spatially restricted coregulators, we found expression territories that do not correlate with expression of spatially restricted NRs. For example, HDAC4 is expressed in trigeminal ganglia and PGC1 and RIP140-A are expressed in several cranial ganglia, whereas RIP140-B is specifically expressed in the habenula. Some of the coregulators, namely HDAC4, Cited3, CARM1, SMRT, and RIP140-B, also show restricted expression in the retina. It should be noted that TRIP7 is expressed in the lens, where only EAR2-B is expressed in a restricted manner. We also observe expression of SMRT at 5 dpf in the thymus, where we did not find any expression of spatially restricted NR genes. These data support in vivo the notion that coregulators mediate the action of transcription factors other than NRs.
Our systematic analysis revealed extensive similarities of expression patterns between NRs and their coregulators. For example, in the p160 family, which contains three members (SRC1, SCR2, and SRC3), SRC3 shows a restricted expression that is reminiscent of that of RXRs and RARs (Figure S5) [30]. This gene is mainly expressed in anterior spinal cord, posterior branchial arches, and tail bud, suggesting possible RAR/RXR interactions with SRC3 in these territories.
PGC1 is another coactivator showing a striking correlation of expression with certain NR genes. This gene was identified by its direct interaction with PPARγ and was later shown to be important for other NRs, including ERRα, TRs, and RXRs (for a review, see [47,48]). In zebrafish, PGC1 shows a very specific expression pattern in adaxial cells, pronephric ducts, and mucous cells during somitogenesis, and in the epiphysis, olfactory bulb, diencephalic nuclei, hindbrain, heart, pronephric ducts, mucous cells, and slow muscle fibers at 24 hpf. Overall, this expression pattern overlaps extensively with those of the ERR genes (Figure 5). During somitogenesis stages, ERRα is expressed in adaxial cells, pronephric ducts, and mucous cells, ERRβ and ERRγ are expressed in pronephric ducts, while ERRβ/γ is expressed in mucous cells. At 24 hpf, PGC1 expression overlaps with that of ERRα in pronephric ducts, in slow muscle fibers, of ERRβ in pronephric ducts, epiphysis and in diencephalic nucleus, of ERRγ in epiphysis and diencephalic nucleus and of ERRβ/γ in the mucous cells. In the mouse, no complete embryonic expression pattern of PGC1 has been reported, but complex expression in adult brain was observed in rat [49]. In mouse, PGC1 is preferentially expressed in slow muscle fibers, a situation that we also found in zebrafish [50]. This is consistent with the notion of specific needs for PGC1 in mediating transcriptional activity of ERRs during embryogenesis and with reports highlighting the functional importance of the PGC1/ERR hub [51].
In addition, we identified two other groups of genes (Rev-erb/ROR and COUP-TF) sharing extensive similarity of expression suggestive of common functions. Nine of the ten Rev-erb/ROR genes are expressed in retina, optic tectum, hindbrain, and/or epiphysis. We also found that the expression patterns of three coregulators, RIP140-B, SMRT, and HDAC4, largely overlap with those of Rev-erb/ROR. These expression data strongly suggest that in vivo these genes are regulated in a similar way. In accordance with this notion, we recently observed that Rev-erbα expression is under the control of Rev-erbs and RORs both in vitro and in vivo [52,53]. These expression patterns are fully consistent with the important role played by these genes in the generation and control of circadian rhythm [54–56]. Interestingly, SMRT has been shown to interact with Rev-erbs in mammalian cells [57]. Taken together, these observations suggest that the roles played by RIP140-B, SMRT, and HDAC4 in circadian rhythm should be more carefully examined in the future. Similarly, among the six members of the COUP-TF group, COUP-TFα-A, COUP-TFα-B, COUP-TFβ, COUP-TFγ, and EAR2-B are expressed in a similar and complex expression pattern in the CNS (Figure S6). Once again, this is congruent with the known role of these genes in nervous system development in zebrafish and more generally in vertebrates.
We performed hierarchical clustering of regionalized NR and coregulator genes and the anatomical structures expressing them using a binary matrix that quantifies expression pattern divergence between genes (Tables S2 and S5; Figure 6). This clustering analysis revealed the existence of a higher-order network relating NR genes, their coregulators, and development according to space and time. The anatomical structures expressing NRs and coregulators reveal a clear organization into three clusters (Figure 6): expression in nervous system at late stages (I), early embryonic expression (II), and late expression in non-nervous system structures (III). Cluster I can be further subdivided: retina and optic tectum (Ia), spinal cord (Ib), and brain structures (Ic). Similarly, cluster II can be divided into an early nervous system (IIa) and an early non-nervous system organs (IIb) subcluster. These results suggest that during development, NR genes and their coregulators can be categorized depending on their timing of expression (early/late) and their expression in nervous or non-nervous tissues.
NR and coregulator genes are split into seven clusters (shown on the vertical axis of Figure 6) that follow the previously discussed organ clustering. The genes that we defined above as coexpressed at several developmental stages are clustered within this hierarchy. SRC3 is found in cluster 4 with RARα-A, RARα-B, RARγ-A, RXRα-B, and RXRγ, since they are expressed early (organ subcluster IIa) and late (organ subcluster Ib) in the spinal cord, a situation illustrated in Figure S5. Similarly, PGC1 belongs to cluster 6 as ERRβ and ERRγ. Several members of the COUP-TF family (COUP-TFα-A, COUPTFα-B, COUP-TFβ, COUP-TFγ, and EAR2B) are grouped in clusters 3 and 8, and the ten Rev-erb and ROR genes are found together in cluster 5, since they are expressed late in the retina and in the brain. Furthermore, these genes are never expressed in the spinal cord, a situation explaining their inclusion in cluster 5.
Therefore, this clustering reveals an underlying hierarchy of NR and coregulator genes and suggests that several transcriptional networks are differentially deployed in a spatiotemporal manner during zebrafish development.
Our expression dataset gives us the opportunity to analyze the evolution of NR gene expression after duplication. We found in zebrafish 19 pairs of genes specifically duplicated in actinopterygians that account for the increased number of NR genes when compared to tetrapods.
According to the Duplication–Degeneration–Complementation (DDC) model [58], duplicated genes have three main fates: in the majority of cases, one of the copies is lost (64% for zebrafish NR genes), in some cases both duplicated genes are subfunctionalized (i.e., they share the function of their nonduplicated ancestor), and in other cases one of the copies undergoes neofunctionalization (i.e., it acquires a new function), while the other retains the function of the ancestor gene. Sub- or neofunctionalization can occur at the level of the expression patterns of the duplicated genes or at the level of their protein coding sequence.
Taking into account that we have no expression data from a basal actinopterygian fish that was not subjected to the genome duplication, expression divergence after duplication can only be inferred by comparison with other vertebrates. Of the 19 duplicated couples that we have studied, we found four cases indicative of neofunctionalization at the level of their expression patterns (RARγ, RORα, RORγ, and GCNF). GCNF provides a clear example of such a case: GCNF-A has an expression pattern that is reminiscent of Xenopus and mouse GCNF [59,60], whereas GCNF-B expression is very divergent, with expression observed in head, lateral line neuromasts, and branchial arches. Therefore, it seems that GCNF-A has kept the ancestral expression pattern, whereas GCNF-B has acquired a new one.
The acquisition of a new function can be achieved by fixing advantageous mutations within one of the duplicated genes. The neofunctionalized gene will then evolve under positive selection, significantly faster than the other gene in the pair, which will retain the ancestral role and thus evolve under purifying selection (elimination of deleterious mutations). Asymmetric evolution between gene duplicates may thus be interpreted as a sign of neofunctionalization [61,62]. We compared the protein sequences of the 19 NR gene pairs to the protein sequence of a nonduplicated outgroup (Homo sapiens) and found that the ratios of the evolution rates of the duplicated proteins varied from 1.01 (i.e., similar rates) to 6.1. Because the outgroup is very distant, only strong differences in the evolution rate can be detected and evaluated as statistically significant, making our results conservative. We found a significant acceleration of the protein evolution rate (i.e., a ratio significantly different from 1), relative to the nonduplicated sequence of the outgroup, for eight out of the 19 gene pairs (p-values < 0.01 in seven out of the eight cases and a p-value = 0.03 in the remaining one). An alternative explanation for the asymmetry in the evolution rates would be the genomic context, as proposed by Zhang and Kishino [63,64]. When two copies have different recombination rates, the copy in the low recombination context accumulates deleterious substitutions because of Hill–Robertson effects (degeneration) and thus will evolve faster than the copy in the high recombination context. We have controlled for this effect by estimating, when possible, the recombination rates of the two genes in each pair (Table S4). The recombination rates were estimated by comparing genetic and physical maps of the zebrafish genome (A. Popa, personal communication). In three out of the eight cases of asymmetrical evolution rates between duplicates, this estimation was not possible at least for one of the genes. Out of the five remaining pairs, only one presented a difference in the recombination rates of the duplicates compatible with the asymmetry in their evolution rates (SHP-A/SHP-B), which suggests that the vast majority of the asymmetrically evolving pairs truly evolved through the neofunctionalization model.
We then looked further into this asymmetric evolution of the duplicates by evaluating their expression pattern divergence. Doing this in a quantitative manner allowed us to investigate if there was any correlation between sequence evolution and the evolution of the expression patterns after duplication. The divergence of the expression patterns of the duplicates varied from 0 (same expression pattern found for both genes, e.g., RXRβ, an almost ubiquitously expressed pair detected in 162 out of 165 organs considered in the analysis or PPARα, a nondetected pair) to 1 (almost completely different expression patterns of the two genes; e.g., SHP-A is detected in only four of the 165 organs and SHP-B is not detected, see Table S2). We computed the sequence divergence between duplicates by calculating the ratio between nonsynonymous to synonymous substitutions (Ka/Ks) between the coding sequences. The Ka/Ks ratio can only be calculated for 17 of the 19 pairs of genes because in two cases (SHP and RORγ) the Ks was saturated. Because all the gene duplicates are from the same duplication event (fish-specific genome duplication), differences in Ks values reflect different mutation rates within the genome. By dividing Ka by Ks we corrected for the influence of these mutation rate differences in the evolution of the coding sequence.
Strikingly, we observed a significant positive correlation (Pearson correlation factor R2 = 0.69 and p-value = 0.04) between the expression divergence and the sequence divergence of the duplicates belonging to the pairs (six) where a neofunctionalization is suggested by the asymmetrical evolutionary rates of the proteins (Figure 7B). This means that the divergence of the coding sequence was accompanied by a divergence of the regulatory sequences. No significant correlation between the expression divergence and the sequence divergence was found for the pairs (11) with similar evolutionary rates (a positive but nonsignificant correlation may be observed in Figure 7B).
Taken together, our results show that for duplicated NR genes, neofunctionalization occurred in almost half of the cases, both at the protein and RNA expression patterns.
Several systematic analyses of the NR superfamily at the gene expression level have recently been reported. Sullivan and Thummel [23] have conducted a northern blot analysis of all 21 Drosophila melanogaster NRs from egg to adulthood. A systematic quantitative PCR analysis of expression of 49 NR genes in 39 adult tissues and at several circadian times has been reported in the mouse [21,22]. These studies revealed NR gene coordinated transcriptional programs in developmental and physiological pathways. Analyzing transcript expression at the tissue level with quantitative PCR or northern blots has the advantage of providing a quantitative measure of transcript abundance. Coupled with hierarchical clustering of the data, this allowed the division of the NR regulatory network in the mouse into two main processes: reproduction, development, and growth on the one hand, and nutrient uptake, metabolism, and excretion on the other. Our analysis of embryonic and larval expression patterns, studied by whole-mount in situ hybridization, allows a direct visualization of the spatiotemporal dynamics of the NR superfamily during development. Our study thus nicely complements these previous global analyses by providing, with unprecedented details, a complete dataset of the embryonic territories where NR-mediated regulation is likely to be deployed.
Our data also allow the definition at the global scale of groups of genes expressed in similar locations at several developmental stages and thus highlight the potential transcriptional hierarchies of NRs and coregulators that occur during development. Clustering of the tissues expressing NR and coregulator genes into three main groups according to developmental timing and nature (neural/nonneural) of the tissue supports the notion that NR regulation is used differently during embryonic development. There is no extensive overlap between the seven clusters we defined and those found by Bookhout and colleagues [21]. This suggests that the underlying logic of NR deployment during embryonic development in zebrafish and in the adult mouse is different. Nevertheless, one should keep in mind that the two datasets are different (qualitative versus quantitative data and embryonic versus adult stages) and are thus difficult to compare. The detection of groups of coexpressed genes suggests that some crossregulation might occur between NR genes and/or their coregulators. The ERR-PGC1 and RAR-RXR-SRC3 groups provide good examples of these potential hubs. Future comparison of the expression patterns reported here with those issued from large-scale gene expression analyses will undoubtedly provide relevant information on NR-regulated networks that control embryonic development.
Our exhaustive expression screen reveals that many NRs known to be tightly linked to the control of metabolism in adults are expressed during embryogenesis (e.g., PXR, HNF4α, RXRs, COUP-TFs, and ERRs as well as several coactivators such as PGC1, CITED3, and RIP140).
It is important to stress that most of the expression patterns we describe here are conserved in vertebrates. Given that the methods used to determine expression during development differ from one model organism to another (e.g., tissue sections in mouse, whole-mount in situ in zebrafish, and Xenopus), and that only a minority of these NR genes have been studied in several organisms, an exhaustive global comparative analysis of the expression patterns is not yet feasible. Nevertheless, of 26 genes for which data are available, we found 22 cases of complete (TRα-A, TRα-B, PPARβ-A, PPARβ-B, VDR, HNF4α, RXRβ-A, RXRβ-B, TLL, NURR1, SF1-A, SF1-B, LRH1, and GCNF-A) or partial (TRβ, RARα-A, RARα-B, RARγ-A, RARγ-B, RXRα-A, RXRα-B, and COUP-TFβ) conservation of expression, whereas in only four cases (PPARα-A, PPARα-B, PXR, and RXRγ) we found very different expression patterns between zebrafish and other vertebrates. Therefore, we are confident that most of the data we generated will be transferable to mammals and will thus be relevant for the study of human diseases.
Both epidemiological and clinical evidence suggests that prenatal factors play a role in the origin of the metabolic syndrome and its components: hypertension, insulin resistance, obesity, and dyslipidemia (reviewed in [65]). Experimental studies demonstrate that an adverse embryonic or fetal environment can induce structural and functional abnormalities in pancreatic islet cells and can lead to permanent changes in insulin sensitivity [66]. Thus, any developmental perturbation that would affect NR expression and/or the production of NR ligands may be transferred to the NR gene regulatory hierarchy and may impact embryonic development and later on adult physiology and metabolism. Indeed, it is easy to induce insulin resistance and symptoms of the metabolic syndrome by manipulating maternal nutrition (an event that could easily affect NR ligand production) or by exposing the mother to synthetic glucocorticoids [67–69]. Therefore, relating the embryonic expression of NRs, including classical pharmacological targets like TR, RAR, RXR, and PXR, to specific developmental processes will help to better understand the mechanisms of the development of metabolic syndrome. Our data provide a unique basis from which to begin such an analysis.
Our expression analysis can also be used to identify roles of certain NR or coregulator genes in specific human diseases. For example, since an unexpected number of them are expressed in retina, it could be fruitful to search for their implication in the development of retinal diseases. There are still a large number of mapped but unidentified Mendelian human retinal diseases, some of which match to the chromosomal location of the NR genes, which we found expressed in the retina. For example, we found both RXRα and Rev-erbα in the retina and both have a chromosomal location in humans (17q) that corresponds to the one detected for a specific retina disease, CORD4 (Cone Rod Dystrophy 4) [70].
In sum, this expression screen, performed on a species that resembles humans on the level of organization and physiology and on a protein superfamily that can easily be targeted by drugs, will provide important new information for the identification of interesting targets for drug discovery.
The importance of neofunctionalization following gene duplication has been continuously discussed in the literature since Ohno proposed that it was the main mechanism allowing phenotypic diversity [71]. There is no doubt now that subfunctionalization plays an equally or even more important role in the functional evolution of gene pairs [58,72]. In contrast, the relative contribution of both mechanisms for functional diversification between gene duplicates is still an open question. Different factors must be taken into account when analyzing gene evolution after duplication, including population characteristics of the species studied [73]. Asymmetric evolutionary rates of duplicates, which may be interpreted as a sign of neofunctionalization [61–64], have been shown to affect 10% to 56% of duplicated genes analyzed in various species from yeast to fish [62]. In teleost fish, differences in evolution rates were found in 37% of the duplicated genes analyzed [74,75]. Here, our analysis revealed that 42% of the 19 NR gene pairs analyzed evolved at different rates (when compared with an orthologous single copy outgroup). Furthermore, the retention of gene duplicates among the NR family (36%) is also higher than the one estimated for the whole genome after the fish whole-genome duplication (15% [74]). This is consistent with a higher gene retention after duplication and the presence of neofunctionalization, both of which have been reported in regulatory/development-implicated gene families [74,76–78] (e.g., NRs) compared with other functional classes of genes.
Finally, we also observed a significant positive correlation between coding sequence divergence and expression pattern divergence for the asymmetrical evolving gene pairs. Coupled evolution between coding and regulatory sequences was previously found for single-copy genes, between orthologs of D. melanogaster and D. yakuba [79] and of C. elegans and C. briggsae [80]. In our case, this parallel evolution between coding and regulatory sequences suggests that neofunctionalization affected both the protein function and the expression pattern of the gene. For instance, the evolution rate of GCNF-B is more than two times that of GCNF-A, suggesting that GCNF-B evolved under positive selection, thus acquiring a new function. This is consistent with the divergence of GCNF-B expression patterns suggestive of neofunctionalization: as is the case for the protein sequence, it seems that GCNF-A has kept the ancestral expression pattern, whereas GCNF-B has acquired a new one. It can be hypothesized that following expression divergence of a pair of duplicated genes, the gene that is expressed in novel embryonic territories will accumulate mutations in its coding region more rapidly, because the cognate protein will be exposed to a novel set of interaction partners.
One of the striking results of our screen is the widespread expression of NR genes in the nervous system: at 36 hpf, 70% of the spatially restricted NRs are expressed in the CNS, whereas 40% of them are expressed in the retina. This represents an underestimation, because ubiquitously expressed NR genes may also play an important role in these organs. Indeed, the expression of the zebrafish HDAC1 gene is widespread in the embryo at all stages of development, whereas this gene plays an important role in the anterior CNS by maintaining neurogenesis [81]. The developmental role played by these genes is perhaps not connected to their adult function in regulating metabolism, but it has to be emphasized that many other observations focus on an unanticipated link between the control of metabolism and nervous system development. In fact, several large-scale expression screens have revealed expression of metabolic enzymes, cholesterol and fatty acid transport proteins, and hormonal receptors in embryos, even during early embryogenesis. In zebrafish, the brain-type fatty acid binding proteins FABP7a and FABP7b, which intracellularly bind to docosahexaneoic acid (DHA), an RXR ligand [82], are distributed in the early developing CNS, retina, pharynx, and swim bladder [83]. Similarly, a fatty acid hydroxylase (FA2H) is expressed in enveloping layer, pronephric ducts, nose, pharynx, liver, and gut during embryonic development [84]. In a recent genome-scale analysis of genes expressed during mouse retina development, prominent expression of metabolic enzymes has been observed in specific cell types, such as the Müller glia [40]. The reasons for such a widespread spatiotemporal control of metabolic genes may be linked to a variable metabolic demand of developing organs or cell compartments related to differential proliferation or differentiation. Alternatively, metabolic proteins could play a specific developmental role. In the case of NR genes, we have at present no specific indication that, for example, the restricted expression of PXR in specific areas of the zebrafish CNS is linked to its detoxification function in adult liver. Another possibility is that metabolic enzymes may be implicated in the production or delivery of signaling molecules. This is of course the case for the CYP26, retinaldehyde dehydrogenases, CRBP, and CRABP, the molecules implicated in retinoid metabolism and transport in vertebrate embryos. Clearly, the evidence that continues to accumulate from various experimental model systems suggests that metabolism should no longer be disconnected from the study of embryonic development.
Given the unknown expression patterns of most of NR genes in zebrafish, we used total RNA extracted from various adult tissues (muscle, gills, liver, etc.) as well as from embryos at different developmental stages. RNA was extracted from frozen tissues using TRIZOL reagent (Life Technologies). The RNA samples were treated with RQ1 deoxyribonuclease, extracted using phenol/chloroform/isoamylic alcohol (25:24:1) and chloroform/isoamylic alcohol (24:1), and finally precipitated with ethanol.
Degenerate or specific primers were designed using an alignment of all published nucleotide sequences for homologs from other vertebrate species according to previously described methods [85] or using available sequences. Many of the primers are degenerate and were used in a touchdown PCR assay [85]. PCR products were cloned into the PCR2.1-TOPO vector (Invitrogen) and subcloned in pBSK+ or pBKS+ to allow synthesis of sense and antisense probes. A list of studied genes and their sequence accession numbers is given in Table S1.
Predicted amino acid sequences were aligned automatically using ClustalW [86] with manual correction in Seaview [87]. Phylogenetic reconstruction was done using amino acid alignments of the longest sequences found for each gene. Only complete sites (no gap) were used. To separate orthologs and paralogs for each sequence, trees were constructed for each group (see Figure S1) with the Phylo_win program [87] using the neighbor-joining method [88] with Poisson-corrected distances on amino acids. Reliability of nodes was estimated by 1,000 bootstrap replicates [89]. Alignments of amino acids were also used to calculate the level of sequence similarities with other vertebrate sequences.
Whole-mount in situ hybridization was performed as previously described [25]. Several stages were used: gastrula (G), early somitogenesis (ES, 3–6 somites); mid-somitogenesis (MS, 14–18 somites); and 24, 36, and 48 hpf [28]. For several genes, expression was also studied at 5 d post-fertilization. Sense and antisense RNA probes for each gene tested were prepared from partial cDNA. Probes were made against internal coding regions for most NRs, allowing detection of the different 5′ and 3′ isoforms.
After in situ hybridization, embryos were mounted on slides in 100% glycerol. Pictures were taken with a Leica M420 Macroscope or with a microscope (Leica DM RA2) with differential interference contrast using a digital camera (Coolsnap CCD, Roper Scientific). Digital pictures were saved as TIFF files, then adjusted for contrast, brightness, and color balance using Adobe Photoshop software and stored as such or after conversion to JPEG format to reduce the file size.
To analyze retinal expression in more detail, embryos previously hybridized with a specific probe were postfixed overnight at 4 °C in 4% paraformaldehyde, 3% glutaraldehyde, and phosphate buffer 0.1 M pH 7.4; dehydrated in graded ethanol and propylene oxide; embedded in a mix of araldite and epon; and sectioned (3.5 μm) on a microtome using standard techniques.
The expression patterns were further coded in a binary matrix to quantify their divergence (see Table S2). In this table, all organs in which at least one gene is expressed, are listed (a total of 165 organs for the whole set of developmental stages), and the presence or absence of each gene transcript in each organ is indicated respectively by a “1” or a “0.” All the organs or anatomical structures were labeled with “1” for ubiquitously expressed genes, whereas all organs were marked with “0” for nonexpressed genes.
Starting from this matrix, expression divergence between the duplicates was calculated as the number of gene expression differences (i.e., the number of organs where only one gene in the pair is detected) over the total number of organs where at least one of the genes in the pair is expressed. This means that the same number of differences will give a stronger divergence if the genes concerned have a restricted expression pattern (i.e., if the pair is expressed in only a few organs) than if they are broadly expressed.
Hierarchical clustering analysis was performed using the binary matrix (101 genes versus 166 anatomical structures; Table S2). We excluded 13 genes for which no expression was detected in the 166 organs, and 31 genes ubiquitously expressed in all structures (except in the yolk syncytial layer). Thus, only genes with regionalized expression (detected here in a number of organs between 1 and 41) were included in the analysis. We have verified that the inclusion of ubiquitous and undetected genes in the analysis does not modify the overall conclusions of the hierarchical analysis. Similarities between the expression patterns of the 57 genes and also between the patterns of anatomical structures were computed as Jaccard's coefficient, which is classically employed for species presence–absence data in ecology [90]. Jaccard's coefficient is an asymmetrical binary coefficient, which does not take into account the case of absence/absence in the degree of similarity between two binary patterns. It is suitable in the framework of expression data, because the presence (i.e., the detection) of a gene in an organ is more informative in terms of expression or not than its absence due to the existence of detection thresholds. Distances between the expression patterns of genes and between the patterns of organs were calculated as d = sqrt(1 − s), with s being the similarity coefficient. Dendrograms were built using the two sets of distances (genes and organs) by hierarchical clustering following the Ward's method. We performed all analyses with the R software (http://www.R-project.org) using the package ade4 [91] to compute distances between expression patterns.
The protein sequences of each pair of actinopterygian-specific paralogs were aligned with the orthologous nonduplicated protein sequence of the outgroup using ClustalX [86]. We used the closest appropriate outgroup (having diverged before the actinopterygian genome duplication) being completely sequenced (H. sapiens). We used RRTree [92] on these protein alignments to make relative rate tests and thus evaluate differences in protein evolution rates of the duplicates. Nucleotide alignments of the corresponding coding sequences were obtained based on the protein alignments. We used Gestimator (analysis-0.6.6 by K. Thornton) to compute the Ka/Ks ratios for each pair of duplicates with Comeron's method [93].
|
10.1371/journal.pntd.0005009 | Plasmodium vivax VIR Proteins Are Targets of Naturally-Acquired Antibody and T Cell Immune Responses to Malaria in Pregnant Women | P. vivax infection during pregnancy has been associated with poor outcomes such as anemia, low birth weight and congenital malaria, thus representing an important global health problem. However, no vaccine is currently available for its prevention. Vir genes were the first putative virulent factors associated with P. vivax infections, yet very few studies have examined their potential role as targets of immunity. We investigated the immunogenic properties of five VIR proteins and two long synthetic peptides containing conserved VIR sequences (PvLP1 and PvLP2) in the context of the PregVax cohort study including women from five malaria endemic countries: Brazil, Colombia, Guatemala, India and Papua New Guinea (PNG) at different timepoints during and after pregnancy. Antibody responses against all antigens were detected in all populations, with PNG women presenting the highest levels overall. P. vivax infection at sample collection time was positively associated with antibody levels against PvLP1 (fold-increase: 1.60 at recruitment -first antenatal visit-) and PvLP2 (fold-increase: 1.63 at delivery), and P. falciparum co-infection was found to increase those responses (for PvLP1 at recruitment, fold-increase: 2.25). Levels of IgG against two VIR proteins at delivery were associated with higher birth weight (27 g increase per duplicating antibody levels, p<0.05). Peripheral blood mononuclear cells from PNG uninfected pregnant women had significantly higher antigen-specific IFN-γ TH1 responses (p=0.006) and secreted less pro-inflammatory cytokines TNF and IL-6 after PvLP2 stimulation than P. vivax-infected women (p<0.05). These data demonstrate that VIR antigens induce the natural acquisition of antibody and T cell memory responses that might be important in immunity to P. vivax during pregnancy in very diverse geographical settings.
| Naturally-acquired antibody responses to novel recombinant proteins and synthetic peptides based on sequences from P. vivax VIR antigens were evaluated in women from five distinct geographical regions endemic for malaria, during and after pregnancy. Levels of IgG to VIR antigens were indicative of cumulative malaria exposure and increased with current P. vivax infection and P. falciparum co-infection. Antibody data were consistent with levels of malaria endemicity and current prevalence in the diverse geographical areas studied. In addition, the magnitude of IgG response to two VIR antigens at delivery was associated with higher birth weight. Furthermore, T cell responses to VIR antigens were naturally induced and their magnitude varied according to P. vivax infectious status. Peripheral blood mononuclear cells from uninfected pregnant women from a highly endemic area produced higher TH1 (IFN-γ) and lower pro-inflammatory cytokines (TNF and IL-6) upon stimulation with a long synthetic peptide representing conserved globular domains of VIR antigens than P. vivax-infected women. Data suggest that further investigation on these antigens as potential targets of immunity in naturally-exposed individuals is warranted.
| Neglected for a long time, P. vivax malaria is raising more attention lately due to the increased recognition of its burden [1–4] and the renewed call for malaria elimination in endemic areas where P. vivax is an important source of malaria. Firstly, P. vivax is the most widely-spread of the human malaria parasites, with an at-risk population of 2.65 billion people [5]. Secondly, P. vivax infection is not as benign as traditionally believed, with severe malaria affecting a variety of population groups, including pregnant women in whom P. vivax infection has been associated with poor outcomes such as anemia, low birth weight (LBW) or congenital malaria [6–13].
The adverse consequences of malaria during pregnancy, the presence of parasites in the placenta and the molecular mechanisms of sequestration (parasite ligand and host receptor) have been well characterized in P. falciparum but to a lesser degree in the case of P. vivax infection. In P. falciparum infection during pregnancy, parasites may adhere to placental chondroitin sulphate A (CSA) through VAR2CSA, a member of the P. falciparum erythrocyte membrane protein 1 (PfEMP-1) family [14,15]. Thus, susceptibility to placental malaria has largely been attributed to a set of P. falciparum strains expressing VAR2CSA. Host immunity to this particular parasite protein has been associated with exposure to or protection against P. falciparum infection during pregnancy [16,17]. There is controversy about P. vivax cytoadherence properties, although we have reported placental P. vivax monoinfections in Papua New Guinea (PNG) with no signs of placental inflammation [18]. Rosetting seems a frequent cytoadhesive phenotype during P. vivax infections, which may contribute to the development of anemia in pregnancy [19,20]. Nevertheless, a P. vivax orthologue of the PfEMP-1 gene family and of VAR2CSA has not been described in parasites infecting pregnant women.
Like P. falciparum, the P. vivax genome contains subtelomeric multigene families. This includes the variant vir superfamily [21–23] with 295 vir pertaining to 10 subgroups [22,23]. From a structural point of view, vir genes differ greatly in size (156–2,316 bp in length) and number of exons (1–5). Unlike PfEMP-1, VIR proteins represent an extremely diverse family clustered in subgroups, which suggests different subcellular localizations and functions. These functions may include immune evasion [22], although P. vivax vir genes do not undergo allelic exclusion in contrast to the clonal variant expression of P. falciparum var genes [24,25]. Moreover, VIR proteins can localize to the surface of infected reticulocytes [21,26] and induce the natural acquisition of antibodies after infection [24,27]. Nevertheless, the host immune responses to VIR proteins and their association with malaria outcomes have not yet been extensively characterized, even less in pregnancy, partly due to the extent of their diversity and the difficulty to express them as recombinant proteins for immunoassays.
We have partially overcome these two problems by using the wheat germ cell-free expression system and by producing two long synthetic peptides containing conserved VIR sequences (PvLP1 and PvLP2) based on the P. vivax line Sal-I. This strain is originally from El Salvador, which was monkey-adapted. To overcome the sequence polymorphisms, we determined conserved globular domains of presently unknown function to synthesize PvLP1 and PvLP2 for testing in immune-epidemiological field studies with parasites from different origins.
A recent meta-analysis has highlighted the necessity of cohort studies representing diverse geographical regions in the field of P. vivax infections, to increase the body of evidence for protective immunity [28]. As part of the PregVax project, a multicenter study aimed at describing the burden of P. vivax malaria in pregnancy, we set out to study naturally acquired immune responses to VIR proteins during pregnancy. Women from five different P. vivax endemic countries in America (Guatemala, Colombia, Brazil), Asia (India) and South Pacific (PNG) were enrolled and antigen-specific immune responses assessed. We used VIR-based recombinant proteins as well as PvLP1 and PvLP2 for antibody and cellular immunoassays. We demonstrate that despite the large diversity of vir sequences, women from all regions mounted antibody responses to the VIR antigens that increased with P. vivax infection and past exposure. Moreover, women from the highest endemic region (PNG) had detectable VIR-specific cellular memory immune responses with distinct patterns according with P. vivax infection status. Altogether, data indicate that VIR antigens might be targets of immunity to P. vivax during pregnancy.
A total of 16 vir genes were selected to be cloned and expressed. Twelve one-exon genes were selected for practical reasons as genomic DNA could be used as template (S1 Table). In addition, four vir genes were selected after a protein BLAST against VAR2CSA domains, presenting 18.8–30.6% protein identity (S2 Table). Because the var2csa-homology regions of vir genes were always located in exon 2, only this exon was cloned. Of these 16 vir genes, four one-exon vir genes were discarded for protein expression: two of them (PVX_006080 and PVX_241290) could not be cloned as the PCR reaction did not work and another two (PVX_045190 and PVX_106220) did not present the expected sequence after cloning. With the classical E. coli expression system, PVX_086890 and PVX_069690 were poorly induced; and PVX_015640, PVX_067190, PVX_090290 and PVX_115485 were insoluble. Attempts to purify the six remaining partially soluble VIR proteins expressed in E. coli resulted in very low yields and not completely clean proteins. Therefore, these six VIR proteins were cloned in the pIVEX vector and further expressed in the cell-free wheat germ system but PVX_112125 showed mutations in the sequence. Thus, five vir genes were successfully expressed in the wheat germ cell-free expression system: vir25 (vir25-related, PVX_001610, group one-exon); vir14 (vir14-related, PVX_101615, group one-exon); vir2 (vir2/15-like, PVX_107750, group var2csa homology), vir24 (vir24-like, PVX_093720, group var2csa homology) and vir5 (vir5-related, PVX_124715, group var2csa homology). Soluble proteins were obtained of predicted sizes as detected by SDS-PAGE (Fig 1). Because expression of VIR proteins was not very productive, we designed two peptides containing conserved VIR sequences in order to perform the immunological assays.
A total of 1,511 peripheral blood samples collected at different time points (recruitment, delivery or postpartum) corresponding to 1,056 women, were analyzed for antibody responses. Unfortunately many of our samples were not paired due to low follow-up rates. The study population characteristics at baseline by country are provided in S3 Table. The infection rates by country and time-point are provided in Table 1. The amount of VIR14, VIR2 and VIR24 proteins produced was not sufficient to measure antibodies to them in all samples, therefore enrolment-delivery-postpartum matching samples were prioritized. The numbers of plasma samples per country for which anti-VIR antibody data were generated against each antigen and at different time point are summarized in S4 Table. For cellular assays, 53 samples (any gestational age including delivery, 18 P. vivax PCR negative, 28 P. vivax PCR positive, 7 unknown infection status) from the PNG pregnant cohort were included in the analyses.
Antibody responses to all VIR antigens were detected (value above negative control cutoff) in all sites and timepoints, except VIR25 and VIR5 at postpartum in India (IN) (Fig 2). IgG levels to all VIR antigens differed among countries at all timepoints (except VIR25 at postpartum) (one-way ANOVA p<0.05). PNG presented the highest magnitudes and prevalence, followed by Guatemala (GT).
Overall, VIR25 appeared to be the most broadly recognized antigen, with significant responses across all endemicities, even in countries like Brazil (BR) and IN where IgG responses to the other VIR antigens were very low (Fig 2). In addition, VIR14 and VIR2 showed consistent and comparable responses in PNG and GT at all timepoints, stronger than those to VIR25, and for VIR2 a peak was also detected at postpartum in Colombia (CO). Finally, VIR5 and PvLP2 only appeared to be considerably recognized by plasma from PNG, and PvLP2 (and PvLP1) also in CO at postpartum. The lowest responses were measured for VIR24, only detected at moderate levels for PNG. Of note, anti-VIR seroprevalence was in range with other P. vivax antigens such as Pv200L: BR: 7%; CO: 35%; GT: 40%; IN: 5%; PNG: 76%.
At recruitment, IgG responses to VIR proteins were closely correlated (Table 2), but they correlated poorly with antibody responses to VIR synthetic peptides or other P. vivax antigens such as Pv200L, which corresponds to a fragment of the merozoite surface protein 1. PvLP2 presented the highest correlation with Pv200L (Table 2) and other P. vivax antigens.
We considered whether anti-VIR responses were due to cross reactivity with other Plasmodium antigens. To assess this, we studied the correlation between anti-VIR responses and antibody responses to 9 P. vivax and 6 P. falciparum additional antigens. Of note, low correlations were found between anti-VIR responses and other anti-Plasmodium responses, suggesting that there was no cross-reactivity (S5 table).
There were higher levels of anti-VIR24 IgGs at delivery, and more anti-VIR2, anti-VIR25 and anti-PvLP1 antibodies at postpartum, compared to recruitment levels, although overall differences using the Wald test were only significant for PvLP1 (S6 Table).
We assessed how different pregnancy variables affected the IgG responses to VIR antigens. Antibody levels to VIR5 were significantly associated with gravidity (proportional differences by [category group of previous pregnancies] [0]: 1; [1–3]: 1.43, 95% CI: 0.91–2.25; [4+]: 0.71, 95% CI: 0.34–1.48, p=0.033). IgG responses to PvLP1 and PvLP2 were significantly associated with present malaria infections (Table 3). Of note, the association with co-infections (P. vivax and P. falciparum) was higher than with mono-infections (P. vivax alone). However, sample size for co-infections was small, especially at delivery, and these results should be considered cautiously. The magnitude of VIR-specific IgG response did not show associations with age and gestational age (p>0.05).
We also analyzed the association between antibody levels at recruitment and future infection (at delivery). Women with higher PvLP1 antibody levels at recruitment had a higher probability of having a P. vivax infection at delivery (per doubling antibody levels, OR=1.84, 95% CI=1.11; 3.04, p=0.017, adjusted analysis).
Finally, we studied the association between antibody levels and pregnancy outcomes, i.e. hemoglobin (Hb) levels and birth weight. A borderline significant positive association between PvLP2 antibody levels at recruitment and birth weight was observed by unadjusted regression analyses (Table 4). At delivery, IgG responses against two VIR proteins with homology to VAR2CSA (VIR2 and VIR24) were positively associated with birth weight in the adjusted analysis (Table 4). No associations were found between antibody levels and Hb levels at delivery (p>0.05).
Peripheral blood mononuclear cells (PBMC) from PNG women with current P. vivax infection had a significantly lower percentage of IFN-γ-producing CD4+ and CD8+ T cells than uninfected women when stimulated with PvLP2, as assessed by intracellular cytokine staining by flow cytometry (Fig 3). No differences in % IFN-γ+ CD4+ T cells between infected and non-infected women were observed when stimulating PBMCs with PvLP1 or in the medium and anti-CD3 controls, and significant but much lower differences compared to PvLP2 stimulus were observed in % IFN-γ+ CD8+ T cells for PvLP1 and medium control (Fig 3).
We also measured the concentration of cytokines, chemokines and growth factors secreted in PBMC cultured either with medium, PvLP1 or PvLP2. Infected women produced more G-CSF and IL-4 than those from uninfected women, independently of the stimulus (Table 5). In addition, supernatants contained more IFN-γ when PBMC were cultured with medium or PvLP2, although the median value was the same, suggesting that differences were not very high. Of note, PBMCs stimulated only with PvLP2 secreted specifically more pro-inflammatory cytokines TNF, IL-6 and regulatory cytokine IL-10 in the infected group than in the uninfected group, although the difference did not reach statistical significance for IL-10 (p=0.062). No differences between the infected and uninfected cohorts were observed in the anti-CD3 control, although overall values in this positive control were higher than in the other three culture conditions.
Considering the large genetic diversity of P. vivax strains [29] and the effect that polymorphisms in host genes such as HLA can have on immune responses to certain antigens [30], it is important to evaluate antibody and cellular immune responses to potential targets of immunity in different geographical populations. Here, significant levels of anti-VIR antibodies were detected in pregnant women from five countries with very diverse endemicity and transmission rates, further supporting the immunogenic properties of VIR antigens previously reported in non-pregnant Brazilian women [21,24,27]. This is remarkable if we consider that the Sal-I genome was used as a template for the production of all recombinant VIR proteins and suggests that despite the high sequence variability in the VIR proteins and the P. vivax circulating strains, B cell epitopes might be sufficiently conserved. This is further supported by the immunogenicity of long synthetic peptides representing conserved globular domains of VIR proteins, particularly PvLP2. We cannot, however, exclude the possibility that low responses to some VIR proteins in particular settings are due to lack of VIR expression or that they contain less B-cell epitopes as opposed to the inability to develop a VIR-specific immune response upon exposure to that variant.
Expressing recombinant Plasmodium proteins using different expression vectors has shown to be a challenging endeavor [31], especially achieving expression of soluble and correctly folded proteins is even more difficult. The cell-free wheat germ expression system used here has proven to be an excellent system to produce soluble and correctly folded proteins [32,33]. In fact, expression of enzymes from the human genome consistently showed that they retain enzyme activity [34]. In spite of these advantages, vir genes are highly AT-rich and several different attempts to express all the genes listed in S1 and S2 Tables using this or other systems such as cell-free and cell-based E. coli have failed.
PNG presented the highest intensity and prevalence of antibody responses against all antigens, despite P. vivax infection rates not being much higher at the time of the study within this cohort than in the other five countries [35]. The fact that asymptomatic infected women were not given treatment in PNG does not explain this difference, as the prevalence of P. vivax infection by microscopy in PNG was only 1%. Nevertheless, antibodies are a reflection of cumulative exposure, and PNG is indeed the country among the five with the highest malaria endemicity historically, even if during the PregVax study P. vivax prevalence was lower than in the past. In addition, regression analyses showed that co-infections with P. falciparum had a higher positive association with PvLP1 and PvLP2 antibody levels than P. vivax mono-infections. PNG had the highest P. falciparum microscopic infection rate in this cohort, suggesting these may boost anti-P. vivax responses. In our cohort we could rule out mostly although not totally undiagnosed submicroscopic co-infection and vir genes do not have orthologues in P. falciparum. It might be that this co-infection boosting effect is due to P. vivax-specific B cell bystander activation by noncognate T cells, which could be induced under conditions of persistent priming by P. falciparum antigens [36,37]. If this was the case, it may be interesting to consider this effect in programmatic terms regarding the search of a malaria vaccine. However, it is also possible that co-infections and higher antibody levels are just two parallel markers of higher previous exposure in some women.
Plasmas from GT also presented significant levels of IgG antibodies to various VIR proteins. This is consistent with P. vivax positivity rates by PCR at the population level, which were the highest in GT and PNG in the whole PregVax cohort. Infections in GT were largely submicroscopic but sufficient to induce detectable antibody responses. There was heterogeneity with regards to recognition of the VIR antigens among countries. VIR25 was the most broadly immunogenic, being recognized in distant geographical regions, suggesting the presence of conserved and/or cross-reactive epitopes within its sequences. However VIR14 and VIR2 induced the highest levels of antibodies, though restricted to the two most endemic countries. Those three proteins (VIR25, VIR14 and VIR2) appeared to induce longer-lived antibodies as they were clearly detected in populations with high infection rates in the past but low at the time of sampling. Antibodies to VIR24, VIR5 and PvLP2 were only clearly present in PNG, and this might indicate geographical diversity in immunogenicity of epitopes and/or their even longest-living nature. IgG antibodies against PvLP1 and PvLP2 were detected in all countries and timepoints but not at high levels. A peculiar pattern was observed in CO, where a significant increase in responses to most VIR antigens, but particularly VIR2 and PvLP2, occurred at postpartum. This likely reflects increased parasite prevalence at a population level at this time rather than a booster of VIR responses after pregnancy. However, at an individual level we did not find an association between antibodies to VIR proteins and infection status, which probably reflects the diversity of vir genes in the P. vivax genome [38]. In contrast, levels of PvLP1 and PvLP2 were associated with present vivax malaria infections. Both antibody levels but specially anti-PvLP2 correlated well with other markers of malaria exposure. This suggests that the design of these peptides (based on conserved sequences) might have helped overcome the problem of having a large and variable gene family. Thus, collectively the data showed that VIR antigens could be markers of exposure at a population level.
We also assessed association between antibodies and protection at the individual level, although this is often difficult as heterogeneity of exposure is not properly assessed and accounted for in field designs. In fact, higher PvLP1 antibodies at recruitment were associated with more risk of infection at delivery, being a correlate of risk rather than of immunity. We have previously reported that higher levels of antibodies in some individuals may indicate those who have had previous malaria episodes and are at higher risk of future episodes if past exposure is not well adjusted for [39,40]. Nevertheless, we also found some indications for a potential protective role of VIR antibodies in malaria in pregnancy outcomes: (i) a borderline significant positive association between PvLP2 antibody levels at recruitment and birth weight and (ii) a positive association of antibody levels to VIR2 and VIR24 (of partial sequence homology to P. falciparum VAR2CSA domains) at delivery and birth weight. Placental P. vivax infection has been reported [18], as well as P. vivax adhesion to CSA [19] and inhibition of P. vivax cytoadhesion using soluble CSA [41]. However, whether VIR2 and VIR24 proteins also bind CSA and are implicated in vivax malaria during pregnancy remains speculative. Unfortunately, our study was not designed to demonstrate any protective role of antibody responses to VIR antigens in vivax malaria and therefore we cannot draw any conclusion.
We present some evidence supporting a relationship between antigen-specific cytokine responses, infection and immunity in PNG pregnant women. Our data show lower PvLP2-specific IFN-γ+ CD4+ T cell frequencies and higher secretion of TNF, IL-6 and IL-10, in P. vivax-infected pregnant women compared to uninfected women and this was not seen for PvLP1 or the control stimuli. IFN-γ has been shown to be essential for controlling experimental malaria infections in mice (reviewed in [42]) and clinical P. falciparum infections in humans [43,44], and IL-10 is a key regulatory cytokine that prevents excessive inflammation but might contribute to the lack of control of infections. Thus, the fact that non-infected women had higher PvLP2-specific TH1 cell frequencies and lower IL-10 production could mean that cellular responses induced by this antigen (for instance by a potential vaccine) could help in controlling vivax infections. Nevertheless, we also observed PvLP2-specific increase of pro-inflammatory cytokines IL-6 and TNF in infected women. IL-6 has been shown to skew T cell differentiation towards TH2 and TH17 [45], which would explain why we observe a decrease of TH1 frequencies. Thus we can assume that VIR epitopes present in PvLP2 trigger the natural acquisition of cellular memory immune responses, but whether these are protective or just markers of exposure can not be concluded from the data presented.
This study presents some limitations. First, samples were not fully paired and sample size was different for some antigens/analyses. Unfortunately, many of these women lived in rural areas far from the hospital. It is highly complex to get full attendance to all antenatal clinics and, after puerperium, is even more complicated. In spite of this, we believe the cohort is quite unique and very valuable to demonstrate the immunogenicity of VIR antigens in different geographical settings. Second, due to its exploratory nature, we did not have statistical power to demonstrate strong associations between anti-VIR immune responses and protection against infection nor poor outcomes, as it was designed to be a first descriptive investigation of adaptive immune response (antibody and T cells) to VIR antigens during pregnancy. Third, multiple comparisons were not corrected for in all statistical assays and results are interpreted for internal coherence and biological plausibility.
In summary, we present the first comprehensive study on immune responses to VIR antigens demonstrating that VIR sequences are the target of the natural acquisition of antibody and cellular responses affected by exposure to malaria infection in five distinct endemic areas. VIR 25 seems to be broadly recognized and we demonstrate that PvLP1 and PvLP2 can be used to profile antibody and cellular immune responses to VIR sequences, overcoming the problem of the large number of diverse VIR proteins. Based on our findings and the large burden of vivax malaria, we believe that larger prospective cohort immune epidemiological studies are needed to specifically address whether VIR-based antigens are targets of protective immunity against the neglected P. vivax parasite and could be considered as candidates for vaccine development towards malaria elimination.
This study was performed in the context of the PregVax project (FP7-HEALTH-201588, www.pregvax.net), a health facility-based cohort study of pregnant women to describe the burden and impact of P. vivax in pregnancy, conducted between 2008 and 2012 in five endemic countries: BR, CO, GT, IN and PNG. Approximately 2,000 women per country were enrolled at the first antenatal visit (recruitment), and followed up until delivery. Symptomatic Plasmodium spp. infections at any time during pregnancy were also recorded though passive case detection. A random subpopulation corresponding to 10% of the entire PregVax cohort was allocated to the “immunology cohort” and was further followed up until at least 10 weeks after delivery (postpartum group). In all visits, Hb levels, P. vivax and P. falciparum parasitemias by blood smear and malaria symptoms were assessed. Giemsa-stained thick and thin blood slides were read onsite following WHO standard quality-controlled procedures to establish parasite presence. External validation of a subsample of blood slides from each country was done at the Hospital Clinic and at the Hospital Sant Joan de Deu, in Barcelona, Spain. Birth weight was recorded. Women with a positive smear were treated according to national guidelines, except in PNG where blood smears could not be read at the moment of the visit for logistical reasons (only symptomatic women were thus treated after confirmation of infection by rapid diagnostic test). The protocol was approved by the national and/or local ethics committees of each site, the CDC IRB (USA) and the Hospital Clinic Ethics Review Committee (Barcelona, Spain). Written informed consent was obtained from all study participants.
A venous blood sample (5–10 mL) was collected aseptically in heparinized tubes from the “immunology cohort” at recruitment, delivery and postpartum visits. Submicroscopic P. vivax and P. falciparum infections were also determined by real time-polymerase chain reaction (PCR), except for the Indian samples, where only P. vivax infection was examined. Submicroscopic infections were only analyzed in a random sub-sample of the cohort. Additionally, blood samples (10 mL) were collected from 39 malaria naïve donors at the blood bank in Hospital Clinic (Barcelona, Spain), and used as negative controls.
Plasma was separated by centrifugation and stored at -80°C. Blood cells from PNG were further fractioned in a density gradient medium (Histopaque-1077, Sigma-Aldrich) to obtain PBMCs and stored in liquid nitrogen. Samples from GT, CO, BR and PNG were analyzed at ISGlobal (Barcelona, Spain) while plasmas from IN were analyzed in Delhi.
Vir genes were amplified from genomic DNA (Sal I strain) by PCR using “PCR Supermix” (Life Technologies). PCR products were introduced in the pIVEX1.4d vector (Roche) previously modified by inserting glutathione S-transferase (GST) after the 6xHis tag sequence. Authenticity of all clones encoding GST-VIR fusion proteins was confirmed by double-strand sequencing before expression in the wheat germ system. Thus, GST-fusion proteins contain open reading frames encoding the predicted VIR proteins.
Primers used for gene amplification are listed in S1 and S2 Tables.
Proteins were expressed with a GST tag using the wheat germ cell-free system as described [46]. Expressed proteins were purified on GST SpinTrap purification columns (GE Healthcare), and eluted proteins were dialyzed in phosphate buffered saline (Tube-O-DIALYZER, GBiosciences). GST was also expressed separately for immune-reactivity control. Pv200L (P. vivax merozoite surface protein 1, fragment 121–416) was produced as previously described [47]. The rest of Plasmodium antigens were produced as described previously [48]. The design and synthesis of P. vivax long synthetic peptides (PvLP) representing conserved central core (PvLP1) and C-terminal (PvLP2) VIR motifs, has been reported previously [26]. The sequences are detailed in S7 Table.
Measurement of plasma IgG antibodies was performed by multiplex suspension array using the Luminex technology, as described [46]. MagPlex magnetic carboxylated microspheres (Luminex Corporation, TX, USA) were covalently coated with 3 μg of protein/peptide per 1.1–1.4 million beads following manufacturer’s instructions. Beads were quantified in a Guava Flow Cytometer (Millipore) and mixed in equal amounts. A unique batch of microspheres was prepared for the whole study, including the samples analyzed in IN. Circa 1000 beads per analyte were incubated with plasma (1:100 dilution) in duplicates, and subsequently with anti-human IgG-biotin (Sigma-Aldrich), followed by streptavidin-conjugated R-PE (Fluka, Madrid, Spain). Beads were acquired on the BioPlex100 system (Bio-Rad, Hercules, CA), and results expressed as median fluorescence intensity of duplicates. Value against GST alone was subtracted for VIR proteins. Raw GST values are presented in S1 Fig. Cross-reactivity was ruled out in a pilot study analyzing a subset of plasmas in singleplex and multiplex. Samples in IN were analyzed with identical protocols and instruments.
Except where indicated, all reagents were purchased from BD Biosciences. PBMCs were thawed, rested for 10–12 h and viability assessed with Guava ViaCount Reagent (Millipore). Only samples with viability >70% were used for assays. Half a million cells per well were resuspended in RPMI-1640 medium plus 10% fetal bovine serum (culture medium) and incubated with PvLP1 or PvLP2 (5 μg/mL). Culture medium was used as negative control and anti-CD3 as the positive control. After 12 h, an aliquot of 30 μL of culture medium supernatant was collected to measure secreted cytokines, while an equal volume of media containing GolgiPlug was added for additional (4 h) incubation. PBMCs were stained with LIVE/DEAD Fixable Violet Dead (Life Technologies), anti-CD14 Pacific Blue, anti-CD19 Horizon V450, anti-CD4 allophycocyanin (APC) and anti-CD8 Peridinin Chlorophyll Protein Complex (PerCP). After washing, cells were fixed and permeabilized with Cytofix/Cytoperm, and incubated with anti-CD3 phycoerythrin (PE)-Cy7, anti-interferon (IFN)-γ PE and anti-CD69 fluorescein isothiocyanate (FITC). Cells were acquired in a LSRFortessa flow cytometer and data were analyzed by FlowJo (FlowJo LLC, OR, USA). Gating strategy is provided in S2 Fig. Supernatants were frozen at -80°C until Luminex analysis with the Cytokine Magnetic 30-Plex Panel (Invitrogen), according to manufacturer’s instructions.
Samples from BR, CO, GT, and half of the samples from PNG were analyzed at the Istituto Superiore di Sanità (Rome, Italy), as described [12]. The threshold for positivity for each species was established as a cycle threshold<45, according to negative controls. P. vivax diagnosis for IN samples was performed in Delhi following Rome’s protocol adapted for the instrument sensitivity (3rd step amplification 72°C for 25 sec instead of 72°C for 5 sec). Approximately half of the PNG samples were analyzed for submicroscopic infections in Madang, following a similar protocol to Rome’s [49], except that the threshold for positivity for each species was established as cycle threshold<40, according to negative controls. DNA was extracted from whole blood-spot filter paper.
Any Plasmodium infection was defined as a positive smear by microscopy and/or positive PCR. One-way ANOVA test was used to evaluate the differences in antibody levels among countries, and Chi-squared or Fisher’s exact tests to evaluate the differences in percentages of individuals with a positive antibody response (values above the mean plus 3 standard deviations [SD] of Spanish controls, cutoff). To assess the differences on antibody levels between non-pregnant and pregnant women at recruitment and delivery, multilevel mixed-effects linear regressions were estimated with the samples from the five countries. Timepoint (recruitment, delivery and postpartum) was the fixed independent variable, while inter-site (country of origin) and inter-subject variability were estimated as random parts. To study the association between antibody levels and pregnancy variables, univariate (only adjusted for country of origin) and multivariate linear regression models were estimated with the variables country, age, gestational age, gravidity (number of previous gestations) and P. vivax or P. falciparum infection during pregnancy (only accounted past or present infections but not future infections). The correlation between IgG responses to different antigens was evaluated with the Spearman's rank test. The association between IgG levels at enrolment and future malaria infections was evaluated with logistic regression models. The association between antibody responses at enrolment and delivery, and Hb levels at delivery and birth weight, were analyzed using univariate and multivariate linear regression models, adjusted by country, Hb at recruitment, gestational age at recruitment, age, gravidity and past or present Plasmodium infection during pregnancy. For the cellular and cytokine analyses, deviation from normality was tested using the Skewness and kurtosis test. Because none of the variables except IL-13 presented a normal distribution, data was presented as medians and comparisons between groups were done using the U-Mann-Whitney test. Cytokine/chemokine production in culture supernatants of unstimulated samples (medium) was not subtracted from the stimulated samples but shown side by side as it is possibly biologically relevant. Significance was defined at p<0.05. Crude p values are interpreted for internal coherence, consistency of results and biological plausibility. Analyses were performed using Stata/SE 10.1 (College Station, TX, USA).
|
10.1371/journal.pntd.0006580 | The impact of the Ebola virus disease (EVD) epidemic on agricultural production and livelihoods in Liberia | There is unequivocal evidence in the literature that epidemics adversely affect the livelihoods of individuals, households and communities. However, evidence in the literature is dominated by the socioeconomic impacts of HIV/AIDS and malaria, while evidence on the impact of the Ebola virus disease (EVD) on households’ livelihoods remains fragmented and scant. Our study investigates the effect of the EVD epidemic on the livelihoods of Liberian households using the Sustainable Livelihood Framework (SLF). The study also explores the effect of the EVD epidemic on agricultural production and productive efficiency of farm households using Spatial Stochastic Frontier Analysis (SSFA). We collected data from 623 households across Liberia in 2015, using a systematic random sampling design. Our results indicated that the annual income of sample households from communities where EVD occurred did not differ from the annual income of households from communities where EVD did not occur. Nonetheless, the majority of sample households reported a decrease in their income, compared to their income in the year before the survey. This suggests that the impact of the EVD epidemic might not only have been limited to communities directly affected by the epidemic, but also it may have indirectly affected communities in areas where EVD was not reported. We also found that the community-level incidence of EVD negatively affected crop production of farm households, which may have exacerbated the problem of food insecurity throughout the country. Moreover, we found that the EVD epidemic weakened the society’s trust in Liberian institutions. In a nutshell, our results highlight that epidemics, such as the recent EVD outbreak, may have long-lasting negative effects on the livelihoods of a society and their effect may extend beyond the communities directly affected by the epidemics. This means that the nation’s recovery from the impact of the epidemic would be more challenging, and the social and economic impacts of the epidemic may extend well beyond the end of the health crisis.
| Epidemics such as HIV/AIDS, malaria and Ebola virus disease (EVD) may adversely impact the livelihoods of the society affected by the epidemics. Nonetheless, the mechanism behind the effects of the epidemics may differ depending on different factors, such as the transmission mechanisms, latency, and mortality rates associated with the diseases, which requires specific research to investigate the effect of each epidemic. In light of this, we analyzed the impact of the recent EVD epidemic on the agricultural production of farm households and its impact on the livelihoods of Liberian society. We collected data from 623 households throughout Liberia during the EVD crisis in 2014–2016, and found that there was no significant difference in the annual income of sample households from communities where EVD occurred and did not occur. Nonetheless, the majority of the sample households reported a decrease in their income compared to the year before our survey. We also found that the community level incidence of EVD had a significant negative effect on crop production of farm households, which might have exacerbated food insecurity in the country. Moreover, the EVD epidemic negatively affected the Liberian society’s trust in Liberian institutions. Our results underline that epidemics, like EVD, might have long-lasting negative effects on the livelihoods of a society, and they may have adverse effect beyond the communities directly affected by the epidemics.
| There is a plethora of evidence in the literature that epidemics such as HIV/AIDS and malaria have profound implications for the livelihoods of the affected society. The impact of HIV/AIDS on livelihoods has been intensively investigated and there is universal consensus that HIV/AIDS adversely affects the livelihoods of individuals, households and communities [1–2], and has macro-level implications for poverty, economic growth, unemployment and political stability [3–7]. Similarly, malaria has been found to have a strong negative effect on the socioeconomic status of households [8–11]. In contrast, the socioeconomic impacts of the Ebola virus disease (EVD) epidemic, specifically the most recent and largest ever EVD epidemic recorded in West Africa from 2014–2016, have not been systematically analyzed. Due to the differences in transmission mechanisms, latency, and mortality rate between EVD and other infectious tropical diseases, such as HIV/AIDS and malaria [12–14], EVD outbreaks likely impact the livelihoods of a society differently. For example, EVD can wipe out an entire family or village within a relatively short period of time. In areas affected by EVD, economic activities may cease completely, as people no longer work on their fields, nor trade or even travel (because of check points and travel restrictions) [15]. HIV/AIDS infections and resultant mortalities, on the other hand, occur over a longer time period; and as such, their effects on livelihoods and the economy are more subtle at first. HIV/AIDS and malaria result in higher costs in terms of the opportunity cost of the time spent caring for the sick household member [16–18], medical expenses and, for the unlucky ones, funeral expenses. In contrast, EVD lowers livelihood outcomes by weakening the ability of the households to earn their living rather than by increasing the expenditure on the sick person and funeral ceremonies. For example, the medical costs of the EVD epidemic in Liberia were mostly covered by the government and the international community, as the epidemic presented a global health emergency. These and other disease-specific characteristics necessitate specific research to investigate the effect of different infectious diseases on the livelihood of the affected societies. In light of this, we analyze the impact of the recent EVD epidemic on the livelihoods of the Liberian society.
There are few studies that explored the impact of the EVD epidemic on the agricultural sector in Liberia [19–20]. These studies reported that the EVD epidemic negatively affected employment in the agricultural sector. At the peak of the epidemic, almost half of the country’s labor force was out of work [19–20]. Farmers were less likely to work on their farms during the EVD epidemic [19]. These studies found that most of the households returned to their farms during the survey, which was conducted from October, 21 to November, 7, 2014, and concluded that the impact of the EVD crisis on the agricultural sector may not have been as severe as predicted [19–20]. Nonetheless, these studies used phone surveys to collect data, which could have resulted in selection bias, as households without mobile phones were systematically excluded from the survey. In addition, the emphasis of the studies was focused on the impacts of the EVD epidemic on the employment in the agricultural sector. However, employment and whether or not households were working on their farm tell only part of the story. Even if households were working on their farms during the EVD epidemic, the productivity of their agricultural inputs, and hence their efficiency may have been compromised by the epidemic. Therefore, our study provides evidence on how the Ebola crisis affected the efficiency of farm households, and the concomitant effects on agricultural production in Liberia. Moreover, our study explores the impact of the EVD epidemic on livelihoods of the Liberian society using the sustainable livelihood framework (SLF) [21–22].
The EVD epidemic affects the livelihoods of individuals, households and communities by weakening the household assets upon which the households’ ability to enhance their livelihood, depends [21]. These assets can broadly be categorized into five categories: physical capital (e.g., infrastructure, tools, equipment), human capital (e.g., knowledge and ability to work), financial capital (e.g., available stocks, access to financial services, regular inflows of money), social capital (e.g., networks for cooperation, trust, support) and natural capital (e.g., land, forests, water) [21–24]. Shocks, such as epidemics, that weaken some or all of these household assets, negatively affect livelihood [2, 25]. Therefore, a complete understanding of the effects of epidemics, such as EVD, on the livelihood of households requires the investigation of their influence on the assets owned by the households.
The EVD epidemic may have weakened the resource base of the Liberian society for various reasons. First, the incidence of the epidemic in the households and/or their communities may have weakened the different categories of assets owned by households directly [2, 24]. Second, measures taken by the Liberian government to contain the spread of the disease may have further dampened the households’ assets and affected their welfare [15, 26, 27]. For example, the government declared a state of emergency and established quarantine zones in most of the affected communities. Schools and markets were closed in several districts and communities. Restrictions on domestic and international travels were imposed [15, 19, 26, 27]. Thus, the mobility restrictions and complete closure of markets might have considerably hampered the livelihoods of individuals, households and communities by reducing their access to different livelihood assets [15, 26]. Third, fear of contracting the disease may have coerced people into avoiding social gatherings and participation in different activities and organizations, thereby weakening the social capital that the Liberian society possesses [27]. Furthermore, during crises, people may increasingly seek support from different social networks such as friends, families, the community, and the government, but the prospect of receiving the needed help may have been significantly hampered by an epidemic, which strains the social cohesion. Stigmatization of survivors of the disease may also contribute to the degradation of the social capital.
Our study uses a systematic nationwide random sampling design to explore the effect of the EVD epidemic on the livelihood assets possessed by Liberian households, and the livelihood outcomes they achieved during the EVD epidemic. In addition, we emphasize the effect of the epidemic on the agricultural sector. We are interested in the effect of the EVD epidemic on agricultural production for two main reasons. First, the agricultural sector plays a crucial role in Liberia and contributes more than 35% of the country’s GDP [28]. Moreover, the majority of Liberians live in rural areas (50%), and are primarily engaged in the agricultural sector to earn a living [20]. Almost 80% of rural households and 18% of urban households are agricultural households in Liberia [29]. Second, we are not aware of any study that examined the effect of the recent EVD epidemic on agricultural production in Liberia or elsewhere accounting for the productive efficiency of farm households.
Liberia is one of the poorest countries in Africa with per capita GDP of $457.9 as of 2014 [30]. Nevertheless, after the end of the civil war in 2003, the country’s economy has been steadily growing. For instance, the economy grew by an annual growth rate of 8% in 2006 and 8.7% in 2013 [30]. Although most of the population lives in poverty (68.6% of the population lives on less than $1.9 a day), Liberia had one of the fastest growing economies in Africa over the past 10 years [31]. This gain of momentum in terms of macro-economic performance was disrupted in 2014 by the epidemic of EVD [32], which affected the country from 2014 to 2016 and resulted in the tragic loss of 4,809 lives (45% of reported cases) [33]. All 15 counties reported incidences of the disease, but the severity of the epidemic varied from place to place [33]. For example, based on the number of EVD cases per 1,000 inhabitants, counties such as Margibi, Montserrado, Grand Cape Mount, and Bomi were more severely affected by the epidemic than Grand Gedeh, River Gee, Sineo and Maryland [33] (Fig 1).
We conducted a systematic nationwide household survey from February to June, 2015. The survey was administered in person by trained Liberian enumerators. Sample households were randomly selected following a “random walk” procedure [e.g., 34, 35]. Starting from the center of a village/town/city (hereafter ‘interview location’), the enumerators walked in different directions and randomly selected households to be interviewed. We aimed at interviewing 5–10 households per location, depending on their size (i.e., more interviews in larger locations). The random walk technique was used to reduce non-response rates, as the enumerators would walk until they found enough households that were willing to participate in the interview. This method is particularly useful in sensitive times like the EVD crisis, when people may have been more reluctant to interact with strangers out of fear of contracting the EVD. Nonetheless, we acknowledge that in larger locations, the random walk method may have resulted in some biases, as households closer to the center of locations were more likely to be sampled than those living farther away from the centers.
Whenever possible, household heads were selected and interviewed. A total of 623 sample households were interviewed. We collected data through face-to-face interviews with sample households across Liberia using a questionnaire (S6 Appendix). We were granted permission to conduct the survey by Liberian authorities after careful evaluation of staff safety, data collection procedures and agreements on data sharing (see S7 Appendix). Additional data on EVD deaths and cases were obtained from the World Health Organization (WHO) and the Liberia Institute of Statistics and Geo-information Services (LISGIS).
To explore the impact of EVD on the livelihoods of Liberian society, we applied the Sustainable Livelihoods Framework (SLF) used by the Department for International Development-United Kingdom (DFID-UK) [21–22]. We used the SLF as it enables us to understand not only the effects of the EVD epidemic on livelihood outcomes, but also the mechanisms driving the effects of the epidemic [21–22]. Thus, our study provides important insight into policies aiming to avert or reduce the impact of future epidemics on livelihoods by addressing the important factors that drive these effects. Although there is a scarcity of studies that applied this framework to investigate the impact of EVD, several studies employed the SLF to explore the livelihood effects of other epidemics, mainly HIV/AIDS [1, 2].
Traditionally, the SLF has been applied to understand differential capabilities of rural families to cope with stresses or shocks [36] and their ability to achieve sustainable livelihoods. A livelihood is defined as a means of living, and the assets required to achieving it [21–22]. The different types of assets that a household needs to achieve better livelihood outcomes are broadly categorized as human, physical, financial, social and natural capitals [21, 22, 24]. Hence, the likelihood of a household to achieve improved livelihood outcomes (such as income, food security and others) depends on its access to different livelihood assets. The livelihood of a household is deemed sustainable when it copes with, and recovers from stresses without compromising the abilities of future generations [21]. Factors that weaken some or all of the livelihood assets of households, adversely affect their livelihood [2, 25]. Therefore, a complete understanding of the effect of shocks, such as epidemics, on the livelihood of households necessitates the investigation of their influence on the assets owned by the households. Employing the SLF, we investigate the impact of the EVD epidemic on different categories of assets possessed by Liberian households, and the livelihood outcomes they achieved during the EVD epidemic (i.e., in the 12 months preceding our survey). Here, we focused on the effect of the EVD epidemic on three categories of household capitals (natural, financial, and social capitals), and their total income and agricultural production. See Table 1 for the definition of the household assets included in our study.
To analyze the data, we used descriptive statistics, a factor analysis and regression models. We used Stata 10.1 [53] and R 3.2.4 [54] for our analysis. In our analysis self-reported incidence of EVD in the community of the respondent was used as a proxy for EVD occurrence. Respondents were asked whether they knew anybody in their own or nearby communities who had contracted EVD. We used self-reported incidence of EVD in one’s community instead of EVD cases reported by WHO, as the publicly available WHO reports are at the spatial resolution of the county-level and not community-level. In our understanding, community-level incidence might be more relevant for household-level decisions, because the EVD incidence at the community spatial scale may have a larger impact on household livelihood than the EVD occurrence at the county spatial scale.
Our analysis is divided into two parts. First, we analyzed the effect of the EVD epidemic on livelihood outcomes, such as total annual household income and agricultural production, using descriptive statistics and a regression analysis. Second, we analyzed the impact of the epidemic on the households’ assets. Here, we used descriptive statistics to summarize the effect of the EVD epidemic on the resource bases of households, and a regression analysis to analyze changes in social capital. For the summary of the methods used see S1 Appendix Table A.
In our regression analysis, we employed spatial econometric models to account for potential spatial dependencies in the data using the “spdep” package in R [55]. Before we ran the models, we tested for the significance of the spatial correlation of the residuals obtained from the mixed-effects models (with random slopes and intercepts) using Moran’s test (see S1 Appendix Table D). The test results suggested that there was a significant spatial correlation in our data, justifying the use of spatial econometric models. We further checked for the suitability of spatial-lagged models and spatial-error models, but found no significant difference between these models. Hence, we report results from spatial-lagged models here and include the results from the spatial-error models in the appendix (see S1 Appendix Table E Models 3 and 6; and Table G model 3).
To estimate the effect of the EVD crisis on agricultural production, we used a spatial stochastic production frontier model with the “ssfa” package in R [39]. The application of the stochastic frontier model (instead of the classical linear regression model) was justified after testing for the significance of the existence of inefficiencies in our agricultural production data using the likelihood ratio test (LR test: chi-bar square (1) = 6.396, p = 0.006).
Finally, we conducted power analyses and found that the power of our tests ranged from 73–98%, indicating a relatively low probability of conducting Type II error.
The majority of our respondents (90%) were male with an average age of 43 years. Almost 64% of the respondents were literate, which is comparable to 66.7% reported by LISGIS [56]. Respondents had attended school for an average of six years. Sample farm households owned, on average, 1.53 hectares of farm land, which is similar to the average farm size of 1.54 hectares reported by the FAO [57]. Forty-two percent of the sample respondents were from urban areas and 58% from rural areas (for more detail see S1 Appendix Table B). Thirty-one percent of the sample households reported the incidence of EVD in their community or nearby communities and almost 20% of these were households that relied heavily on farming for their livelihoods.
Our study offers important insights on the effect of the EVD epidemic on the livelihoods of Liberian society and the mechanisms underlying them. We found that the incidence of EVD did not influence the total annual household income depending on whether or not the households were located in/near communities where EVD occurred. However, the majority of the sample households reported that their income was lower during the EVD crisis, as compared to their income before the outbreak. This suggests that the effects of the EVD epidemic were not limited to the communities where EVD occurred, but that the EVD crisis affected communities throughout the country. These results are in line with the findings of Bowles et al. (2016) [63] who reported that during the EVD epidemic, there was a remarkable decline in economic activities across Liberia, but that in most cases, there was little association between the decline in economic activities and the number of Ebola cases. Thus, post-epidemic rehabilitation measures should not only be limited to communities directly affected by EVD, but should also target those indirectly affected by the epidemic. We also found that the incidence of Ebola significantly reduced total crop production of farm households, which is in line with other studies [see also 15, 26, 58, 64]. As farm households in our sample had consumed about 85% of their own production, they heavily depended on their own agricultural production for survival, which is a typical characteristic of subsistence farmers. Thus, the reduction in their agricultural production likely had an adverse effect on their food security, which is in line with the existing literature [15, 20, 26, 35, 58, 64]. Most of these studies reported that the agricultural sector was one of the most severely affected sectors by EVD, and food security was significantly hampered by the epidemic in Liberia.
Our results also revealed that the incidence of EVD negatively affected the trust of the citizens in Liberian institutions. Respondents who reported the incidence of EVD in/near their community, were more likely to report a decrease in their trust in the government and village chief(s). Our results suggest that in the long term, the deterioration in the social capital resulting from the EVD epidemic, may have adverse effects on the stability of the country’s political system [60]. Degradation of social capital may increase the likelihood of social conflict and crime. Moreover, distrust in state institutions may render recovery efforts more challenging and make the country more susceptible to future outbreaks, as citizens may no longer comply with the recommendations of the state institutions [27].
Finally, although we controlled for most of the relevant factors in our analysis, there may still be some confounding effects, as our results were based on cross-sectional data collected at a single point in time. In addition, as it is customary to the surveys of our type, there could be limitations associated with retrospective memory, as respondents may not have accurately recalled information from the previous year, though we believe that one year is short enough for respondents to accurately report the events in their household. We believe that our results could help Liberia and other countries in the developing world with similar socioeconomic conditions, as well as the international community, to be better prepared for future crises and distribute livelihood rehabilitation efforts more effectively, thereby facilitating the affected nation’s speedy recovery after such crises.
|
10.1371/journal.pgen.1006946 | Plant microRNAs in larval food regulate honeybee caste development | The major environmental determinants of honeybee caste development come from larval nutrients: royal jelly stimulates the differentiation of larvae into queens, whereas beebread leads to worker bee fate. However, these determinants are not fully characterized. Here we report that plant RNAs, particularly miRNAs, which are more enriched in beebread than in royal jelly, delay development and decrease body and ovary size in honeybees, thereby preventing larval differentiation into queens and inducing development into worker bees. Mechanistic studies reveal that amTOR, a stimulatory gene in caste differentiation, is the direct target of miR162a. Interestingly, the same effect also exists in non-social Drosophila. When such plant RNAs and miRNAs are fed to Drosophila larvae, they cause extended developmental times and reductions in body weight and length, ovary size and fecundity. This study identifies an uncharacterized function of plant miRNAs that fine-tunes honeybee caste development, offering hints for understanding cross-kingdom interaction and co-evolution.
| How caste has formed in honeybees is an enduring puzzle. The prevailing view is that royal jelly stimulates the differentiation of larvae into queen. Here, we uncover a new mechanism that plant miRNAs in worker bee’s food postpone larval development, thereby inducing sterile worker bees. Thus, the theories about honeybee caste formation need to be re-examined from a new angle besides the traditional focus on royal jelly and its components. Furthermore, since miRNAs are transmitted between species of different kingdoms and can contribute to the phenotype regulation, this new model of horizontal miRNA transfer may open up a new avenue to further study the molecular mechanisms underlying cross-kingdom interaction and co-evolution.
| Caste development in social insects represents a major transition from one level of organization to another in evolution and is believed to be central to the ecological success of social insects [1]. How castes evolved is an enduring puzzle that has long fascinated scientists but currently has no satisfactory answers. Honeybees (Apis mellifera) represent a principal example of caste development. Female honeybees develop into two castes, queens and workers, which differ in morphology, physiology and social function [1, 2]. The queens are reproductive, have a larger body size, develop faster and live longer, whereas workers are characterized by the opposite traits and are mostly sterile helpers that nourish larvae and collect food [3]. This dimorphism is not a consequence of genetic differences but is mainly determined by larval feeding: female larvae receiving a rich diet of royal jelly develop into queens, whereas a less sophisticated diet named “beebread” leads to the worker bee fate [4, 5]. However, it is still not fully understood how different diets modify the developmental trajectory of honeybees to such a thorough extent. While several components of the larval diet, such as specific royal jelly proteins, sugars, p-coumaric acid and fatty acids, have been independently shown to influence caste development in honeybees [6–10], they still cannot account for the full impact of larval food on honeybee development. In this study, we investigated a largely overlooked component of larval food, microRNA (miRNA), and examined its effect on caste development.
miRNAs are a class of 19–24-nucleotide-long non-coding RNAs that act as post-transcriptional regulators of gene expression in eukaryotes [11]. Recently, we reported an unexpected finding that plant miRNAs that are ingested from plant food sources can pass through the gastrointestinal tract, enter into the blood, accumulate in tissues and regulate endogenous gene expression in animals [12]. Other studies have also documented the importance of small RNAs that are transmitted from one species to another and facilitate cross-talk and interspecies communication [13–16]. Moreover, multiple studies have proven that dietary exogenous miRNAs are detectable in consumed animal blood and tissues [17–20]. These studies furnish an additional layer of gene regulation: cross-kingdom regulation mediated by exogenous miRNAs. It is very tempting to speculate that small RNAs in larval food may be an active component that influences honeybee development.
Because beebread is a mixture of pollen and honey, while royal jelly is a glandular secretion of nurse bees [4], the main food sources of worker- and queen-destined larvae are, in theory, plant- and animal-derived, respectively. Thus, we hypothesize that different miRNA contents from larval food of different origins may have distinct impacts on honeybee development. In agreement with this hypothesis, it has been well established in the literature that insects, including honeybees and fruit flies, can ingest small RNAs and that ingested small RNAs can regulate the expression of insect genes, thus reshaping the insects’ phenotypes [21–24]. In this study, we provide evidence for a previously uncharacterized regulatory mechanism of worker bee development, which can be partially attributed to the plant miRNAs enriched in beebread and pollen fed to young larvae.
First, we analysed the small RNA components in royal jelly, honey, beebread and pollen using Illumina deep-sequencing technology. To investigate pollen as a larval food source under natural conditions, we used bee pollen collected and packed by worker bees. The royal jelly, honey, beebread and pollen were collected during the cole (Brassica campestris) flowering stage. Consistent with previous reports [25, 26], the lengths of small RNAs in pollen were concentrated in a range from 19 to 24 nucleotides (S1 Fig). However, the lengths of small RNAs in royal jelly, honey and beebread were distributed over a wider range, from 13 to 28 nucleotides, probably due to degradation products from longer RNAs during their processing within the beehive. Next, total small RNAs were mapped to the reference transcriptome assemblies of honeybee and cole and were further assigned to different classes of small RNAs. In agreement with the hypothesis that royal jelly RNA is mainly animal-derived and beebread RNA is plant-derived, honeybee small RNAs were present at a far higher level in royal jelly than in beebread and pollen, while the abundance of cole small RNAs gradually increased from royal jelly to honey to beebread and pollen (Fig 1A).
A large proportion of the small RNAs were annotated as miRNAs and as the degradation products of tRNAs, rRNAs and mRNAs. By aligning small RNA reads to known miRNAs in the miRBase database 21.0, a total of 46, 39, 14 and 15 annotated bee miRNA types were detected in royal jelly, honey, beebread and pollen, respectively (S1 Table). Most of the bee miRNAs had less than 10 sequence reads in the samples, but they had much higher reads in royal jelly than in honey, beebread and pollen (Fig 1B). On the other hand, there were 41, 71, 58 and 53 annotated plant miRNA types in royal jelly, honey, beebread and pollen, respectively (S1 Table). These plant miRNAs were present at far higher concentrations than animal miRNAs, and their concentration in beebread and pollen was invariably much higher than that in royal jelly and honey (Fig 1B). The differential enrichment of plant miRNAs in beebread and animal miRNAs in royal jelly is clearly shown in S2A Fig. In contrast, the miRNA compositions of beebread and pollen showed high similarities to each other, with a Pearson’s correlation coefficient (R) close to 1 (S2A Fig). The 16 representative plant miRNAs (miR156a, miR157a, miR158a, miR160a, miR162a, miR166a, miR166g, miR167a, miR168a, miR172a, miR172c, miR390a, miR397a, miR403, miR824 and miR845a) with the highest concentrations in beebread and pollen of cole are listed in Fig 1C.
Given the diversity of pollen that is collected by honeybees, plant miRNAs might not be uniformly present in pollen from different sources. Therefore, it is essential to analyze the small RNA components in beebread and pollen collected from different geographical and botanical sources. We performed deep sequencing on royal jelly, honey, beebread and pollen collected during the camellia (Camellia japonica) flowering stage. The results revealed again that the plant miRNAs were more abundant in pollen and beebread than in royal jelly and honey (S2 Table). Likewise, the miRNA profiles were quite similar between beebread and pollen and widely different between beebread and royal jelly (S2B Fig). Interestingly, the plant miRNA profiles of cole and camellia beebread showed similarity to each other, especially for many plant miRNAs that are evolutionarily conserved across the major lineages of plants. For example, 13 of the 16 plant miRNA species enriched in cole beebread were also present in camellia beebread (Fig 1D). Thus, the global components of plant miRNAs in beebread and pollen may not be very diverse between different sources.
However, because deep sequencing is inferior to the more commonly used qRT-PCR for miRNA quantification [27], we performed qRT-PCR assays with a standard curve set using synthetic oligonucleotides of known concentrations to determine the actual concentrations of plant miRNAs in royal jelly, honey, beebread and pollen. All 16 representative plant miRNAs except miR166g (whose qRT-PCR primer was not commercially available) could be readily detected using qRT-PCR in beebread and pollen of cole but were nearly undetectable in royal jelly and honey (generally < 0.1 fmol per μg total RNA) (Fig 1E). It should be noted that we used two normalization strategies for cross-sample comparisons of miRNAs in royal jelly and beebread, and both strategies showed that each plant miRNA was much more abundant in beebread than in royal jelly (S2C Fig). Moreover, northern blotting, which can determine the sizes and concentrations of RNAs, produced the same differences described above for plant miRNA concentrations and showed that miR156a, miR162a and miR168a were detectable in beebread and pollen but not in royal jelly and honey (Fig 1F).
To investigate the effects of plant RNAs, and particularly miRNAs, on honeybee phenotypes, we removed the larvae from the colony setting and reared them on a laboratory diet with or without the addition of plant RNAs or miRNAs. To avoid overfeeding and generating supra-physiological effects, our pilot study first determined the amounts of the 16 representative plant miRNAs that were contained in natural beebread (Fig 1E and S2C Fig). Since the plant miRNA composition enriched in natural beebread is very similar to that in pollen (S2A Fig and S2B Fig), we added total RNA purified from cole pollen to the laboratory diet at the same level as determined based on miRNA levels to reconstitute a close mimic of natural beebread in terms of its miRNA components (“beebread mimic” in S3 Fig). When developing larvae were fed with this beebread mimic, 2-fold of beebread mimic dramatically suppressed the growth of the developing larvae and even caused some larvae to die, whereas 0.5- or 1-fold of beebread mimic reduced larvae growth but had little effect on their survival (S4 Fig).
Next, the effects of plant RNA supplements were characterized based on the developmental time, weight, length and ovary size of adult bees immediately upon emerging from the pupal stage (Fig 2A). Feeding larvae with beebread mimic increased the whole-body accumulation of the 16 representative plant miRNAs (S5A Fig). We did not distinguish the particular tissues where the ingested plant miRNAs were located but instead investigated the effects of plant miRNAs on the whole body as the uptake of exogenous small RNAs from the insect gut has been frequently observed [21, 22, 28]. As a result of the plant RNA supplements, larvae grew relatively slowly during their development and emerged as adults with more of a worker morphology (S6 Fig) characterized by a prolonged developmental time (on average 0.49 days longer, p = 0.0444), reduced weight (on average 14.81% lighter, p = 0.0008) and size (on average 6.55% shorter, p = 0.0005) at adult emergence and a decreased ovary size (on average 21 fewer ovarioles, p = 0.0358) (Fig 2B–2E).
To validate the contribution of plant miRNAs to the observed honeybee phenotypes, we synthesized the 16 plant miRNAs enriched in beebread and pollen, and then the synthetic miRNA pool was added to the larval diet at levels equivalent to those in natural beebread. Compared to the control group, honeybees that were fed a diet containing the miRNA pool showed an increased accumulation of corresponding plant miRNAs within their body (S5B Fig) and developed worker bee-like characteristics, i.e., reduced sizes at adult emergence (10.27% lighter and 4.01% shorter, p = 0.0194 and p = 0.0264, respectively), extended pre-adult developmental time (0.58 days longer, p = 0.0254) and decreased ovary sizes (38 fewer ovarioles, p = 0.0094) (Fig 2F–2I).
Next, we performed bioinformatics analysis to dissect the potential functions of the plant miRNAs in honeybee’s food. Two bioinformatic algorithms (RNAhybrid and miRanda) were used in combination to scan honeybee mRNA sequences for potential binding sites for the 16 representative plant miRNAs. A total of 96 honeybee genes were predicted by both RNAhybrid and miRanda algorithms as the target genes of the 16 plant miRNAs. Most of the 96 genes were predicted to be targeted by only one plant miRNA, whereas a few genes were common targets of 2–3 plant miRNAs. We then used Gene Ontology (GO) analysis to look for biological processes that might be associated with the 96 target genes of the 16 plant miRNAs based on the strategy of a previous study [29]. Significant enrichment of GO functional categories related to “development” was observed (S7 Fig and S3 Table), suggesting again that the plant miRNAs specifically enriched in beebread and pollen may be involved in regulation of the development process of honeybees. Among the 96 target genes, some genes known to influence the developmental fate of honeybees were specially selected and listed in S4 Table.
Subsequently, plant miRNAs targeting Apis mellifera TOR (amTOR) were analysed, as previous studies have demonstrated that amTOR plays a stimulatory role in caste development: the queen fate is associated with elevated amTOR activity, and the inhibition of amTOR causes developmental changes towards worker characteristics in queen-destined larvae [22, 30, 31]. To screen for plant miRNAs that could directly regulate amTOR expression, luciferase reporter assays were conducted. Each plant miRNA binding site in the amTOR gene was fused separately into a position downstream of the firefly luciferase gene in a reporter plasmid. The resulting plasmids were co-transfected into a cell line in combination with above-mentioned plant miRNAs. Among 9 plant miRNAs that could potentially target amTOR, miR162a resulted in a 72% decrease in luciferase activity (Fig 2J), whereas miR156a showed a 14% reduction and other 7 plant miRNAs did not affect luciferase activity (S8 Fig), suggesting that miR162a specifically recognizes amTOR and mediates the post-transcriptional inhibition of this gene. In addition, the amTOR/miR162a hybrid is illustrated in Fig 2K, and its free energy was -26.4 kcal/mol, which was well within the ranges of genuine miRNA-target pairs (-17 kcal/mol is a cutoff value of free energy) [32]. However, when point mutations were introduced into the predicted “seed site” in the amTOR gene, the fused luciferase reporters were no longer affected by miR162a (Fig 2J).
Subsequently, to determine the potential effects of miR162a alone on amTOR expression and the corresponding phenotypes, honeybee larvae were reared with a diet to which either synthetic miR162a or scrambled RNA was added. Notable increases in the amount of ingested miR162a (S5C Fig) and decreases in the level of amTOR mRNA (Fig 2L) were detected in honeybees that were reared with a diet containing miR162a. Similarly, amTOR mRNA was downregulated in honeybees reared on a diet containing either total pollen RNA or the synthetic miRNA pool (S9A and S9B Fig). In contrast to the scrambled RNA, which had no effect on any of the tested morphological characteristics, miR162a supplied in the larval food significantly reduced the body weights (7.87% lighter, p = 0.0292) and lengths (4.49% shorter, p = 0.0103) and ovary sizes (29 fewer ovarioles, p = 0.0301) of newly emerged adults but did not significantly increase the developmental time (0.17 days longer, p = 0.1755) of the adult bees (Fig 2M–2P).
To further investigate the evolutionary dynamics of the molecular mechanisms underlying social development between solitary and eusocial species, we tested plant RNA and miRNAs on a non-social model insect, Drosophila melanogaster. Although there is no caste differentiation in Drosophila, there is evidence that molecular pathways involved in establishing caste dimorphism are also conserved in the individual development of Drosophila [6]. Thus, we investigated the mechanism underlying honeybee caste differentiation in Drosophila (Fig 3A).
First, we ruled out the possibility that the residual chemicals from RNA isolation might block larval development as the mock group of Drosophila larvae fed the same chemical residues developed normally (S10 Fig). In accordance with the observation that the beebread mimic postpones queen differentiation in honeybees, Drosophila larvae reared with medium containing total pollen RNA had longer developmental times (p<0.0001), were smaller (7.83% lighter and 2.99% shorter in females, p = 0.0038 and p = 0.0014, respectively; 7.32% lighter and 4.33% shorter in males, p = 0.0144 and p<0.0001, respectively), had fewer ovarioles (1.6 fewer ovarioles, p = 0.0008) and showed reduced fecundity (a total of 28.04% eggs fewer) compared to those reared with the control medium (Fig 3B–3F). Similarly, plant miRNAs also accumulated in Drosophila larvae (S5D Fig). Subsequently, to narrow down the active components in plant RNA, small RNAs were enriched from total pollen RNA, and the effects of small RNAs on Drosophila phenotypes were examined in the same manner as described above. Small plant RNAs also delayed Drosophila development (p<0.0001) and reduced the final adult size (6.84% lighter and 3.45% shorter in females, p = 0.0002 and p = 0.0005, respectively; 7.50% lighter and 3.98% shorter in males, p = 0.0030 and p = 0.0010, respectively), ovary size (0.9 fewer ovarioles, p = 0.0288) and fecundity (25.13% eggs fewer) of Drosophila as effectively as total RNA (S11 Fig).
Similarly, when the miRNA pool was fed to developing Drosophila larvae, we observed an increase in plant miRNA levels (S5E Fig) and corresponding decreases in final adult size (5.88% lighter and 7.73% shorter in females, p = 0.0208 and p<0.0001, respectively; 6.82% lighter and 3.39% shorter in males, p = 0.0111 and p = 0.0013, respectively), ovary size (1.6 fewer ovarioles, p = 0.0004) and fecundity (19.96% eggs fewer) (Fig 3H–3K). However, developmental times did not change in Drosophila reared with medium containing the miRNA pool (p = 0.768) (Fig 3G). To determine the specificity of the inhibitory effects of plant miRNAs on Drosophila development and to exclude the possibility that the phenotypic changes were caused by components other than plant miRNAs, an miRNA antisense pool against the above-mentioned 16 miRNAs was synthesized and added to the Drosophila larval medium together with small pollen RNAs to abolish the function of these plant miRNAs. The inhibitory effects of plant RNAs on the adult size, ovary size and fecundity of Drosophila were completely reversed by the addition of the antisense pool to the larval diet (Fig 3L–3O).
Next, a similar miR162a binding site in the Drosophila melanogaster TOR (dmTOR) gene was identified (Fig 4A). When this binding site was fused into the luciferase reporter plasmid, miR162a also reduced luciferase activity (Fig 4B). However, when a point mutation was introduced into the miR162a binding site in the dmTOR gene, the mutated luciferase reporter was unaffected by miR162a (Fig 4B). The correlation between miR162a and dmTOR was further examined by evaluating dmTOR protein expression in Drosophila Schneider 2 cells (S2 cells) after the induction of miR162a. The expression of the dmTOR protein was significantly inhibited by miR162a in S2 cells (Fig 4C). We further performed a biotin-avidin pull-down assay to assess the direct binding of miR162a to dmTOR mRNA. miR162a was only enriched in the pull-down product precipitated by the anti-dmTOR probe and was undetectable in the products that were precipitated by a random probe or no probe (Fig 4D), suggesting that miR162a directly binds to dmTOR mRNA in S2 cells.
Moreover, Drosophila larvae reared with synthetic miR162a supplied in the medium showed increased whole-body accumulation of miR162a (S5F Fig) and reduced whole-body expression of dmTOR mRNA (Fig 4E). A similar reduction in dmTOR mRNA levels was observed in Drosophila reared with medium containing total pollen RNA or the synthetic miRNA pool (S9C and S9D Fig). Consequently, Drosophila reared with miR162a in the medium exhibited a decrease in body weight (8.82% and 8.75% lighter in females and males, p = 0.0013 and p = 0.0003, respectively), length (4.45% and 6.04% shorter in females and males, p<0.0001 and p<0.0001, respectively), ovary size (0.9 fewer ovarioles, p = 0.0050) and fecundity (21.79% eggs fewer) but had equal developmental times compared to the control larvae (p = 0.323) (Fig 4F–4J). In contrast, Drosophila larvae reared with the scrambled RNA in the medium showed no such phenotypes.
Finally, the correlation between miR162a and dmTOR was analysed using transgenic Drosophila expressing a GFP reporter transgene with an miR162a binding fragment of dmTOR inserted downstream. In association with the observed phenotype of reduced fecundity in Drosophila reared with miR162a in the medium (Fig 4J), decreased GFP levels in both nurse cells and follicle cells in the egg chambers were observed when miR162a was added into the larval diets of the transgenic line, while the addition of scrambled or seed-mutant miR162a mimics had no effect on GFP levels (Fig 4K). These results suggest that miR162a in larval food was sufficiently delivered to Drosophila ovaries and that it suppresses endogenous dmTOR expression.
Caste differentiation of honeybees is a complex developmental process influenced by genetic, epigenetic and environmental variations. The prevailing view is that the nutrients in royal jelly (primarily proteins, sugars and fatty acids) drive queen development [6, 10, 33, 34]. However, the active components that determine the developmental fate of honeybees remain elusive and even controversial [35]. Recent studies provide new insights into the relationship between epigenetic regulation and caste differentiation in insects [36–38]. In this study, we identified that plant miRNAs are significantly enriched in beebread and pollen and not in royal jelly. This striking difference prompted us to hypothesise that miRNAs, acting as important epigenetic regulators, may be transferred from the food of worker-destined larvae to their bodies and negatively regulate larval development; in contrast, miRNAs in the royal jelly are not sufficient to reach a functional level and to have biological relevance, therefore queen-destined larvae consuming royal jelly evade miRNA regulation. To test this hypothesis, we elucidated the effects of the plant RNAs and miRNAs that were enriched in beebread and pollen on honeybee phenotypes, and uncovered a previously unrecognized role for RNA as an environmental determinant of honeybee caste development. Furthermore, we investigated phenotypic changes in Drosophila caused by food supplemented with plant RNAs and miRNAs and observed larvae developing into adults with phenotypes similar to those of worker bees. We verified that these effects on the development of honeybees and Drosophila were caused by plant RNAs and specific miRNAs and excluded the possibility of a general effect of RNAs, because a synthetic scrambled RNAs added to the larval diet did not cause any phenotypic changes in honeybees or Drosophila. We also ruled out the possibility that the potentially toxic effects of chemical residues from the RNA isolation process caused the observed phenotypic changes in Drosophila, because a mock diet (H2O instead of pollen was processed for RNA isolation and added to the diet) with similar chemical residues had no effects on larval development. As a next step, we investigated whether honeybee development was regulated by variations in specific genes that are targeted by specific plant miRNAs. Mechanistic studies revealed that the blocking of the queen fate was, at least in part, due to amTOR knockdown by miR162a. Overall, our study revealed that the development of worker bee caste may be, at least in part, attributed to a previously uncharacterized effect executed by the transfer of enriched plant miRNAs in beebread and pollen to the young larvae.
The mobility of small RNA molecules (e.g., siRNA and miRNA) from one species to another is a newly discovered mechanism for the spread of gene-silencing signals and for facilitating cross-talk between different organisms, even between species of different kingdoms [39]. The cross-species transfer of small RNAs has been frequently reported to occur between interacting organisms: from bacteria to nematodes [40], from fungal pathogens to plants [14], from plants to pathogenic and symbiotic microbes [41–44], from plants to nematodes [45], and from plants to insects [46]. For example, transgenic plants engineered to produce siRNAs against essential pest genes are more resistant to pest attack [46]. In this study, we sought to broaden the understanding of the existence of small RNA transfer between representative species in the natural world: honeybees and plants. Our evidence indicates that ingested plant miRNAs affect gene expression and can reshape honeybee phenotypes, and it may provide additional support for the concept of horizontal small RNA transfer. We focused on the phenomenon of plant miRNA uptake and function but did not uncover a clear molecular mechanism accounting for the entrance and transfer of miRNAs within honeybees. We propose that systemic RNAi, which allows small RNAs to be transported across cellular boundaries and to spread throughout the whole body of insects [21, 22, 28], might be a possible transport mechanism. However, this mechanism, which is mediated through SID-1 transmembrane protein activity [47, 48], has only been intensively characterized in C. elegans. Whether SID-1 homologues are present in honeybees and play equivalent roles in small RNA transport requires further investigation. Another open question is how honeybees make use of the available dosage of plant miRNAs to control their development. In our experiments, the same amount of plant miRNAs as is found in natural beebread was used, and this dose produced similar effects to those seem in nature. It is largely unknown if honeybees possess an amplification pathway as is found in C. elegans [49] to allow a small amount of RNA taken up from the environment to generate abundant secondary RNAs and to trigger strong responses within the body. In addition, plant miRNAs tend to induce mRNA cleavage through perfect or near-perfect complementarity with their target sequences, while animal miRNAs generally cause translational repression through partial complementarity [11, 50–52]. The observation that miR162a decreased amTOR mRNA levels in vivo indicates that it behaves, at least in some ways, similarly to a plant miRNA. However, miR162a shows non-perfect complementarity with its target sequence, even with a G:U wobble in the seed region, indicating a regulatory action of animal miRNA. It is also unclear how plant miRNAs are incorporated into the honeybee’s Argonaute complexes. Because the ingested plant miRNAs should be mature single-stranded RNAs, it is not clear how these single-stranded small RNAs are loaded into Argonaute proteins to produce a functional miRNA form. Nevertheless, because miRNAs and other small RNAs have been frequently detected to be transported between species and hijack the RNAi machinery of host cells to exert biological functions [14, 40–44], it would be interesting to analyse the mode of action of plant miRNAs in honeybee cells. However, these questions are beyond the scope of this study.
Protocols have been developed for rearing honeybee since 1927 [53–55]. The diet of a mixture of fresh royal jelly, fructose, glucose, yeast extract and H2O has been proven to be the optimal food for honeybee larvae [56–58]. In this study, it should be noted that one direct test could be to feed honeybees with beebread in which plant miRNAs have been eliminated. In fact, we have attempted to rear honeybees with pollen or beebread supplemented with antisense miRNAs without royal jelly. Unfortunately, all of the larvae died during cultivation. This result is consistent with previous observations that royal jelly is indispensable for the rearing of honeybee larvae in vitro [54, 55, 59]. Alternatively, we added plant RNAs or miRNAs to the larval diet of honeybees, which can defer the queen bee fate even in the presence of royal jelly and therefore supports our arguments. In fact, queen development is not the default trajectory in honeybees and royal jelly is needed to act on the endocrine system to direct larvae differentiation into a queen fate. The pathways controlling body size, developmental duration and fertility are anyway downregulated in worker-destined larvae [22, 30, 31]. According to our study, we suggest that the negative effects of beebread and pollen on larval development may be a part of the causation. Additionally, lab-reared honeybees largely develop with intermediate characteristics between a worker and queen, i.e., with more ovarioles than natural worker bees [60]. This phenomenon implies that an essential ingredient may be missing from the larval diet used for in vitro cultivation that impairs the differentiation of worker bees. We suggest that the plant RNA enriched in natural beebread is a very likely candidate, although we cannot rule out other possibilities.
Caste development is a complex process that involves multiple regulatory factors. Although this study largely focused on how plant miRNAs negatively affect the development of honeybees, we do not claim that plant miRNAs are the sole factor regulating honeybee development, and thus, removing plant miRNAs alone is not sufficient to disrupt the development of all phenotypes related to caste differentiation. Likewise, we do not expect that plant miRNAs can completely reverse the developmental fate, i.e., turn worker into queen or queen into worker. It is worth noting that the inhibitory effects of plant miRNAs on honeybee development were gradually reduced from treatments with total pollen RNA to the miRNA pool and to only miR162a. For example, total pollen RNA prolonged the developmental time in honeybees and Drosophila, while miR162a did not. This phenomenon indicates that miR162a is not the sole active component, and other miRNAs, even larger RNAs, may also contribute to developmental regulation. Indeed, miR162a is only one of the multiple plant miRNAs enriched in beebread, and these miRNAs are only a portion of all classes of small RNAs, which themselves account for only a small fraction of total RNAs. Therefore, we propose that a single miRNA (i.e., miR162a) does not operate as an all-around regulator of caste development; instead, more plant RNA components likely function in a cooperative manner in the regulatory network leading to caste development. Similarly, the miR162a-amTOR pair is only one of the pathways that participate in this cross-kingdom regulation. The involvement of other regulatory pathways (e.g., those indicated in the bioinformatic analysis summarized in S7 Fig and S3 and S4 Tables) in honeybee development requires further investigation. In summary, the development of queens and workers is not determined by a single compound but, instead, is driven by the cooperation of multiple components in the larval food, which may include proteins, sugars, fatty acids and plant RNAs. However, why honeybees use such a sophisticated and intricate mechanism to regulate the queen-worker dimorphism is a fascinating question. For larvae that are destined to become queens, royal jelly is fed in copious amounts to drive the development of royal phenotypes. For worker-destined larvae, substantial quantities of plant miRNAs are absorbed when consuming beebread and pollen, thereby negatively influencing the larval development and inducing sterile worker bees. Reliance upon beebread and pollen as the exclusive food for sterile workers may have evolved in concert with the exploitation of plant miRNAs for caste regulation via a form of “RNAi castration”. The positive effects of royal jelly and the negative effects of beebread may maintain the stability of the colony’s social order and contribute to the survival of the colony in a coordinated manner. However, an opening question is raised regarding whether the plant miRNAs that reduce the development and fertility in honeybees and Drosophila have similar influences on solitary bees and bumblebees that would be exposed to the same plant miRNAs. Another opening question is about the widespread apicultural use of artificial pollen substitutes (commonly consist of protein sources derived from soy, wheat or lentils) in agricultural systems. Although the supplemental protein diets offset the poor nutritional conditions in honeybee colonies, long-term consumption of protein as the sole nutrition may compromise the ability of plant miRNAs to fine-tune honeybee development. Indeed, previous studies had explore the influence of natural pollen and artificial pollen substitutes on the cellular immunity, survival and parasite infection in honeybees and shown that the change from a natural to an artificial high nutritious diet in terms of protein content is not sufficient to promote healthy bees [61, 62]. If consumption of natural or artificial diets did produce varying levels of plant miRNAs in honeybees and impact the survival and breeding of honeybees deserves further investigation. Overall, our study uncovered a new layer of caste regulation in which plant RNAs are transmitted between species of different kingdoms, offering hints for understanding cross-kingdom interactions and co-evolution.
The pollen used for this study was bee pollen, which are pollen pellets compressed and packed into corbicula on the outer surfaces of the hind legs after collection by forager bees. The pollen was separated using a specific collection device when bees come back to the comb. The royal jelly, honey, beebread and pollen were obtained in the cole or camellia flowering stage. All of the samples were stored at -80°C immediately after collection. Total RNA was extracted from royal jelly, honey, beebread and pollen using TRIzol Reagent (Invitrogen, Carlsbad, CA, USA). Small RNAs were extracted from royal jelly, honey, beebread and pollen using the MirVana Protein and RNA Isolation System (Ambion, Austin, TX, USA). Synthetic plant miRNA mimics and inhibitors and scrambled negative control RNAs were purchased from Invitrogen.
The diets (V.S. diet, D-1 diet and D-2 diet) for laboratory rearing of honeybee larvae have been described previously [22]. The V.S. diet for the first 3 days was as follows: 50% fresh royal jelly, 6% fructose, 6% glucose, 1% yeast extract and 37% dd-H2O. The D-1 diet for the next 2 days was as follows: 53% fresh royal jelly, 6% fructose, 6% glucose, 1% yeast extract and 34% dd-H2O. The D-2 diet for the following days and until pupation was as follows: 53% royal jelly, 7.8% fructose, 7.8% glucose, 1% yeast extract and 30.4% dd-H2O. A healthy colony was chosen for egg laying, and the queen was caged in an empty comb from 6:00–18:00. After 72 h, the hatched larvae were moved to 48-well plates, and total pollen RNA, small pollen RNAs, synthetic miRNA pool and synthetic miR162a were added to the diets. The detailed experimental procedure for preparation of the diets with added plant RNA (total pollen RNA, synthetic miRNA pool or synthetic miR162a) is shown in S12A Fig. DEPC-H2O was added to the diet as a control. The larvae were transferred to new plates with fresh diets every 12 h. The plates were kept in a crisper with 15.5% glycerine (90% relative humidity), and the crisper was placed in an incubator (33°C) during the larval period. Defecating larvae were transferred into new 24-well plates, and each well contained a piece of filter paper. The plates containing defecating larvae were kept in a crisper with a saturated sodium chloride solution (70% relative humidity), and the crisper was left in an incubator (33°C). Then, the newly emerged adults were collected, and their characteristics were measured. At the beginning, we moved 48 larvae into the plates for each group and generally got 25–30 emerged adults due to the mortality during in vitro rearing. The honeybee larvae cultivated in this laboratory conditions largely developed to intermediates with characteristics between a worker and queen. For example, they generally had ovarioles (30–80 ovarioles) more than natural worker bees (< 10 ovarioles) but less than queens (> 150 ovarioles).
A total of 20–30 pairs of Drosophila were caged in a tube containing ~15 mL of medium from 10:00–16:00 for egg laying (10–15 tubes for each experimental group). Total pollen RNA, small pollen RNAs, the synthetic miRNA pool and synthetic miR162a were added to the medium. The detailed experimental procedure for preparation of the medium with added plant RNA (total pollen RNA, small pollen RNAs, synthetic miRNA pool or synthetic miR162a) is shown in S12B Fig. DEPC-H2O was added to the medium as a control. Approximately 8–9 days later, newly enclosed adults were collected, and their characteristics were measured. We generally got 25–35 enclosed adults at this stage. On day 5 after eclosion, 5 pairs of Drosophila were placed in a custom tube for fertility analyses (10–15 tubes for each experimental group). The eggs that were laid by the 5 pairs of Drosophila were counted every day for 5 days. The culture environments for each parallel test were carefully controlled, and we only compared results obtained in the same parallel test, which excludes confounding environmental factors that may otherwise affect experimental results.
The sequencing procedure was conducted as previously described [15]. Briefly, fresh samples of royal jelly, honey, beebread and pollen were collected from colonies of Italian honeybees. Total RNA was extracted from 10 g of these samples using Trizol Reagent (Invitrogen, Carlsbad, CA, USA) according to the manufacturer’s instructions. Then, equal amounts of total RNA were analysed using Illumina deep-sequencing technology, and the sequencing procedure was performed by BGI (Shenzhen, China). After masking the adaptor sequences from the raw data and removing short and low-quality reads, a total of 9,548,986, 13,683,503, 9,559,836 and 9,561,153 reads from royal jelly, honey, beebread and pollen of cole and 8,996,733, 12,160,200, 15,237,283 and 16,690,115 reads from royal jelly, honey, beebread and pollen of camellia were obtained, respectively. The clean reads were aligned to the transcript sequences using bowtie 1.1.2 (http://bowtie-bio.sourceforge.net) with perfect match. Transcript sequences of Apis mellifera (assembly Amel_4.5) and Brassica napus (assembly Brassica_napus_assembly_1.0) were downloaded from the NCBI genome database (https://www.ncbi.nlm.nih.gov/genome). Clean reads were also compared to the known miRNA precursors in the miRBase database 21.0 based on the Smith-Waterman algorithm. Only candidates with no mismatches and no more than 2 shifts were counted as miRNA matches. For normalization, the total sequencing frequency of each sample was normalized to 10,000,000. Data for Illumina deep-sequencing have been deposited at GEO with the accession code GSE76286 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?token=yrevmigijrynjsl&acc=GSE76286).
To determine the plant miRNA levels in honeybee larval food, total RNA was extracted from royal jelly, honey, beebread and pollen using Trizol Reagent (Invitrogen) according to the manufacturer’s instructions. To determine the amTOR, dmTOR and miR162a levels in honeybees or Drosophila, newly emergence adults were collected, and total RNA was extracted using Trizol Reagent (Invitrogen).
Assays to quantify mature miRNAs were performed using TaqMan miRNA probes (Applied Biosystems, Foster City, CA) according to the manufacturer’s instructions. Briefly, 1 μg of total RNA was reverse-transcribed to cDNA using AMV reverse transcriptase (TaKaRa, Dalian, China) and a stem-loop RT primer (Applied Biosystems). The following reaction conditions were used: 16°C for 30 min, 42°C for 30 min, and 85°C for 5 min. Real-time PCR was performed using a TaqMan PCR kit on an Applied Biosystems 7500 Sequence Detection System (Applied Biosystems). The reactions were incubated in a 96-well optical plate at 95°C for 5 min, followed by 40 cycles of 95°C for 15 sec and 60°C for 1 min. All of the reactions were run in triplicate. After the reactions, cycle threshold (CT) values were determined using fixed threshold settings, and the mean CT of triplicate PCRs was determined. To calculate the absolute expression levels of the target miRNAs, a series of synthetic miRNA oligonucleotides at known concentrations were reverse transcribed and amplified. The absolute amount of each miRNA was then calculated in reference to the standard curve. For cross-sample comparisons of miRNAs in royal jelly, honey, beebread and pollen, miRNA levels were normalized to the total amounts of RNA or to the total mass of the samples.
To quantify amTOR and dmTOR mRNA, 1 μg of total RNA was reverse-transcribed to cDNA using a specific reverse primer and AMV reverse transcriptase (TaKaRa) under the following conditions: 16°C for 15 min, 42°C for 60 min, and 85°C for 5 min. Subsequently, real-time PCR was performed using the RT product, SYBR Premix Ex Taq (Takara, Dalian, China) and specific primers for amTOR and dmTOR. The primers that were used in this study were as follows: amTOR-forward, 5’-TTGGTTGGGTACCGCATTGT-3’; amTOR-reverse, 5’-AACCTGGGGCCATTCTTAGC-3’; dmTOR-forward, 5’-CTCTTACATGAATCCGATCCTCA-3’; and dmTOR-reverse, 5’-CGGAGCCTCCATTAACCT-3’. The reactions were incubated at 95°C for 5 min, followed by 40 cycles at 95°C for 15 sec, 55°C for 30 sec, and 72°C for 30 sec. After the reactions were complete, CT values were determined using fixed threshold settings. The relative amounts of amTOR and dmTOR were normalized to amActin and dmActin, respectively. The primers for amActin and dmActin were as follows: amActin-forward, 5’-TGCCAACACTGTCCTTTCTG-3’; amActin-reverse, 5’-AGAATTGACCCACCAATCCA-3’; dmActin-forward, 5’-CGCGATTTGACCGACTACCT-3’; and dmActin-reverse 5’-TTGATGTCACGGACGATTTCA-3’.
Small RNAs were extracted from royal jelly, honey, beebread and pollen using the MirVana Protein and RNA Isolation System (Ambion, Austin, TX, USA). The northern blot analysis was carried out using miRCURY LNA microRNA Detection Probes with DIG-labelling (Exiqon, Woburn, MA, USA) and a DIG luminescence detection kit (Roche, Indianapolis, IN, USA) according to the manufacturer’s instructions. Briefly, samples of small RNAs (15 μg) and synthesized size markers (Invitrogen) were added to Gel Loading Buffer II (Ambion) and denatured at 95°C for 5 min. A 15% TBE-urea gel was pre-run at 250 V for 60 min, and the samples and size markers were added to the gel and run at 250 V until the bromophenol blue (BPB) from the loading solution reached approximately 1 cm above the bottom of the gel. Generally, BPB and cyanol from the loading solution run at approximately 15 bases and 60 bases, respectively. RNA was then transferred onto a nylon membrane (Hybond N+, Amersham Biosciences) via electroblotting at 250 mA in 0.5× TBE (Tris-borate-EDTA) buffer for 1 h. After UV-crosslinking at 1200 mJ, a prehybridization step was performed by incubating the membrane with 40 mL of ULTRAhyb-Oligo solution (Ambion) pre-heated to 50°C. Prehybridization was performed for 30 min at 50°C in a standard rotating hybridization oven. DIG-labelled LNA probes were hybridized to the membranes overnight at 50°C with slow rotation. The next day, the membrane was washed twice for 15 min each in NorthernMax Low-Stringency wash solution no. 1 (Ambion) at 50°C, briefly rinsed for 10 min with Washing Buffer from the DIG wash and Block Buffer Set (Roche), blocked for 30 min in 1× Blocking Solution (Roche), incubated for 30 min in antibody solution (anti-DIG-AP 1:10,000 in 1× Blocking solution, Roche), washed twice for 15 min each with Washing Buffer and incubated for 2–5 min with 1× Detection Buffer (Roche). Then, the membrane was incubated with CSPD, the chemiluminescent substrate for alkaline phosphatase (Roche) and exposed to Amersham Hyperfilm ECL (GE Healthcare Life Sciences, Piscataway, NJ) following the instructions of the DIG Luminescent Detection Kit (Roche).
Sequence information of honeybee mRNAs was collected from the NCBI database. Two bioinformatic algorithms, RNAhybrid and miRanda [63, 64], were used in combination to scan honeybee mRNAs for potential binding sites for plant miRNAs. The gene lists generated by miRNA target prediction were assigned to orthology groups with Drosophila melanogaster genes on the basis of BLAST match, and GO terms were assigned to bee genes based on annotation of Drosophila genes. GO functional terms and Drosophila gene GO annotations were downloaded from the GO database. Counts of genes in specific categories were performed by using PANTHER, a gene functional classification tool. χ2 tests were performed in R, and differences were considered statistically significant at p < 0.05. Cytoscape was used to build the GO network associations.
We utilized the processing machinery of pri-dme-mir-184 to express miR162 in Drosophila S2 cells. The S2 cell line was cultured at 28°C with Schneider’s Drosophila medium containing 10% heat-inactivated FBS. The miR162a sequence was substituted into a 300-bp pri-dme-mir-184 backbone with structurally conserved nucleotide changes to maintain pairing. The 300-bp pri-dme-mir-184 was GTTTTCTATTCACGCTTTAGTGCACTTATTTACTCGATTGTATGATCCAAAGCTCCTCTTTGACTCGCCGAATTCCTGTCGATTCAATGGGTATTGGTTTGGTTGGCCGGTGCATTCGTACCCTTATCATTCTCTCGCCCCGTGTGCACTTAAAGACAACTGGACGGAGAACTGATAAGGGCTCGTATCACCAATTCATCCTCGGGTCAGCCCAGTTAATCCACTGATTTGCACACTTTTCTTTATACATACGAGGATACTTACCCCACGTTTCGATTACGCGCATCAATCAATCAATCA, and the underlined parts were replaced with TCGATAAACCTCTGCATCCAG and AATGAATGAGAGGCTTTATCGA, respectively. The 300-bp fragment containing the miR162a sequence was synthesized directly and cloned into a pAc5.1 vector. Cultured cells were prepared for transfection by seeding 1×106 cells/mL in a 24-well plate. After culturing the cells for 12–18 h, transfection was performed with Effectene transfection reagent (Qiagen, Valencia, CA, USA). The transfection mixture per well contained 6 μL of Effectene reagent only, 6 μL of Effectene reagent and 0.3 μg of miR162a expressing plasmids, or 6 μL of Effectene reagent and 0.3 μg of pAc5.1 vectors without any insert. The cells were collected 48 h after transfection and used for western blotting analysis.
Plasmids expressing miR162a were transfected into S2 cells using Effectene (Qiagen) according to the manufacturer’s instructions. The cells were lysed in RIPA buffer (0.5% NP-40, 0.1% sodium deoxycholate, 150 mM NaCl, 50 mM Tris-HCl (pH 7.5)). The lysates were resolved via 6% SDS-PAGE (for the dmTOR protein) or 10% SDS-PAGE (for internal control GAPDH protein), transferred to a PVDF membrane (Millipore, Bedford, MA, USA) and probed with anti-dmTOR or anti-GAPDH antibodies (Santa Cruz Biotechnology, CA, USA). Anti-dmTOR antibodies were polyclonal antibodies that were custom-made by GenScript USA Inc. (Nanjing, China). The epitope was predicted using the GenScript OptimumAntigen design tool, and the peptide antigen was then synthesized. After the coupling reaction and mixing with complete adjuvant, the coupled antigen was used once for a subcutaneous injection. The host strain was a New Zealand rabbit. Then, the coupled antigen was mixed with incomplete adjuvant and injected into the rabbit. Subsequently, serum was taken from the immunized rabbit, and the antibody was purified.
A DNA probe complementary to dmTOR was synthesized with 5’ and 3’ terminal biotin labels. The probe was dissolved in a wash/binding buffer (0.5 M NaCl; 20 mM Tris-HCl, pH 7.5; 1 mM EDTA) to a concentration of 8 pmol/μL. Then, the probe was incubated with streptavidin magnetic beads (New England Biolabs) at room temperature for 1 h with occasional agitation. After incubation, the probe-coated beads were washed twice and captured with a magnet to remove the supernatant. The total RNA that was extracted from miR162a-transfected S2 cells (50~100 μg) was pretreated with DNaseI and then heated at 65°C for 5 min, followed immediately by an ice bath. Then, the RNA was incubated with the prepared probe-coated beads at 37°C for 3 h with occasional agitation, and the beads were washed twice with wash/binding buffer and once with a cold low-salt buffer (0.15 M NaCl; 20 mM Tris-HCl, pH 7.5; 1 mM EDTA). After each wash, a magnet was applied to the tube, and the supernatant was removed. Finally, the RNA was eluted from the probe-coated streptavidin beads with Elution Buffer (10 mM Tris-HCl, pH 7.5; 1 mM EDTA) prewarmed to 90°C and then analysed via qRT-PCR. The following probe sequences were used: anti-dmTOR pull-down probe 5’-CTAGAGCCCAAGTCTGCATTGAA-3’ and random pull-down probe 5’-GGCAGCTAACCTATATGACATGC-3’.
Drosophila were cultured following standard procedures at 25°C except for the transgenic lines, which were cultured at 29°C. Strain w1118 was obtained from the Bloomington Drosophila Stock Center. To generate the transgenic line, the miR162a binding sequence in the dmTOR gene was cloned into a pUbi-GFP expression vector, and the pUbi-GFP-dmTOR transgenic line was obtained via embryo injection according to standard procedures. After miR162a or mutant miR162a was added to the larval diets of the transgenic Drosophila, the ovaries of transgenic Drosophila were dissected in PBS and then fixed in a devitellinizing buffer (100 μl, 7% formaldehyde) and heptane (600 μl) mixture for 10 minutes. After 3 washes in PBS for 10 min each, ovaries were incubated in blocking solution (PBT, 10% goat serum) for 30 min. GFP levels were observed and compared between different groups.
The analyses were performed using IBM SPSS Statistics 19. One-way ANOVAs and two-tailed Student’s t-tests were used for the analyses. The data are presented as the means ± SEM of at least three independent experiments, and differences were considered statistically significant at p < 0.05.
|
10.1371/journal.pgen.1003285 | Drosophila Yemanuclein and HIRA Cooperate for De Novo Assembly of H3.3-Containing Nucleosomes in the Male Pronucleus | The differentiation of post-meiotic spermatids in animals is characterized by a unique reorganization of their nuclear architecture and chromatin composition. In many species, the formation of sperm nuclei involves the massive replacement of nucleosomes with protamines, followed by a phase of extreme nuclear compaction. At fertilization, the reconstitution of a nucleosome-based paternal chromatin after the removal of protamines requires the deposition of maternally provided histones before the first round of DNA replication. This process exclusively uses the histone H3 variant H3.3 and constitutes a unique case of genome-wide replication-independent (RI) de novo chromatin assembly. We had previously shown that the histone H3.3 chaperone HIRA plays a central role for paternal chromatin assembly in Drosophila. Although several conserved HIRA-interacting proteins have been identified from yeast to human, their conservation in Drosophila, as well as their actual implication in this highly peculiar RI nucleosome assembly process, is an open question. Here, we show that Yemanuclein (YEM), the Drosophila member of the Hpc2/Ubinuclein family, is essential for histone deposition in the male pronucleus. yem loss of function alleles affect male pronucleus formation in a way remarkably similar to Hira mutants and abolish RI paternal chromatin assembly. In addition, we demonstrate that HIRA and YEM proteins interact and are mutually dependent for their targeting to the decondensing male pronucleus. Finally, we show that the alternative ATRX/XNP-dependent H3.3 deposition pathway is not involved in paternal chromatin assembly, thus underlining the specific implication of the HIRA/YEM complex for this essential step of zygote formation.
| Chromosome organization relies on a basic functional unit called the nucleosome, in which DNA is wrapped around a core of histone proteins. However, during male gamete formation, the majority of histones are replaced by sperm-specific proteins that are adapted to sexual reproduction but incompatible with the formation of the first zygotic nucleus. These proteins must therefore be replaced by histones upon fertilization, in a replication-independent chromatin assembly process that requires the histone deposition factor HIRA. In this study, we identified the protein Yemanuclein (YEM) as a new partner of HIRA at fertilization. We show that, in eggs laid by yem mutant females, the male pronucleus fails to assemble its nucleosomes, resulting in the loss of paternal chromosomes at the first zygotic division. In addition, we found that YEM and HIRA are mutually dependent to perform chromatin assembly at fertilization, demonstrating that they tightly cooperate in vivo. Finally, we demonstrate that the replication-independent chromatin assembly factor ATRX/XNP is not involved in the assembly of paternal nucleosomes. In conclusion, our results shed new light into critical mechanisms controlling paternal chromosome formation at fertilization.
| Assembly of octameric nucleosomes in eukaryotic chromatin is a stepwise process where deposition of a histone H3-H4 heterotetramer precedes incorporation of two H2A-H2B dimers [1]. While the bulk of de novo chromatin assembly occurs during genome replication and mainly involves canonical histone H3, alternative, replication-independent (RI) chromatin assembly pathways use the conserved histone H3 variant H3.3 [2], [3]. Canonical (or replicative) H3s (H3.1 and H3.2 in mammals, H3.2 in Drosophila) are synthesized in early S phase and deposited at DNA replication forks by the trimeric CAF-1 (Chromatin Assembly Factor-1) complex [4]. In contrast, H3.3 is expressed throughout the cell cycle and is deposited at various genomic regions in a DNA-synthesis independent manner [5]–[8]. During the past decade, research on H3.3 has largely focused on the ability of this histone to be deposited at transcribed genes, opening the possibility that H3.3 could constitute an epigenetic mark of active chromatin [9]–[13]. Recent advances in the field have let emerge a more complex view of H3.3 biology. Although H3.3 is indeed enriched at transcribed gene bodies, it is now established that this histone is also deposited at various chromatin regions, such as regulatory elements, mammalian telomere repeats or satellite DNA blocks [5]–[7], [14]–[18]. This surprising versatility of H3.3 could simply reflect its ability to be deposited in regions that are subjected to nucleosome depletion or rapid histone turnover [5], [7], [19].
In metazoa, H3.3 is also implicated in a variety of nuclear processes that specifically occur in germ cells and in early embryos [7], [20]–[22]. In mouse spermatocytes, for instance, H3.3-containing nucleosomes are assembled on sex chromosomes during their inactivation and accumulate over the whole sex body [23]. Moreover, an insertion mutation in the mouse H3.3A gene induces male subfertility, among other phenotypes [24]. Certain lysine residues of H3.3 are also important for the establishment of heterochromatin during reprogramming in mouse zygotes [25]. Recently, knock-down experiments in Xenopus laevis demonstrated a specific and critical requirement of H3.3 during embryo gastrulation [26]. In Drosophila, H3.3 deficient animals are viable but are both male and female sterile [27], [28]. H3.3 is notably required for the proper segregation of meiotic chromosomes in spermatocytes [28] and for the global organization of early spermatid chromatin [28], [29].
A remarkable H3.3 deposition process also occurs during the decondensation of the male pronucleus at fertilization [21]. This unique, genome-wide assembly of H3.3 nucleosomes follows the rapid removal of sperm-specific nuclear basic proteins (SNBPs) from the fertilizing sperm nucleus, after its delivery in the egg cytoplasm. In many animal species, during spermiogenesis, histones are progressively replaced with SNBPs, such as the well-characterized protamines [30]–[32]. The nature and extent of this replacement is highly variable in metazoans [32]. In Drosophila, protamine-like proteins are encoded by two paralogous genes named Mst35Ba and Mst35Bb [33], [34]. In this species, the vast majority of sperm DNA is packaged with protamines and with other non-histone SNBPs [21], [35], implying that de novo assembly of paternal nucleosomes at fertilization after SNBP removal must occur over the entire male genome.
We had previously shown that this unique RI assembly requires the conserved H3.3 histone chaperone HIRA [36], [37]. Indeed, loss of function mutations in Hira are viable in Drosophila, but nucleosome assembly in the male pronucleus is completely abolished in eggs laid by mutant females, resulting in the loss of the paternal set of chromosomes and the development of gynogenetic haploid embryos [36], [37]. In mice, HIRA is present in the decondensing male nucleus [38] and is most likely responsible for the strong paternal H3.3 enrichment observed in the zygote [38], [39]. Recently, HIRA has been implicated in the formation of the male pronucleus in the crucian carp [40], confirming the widespread role of this histone chaperone in paternal nucleosome assembly at fertilization.
The Hir/HIRA complex is composed of a small number of proteins that are conserved between yeast and human. In S. cerevisiae, the Hir chromatin assembly complex includes the HIRA-related proteins Hir1 and Hir2, Asf1 (Anti Silencing Factor 1), Hir3 and Hpc2 [41]–[43]. Hir3 is a poorly conserved protein related to Hip3 (S. pombe) and human CABIN1, but which does not seem to have an ortholog in Drosophila [43]–[45]. Hpc2 is functionally related to Hip4 in fission yeast and to the HIRA-associated proteins Ubinuclein 1 and Ubinuclein 2 (UBN1/UBN2) [8], [44], [46]–[48]. Interestingly, the strongest conservation between Hpc2 orthologs resides in a ∼50 amino-acid domain called HRD (Hpc2-Related Domain) or HUN (Hpc2-Ubinuclein-1) domain [44], [48] and to a smaller domain called NHRD [49]. In Drosophila, Yemanuclein (YEM; also named Yemanuclein-α [50], [51]) is the only protein with a HRD domain [44]. The yem gene has a strong ovarian expression and encodes a nuclear protein that accumulates in the germinal vesicle of growing oocytes [51]. Recently, a mutant allele of yem (yem1) has been characterized as a V478E replacement, which results in female sterility [52]. In this first report on YEM function, YEM was implicated in the segregation of chromosomes during the first female meiotic division but the sterility of mutant females suggested the existence of yet unknown roles for YEM [52]. In this paper, we have explored the implication of YEM in HIRA-dependent RI nucleosome assembly in the zygote. We show that the cooperation of YEM and HIRA in vivo is critical for the assembly of H3.3-containing nucleosomes in the male nucleus at fertilization.
The original yem1 point mutation causes a single amino-acid replacement (V478E) in YEM protein (Figure 1A) [52]. This mutation induces female sterility but has no detectable effect on the level of yem transcripts in ovaries nor on the accumulation of YEM protein in the oocyte nucleus (or germinal vesicle, GV) (Figure 1B, 1C). To obtain a more severe mutant allele of yem, we mobilized a P-element inserted near the transcriptional start site of the yem gene (Figure 1A). One of the imperfect excisions of this P-element generated a 3180 bp deletion (named yem2) that spans the 5′ UTR and most of the coding region of yem. Accordingly, the yem2 allele induced female sterility in association with yem1 or with the large non-complementing deficiency Df(3R)3450 (Table 1). In yem2/Df(3R)3450 females, yem transcripts (corresponding to a region of the gene not covered by the yem2 deletion) were greatly reduced compared to yem1/Df(3R)3450 females, and the YEM protein was not detected in the oocyte nucleus (Figure 1B, 1C). Finally, the female sterility of both yem mutant alleles was rescued by expressing a transgenic YEM protein tagged in its C-terminus with the Flag peptide (YEM-Flag) (Table 1). Taken together, these data suggest that yem2 is a null or at least a strong loss of function allele of yem.
The YEM protein has been previously detected in a HIRA complex purified from embryonic nuclear extracts [53], suggesting that it could represent the Drosophila ortholog of UBN1/Hpc2. To more directly test the interaction of HIRA and YEM, we performed co-immunoprecipitation experiments using functional Flag-tagged and GFP-Flag-tagged transgenic versions of YEM and HIRA proteins, respectively. We confirmed that, in ovarian protein extracts, HIRA was able to co-immunoprecipitate with YEM, and vice versa (Figure 2A). In the same experiments, however, the ATP-dependent chromatin remodeling factor CHD1 was not detected in the HIRA immune complex, in contrast to what was previously reported [54]. Although the reason for this apparent discrepancy with the study by Konev et al. is not clear, it reinforces the fact that, in our experimental conditions, HIRA and YEM show reproducible and specific interaction, confirming that these proteins are subunits of a common complex.
HIRA and YEM were previously shown to display a remarkable and specific accumulation in the nucleoplasm of the GV throughout oogenesis [36], [51]. Similarly, immunodetection of the Flag-tagged versions of HIRA and YEM recapitulates their endogenous accumulation in the GV, where both proteins co-localize (Figure 2B). The oocyte nucleus is a large nucleus that essentially contains nucleoplasm, as the oocyte chromosomes remain confined within a small, compact structure called the karyosome [55]. Surprisingly, we observed that HIRA-Flag accumulation in the GV was completely abolished in yem2/Df(3R)3450 mutant oocytes. Conversely, we found that YEM-Flag was undetectable in the GV of about half of null HiraHR1 mutant oocytes (Figure 2C). These effects could not be explained by reduced protein levels in mutant flies, as HIRA-Flag and YEM-Flag expression were apparently not affected in yem2/Df(3R)3450 an HiraHR1 mutants, respectively (Figure 2D). These results indicate that YEM and HIRA are mutually required for their localization or for their stabilization in the oocyte and suggest that these proteins interact prior to their release in the egg cytoplasm, after GV breakdown. Taken together, these results confirm that YEM and HIRA belong to the same complex in vivo.
The female sterility associated with yem1 or yem2 mutations actually results from a maternal effect embryonic lethality phenotype. Indeed, eggs from yem1/Df(3R)3450 or yem2/Df(3R)3450 females (referred to as yem mutant eggs for simplicity) are normally fertilized and they initiate development, but the embryos systematically die before hatching (Table 1 and not shown). These features are reminiscent of the maternal effect embryonic lethality phenotype of Hira mutants, where embryos develop as non-viable gynogenetic haploids after the loss of paternal chromosomes during the first zygotic division [36], [37], [56]. We thus examined male pronucleus formation in yem mutant eggs. In wild-type eggs, shortly after fertilization, while maternal chromosomes complete meiotic divisions, the decondensing male nucleus is strongly and specifically stained with an antibody recognizing acetylated histone H4, a mark of newly assembled chromatin [36], [37]. Strikingly, we observed that in yem mutant eggs, acetylated H4 was practically not incorporated in the male pronucleus (Figure 3A). At pronuclear apposition, male pronuclei in yem mutant eggs always appeared round and condensed (Figure 3B), in a way identical to the male nucleus in Hira mutants [36], [37], [57]. Paternal chromosomes subsequently failed to integrate the first zygotic division in yem eggs (Figure 3C), resulting in gynogenetic haploid development and embryonic lethality (Figure 3D). It should be mentioned however that exceptional gynogenetic development of adults can occur if the female pronucleus is diploid as the result of defective meiosis [52].
While the yem-flagHPF16 transgene efficiently rescued yem female sterility, another insertion of the same construct (yem-flagHPF1) only restored fertility to very low levels, likely because of its weak expression (Table 1). Interestingly, in eggs laid by yem1/Df(3R)3450; yem-flagHPF1 females, the male pronucleus still appeared round and condensed but consistently incorporated significant levels of acetylated histone H4 (Figure 3B). This suggests that the level of maternal YEM protein is limiting for both nucleosome assembly and male pronucleus decondensation.
We have previously shown that HIRA-dependent nucleosome assembly in the male pronucleus exclusively uses the histone H3 variant H3.3 [36], [37]. To observe H3.3 deposition in the male pronucleus, we used a previously described, maternally expressed Flag-tagged transgenic version of H3.3 (H3.3-Flag) [37]. In contrast to control eggs, H3.3-Flag was not incorporated in paternal chromatin of yem1 eggs, similarly to Hira mutants (Figure 4A). However, the female pronucleus in yem eggs still incorporated low levels of H3.3-Flag during the first round of DNA replication, arguing that, like HIRA, YEM does not participate to the limited S phase deposition of H3.3 which occurs in replicating nuclei of early embryos [36] (Figure 4A). As a complementary approach, we analyzed the yem mutant phenotype using a commercially available monoclonal anti-H3.3 antibody. In wild-type fertilized eggs, the antibody specifically stained the decondensing male pronucleus, but not the maternal chromosomes, thus confirming its specificity for H3.3 (Figure 4B). In agreement with the results obtained with H3.3-Flag, no staining was detected above background when Hira and yem mutant eggs were stained with the anti-H3.3 antibody (Figure 4C). Altogether, these results demonstrate the critical requirement of YEM for the assembly of H3.3-containing nucleosomes on paternal DNA.
Although mutant yem1/Df(3R)3450 and yem2/Df(3R)3450 adults were viable, survival rates were reduced for yem2/Df(3R)3450 individuals (Table S1) indicating that YEM also functions in somatic cells. Interestingly, the partial lethality of yem2 mutant individuals was not aggravated when combined with the HiraHR1 null allele. Thus, HIRA and YEM do not have redundant functions but, instead, are obligate partners not only for male pronucleus chromatin assembly but presumably also for other somatic RI nucleosome assembly processes.
Consistent with its critical role in paternal chromatin assembly, maternally expressed HIRA is recruited to the male nucleus shortly after fertilization in both Drosophila and mouse [37], [38]. Strikingly, while robust HIRA-Flag staining is observed in the decondensing male nucleus in control eggs, HIRA-Flag was not detected in eggs from yem1 and yem2 females (n>20; Figure 5D). Thus, YEM is required for the recruitment or for the stabilization of HIRA in the male nucleus. As expected, maternal YEM-Flag was also detected in the decondensing male nucleus before pronuclear apposition (Figure 5B). However, in contrast to the homogeneous distribution of HIRA-Flag in the male nucleus, YEM-Flag appeared also enriched in a small number of discrete foci of unknown nature (Figure 5A). We verified that these foci localized neither to the centromeres nor to the telomeres of the male pronucleus (Figure 5C). Interestingly, the formation of these YEM-Flag foci appeared largely independent of HIRA, whereas the rest of YEM-Flag was not detected in a large majority of Hira mutant eggs (Figure 5B). Thus, with the exception of these discrete regions, our experiments demonstrate that HIRA and YEM are interdependent for their localization within the male pronucleus and for paternal chromatin assembly.
Several groups have recently established that in mammalian cells, RI H3.3 deposition is mediated by at least two distinct protein complexes. HIRA and its partners are involved in the enrichment of H3.3 at active genes and at upstream regulatory elements of both active and repressed genes [6]. In contrast, ATRX, a member of the SNF2 family of ATP-dependent chromatin remodeling factors and the histone chaperone DAXX (Death-Associated protein) are essentially responsible for the enrichment of H3.3 nucleosomes at heterochromatin loci [6], [58]–[60]. In Drosophila, the ATRX homolog XNP (or dATRX) colocalizes with H3.3 throughout the chromatin of somatic cells [16]. To investigate the potential involvement of this chromatin remodeler in the assembly of paternal nucleosomes in the newly fertilized egg, we first determined its distribution in oocytes and eggs using a specific antibody recognizing both XNP isoforms [16]. Interestingly, XNP was found to accumulate in the oocyte nucleus, in a way remarkably similar to HIRA and YEM (Figure 6A). However, XNP was not observed in the decondensing male nucleus at fertilization (n>20) and the protein remained absent from early cleavage nuclei until their migration to the embryo periphery, at the syncytial blastoderm stage (Figure 6B and not shown). In addition, we observed that chromatin assembly in the male nucleus occurred normally in eggs from xnp2/xnp3 mutant females (n>20; Figure 6C). Finally, females homozygous for the semi-lethal allele xnp3, which abolishes the expression of the long XNP isoform [61], produced a limited amount of eggs that nevertheless hatched (not shown). We conclude that dATRX/XNP is most likely not involved the assembly of paternal nucleosomes at fertilization.
In human cells, the HIRA core complex is composed of at least three subunits, including HIRA, UBN1 and CABIN1 [43]. This complex is functionally involved in a large diversity of cellular and developmental processes that require dynamic histone turnover or de novo assembly of nucleosomes, independently of DNA synthesis. Although the HIRA complex mediates the deposition of the highly conserved H3.3 histone variant, its subunits display a comparatively weak overall conservation in animals. For instance, Drosophila does not seem to have any CABIN1 homolog and the highest conservation between UBN1 and YEM is mainly restricted to the small HRD domain. Despite this poor conservation, our work establishes Yemanuclein as a bona fide ortholog of Ubinuclein, by demonstrating its physical interaction with the HIRA histone chaperone and its critical requirement for H3.3 deposition during male pronucleus decondensation.
In contrast to the knock-out of the Hira gene in mouse, which is zygotic lethal in early embryos [62], null mutants of Drosophila Hira are viable but homozygous females are completely sterile [36]. This indicates that only the maternal contribution of Hira is essential, at least to form the male pronucleus. Our characterization of a null yem2 allele allowed us to reach the same conclusion for YEM. Remarkably, the phenotype of the male pronucleus in eggs laid by yem mutant females appeared indistinguishable to what we previously reported for Hira mutants. In both cases, RI deposition of H3.3-containing nucleosomes is practically abolished, typically preventing the full decondensation of the male nucleus and its integration into the zygotic nucleus. Thus, YEM and HIRA are equally required to assemble paternal nucleosomes at fertilization. This unique and major function of the HIRA complex is most likely conserved in animal groups where histones, and notably H3 and H4, are replaced with SNBPs in sperm. This is for instance the case of mammals, where protamines package about 95% and 85% of mouse and human sperm DNA, respectively [30], [32]. In fact, HIRA has been previously detected in the decondensing male nucleus at fertilization in mouse, which incorporates H3.3 before the first round of DNA replication [38], [39]. We thus expect Ubinuclein1/2 to be also involved in paternal chromatin assembly in mammals. In apparent contradiction with this prediction, a transgene expressing human UBN1 in the female germline could not rescue the sterility of yem mutant females (Figure S1 and not shown). However, this absence of complementation of YEM and UBN1 can be explained by the strong divergence of these orthologous proteins at the primary sequence level and it suggests that UBN1 can only function within its native, human HIRA complex. The apparent lack of a CABIN1 homolog in Drosophila also underlines the central role played by the HIRA-UBN1/YEM pair in the complex. Interestingly, while the implication of HIRA and UBN1 for RI deposition of H3.3 in vivo was recently demonstrated in human cells, CABIN1 seemed to play only an auxiliary role in this context [63]. Possibly, CABIN1 could be important for human-specific functions of the HIRA complex, such as the formation of senescence-associated heterochromatin foci [45], [64].
We had previously shown that HIRA specifically accumulates in the sperm nucleus shortly after its delivery in the egg cytoplasm [37]. Here, we have established that maternally expressed YEM similarly accumulates in the male nucleus at fertilization and until pronuclear apposition. Strikingly, we have also shown that HIRA and YEM are mutually dependent for their targeting to the male nucleus, strongly suggesting that these proteins physically interact during the assembly of paternal nucleosomes. However, nothing is known about the mechanism responsible for their rapid and specific localization in the fertilizing sperm nucleus, which is delivered in the cytoplasm of the gigantic egg cell. We had previously established that the HIRA-dependent assembly of paternal nucleosomes occurs after the removal of sperm protamines [36]. This opens the simple possibility that the HIRA complex could recognize exposed sperm DNA immediately after the removal of SNBPs. Interestingly, pioneer work on YEM by Aït-Ahmed et al. had established that this maternal protein was able to bind DNA in vitro [51]. This property could be important to efficiently target the HIRA complex to sites of de novo nucleosome assembly in the decondensing male nucleus. This hypothesis has recently received indirect experimental support in human cultured cells [63]. In their study, Ray-Gallet et al. established that HIRA, UBN1 and CABIN1 were all individually able to bind DNA in vitro and they proposed that this remarkable property could allow the HIRA complex to target naked DNA for H3.3 deposition. Accordingly, this HIRA-dependent nucleosome gap-filling mechanism has been shown to participate in the maintenance of genome integrity [63], but could also be employed, at the genome-wide scale, for de novo assembly of paternal chromatin at fertilization.
Finally, the observation that YEM accumulates in discrete nuclear regions in both the male nucleus (this study) and the oocyte karyosome [52] opens the possibility that YEM could perform additional roles not related to nucleosome assembly.
Despite its expression in the female germline, we found that Drosophila ATRX/XNP is not targeted to the male nucleus and does not seem to play any role in male pronucleus formation. Among the 17 SNF2 type chromatin remodelers present in Drosophila [16], the Chromodomain-helicase-DNA-binding protein 1 (CHD1) is the only one that has been implicated in the remodeling of paternal chromatin at fertilization [21], [54]. In contrast to Hira and yem, mutations in chd1 do not drastically affect H3.3 incorporation in paternal chromatin but still severely compromise the decondensation of the male nucleus, which appears aberrant in shape [21], [54]. In contrast to the HIRA/CHD1 interaction reported by Konev et al. [54], we could not detect any interaction between these proteins in ovaries, using experimental conditions that permitted co-immunoprecipitation of HIRA and YEM. Our results thus suggest that the role of CHD1 in the male nucleus is distinct from the nucleosome assembly process mediated by the HIRA complex.
Although the implication of the HIRA histone chaperone in paternal chromatin assembly was firmly established a few years ago, it has remained unclear until now if this highly specialized RI assembly process also involved other subunits of the HIRA complex or other histone deposition pathways. In fact, we have previously reported that the histone chaperone ASF1 [65], which is known to interact with both the CAF1 and HIRA complexes, was actually absent from the decondensing male nucleus [36]. Although the role, if any, of ASF1 in paternal chromatin assembly awaits a proper functional characterization, we do not expect this histone chaperone to be directly involved in the assembly of nucleosomes on paternal DNA. Accordingly, ASF1 has been previously shown to be dispensable for direct de novo RC or RI histone deposition in Xenopus egg extracts [66].
The complete failure of the male nucleus to assemble its chromatin in Hira or yem mutant eggs demonstrates that no other nucleosome assembly machinery can substitute for the HIRA-YEM complex in this peculiar context. However, the functional requirement of H3.3 itself in this process is not known. In Drosophila, H3.3 is not absolutely required for survival but it is essential for both male and female fertility [27], [28]. Viability of His3.3A; His3.3B double null mutants could be explained by the fact that, in the absence of H3.3, canonical H3 can be assembled in a RI manner [28]. Although the mode of RI deposition of replicative H3 in these mutants is not known, it opens the possibility that HIRA could use canonical H3 in certain critical circumstances, such as a limiting availability of H3.3. This compensatory mechanism, however, is apparently not possible in Drosophila spermatocytes, where H3.3 is required for the correct segregation of chromosomes during meiotic divisions, underlining the importance of this variant for sexual reproduction [28]. Similarly, future work should aim at determining whether H3.3 is specifically required for the assembly of paternal nucleosomes at fertilization.
Both HIRA and YEM proteins, which are presumably expressed from germinal nurse cells, display a remarkable accumulation in the oocyte nucleus during oogenesis [36], [51]. Most of the volume of the large germinal vesicle is devoid of DNA as the maternal genome is tightly packaged within the karyosome. The presence of HIRA and YEM in the nucleoplasm of the GV is thus not related to nucleosome assembly. However, the fact that HIRA and YEM are mutually dependent for their accumulation in the GV suggests that they are stored in this compartment as a complex. In contrast to the null alleles, point mutations do not affect HIRA/YEM localization in the GV, suggesting that the mechanisms controlling their recruitment to the GV or to the male pronucleus are distinct. This could reflect the fact that the HIRA complex is active in the male pronucleus where these proteins are in a chromatin environment in contrast to their nucleoplasm distribution in the GV. Whether or not this transient accumulation of HIRA/YEM in the GV plays any role in the maturation of the complex before paternal chromatin assembly at fertilization remains to be tested. Interestingly, it has been proposed that in human cells, formation of senescence-associated heterochromatin foci by HIRA requires its prior localization to promyelocytic leukemia nuclear bodies, suggesting that these structures could participate in the formation of the HIRA complex before its translocation to chromatin [48], [67]. It should be mentioned, however, that dATRX/XNP also accumulates in the GV despite its dispensability for paternal chromatin assembly. A recent study [68] reported the presence of several nuclear proteins in the GV with no known function in the oocyte, suggesting that this structure could serve as a storage compartment for a large number of nuclear proteins.
In conclusion, our characterization of Drosophila Yemanuclein demonstrates that this protein is a functional partner of HIRA in vivo. It also establishes that HIRA and YEM directly cooperate in the male nucleus for the genome-wide replacement of sperm protamines with H3.3-containing nucleosomes. The specific requirement of the HIRA complex in this unique developmental chromatin assembly process implies the existence of specific properties not shared with other H3.3-deposition pathways. In this regard, future work should explore the potentially conserved DNA binding property of the HIRA complex [51], [63] and its potential role in targeting the fertilizing sperm nucleus in animals.
Flies were grown in standard conditions at 25°C. The w1118 stock was used as a wild-type control in all experiments. The Hirassm and HiraHR1 alleles and the Hira-flag transgenic constructs have been described earlier [36], [37]. For the construction of the Hira-GFP-FLAG fusion gene, the eGFP coding sequence was inserted between the Hira and Flag tag sequences of PW8-Hira-3xflag [37]. The yem1 mutation is a T>A substitution falling in the fifth exon of yem which results in a V478E mutation [51], [52]. The GFP-K81 transgene is described in [69]. To mark paternal telomeres we used w1118/Y; 5′K81-GFP::K81; K812 males [69]. The w; P[w+, g-EGFP-cid]III.2 [70] stock has been kindly provided by Stefan Heidmann. The Df(3R)3450 deficiency, the P{EPgy2}EY23024 insertion and the xnp2 and xnp3 mutant alleles [61] were obtained from the Bloomington Drosophila Stock Center.
The yem2 mutation was isolated after standard remobilization of the P{EPgy2}EY23024 element and selected for its non-complementation of the yem1 chromosome. yem2 is a 3180 bp deletion from position +2 in the 5′UTR (positions 24945416 to 24948596 in the genome), uncovering the first 5 exons and part of exon 6 of the yem gene. Note that we only refer in this study to the original gene model [51] identified as RA in Flybase (Flybase ID# FBtr0085415) and not to the recently predicted longer RB transcript (see Flybase.org).
Total RNAs were extracted with the Trizol method (Invitrogen) from at least 50 whole adults, ovaries or carcasses. Reverse transcription was performed using oligo(dT) primers and the SuperScript First-Strand Synthesis system for RT-PCR (Invitrogen). For the yem and RP49 PCR reactions, the following primers were used YEMAPRIMER15/YEMAPRIMER16 and RP49FWD/RP49REV (see primers section).
YEMAPRIMER2: TGCGAAAACCGCGACCAGTG
YEMAPRIMER9: GGGCAGTTGTTGCGTGGATG
YEMAPRIMER15: GGATCCCATTCCTCCGCTTG
YEMAPRIMER16: CTCAGGCAGCAGCACTCAAT
RP49FWD: AAGATCGTGAAGAAGCGCAC
RP49REV: ACTCGTTCTCTTGAGAACGC
OA37: ACGTCCAAGCAGCTAGCTGCCA
OA38: GAATCTAGACTTGTCATCGTCGTCCTTGTAGTCTTGGCGCGTGGGCGTACT
Eggs were collected, dechorionated, devitellinized and fixed in methanol as described [56]. Eggs were then rehydrated in TBS-Triton 0,15% and incubated with primary and secondary antibodies at the indicated dilution. Finally, eggs were incubated in a 2 mg/ml RNAse A solution for 1 h at 37°C and were mounted in a mounting medium (DAKO S3023) containing 5 µg/ml propidium iodide. For anti-YEM AS2 antibody staining, ovaries were dissected in PBS-Triton 0,1% and were immediately incubated with the antiserum without fixation, stained with DAPI and mounted, as described [74]. For other experiments, ovaries were dissected in PBS-Triton 0,1% and fixed at room temperature in 4% PFA in PBS for 25 minutes. Ovaries were then stained with propidium iodide and mounted as described above. Slides were observed under an LSM 510 META confocal microscope (Zeiss). Images were treated with LSM image browser, Image J or Photoshop CS2 (Adobe).
We used the following antibodies: AS2 anti-YEM antibody (1/100; [51], [74]), M2 monoclonal anti-Flag antibody (1∶500 in ovaries, 1∶1000 in embryos; Sigma), anti-polyacetylated histone H4 (1∶200; Millipore 06-589), monoclonal anti-H3.3 (H3F3B) (1∶800, Abnova), anti-XNP [16] (1∶5000) and anti-UBN1 (1∶200) [75]. Secondary antibodies were Alexa488 goat anti-mouse or goat anti-rabbit (1∶1000, Invitrogen) and Cy3 donkey anti-rabbit (1∶800, Millipore).
50 µl of ovaries were homogenized in lysis buffer (15 mM Hepes (pH 7.6); 10 mM KCl; 5 mM MgCl2; 0.5 mM EDTA; 0.5 mM EGTA; 350 mM Sucrose; 1 mM DTT) with protease inhibitors (Halt Protease Inhibitor Single Use Cocktail, Thermo Scientific; 1 mM PMSF). The protein extract was centrifuged, isolated from debris and stocked in half volume of glycerol at −80°C. SDS-Page electrophoresis was carried out on 8% acrylamide gels and western blot was performed using standard procedures using Pierce ECL Western Blotting Substrate (Thermo Scientific). The following antibodies were used: M2 anti-Flag (1∶1000; Sigma), anti-Tubulin (1/1000; Sigma), Peroxydase-coupled goat anti-mouse (1∶10000; Beckman).
For co-immunoprecipitation experiments, we essentially used the protocol described in Jäger et al., 2001 [76] with some modifications as indicated. A hundred ovaries were dissected manually in 250 µl lysis buffer on ice. Lysis buffer was as described [76] to the exception of the protease inhibitors. In our conditions, Roche tablets of EDTA-free protease inhibitor cocktail were used as recommended by the supplier. PMSF was also added to a 1 mM final concentration. Before homogenization 250 µl ice-cold lysis buffer were added. The homogenates were cleared by centrifugation and the supernatant was adjusted to 1 ml in lysis buffer. The protein extracts were then submitted to the immunoprecipitation procedure after 2×30 µl were set aside to be used as input in western blot experiments. G-Sepharose beads (Sigma) were used as recommended by the supplier with the following antibodies at a 1/250 dilution: mouse monoclonal Flag M2 (Sigma) for HIRA and the AS2 rabbit polyclonal for YEM. Rabbit preimmune serum was used as negative control. Gel separation and western blots analysis were performed as indicated above. The rabbit CHD1 antibody (a gift from A. Lusser) was used at a 1/250 dilution. Secondary antibodies were goat peroxydase-coupled anti-mouse and anti-rabbit antibodies (1∶10000; Beckman). Revelation was performed with the Millipore Immobilon Western Chemiluminescent substrate as recommended by the supplier.
|
10.1371/journal.pntd.0002906 | Cytokine Production but Lack of Proliferation in Peripheral Blood Mononuclear Cells from Chronic Chagas' Disease Cardiomyopathy Patients in Response to T. cruzi Ribosomal P Proteins | Trypanosoma cruzi ribosomal P proteins, P2β and P0, induce high levels of antibodies in patients with chronic Chagas' disease Cardiomyopathy (CCC). It is well known that these antibodies alter the beating rate of cardiomyocytes and provoke apoptosis by their interaction with β1-adrenergic and M2-muscarinic cardiac receptors. Based on these findings, we decided to study the cellular immune response to these proteins in CCC patients compared to non-infected individuals.
We evaluated proliferation, presence of surface activation markers and cytokine production in peripheral blood mononuclear cells (PBMC) stimulated with P2β, the C-terminal portion of P0 (CP0) proteins and T. cruzi lysate from CCC patients predominantly infected with TcVI lineage. PBMC from CCC patients cultured with P2β or CP0 proteins, failed to proliferate and express CD25 and HLA-DR on T cell populations. However, multiplex cytokine assays showed that these antigens triggered higher secretion of IL-10, TNF-α and GM-CSF by PBMC as well as both CD4+ and CD8+ T cells subsets of CCC subjects. Upon T. cruzi lysate stimulation, PBMC from CCC patients not only proliferated but also became activated within the context of Th1 response. Interestingly, T. cruzi lysate was also able to induce the secretion of GM-CSF by CD4+ or CD8+ T cells.
Our results showed that although the lack of PBMC proliferation in CCC patients in response to ribosomal P proteins, the detection of IL-10, TNF-α and GM-CSF suggests that specific T cells could have both immunoregulatory and pro-inflammatory potential, which might modulate the immune response in Chagas' disease. Furthermore, it was possible to demonstrate for the first time that GM-CSF was produced by PBMC of CCC patients in response not only to recombinant ribosomal P proteins but also to parasite lysate, suggesting the value of this cytokine to evaluate T cells responses in T. cruzi infection.
| Chronic Chagas' disease Cardiomyopathy (CCC) is the most frequent and severe consequence of the chronic infection by protozoan parasite T. cruzi. Patients with CCC develop high levels of antibodies against ribosomal P proteins of T. cruzi, called P2β and P0. These antibodies can cross-react with, and stimulate, the β1-adrenergic and M2 muscarinic cardiac receptors, inducing a functional and pathological response in cardiomyocytes. In this study, we focused on the cellular immune response developed by CCC patients in response to T. cruzi ribosomal P proteins. Peripheral blood mononuclear cells (PBMC) from CCC patients stimulated with both proteins neither proliferated nor induced the expression of activation markers on CD4+ and CD8+ T cells. However, these cells responded by the secretion of IL-10, TNF-α and GM-CSF, giving evidence that there is indeed a pool of specific T cells in the periphery responsive to these proteins. Interestingly, the cytokines profile was not related with those described to whole parasite lysate or other recombinant proteins, suggesting that each parasite protein may contribute differently to the complex immune response developed in patients with Chagas' disease.
| Trypanosoma cruzi, the etiological agent of Chagas' disease, affects approximately 8–10 million people, and its infection is one of the major human health problems in Central and South America, being extended now to Europe (especially Spain and Portugal), the United States, Canada, Japan and Australia [1], [2], [3]. Upon exposure to the parasite, the humoral and cellular immune responses elicited by the host, keep acute parasitemia under control [4], [5]. However, approximately 30–40% of the infected individuals, several years after initial exposure, develop clinical symptoms of visceral damage, which may include cardiac lesions, digestive alterations or both manifestations (cardiac plus digestive) [5]. Chronic Chagas' disease Cardiomyopathy (CCC), the most frequent and severe consequence of the chronic infection by T. cruzi, is manifested predominately as an arrhythmogenic cardiomyopathy [6]–[9].
Up to now, the mechanisms of the pathophysiology of Chagas' disease are not completely elucidated and two main hypotheses have been proposed. The first one is based on the inflammatory reaction elicited by the parasite leading to tissue damage, while the second argues for an autoreactive process resulting from an impaired immune response associated with molecular mimicry [10]–[13]. However, it is currently accepted that both mechanisms are not mutually exclusive and that Chagas' disease is the result of both, parasite persistence in the chronic phase and the presence of autoantibodies/self-reactive T cells to host molecules [14], [15]. As supporting evidence for the autoimmune hypothesis, previous work in our laboratory demonstrated the presence of circulating antibodies against ribosomal P proteins of T. cruzi (anti-P Abs) with agonist-like properties on cardiac receptors in patients with CCC [16]–[24]. Those Abs predominantly recognized the C-terminal end of P2β (peptide R13, EEEDDDMGFGLFD) or P0 proteins (peptide P015, EEEDDDDDFGMGALF), which bear structural similarity to the acidic motif, AESDE, located on the second extracellular loop of the cardiac receptor [19], [20], [22]. Several studies including patients with CCC as well as experiments performed in mice immunized with recombinant P2β or P0 protein demonstrated a correlation between the presence of anti-P Abs and cardiac disorders [21], [22]. These findings were confirmed by the generation of anti-R13 monoclonal Ab, mAb 17.2, which not only induce a dose-dependent increase on the beating frequency of rat cardiomyocytes in culture that is abolished by bisoprolol, a specific β1-adrenergic receptor antagonist [25], but also provoke apoptosis in the murine cardiac cell line HL-1 by its long-lasting β1-AR stimulatory activity [24]. The humoral immune response against ribosomal P proteins has been largely studied in patients with CCC; however, little is known about their recognition by T cells.
Most studies concerning the T cell immune response in Chagas' disease, have been performed using freshly isolated peripheral blood mononuclear cells (PBMC) but stimulated with epimastigote (the replicative form found in the midgut of insect vectors) or trypomastigote (the infective form found in the bloodstream and other human extracellular fluids) lysate [26]–[29]. Few investigations have been focused on the reactivity of T cells against purified antigens of the parasite [30]–[40]. To date, studies performed with recombinant parasite proteins, such as the cytoplasmatic repetitive antigen (CRA), B13, trans-sialidase, and paraflagellar rod proteins on PBMC and cruzipain on T cells lines revealed that patients with CCC produced significant amount of IFN-γ upon stimulation, which is in line with the typical pattern of inflammatory response described for T. cruzi lysate [34]–[40]. However, Lorena et al. also reported that the flagellar repetitive antigen (FRA) induced proliferation of PBMC by thymidine incorporation, but no difference was observed in IFN-γ and TNF-α secretion between patients with CCC and non-infected individuals [37]. The aim of this study was to analyze the cellular immune response developed in patients with CCC against T. cruzi ribosomal P proteins, knowing the existence of a cross-reactive component at the humoral level. The specificity of the response was analyzed by proliferation and cytokine production using multiplex technology because it allows to quantify a large spectrum of cytokines in the same cell culture supernatant. Results showed that T. cruzi ribosomal P proteins, specifically P2β and the C-terminal portion of P0 (CP0, 110 aa), did not induce the proliferation of PBMCs from CCC in a different manner than non-infected individuals. However, these antigens were able to induce the secretion of IL-10, TNF-α and GM-CSF by PBMC as well as both CD4+ and CD8+ T cells in patients with CCC. Surprisingly, ribosomal P proteins did not stimulate but actually reduced the secretion of IFN-γ in cardiac patients. Furthermore, our results demonstrate for the first time that GM-CSF is produced in response not only to parasite lysate but also to ribosomal P proteins. These findings suggest that GM-CSF production could be included in the future to evaluate whole parasite and parasite protein specific T cell responses in Chagas' disease.
The research protocols followed the tenets of the Declaration of Helsinki and were approved by the Medical Ethics Committee of Ramos Mejía and Fernández Hospitals. All enrolled patients gave written informed consent, according to the guidelines of the Ethical Committee of the Hospitals, before blood collection and after the nature of the study was explained.
Patient selection was conducted at the Cardiovascular Division of the Ramos Mejía and Fernández Hospitals, Buenos Aires, Argentina. Positive serology for Chagas' disease was determined by two or more tests (indirect immunofluorescence, enzyme-linked immunosorbent assay [ELISA], indirect hemagglutination, or complement fixation). Patients who had at least two of three tests were considered positive for Chagas' disease. Patients underwent a complete clinical and cardiologic examination that included medical history, physical examination, electrocardiogram (ECG) at rest, laboratory and chest X-ray analysis, and echo doppler cardiography evolution. The exclusion criteria included the presence of systemic arterial hypertension, diabetes mellitus, thyroid dysfunction, renal insufficiency, chronic obstructive pulmonary disease, hydroelectrolytic disorders, alcoholism, history suggesting coronary artery obstruction and rheumatic disease, and the impossibility of undergoing the examinations. The study population consisted of 27 patients who completed the screening protocol and were diagnosed with Chronic Chagas' disease Cardiomyopathy.
Twenty non-infected individuals (NI), within the same age range (30–70 years old) and showing negative serological tests for Chagas' disease, were included as control group.
Due to its predominant clonal proliferation, the T. cruzi species is composed by multiple strains showing extensive genetic diversity, which were recently grouped into 6 evolutionary lineages or discrete typing units (DTUs) known as TcI to TcVI [41].
Gluthatione S-transferase (GST)-fusion proteins bearing the entire TSSA from Sylvio X-10/1 strain (henceforth TSSA Sy, representative of TSSA isoforms from DTU TcI parasites) and CL Brener strain (henceforth TSSA CL, representative of TSSA isoforms from DTUs TcII/TcV/TcVI parasites) were expressed in Escherichia coli BL21 strain and purified as described [42]. Briefly, supernatants of bacterial cultures transformed with the indicated construct were induced for 3 h at 28°C with 0.250 mM isopropyl—β-D-thiogalactopyranoside, purified by glutathione-Sepharose chromatography and extensively dialyzed against PBS. The purity and integrity of GST-TSSA samples was assessed with silver-stained SDS-PAGE gels [42].
Whole antigenic lysate from T. cruzi epimastigotes was prepared as described previously [43]. Briefly, fresh epimastigotes (CL Brener, DTU Tc VI) cultured in a liquid medium (liver infusion tryptose), were collected by centrifugation and washed three times with PBS. After centrifugation at 500xg during 5 min, the parasites were resuspended in lysis buffer (PBS, EDTA 1 mM, β-mercaptoethanol 5 mM, 0.1% SDS and protease inhibitors cocktail) and submitted to three cycles of freezing-thawing. The parasite lysate was diluted with PBS at 1 mg/ml, filter sterilized on 0.2 µm-pore-size membranes, assayed for protein concentration, aliquoted, and stored at −80°C until use.
The T. cruzi recombinant proteins selected for this study were P2β-His and CP0-His; this last one corresponds to the C-terminal portion of P0 (110 aa). The ribosomal P proteins were obtained and purified by means as His6-tag as described [44]. The purity and specificity of the recombinant proteins were analyzed by SDS-PAGE gels and Western-blot with a pool of chagasic and non-infected sera. Protein concentration was determined by Bradford (BioRad, Hercules, CA, USA), using BSA (Sigma, St Louis, MO, USA) as standard protein.
Peptides were prepared by solid-phase method of Merrifield as described by Müller et al. with a semi-automatic multi-synthesizer NPS 4000 (Neosystem, Strasbourg, France) [45]. Their purity was assessed by High Performance Liquid Chromatography (HPLC) and identified by mass spectrometry. Peptide R13 (EEEDDDMGFGLFD) was derived from the 13 C-terminal amino acids of P2β, P015 (EEEDDDDDFGMGALF) from 15 C-terminal region of P0 protein, and peptide H13 (EESDDDMGFGLFD) was derived from the corresponding region of mammalian ribosomal P proteins [46]. For ELISA, these peptides were coupled at a molar ratio of 1∶30 to BSA (Sigma, St Louis, MO, USA) with 0.05% glutaraldehyde as previously described [45]. The products were assessed by analytical HPLC and amino acid analysis was used to calculate the peptide–BSA molar ratio.
Microwell plates (Nunc Maxisorp) were coated overnight at 4°C with 50 ng protein/well of T. cruzi lysate, 2 µg/well of recombinant proteins P2β-His and CP0-His or 2 µM of synthetic peptide in 50 µL of 0.05 M carbonate buffer pH = 9.6. Plates were washed with PBS containing 0.1% Tween-20 (PBST) and then blocked with PBST containing 2.5% non-fat dry milk (PMT) for 1 h at 37°C. After washing, 50 µL of each diluted human serum (dilution 1/200 in PMT) was loaded onto plates and incubated for 1 h at 37°C. Following washing, plates were incubated with 50 µl of peroxidase-conjugated goat anti-human IgG (dilution 1/3,000 in PMT) (Sigma, St Louis, MO, USA). Enzyme activity was revealed with TMB and, OD was read at 415 nm with an Automated Plate Reader (Molecular Devices, CA, USA). All samples were tested in duplicate. Sera from 8 non-infected individuals were also included on the plate to determine the baseline level, as the OD mean value +3 SD. Antibody level is expressed as Reactivity index which was determined as the OD mean value of each serum sample/baseline value.
Peripheral blood mononuclear cells (PBMC) were isolated from heparinized blood by Ficoll-Hypaque density gradient centrifugation (GE HealthCare, Uppsala, Sweden), washed once and resuspended in RPMI-1640 medium containing 100 U/ml penicillin, 100 mg/ml streptomycin, 2 mM L-glutamine and 5% of AB Rh-positive heat-inactivated normal human serum (Sigma, St Louis, MO, USA). Cell suspensions (200 µl) were cultured as triplicates in the presence or absence of different stimuli for 4 or 6 days at a density of 2.5×105 cells/well in 96-well sterile plates (round bottom). Stimuli used in the cultures included T. cruzi lysate, P2β-His, CP0-His (at a final concentration of 10 µg/ml for 6 days), peptides R13, P015 and H13 (at a final concentration of 5 µg/ml for 6 days) while PHA (Phitohemaglutinin, Sigma, at a final concentration of 5 µg/ml for 4 days) was used as positive control. All concentrations were determined by performing titration experiments. After the incubation period, cultures were exposed to 1 μCi/well of 3H-thymidine (3H-TdR, specific activity, 2 Ci/mmol, Amersham, Arlington Heights, IL) for 6 h and then harvested on glass fiber filters. The incorporated radioactivity was determined by liquid scintillation counting. All cultures were performed in triplicate. Results are expressed as Stimulation Index, calculated as the mean cpm of stimulated cultures/mean cpm of non-stimulated (culture medium only) cultures.
2.5×106 cells were cultured in 24-well plates in 1 ml cultures for 6 days with either medium alone, or T. cruzi lysate, P2β-His, CP0-His (at a final concentration of 10 µg/ml). After centrifugation, cells were washed, resuspended in ice-cold PBS, stained for 30 min at 4°C with the following fluorescent-labeled monoclonal antibodies: allophycocyanin (APC) conjugated anti-CD3 + phycoerythrin-cyano dye Cy5 (PE-Cy5) conjugated anti-CD4 + phycoerythrin (PE) conjugated anti-HLA-DR + fluoresceinisothiocyanate (FITC) conjugated anti-CD25, or APC anti-CD3 + PE-Cy5 anti-CD8 + PE anti-HLA-DR + FITC anti-CD25. Cells were then fixed with 4% formaldehyde in PBS and kept at 4°C until analyzed by flow cytometry. In all cases, 10,000 to 15,000 events in the lymphocyte gate were acquired using a FACSAria flow cytometer (Becton Dickinson). Phenotypic analyses were carried out with FlowJo flow cytometric analysis software (TreeStar), selecting the small lymphocyte population. PBMC stained with FITC, PE-, APC- and PE-Cy5- labeled Ig control Abs were included in all experiments for background fluorescence. All Abs were purchased from BD Biosciences (San Diego, CA, USA).
CD8+ T cells were isolated from PBMC by positive selection using EasySep CD8 Selection Kit (StemCell Technologies, Inc., Vancouver, Canada), while CD4+ T cells were separated from CD3+CD8neg T cells by negative selection (EasySep CD3 Selection Kit, StemCell Technologies). The purity of both populations was assessed by flow cytometry using specific conjugated mAb (see “Phenotypic analysis of PBMC”) and, it was shown to be higher than 90% for both T cells subsets.
IL-2, IL-4, IL-10, IL-13, IL-17, IFN-γ, GM-CSF and TNF-α were measured in the supernatants of whole PBMC cultures stimulated in the presence or absence of the indicated antigens and collected on days 1, 2 and 6 after stimulation. In addition, the same cytokines were quantified in cultures of isolated CD4+ or CD8+ T cells (5×105 cells) co-cultured with irradiated CD3neg T cells (ratio 1∶1) in the presence or absence of antigen after 6 days of stimulation. Cytokines were measured by using MILLIPLEX MAP Human Cytokine/Chemokine Kit (for 8 cytokines) following the manufacturer's directions (Millipore, St Charles, MO) and Luminex instrument and Beadlyte software were used for analysis. All samples were tested in duplicate. Results are expressed in ng/ml or Fold increase (FI) which was determined as [(cytokine in stimulated culture) - (cytokine in NS culture)]/(cytokine in NS culture), where NS denotes non-stimulated cultured PBMC.
Statistical analysis was performed with GraphPad Prism statistical software (GraphPad Software). The nonparametric Mann-Whitney U test was used to generate P values comparing the median experimental values between groups each of the multiple sets of experimental data. Within each experiment, overall statistical significance of each result at both 10% and 5% significance was determined using Holm-Bonferroni Correction. Differences were considered statistically significant at P<0.05.
Patients included in this study were all born in endemic areas from Argentina and Bolivia, and at the time of the enrollment they have been living in Buenos Aires (where no vectorial transmission occurs) for more than ten years, in average. The mean age was 54.2±10.1, and 57% were female.
All T. cruzi-infected subjects were in the chronic phase of Chagas' disease, involving only cardiac alterations. According to the New York Heart Association (NYHA) functional classification system, patients were classified as Class I, II, III/IV. Patients with no functional limitations but with some electrocardiographic alterations were classified as Class 0 [5].
Blood samples yielded negative results for currently used PCR protocols targeting parasite DNA [47], which is frequently the case in chronic chagasic patients due to low parasitemia. Taking this into account, we analyzed the profile of the humoral anti-TSSA (trypomastigote small surface antigen) response in our study patients as an indirect means of identifying the genotype of the infecting strain(s) [48], [49]. To carry out this analysis, we evaluated the reactivity of serum samples against either TSSA Sy (the TSSA isoform from DTU TcI) or TSSA CL (the TSSA isoform from DTUs TcII/V/VI) in conventional ELISA and dot-blot (see Text S1 for details and Figure S1).
The main characteristics of the study population are summarized in Table 1.
To characterize the humoral response in the subject population included in this study, the antibody reactivity against T. cruzi lysate, ribosomal P proteins, P2β and CP0, together with their C-terminal peptides R13 and P015 was determined in sera of CCC patients and non-infected individuals by ELISA. The reactivity against peptide H13, which corresponds to the C-terminal region (residues 102–115) of the human ribosomal P protein was also measured.
Results showed that sera from CCC patients presented reactivity against T. cruzi lysate, with titers ranging from 1/200 to 1/20,000 (Table 1). Only patient P19 showed a titer against parasite proteins similar to those detected in non-infected individuals (<1/200 at OD = 1). Although antibodies in the serum sample from this patient were not detected by in-house ELISA, two of three serological tests for T. cruzi infection, together with clinical and cardiological examinations confirmed patient P19 to have CCC. In addition, sera of all patients, including P19, reacted with a broad range of T. cruzi proteins as determined by Western-blot (data not shown).
The majority of CCC patients (24/27) showed reactivity (Reactivity index >1.7) to ribosomal P2β protein and its peptide R13. The level of anti-CP0 antibodies was also elevated in the chagasic patients (17/27) compared to non-infected individuals, but the overall reactivity was lower than that observed for P2β protein (Figure 1). On the other hand, only marginal differences were determined in the median of the Reactivity Index for the anti-P015-antibodies in cardiac patients in comparison with non-infected subjects. No difference was observed against peptide H13 (human P ribosomal protein derived) between both groups of individuals (Figure 1). Together, these results showed that CCC patients mount a significant antibody response to ribosomal proteins as well as to peptides R13 and to a lower level to P015 in comparison to non-infected subjects.
In order to investigate the cellular response to ribosomal P proteins, PBMC from CCC patients and non-infected individuals were tested for their proliferative capacity in response to different T. cruzi antigens. To determine the optimal protein and peptide concentration yielding the most consistent results, the proliferative response was initially assayed in PBMC cultures from 4 cardiac patients non-included in this study. The results showed that 10 µg/ml of T. cruzi lysate or ribosomal P proteins and 5 µg/ml of the peptides were optimal to trigger proliferative responses, and so these concentrations were used in the studies presented here.
As shown in Figure 2, the majority of PBMC from CCC patients proliferated upon stimulation with T. cruzi lysate (Stimulation index median: 4.45) compared to PBMC from non-infected individuals (Stimulation index median: 1.07; P<0.001). On the contrary, the stimulation index of PBMC from cardiac patients and control subjects in response to ribosomal P proteins (Figure 2) as well as to peptides R13, P015 and H13 was not significantly different (data not shown). PBMC from all subjects proliferated in response to PHA and the responses were not significantly different between the cardiac and non-infected individuals (data not shown).
To characterize the phenotype of the cells after the stimulation with the different stimuli, cells were stained with different T cell markers and analyzed by flow cytometry. The forward vs side scatter dot plots revealed that the frequency of lymphocyte population in non-stimulated cultures was significantly lower in cardiac patients compared with non-infected individuals (48±13% vs 62±10%, respectively; P<0.001). However, the CD3+CD4+:CD3+CD8+ ratio was approximately 2∶1 in both groups. Interestingly, results showed that CCC patients present higher subsets of CD25 and HLA-DR positive cells on both CD3+CD4+ and CD3+CD8+ populations upon T. cruzi stimulation (Figure 3). However, the expression of these markers was similar in T cells from cardiac patients and non-infected individuals when cells were stimulated with ribosomal P proteins (Figure 3).
Given the lack of proliferative response to ribosomal P proteins in the CCC patients, T cell activation was studied by analyzing cytokine secretion. Thus, PBMCs from 10 cardiac patients with different disease severity, and 8 non-infected donors were stimulated with P2β and CP0 proteins and T. cruzi lysate as well as PHA as positive control. Supernatants after 1, 2 and 6 days post-stimulation were collected and multiplex analysis was performed to evaluate the levels of GM-CSF, IFN-γ, IL-10, IL-13, IL-17, IL-2, IL-4 and TNF-α. Despite the fact that cytokine responses have been studied by others after T. cruzi stimulation in patients with Chagas' disease [50], [51], reports have used different assays and stimulation/culture conditions making the direct comparison of all the cytokines difficult to achieve. In this study, we aimed to simultaneously evaluate the kinetic responses of multiple cytokines in the same culture well. Figure 4 shows the maximum fold increase detected for each cytokine and in each subject among day 1, 2 and 6 determinations. The fold increase was determined by the difference between cytokine production (in pg/ml) in stimulated wells and the cytokine production in non-stimulated control wells divided the cytokine production in non-stimulated control wells. The actual fold increase for each of the days and the background production in pg/ml of each of the cytokines in non-stimulated wells are shown in Figures S2 to S5 and S6, respectively.
Upon stimulation with ribosomal P proteins, GM-CSF, IL-10 and TNF-α were secreted at higher levels in cardiac patients compared with non-infected individuals (Figure 4 and Figure S2 and S3). However, both proteins induced similar levels of IFN-γ production in PBMC from cardiac patients and non-infected subjects (Figure 4). Furthermore, the fold increase of IFN-γ production in response to both proteins was lower and statistically significant in the cardiac group after only the first days post-stimulation (Figure S2 and S3). The level of IL-2, IL-4, IL-13 and IL-17 secreted after stimulation with the ribosomal P proteins was very low or null at any of the 3 time points analyzed and, it was found to be similar between CCC patients and non-infected individuals (Figure 4 and Figure S2 and S3).
A larger number of cytokines were produced in response to T. cruzi lysate or the universal stimulus PHA than in response to the individual ribosomal P proteins (Figure 4). Indeed, PBMC from cardiac patients in response to T. cruzi lysate also secreted statistically significant and higher levels of IFN-γ, IL-2 and IL-13 compared with non-infected individuals. IFN-γ and IL-13 were also increased in CCC patients vs non-infected individuals when PHA was used for stimulation. These results indicate that although the cells were capable of producing IFN-γ and IL-13 in response to whole parasite or PHA, their production was not detected when the ribosomal P proteins were used as stimulus. The kinetic cytokine profile for T. cruzi lysate and PHA is shown in Figure S4 and S5.
The results presented above revealed a cytokine signature expression upon stimulation with ribosomal P proteins and T. cruzi lysate in whole PBMC. To better understand the specific contribution of the T cells to this profile, CD3+CD4+ and CD3+CD8+ T cell subsets from three cardiac patients were enriched from PBMC and stimulated with the antigens in the presence of autologous antigen-presenting cells. Samples from patients RM11, RM12 and RM14 were chosen since they were among those that showed clear cytokines response after ribosomal P proteins stimulation.
As shown in Figure 5, GM-CSF was overall produced by both, CD4+ and CD8+ subsets by the 3 patients in response to the proteins and T. cruzi lysate. In general, IFN-γ was produced at very low levels by CD4+ and CD8+ T cells in all patients in response to the proteins, but enough to be different from the non-stimulated wells in the case of CD4+ T cells (Figure 5). IL-10 was found to be secreted most frequently by both T cell subsets. IL-13 was not produced by CD8+ T cells in any of the 3 patients analyzed and in response to all the stimuli tested. However, IL-13 was produced by CD4+ T cells in response to T. cruzi lysate and/or the proteins in the 2 of the 3 patients (RM11 and RM14). TNF-α was produced by both, CD4+ and CD8+ T cells and its production was higher in response to the proteins than to T. cruzi in 2 of the 3 patients. IL-2, IL-4 and IL-17 were not detected in response to any of the stimuli (data not shown).
Since it has been widely demonstrated the relevance of antibodies directed to ribosomal P proteins in the pathophysiology of Chagas' disease [21], [23], [24], this study aimed to further understand the cellular immune response raised against these proteins in CCC patients. Our results showed that PBMC did not proliferate upon in vitro stimulation with P2β and CP0 proteins. Additionally, the lack of proliferation in response to the proteins was associated with the absence of the expression of activation markers CD25 and HLA-DR on CD4+ and CD8+ T cell populations. These findings were also protein-specific, since T. cruzi lysate provoked an augmentation of both markers on the surface of T cells in agreement with data published by others [50], [51]. Interestingly, the percentage of both T cell subtypes, CD3+CD4+ and CD3+CD8+ in PBMC were similar in cardiac patients and non-infected individuals independently of the stimulus. These results suggest that the lack of proliferative response was not due to an overall decrease on the size of the T cell population, nor to a shutdown of the proliferative capacity in these patients since the same cells responded to T. cruzi lysate and a T cell specific universal mitogen such as PHA. However, it was possible to speculate that T cells specific to these proteins have been deleted by negative selection due to the similarity to the host specificities. In this regards, the analysis of the T cell response by cytokine release discarded this possibility since indeed, several cytokines were expressed in response to ribosomal T. cruzi proteins.
The use of multiplex technology allowed us to simultaneously analyze 8 cytokines, namely, IL-2, IL-4, IL-10, IL-13, IL-17, IFN-γ, TNF-α and GM-CSF, corresponding to well-described CD4+ and CD8+ associated cytokines. In particular, GM-CSF was included because not only its production has been associated to antigen mediated activation of T cells by us and others but also, the threshold of antigen requirement for its production is lower than for other cytokines as TNF-α, IL-4 or IFN-γ [52]–[54].
Our results showed that PBMC from CCC patients secreted high levels of GM-CSF, IL-10 and TNF-α in response to P2β and CP0 proteins. Interestingly, the secretion of IFN-γ at day 1 and 2 post-stimulation with ribosomal P proteins was similar or lower in cardiac patients vs non-infected individuals. Moreover, our data demonstrated that patients with CCC developed a different cytokine profile in response to T. cruzi and PHA stimulation than non-infected subjects.
Even though the secretion of GM-CSF, IL-10 and TNF-α in response to the proteins was significantly higher in cardiac vs non-infected individuals (P<0.05, nonparametric Mann-Whitney U Test), these P values nonetheless did not stay significant at the 5% level when a multiple comparison (all 32 cytokine/stimulus pairings) was performed by using Holm-Bonferroni correction. In contrast, the P values for these cytokines in response to T. cruzi lysate did reach statistical significance at the 5% level. This difference could be explained by the fact that the frequency of single specific parasite protein T cells within the bulk population is lower than the frequency developed in response to whole T. cruzi lysate and therefore it leads to lower cytokine secretion levels. However, it is important to remark that were the P values distributed at random amongst the proteins data, there would be only a chance of the three exact same cytokines (GM-CSF, IL-10 and TNF-α) being secreted in response to both proteins, demonstrating that the difference observed between cardiac patients and non-infected individuals was not a mere coincidence.
Following with T. cruzi lysate response, we observed that all studied cytokines were elevated and significantly different in the supernatants of cultured PBMC from cardiac patients with exception of IL-4 and IL-17. Upon PHA stimulation, PBMC from cardiac patients secreted higher amount of GM-CSF, IFN-γ, IL-10, IL-13, and TNF-α; similar production was observed for IL-2, IL-4 and IL-17 between both groups of individuals. In addition, and independently of the stimulus, our results also showed that these cytokines were secreted by both T cells populations, except for IL-13 which was predominantly produced by CD4+ T cells. Despite this finding, it is well-known that non-T cells, such as monocytes or B cells, also participate in the secretion of these cytokines. Indeed, Gomes et al. [28], by intracellular cytokine staining, reported that the majority of the IL-10-producing cells are monocytes (CD14+ cells) in asymptomatic patients, and the same group recently demonstrated that CD19+ B cells is another important source of this cytokine in cardiac patients [55].
Furthermore, the spontaneous release of cytokines in non-stimulated PBMC, which provides information about the basal level of cytokine production in vivo, showed a lower level for IFN-γ, IL-10, IL-13, and TNF-α in CCC patients (Figure S6). It should be mentioned that Dutra et al. demonstrated that the expression of IFN-γ, IL-10, IL-13 mRNAs was increased in PBMC from chagasic patients [33]. However, this discrepancy could depend either on the use of ex vivo PBMC or on the methodology used to determine cytokine expression. Our data, together with those reported by Giraldo et al. [56], may suggest that T. cruzi persistence provokes a general dysfunction in peripheral T cell response.
The high levels of pro-inflammatory cytokines, like IFN-γ and TNF-α, together with undetectable IL-4 production in response to PHA and T. cruzi stimulation suggest that there is a shift towards polarized Th1-type of cytokine response in CCC patients. Although IL-10 was first described related to Th2 cells, now is known that is produced by all T cells, including Th1 and a regulatory T cell subsets, called Tr1 cells or IL-10-producing cells [57]. Recent studies with an experimental murine model revealed not only the protective role of IL-10 against fatal myocarditis, but also demonstrated that this cytokine was produced by both CD4+ and CD8+ subsets of IFN-γ+IL-10+ double-producing T cells [58]. Similar data were obtained in studies by Belkaid et al. [59], where the main source of IL-10 in dermis and draining nodes of mice infected with Leishmania major is a subset of CD4+ T cells that produce both IL-10 and IFN-γ.
Studies performed with others recombinant parasite proteins demonstrated that the majority of chagasic patients develop a strong humoral and cellular immune response with a tendency to the typical pattern of inflammatory response described for T. cruzi lysate [34]–[40]. On the contrary, the cytokines released upon ribosomal P proteins stimulation made difficult to set a specific Th cells responsible for their secretion. This mixed cytokine profile which could be involved in balancing heart tissue damage and parasite persistence during chronic disease, strengthens in part the fact that B cells, through antibodies directed against P2β and CP0 and not T cells, would have the major role in the development of cardiac symptoms by their interaction with β1-adrenergic and M2 muscarinic receptors.
Interestingly, GM-CSF was secreted at high levels by PBMC from CCC patients when T. cruzi lysate, and both ribosomal P proteins were used as stimulus. To our knowledge, this is the first time that GM-CSF is used to evaluate the T. cruzi specific response of stimulated PBMC from cardiac patients. Instead, GM-CSF has been associated to a decrease in the rate of infection of both non-activated and IFN-γ activated macrophages infected with T. cruzi [60]. Moreover, Olivares Fontt et al. reported that the administration of exogenous recombinant GM-CSF improved the deficient immune response of chronically infected mice or, if neutralized by Ab anti-GM-CSF, it aggravated infection increasing parasitemia and host mortality in T. cruzi infected BALB/c mice [61]. In the aforementioned report, the role of GM-CSF was studied by correlating the outcome of infection with the titer of GM-CSF in plasma levels [61]. Even though it was not defined which cells were involved in GM-CSF secretion, it was speculated that lymphocytes could be in part contributing to the low but sustained amount of GM-CSF levels in infected mice. In our experiments, CD4+ and CD8+ T cells contributed almost equally with the secretion of this cytokine, independently of the stimulus, but it is not possible to discard that other cells as part of the PBMC pool also produced this cytokine.
While many questions remain regarding the pathogenesis of Chagas' disease, this study represents one of the most comprehensive about the cytokine profile in response to T. cruzi and two recombinant proteins, like P2β and CP0. The results show that a pool of PBMC in CCC patients has specificity for T cruzi proteins and that this specificity is revealed by a Th1-cytokine dominant milieu, combined with regulatory cytokines like IL-10 and IL-13. This observation reinforces the idea that a delicate cytokine equilibrium prevails during the chronic phase of the disease. Interestingly, another cytokines, namely GM-CSF, were found significantly increased in cardiac compared to non-infected individuals, tempting us to suggest that this cytokine may be further applied for studying antigen responses at different stages of the disease.
Finally, due to the limited number of patients infected with TcI (3/27) compared with TcVI (20/27), it was not possible to determine a correlation between the intensity of humoral and cellular immune response and the T. cruzi lineage detected by TSSA reactivity. Further studies in that sense would provide valuable information on the role and contribution of genetic variability of T. cruzi to the immune response developed in humans.
As a whole, our findings also demonstrate that not all parasite proteins provoke a strong T cell activation combined with a pattern of cytokines similar to those described to T. cruzi lysate or by infection with trypomastigote in CCC patients. In addition, as it was recently reported in the context of B cell-T cell recognition for other molecule-specificities [62], [63], it is possible to hypothesize that B cells specific for ribosomal P proteins could obtain help from T cells exhibiting different antigen reactivity. However, the interacting elements for T cell help recognition and activation may be the same for P2β and CP0, since a positive correlation was observed between the cytokines secreted by each of them (Figure S7). Currently, we are in the process of analyzing the immunoprevalence of recognition of these ribosomal P proteins and novel specificities involved in the immune response to T. cruzi infection in a large number of infected subjects at different stages of the disease.
|
10.1371/journal.pgen.1002824 | Genome-Wide Association Analysis in Asthma Subjects Identifies SPATS2L as a Novel Bronchodilator Response Gene | Bronchodilator response (BDR) is an important asthma phenotype that measures reversibility of airway obstruction by comparing lung function (i.e. FEV1) before and after the administration of a short-acting β2-agonist, the most common rescue medications used for the treatment of asthma. BDR also serves as a test of β2-agonist efficacy. BDR is a complex trait that is partly under genetic control. A genome-wide association study (GWAS) of BDR, quantified as percent change in baseline FEV1 after administration of a β2-agonist, was performed with 1,644 non-Hispanic white asthmatic subjects from six drug clinical trials: CAMP, LOCCS, LODO, a medication trial conducted by Sepracor, CARE, and ACRN. Data for 469,884 single-nucleotide polymorphisms (SNPs) were used to measure the association of SNPs with BDR using a linear regression model, while adjusting for age, sex, and height. Replication of primary P-values was attempted in 501 white subjects from SARP and 550 white subjects from DAG. Experimental evidence supporting the top gene was obtained via siRNA knockdown and Western blotting analyses. The lowest overall combined P-value was 9.7E-07 for SNP rs295137, near the SPATS2L gene. Among subjects in the primary analysis, those with rs295137 TT genotype had a median BDR of 16.0 (IQR = [6.2, 32.4]), while those with CC or TC genotypes had a median BDR of 10.9 (IQR = [5.0, 22.2]). SPATS2L mRNA knockdown resulted in increased β2-adrenergic receptor levels. Our results suggest that SPATS2L may be an important regulator of β2-adrenergic receptor down-regulation and that there is promise in gaining a better understanding of the biological mechanisms of differential response to β2-agonists through GWAS.
| Bronchodilator response (BDR) is an important asthma phenotype that measures reversibility of airway obstruction by comparing lung function before and after the administration of short-acting β2-agonists, common medications used for asthma treatment. We performed a genome-wide association study of BDR with 1,644 white asthmatic subjects from six drug clinical trials and attempted to replicate these findings in 1,051 white subjects from two independent cohorts. The most significant associated variant was near the SPATS2L gene. We knocked down SPATS2L mRNA in human airway smooth muscle cells and found that β2-adrenergic receptor levels increased, suggesting that SPATS2L may be a regulator of BDR. Our results highlight the promise of pursuing GWAS results that do not necessarily reach genome-wide significance and are an example of how results from pharmacogenetic GWAS can be studied functionally.
| Asthma is a chronic respiratory disease that affects over 20 million Americans and 300 million people worldwide [1], [2]. A hallmark characteristic of asthma is reversible airway obstruction, which is commonly measured via a bronchodilator response (BDR) test, in which the reduction of bronchoconstriction after administration of a short-acting reliever drug is quantified [3]. β2-agonists, the most common short-acting reliever drugs used during BDR tests and for asthma therapy, act in part by stimulating β2-adrenergic receptors (β2ARs) on airway smooth muscle cells to reduce bronchoconstriction via subsequent increases in cyclic adenosine monophosphate (cAMP) and protein kinase A (PKA) [3]. Although a comprehensive pathophysiologic understanding of BDR has not been obtained, it is a complex trait involving interactions among various tissues and cells, including inflammatory [4], airway epithelium [5], smooth muscle [6], and the autonomic nervous system [7]. In addition to being used for the diagnosis of asthma, BDR tests can be used to measure whether inhaled β2-agonists are effective in patients. Although short-acting β2-agonists are widely used clinically as asthma rescue medications, they are variably efficacious among patients [8]. Studying BDR may thus provide information regarding both the pathophysiology and pharmacogenetics of asthma.
The search for genetic variants that modify asthma susceptibility has resulted in the most recent multi-center asthma genome-wide association studies (GWAS) providing strong statistical evidence for the association of many genes, including the IKZF3-ZPBP2-GSDMB-ORMDL3 locus, HLA-DQ, IL1RL1, IL18RL1, IL33, TSLP, SLC22A5, SMAD3, and RORA, with asthma [9], [10]. Functional experiments to identify the role that these genes play in asthma pathophysiology are hindered by the complexity of the asthma phenotype. Familial aggregation [11] and genetic association studies [12] have provided suggestive evidence for a genetic contribution to interindividual differences in BDR. Candidate genes reported to be associated with BDR include β2-adrenergic receptor (ADRB2) [13], [14], adenylyl cyclase type 9 (ADCY9) [15], corticotrophin-releasing hormone receptor 2 (CRHR2) [16], and arginase 1 (ARG1) [17], [18]. While BDR is a complex phenotype, functional studies of BDR candidate genes are simpler than those for a general asthma phenotype because this pharmacogenetic phenotype can be readily simulated in vitro via stimulation of cells with β2-agonists.
In this study, we performed a GWAS of BDR in 1,644 non-Hispanic white asthmatics and found that the strongest evidence of association with BDR was at variants near the Homo sapiens spermatogenesis associated, serine-rich 2-like (SPATS2L) gene. We attempted to replicate the primary findings in two independent populations and investigated the function of SPATS2L via mRNA knockdown experiments and found evidence to support its involvement in BDR.
Figure 1 is an overview of our study design. Characteristics of the subjects used in the primary GWAS are provided in Table 1. We utilized 1,644 non-Hispanic white subjects from six clinical trials to measure the association of SNPs to BDR. After QC filters, 469,884 SNPs genotyped in CAMP/LOCCS/LODO/Sepracor and either genotyped or imputed in CARE and ACRN were used to test for the association of SNPs to BDR. We utilized genotyped SNPs for CAMP/LOCCS/LODO/Sepracor because these cohorts, who were all genotyped using Illumina platforms, had the largest sample size. Due to the poor overlap of Illumina and Affymetrix platform SNPs, we utilized HapMap Phase 2 imputed SNPs for CARE and ACRN, so that the maximal number of SNPs in all cohorts could be analyzed. The quantile-quantile (QQ) and Manhattan plots revealed that the distribution of association P-values was similar to that expected for a null distribution and that no P-values met genome-wide statistically significant levels (Figures S1 and S2). To expand the primary association results further, all SNPs available in the June 2010 release of the 1000 Genome Project (1000GP) data were imputed using MaCH in each of the three primary groups of genotype data and overall BDR GWAS results were re-computed. Among SNPs contained in the primary GWAS, imputed and genotyped P-values were similar, particularly for those with low P-values (Figure S3). Some imputed regions had P-values lower than those of the primary GWAS, but the results in most of these regions were not supported by primary GWAS data (Figure S2). Thus, we proceeded to attempt to validate the primary GWAS findings based on the combined genotyped CAMP/LOCCS/LODO/Sepracor SNP results and HapMap Phase 2 imputed CARE and ACRN SNP results. The top five primary GWAS SNPs with P-value<1E-05 are in Table 2. Further details on these regions and all primary GWAS SNPs with P-value<1E-04 are in Tables S1 and S2. Further details on all 1000GP imputed GWAS SNPs with P-value<1E-05 are in Tables S3
We attempted to replicate in SARP all of the SNPs with primary GWAS P-values<1E-04 (Table S5). Three had nominally significant P-values (i.e. <0.05), and two of these associations supported the top 5 primary GWAS associations (Table 3). The lowest combined P-value for all primary GWAS plus SARP data was 7.7E-07 for rs295137. The region of BDR association spanning this SNP was in the 5′UTR region of SPATS2L, a gene of unknown in vivo function and paralog of SPATS2 (Figure 2). The effect of the rs295137 genotype on BDR is shown in Figure 3, and a plot of the residuals of the linear regression fit of BDR while adjusting for age, sex, and height is shown in Figure S4. We sought further evidence of association for the two SPATS2L SNPs with lowest P-values in our primary GWAS in a second independent population: DAG. There was no evidence for association in this cohort (rs295137 P-value = 0.21; rs295114 P-value = 0.21), and combined P-values for these two SNPs across all cohorts were 9.7E-07 and 1.6E-06 (Table 3). To investigate whether our top combined association represented a biologically significant finding, we sought experimental evidence that SPATS2L was involved in bronchodilator response.
We found one public gene expression array experiment (GSE13168) that would help to address the question of whether SPATS2L is differentially expressed in response to changes in the BDR pathway. We compared the levels of expression of two SPATS2L and one SPATS2 probes in human airway smooth muscle (HASM) cells that stably expressed a PKA inhibitor vs. a GFP control at baseline and when stimulated with the pro-asthmatic cytokines interleukin-1β (IL1β), epidermal growth factor (EGF), or both. A trend of differential expression was observed for the SPATS2 and one SPATS2L probes, but not a second SPATS2L probe (Figure S5). None of the comparisons met a Benjamini-Hochberg adjusted significance threshold, but nominally significant P-values were obtained for the SPATS2 probe under all conditions and for one of the SPATS2L probes under the condition of EGF and IL1β stimulation (Table S6). According to the Gene Enrichment Profiler, the two SPATS2L probes are highly expressed in lung, and all three probes are highly expressed in smooth muscle, especially the SPATS2 one. Overall, there are strikingly different tissue-specific expression patterns for each probe (Figure S6).
We further investigated the involvement of SPATS2L in the β2-adrenergic response pathway by knocking down SPATS2L mRNA using two different small interfering RNAs (siRNA) and measuring subsequent changes in β2AR protein levels. The knockdown efficiency of the siRNAs was >80% reduction of SPATS2L mRNA as measured by qRT-PCR, and the corresponding increases in β2AR (normalized against the control β-actin protein) levels were 1.88- (SD 0.41) and 1.86- (SD 0.30) fold for the two SPATS2L siRNAs (Figure 4).
The association of SNPs with BDR at SNPs in/near genes (i.e. ADRB2, ADCY9, CRHR2, ARG1) previously reported as being associated with BDR was measured in our primary GWAS imputed data (Figure S7). Nominally significant (P-value<0.05) SNPs were found in ADCY9, CRHR2, and ARG1, but not in ADRB2 (Tables S7 and S8). The SNPs with lowest P-values within 50 kb of these genes were: rs2531988 for ADCY9 (3.2E-03), rs12533248 for CRHR2 (0.029), and rs6929820 for ARG1 (0.012).
In recent years, many GWAS that have successfully identified risk-modifying loci for a wide range of complex diseases have been published, but progress toward understanding how the loci and genes identified are functionally related to diseases has been slow [19]. The relationship of genes and gene variants to pharmacogenetic traits is often easier to test functionally than that for complex diseases because pharmacogenetic traits are more amenable to in vitro testing. However, compared to GWAS of complex diseases, GWAS of pharmacogenetic traits have been challenged by the relatively small size of drug clinical trials, which has caused many studies to be underpowered for obtaining genome-wide significant associations [20]. Nonetheless, successful pharmacogenetic GWAS have led to the identification of loci involved in modulating response to inhaled corticosteroids among asthma patients [21], warfarin dose [22], and lipid-lowering response to statins [23].
One of the difficulties specific to BDR GWAS is the complexity of the BDR phenotype. Regardless of how it is quantified, BDR is highly variable among asthma patients due to the time-dependent variation in baseline FEV1 and the influence of external environmental factors [24]. BDR can be quantified in various ways, with slightly varying resulting classification of patients as responders or non-responders. For our study, we selected the definition most widely utilized in clinical and human asthma research settings: percent change in baseline FEV1 following administration of a standard dose of short-acting inhaled β2-agonist [25]. We have attempted to control for the known relationship between baseline lung function and BDR [24], [26] by (1) selecting a definition of BDR that standardizes the change in FEV1 by dividing by baseline FEV1 and (2) by using age, sex, and height, which together account for a large portion of the variability in baseline lung function, as covariates in our statistical models. Because BDR tests are routinely performed during asthma clinical trials to use as inclusion criteria and to monitor outcomes among patients, we were able to utilize subjects from several diverse asthma clinical trials that were not specifically designed to study the pharmacogenetics of BDR. Most of these trials included a wash-out period that reduces modification of BDR due to concomitant medication administration, but LOCCS and some CARE and ACRN subjects were administered BDR tests at a time when they were not necessarily off of medications (Table 1). Subjects without a placebo washout, and especially those who were on ICS (e.g. LOCCS subjects), may be expected to have less BDR than those on placebo. The relationship of the magnitude of BDR to the various gene loci could therefore be blunted and show a less significant relationship than would be expected if all studies had incorporated a placebo washout.
In addition to variable washout periods, the cohorts had other significant differences in their design. Two trials consisted of children with asthma (i.e. CAMP, CARE), while the others consisted mostly of adults. The gender composition varied from 25% to 62% male. We attempted to control for age, gender and height, all of which are known to influence BDR, by including them as covariates in the association analysis. The mean and range of BDR also varied among trials. Of most significance, because the Sepracor trial used BDR greater than 15% as a criterion for inclusion, its subjects had markedly greater BDR than those of other trials. We attempted to control for this difference and any other trial specific differences among the cohorts that were pooled in the primary analysis by including trial as a covariate in the association model. There were additional differences among trials that were not taken into account. For example, SARP and Sepracor were composed of subjects with more severe asthma than those of other cohorts. Some ACRN, CARE, and SARP subjects were administered a different amount of albuterol during their BDR tests than those of CAMP, Sepracor, LOCCS, and LODO. DAG subjects were administered a different beta-agonist (i.e. Salbutamol) at a different concentration than that used with subjects of all other trials. DAG and SARP subjects were not participants of clinical trials, so there was greater heterogeneity of subjects within those cohorts. Despite the heterogeneity among trials, we utilized as many subjects as possible in an attempt to increase our statistical power to detect associations of SNPs with BDR. We reasoned that any associations detected despite the heterogeneity of the trial populations would be those most likely to generalize to all asthma patients. Another expected consequence of the trial heterogeneity is that our association results do not replicate in all cohorts. While having the largest number of subjects provides the greatest statistical power to detect statistically significant associations that are most generalizable across the clinical trials, we may be missing associations that are specific to the individual trials. For example, the subjects within clinical trials representing different ranges of asthma severity, age, and baseline characteristics may have genetic associations that are unique to subjects with their specific trial characteristics. The small sample size of each individual clinical trial makes detection of trial-specific associations more challenging. Despite the cohort heterogeneity, our meta-analysis identified a strong association that suggests a novel gene is involved in BDR.
Our top association was at SNP rs295137, with a combined P-value across all cohorts of 9.7E-07. This P-value does not meet conventional genome-wide significance thresholds (e.g. Bonferroni corrected minimally significant P-value would be 0.05/469,884 = 1.1E-07), but performing searches through public data sources and the fact that other pharmacogenetic GWAS have discovered biologically important results without genome-wide significant associations led us to pursue our top association further. The region of association surrounding rs295137 is in the 5′UTR of SPATS2L (Figure 2). This gene maps to chromosome 2 at 2q33.1, covering 176.78 kb from 201170592 to 201347368 (NCBI 37, August 2010). According to data gathered via the AceView [27] tool, SPATS2L is a complex locus that may have at least 30 spliced variants, its in vivo function is unknown, and it is a highly expressed gene in many tissues, with the greatest number of GenBank accessions belonging to lung. In gene-trap experiments in myoblasts, SPATS2L (a.k.a. SGNP) was found to be involved in ribosomal biogenesis and translational control in response to oxidative stress [28].
The availability of one public expression array experiment that utilized HASM cells expressing a PKA inhibitor (PKI) to modify the β2-adrenergic pathway allowed us to perform a preliminary search for evidence that SPATS2L may be involved in BDR. We found that a probe for SPATS2, the paralog of SPATS2L, was significantly differentially expressed in PKI vs. control cells at baseline and when stimulated with pro-inflammatory cytokines (EGF, IL1β, or both). One SPAST2L probe followed this trend but had a nominally significant P-value only under the condition of stimulation with both EGF and IL1β, while the other SPATS2L probe did not exhibit any changes. As illustrated in Figure S6, the tissue-specific expression patterns of the three probes varied widely. While all were expressed in smooth muscle, the SPATS2 probe's relative expression in this tissue was markedly greater than that of the SPATS2L probes. Taken together, the expression patterns are consistent with tissue and isoform dependent changes in SPATS2L gene products. While the public dataset SPATS2L results were inconclusive based on the differences among probes, they suggested that SPATS2L expression may change when PKA is inhibited in HASM cells.
Knockdown of SPATS2L in HASM cells resulted in significantly increased β2AR protein levels, suggesting that SPATS2L may affect BDR by directly modulating β2AR protein expression. In HASM, β2-agonists exert their effects exclusively via the β2AR [6]. The relaxation of HASM occurs after the binding of β2-agonists to β2ARs via increased levels of cAMP followed by PKA activation. PKA activation leads to changes in gene transcription via activation of cAMP response element binding protein (CREB). Because β2ARs are the gateway to the effects of β2-agonists in HASM cells, modulations, such as SPATS2L inactivation, that increase the levels of β2ARs in HASM cells may lead to both greater relaxation in response to β2-agonists in the short term and greater differences in gene transcription in the longer term. Further study is needed to elucidate the precise mechanism by which SPATS2L regulates β2AR and consequently modifies BDR. Among our primary GWAS subjects, those whose SPATS2L SNP rs295137 has the TT genotype have greater BDR than those with CT or TT genotypes (median BDR 16.0 vs. 10.9). In one of the simplest scenarios, it is possible that the increased BDR among subjects with the TT genotype results from this genotype playing a direct role in decreasing transcription of SPATS2L, which in turn results in increased β2AR levels. Further work is required to understand how specific SNP associations in/near SPATS2L affect SPATS2L function and/or expression and how such effects impact β2AR signaling and BDR. Because the observed influence of our most strongly associated SNP genotype on BDR is relatively small (Figure 3), our current data do not support the development of any personalized therapeutics based solely on variants in/near SPATS2L.
In addition to studying top primary GWAS SNPs, we attempted to replicate findings from previous BDR candidate gene association studies. Specifically, we measured association between BDR and ADRB2 [13], [14], ADCY9 [15], CRHR2 [16], and ARG1 [17] variants. Notably, these previous findings are not entirely independent of those from the current GWAS: CAMP was a primary population utilized to identify associations in ADRB2, ADCY9, CRHR2, and ARG1 in previous reports; LODO and Sepracor were replication populations in the CRHR2, and ARG1 reports; and LOCCS was a replication population in the ARG1 report. At a nominal significance level, we replicated gene-level associations for all of the candidate genes other than ADRB2. This gene, which encodes the β2AR, is the most studied gene related to BDR and SNPs and haplotypes in this gene have been related to decreased pulmonary function [29], response to β2-agonist treatment [30], an increased frequency of asthma exacerbations [31], and BDR [13], [14]. Initial reports of ADRB2 associations were very promising and suggested that variants of this candidate gene would be reliable markers of BDR pharmacogenetics. However, a meta-analysis of 21 studies of ADRB2 polymorphisms found that most of the positive associations identified in individual studies cannot be compared to findings in other studies due to different study designs, phenotypes considered and selective reporting, making the number of variants with true replications very small and questioning the utility of ADRB2 polymorphisms for generalizable pharmacogenetic tests [32]. Our inability to find associations with ADRB2 variants is consistent with the view that BDR genetics are complex: no individual SNPs or genes are responsible for a large proportion of BDR variability observed among all asthmatics. Our results suggest that genes other than the previously identified candidate genes are more strongly associated with BDR and that functional studies of these regions may yield important insights into BDR biology despite not having strong effects or generalizing to all populations.
In summary, a BDR GWAS among asthma patients from eight cohorts found that the most strongly associated SNP, rs295137, had a combined P-value of 9.7E-07. This association led us to SPATS2L, a gene of unknown in vivo function that we showed may be involved in BDR via the down-regulation of β2AR levels. Our results support the notion that there is promise in pursuing GWAS results that do not necessarily reach genome-wide significance and are an example of the way in which results from pharmacogenetic GWAS can be studied functionally.
Each study was approved by the Institutional Review Board of the corresponding institution, which ensured that all procedures followed were in accordance with the ethical standards of the responsible committee on human experimentation. Informed consent was obtained for all study participants.
The primary group of subjects consisted of 1,644 non-Hispanic white asthmatics from the following drug clinical trials: Childhood Asthma Management Program (CAMP) [33], Leukotriene Modifier or Corticosteroid Salmeterol study (LOCCS) [34], Effectiveness of Low Dose Theophylline as an Add-on Treatment in Asthma trial (LODO) [35], a medication trial conducted by Sepracor, Inc. [36], [37], and subsets of clinical trials within the Childhood Asthma Research and Education (CARE) network [38], and the Asthma Clinical Research Network (ACRN) [39] participating in the NHLBI SNP Health Association Resource (SHARe) Asthma Resource project (SHARP). Some basic characteristics of these cohorts are in Table 1 and further details are provided in Text S1. BDR tests were performed according to American Thoracic Society criteria with Albuterol as the β2-agonist [25], unless otherwise noted. Baseline BDR measures were utilized, and BDR was quantified as the percent change in FEV1 in response to administration of a β2-agonist [i.e. (post-BD FEV1 – pre-BD FEV1)/pre-BD FEV1].
Genome-wide genotyping for CAMP subjects (n = 546) was performed on the HumanHap550 Genotyping BeadChip or Infinium HD Human610-Quad BeadChip by Illumina, Inc (San Diego, CA) at the Channing Laboratory. LOCCS (n = 135), LODO (n = 114), and Sepracor (n = 401) subjects were genotyped at the Riken Center for Genomic Medicine using the Infinium HD Human610-Quad BeadChip. CARE (n = 207) and ACRN (n = 241) subjects were genotyped on Affymetrix 6.0 genotyping chip by Affymetrix, Inc. (Santa Clara, CA). Data from those subjects genotyped using Illumina technologies was combined into a primary dataset with 469,884 overlapping SNPs having missingness <1%, passing HWE (P-value threshold of 1E-03), and having minor allele frequency (MAF)>0.05. EIGENSTRAT was used to identify 23 outliers (not included in counts above) based on being outside of six standard deviations of the top four principal components during five iterations [40]. The genomic inflation factor (λGC) of the remaining 1,196 subjects was 1.002, demonstrating minimal population stratification. CARE and ACRN dataset quality control (QC) also included the removal of four related subjects (i.e. CARE siblings), SNPs with MAF<0.05, missingness >5%, or not passing HWE based on a threshold of 1E-03. The λGC for CARE and ACRN genotype data were 1.02 and 0.98, demonstrating minimal population stratification among subjects within each group. Comprehensive genotyping and QC measures are provided in Text S1.
Due to the poor overlap among SNPs genotyped on the Illumina and Affymetrix platforms, imputation of all SNPs available in HapMap Phase 2 Release 22 CEU data using the Markov Chain Haplotyping software (MaCH) [41] was performed for ACRN and CARE genotyped data. The primary GWAS consisted in the set of 469,884 SNPs that were successfully genotyped in those cohorts using Illumina arrays (i.e., CAMP/Sepracor/LOCCS/LODO) and imputed with HapMap Phase 2 data in those cohorts genotyped with Affymetrix arrays (i.e., ACRN and CARE) with a ratio of empirically observed dosage variance to the expected (binomial) dosage variance greater than 0.3, indicating good quality of imputation.
To expand the association results, imputation of all SNPs available in the June 2010 release of the 1000 Genome Project (1000GP) data using MaCH was performed for each of the three primary groups of genotype data. An overlapping set of 4,571,615 imputed SNPs had a MAF>0.05 and ratio of empirically observed dosage variance to the expected (binomial) dosage variance greater than 0.5, indicating good quality of imputation.
The association of SNPs with BDR was measured with a linear regression model as implemented in PLINK [42] in the three sets of data: 1) CAMP/Sepracor/LOCCS/LODO, 2) ACRN, 3) CARE. Association of imputed SNPs was carried out using dosage data. Covariates for the CAMP/Sepracor/LOCCS/LODO group included age, gender, height, and study. Covariates for the CARE and ACRN groups included age, gender, and height. To get the primary GWAS results, CARE and ACRN P-values were combined with those of the CAMP/Sepracor/LOCCS/LODO group by using Liptak's combined probability method [43] where hypothesis tests in CARE and ACRN had one-sided alternatives, based on the direction of the association in CAMP/Sepracor/LOCCS/LODO, so that SNPs with association tests in opposite directions would not produce inappropriately small P-values. The overall λGC was 1.002 in the primary set of GWAS results and 1.000 in the 1000GP imputed data GWAS.
Plots of association results near specific genes were created using LocusZoom with the hg18/1000 Genomes June 2010 CEU GenomeBuild/LD Population [44].
The publicly available Gene Expression Omnibus (GEO) dataset, GSE13168, corresponding to an experiment in which human airway smooth muscle (HASM) cell cultures were generated from four donor trachea to test for the effects of glucocorticoids and PKA inhibition on the HASM transcriptome using the Affymetrix Human Genome U133A platform was used [48]. We tested for the involvement of our top primary GWAS gene in the β2-adrenergic pathway by comparing the differential expression of genes in cells stably expressing a PKA inhibitor (PKI) vs. control at baseline and in the presence of pro-inflammatory cytokines interleukin-1β (IL1β), epidermal growth factor (EGF), or both. The expression array contained two SPATS2L probes (i.e., 215617_at, 222154_s_at) and one SPATS2 probe (i.e., 218324_s_at). The probe for the paralog of SPATS2L was included to account for the possibility of non-specific binding of SPATS2L mRNA to the SPATS2 probe. Analyses were conducted in R [46]. Pre-processing of raw signal intensities was performed with RMA [49] and differential expression was quantified using the limma package [50]. Tissue-specific expression of these probes was assessed using 557 microarrays from 126 human normal primary tissues in the Gene Enrichment Profiler [51].
Primary HASM cells were isolated from aborted lung transplant donors with no chronic illness. The tissue was obtained from the National Disease Resource Interchange (NDRI) and their use approved by the University of Pennsylvania IRB. HASM cell cultivation and characterization were described previously [52], [53]. Passages 4 to 7 HASM cells maintained in Ham's F12 medium supplemented with 10% FBS were used in all experiments. 2×105 HASM cells were grown overnight and then transfected with 50 nM siRNA by using DharmaFECT 1 reagent (Thermo Scientific, Lafayette, CO, USA). About 72 h post transfection, cells were washed with PBS and lysed with NP-40 lysing buffer (50 mM Tris-HCl pH7.5, 150 mM NaCl, 0.5% Nonidet P-40) containing protease inhibitor cocktail (Roche Diagnostics Corporation, Indianapolis, IN, USA). Protein samples were denatured 10 min at 50°C, separated on NuPAGE 4–12% Bis-Tris gels (Invitrogen, Grand Island, NY, USA) and transferred to PVDF membranes (Bio-Rad Laboratories, Hercules, CA, USA). Immunoblot signals were developed using SuperSignal West Pico (Pierce Protein Research Products, Thermo Fisher Scientific, Rockford, IL, USA) and quantified by ImageJ software. Non-targeting control siRNA, SPATS2L-specific siRNA 1 (sense 5′- guc agu cca uug auu guc u(dT)(dT)-3′, antisense 5′- aga caa uca aug gac uga c(dT)(dT) -3′) and SPATS2L-specific siRNA 2 (sense 5′-caa ccu gug uug uag cag u(dT)(dT)-3′, antisense 5′- acu gcu aca aca cag guu g(dT)(dT) -3′) were obtained from Sigma-Aldrich (Mission siRNA; St. Louis, MO, USA). Antibodies for β2AR (H20) and β-actin were from Santa Cruz Biotechnology, Inc. (Santa Cruz, CA, USA). Experiments were performed in triplicate.
|
10.1371/journal.pgen.1003742 | Type I-E CRISPR-Cas Systems Discriminate Target from Non-Target DNA through Base Pairing-Independent PAM Recognition | Discriminating self and non-self is a universal requirement of immune systems. Adaptive immune systems in prokaryotes are centered around repetitive loci called CRISPRs (clustered regularly interspaced short palindromic repeat), into which invader DNA fragments are incorporated. CRISPR transcripts are processed into small RNAs that guide CRISPR-associated (Cas) proteins to invading nucleic acids by complementary base pairing. However, to avoid autoimmunity it is essential that these RNA-guides exclusively target invading DNA and not complementary DNA sequences (i.e., self-sequences) located in the host's own CRISPR locus. Previous work on the Type III-A CRISPR system from Staphylococcus epidermidis has demonstrated that a portion of the CRISPR RNA-guide sequence is involved in self versus non-self discrimination. This self-avoidance mechanism relies on sensing base pairing between the RNA-guide and sequences flanking the target DNA. To determine if the RNA-guide participates in self versus non-self discrimination in the Type I-E system from Escherichia coli we altered base pairing potential between the RNA-guide and the flanks of DNA targets. Here we demonstrate that Type I-E systems discriminate self from non-self through a base pairing-independent mechanism that strictly relies on the recognition of four unchangeable PAM sequences. In addition, this work reveals that the first base pair between the guide RNA and the PAM nucleotide immediately flanking the target sequence can be disrupted without affecting the interference phenotype. Remarkably, this indicates that base pairing at this position is not involved in foreign DNA recognition. Results in this paper reveal that the Type I-E mechanism of avoiding self sequences and preventing autoimmunity is fundamentally different from that employed by Type III-A systems. We propose the exclusive targeting of PAM-flanked sequences to be termed a target versus non-target discrimination mechanism.
| CRISPR loci and their associated genes form a diverse set of adaptive immune systems that are widespread among prokaryotes. In these systems, the CRISPR-associated genes (cas) encode for proteins that capture fragments of invading DNA and integrate these sequences between repeat sequences of the host's CRISPR locus. This information is used upon re-infection to degrade invader genomes. Storing invader sequences in host genomes necessitates a mechanism to differentiate between invader sequences on invader genomes and invader sequences on the host genome. CRISPR-Cas of Staphylococcus epidermidis (Type III-A system) is inhibited when invader sequences are flanked by repeat sequences, and this prevents targeting of the CRISPR locus on the host genome. Here we demonstrate that Escherichia coli CRISPR-Cas (Type I-E system) is not inhibited by repeat sequences. Instead, this system is specifically activated by the presence of bona fide Protospacer Adjacent Motifs (PAMs) in the target. PAMs are conserved sequences adjoining invader sequences on the invader genome, and these sequences are never adjacent to invader sequences within host CRISPR loci. PAM recognition is not affected by base pairing potential of the target with the crRNA. As such, the Type I-E system lacks the ability to specifically recognize self DNA.
| There are several prokaryotic defense systems that confer innate immunity against invading mobile genetic elements, such as receptor masking, blocking DNA injection, restriction/modification (R-M) and abortive infection (reviewed in [1]–[3]). In addition, half of the bacteria, and most of the archaea, contain CRISPR-Cas (Clustered Regularly Interspaced Short Palindromic Repeats/CRISPR-associated) defense systems, unique in being the only adaptive line of prokaryotic defense (reviewed in [4]–[7]). CRISPR-Cas systems provide adaptive immunity to the host by incorporating invader DNA sequences into chromosomal CRISPR loci [8]–[11]. The 30–40 nt invader-derived DNA sequences are separated by host-derived similarly-sized repeat sequences. Adjacent to a CRISPR locus, a set of cas genes can often be found that encode the protein machinery essential for CRISPR-immunity. The cas genes occur in characteristic combinations that serve as a classification criterion of CRISPR-Cas systems into three major types [12]. In Type I and Type III systems the long precursor CRISPR RNA (pre-crRNA) is processed by CRISPR specific endoribonucleases into small CRISPR RNAs (crRNAs) that contain a repeat sequence flaked by portions of the adjacent CRISPR repeat sequence [13]–[18]. In some CRISPR-Cas subtypes the crRNA undergoes further processing at the 3′ end [19], [20]. In Type II CRISPR-Cas systems the pre-crRNA is processed by RNase III [21]. The processed crRNA molecules then remain bound to one or more Cas proteins to guide recognition and cleavage of complementary nucleic acid sequences [22]–[27].
With the exception of Type III-B CRISPR-Cas systems, which cleave RNA [23], [24],[28], all other characterized CRISPR-Cas systems appear to target DNA [27], [29]–[32] and hence require a mechanism to avoid aberrant cleavage of genomic DNA, i.e. a mechanism to discriminate the genomic “self” DNA of a CRISPR cassette from the invader “non-self” DNA. The absence of such discrimination leads to a suicidal autoimmune response [33]–[35]. In R-M systems this problem is solved by modification of the genomic DNA and cleavage of unmodified invader DNA only (reviewed in [3]). For CRISPR-Cas systems on the other hand, the mechanism(s) of self versus non-self discrimination is only partially understood.
For the Type III-A system of Staphylococcus epidermidis autoimmunity is prevented through a mechanism that relies on sensing base pairing between the 5′-handle (the repeat-derived sequence at the 5′-end of the crRNA) and the corresponding portion of CRISPR repeat [36]. The Type III-A CRISPR-Cas system consists of nine cas genes (cas1, cas2, cas10, csm2, csm3, csm4, csm5, csm6, cas6) and a CRISPR with type-8 repeats [37]. After a primary processing step of the pre-crRNA, the resulting crRNAs are further matured through ruler-based cleavage from the 3′ end, yielding 43 and 37 nt crRNA species [20]. These mature crRNA species guide one or more Cas proteins (possibly a Csm-complex) to target DNA [32], presumably through base pairing between the crRNA spacer sequence and the complementary protospacer sequence. However, CRISPR-interference is inhibited when, in addition to base pairing over the spacer sequence, the 5′-handle also base pairs with the protospacer-flanking sequence of the target DNA [36]. In this manner, self-targeting of the CRISPR locus is avoided by default, since self-targeting inevitably leads to full base pairing of the 5′-handle of the crRNA with the CRISPR repeat sequence from which it is transcribed. In particular, the presence or absence of base pairing at three positions downstream of the protospacer (positions −2, −3, and −4 relative to the 3′-end of the protospacer) is decisive in discriminating self from non-self [36]. The molecular details of how base pairing at positions downstream of the protospacer are sensed, and whether it involves Cas proteins, is currently unknown.
Intriguingly, Type I systems contain di- or tri-nucleotide conserved motifs (protospacer adjacent motifs (PAM)) downstream of protospacers opposite of the crRNA 5′-handle [38]–[40] (Figure 1A and 2A). In the Type I-E CRISPR-Cas system, PAM sequences are recognized by ribonucleoprotein complex Cascade during target DNA binding [29], [41]. The Type I-E system of Escherichia coli K12 consists of 8 cas genes (cas3, cse1, cse2, cas7, cas5, cas6e, cas1, cas2) and two CRISPR loci with type-2 repeats [37]. The ribonucleoprotein complex Cascade is composed of a 61 nt crRNA, and five different Cas proteins in an uneven stoichiometry: Cse11Cse22Cas76Cas51Cas6e1 [22]. Cascade efficiently binds target DNA through an R-loop formed between the 32 nt spacer sequence of the crRNA and the protospacer sequence [22] (Figure 1A), with a binding affinity that is strongly dependent on the presence of one of the four functional PAM sequences [29], [41]. Whereas R-loop formation by Cascade involves the entire protospacer sequence [22], it is unknown whether the PAM nucleotides can participate in base pairing with the crRNA and, if so, how this influences CRISPR interference. Due to the fact that the last nucleotide from the repeat is derived from the PAM sequence during spacer acquisition [8], [11], [42], this nucleotide in the crRNA invariably has the potential to base pair with the −1 position of the PAM, and therefore might be involved in R-loop formation [8]. In contrast, the −2 and −3 positions of the PAM lack base pairing potential with the 5′-handle of the crRNA (Figure 2A). The 5′-handles of other Type I systems and 3′-handles of Type II also display limited base pairing potential with their cognate PAMs (Table S1), in principle allowing for a differential base pairing mechanism that defines self versus non-self. For Type I-F CRISPR-Cas systems, potential base pairing between PAM sequences and the 5′-handle of the crRNA was recently shown to affect CRISPR interference [43], suggesting that self versus non-self discrimination in this subtype may depend both on sensing PAM identity and on sensing differential base pairing with the crRNA repeat.
In Type I-E systems it has been shown that a loop structure (L1) of the Cse1 subunit of Cascade specifically interacts with the PAM sequence, a process that is thought to destabilize the double-stranded DNA of the target to allow for strand invasion during R-loop formation [44]. Since self DNA of the CRISPR locus does not contain PAM sequences, this mechanism would specifically direct Cascade to target DNA only. However, the observation that target DNA containing a PAM mutant triggers Cascade-dependent primed spacer acquisition in vivo suggests that PAM authentication may not be absolutely required for R-loop formation [11]. Indeed, negatively supercoiled DNA containing a protospacer with a mutant PAM can still be bound by Cascade, albeit with a lower affinity than the same target with wild-type PAM [29]. In line with this, it was suggested that during phage infection Cascade can overcome the absence of a bona fide PAM when Cascade expression levels are high and that the target flanking sequences could participate in this discrimination event [44]. This suggests that a differential base pairing mechanism may play a role in self versus non-self discrimination by Type I-E CRISPR-Cas systems. In agreement with this, it was suggested that complementarity between the crRNA repeat and the protospacer flanking sequence inhibits CRISPR-interference in the Type I-E system of Streptococcus thermophilus [45]. The mechanistic basis of such a differential base pairing mechanism could lie in a perturbation of Cse1-mediated PAM recognition by base pairing interactions between crRNA repeats and the PAM.
To study whether a differential base pairing mechanism plays a role in self versus non-self discrimination by the Type I-E system of E. coli K12, we have systematically mutated both the crRNA repeats and the protospacer-flanking sequences and determined the effects of these mutations and their combinations on CRISPR interference in vivo and target binding in vitro. The results of our analysis demonstrate that discrimination of self from non-self by Type I-E CRISPR-Cas systems occurs through a mechanism that is independent of base pairing between these sequences. Hence, the principal mechanism by which Type I-E systems discriminate self from non-self appears to be solely Cse1-mediated and as such is fundamentally different from the differential base pairing mechanism employed by Type III-A systems. While the mechanism employed by Type III-A is best described as being based on self-recognition (self versus non-self), the mechanism of Type I-E systems is instead based on target-recognition (target versus non-target). While Type III systems can differentiate between targets and non-targets in the absence of a PAM, Type I-E systems are fully PAM-dependent and discrimination cannot take place in the absence of a PAM.
Self versus non-self discrimination by the Type III-A CRISPR-Cas system of S. epidermidis has been shown to rely on a differential base pairing mechanism [36]. As a result CRISPR-interference is specifically inhibited when protospacer sequences are flanked by CRISPR repeat sequences. To test whether this mechanism also applies to the Type I-E CRISPR-Cas system of E. coli K12, CRISPR-interference was tested against targets containing protospacers flanked by CRISPR repeat sequences. For these analyses, we have cloned the previously described g8 protospacer, from phage M13 [41], into the pUC19 plasmid and systematically mutated sequences adjacent to the protospacer. E. coli cells expressing Cascade, a g8 crRNA and Cas3 are resistant against transformation by a plasmid in which the g8 protospacer is flanked by a CAT PAM (Fig. 1B, pWUR690, approximately 1000-fold lower efficiency of transformation than a control pUC19 plasmid). In contrast, these cells are susceptible to plasmid transformation by plasmid pWUR687 in which the g8 protospacer is flanked by CRISPR repeat sequences (Figure 1B). However, the plasmid resistant phenotype can be restored by introducing a CAT PAM in the CRISPR repeat sequence flanking the protospacer (pWUR688), which alters the base pairing potential only at the −2 and −3 positions (Figure 1B). Plasmid pWUR689, which has the potential to base pair with g8 crRNA at positions −1, −2 and −3 (protospacer adjacent sequence is CGG) escapes CRISPR-interference from wild-type g8 crRNA expressing E. coli (Figure 1B). The observation that protospacer adjacent sequences complementary to the crRNA at positions −1, −2, and −3 avoid Cascade targeting suggest that base pairing at these positions may play a role in self avoidance.
To investigate whether avoidance of targeting is due to decreased binding affinities of Cascade for protospacers with mutations at the −1, −2, and −3 positions, we performed Electrophoretic Mobility Shift Assays using purified g8 crRNA-loaded Cascade. While high affinity binding could be demonstrated to dsDNA containing the g8 protospacer flanked by the CAT PAM (Figure 1B and S1), protospacers flanked by either CRISPR repeat sequences or a repeat-derived CGG sequence were bound with low affinity (Figure 1B and S1). This indicates that target versus non-target discrimination occurs at the level of Cascade affinity for dsDNA target sequences. Furthermore, the data also indicate that “self” DNA recognition may occur, as observed in Type III-A systems, through sensing differential base pairing between protospacer adjacent sequences and the 5′ handle of the crRNA.
To investigate if base pairing between the three nucleotides from the 5′-handle of the crRNA and the PAM is involved in discriminating self from non-self DNA we systematically mutated the corresponding nucleotides in the 5′-handle (i.e., −1, −2, and −3), and analyzed how these mutations affect CRISPR-based immunity against DNA targets flanked by various PAM sequences. Previously [29], four PAM sequences (CAT, CTT, CCT and CTC), have been reported to confer immunity on wild-type g8 crRNA expressing E. coli against phage M13 infection in vivo, and to give rise to high affinity DNA binding by g8 crRNA-bound Cascade in vitro (Figure 2B and Figure S2A). The last nucleotide of the 5′-handle of the crRNA (the −1 position) invariably has the potential to base pair with the PAM [8], while the −2 and −3 positions lack such base pairing potential (Figure 2A). The resulting configuration is distinct from the fully base-paired configuration that would form if base pairing in this region were the basis of self versus non-self discrimination.
To analyze whether base pairing at position −1 is required for CRISPR interference, a mutant CRISPR was constructed, yielding a g8 crRNA that lacks base pairing potential with the PAM at this position. This CRISPR, denoted g8G-1T carries a G-to-T substitution at position −1, within the repeat sequence. SDS-PAGE analysis of purified Cascade complexes containing either mutant or WT crRNA shows that these complexes have the same apparent stoichiometry, thereby confirming the integrity of the complex (Figure S4A). In addition, isolation of crRNA from these protein complexes shows that crRNA biogenesis is unaffected by the introduced mutation (Figure S4B). Interestingly, despite the absence of base pairing at the −1 position, cells expressing the mutant crRNA maintain the ability to block infection by M13 phages containing each of the four functional PAM sequences (Figure 2C). Consistently, high affinity binding by g8G-1T crRNA-containing Cascade to targets containing the g8 protospacer and the functional PAM variants was observed (Figure 2C and Figure S2B). However, as previously observed for the WT g8-crRNA-Cascade complex [29], a mutation at the −2 position of the PAM (i.e., CGT) neither confers resistance in vivo (efficiency of plaquing (e.o.p.) = 1) nor gives rise to high affinity DNA binding in vitro (Figure 2C, and Figure S2B). This PAM mutant potentially yields an additional base pair with the −2 position of the 5′-handle, both in the WT g8-crRNA-Cascade and the g8G-1T mutant complex (Figure 2BC). Hence, it appears that a base pair at position −2 may be the signal that a protospacer is located in “self” DNA and therefore should not be targeted.
To specifically test the role of base pairing at position −2 in CRISPR-immunity, we designed a synthetic CRISPR locus containing a C to A substitution at the −2 position of a CRISPR locus containing spacer sequences that target the g8 protospacer from M13 phage. The g8C-2A CRISPR mutation results in a slight effect on Cascade assembly, as the bands corresponding to Cse1 and Cse2 have modestly lower and higher intensities on an SDS-PAGE, respectively, as compared to wild-type g8-crRNA-Cascade (Figure S4). However, g8C-2A CRISPR RNA processing is unaffected (Figure S4). Importantly, the g8C-2A crRNA-guided Cascade complex has a slightly reduced affinity (60±12 nM) for dsDNA targets that have a canonical CTT PAM sequence, which has the potential to base pair at the −2 position of the mutant crRNA (Figure 3A, white PAM). Despite the potential of the mutant Cascade complex to establish an additional base pair, a partially resistant phenotype (e.o.p.∼10−2) is observed against phages carrying the canonical PAM (Figure 3A), which is consistent with the in vitro DNA binding experiments (Figure 3A and Figure S3A). Targets containing non-canonical PAM sequences are bound with more reduced affinities by the g8C-2A crRNA-guide Cascade complex and are not subject to CRISPR-interference in vivo (Figure 3A). The partial resistant phenotype of the g8C-2A mutant that is observed in combination with the canonical PAM indicates that potential base pairing at both positions −1 and −2 does not serve as a trigger for a non-targeting response.
To probe the importance of base pairing at the −3 position, an additional CRISPR mutant was designed, denoted g8C-3G, which carries a C to G mutation at the −3 position of the CRISPR repeat. Again, complex formation and crRNA biogenesis were unaffected by the mutation (Figure S4). Although the potential for base pairing with most PAM sequences remains the same, a dramatic decrease in both resistance against M13 phage in vivo and DNA binding by g8C-3G-Cascade in vitro is observed (Figure 3B and Figure S3B).
The combined results obtained with the three CRISPR mutants indicate that the repeat sequence itself rather than its base pairing potential with the protospacer flanking sequence affects PAM recognition. In order to have a more complete and unbiased analysis of the effects of adding or removing base pairing potential at positions −1, −2 and −3, we constructed 26 different PAM sequences adjacent to the g8 protospacer in the M13 phage genome (Figure 4A, white text on black background). All phages were viable as judged by their ability to infect host bacteria lacking the M13-targeting CRISPR (data not shown). The phages were tested for their ability to infect cells expressing each of the 21 different g8 crRNAs with mutated repeat sequences at positions −1, −2 and −3. Northern blot analysis showed that processing of mutant g8 crRNAs was unaffected (data not shown). The results reveal that only a small subset of CRISPR repeat mutants confer full phage resistance, and only in conjunction with the four previously validated functional PAM sequences (Fig. 4). When resistance was observed, it was independent of crRNA-PAM base pairing patterns, but rather appeared to be constrained by a limited number of allowed nucleotides at the −1, −2 and −3 positions of the 5′-handle, and a fixed number of PAM sequences.
Many 5′-handle mutants show a lack of resistance despite the presence of a bona fide PAM in the target and irrespective of the base pairing pattern (Figure S5). Efficient CRISPR-interference requires the presence of a cytosine at the −2 position of the crRNA repeat (Figure 4). Substitution of this position to guanidine or uracil interferes with CRISPR-defense. When this position is mutated to an adenosine, a partially resistant phenotype is observed during phage infection in conjunction with the canonical PAM, which is bound with the highest affinity by Cascade in vitro. Presumably this high affinity binding can compensate for the negative effects on DNA binding caused by mutations at the −2 position of the 5′-handle, leading to a partially phage resistant phenotype. Furthermore, CRISPR-mediated phage resistance requires a cytosine at the −3 position. The most likely explanation for the fact that some repeat mutants are not tolerated is that the Cascade subunits involved in binding the 5′-handle exhibit a level of sequence specificity.
Although combinations of fully complementary 5′-handles and protospacer flanking sequences do not lead to phage resistance in vivo, this appears to be base pairing independent (Figure S5), as restoring the wild-type base pairing pattern by altering protospacer flanking sequences fails to rescue the phage-sensitive phenotype. For example, the g8C-3A, C-2T CRISPR fails to provide resistance either against M13 phage with a fully complementary CAT PAM (Figure 4B) or against a CTC PAM mutant phage, which is complementary at the −1 position only (Figure 4C). A similar result is obtained when g8C-3A, C-2A CRISPR expressing cells are infected with CTT or CTC PAM phages (Figure 4D and E), indicating that the repeat sequence itself is affecting CRISPR-interference in these instances. Altogether, these data exclude the possibility that the Type I-E system makes use of a differential base pairing mechanism to inhibit self-targeting. The finding that the specificity of PAM recognition is unaffected by its potential to base pair with the 5′-handle is consistent with Cse1 being the only factor involved in PAM recognition [44].
To rule out the possibility that the specificity of PAM recognition by g8-Cascade variants depends on the expression levels of CRISPR-Cas components, the same analyses were performed with an engineered M13 targeting E. coli strain with cas genes fused to inducible promoters [12]. When repeat mutations were introduced into the genomic CRISPR cassette in this strain, identical results were obtained (Figure S6), showing that the data described here are expression level independent.
Previous studies on the S. thermophilus Type II-A CRISPR1/Cas system have revealed differences in PAM specificity and effectivity in either plasmid or phage interference assays [30],[45]. To test whether the Type I-E CRISPR/Cas system also displays assay-dependent differences in PAM utilization, we generated plasmids carrying the g8 protospacer (pG8) flanked by any of the 26 PAM mutants tested in the phage assays. Transformation of the pG8 variants into E. coli cells expressing Cascade, a g8 crRNA and Cas3 show that the four PAMs (CAT, CTT, CCT, and CTC) that provide interference during phage infection also affect plasmid transformation (resulting in a more than 1000-fold decrease in efficiency of transformation (e.o.t.)). Apart from these four PAMs, a non-consensus TTT PAM also yields a full resistance phenotype (Figure S7; >1000-fold decrease in e.o.t.), as has been observed before [8], while M13 phage carrying this non-consensus TTT PAM sequence escape interference (Figure 4A). In addition, ten non-consensus PAMs give rise to a partial resistance phenotype (Figure S7; e.o.t. <10−1 for CCA, CAA, GAT, CTG, and AGA PAMs; e.o.t. <10−2 for CTA, GTT, TAT, ATT and TTC PAMs), which is in line with previously reported partial resistance in S. thermophilus against transformation with a target plasmids carrying non-consensus PAMs [30]. The data show that PAM authentication during CRISPR-based protection is more promiscuous during plasmid transformation than during phage infection.
CRISPR-Cas systems are the only prokaryotic adaptive immune systems described to date. Although initially thought of as a single system, we now know that these systems are structurally and mechanistically diverse. Here we have investigated whether a differential base pairing mechanism to discriminate self from non-self, as described for the Type III-A system of S. epidermidis, also applies to the Type I-E CRISPR-Cas system of E. coli K12. By systematically mutating the crRNA repeat sequence and the PAM positions, we demonstrate that this Type I-E system does not utilize the potential for base pairing between the 5′-handle and the protospacer flanking sequences to avoid self targeting.
The −1 position of crRNA has recently been shown to be invader-derived and hence invariably has the potential to base pair with cognate DNA, both in E. coli [8], [11], [42] and in S. thermophilus [45], [46]. This discovery suggested that base pairing at the −1 position would be critical for target recognition by Cascade, in the same way that nucleotides in the seed region (nucleotides +1 to +5, +7 and +8) are essential for target recognition [41]. However, our results clearly show that base pairing at position −1 is not essential for CRISPR-interference. It has recently been suggested that the −1 position of the CRISPR repeat could be considered part of the spacer [42]. However, this does not seem appropriate since this nucleotide does not appear to be involved in base pairing with the invading target sequence. The absence of a base pairing requirement for the −1 position might suggest that this position is not available for base pairing due to structural constraints.
The −2 position of the crRNA repeat requires the presence of a cytosine for efficient CRISPR-interference (Figure 4). When this position is mutated to an adenosine, a partially resistant phenotype is observed during phage infection in conjunction with the canonical PAM. Substitution of the −2 position to a guanidine or uracil renders the CRISPR-interference pathway non-functional. Interestingly, mutation of the −2 position to adenosine causes an apparent structural alteration of the Cascade complex. While most subunits are present in the same apparent stoichiometry in the mutant g8C-2A-Cascade as in the wild-type complex, the Cse1 subunit is underrepresented. This might suggest that Cse1 interacts with the −2 position of the repeat and that interaction with this base is important for efficient incorporation of Cse1 into the complex. Like the −2 position, the −3 position requires a cytosine for CRISPR-mediated phage resistance to be manifested. However, complex formation is unaffected in g8C-3G-Cascade (Figure S4A).
The −3, −2 and −1 positions are among the most conserved bases of type 2 repeats [37]. Although the current resolution of the Cascade structure does not allow us to confidently pinpoint the location of the −2 and −3 bases of the 5′-handle of the crRNA, these bases appear to be part of a 5′ hook-like structure that is primarily cradled by the last subunit of the Cas7 hexamer (i.e., Cas76) [47]. The arch of the crRNA may position the 5′ terminal nucleotides within bonding distance to residues in loop-1 of Cse1, which is consistent with the assembly defects reported for L1 mutations [44]. However, the resolution of the current Cascade structure and absence of density for L1 in the X-ray crystal structures of Cse1 prevent confident assignment of these interactions. Higher-resolution structures of the Cascade will be critical for a precise understanding how the crRNA and the Cas proteins are arranged in this complex.
In some CRISPR systems PAM sequences play an important role during different stages of CRISPR defense. In the Type I-E system of E. coli, PAM sequences are recognized by Cas1 and/or Cas2 during the selection of pre-spacers for integration into the CRISPR [9]. PAM motifs allow the CRISPR adaptation machinery to correctly orient newly acquired spacers into the CRISPR array [38], [48]–[50]. Interestingly, in Type I-E systems, the PAM selectivity of the CRISPR-adaptation machinery has co-evolved with that of the CRISPR-interference machinery, as the preference for the CTT PAM is observed both during Cas1/Cas2-dependent spacer integration [9] and during target DNA binding by Cascade [29]. In contrast, the E. coli I-F integration machinery appears to select for a PAM that overlaps but differs from the motif that yields optimal interference levels [43]. In this E. coli I-F subtype the PAM was found to be a GG motif at the −1 and −2 positions relative to the protospacer, while an overlapping, but different, motif (GG at the −2 and −3 positions) provided optimal interference levels [43]. The presence of a G at position −2 was both required and sufficient for interference. The I-F subtype of Pectobacterium atrosepticum on the other hand requires a GG motif immediately flanking the protospacer for interference, and mutagenesis of the G at position −1 to a T (which potentially base pairs with the repeat) gives rise to an escape phenotype [35]. Recently, a new nomenclature has been proposed that takes into account the differences in motif selectivity during spacer integration and CRISPR-interference [51].
PAMs have been shown to be important for CRISPR interference in various Type I and Type II CRISPR-Cas subtypes (e.g. Type I-A systems in S. solfataricus [40], Type I-B in Haloferax volcanii [39], Type I-E in E. coli [29], Type I-F in P. aeruginosa [52], E. coli [43] and P. atrosepticum [35], as well as in Type II-A and II-B systems of Streptococcus pyogenes and S. thermophilus [27], [30], [50], [53], [54]). Recently published x-ray crystal structures of the Cse1 subunit of Cascade [44], [55] have provided detailed insights into the molecular mechanism of Cascade-mediated recognition of the PAM. The well-conserved L1 loop of Cse1 was shown to directly interact with the PAM sequence and to enhance target DNA affinity in the presence of a bona fide PAM [44]. As such, the Cse1 subunit plays a crucial role in PAM authentication in Type I-E systems [44]. Our data indicate that PAM authentication occurs without the formation of base pairs between the 5′ handle of the crRNA and the PAM.
While Cascade-like complexes appear to be common components of Type I systems, the PAM-authenticating protein, Cse1, is unique to Type I-E systems. This could mean that other Cascade-like complexes, such as the aCascade (IA-Cascade) [25], IC-Cascade [17] the as yet unidentified ID-Cascade, and the Csy-complex (IF-Cascade) [26] may have their own specialized PAM-sensing proteins. It has been hypothesized that the large subunits of Type I systems (Cas8a1 and Cas8a2 (Type I-A), Cas8b (Type I-B), Cas8c (Type I-C), Cas10d (Type I-D), Cse1 (Type I-E), Csy1 (Type I-F)) are homologous to Cas10 proteins associated with the Type III systems [56], but these predictions await experimental verification. If these predictions are correct they may suggest that PAM recognition is carried out by the large subunit of other CRISPR-Cas subtypes.
Under native-like expression levels, the change in affinity of Cascade for a target resulting from the presence or absence of a PAM sequence appears to be sufficient to serve as a robust mechanism to discriminate non-self target sequences (i.e. protospacers flanked by a PAM) from non-target sequences (i.e. protospacers without PAM) in vivo [44]. Given the absence of PAM sequences in the CRISPR array, self DNA automatically falls into the non-target category and is not subject to interference. For Type III systems, on the other hand, no PAMs have yet been found, suggesting that these systems lack PAMs [23], [36]. For Type III-A systems it has been shown that differentiation between self DNA and non-self DNA relies on sensing differential complementarity between the 5′-handle of the crRNA and the protospacer-flanking sequence (Figure 5A) [36]. This discrimination mechanism is based on specific recognition of self DNA, and is therefore best described by the term self versus non-self discrimination (Figure 5A). Here we demonstrate that self-avoidance by the Type I-E system does not rely on potential base pairing between crRNA repeats and protospacer flanking sequence. Therefore, Cascade lacks the ability to specifically recognize self and relies on specific target DNA recognition through PAM authentication. We argue that PAM authentication is a “target versus non-target” discrimination mechanism (Figure 5B), which is fundamentally different from the “self versus non-self” discrimination mechanism employed by Type III-A systems. Either mechanism is sufficient to avoid targeting of the CRISPR locus on the host genome. In target versus non-target discrimination, self sequences within the CRISPR locus (i.e. spacers) automatically belong to the non-target class, since PAM sequences are absent in the CRISPR repeat. Likewise, in self versus non-self discriminating systems target sequences fall in the non-self class. It appears likely that PAM-sensing CRISPR-Cas systems all make use of target versus non-target discrimination. Unlike Type III systems, discrimination between targets and non-targets by Type I-E systems cannot take place in the absence of a PAM.
Both discrimination mechanisms, however, are not mutually exclusive. The Type I-F system of E. coli LF82 has been speculated to utilize both target versus non-target discrimination and self versus non-self discrimination [43], although this hypothesis awaits experimental verification by testing the effect of crRNA repeat mutagenesis on CRISPR interference. By having both mechanisms in place an additional level of security against self-targeting of the host genome could be warranted. The requirement for a more stringent protection against self-targeting could be related to the constitutive gene expression of the Type I-F in E. coli LF82 [43], whereas the expression of the Type I-E system of E. coli K12 is repressed under laboratory growth conditions [57], [58], [59].
The distinct mechanisms of self versus non-self discrimination of Type III-A and target versus non-target recognition of Type I-E have implications for the route that invaders can take to escape CRISPR-interference. While both systems can be evaded by making point mutations in the protospacer [41], [60], only the Type I-E system can be evaded by mutations outside the protospacer, specifically in the region containing the PAM. In contrast, escape from Type III-A interference through mutations outside the protospacer seems rather unlikely, as it would typically require three mutations to establish base pairing between the 5′ handle and the protospacer flank [36].
E. coli BL21 (DE3) strains were used for Cascade purification. Novablue (DE3) cells supplemented with CRISPR plasmid and plasmids expressing cas genes and engineered K12 strains with cas genes fused to inducible promoters were used for phage sensitivity tests and transformation assays. A description of the plasmids and the strains used in this study can be found in the Supplementary Information (Table S1).
Wildtype M13-Cascade was expressed in E. coli BL21 (DE3) and purified as described before [29], from pWUR408, pWUR514 and pWUR615 (Table S1). g8G-1T-Cascade, g8C-2A-Cascade, g8C-3G-Cascade, were expressed from pWUR408, pWUR514 and either pWUR680, pWUR682, or pWUR684, respectively (Table S1). pWUR680, pWUR682, and pWUR684 were generated by subcloning a synthetic CRISPR (Table S3 and Table S4, Geneart) into pACYC using EcoNI and Acc65I restriction sites. Although BL21 (DE3) contains genomic CRISPR loci, previous analyses by Mass Spectrometry have demonstrated that these expression and purification conditions yield homogeneous Cascade complexes loaded with crRNA species from the overexpression plasmids, and not from the chromosme [22].
Purified Cascade was separated on a 12% SDS-PAGE as described before [22], and stained using Coomassie Blue overnight, followed by destaining in Millipore water. Nucleic acids were isolated from purified Cascade complexes using an extraction with phenol∶chloroform∶isoamylalcohol (25∶24∶1) equilibrated at pH 8.0 (Fluka) and separated on a 6M urea 15% acrylamide gel, as described in [22], followed by staining with SybR safe (Invitrogen) in a 1∶10000 dilution in TAE for 30 minutes. Electrophoretic Mobility Shift Assays were performed as in [29], using the PAGE-purified oligonucleotides listed in Table S2, which were annealed and 5′-labeled with 32P γ-ATP (PerkinElmer) using T4 polynucleotide kinase (Fermentas). Determining the Kd of the Cascade target DNA interaction was performed as described in [41]. Briefly, the signals of unbound and bound probe were quantified using Quantity One software (Bio-Rad). The fraction of bound probe was plotted against the total Cascade concentration, and the data fitted by nonlinear regression analysis to the following equation: Fraction bound probe = [Cascade]total/(Kd+[Cascade]total).
Mutations of PAM sequence preceding the g8 protospacer were introduced into the M13 phage genome by QuickChange Site-Directed Mutagenesis Kit (Stratagene) as described previously ([41]).
Repeat mutant library was generated by QuikChange Site-Directed Mutagenesis Kit (Stratagene) according to manufacturer's protocol. The g8 CRISPR cassette plasmid targeting the M13 phage gene 8 (pWUR477-g8, described in [41]) was used as template. Mutations were introduced at positions −3, −2, or −1 of the repeat preceding the g8 spacer.
Cells sensitivity to wildtype and mutant M13 phages was determined by a spot test method as described [41] or using standard plaquing assay. Efficiency of plaquing was calculated as a ratio of the plaque number formed on a lawn of tested cells to the number of plaques on sensitive (non-targeting) cell lawn.
K12 strains with cas genes fused to inducible promoters and g8 spacer in CRISPR were transformed with 10 ng of plasmid DNA by electroporation. Transformation efficiency was determined as colony forming units for transformants of targeting strain BW40119 (Table S1) per µg DNA. Plasmids containing the g8 protospacer and PAM mutants were ordered synthetically at Geneart, Germany.
|
10.1371/journal.pcbi.1003957 | Live Imaging-Based Model Selection Reveals Periodic Regulation of the Stochastic G1/S Phase Transition in Vertebrate Axial Development | In multicellular organism development, a stochastic cellular response is observed, even when a population of cells is exposed to the same environmental conditions. Retrieving the spatiotemporal regulatory mode hidden in the heterogeneous cellular behavior is a challenging task. The G1/S transition observed in cell cycle progression is a highly stochastic process. By taking advantage of a fluorescence cell cycle indicator, Fucci technology, we aimed to unveil a hidden regulatory mode of cell cycle progression in developing zebrafish. Fluorescence live imaging of Cecyil, a zebrafish line genetically expressing Fucci, demonstrated that newly formed notochordal cells from the posterior tip of the embryonic mesoderm exhibited the red (G1) fluorescence signal in the developing notochord. Prior to their initial vacuolation, these cells showed a fluorescence color switch from red to green, indicating G1/S transitions. This G1/S transition did not occur in a synchronous manner, but rather exhibited a stochastic process, since a mixed population of red and green cells was always inserted between newly formed red (G1) notochordal cells and vacuolating green cells. We termed this mixed population of notochordal cells, the G1/S transition window. We first performed quantitative analyses of live imaging data and a numerical estimation of the probability of the G1/S transition, which demonstrated the existence of a posteriorly traveling regulatory wave of the G1/S transition window. To obtain a better understanding of this regulatory mode, we constructed a mathematical model and performed a model selection by comparing the results obtained from the models with those from the experimental data. Our analyses demonstrated that the stochastic G1/S transition window in the notochord travels posteriorly in a periodic fashion, with doubled the periodicity of the neighboring paraxial mesoderm segmentation. This approach may have implications for the characterization of the pathophysiological tissue growth mode.
| Cell cycle progression is considered to involve a cellular time-counting machinery for proper morphogenesis and patterning of tissues. Therefore, it is important to understand the regulatory mode of cell cycle progression during physiological and pathological tissue growth, which will benefit tissue engineering therapy and tumor diagnosis. Since cell cycle progression is a highly variable process, it is a challenging task to retrieve the spatiotemporal regulatory mode hidden in heterogeneous cellular behavior. To overcome this issue, we took advantage of live imaging analyses with a fluorescence cell cycle indicator, Fucci technology, and mathematical modeling of developing zebrafish fish embryo as a model system of growing tissue. Our result demonstrated that the developmental growth of embryonic axis progressed in a rhythmic fashion. The presented analyses will benefit the characterization of the regulatory mode of tissue growth, in both physiological and pathological development, such as that involving tumor formation, thus may contribute to cancer diagnosis.
| The development of multicellular organisms is a highly coordinated process, in which cell proliferation and sequential changes in cellular identities are spatiotemporally regulated, through which patterned tissues and organs are ultimately formed. As a system to ensure the precision and reproducibility of the developmental process, the concept of “intrinsic time” has been postulated [1], [2]. Cell cycle progression has long been considered to involve a cellular time-counting machinery for proper morphogenesis and patterning of tissues. This notion is fundamentally supported by observations of increased mitotic activity in populations of cells that transiently appear during the developmental process. The presence of temporal waves of mitotic activity in the developing limb mesenchyme is reported to correlate with a segmented skeletal pattern, thus possibly accommodating the positioning of bones and joints in limbs [3]. In addition, a clustered mitotic activity observed in the endoderm are proposed to be responsible for morphogenetic folding to form the digestive tract [4]. Furthermore, periodic surges of mitotic activity in the paraxial mesoderm have been repeatedly observed in concert with reiterate somite formation in embryonic tissue [5]–[7]. Since somites principally endow a segmented architecture to the axial skeleton and its associated muscles and neurons, timed machineries of somite formation provide a fundamental system for body plan and anatomical structure [8]–[10]. The periodic formation of somites is regulated by the segmentation clock, which exhibits an oscillatory expression of signaling molecules related to Notch, Wnt and Fgf [9], [11]–[14]. Though it has been proposed that cell cycle progression regulates periodic somite formation, as described above, the current findings argue against the idea that the cell cycle clock is involved as an oscillator of the segmentation clock [15]–[17].
It is well established that cell cycle progression is a highly variable process. A phenomenological description of the stochastic cell cycle progression has been reported [18]–[22], and mathematical models that account for the variable transition timing in cell cycle progression have been proposed [19], [23]–[30]. Based on in vitro observations of mammalian cell cultures, a conceptual framework of “the restriction point” of the G1/S transition has been proposed [22]. The restriction point divides the G1 phase into the G1-postmitosis phase (G1-pm) and the G1-pre S phase (G1-ps), in which cells are able to proliferate dependent and independent of mitotic stimuli, respectively. G1-pm is highly constant in time length (approximately three hours), while the duration of G1-ps varies considerably. The restriction point is currently understood to extend the timing of phosphorylation of Rb proteins by Cyclin D1, thus releasing E2F in order to initiate S phase entry. Mathematical modeling analyses have also suggested a bistable mechanism to control the restriction point in the mammalian G1/S transition [31]–[35]. Yao et al., experimentally demonstrated bistable E2F activation that directly correlated with the ability of a cell to traverse the restriction point by temporally monitoring the E2F transcriptional activity with stimuli of various magnitudes, thus validating that the RB-E2F pathway involving multiple positive feedback loops can generate bistability; namely, by forming the Rb-E2F bistable switch [36]. This Rb-E2F bistable switch is further extended to work even when subjected to noise, which supported the proposed models to account for the temporal variability in the G1-S transition [37]. In this stochastic model, both cellular intrinsic and extrinsic noise can be taken into account. The intrinsic noise results from the stochastic nature of biochemical interactions due to the stochastic gene expression levels in each single cell, while the extrinsic noise arises from heterogeneous properties of a cell, such as the cell's size, shape, cell cycle phase and cell division [38]–[43].
Generally, during in vivo tissue development, biochemical phenomena are intrinsically associated with stochasticity, in which fluctuations in cellular responses are observed in populations of cells exposed to the same environmental conditions [38], [44]–[47]. In multicellular organism development, heterogeneous cellular behavior, such as the G1/S transition, possibly blinds regulatory events. Therefore, it is relevant to develop novel approaches to numerically estimate noise strength or the probability of a stochastic cellular response in multicellular organisms, which would unveil a regulatory mode of tissue growth associated with morphogenesis and patterning. Our research group previously reported the findings of whole-view imaging of cell cycle progression during embryonic development obtained using fluorescence live imaging of Cecyil, a zebrafish line that genetically carries a fluorescence cell cycle indicator, Fucci [48]. The Fucci system applies ubiquitin-based cell cycle phase-specific oscillators that are fused to distinct colored fluorescence proteins [49], which demarcates cell nuclei in the G1 and S/G2/M phases in red and green, respectively (Figures 1A, B). By taking advantage of the Fucci system, this study aimed to establish a method to clarify the regulatory mode of cell cycle progression veiled in the stochastic behavior of developing tissue. Quantitative analyses of our live imaging data combined with mathematical modeling demonstrated that the stochastic G1/S transition window in the notochord travels posteriorly in a periodic fashion with doubled periodicity of neighboring paraxial mesoderm segmentation. The results of our analyses are expected to benefit the characterization of the regulatory mode of tissue growth, in both physiological and pathological development, such as that involving tumor formation.
Mid-sagittal optical sections of fixed Cecyil embryos counterstained with phalloidin (Figure 1C) and live observation of fluorescence and DIC (differential interference contrast) images (Figure 1 D to K, Movie S1) demonstrated that morphological differentiation of notochordal cells is associated with cell cycle progression. Newly formed notochordal cells in the posterior region expressed intense red fluorescence signals in their nuclei, being arranged to form two arrays of red cells. These cells arose from their precursor pools of chord neural hinges that comprised a mixture populations of dark-red cells and green cells (indicated by the sky blue arrowheads in Figure 1 D to K), suggesting that G1 entry is tightly related to initial notochord formation. The second color switches of the Fucci signals (G1 red to S/G2/M green, indicated by the red and green arrowheads in Figure 1 D to K) were observed prior to the initiation of cell vacuolization (Figure 1C); therefore, this initial G1/S transition is possibly required for notochord differentiation. In most of the observations, we observed a mixed population of green and red cells between the anterior green cells and posterior red cells, possibly suggesting the stochastic behavior of the G1/S transition in each cell. Therefore, we termed this mixed population of green and red cells the G1/S transition window. As embryonic development progressed, the G1/S transition window moved posteriorly by following the direction of body axis elongation (Figure 1 F to K). Comparing the DIC and fluorescence images, we mapped the position of a newly formed somite (indicated by the blue arrowheads in Figure 1 F to K) and that of the G1/S transition window (red and green arrowheads). This observation indicated that the initial G1/S transition is always observed at a position posterior to newly formed somites (Figure 1D). These findings suggest that the regulatory mode of the G1/S transition in notochordal cells is related to progressive modes of embryonic body axis elongation and somite segmentation.
In order to quantitatively analyze the mode of G1/S transition in notochordal cells, we digitalized two cellular states, the G1 and S/G2/M phase, using image processing. We initially tracked each single notochordal cell by measuring the fluorescence intensity of the red and green signals in live imaging data. Figure 1J shows the representative tracking data of a single cell indicated by the white arrows in Figure 1 G to H. The time point of the G1/S transition was determined as the crossing point of the red and green lines of the fluorescence intensity, which occurred at 115 minutes in Figure 1J. For details, please see the Methods section.
Based on this method, we next plotted the binary valued cellular states of the upper (dorsal) and lower (ventral) columns of cells in a single dimension of the anterior to posterior position (Figure 2A, Movie S2). This plotting was then applied to succeeding time lapse images. Therefore, the second dimension was introduced in order to demonstrate the temporal dynamics of the G1/S transition (Figure 2B). Notably, the spatial positions of the G1/S transition were observed to overtly progress posteriorly, although the transition in each cell did not occur in the exact order of their position, but rather in a stochastic fashion. When the total number of green and red cells was scored at each time point, it was obvious that the stochastic G1/S transition progressed posteriorly as time progressed (Figure 2C), although the modes of the upper and lower cells were not entirely synchronized.
In order to further investigate the mode of this stochastic G1/S transition in more detail, we focused on spatiotemporal changes in a mixed population of red and green cells that demarcated the stochastic G1/S transition window. We scored the positions of the anterior-most red cell (ARC) and posterior-most green cell (PGC), which demarcate the anterior and posterior positions of the stochastic G1/S transition window, respectively, at each time point of observation (Figure 3A, Figure S1). Scoring of the position of the ARC and PGC showed a pattern of step-wise progression in which the position of the ARC (indicated by the red lines in Figure 3B and Figures S2A and S3A) followed and occasionally caught up to that of the PGC (green lines in Figure 3B and Figures S2A and S3A). Therefore, the stochastic G1/S transition window appeared to repeatedly widen and shorten its width (see the space enclosed by the green and red lines in the graphs shown in Figure 3B and Figures S2A and S3A).
We applied another method of scoring in order to further analyze the spatiotemporal dynamics of the G1/S transition window, in which the total number of green cells in seven cells anterior to the PGC, including the PGC itself, (total: 8 cells) at each time point was scored (Figure S1). Interestingly, this scoring system demonstrated that the number of green cells in this defined area synchronously oscillated in the upper and lower columns over a period of 60 minutes or more (Figure 3C, Figure S2B and S3B).
We next investigated whether the state of the cell cycle phase affects the G1/S transition of neighboring cells. The total number of green cell pairs (the upper and lower cells at the same position are both labeled in green) in seven pairs anterior to the PGC and a pair of cells including the PGC was scored (Figure 3D, red ‘+’ markers). Consistent with the data presented in Figure 3C, the temporal changes in the number of green cell pairs also exhibited an oscillatory behavior. We next compared these data with the random or biased distribution patterns generated by in silico simulation (see the Methods section). The expectation values of the number of green cell pairs for the random and biased cases are shown in Figure 3D in black and blue lines, respectively. The experimental data shown by the red ‘+’ markers exhibited a better fit to the data obtained with the random simulations (Figure 3D, Figure S2C and S3C), suggesting that the G1/S transition in notochordal cells progresses independently with the cell cycle phase of neighboring cells.
In order to unveil a possible regulatory mode of the G1/S transition in notochordal cells, we employed a mathematical modeling approach. The model was constructed according to a Markov process describing the stochastic transition from the G1 phase to the S phase. At each time step t, a cell undergoes the G1/S transition with a probability αΔt over a short time interval Δt (Figure 4A). The spatial position of the notochordal cells is represented as a one-dimensional lattice in the order of anterior to posterior (left to right, respectively, as shown in Figure 4B). Each cell in the lattice is identified by an index i (i = 1,…,n). Based on our biological observations described in Figure 1, we assumed the presence of a regulatory wave that conveys a signaling cue to promote the G1/S transition in notochordal cells in a stochastic fashion. This signaling cue on a single dimensional axis is controlled by the signal transmitting function f(i,t) of t and i (Figure 4C). By introducing a tuning parameter z in f(i,t), we were able to examine how the probability affects the distinct mode of the regulatory cue (for details, see the Methods section). In other words, z corresponds to the step size, which is defined by the width of a given number cells. For example, in the model of z = 8, the step size is the total width of eight cells. Based on our observations described above, we assumed modes of continuously traveling waves (continuous mode) and periodically traveling waves (periodic modes) with different periodicity.
In the deterministic cell cycle progression, i.e. αΔt = 1, without fluctuation, the continuous wave model (z = 1) exhibited a linear mode of progression in the G1/S transition (Figures S4A and S4C). Meanwhile, in the periodic models (a model of z = 8 with 30 minutes periodicity is shown in Figure 4C as an example), the G1/S transition progressed in a step-wise mode (Figures S4B and S4C). Under these conditions, the spatial positions of the ARC and PGC always coincided with each other (Figure S4D).
In order to introduce stochasticity into the model, i.e. αΔt<1, we estimated the range of probability, αΔt. Based on our analysis shown in Figure 3D, the timing of the G1/S transition is not tightly affected by the cell cycle phase of neighboring cells. Therefore, the time differences of S phase entry between the upper and lower cells allowed us to estimate the probability range. A set of time differences in S phase entry was computed for all tracked cells in three individual specimens, and the probability distribution was subsequently obtained by summing all of the data (Figure 4D, histogram). For curve fitting, we theoretically derived the probability distribution function for the time difference based on two independent Bernoulli processes (see the Methods section). By fitting the parameter with experimental data, we estimated that αΔt = 0.1101 (Figure 4D, red line), suggesting that the plausible parameter space is located around αΔt = 0.1, which specifically means that approximately 10% of cells would exhibit S phase entry in Δt = (2.5 minutes) of a single time lap of the imaging transition (see the Methods section). Therefore, we hereafter set the parameter as αΔt = 0.1 and examined the system behavior around this value.
With this parameter setting of G1/S transition probability, the stochastic modes of the G1/S progression were seen in both the continuous and periodic models (Figures 5A and 5B, Movie S3 and S4). Quantification of the temporal changes in the total number of green cells under the continuous (z = 1) and periodic (z = 8) modes did not show any obvious distinction due to the stochasticity (Figure 5C), as observed in the experimental data analyses shown in Figure 2C.
In order to dissociate the continuous mode from periodic modes regulating the stochastic G1/S transition, we next quantified the positions of the PGC and ARC in these in silico simulations. Consequently, the temporal changes in the position of the PGC under the periodic mode appeared to retain its step-wise progression pattern compared to that observed under the continuous mode (Figure 5D, green lines and Figure S5A and S5B), while the position of the ARC exhibited a largely fluctuating pattern under both of these regulatory waves (Figure 5D, red lines, Figure S5C and S5D). Therefore, we decided to perform a detailed analysis of the progressive mode of the PGC position by measuring the time interval from the appearance time of a PGC to that of the next PGC. An example of the calculation procedure of the time interval of the PGC is illustrated in Figure S6. A set of the interval is first computed from a simulation result, after which the probability distribution is obtained. The probability distribution of three different simulations, the continuous model (z = 1) (Figure 6A), periodic model (z = 8) (Figure 6B) and two-fold periodic model (z = 16) (Figure 6C) are subsequently calculated. In the continuous model, the time interval is concentrated on a smaller range (<20 min), and the frequency decreases as the time interval increases. The probability extends to a larger range around 20–40 minutes in the periodic model and extends further and persists up to 60 minutes in the two-fold periodic model. On the other hand, the same analyses of the ARC exhibited no apparent differences between the distributions obtained from three distinct simulation models (Figure S7). Therefore, these analyses indicated that measuring the time interval of PGC reflects the regulatory mode of the stochastic G1/S transition and that the probability distribution function can be used to provide discrimination information in order to analyze the regulatory mode of stochastic changes.
We next obtained the probability distribution from the experimental time-lapse imaging data. A set of the time interval was computed from each time-lapse image sequence, and the probability distribution was then calculated by summing all three sets of time-lapse imaging data (Figure 6D). It is obvious that the probability distribution obtained from the experimental data resembles that obtained from the simulated data. In order to quantitatively compare datasets from experiments and simulations, we used the Kullback-Leibler (KL) distance, which measures the difference between two probability distribution functions p(x) and q(x) (see the Methods section). In order to compare the KL distance between the experiment and simulation, we prepared the probability distribution of the experimental data as p(x) and that of the simulated data as q(x), where x denotes the time interval of PGC. We computed the KL distance repeatedly, 300 times for four different simulation models, including the continuous (Figure S8A), periodic (Figure S8B), two-fold periodic (Figure S8C) and three-fold periodic (Figure S8D) models, and then calculated the mean and standard deviation of the KL distance (Figure 6E). The mean KL distance values of the periodic models were smaller than those of the continuous model, indicating that the hypothesis of the periodic G1/S transition is likely. Among these periodic models, the two-fold periodic model had the lowest KL distance value. In order to validate our simulations, we carried out sensitivity analyses varying the transition probability αΔt, with a range from 0.07 to 0.12 (Figure 6F) and the number of cells constituting a single somite width modified from eight to seven cells (Figure S9); the obtained results did not affect the conclusions. These findings suggest that the G1/S transition progresses under the influence of a periodic regulatory mode and that its periodicity is the most likely to be the two-fold period of somite segmentation.
In order to provide credible evidence for the periodicity of the regulatory mode of the G1/S transition in notochordal cells, we enumerated the number of green cells in these three distinct periodic simulation models, as we did in the experimental observation shown in Figure 3C. In each simulation model, we performed two independent simulations, which were assumed to involve the upper and lower columns of notochordal cells. According to the continuous model simulation results, the upper and lower columns of cells did not appear to show any obvious synchronized behavior (Figure 7A). In the simulations with the periodic and the two-fold periodic models (Figure 7B and C), two columns of cells exhibited synchronized oscillation. However, in terms of the stability of oscillation, the two-fold periodicity model was more robust than that of the periodic model. Furthermore, the two-fold periodic model appeared to be most suitable for the experimental model shown in Figure 3C.
As an inventive application of Fucci technology [48]–[50], the integrative approach we applied in this study was comprised of three parts, including quantitative data acquisition from live imaging, model establishment and model selection (Figure 8A), and revealed that the progressive mode of the G1/S transition window travels in a periodic fashion in newly formed notochordal cells during embryonic axis elongation, as schematically demonstrated in Figure 8b. Once notochordal cells are formed from their precursor pool located at the tip of an embryo, the cells in the G1 phase shown in red are arranged in a line. In the next step, a group of G1 cells enters the S phase. This G1/S transition is temporally stochastic; therefore, a mixed population of green and red cells is established between a group of posterior red cells and a group of anterior green cells, which we described as the G1/S transition window (Figure 8B i). In this window, an increasing number of G1 cells enters the S phase shown in green; therefore, this region is eventually filled with green cells (Figure 8B ii). In the subsequent cycle, the next G1/S transition window shifts a width of 16 cells (two somite widths) posteriorly, thus establishing a new window (Figure 8B iii). The G1 cells in this window enter the S phase in a stochastic manner, finally filling the window (Figure 8B iv).
Based on the live imaging data of the developing Cecyil embryos, numerical conversion of the G1/S transition in notochordal cells was carried out using concomitant tracking of the cell fates and changes in the fluorescence intensity of the red and green Fucci signals. As previously described, live imaging of Cecyil embryos provides an almost entire view of morphogenetic cellular movement and cell cycle transition [48]. With careful observation and analyses of morphogenetic cellular events in Cecyil embryos, we noticed that the developing notochord is an appropriate organ for our purposes in establishing a modeling approach. Once notochordal cells are formed, they are arrayed as two columns of cells in the anterior to posterior axis in mid-sagittal optical sections. Their cellular movement is much less active than that of other mesodermal tissues; therefore, the precision of cell tracking and enumeration of spatiotemporal mapping of the G1/S transition are highly ensured.
In order to establish a reliable mathematical model to extract the regulatory mode of the G1/S transition in developing multicellular tissue, we employed a transition probability model in which cellular states are simply represented as binary states, the G1 or S phase, and temporal variability in S phase entry originates from a random transition with a fixed probability [18]–[22]. The phenomenological model is advantageous, with very few parameters to be measured or inferred in an appropriate manner, although it lacks details in terms of molecular mechanistic insight. The molecular mechanisms underlying the random G1/S transition are currently understood to involve stochasticity in Rb-E2F dynamics, as described by employing stochastic differential equations [37]. This model demonstrates that the stochastic dynamics in the Rb-E2F model can be quantitatively mapped into phenomenological models. Therefore, we decided to apply a simple transition probability model in order to analyze the dynamics of the G1/S transition based on live imaging data of developing multicellular tissue.
We next investigated the independency or dependency of the timing of S phase entry in cells in the G1/S transition window using a linkage analysis. Comparisons of in silico simulations and scoring from the imaging data indicated that the G1/S transition in notochordal cells progresses independently from the cell cycle state of neighboring cells. This analysis interestingly suggested that stochasticity in the cellular response is highly intrinsic, even in populations of developing multicellular tissue affected by regulatory surges. This finding firstly confirms the phenomenological concept of the stochastic G1/S transition in in vivo tissue development and helped us to estimate the probability of timing of S phase entry, as discussed in the following paragraph.
We then constructed a mathematical model to describe the regulatory mode of the stochastic G1/S transition in notochordal cells. Seeing that the spatial position of the G1/S transition window is always located between newly formed somites and the posterior tip of the embryo, the presence of a sort of progressive developmental wave that triggers S phase entry was postulated and mathematically represented by introducing the signal transmitting function. Our model here has only one free parameter, that is, the probability of timing of the G1/S transition. This parameter was successfully estimated by fitting the theoretically derived probability distribution function with data obtained from live imaging. The notochord structure is profitable in that we simply plotted the time differences in S phase entry between the upper and lower cells in order to estimate the probability range based on the assumption that the developmental wave travels from anterior to posterior and equally to dorsal and ventral cells. Our strategy of model construction with probability estimation is applicable to cell populations in which the timing of the G1/S transition is tightly affected by the cell cycle state of neighboring cells. However, in that case, an alternative method for parameter estimation is required.
With the probability estimation, we conducted model selection by comparing the results obtained from the models with those obtained from the experimental data. We tested several hypothesized models assuming the presence of continuous wave or periodic waves with different periodicity, as reflected in the choice of the tuning parameter z of the signal transmitting function. Due to the stochasticity, the boundary of the G1 and S phase cells fluctuated, presenting some difficulty in dissociating the continuous and periodic regulatory mode. For instance, merely counting the temporal changes in the total number of green and red cells did not provide any information about the spatial progressive mode of the G1/S transition window. It is probable that the noise strength numerically estimated in this work completely veiled the spatial information related to a possible regulatory mode in this analysis. Therefore, we focused on the PGC (posterior-most green cell) and ARC (anterior-most red cell). After conducting several trials, the probability distribution in the time interval of the PGC, but not the ARC, was found to provide information regarding a possible regulatory mode. Once the G1/S transition window opens at a given time point of embryonic development, all cells within the window equally acquire the chance to enter the S phase. Among these cells, the cells that immediately traverse the G1/S transition demarcate the spatial position of the newly established window. Therefore, the PGC tends to indicate the posterior end of the window, although this is not always precise, thus implying some information about the possibility of retrieving the progressive mode of the window shift. In contrast to the PGC, the ARC comprises the last cells to traverse the G1/S transition affected by the window at a given time, thus losing the spatio-temporal information of the window. This finding highly contributed to our model selection, which suggests that analyses of immediately responding cells provide clues for inferring possible regulatory modes under stochastic cellular behavior. In order to statistically determine which model represents the most likely scenario, we applied the KL distance, a measurement of the difference between two probability distributions [51]–[53]. The KL distance is applied as a discriminant function; therefore, it is often used as a basic tool for model inference and selection. This use of the KL distance enables hypothesized models to be ranked from likely to unlikely, thus allowing for the selection of the most possible regulatory mode. Consequently, the G1/S transition in notochordal cells was found to progress under the influence of periodic developmental waves, and its periodicity is most likely to involve the two-fold period of somite segmentation. We validated this result by employing another method to score the results of the in silico simulation. Enumeration of the total number of green cells in the G1/S transition window was applied, as performed in the linkage analysis to analyze the independency of the timing of S phase entry in each cell. The analysis also demonstrated that only periodic waves of the two-fold period of somite segmentation showed stable synchronized oscillation.
Our analyses also clarified that the G1/S transition in notochordal cells posteriorly travels in a stepwise fashion, which appears to be accompanied by embryonic body axis elongation and neighboring paraxial mesoderm segmentation. This finding implies that the G1/S transition regulatory window is reminiscent of the bistability window of the mutual inhibition of FGF and RA signaling, which has been proposed to provide a poised state of segmentation of the paraxial mesoderm as well as initial differentiation of neuronal cells from stem cells [54]–[58]. Within the bistability window, cells can be triggered to switch between either of the two steady states. This trigger is provided by the periodic signal of the segmentation clock in the paraxial mesoderm. In fact, the occurrence of periodic surges of the G1/S transition of paraxial mesoderm cells at the exact position of the bistability window in chick development has been proposed [7], although we did not observe a similar phenomenon in the zebrafish paraxial mesoderm [48]. However, the periodicity operated in the G1/S transition window is two-fold longer than that of the zebrafish segmentation clock. None of the cyclic patterns of the gene expression have been observed in early notochordal cell development. It has also not yet been elucidated how such an extrinsic bistable switch comprised by FGF and RA gradients affects the intracellular machinery. Therefore, the molecular machinery of the cyclic regulation of the G1/S transition in notochordal cells remains to be elucidated.
As it has been reported that a component of the segmentation clock oscillates during chick fore-limb development with a six-hour periodicity that is four-fold longer than that of chick somite segmentation [59], a mechanism that multiplies the periodicity composed by Notch signals may exist in developing tissues, potentially linked to periodic surges of cell division, as previously described [3]. Given that intracellular bistable switches have also been proposed to be involved in the cell cycle phase transitions [30], [36], it is interesting to consider how distinct levels of multiple bistable switches interact to induce dynamic changes in the cellular properties, which may cause phenomenologically described complex modes of periodic entrainment. It would be worthwhile to analyze a model organism in which the activity of intracellular molecules composing a bistable switch, such as Rb and E2F, are genetically modified.
Recently, a Doppler effect has been reported to be involved in zebrafish segmentation [60]. Time-lapse observations of the transcriptional activity of her1, a component of the segmentation clock, during embryonic development revealed that its oscillation in the posterior portion is slower than that in the anterior part. Since the anterior oscillation is directly linked to the pace of segmentation, the shortening rate of the presomitic mesoderm due to a gradual slowdown in the rate of embryonic body axis elongation may affect the segmentation pace through a Doppler effect. It is possible that the stepwise posterior shift of the G1/S transition window observed in this work was also affected by this Doppler effect. However, this is less obvious, because the G1/S transition window is always located between newly formed segments and the tail tip throughout embryonic development.
It is relevant to consider how the stochastic G1/S transition would benefit organ development. Vertebrate embryos employ a robust system for developmental pattern formation, with the primary function of establishing a reproducible pattern in a proliferating population of cells. Biological robustness is proposed to be a characteristic required to maintain cell functions in the face of external and internal perturbations [61]. The stochasticity observed in molecular and cellular functions can be considered an opposite concept to robustness. For instance, cell proliferation has been reported to serve as a source of noise for the synchronized expression of genes of the segmentation clock [16]. However, stochasticity in the cell cycle also provides the mechanical flexibility necessary to adapt to a fluctuating environment and/or respond to sudden changes in environmental cues. Therefore, it is possible that the stochasticity of the G1/S transition in notochordal cells is required to maintain the tissue structure and provide a poised state for subsequent tissue development.
Correlations of cell cycle progression in embryonic morphogenesis and organ development have long been proposed based on counts of the number of cells exhibiting mitotic nuclei in tissue sections or three-dimensionally reconstructed fixed tissues. By exploiting fluorescent live imaging of Cecyil embryos with subsequent scoring and mathematical modeling, we successfully extrapolated an unprecedented regulatory pattern of the G1/S transition during early notochord development. It is possible that periodic regulation of cell cycle progression in a group of cells is a common phenomenon and that the stochastic nature of cell cycle progression in developing tissue hinders an unobserved regulatory system. Our approach provides a way to infer the regulatory mode of the stochastic G1/S transition in tissue and organ development. The potential utility of this approach may be applicable, not only for understanding physiological development, but also clarifying mechanisms of pathological tissue development, such as that involving carcinogenesis. Generally, cancer cells are associated with the affected cell cycle mechanism, which may directly or indirectly influence the bistable switch of cell cycle transitions. Therefore, tumor tissue growth can increase the occurrence of the complex or de-regulated mode. The gold standard for cancer classification fundamentally involves microarray-based gene profiling to characterize the cancer signature. The utility of mathematical models of stochastic molecular pathways is also proposed to be applicable for cancer diagnosis [37]. Mathematical models based on experimental studies are proposed to bridge genetics and tumor behavior, with the potential to provide better personalized cancer treatment [62], [63]. The development of live imaging-based mathematical models to infer both the heterogeneity and dynamics of cellular behavior will provide further insight into the characteristic features of both tumor and physiological tissue development.
All experimental procedures using zebrafishes were conducted with following ethical regulations of the Experimental Animal Committee of RIKEN BSI.
In vivo time-lapse imaging was carried out as previously described [48], with some modifications. Dechorionated embryos were embedded in small holes molded by glass beads (Iuchi BZ-1) on the surface of 1% agarose (Takara L03) solution in E3 medium (5 mM NaCl, 0.17 mM KCl, 0.4 mM CaCl2 and 0.16 mM MgSO4) and then covered with 0.3% agarose. The chamber encasing the embedded embryos was filled with E3 solution containing Tricaine. Time-lapse 3D imaging was subsequently performed in the xyz-t mode using a confocal upright microscope EZ-C1 system (Nikon, Tokyo) equipped with a water-immersion 16× objective (N.A. 0.8). Two laser lines, 488 nm and 561 nm, were used. Differential interference contrast microscope (DIC) images were concomitantly acquired. The recording interval was 2.5 minutes. At each time point, optical sagittal sections of 27∼34 confocal images along the z axis were acquired to access developing notochord tissue located in the deepest embryonic layer. In order to avoid cross-detection of green and orange signals, the images were acquired sequentially at 488 nm and 561 nm. Total eighteen time-lapse imaging data of distinct embryos were obtained. Seven of them were subjected to image processing, and representative data of three distinct embryos were used for modeling analyses.
Dechorionated Cecyil embryos were fixed with 4% PFA (pH 7.4) in PBS for one hour at room temperature, then washed in 0.1% Triton in PBS (PBT). The embryos were then incubated one hour at room temperature in PBT containing 660 nM of Alexa Fluor 647-Phalloidin and washed in 0.1% Triton in PBS (PBT). Image acquisition was performed using an EZ-C1 (Nikon) confocal upright microscope system equipped with 488 nm, 561 nm and 638 nm laser lines. More than twenty distinct fixed embryos were observed with comparing to time-laps data.
Optical slices focusing on tissue layers containing notochord tissue at each time point were manually selected, and the time sequence was reconstructed for further image processing. Cell tracking was performed using the Manual Tracking plugin in the ImageJ software program. To ensure recognition of the final position of each notochordal cell, the tracking was conducted in a temporally backward and forward manner. In order to monitor temporal changes in the green and red fluorescence intensity of each cell, we developed a program that automatically averages the signal intensity of a defined area of the nucleus at each time point on sequential images. This program was written in C/C++ for use under Win32 environments. In order to define the timing of the G1/S transition of each cell, the time course of the signal intensity of the green and red fluorescence signals was plotted. The time point of transition was defined as the time at which the intensity of the green signal increased and exceeded that of the red signal continuously five times points.
An algorithmic description for the random and biased simulations of the binary valued cell distribution was provided. For the random case, we spatially distributed the upper and lower green cells randomly, while keeping the same number of green cells as observed in the experiments. The initial state was set to all red. With respect to the randomly chosen site i, only if the chosen site was red, the site was set to green, and this procedure was repeated until the total number of green cells reached the experimentally observed number. The expectation value of the number of green cell pairs was calculated by repeating the simulation 10,000 times. For the biased case, we first distributed the upper green cells randomly as performed in the random case. We then distributed the lower green cells according to which (randomly chosen) site was set to green, with a probability q (here q = 0.9 is taken) when the corresponding upper cell was green and a probability 1-q ( = 0.1) when the corresponding upper cell was red. The expectation value of the number of green cell pairs was calculated by repeating the simulation 10,000 times.
In order to describe the stochastic cell cycle transition, we developed a mathematical model based on the Markov process [64], [65]. When only a single cell is considered, at each time step t, a cell undergoes the G1/S transition with a probability αΔt within a short time interval Δt (Figure 4A). The backward process of the transition, such as the green (S) to red (G1) transition, is not assumed. The time t is measured in units corresponding to the time-lapse interval Δt = 2.5 minutes, and the parameter α represents the probability of the transition per unit time Δt. In a developing notochord, the cells are arrayed along the anterior-posterior axis, and the cell cycle transition progresses from anterior to posterior in a time-dependent manner. In order to model this time-dependent stochastic process on the one-dimensional axis, we introduced the signal transmitting function f(i,t) of t and i, where i denotes the cell position along the anterior-posterior axis. The stochastic process is designed such that each cell independently undergoes the G1/S transition with a probability αΔt, if the condition f(i,t)≤0 is satisfied (Figure 4B). The signal transmitting function f(i,t) can be written in the following form:where z is a positive integer, and tk is taken as tk = k Lc z/(λΔt). Lc is the diameter of notochordal cells and λ is the posterior elongation speed, both of which are estimated from time-lapse imaging data as Lc = 5 µm and λ = 4/3 µm/min, respectively. Therefore, the number of cells constituting a single somite width is calculated as T*λ/Lc = 8, where T ( = 30 min) is a period of the segmentation clock. The time-dependent function (tk≤t) is a step function satisfyingIn the numerical simulations, tm should be greater than the maximum simulation time tmax; therefore, m should be taken so as to be m>mc = λΔt tmax/(Lcz).
Hypothesized models can be obtained by modulating the parameter z. We categorized the models into two basic types, “the continuous model” and “the periodic model.” If we take z = 1, we can obtain the continuous model in which the external cell cycle cue comes continuously in a spatiotemporally correlated manner. The periodic models are further divided into subtypes based on the cue period. In this case, z can be interpreted as the number of cells constituting a single somite width. Hence, if we take z = 8, we can obtain the (normal) periodic model in which the signaling cue comes periodically, in concert with the cycle of the segmentation clock. When z is taken to be equal to the integral multiple of 8, we have a multifold periodic model in which the period is a multiple of the cycle of the segmentation clock. In this study, we considered the two-fold periodic model (z = 16) and three-fold periodic model (z = 24). Regions satisfying f(i,t)≤0 for both the continuous and periodic models are depicted in Figure 4C.
We described how the probability distribution function for the difference in the S phase entry time between two independent cells was derived. We considered a Bernoulli trial with a probability p of success. A variable x is represented as x = (0,1), thus x takes 1 (success) with probability p and x takes 0 (failure) with probability 1-p. The initial value of x is set to 0, and the trial is repeated until we have the first success x = 1. The probability distribution function f(t) that x becomes 1 at the trial time t is given bywhere f(t) satisfies
We next considered the joint probability distribution f(t1,t2) of two independent trials,The two independent processes correspond to experimentally obtained processes of the G1/S transition for an upper and lower cell. In order to derive the distribution function for time difference t = t1−t2, we summed f(t1,t2) fixing t with respect to t1 and t2,where δt, t′ is the Kronecker delta. With a short calculation, we obtained the analytical expression of g(t)This expression was used to estimate the parameter p = αΔt, the probability of the G1/S transition per unit time Δt (Figure 4D).
We utilized the Kullback-Leibler (KL) distance to measure the difference between the experimental and simulation data and determine which hypothesis best reflects the system. The KL distance [51] is defined aswhere p(x) and q(x) denote probability distribution functions. In our calculations, p(x) represents the probability distribution function of time interval for PGC obtained from the experimental data and q(x) represents that obtained from the simulation data. By definition, if p(x) = q(x), obviously D(p,q) = 0.
The numerical simulations and statistical analyses were carried out using the MATLAB software program (The Mathworks, Inc.). Parameter fitting was performed with the MATLAB software using the fminsearch function. The target function was chosen as the residual sum of squares between theoretically estimated values and the experimentally obtained values.
|
10.1371/journal.pntd.0004417 | Climate and the Timing of Imported Cases as Determinants of the Dengue Outbreak in Guangzhou, 2014: Evidence from a Mathematical Model | As the world’s fastest spreading vector-borne disease, dengue was estimated to infect more than 390 million people in 2010, a 30-fold increase in the past half century. Although considered to be a non-endemic country, mainland China had 55,114 reported dengue cases from 2005 to 2014, of which 47,056 occurred in 2014. Furthermore, 94% of the indigenous cases in this time period were reported in Guangdong Province, 83% of which were in Guangzhou City. In order to determine the possible determinants of the unprecedented outbreak in 2014, a population-based deterministic model was developed to describe dengue transmission dynamics in Guangzhou. Regional sensitivity analysis (RSA) was adopted to calibrate the model and entomological surveillance data was used to validate the mosquito submodel. Different scenarios were created to investigate the roles of the timing of an imported case, climate, vertical transmission from mosquitoes to their offspring, and intervention. The results suggested that an early imported case was the most important factor in determining the 2014 outbreak characteristics. Precipitation and temperature can also change the transmission dynamics. Extraordinary high precipitation in May and August, 2014 appears to have increased vector abundance. Considering the relatively small number of cases in 2013, the effect of vertical transmission was less important. The earlier and more frequent intervention in 2014 also appeared to be effective. If the intervention in 2014 was the same as that in 2013, the outbreak size may have been over an order of magnitude higher than the observed number of new cases in 2014.The early date of the first imported and locally transmitted case was largely responsible for the outbreak in 2014, but it was influenced by intervention, climate and vertical transmission. Early detection and response to imported cases in the spring and early summer is crucial to avoid large outbreaks in the future.
| Dengue has not been considered to be a major problem in China since it is recognized as an imported disease and only 8,058 cases were reported from 2005 to 2013. However, in 2014 alone, 47,056 new cases were reported. In this study, a mathematical model was developed to determine the possible cause of this outbreak. The most important parameters found to underlie the pattern of a small outbreak in 2013 and a much larger one in 2014 was the timing of the first imported and locally transmitted case. The importance of precipitation and temperature was also confirmed by the simulation results under different climate scenarios. The model also suggests that the earlier and more frequent control interventions in 2014 targeting immature mosquitoes, such as emptying water containers, and adult control, were effective in preventing larger outbreaks. Furthermore, more attention should be paid to imported cases occurring between March 1st and July 1st to prevent early and prolonged transmission. Without early detection and response, the final outbreak size might otherwise be an order of magnitude or more the size when the imported case occurred outside this time period.
| Dengue is a febrile illness caused by the dengue virus which is further classified into 4 serotypes (DENV 1–4), and transmitted by Aedes aegypti and Aedes albopictus mosquitoes. Classically, dengue virus infection produces mild flu-like fevers but can also result in lethal dengue hemorrhagic fever (DHF) and dengue shock syndrome (DSS) when infected a second time with a different serotype [1]. According to the World Health Organization (WHO), dengue is the fastest growing vector-borne disease in the world with only one thousand cases reported in the 1950s to more than 90 million cases in the 2000s [2]. Estimated from a systematic literature search, there were 96 million apparent dengue infections globally in 2010; however, an additional estimated 294 million infections were asymptomatic [3].
Dengue is believed to be an imported disease in mainland China, and 55,114 cases were reported from 2005 to 2014. Approximately 94 percent of the indigenous cases that occurred in this period were reported in Guangdong Province, and 83 percent of these Guangdong cases were in Guangzhou City [4]. In 2014, an unprecedented dengue outbreak hit Guangzhou, with 37,341 new cases contributing to 94 percent of the new cases from 2005 to 2014 in Guangzhou. The annual new cases in Guangzhou were normally lower than 150 except for the 765 in 2006, 1,249 in 2013 and 37,341 in the 2014 outbreak.
Guangzhou differs significantly from other dengue transmission areas with Ae. albopictus as the sole vector rather than Ae. aegypti [5]. Unlike Ae. aegypti, Ae. albopictus adapts to the cold winter in temperate and subtropical areas by diapausing, which gives it the ability to expand to higher latitudes. Normally, the adults cannot survive the low temperature in winter, but they can produce diapause eggs when the temperature becomes lower and the day becomes shorter [6,7]. These diapause eggs will not hatch until the next spring, when the temperature and water condition become favorable again. Moreover, the vertical transmission of dengue virus in Ae. albopictus is more efficient [8], with approximately 0.5 to 2.9 percent of the eggs laid by infected mosquitoes being infected [8–10]. When the vertical infected diapause eggs develop to adults in the next spring, they have the ability to infect humans immediately without biting infected humans, even causing a significant outbreak if there were sufficient infected eggs in the past year. This pathway might allow dengue to become endemic in Guangzhou. The other possibility for dengue to be endemic is through overwintering infected adults, especially when global warming increases the temperature in the winter. However, the daily mean temperature from December to February of the 30-yr average (the coldest three months) was 14.8°C, and that of 2013 was 14.4°C. Thus the possibility for infected adults to live through the winter of 2013 is relatively low, considering that the temperature in the winter of 2013 was not abnormally high. Mathematical models suggest that vertical transmission can increase the endemic level of the vector population and human population significantly [11]. However, Ae. albopictus is less efficient in transmitting dengue virus. Typical explosive DHF epidemics have not been found in the places where Ae. albopicus predominates over Ae. aegypti, such as parts of China, the Seychelles Islands, La Reunion Island, the Maldive Islands, historically in Japan and most recently in Hawaii [12,13].
Another possible causal factor for the 2014 outbreak in Guangzhou was the abnormally high precipitation in May and August which provided more breeding sites and increased the environmental carrying capacity for Ae. albopictus [14]. A third possibility was the early starting date of the outbreak, with the earlier imported cases occurring in the late spring and early summer leading to the greater final size of the epidemic as a result of the lengthened infection season before the decrease of Ae. albopictus abundance in the winter [15]. A multivariate Poisson regression analysis of the Guangzhou outbreak data was recently published that showed the number of imported cases, minimum temperature with a one-month lag and cumulative precipitation with a three month lag predicted the outbreak in 2013 and 2014 [16]. Here we use a mathematical model rather than statistical model to further explore the factors underlying these outbreaks since the structure of our model is based on mechanistic factors controlling both mosquito population dynamics and the dynamics of viral transmission explicitly and, therefore, should allow greater confidence in making predictions in the presence of environmental change [17].
There were 99 dengue transmission models cited in literature from 1970 to July 2012, most based mainly on the Ross-Macdonald model of malaria transmission, the classical theoretical framework for modelling mosquito-borne diseases [18]. However, the core assumptions of some of these models differs from Ross-Macdonald model in various ways. Examples include those explicitly modelling the mosquito immature stage population dynamics [19,20], the temperature-dependent extrinsic incubation period (EIP) [21], vertical and mechanical transmission [11,22], spatial heterogeneity [15,23], control strategies [24,25], or multiple dengue virus serotypes [26]. Stochastic models [15] or agent-based models [27] have also been developed to simulate the transmission dynamics of dengue virus. To emphasize local characteristics, we included only immature stage population dynamics, vertical transmission, control strategies, and temperature-dependent adult mosquito mortality rate, biting rate and the EIP. Coinfection, multiple pathogen types, and temporary immunity were not considered here since dengue virus 1 (DENV-1) has been the only predominant serotype found in Guangzhou since the 1990s [4]. Though the test results for 2014 are not ready, in the 1,249 cases of 2013, 1,243 are DENV-1 cases and only 6 are dengue virus 2 (DENV-2) cases. Spatial distributions of the mosquitoes, and heterogeneous biting were not considered here mainly because of limited data availability.
In this paper, a population level deterministic mathematical model including explicitly modelled water level and mosquito population in different life stages was developed. Then the parameters in the model were estimated via successive cycles of fitting. Observed monthly mosquito index was used to validate the mosquito submodel. Finally, different scenarios were created to investigate the important mechanisms responsible for the unprecedented outbreak of dengue in 2014 in Guangzhou City.
The study was reviewed and approved by the Ethics Committee of the Guangzhou Center for Disease Control and Prevention. All the patient data were de-identified and the data were analyzed anonymously.
Guangzhou is the capital and largest city of Guangdong Province, with a total area of 7,434 square kilometers [28] and a population of 13 million at the end of 2013 [29]. (Fig 1). It is one of the most urbanized areas and the center of China's economic growth. With the Tropic of Cancer crossing just north of the city, Guangzhou has a humid subtropical climate with hot and wet summers and mild and dry winters. The annual average temperature is approximately 21.5°C. January is the coldest month with an average temperature of 13.0°C while the hottest is July at 28.5°C. Annual rainfall varies from 1,612 to 1,909 mm, with more than 80 percent occurring between April and September [30]. The wet and warm climate is favorable for the growth of Ae. albopictus, which is the secondary vector for dengue virus in the world but the sole vector in Guangzhou [31,32]. Though dengue is not endemic in Guangzhou, more than 0.30 million travelers from dengue endemic countries such as Malaysia, Singapore, Indonesia, Thailand and India visit Guangzhou each year. These countries are also the top choices for outbound travelers from Guangzhou [33]. Since the natural and socio-economic conditions in Guangzhou are conducive to mosquito growth and reproduction, high densities of Ae. Albopictus together with dengue-infected travelers present a high potential for initiating local spread of the disease [31].
Dengue is a notifiable disease in China which means that, once diagnosed, cases must be reported to the web-based National Notifiable Infectious Disease Reporting Information System (NIDRIS) within 24 hours. All case reports used in this analysis were diagnosed according to the National Diagnostic Criteria for Dengue Fever (WS216-2008) published by the Chinese Ministry of Health [34]. In addition, active case detections was carried out through field investigations in the communities with confirmed dengue cases [14]. Cases were then divided into indigenous and imported cases based on whether the patient traveled to a dengue endemic area and was bitten by mosquitoes there within 15 days of the onset of illness [14]. A list of daily reported new cases for 2013 and 2014, obtained from Guangzhou Center for Disease Control and Prevention (Guangzhou CDC), was used to calibrate the model. This dataset was published online in the transmission season on the website of the Health Department of Guangdong Province (http://www.gdwst.gov.cn/). There were a total of 1,249 and 37,341 reported cases for 2013 and 2014, respectively.
Monthly mosquito surveillance reports consisting of the Breteau Index (BI) and the Mosquito Ovitrap Index (MOI) in 2013 and 2014 were also obtained from Guangzhou CDC and used to validate the mosquito submodel (S1 Table). BI is the number of positive containers with Ae. albopictus larva per 100 houses inspected, and is considered to be the best single index for Aedes density surveillance [14]. MOI is the percentage of Ae. albopictus positive ovitraps in all ovitraps collected from a specified area, and reflects the abundance of the adults [35].
Daily temperature, rainfall and evaporation data for Guangzhou from 2012 to 2014, which were used as inputs to the model, were downloaded from the China Meteorological Data Sharing Service System (CMDSSS) (http://cdc.nmic.cn/). In addition, climate data from 1985 to 2014 were also retrieved from CMDSSS to calculate 30-year daily average values.
Population data for the human submodel was obtained from the Guangdong Statistical Yearbook on China Infobank (http://www.bjinfobank.com/). This data was also used to estimate human birth rate and death rates in Guangzhou [36–38].
A deterministic mathematical model was developed to interpret the transmission of dengue in Guangzhou city based on the Ross-Macdonald model [39,40], which is a basic framework widely used to study the dynamic transmission of mosquito-borne diseases. Fig 2 presents the structure of our model with Table 1 showing the definition for each symbol in this figure. Temperature can influence the development rate, death rate of immature mosquitoes, average duration of and number of eggs laid each gonotrophic cycle, biting rate and the EIP of dengue virus [41–43]. The form of temperature-dependent functions were based on [20,41], and the coefficients were estimated from experiments on Ae. albopictus strains from Guangzhou and adjacent areas [42,44]. Density of the larvae also plays an important role in the development rate of eggs and larvae, and the death rate of larvae. The form of density forcing rates were taken from [27]. More detailed information about the parameters, temperature or density forcing functions for Ae. albopictus development and death rates, and the differential equations for the model can be found in the S1 File.
The model includes several modifications to the Ross-Macdonald framework to incorporate the influence of climate factors, vertical transmission and local interventions. First, the immature aquatic phases of Ae. albopictus were modeled explicitly since the development rate of eggs, larva, and pupa, as well as the mortality of larva and pupa can be influenced by temperature and density. Second, a SEI (Susceptible, Exposed, and Infected) model was used for mosquito submodel instead of a SI (Susceptible and Infected) model to capture the temperature-dependent pathogen latency in Ae. albopictus. Thirdly, an element to reflect mosquitoes infected by vertical transmission was added, because Ae. albopictus has the ability to transmit dengue virus vertically through eggs, with a filial infection rates ranging from 0.5 to 2.9% for Dengue-1 virus [8]. Fourthly, we explicitly modeled the water availability by including evaporation, rainfall, and maximum and minimum water level (See details in S1 File). The environmental carrying capacity for mosquitoes will increase when the water level rises, and the density-dependent death rate will decrease in a short period. Furthermore, a spillover effect is triggered when there is an extreme rainfall event and the water level is close to the maximum water level, resulting in a loss of immature mosquitoes. The ideal death rate of larva and the development rate of eggs and larva depend only on temperature. However, the real death rate also depend on the water-level or density of the larva (See S1 File for more information). Similarly, the control intervention to empty water containers can also remove a fraction 1-μi of water and immature mosquitoes, while ultra-low-volume (ULV) aerosol applications of insecticides can kill a fraction 1-μa of adult mosquitoes. In addition, temperature-dependent biting rate and the number of eggs per gonotrophic cycle were incorporated to better represent the effects of climate on mosquito population dynamics. Since dengue is still considered as a non-endemic disease in China, which means new autochthonous cases occur only after imported cases, an imported case input was added to the system at day β2013 and β2014 (January 1st, 2012 as day 1) to initiate the outbreak in 2013 and 2014, respectively. Instead of using the date of the first reported imported case, we treated the timing of the first imported case as a parameter, since the outbreak may be started by an unreported or asymptomatic case. We only added the first imported case to the system and left out all the other subsequent imported cases, because it was a small number when compared with the number of infectious people after the rapid local transmission began, and was reasonable to be ignored. And because Ae. albopictus will survive adverse winter temperatures as diapausing eggs, the development rate from eggs to larva is assumed to be zero from late October to early March [53]. The reporting rate φ was also included to account for the asymptomatic and unreported dengue infections.
In summary, a SEI model was used for the vector submodel and SEIR (Susceptible, Exposed, Infected and Recovered) model for the human submodel (Fig 2). Five different life stages for mosquitoes were considered: three aquatic stages (E, eggs; L, larva; P, pupa), one emerging adult stage (Ae), and one biting and reproductive adult stage (A). Subscripts u and i were used to represent uninfected and vertical infected aquatic phases and emerging adults; while s, e, and i were used to denote the populations of susceptible, exposed, and infected adults. Analogously, the human population was divided into four subclasses: Hs, He, Hi, and Hr, which stands for susceptible, exposed, infected and recovered humans, respectively.
All the analyses were conducted in R 3.2.0 [54], and the differential equations in the model were solved by R package deSolve [55]. The model was run over the period 2012 to 2014, though the focus is on simulating the dengue outbreak in 2013 and 2014. The mosquito abundance for only the first simulated year is affected by the initial value for eggs, and the following years showed no memories from previous years [56], so an extra year was needed to achieve a stable mosquito population for 2013 and 2014. However, for simplicity, we assumed that there was no imported cases in 2012. The only possibility for dengue cases in 2012 to affect the next two years was through vertical transmission. Taking into account the low vertical transmission rate and the small number of dengue cases in 2012 (139 cases), we assumed that the influence of 2012 on the next two years was negligible.
Typical of the class of mechanistic disease transmission models used here, there are a large number of parameters with substantial uncertainty in their values. In addition, in this analysis we place considerably more confidence in the timing and pattern of the field data describing human cases and mosquito infection and abundance in Guangzhou in 2013 and 2014 than in the precise numbers reported on any day. As a result, we chose to address the issue of parameter estimation using a strategy that has been called regional sensitivity analysis or RSA [57]. This approach begins with the specification of a region of parameter space thought to include the range of feasible values of each parameter with high probability (As the typical value for each parameter in Table 1). Monte Carlo simulation runs are then conducted to assess the performance of the model over this parameter space. Here we define this space by specifying the univariate marginal distributions of the model parameters need to be estimated, as given in Table 1, each of which we assume to be independent.
Classification criteria are then defined and applied to the output of the model to determine if a particular realization captures the essential features of the pattern of daily case reports. Fig 3 shows the specific criteria for the 2013 and 2014 Guangzhou dengue case reports (See the detailed criteria in S1 File). If a particular model run results in a case report trajectory passing through all six of the shaded windows, the model is classified as a “pass”, that is as having adequately mimicked the pattern of the field data used for calibration.
Passing and failing parameter vectors are then collected for subsequent analysis. Usually the first simulation experiments using the RSA approach result in a very small fraction of passes and these vectors extend over almost the entire range of any of the univariate prior distributions. This is a result of the fact that there are many parameter combinations that can produce the same patterns of model response and their correlation structure is usually very complex in the high dimensional parameter space being sampled. Non-uniqueness of model parameterization of this sort is an issue about which there is a substantial literature particularly in the field of hydrology [57]. In the present case, 74 passing parameter vectors were obtained in 410,594 initial simulation runs which we term Cycle 1. In the S2 File the sample cumulative distribution functions are shown for each parameter for passes and fails. As shown there, some parameter distributions that differ little between passes and fails which gives little clue as to parts of the range of that parameter where passes are more likely. The value of dm,n, the Kolomogorov statistic, is a measure of the maximum difference between the two distributions and can be used as a rough index of sensitivity. Very large differences can be seen for some parameters in Cycle 1, for example β2013, μi, μem and ωmax.
In view of the very low pass rate of the first set of simulations, we chose to use the outcome of the Cycle 1 experiments to seek a subspace in which passing parameters were more likely to be found. This was done by trimming the ranges of parameters with large values of dm,n. Trimming was an ad hoc procedure based on trimming the range of parameters with few passes at either the high or low end of the sample distribution function of passes. A total of 4 trimming cycles were conducted resulting in a pass rate of 3.19% in the final subspace, Space 5, an increase of about 175 times over the initial space, Space 1. The marginal distributions of the parameters in Space 5 now show much reduced differences between passing and failing distributions as discussed below. We regard a highly trimmed parameter range and a large decrease in dm,n between spaces 1 and 5 as evidence of the importance of a parameter in producing simulations meeting the pass criteria. However, we note that a parameter may be very important, but if initial uncertainty in its value is small, that is the prior range is narrow, there may be little difference in the marginal distributions under passes versus fails. A second situation in which a parameter can show little difference in its pass/fail marginal distributions yet be important can occur if there are interactions with other parameters not reflected in the marginal distributions. The pairwise correlation matrix can give some clues to such situations and will be discussed below for Space 1 and Space 5. (See S2 Table for the pairwise correlation matrices for parameter values Cycle 1 and Cycle5)
Fig 4 summarizes the case report data for 2013 and 2014 and shows the envelope of 637 passing simulation trajectories from the Space 5 parameter distributions. The daily number of new cases output by the model was calculated as the number of individuals entering the compartment Hi times the reporting rate φ. The median final epidemic size for 2013 and 2014 was 1,044 and 30,863, respectively, for the 637 passing parameter sets of Cycle 5. Although the envelope of passing simulations contains the observed peak values in both years, the median passing peaks were 16% and 17% lower than the observed peaks respectively.
Fig 5 shows the Cycle 5 simulation results and field data for larvae and adult mosquitoes. We aggregated the monthly average amount of larva and adults from daily model output of the 637 passing simulations, then normalized them to 0 to 1 and plotted them against the normalized BI and MOI data from Guangzhou CDC. Mosquito surveillance data in 2012 was not used in the validation, because the mosquito abundance in the first simulated year can be affected by the initial value for eggs. Entomological surveillance data recorded only the absence/presence not the number of Ae. albopictus in each container, so it is only a proxy of the abundance. The minimum, maximum, mean and standard deviation for Pearson’s correlation between scaled model output larva amount and BI were 0.76, 0.86, 0.82, and 0.02, respectively. And the correlation for scaled model output adult amount and MOI ranged from 0.65 to 0.80, with a mean of 0.74 and standard deviation of 0.03. The BI and MOI data were not used in calibration but the patterns produced by the model, as shown in Fig 5, confirm that the model is producing realistic patterns of mosquito abundance over time.
Table 2 shows the Space 1 versus Space 5 marginal distribution comparisons with RR (range reduction) denoting the fractional reduction of the range in each parameter. The ranges of six parameters were unaltered and nine were reduced by 50% or more of their initial range. The nine fall into three types of parameters, those associated with vector population dynamics and infection (μE, λ, σ, and αhv), those related to the timing and reporting of imported cases (β2013, β2014, and φ), and those associated with the effectiveness of mosquito control interventions (μa and μi).
The pairwise correlation matrices for the passing parameter distributions in Space 1 and Space 5 are shown in S2 Table. For Space 1, the same 9 parameters with large range reductions show high correlations with one or more of others in that group. There is also a very high correlation between ωmin and ωmax, an artifact that is imposed by the model structure. In the Space 5 correlations, all of the high values from Space 1 are lower, and most very much lower, as might be expected. However, some new correlations emerge, notably with πmax, the maximum carrying capacity for immature stages of the mosquito. These correlations are with μem, mortality during adult emergence, and the human to vector, αhv, and vector to human, αvh, transmission probabilities.
We do not believe we have access to additional field data or other information which will allow significant further reduction in Space 5 or point to other areas in the parameter space that might suggest alternative underlying processes to be driving the observed patterns of behavior of the system. Hence, parameter vectors meeting the passing criteria sampled from Space 5 will be used in the subsequent explorations of the key processes underlying the 2014 epidemic.
The year 2013 differed from 2014 in several aspects, notably the date of imported cases (β2013 and β2014), climate, the time and frequency of the interventions, and the number of eggs infected by vertical transmission from the previous year.
We first explore the timing of imported cases, which is not the real timing of the first imported case reported to the NIDRIS, but a parameter we need to estimate, denoting the first imported case that starts the local transmission. The outbreak in 2014 started at June 11th, and peaked around October 1st, with a time interval of 112 days, while the smaller outbreak in 2013 began at July 14th, and peaked around October 19th, with an interval of 97 days. If the force of infection was the same for these two years, the final size of epidemic in 2014 would be significantly higher than that in 2013 on this basis alone. Without interventions, the peak occurs when the temperature drops to cause a sufficient combination of a decrease in biting rate and an increase in the mosquito death rate. Appropriately timed interventions reduce the abundance of mosquitoes, thus reducing the force of infection which results in an earlier peak. To make the two years comparable, we changed the date of the first imported case in 2014, β2014, in the 637 passing parameter sets to β2014 + (Peak time 2013—β2013) (Scenario Postpone 2014). By doing this, we made the time interval between the imported case and the peak in 2014 equal to that in 2013. The case report trajectory for each run was recorded (Fig 6A). Then in another scenario (Scenario Advance 2013), we changed the β2013 to β2013 –(Peak time 2014—β2014), to attempt to produce an outbreak in 2013 with a size similar to the observed number in 2014 (Fig 6B). Furthermore, to investigate the relationship between the date of imported cases and the final epidemic size further, we kept all the other parameters the same as in the 637 passing sets, and only changed β2014 in each set to integers between Day 791 and 1066, that is from March 1st, 2014 to November 30th, 2014 (Scenario Change importing dates), and recorded the final epidemic size for each run (Fig 6F).
The results of these experiments are shown in Fig 6. When the time interval between the imported case and the peak in 2014 was changed to match that in 2013, only 30 parameter sets (4.7% of the original 637) mimicked the pattern of the outbreak in both years. The median final epidemic size of 2014 dropped to 1,474, similar to that of 2013 (Fig 6A). And when the time interval between the imported case and the peak in 2013 was increased to the same as that in 2014, none of the 637 parameter sets produced passing behaviors. As shown in Fig 6B, after the change, the peak number of cases was significantly higher in both years, with new median final outbreak sizes of 158,889, and 137,003 for 2013 and 2014, respectively. In summary, postponing the date of the import case in 2014 produces an outbreak whose scale is similar to that of 2013, and advancing the date of the import case in 2013 produces an outbreak even worse than observed in 2014. In addition, since all other parameters were unchanged except for β2013, the larger than observed outbreak in 2014 is attributable to vertical transmission, that being the only way that the situation in 2013 can influence that in 2014. A separate scenario was created, still by advancing β2013, but removing all infected eggs in the system at the beginning of 2014, as discussed below.
The final experiment on the timing of imported cases involved holding all parameters the same as in the 637 passes, except that of the date of imported case which was varied from March 1st to November 30th. The final epidemic size for each run is plotted in Fig 6F and shows that when the first imported case occurs on April 18th, the median final epidemic size was the highest at 60,158. The final epidemic size became stable after July 1st at approximately 1,350, similar to the observed size in 2013. These experiments clearly suggest that the date of the first imported cases was a crucial determinant of the severity of the 2014 epidemic.
However, the force of infection was not the same in 2013 and 2014. It is affected by mosquito abundance, biting rate, transmission probability from vector to human and transmission probability from human to vector. Though we assumed the transmission probabilities were the same for 2013 and 2014 (αhv and αvh), the biting rate depends on temperature and the mosquito abundance depends on both temperature and precipitation. Hence, different scenarios were created to study the role of climate of the variations in climate depicted in Fig 7. Experiments were conducted in which the precipitation, temperature, and evaporation data of 2014 were replaced by data of 2012, 2013 or of the 30-yr average. Since the temperature in 2014 did not differ significantly from that in other years, while the precipitation in May and August 2014 were much higher, we also ran simulations with actual temperature and evaporation in 2014, but scaled the precipitation to 30 year average.
The new case trajectory for the real climate data was treated as baseline here, so the passing rate was 100 percent for the 637 parameter sets. Table 3 shows the results of the various experiments. As can be seen, the passing rate was relatively low, at only about 28 percent, when 30 year average precipitation was used to replace the real 2014 data (Scenario 2, 5, 6, 7, 8, and 13). Furthermore, the median peak size and final epidemic size were significantly lower than baseline. When the 2014 precipitation was used (Scenario 3, 9, 10, 11 and 12), the passing rate was around 65 percent but it was more than 80 percent when the 2013 or 2014 temperature was used. The median peak outbreak size was higher than baseline when 2014’s precipitation and average temperature were combined together (Scenario 3 and 11). The maximum difference between precipitation in 2014 and the 30-year average occurred in May and August, so we scaled the precipitation in these two months to their 30-year average. The passing rates were 65.0, 61.2 and 35.8 percent when we scaled only May, only August and both, respectively (Scenario 16, 17, and 18). When comparing Scenario 19, 20, 21 with 13, higher passing rate and average outbreak size are observed as a result of increasing the rainfall in May and August above the 30-year average. Rainfall in August seems to be slightly more important.
All the results suggest that the precipitation in 2014 played an important role in forming the outbreak, especially rainfall in May and August. However, the temperature in 2014 was lower than average in the spring and winter months, thus acting as a protective factor. That is, if the temperature in 2014 had been higher, the average outbreak size would have been higher as well.
The peak time of daily new cases is clearly sensitive to the date of interventions and the simulation results suggests that the interventions are very effective. The most common interventions in Guangzhou were emptying water containers and ULV spraying of adulticide, both conducted at neighborhood level and organized by neighborhood committee. Emptying water containers reduces the abundance of the immature stage, water level and environmental carrying capacity, thereby reducing adult abundance. ULV spraying of insecticide decreases the abundance of adults almost instantly. With a reduced vector to human ratio, the force of infection decreases while the recovery rate remains the same resulting in an earlier peak. The interventions in 2013 took place every Friday from October 9th to November 10th, while in 2014 on every Friday from September 24th to November 30th, as well as on July 25th, August 15th, September 4th and 28th, and October 8th.
To determine the effectiveness of these interventions, we set interventions in 2014 the same as those in 2013 and recorded the trajectories (Scenario Change intervention dates). The intervention in 2013 took place later and has a much lower repetition frequency. No passes occurred after the changes, because the median peak size and overall outbreak size increased drastically to 21,808 and 843,430 respectively (Fig 6C), approximately 27 times the baseline value and 23 times the actual reported cases in 2014. The new peak time was October 12th, almost 15 days later than the observed peak, which again shows the importance of the time interval between the imported case and the peak.
In addition, the filial infection rate can also change the characteristic of an outbreak. To investigate the importance of vertical transmission, the number of infected eggs (Evi) was set to zero at the beginning of 2014 (Scenario Remove Evi), because this is the only way that the epidemic in 2013 can influence that in 2014. This change resulted in 490 passing simulations out of 637 runs. The median of peak size and outbreak size were 778 and 2,792, respectively, only slightly lower than the baseline (Fig 6D). In addition, when we investigated the role of timing of the imported case and changed β2013 to make the time interval between the import case and peak in 2013 the same as that in 2014, we found that the peak size and outbreak size in 2014 also increased and attributed this to vertical transmission. A scenario was also run with both an advanced import case date in 2013 and no infected eggs carried over from 2013 to 2014 (Scenario Advance 2013 and remove Evi). The peak and outbreak size dropped to 539 and 15,526, respectively, which suggested that the effect of vertical transmission should not be neglected when the outbreak size in the previous year was large (Fig 6E).
From our analyses, four factors appear to have been principally responsible for the pattern of the moderate outbreak in 2013 and the much larger one in 2014, namely the date of the first imported case, unusually high precipitation in 2014, interventions, and vertical transmission. We found the timing of first imported and transmitting case was the dominant feature responsible for this pattern. Furthermore, once the timing of imported case is fixed, climate significantly affects the dengue transmission dynamics. For example, precipitation in May and August, 2014 were found to have a moderate effect on the size of the outbreak, while temperature in 2014 was less favorable for the outbreak and suggests that if the temperature had been higher in the spring and winter months in 2014, the final outbreak size would have been even greater. Vertical transmission played a minor role in forming the pattern, but it is likely to be significant only when the outbreak size in the previous year is large. In addition, we found that the earlier and more frequent interventions in 2014 proved to be effective, otherwise the outbreak size might have been over an order of magnitude higher than the observed value.
The date of imported case was crucial in producing the outbreak pattern in 2013 and 2014. The date of the first imported case in our analysis is not the exact date of the first imported case, but a dummy variable indicating the time of the imported cases which starts the local outbreak. Since we have no information about which imported case will cause local transmission, the time of imported cases was set to be a parameter to be fitted in the model. Though imported cases occurred in almost every month, indigenous cases were mainly reported from July to November when the mosquito abundance and biting rate are higher, and the EIP is shorter. [4] Temperature and arrival date of the first infectious human also interact since early arrival will occur at lower temperature, but there is a longer time for transmission to increase before the beginning of winter season and thereby produce a larger outbreak [15]. That is, low temperature can increase the EIP as well as reduce the biting and the mortality rate resulting in fewer mosquitoes surviving to be infectious as was also shown in Fig 6F. Considering the tradeoff between higher biting rate and longer transmission season, a case imported around mid-April appears to have triggered the biggest outbreak in 2014. (Fig 6F) In addition, the number of imported cases also matters to the outbreak size [16]. However, we did not take this into account, since in our deterministic model one imported case is sufficient to initiate internal transmission.
Precipitation too can have both beneficial and detrimental effects on the abundance of Ae. albopictus and dengue transmission. Ae. albopictus mainly breed in flower pot trays, bamboo tubes, used tyres, disposable containers and surface accumulated water. Precipitation can change the water level in these containers and thereby affect the density-dependent development and death rate [27]. When the water level is higher, the environmental carrying capacity also increases; hence, the maximum number of mosquitoes the environment can support will also increase. Higher water level will also bring down the death rate and increase the development rate, so the survival rate of mosquitoes increases during such periods, and development from larvae to adult will be faster. On the other hand, heavy rainfall also can destroy breeding sites. When a heavy rain occurs at the time the water level is close to the maximum, some of the immature stage mosquitoes will be washed out of their containers (Spillover effect in Fig 2) making container habitats significantly less attractive to ovipositing females. Both mechanisms can cause population loss of Ae. albopictus [58].
In contrast, a study in France suggested that the heavy rainfall events can increase the risk of chikungunya [59]. In fact, the relationship between precipitation and mosquito abundance is complicated. We increased or decreased different amounts of rainfall in 10-day time windows, and ran the model with these new precipitation profiles. The result showed that the relationship between the amount of precipitation and mosquito abundance or the number of dengue cases was nonlinear, and there was no simple rule to predict the effects of rainfall or heavy rainfall.
According to Table 3, the temperature in 2014 was not as important as precipitation in causing the outbreak pattern because the inter-annual change of temperature is much smaller than that of precipitation. However, temperature plays an important role in controlling various aspects of the seasonal population dynamics of Ae. Albopictus as discussed above.
The vertical transmission rate was less important in determining the outbreak pattern in 2013 and 2014, though experiments have confirmed that adults hatched from infected diapause eggs can transmit dengue virus [60]. Our analysis suggested that with the small number of cases in 2013, it is impossible that the big outbreak size in 2014 was caused by only by vertical transmission, therefore dengue was still imported, not endemic, for the 2014 outbreak, which was also recognized by analyzing seasonality and virus source of dengue cases [4]. The probable sources of dengue virus detected in Guangzhou were mainly Thailand, Philippines, Indonesia, Vietnam, Cambodia, and Malaysia [14,61], all of which are also popular tourist destinations for residents in Guangzhou [33]. However, the influence of vertical transmission should not be neglected if a big outbreak occurred in the previous year. Considering the large amount of infected eggs left over from 2014 to 2015, the effect of vertical transmission in 2015 should be large, even can start a local outbreak without any imported case. However, there is no big outbreak in 2015, with 44 imported cases but only 57 indigenous cases, though the precipitation in May was higher and there are more imported cases than 2014. This is likely to be attributable to the extensive interventions in 2015. After the unprecedented outbreak in 2014, the government paid more attention to early detection of imported cases, early mosquito control (started in April compare with in the end of July, 2014), and the quarantine of every suspicious case. Moreover, residents in Guangzhou have more knowledge about the difference between dengue and influenza after 2014, so they are more likely to go to the hospital when symptoms occur and will be put quarantined immediately after confirmed, which can also reduce local transmission. However, due to the deterministic nature of this model, its use is only appropriate when the scale of the outbreak was big enough to ignore the stochastic effects, but the outbreak size in 2015 was relatively small, therefore we did not simulate the situation in 2015 here.
Currently, there is no effective commercial dengue virus vaccine available. Thus, the prevention of a dengue outbreak relies heavily on vector control. Container emptying and ULV spraying are the most common control strategies in China. Other approaches such as releasing Wolbachia infected male Ae. albopicus and introducing mosquito larvae-eating fish have also been adopted, though to a much smaller extent. Although the efficiency of ULV spray in controlling adult Ae. albopictus has been questioned over the years [24,62], larval source reduction has proven to be successful [62]. Since Guangzhou applied both strategies at the same time, we could not separate them. In addition, though it was argued that mosquito control strategies were often implemented after the peak of transmission and had little or no impact on dengue transmission [62], the first intervention in Guangzhou was timely, two months earlier than the peak, and does appear to have reduced the final epidemic size significantly.
Some studies have suggested positive associations between dengue incidence and the Aedes household index and the BI, [63,64] while others have concluded that there was no significant correlation [65,66]. In our study, we found that the abundance of Ae. albopictus was almost the same for 2013 and 2014 (Fig 5), and there is no relationship between dengue incidence and the mosquito index for Guangzhou in this specific outbreak. However, on the other hand, according to our results when manipulating the climate files, the abundance of mosquitoes can affect the transmission dynamics, though does not appear to be the most important reason for the large 2014 outbreak.
There are, of course, various limitations to our analysis particularly for use in the future. For example, the whole population was considered to be susceptible in 2012 since dengue is not a common disease in Guangzhou. There were 12.70 million people at the beginning of 2012 and only 2,381 cases were reported between 2002 and 2011. In addition, there may be transmission of other serotypes in the future, only one serotype was included in the model because most of the cases have been DENV-1 in recent years. Another limitation of the study may be that the temperature dependent functions employed in the model were based on experiments which were conducted under constant temperature conditions [42,44,67]. Temperature changes from day to day as well as the diurnal temperature range can also change the transmission dynamics [68,69]. The most significant limitation, however, may be that our model does not take spatial effects into account. Further steps should be taken to develop a spatially-explicit individual based model, and to include the spatial heterogeneity and stochasticity of transmission of dengue in Guangzhou. With a stochastic model, we can learn more about the probability of local transmission, which can be combined with the outbreak scale to give us a more practical estimation of the dengue outbreak risk.
With the spread of Ae. albopictus under global warming and increasing numbers of international travelers, dengue poses additional challenges to policymakers, especially when taking into account the antibody-dependent enhancement, which can lead to increased viral replication and higher viral loads [70] when infected by another heterologous strain. A second wave outbreak with a different serotype could bring more serious manifestations of dengue fever like DHF or DSS [71]. Sustained efforts should be taken to control mosquito abundance and to prevent or limit the extent of further outbreaks.
|
10.1371/journal.pgen.1002877 | The Burkholderia bcpAIOB Genes Define Unique Classes of Two-Partner Secretion and Contact Dependent Growth Inhibition Systems | Microbes have evolved many strategies to adapt to changes in environmental conditions and population structures, including cooperation and competition. One apparently competitive mechanism is contact dependent growth inhibition (CDI). Identified in Escherichia coli, CDI is mediated by Two–Partner Secretion (TPS) pathway proteins, CdiA and CdiB. Upon cell contact, the toxic C-terminus of the TpsA family member CdiA, called the CdiA-CT, inhibits the growth of CDI− bacteria. CDI+ bacteria are protected from autoinhibition by an immunity protein, CdiI. Bioinformatic analyses indicate that CDI systems are widespread amongst α, β, and γ proteobacteria and that the CdiA-CTs and CdiI proteins are highly variable. CdiI proteins protect against CDI in an allele-specific manner. Here we identify predicted CDI system-encoding loci in species of Burkholderia, Ralstonia and Cupriavidus, named bcpAIOB, that are distinguished from previously-described CDI systems by gene order and the presence of a small ORF, bcpO, located 5′ to the gene encoding the TpsB family member. A requirement for bcpO in function of BcpA (the TpsA family member) was demonstrated, indicating that bcpAIOB define a novel class of TPS system. Using fluorescence microscopy and flow cytometry, we show that these genes are expressed in a probabilistic manner during culture of Burkholderia thailandensis in liquid medium. The bcpAIOB genes and extracellular DNA were required for autoaggregation and adherence to an abiotic surface, suggesting that CDI is required for biofilm formation, an activity not previously attributed to CDI. By contrast to what has been observed in E. coli, the B. thailandensis bcpAIOB genes only mediated interbacterial competition on a solid surface. Competition occurred in a defined spatiotemporal manner and was abrogated by allele-specific immunity. Our data indicate that the bcpAIOB genes encode distinct classes of CDI and TPS systems that appear to function in sociomicrobiological community development.
| Contact dependent growth inhibition (CDI) is a phenomenon discovered in Escherichia coli in which CDI+ bacteria inhibit the growth of CDI− bacteria upon cell-to-cell contact. CDI is mediated by large toxic “exoproteins” present on the bacterial cell surface. An ‘immunity’ protein protects CDI+ cells from killing themselves. While predicted CDI systems are widespread throughout bacterial genera, the role of these systems in nature has remained elusive. Here we identify a distinct class of CDI system in Burkholderia species. The genes encoding these systems are expressed in a stochastic manner such that only a few cells in the population produce the proteins at any given time when grown in broth. We also show that these systems are required for aggregation on an abiotic surface, suggesting an important role for CDI in biofilm formation. Finally, we show that CDI mediates competition under specific conditions in a precise spatiotemporal pattern when bacteria are grown on a solid surface. Our data suggest that in nature, CDI systems may be used by bacteria to establish complex sociomicrobial communities.
| Microbes rarely live alone. Whether free in the environment or in close association with eukaryotic hosts, microbes typically share their living space with other viral, prokaryotic, and/or eukaryotic microorganisms. Survival under these conditions requires mechanisms for sensing, responding to, and cooperating or competing with neighboring organisms. Contact dependent growth inhibition (CDI) systems are protein toxin delivery mechanisms that appear to be involved in interbacterial competition [1]. CDI was discovered in Escherichia coli strain EC93 due to its ability to inhibit the growth of specific CDI− E. coli strains upon cell-to-cell contact. CDI is mediated by Two–Partner Secretion (TPS) system proteins CdiA and CdiB [1]. TPS systems are widespread amongst Gram-negative bacteria. They export large exoproteins (TpsA family members such as CdiA) across the outer membrane using pore-forming β-barrel proteins (TpsB family members such as CdiB) [2], [3]. Functions attributed to TpsA proteins before the discovery of CDI included adherence to eukaryotic cells, induction of cytolysis in host cells, iron uptake, and autoaggregation [2], [3]. Characterization of CDI in E. coli revealed an additional TpsA-mediated function: inhibition of ‘target’ bacterial cell growth upon contact. CDI+ bacteria are protected from autoinhibition because they produce CdiI, a 79 amino acid ‘immunity’ protein encoded immediately 3′ to cdiA [1].
Research in our lab on Burkholderia pseudomallei led to the discoveries that genes predicted to encode CDI systems are present in a large number of α-, β-, and γ-proteobacteria, that the C-terminal ∼300 aa of CdiA proteins (CdiA-CTs) and CdiI proteins are highly variable, and that CdiA-CTs are sufficient to confer toxicity when produced intracellularly in E. coli [4]. Some CdiA-CTs have been demonstrated to possess nuclease activity, functioning as DNases or tRNases [4], [5]. CdiI proteins bind to cognate CdiA-CT proteins (those encoded by the same cdi locus), blocking their nuclease activity, but not heterologous CdiA-CT proteins (those encoded by different cdi loci) [4], [5]. Experiments with E. coli strains producing chimeric CdiA proteins showed that CdiI proteins provide immunity against interbacterial growth inhibition in an allele-specific manner, conferring protection only towards cognate CdiA-CTs but not heterologous CdiA-CTs [4]. Although chimeric CdiA proteins containing CdiA-CTs encoded by different species of bacteria were effective at mediating interbacterial competition in E. coli, CDI has so far only been shown to function between members of the same species [1], [4]. The current model for CDI states that, upon cell-to-cell contact with a closely related bacterium (possibly by interacting directly with the outer membrane protein BamA [6]), the CdiA-CT from a CDI+ cell is delivered to the cytoplasm of the target cell where it inhibits cell growth by degrading DNA or specific tRNAs. If present in the target cell, the cognate CdiI immunity protein binds to the CdiA-CT, blocking its nuclease activity [4]. While considerable insight has been gained regarding how CDI systems function mechanistically in E. coli, almost nothing is known about when or why these systems are deployed by bacteria in nature.
Initial bioinformatic analysis of available bacterial genomes revealed that putative CDI systems fall into two distinct classes, “E. coli-type,” which include systems found in bacterial genera other than Burkholderia, and “Burkholderia-type,” found only in Burkholderia spp [4]. E. coli-type CDI systems are encoded by genes with the order cdiBAI and predicted CdiA proteins contain a highly conserved VENN motif that separates the conserved (∼2700 aa) N-terminus from the variable (∼300 aa) C-terminus (the CdiA-CT). Burkholderia-type CDI systems are encoded by genes with the order cdiAIB and putative CdiA proteins contain an NxxLYN motif instead of VENN [4]. Whether the Burkholderia proteins actually function as CDI systems has not yet been demonstrated.
Burkholderia spp are Gram-negative soil saprophytes and many are opportunistic pathogens [7], [8], [9]. Burkholderia cepacia complex (Bcc) strains, for example, cause life-threatening respiratory infections in cystic fibrosis patients [10], [11], and B. pseudomallei strains cause melioidosis, a disease that can range from localized wound infections and abscesses to fulminant pneumonia and septicemia [9], [12]. Because it is highly virulent by the aerosol route, resistant to most commonly used antibiotics, and extremely closely related to Burkholderia mallei, which has been used as a bioterrorism agent in the past, B. pseudomallei is an NIAID Category B Biothreat pathogen and select agent [13]. Working with B. pseudomallei in the laboratory requires BSL-3 practices and rigorous security measures. Burkholderia thailandensis is closely related to B. pseudomallei and occupies the same environmental niche (both are endemic to southeast Asia and northern Australia) [14], [15], [16], but is not a human pathogen, is not a select agent, and requires only BSL-1 practices.
Here, we characterize the unique class of CDI systems produced by Burkholderia spp. We show that these systems compose a novel class of TPS system that requires a third protein for the large exoprotein to function, that expression of Burkholderia CDI protein-encoding genes is regulated in a probabilistic manner, that the gene products contribute to biofilm formation, and that CDI-mediated interbacterial competition in Burkholderia occurs on solid surfaces in a unique temporal and spatial pattern.
To investigate the operon structure of the bcp locus in B. thailandensis strain E264, reverse transcriptase (RT) PCR was performed on RNA extracted from bacteria cultured in low salt LB broth (LSLB), the standard medium used for culturing B. thailandensis. Primer sets (Table S3) flanking the junctions between bcpA and bcpI, bcpI and bcpO, and bcpO and bcpB (1, 2, and 3, respectively, in Figure 2C) yielded products of the expected sizes (Figure 2A, top panel), indicating the bcpAIOB genes form an operon. RT-PCR was performed to identify the approximate transcription start site using forward primers annealing 5′ to bcpA with an internal bcpA reverse primer. Products of the expected size were obtained for forward primers annealing 50, 70, 120, and 150 nt 5′ to bcpA (Figure 2A, middle panel), indicating the transcription start site for bcpA is at least 150 nt 5′ of the translation start site. Faint products were obtained for forward primers annealing 200 and 250 nt 5′ to bcpA (Figure 2A, bottom panel), but no product was obtained with the forward primer annealing 300 nt 5′ to bcpA, suggesting the possibility of a second promoter located 250–300 nt 5′ to the translation start site.
To measure expression of bcpAIOB, we constructed a strain containing a bcpA promoter-lacZ fusion (PbcpA-lacZ) inserted at the attTn7 site on the chromosome in B. thailandensis E264 and measured β-galactosidase activity in cells cultured under various conditions. For comparison, we also constructed two additional strains, one in which the promoter of the gene encoding the ribosomal S12 subunit (PS12) was fused to lacZ and one with no promoter (Pneg) 5′ to lacZ. Approximately 2,000 Miller units of β-galactosidase activity were produced in the PbcpA-lacZ fusion strain cultured under all conditions tested (Figure 2B). The PS12-lacZ fusion produced ∼11,000 Miller units of β-galactosidase activity and the Pneg-lacZ fusion produced ∼500 Miller units of β-galactosidase activity in cells cultured in LSLB broth. These data suggest bcpA is transcribed at relatively low levels under each of the culture conditions tested.
We next constructed strains with in-frame deletion mutations in genes in the bcpAIOB operon. While it was possible to construct a ΔbcpO strain and a ΔbcpB strain by allelic exchange, and a strain in which the entire bcpAIOB operon was replaced with a gene encoding kanamycin resistance by natural transformation (Figure 2C), it was not possible to construct a ΔbcpI strain, suggesting bcpI is essential or, analogous to the E. coli CDI system, BcpI is required to protect against BcpA-mediated toxicity. For complementation experiments, we constructed plasmids to deliver bcpO or bcpB, driven by the constitutively active PS12 promoter or the native promoter (PbcpA) to the attTn7 site. All mutant strains grew equally compared to wild type E264 in LSLB medium (data not shown).
Initial attempts to investigate the contribution of bcpO and bcpB to BcpA production were carried out by performing Western blots from cell lysates of bacteria expressing the bcpAIOB genes from their native promoter, PbcpA, and producing BcpA containing a hemagglutinin (HA) epitope (BcpA-HA) N-terminal to the predicted Nx(E/Q)LYN sequence of the mature protein (i.e., immediately C-terminal to F2633, 141 amino acids from the N-terminal side of the Nx(E/Q)LYN sequence). We were unable to detect BcpA in this strain, possibly because bcpAIOB expression was insufficient under the growth conditions used. We therefore constructed B. thailandensis strains in which the bcpAIOB operon was controlled by the constitutively active ribosomal S12 subunit promoter, PS12. In immunoblots of whole cell lysates of otherwise wild type B. thailandensis (E264BcpA-HA::pECG22), anti-HA antibodies recognized a polypeptide with considerably slower mobility than the 250 kDa molecular weight marker (possibly corresponding to 306 kDa, the predicted molecular mass of BcpA), plus three slightly smaller and much less abundant polypeptides (Figure 3). The polypeptide profile detected in the ΔbcpO strain was identical to that of E264BcpA-HA::pECG22. By contrast, only a very low level of the largest polypeptide was detected in whole cell lysates of the ΔbcpB strain (E264BcpA-HAΔbcpB::pECG22) (Figure 3). In other TPS systems, the TpsA protein is undetectable when the strain contains a loss-of-function mutation in the TpsB-encoding gene, presumably because the TpsA protein, which cannot be translocated across the outer membrane, is degraded in the periplasm [18], [19]. Our data suggest that the same is true for bcpA and bcpB. Complementation of the ΔbcpO and ΔbcpB strains with bcpO and bcpB, respectively, expressed from the PS12 promoter did not alter the polypeptide profiles of these strains (Figure 3). However, the fact that BcpA-HA protein was detectable in the ΔbcpO strain and not in the ΔbcpB strain indicates that the ΔbcpO mutation does not have polar effects that abrogate expression of bcpB. Similarly, RT-PCR indicated that bcpB transcription was not abrogated by the in-frame deletion in the ΔbcpO strain (Figure S2).
Upon plating the PbcpA-lacZ fusion strain on solid medium containing X-gal, we found that the colonies were not all the same intensity blue; approximately 5–10% of the colonies were dark blue and 90–95% of the colonies were light blue (Figure 4A). To explore the possibility that bcpA was highly expressed in just a small proportion of cells in the population, we constructed fluorescent reporter fusion strains by delivering a gfp gene fused to PbcpA or PS12 to the attTn7 chromosomal insertion site in B. thailandensis E264. Bacteria were cultured in liquid broth and visualized by confocal microscopy. All bacteria containing the PS12-gfp fusion produced high levels of GFP (Figure 4B). By contrast, only a few fluorescent bacteria were present in cultures containing the PbcpA-gfp fusion. These data suggest bcpA is differentially expressed within the bacterial population.
We next performed flow cytometry to measure bcpA-gfp expression in a large number of bacterial cells when cultured in liquid medium. Events distinct from the PBS control (Figure S3) were identified as bacteria and gated on for subsequent analysis. Approximately 99% of the bacteria in cultures of the PS12-gfp fusion strain were GFP+ (Figure 4C, Table S2), whereas only ∼0.2% of the PbcpA-gfp fusion containing bacteria were GFP+, indicating that bcpA is differentially expressed within a population of bacteria when cultured in liquid medium. The mean relative fluorescence intensity of the PbcpA-gfp GFP+ cells was similar to that of PS12-gfp GFP+ cells (Table S2), indicating expression was very high in the GFP+ PbcpA-gfp bacteria. Taken together, these data show that only a small percentage of bacteria express bcpA when cultured in liquid medium, but those that do express bcpA do so at a high level.
Culturing wild type B. thailandensis in M63 minimal medium resulted in a dramatic autoaggregation phenotype in which bacteria aggregated and adhered to the walls of glass test tubes (Figure 5A). By contrast, the ΔbcpAIOB strain grew as a homogenous suspension of planktonic cells (Figure 5A), indicating that the bcpAIOB genes are required for autoaggregation. The ΔbcpB and ΔbcpO strains also grew planktonically (Figure 5A). Because TpsB family members are required for secretion of TpsA proteins to the bacterial surface and Western blot data confirmed that no BcpA protein could be detected in the ΔbcpB strain (Figure 3), the ΔbcpB strain was expected to have the same phenotype as the ΔbcpAIOB strain. The fact that the ΔbcpO strain failed to autoaggregate, however, indicates that the BcpO protein is required for BcpA function. Because the BcpA protein profile in the ΔbcpO mutant was identical to that of wild type bacteria, BcpO likely contributes to maturation events that occur during or after translocation across the outer membrane. Because CDI systems have so far been demonstrated to function only in interbacterial growth inhibition between CDI+ and CDI− bacteria, these data represent the first demonstration of a phenotype for a strain defective for expression of genes encoding a (putative) CDI system in a homogeneous culture.
Efforts to complement the bcpO and bcpB genes were not successful in restoring the wild type autoaggregation phenotype. Specifically, expression of bcpO and bcpB from the PS12 promoter or the native promoter, PbcpA, at the attTn7 site failed to restore autoaggregation in the ΔbcpO and ΔbcpB mutants, respectively (data not shown). To address the possibility that lack of autoaggregation in the ΔbcpO mutant was due to an unintended mutation other than the ΔbcpO mutation, we tested several independently constructed ΔbcpO mutant strains. All of these strains grew planktonically and failed to autoaggregate. Together, these data indicate that the bcpAIOB genes are required for autoaggregation, and they suggest that expression of the bcpO and bcpB genes from their native locus is critical for proper function of their gene products.
A strain in which expression of the bcpAIOB genes was controlled by the constitutively active PS12 promoter (PS12-bcpAIOB) also autoaggregated when cultured in M63 minimal medium, but with different kinetics and aggregation characteristics compared with wild type B. thailandensis. The PS12-bcpAIOB strain aggregated but was not adherent to the walls of the test tube at 24 hours, and by 48 hours, the adherent/aggregated bacteria had a smoother, more mucoid appearance (Figure 5B). These data indicate that controlled expression of the bcpAIOB genes (high in approximately 0.2% of the population and undetectable in the rest) is important for the wild type autoaggregation phenotype.
Autoaggregation and adherence to the walls of test tubes by B. thailandensis may be a form of biofilm. Because DNA has been shown to be an important component of many bacterial biofilms [20], we sought to determine if DNA contributed to the B. thailandensis autoaggregation phenotype. Addition of 4 U DNaseI decreased the amount of autoaggregation and addition of 10 U DNaseI completely abrogated autoaggregation (Figure 5A), causing the bacteria to grow planktonically. Extracellular DNA is therefore required for the autoaggregation phenotype. To quantify the amount of extracellular DNA in wild type cultures compared to ΔbcpAIOB cultures, supernatants were filter sterilized and analyzed by spectrophotometry. No difference in the quantity of DNA could be detected (data not shown), suggesting there may be variations in the quality of DNA present in the two cultures that mediate autoaggregation and/or that DNA–BcpA interactions are required for autoaggregation.
Our inability to construct an E. coli strain producing the C-terminal ∼350 aa of BcpA from B. pseudomallei 1026b unless the cognate BcpI protein was also produced provided the first evidence that the C-terminal domains of CdiA/BcpA proteins are sufficient to cause toxicity when produced intracellularly [4]. Here, we constructed plasmids to encode the last ∼350 aa (including the Nx(E/Q)LYN motif) of BcpA from B. pseudomallei K96243 (BcpA-CTBpK96243) and B. pseudomallei 1106A-2 (BcpA-CTBp1106A-2) (with an added ATG at the 5′ end) 3′ to the rhamnose inducible promoter PrhaB. Overnight cultures of B. thailandensis containing these plasmids were supplemented with 0.2% glucose (to suppress PrhaB) and diluted into LSLB containing either 0.2% glucose or 0.2% rhamnose. Bacterial viability was monitored after four hours of culture at 37°C by counting colony forming units (cfu). The number of cfu/ml of B. thailandensis strains containing these plasmids cultured in medium containing 0.2% glucose was not altered after four hours (Figure 6A). By contrast, when cultured with 0.2% rhamnose to induce PrhaB, the number of cfu/ml of B. thailandensis strains containing the plasmids encoding BcpA-CTBpK96243 and BcpA-CTBp1106A-2 decreased by 3 and 2 logs, respectively (Figure 6A, left and right panels, respectively). These results indicate BcpA-CTs are sufficient to cause toxicity when produced intracellularly in B. thailandensis.
We next investigated the ability of BcpI to provide immunity to BcpA-CT-mediated intracellular toxicity. We constructed another set of plasmids encoding BcpI proteins from B. pseudomallei K96243 (BcpIBpK96243) and B. pseudomallei 1106A-2 (BcpIBp1106A-2) also under control of PrhaB. B. thailandensis containing these plasmids in combination with the BcpA-CT-encoding plasmids were cultured as described above. Again, bacterial viability was monitored after four hours. The number of cfu/ml of B. thailandensis harboring the plasmids encoding cognate BcpA-CTBpK96243 and BcpIBpK96243 increased 0.5 log when cultured with 0.2% glucose and decreased only slightly when cultured with 0.2% rhamnose (Figure 6B, left panel). Similarly, the number of cfu/ml of B. thailandensis harboring the plasmids encoding cognate BcpA-CTBp1106A-2 and BcpIBp1106A-2 decreased slightly when cultured with 0.2% glucose and decreased less than 1 log when cultured with 0.2% rhamnose (Figure 6B, right panel). While the decrease in cfu/ml of B. thailandensis containing the plasmids encoding BcpA-CTBp1106A-2 and BcpIBp1106A-2 observed when cultured with 0.2% rhamnose was statistically significant (p<0.006), the log fold change of this strain compared to the log fold change of B. thailandensis harboring the BcpA-CTBp1106A-2 encoding plasmid alone was also statistically significant (p<0.0001) when cultured with 0.2% rhamnose. These data indicate cognate BcpI proteins are able to rescue the toxic phenotypes observed when BcpA-CTs are produced intracellularly. By contrast, the number of cfu/ml of B. thailandensis harboring plasmids encoding the non-cognate pair BcpA-CTBpK96243 and BcpIBp1106A-2 decreased by 3 logs when cultured with 0.2% rhamnose, but did not change when cultured with 0.2% glucose (Figure 6C, left panel); and B. thailandensis containing plasmids encoding the non-cognate pair BcpA-CTBp1106A-2 and BcpIBpK96243 decreased by 2 logs when cultured with 0.2% rhamnose, but did not change when cultured with 0.2% glucose (Figure 6C, right panel). These data indicate BcpI proteins mediate immunity to BcpA-CT intracellular toxicity in an allele-specific manner.
Previous work with E. coli demonstrated that E. coli-type CDI systems function in interbacterial competition in liquid medium when the cdiBAI genes are expressed from a constitutive or inducible promoter [1], [4]. Additionally, it was demonstrated that expression of the cognate immunity gene in target bacteria was protective against CDI [4]. To determine if the bcpAIOB genes also mediate interbacterial competition and if bcpI provides protection, similar competition assays were performed. Wild type inhibitor (E264CmR) and ΔbcpAIOB mutant target bacteria were mixed at a 1∶1 ratio and cultured in LSLB broth at 37°C for 24 hours without antibiotic selection. Wild type bacteria were also cultured at a 1∶1 ratio with ΔbcpAIOB bacteria constitutively expressing the cognate immunity gene (E264ΔbcpAIOB::bcpIE264) under the same conditions. After 24 hours, the cultures were serially diluted and plated on selective media to determine the number of cfu of each strain. In the competition between wild type and ΔbcpAIOB mutant bacteria, both strains grew equally (Figure 7A), and therefore no competitive advantage was observed for the wild type strain in this assay. Both strains also grew equally in the competition between wild type and ΔbcpAIOB::bcpIE264 bacteria (Figure 7B). Since our expression data indicated that B. thailandensis bcpAIOB genes are expressed in only about one in one thousand wild type bacterial cells cultured in liquid medium (Figure 4), we hypothesized two non-mutually exclusive reasons for the lack of apparent competition between wild type bacteria and the ΔbcpAIOB mutant. 1) In addition to inhibiting ΔbcpAIOB mutant bacteria, the wild type bacteria expressing their bcpAIOB genes might also inhibit the growth of wild type bacteria not expressing their bcpAIOB genes because these bacteria would not be producing BcpI, and hence both wild type and ΔbcpAIOB bacteria would be inhibited to the same extent. 2) Regardless of the susceptibility of wild type bacteria to growth inhibition by other wild type bacteria, the number of ΔbcpAIOB mutant bacteria inhibited by wild type bacteria may be insignificant because of the low number of wild type bacteria expressing their bcpAIOB genes.
To test the first hypothesis, we constructed a strain constitutively expressing the cognate bcpI gene (from B. thailandensis E264) in a wild type E264 background (E264::bcpIE264). Constitutive expression of bcpIE264 in the wild type strain did not alter the results of the competition assay; both strains again grew equally (Figure 7C), suggesting that wild type bacteria expressing bcpAIOB were not inhibiting wild type bacteria that were not expressing bcpAIOB to an appreciable level. To test the second hypothesis, we performed a competition experiment with the strain expressing bcpAIOB from the PS12 promoter (PS12-bcpAIOB) and ΔbcpAIOB bacteria. Again, both strains grew equally (Figure 7D), suggesting expression of bcpAIOB in every wild type bacterium does not lead to growth inhibition of mutant target bacteria in liquid medium. Collectively, these data indicate that expression of bcpAIOB from either the native promoter or a constitutively active promoter (in single copy on the chromosome) is not sufficient to cause interbacterial competition against ΔbcpAIOB target bacteria when mixed at a 1∶1 ratio in liquid medium.
To determine the contribution of bcpO to BcpAIOB-mediated CDI, we conducted a competition between ΔbcpO bacteria and ΔbcpAIOB bacteria. In the center of the colony biofilm, ΔbcpO bacteria outcompeted ΔbcpAIOB bacteria by ∼1 log at 24 hours (Figure 11, top panel, column I) and along the leading edge of the colony biofilm, ΔbcpO bacteria outcompeted ΔbcpAIOB bacteria by ∼1.5 logs at 24 hours (Figure 11, bottom panel, column I). For both locations, competition by the ΔbcpO strain was severely reduced compared to competition by wild type bacteria (Table 2). To determine if the small competitive advantage displayed by the ΔbcpO strain was in fact due to CDI, we competed ΔbcpO bacteria with ΔbcpAIOB mutant bacteria constitutively expressing the cognate bcpI gene (E264ΔbcpAIOB::bcpIE264) or a heterologous bcpI gene (E264ΔbcpAIOB::bcpIK96243). The competitive index for competition between the ΔbcpO strain and E264ΔbcpAIOB::bcpIE264 was zero in the center and along the leading edge of the colony biofilm at 24 hours (Figure 11, column II), indicating constitutive expression of bcpIE264 in ΔbcpAIOB mutant bacteria is protective against CDI by the ΔbcpO strain. However, constitutive expression of bcpIK96243 in ΔbcpAIOB mutant bacteria was not protective, as ΔbcpO bacteria outcompeted E264ΔbcpAIOB::bcpIK96243 bacteria by ∼1 log in the center and ∼2 logs along the leading edge of the colony biofilm at 24 hours (Figure 11, column III). Complemention of the ΔbcpO strain with a copy of bcpO expressed constitutively (E264ΔbcpO::bcpO) only slightly increased CDI activity in the center of the colony biofilm but fully restored activity at the edge, i.e. ΔbcpAIOB bacteria were completely outcompeted in this location (Figure 11, column IV, Table 2). Lack of restoration of autoaggregation and partial restoration of CDI by the bcpO complementation strain underscores the complexity of the bcpAIOB system and the role of BcpO in interbacterial CDI.
Together, our data indicate that the ΔbcpO strain has a greater than ten–fold defect in CDI-mediated interbacterial competition in the center of the colony biofilm and at least a thousand–fold defect along the leading edge compared to wild type bacteria after 24 hours (Table 2). BcpO, therefore, plays a substantial and critical role in CDI-mediated interbacterial competition in B. thailandensis.
CDI has so far been demonstrated only in E. coli, the species in which it was discovered, and Dickeya dadantii, a phytopathogenic bacterium that infects a variety of crop plants [1], [4]. Although genetic loci in six Burkholderia strains were predicted to encode CDI systems based on the presence of small (∼300 bp) ORFs immediately 3′ to genes predicted to encode TpsA proteins, the gene order in these loci was different than that of the E. coli cdiAIB genes (and all other putative CDI system-encoding loci) and the motif separating the conserved and variable regions of the predicted TpsA protein was NxxLYN rather than VENN [4]. We showed in this study that the Burkholderia bcpAIOB genes do in fact encode proteins that function in many ways like the CDI system of E. coli; the BcpA-CTs are toxic when produced intracellularly, the bcpI genes confer immunity in an allele-specific manner, and wild type B. thailandensis can outcompete a ΔbcpAIOB mutant if the mutant does not express the cognate immunity gene. B. thailandensis is therefore the third species in which CDI has been demonstrated. Moreover, together with the previous observations, our results indicate that Burkholderia bcpAIOB genes define novel classes of both CDI and TPS systems. We also showed in this study that the Burkholderia bcpAIOB genes are required for autoaggregation and adherence to an abiotic surface, and that they are expressed in a probabilistic manner when the bacteria are cultured in liquid medium, two phenotypes not previously ascribed to CDI or CDI system-encoding genes.
The TPS pathway is one of the simplest mechanisms for the secretion of proteins to the surface of Gram-negative bacteria. The paradigm, based primarily on studies of the FHA/FhaC proteins of Bordetella species and the HMW1/HMW1B proteins of Haemophilus influenzae, states that the large β-helical exoprotein (the TpsA family member) is translocated across the cytoplasmic membrane by the general Sec pathway and then requires only one protein, the TpsB family member, for translocation across the outer membrane [2], [3]. Experimental support for the sufficiency of the TpsB protein in outer membrane translocation of the TpsA protein was recently obtained in a study using purified FhaC, liposomes, and a polypeptide corresponding to the N-terminal 370 aa of FHA [21]. Our bioinformatic analysis identified a small ORF located 5′ to bcpB in most Burkholderia-type putative CDI protein-encoding loci, which we named bcpO. BcpO of B. thailandensis E264 is predicted to be a lipoprotein that localizes to the inner leaflet of the outer membrane and deletion of bcpO resulted in loss of autoaggregation (identical to deletion of the entire bcpAIOB operon) and significantly reduced CDI activity. BcpA appeared to be produced and exported across the outer membrane in the bcpO mutant, based on the fact that BcpA was not degraded, as assessed by immunoblot, and was capable of mediating a very low level of CDI. Unfortunately, our attempts to visualize BcpA on the surface of wild type and ΔbcpO mutant bacteria were unsuccessful (data not shown). Based on the phenotypes of the bcpO mutant and the predicted cellular location of BcpO, we hypothesize that BcpO is involved in efficient, export across the outer membrane, maturation of BcpA into a functional protein, release of BcpA from the cell surface, and/or sensing interbacterial interactions. Regardless of its role, our data indicate that the Burkholderia BcpAIOB proteins define a novel class of TPS system that requires an additional small protein, BcpO, to produce a fully functional TpsA protein.
TpsB proteins are members of the Omp85-TpsB superfamily, which includes BamA (the main component of the Bam complex that inserts β-barrel proteins into the outer membranes of Gram-negative bacteria), Tob55/Sam50 (which inserts proteins into the outer membranes of mitochondria), and Toc75 (which inserts proteins into the outer membranes of chloroplasts) [22]. In addition to BamA, the Bam complex contains four lipoproteins, BamB, C, D, and E, that localize to the inner leaflet of the outer membrane and play important but mostly unknown roles in outer membrane protein assembly [22]. The BcpAIOB system of B. thailandensis E264 may function in an analogous manner to the Bam complex, as it appears to require at least one predicted periplasmic lipoprotein for proper secretion or maturation of its substrate. Curiously, some classes of predicted CDI systems in Burkholderia spp contain BcpO proteins that do not have signal sequences and that vary in an allele-specific manner with their cognate BcpA and BcpI proteins. Whether these proteins function similarly to BcpO of B. thailandensis E264 or perform completely different functions is unknown. These systems may represent yet additional variation of the TPS and CDI paradigms.
Our PbcpA-gfp studies showed that when B. thailandensis is cultured in liquid medium (either LSLB or M63), expression of bcpAIOB is high in approximately 0.2% of the bacteria and undetectable in the rest. The only other strain for which expression of CDI system-encoding genes has been demonstrated is E. coli EC93, which, in contrast to other strains that have been investigated, appears to express the cdiBAI genes constitutively [4]. Our result suggests that, in liquid medium under the conditions tested, bcpAIOB gene expression is controlled in a probabilistic manner. Several cellular differentiation processes in organisms ranging from bacteria to humans are controlled probabilistically and transiently [23], [24], [25], one of the best understood being the development of competence in Bacillus subtilis. When starved for nutrients, a minority of B. subtilis cells in a population express genes required for DNA uptake (competence), while the rest commit to sporulation [26]. Regulation of competence genes in B. subtilis involves an excitable core module containing both positive and negative feedback loops [27]. We expect that equally complex regulatory circuits control bcpAIOB expression in Burkholderia, and our future experiments will be aimed at identifying and characterizing the systems involved.
When cultured in M63 minimal medium, wild type B. thailandensis aggregated and adhered to the walls of glass test tubes. This phenotype required BcpAIOB as the ΔbcpAIOB, ΔbcpO, and ΔbcpB mutants grew planktonically. This is the first demonstration of a phenotype for any CDI− strain other than susceptibility to growth inhibition by CDI+ counterpart bacteria. Although we did not formally test for biofilm formation, Schwarz et al. showed that B. thailandensis forms a biofilm in a flow chamber model [28]. Our data suggest that the BcpAIOB proteins contribute to biofilm formation and therefore their true role in nature may not be (just) mediating interbacterial competition. Interestingly, the conditions that promote B. thailandensis autoaggregation are the same conditions in which expression of the bcpAIOB genes occurs in only about 0.2% of the bacterial cells. Treating cultures with DNaseI abrogated autoaggregation, suggesting that DNA may be an essential component of an extracellular matrix that holds aggregated cells together. This observation suggested the intriguing hypothesis that CDI contributes to biofilm formation by killing a small proportion of cells in the population such that their released DNA can be used for extracellular matrix formation. However, the strain expressing bcpAIOB from the strong, constitutive PS12 promoter also aggregated, albeit with different kinetics. In this population, all cells should produce BcpI and therefore should be immune to CDI, although we cannot rule out the possibility that strong, constitutive expression of the bcpAIOB locus alters the functionality of the individual gene products. Moreover, we detected extracellular DNA in cultures of both wild type and ΔbcpAIOB bacteria, indicating that extracellular DNA is required but not sufficient for autoaggregation. These observations underscore the complexity of the aggregation/biofilm phenotype and indicate that determining the role of BcpAIOB in this microbial lifestyle will require an extensive amount of additional investigation.
By contrast to what was observed for B. thailandensis cultured in liquid, BcpAIOB-dependent interbacterial competition was easily observed when the bacteria were grown on a solid surface. Within the colony biofilm, wild type bacteria outcompeted ΔbcpAIOB mutants by ∼2.5 logs in the center and completely (≥4.4 logs) at the edge after 24 hours of co-incubation. Examination of earlier time points indicated that competition occurred as early as 12 hours in the center and six hours at the edges. Live-cell imaging provided an explanation for the difference in competition at the two sites; bacteria in the center of the colony biofilm were just beginning to contact each other between 6 and 12 hours, while bacteria at the edges were in contact immediately upon plating. The efficiency of the competition when bacteria were in contact however, suggested that the bcpAIOB genes must be expressed in more than ∼0.2% of wild type bacteria under these conditions. The fact that bacteria transitioned from coccobacilli to long rods after being moved from stationary phase growth in broth to the solid surface provided evidence for a change in gene expression, but we were unsuccessful in our attempts to detect GFP+ bacteria when the PbcpA-gfp strain was used to form colony biofilms. However, the strain expressing bcpAIOB from the PS12 promoter was able to outcompete the ΔbcpAIOB strain when the colony biofilms were initiated with the bacteria present at a 1∶1 ratio but not when present at a 1∶1,000 ratio. Together, these data suggest that expression of the bcpAIOB operon is induced in a majority of the bacteria within a few hours after plating on agar. The competitive index did not increase between 24 and 96 hours in the center of the colony biofilm where there were still ΔbcpAIOB mutants present, however, suggesting that the early activation of bcpAIOB gene expression was only transient. Moreover, the fact that streaking the strain containing a PbcpA-lacZ fusion resulted in a heterogeneous population of dark and light blue colonies indicates that the signal that induces bcpAIOB gene expression is not solely the solid surface environment. We are currently investigating the possibility that bcpAIOB gene expression is induced in response to both environmental cues and recognition of neighboring bacteria, including sensing whether those neighboring bacteria have BcpA proteins on their surface.
By contrast with our observations with B. thailandensis, CDI in E. coli occurs in liquid medium [1]. In strain EC93, the rat fecal isolate in which CDI was discovered, the cdiBAI genes are expressed constitutively [1]. In human uropathogenic E. coli strain 536, cdiBAI expression was not detected under standard laboratory growth conditions [4]. To measure CDI activity in 536, or in laboratory K12 strains of E. coli, therefore, the cdiBAI genes were expressed from an inducible promoter on a high copy number plasmid [1], [4]. Moreover, it was shown that production of capsule or P or S pili in the target bacteria blocked CDI [1]. These caveats raised questions about the biological relevance of CDI as observed under these conditions. In light of our findings, the possibilities that cdiBAI genes in E. coli and other bacteria are expressed in a probabilistic and transient manner and that they contribute to aggregation and/or biofilm formation are compelling hypotheses to test.
Another apparent difference between Burkholderia-type CDI systems and E. coli-type CDI systems is the presence of one or more “orphan cdiA-CT/cdiI modules” located 3′ to cdiBAI genes [5]. Comparative genome analyses suggest that these modules may move within or between bacteria into the functional cdiBAI locus, changing the allele that is expressed, and hence they may contribute to the diversity of cdiBAI alleles within a population and even within a clonal population [5]. We searched specifically for bcpA-CT/bcpI modules in bacteria containing Burkholderia-type CDI systems but found no evidence for their existence. Diversity of CDI systems amongst Burkholderia must, therefore, occur by a different mechanism.
Our experiments, aimed at providing an initial characterization of the function of the bcpAIOB gene products and determining their contribution to aggregation and interbacterial competition, used wild type and mutant derivatives of B. thailandensis strain E264. As environmental saprotrophs, however, Burkholderia species share their habitats with a plethora of other bacteria, as well as viruses and eukaryotes. How CDI is used in this environment and whether it occurs only in an intra-species manner or also in an inter-species manner is not known. Our bioinformatic analysis identified 58 bcpAI(O)B loci and 41 different alleles. Some strains have multiple alleles and several alleles are present in multiple strains. Notably, one allele that is present in two different B. pseudomallei strains (1106A-1 and BCC215-2) is also present in B. gladioli (strain BSR3-1), suggesting that BcpAIOB-mediated inter-species CDI does occur. It is also notable that many, if not all, bcpAI(O)B genes are located on genomic islands. Several Burkholderia species have been shown to be naturally competent [29] and there is tremendous genomic diversity amongst Burkholderia strains, mediated in part by horizontal gene transfer [30]. These observations raise several questions. For example, is there a hierarchy of potency amongst the various BcpA proteins? What happens if bacteria with different bcpAIOB alleles come into contact? Or if a bacterium containing one allele encounters a bacterium containing two alleles? Is it advantageous for a bacterium to contain multiple alleles? If so, why is the greatest number of alleles identified in a single strain only three? Is there a cost associated with bcpAIOB alleles that limits the number that can be tolerated within a single cell? The question of how diversity amongst bcpAIOB alleles is generated is a perplexing one as nucleotide changes in the region encoding BcpA-CT can presumably be tolerated only if compensatory changes occur in bcpI, and vice versa. Finally, the most important question and one related to CDI in general but perhaps the most difficult to address experimentally: What is the true role of CDI in nature? Is it used for competitive exclusion or for the acquisition of DNA that can be used for nutrition, as a biofilm matrix, or a source of genetic diversity? Perhaps it serves all of these purposes – or none and plays a role that we cannot yet imagine. Because life in a community is the rule rather than the exception for most bacteria, understanding how and why CDI systems function will be broadly relevant.
While this manuscript was in revision, Nikolakakis et al. published a related characterization of CDI systems in Burkholderia [31]. Although these authors used different experimental approaches, their results are consistent with ours and support the conclusion that bcpAIOB genes encode CDI systems that can mediate interbacterial competition and that function in an allele-specific manner. Moreover, Nikolakakis et al. showed that the C-terminal domains of some BcpA proteins possess tRNase activity, similar to what has been demonstrated for the C-terminal domains of CdiA proteins in E. coli type CDI systems [31].
Burkholderia thailandensis E264 is an environmental isolate [14]. Plasmids were maintained in E. coli DH5α and DH5αλpir and mated into B. thailandensis using the donor E. coli strain RHO3 [32].
B. thailandensis was cultured in low salt Luria-Bertani medium (LSLB, 0.5% NaCl) (unless otherwise stated), M63 minimal medium (supplemented with 1 mM MgSO4, 0.2% glucose, and 0.4% glycerol) [29], or M9 minimal medium (supplemented with 2 mM MgSO4, 0.1 mM CaCl2, and 0.2% glucose) supplemented, as appropriate, with 35 µg/ml chloramphenicol, 250 µg/ml kanamycin, 20 µg/ml tetracycline, 50 µg/ml trimethoprim, 40 µg/ml X-gal, or 0.1% chlorophenol alanine. Overnight cultures were aerated for ∼18 h at 37°C to OD600 ∼7–9, centrifuged at 16,000× g for 1 min, washed and resuspended in fresh LSLB broth or sterile phosphate buffered saline (PBS) for subsequent use. E. coli strains were cultured in Luria-Bertani (LB) medium supplemented, as appropriate, with 100 µg/ml ampicillin, 35 µg/ml chloramphenicol, 50 µg/ml kanamycin, 20 µg/ml tetracycline, 50 µg/ml trimethoprim, or 200 µg/ml diaminopamillic acid (for RHO3 strains).
The B. thailandensis ΔbcpO and ΔbcpB strains were created by allelic exchange. Deletion constructs were created by PCR amplifying (PFU Ultra polymerase, Agilent) ∼500 nucleotides 5′ of bcpO and bcpB (including the first three codons of the gene) and the ∼500 nucleotides 3′ of bcpO and bcpB (including the last three codons of the gene) from E264 genomic DNA. DNA fragments were restriction digested and cloned into allelic exchange vector pSM112, which carries a pheS counter-selectable marker. The resulting plasmids, pMA11 and pMA12, were used to create the ΔbcpO and ΔbcpB strains, respectively.
The B. thailandensis ΔbcpAIOB strain was created by natural transformation as described [29]. Briefly, ∼800 nucleotides 5′ of bcpA (including the first three codons of the gene) and ∼800 nucleotides 3′ of bcpB (including the last three codons of the gene) were amplified from genomic E264 DNA and the gene encoding kanamycin resistance (plus ∼500 nucleotides to include the promoter) was amplified (PFU Ultra polymerase, Agilent) from pUC18miniTn7(Km). Overlap PCR was performed to construct a single DNA product in which the kanamycin resistance-encoding gene was flanked by the regions 5′ to bcpA and 3′ to bcpB, which was subsequently transformed into B. thailandensis.
The constitutively active strain (PS12-bcpAIOB) was constructed as follows. Approximately 200 nucleotides 5′ to the ribosomal S12 protein-encoding gene (containing the promoter, PS12) and ∼750 nucleotides 3′ of the bcpA translation start site were PCR amplified (PFU Ultra polymerase, Agilent) from genomic E264 DNA, joined by overlap PCR, and cloned into pEXKm5 [32] to create pECG22, which was then used for cointegration into the chromosome of E264.
Strains used for Western blot analysis were constructed as follows. A DNA fragment corresponding to ∼1000 nucleotides of internal bcpA sequence with an encoding HA-tag located after F2633 of BcpA was constructed by overlap PCR and cloned into pEXKm5 [32]. The resulting plasmid, pECG17, was used for allelic exchange with B. thailandensis E264, E264ΔbcpO, and E264ΔbcpB to generate the strains E264BcpA-HA, E264BcpA-HAΔbcpO, and E264BcpA-HAΔbcpB, respectively. pECG22 was then mated into these strains to yield E264BcpA-HA::pECG22, E264BcpA-HAΔbcpO::pECG22, and E264BcpA-HAΔbcpB::pECG22 such that PS12 would drive expression of an intact bcpAIOB locus in which the encoding BcpA protein contained an HA-tag located after F2633. Complementation of bcpO and bcpB is described below; strains yielded were: E264BcpA-HAΔbcpO::bcpO::pECG22 and E264BcpA-HAΔbcpB::bcpB::pECG22, respectively. The bcpO and bcpB genes were expressed from the constitutive promoter PS12.
Strains complemented with bcpO, bcpI, and bcpB were constructed using a Tn7 transposase as described [33]. Derivatives of pUC18T-mini-Tn7T-Gm were constructed in which the gentamycin resistance-encoding gene was replaced with a gene encoding kanamycin resistance (pUC18miniTn7(Km)), chloramphenicol resistance (pUC18miniTn7(Cm)), or tetracycline resistance (pUC18miniTn7(Tc)). Briefly, ∼200 nucleotides 5′ to the ribosomal S12 protein-encoding gene (containing the promoter, PS12) or 500 nucleotides 5′ to bcpA (containing the promoter, PbcpA) and bcpO, bcpI, or bcpB were amplified from genomic E264 DNA, restriction digested, and cloned into pUC18miniTn7(Km) or pUC18miniTn7(Tc) for insertion at the attTn7 site on the chromosome.
B. thailandensis E264CmR and E264KmR (microscopy and flow cytometery vector control) were constructed using pUC18miniTn7(Cm) and pUC18miniTn7(Km) to insert a gene encoding chloramphenicol or kanamycin resistance, respectively, on the chromosome at the attTn7 site.
All lacZ and gfp reporter strains were created using the Tn7 transposase method as well. Briefly, ∼500 nucleotides 5′ to bcpA (PbcpA), ∼200 nucleotides corresponding to the predicted S12 promoter (PS12), or no DNA (Pneg) were cloned 5′ to a promoterless lacZ gene and ∼500 nucleotides 5′ to bcpA (PbcpA) was cloned 5′ to a promoterless gfp gene and cloned into pUC18miniTn7(Km). Transposition into the chromosome of B. thailandensis at the attTn7 site generated the PbcpA-lacZ, PS12-lacZ, Pneg-lacZ, and PbcpA-gfp strains. The miniTn7-kan-gfp plasmid [34] was used to generate the B. thailandensis PS12-gfp reporter strain in a similar manner.
All strains constructed were verified by PCR and sequencing analysis (Eton BioScience).
Burkholderia-type CDI-encoding loci were identified by using the predicted BcpB protein sequence from B. pseudomallei K96243 as a pBLAST query against the entire NCBI genome database. DNA sequences encoding predicted TPS system-like proteins were analyzed in Vector NTI Advance 11. A locus was considered to encode a Burkholderia-type CDI system if it was composed of a large ORF (∼3000 codons) followed immediately 3′ by a small ORF (∼100 codons), both of which were 5′ to the tpsB homolog. Alignments were performed in Vector NTI and then analyzed in Jalview 2.7 with Taylor residue coloring.
Total RNA was extracted from bacterial cells cultured overnight in LSLB broth using TRIzol Reagent (Invitrogen) according to the manufacturer's protocol. RNA was treated with 4 U of DNaseI (Ambion) for 30 min at 37°C. 2–5 µg of RNA was subsequently used for cDNA synthesis using SuperScript III First-Strand (Invitrogen) according to the manufacturer's protocol. RT-PCR was performed using GoTaq polymerase (Promega); PCR was also performed on E264 genomic DNA with identical primer sets as a control. PCR products were analyzed by 0.8% agarose gel electrophoresis and stained with ethidium bromide for visualization.
Reporter strains were cultured overnight in LSLB at 37°C with or without aeration, at room temperature (∼25°C) with aeration, or in M63 or M9 minimal medium at 37°C with aeration. Pneg-lacZ and PS12-lacZ were cultured in LSLB at 37°C with aeration. Heat shocked samples were obtained by incubating 500 µl aliquots of overnight LSLB (37°C, aerated) culture at 42°C for 5 min. Cultures were normalized to OD600 ∼0.2–0.5 and β-galactosidase activity was measured as described [35]. Two independent assays were performed in triplicate.
Overnight cultures of strains in LSLB medium were diluted to OD600 = 0.2 into 2 ml of M63 minimal medium in glass tubes. Bacteria were cultured with aeration for ∼48 h at 37°C with or without 4 or 10 U DNaseI with 10× DNaseI buffer (Ambion).
WT (E264BcpA-HA::pECG22), ΔbcpO (E264BcpA-HAΔbcpO::pECG22), ΔbcpB (E264BcpA-HAΔbcpB::pECG22), ΔbcpO::bcpO (E264BcpA-HAΔbcpO::bcpO::pECG22), ΔbcpB::bcpB (E264BcpA-HAΔbcpB::bcpB::pECG22), and no tag (E264::pECG22) strains were cultured overnight in LSLB supplemented with kanamycin. Cultures were washed twice in LSLB and diluted into fresh LSLB containing antibiotics. Bacteria were cultured overnight, pelleted, and resuspended in 2× SDS PAGE loading buffer to OD600 ∼50. Boiled samples were separated on a 5% SDS PAGE gel, transferred to nitrocellulose, and probed with mouse monoclonal anti-HA.11 antibody (Covance) at 1∶1,000 and goat anti-mouse IgG conjugated to IRDye680 (Odyssey) at 1∶15,000. Images were acquired on a Li-Cor (Odyssey) with software Odyssey v3.0.
The last ∼350 codons of bcpA (including the Nx(E/Q)LYN motif) from B. pseudomallei strains K96243 and 1106A-2 were PCR amplified (with an added ATG at the 5′ end) from genomic DNA and cloned into pSCrhaB2 [36]. Plasmids carrying bcpI were cloned in a similar manner except the trimethoprim resistance gene of pSCrhaB2 was replaced with a kanamycin resistance gene to allow for selection of these plasmids in combination with the plasmids encoding BcpA-CTs. B. thailandensis containing the BcpA-CT-encoding plasmids alone or in combination with the BcpI-encoding plasmids were cultured overnight in LSLB supplemented with trimethoprim (strains with BcpA-CT-encoding plasmids) or trimethoprim and kanamycin (strains with BcpA-CT- and BcpI-encoding plasmids) and 0.2% glucose. Cultures were washed and diluted in fresh LSLB medium to OD600 0.2–0.6 supplemented with antibiotics and 0.2% glucose or 0.2% rhamnose. Bacteria were cultured with aeration for 4 h at 37°C. Aliquots were taken at 0 h and 4 h, diluted in PBS, and plated on LSLB with antibiotic selection and 0.2% glucose to determine the cfu/ml of each strain. Two independent experiments were performed in triplicate.
Confocal microscopy– Liquid cultured bacteria preparation; bacteria were cultured overnight in LSLB broth and M63 minimal medium and concentrated to OD600 ∼50 in PBS. Solid medium cultured bacteria preparation; bacteria in colony biofilms on agar were cut out from petri dishes (agar included) and placed on a glass slide. Cover slips were added to the top of the colony biofilms. Bacteria were imaged on a Ziess LSM5 Pascal Confocal Laser Scanning microscope, 63× objective with oil immersion. Macroscope– Colony biofilms were imaged with a Leica M420 macroscope, 0.5× objective, at 1, 2, and 4 days post inoculation. Live-imaging microscopy– 500 µl of LSLB agar was added to cover glass bottom dishes, Delta T, (Biotechs) and allowed to solidify. Bacteria were plated onto the agar and imaged on an Olympus IX81 inverted microscope at 20× objective.
Bacteria were cultured overnight in LSLB broth and M63 minimal medium supplemented with kanamycin and diluted to OD600 ∼0.1 in PBS. Bacterial samples were analyzed on a Gallios Flow Cytometer (Beckman Coulter); bacteria were gated based on forward scatter and side scatter (∼1 um diameter and distinct from particles in the PBS only control (Figure S3)) for subsequent analysis. Data were analyzed with Kaluza 1.1 software.
Liquid– bacteria were cultured overnight, washed with PBS, and diluted to OD600 0.2 in fresh LSLB medium. Strains were mixed at a 1∶1 ratio (final volume of 2 ml) without antibiotic selection and cultured with aeration for 24 h at 37°C. Aliquots were taken at 0 h and 24 h, diluted in PBS, and plated on LSLB with antibiotic selection to determine the cfu/ml of each strain in the competition. Two independent experiments were performed in triplicate.
Solid– bacteria were cultured overnight, washed with PBS, and diluted to OD600 0.2 in LSLB. Strains were mixed at a 1∶1 ratio and 20 µl of culture was plated on solid LSLB (1.5% agar) without antibiotic selection. The culture inoculum was plated on LSLB with antibiotic selection to determine the ratio at 0 hours. Agar plates were stored at room temperature (∼25°C) for the duration of the experiment. Bacteria were picked from the colony biofilms with a sterile pipette tip, diluted in PBS, and plated on LSLB with antibiotic selection to determine the cfu of each strain in the competition at a particular dilution. The competitive index for each day was calculated as the ratio of wild type bacteria to mutant bacteria at time X hours divided by the ratio of wild type to mutant at time 0 hours. Two to three independent experiments were performed in triplicate.
|
10.1371/journal.ppat.1005344 | The Epstein-Barr Virus BART miRNA Cluster of the M81 Strain Modulates Multiple Functions in Primary B Cells | The Epstein-Barr virus (EBV) is a B lymphotropic virus that infects the majority of the human population. All EBV strains transform B lymphocytes, but some strains, such as M81, also induce spontaneous virus replication. EBV encodes 22 microRNAs (miRNAs) that form a cluster within the BART region of the virus and have been previously been found to stimulate tumor cell growth. Here we describe their functions in B cells infected by M81. We found that the BART miRNAs are downregulated in replicating cells, and that exposure of B cells in vitro or in vivo in humanized mice to a BART miRNA knockout virus resulted in an increased proportion of spontaneously replicating cells, relative to wild type virus. The BART miRNAs subcluster 1, and to a lesser extent subcluster 2, prevented expression of BZLF1, the key protein for initiation of lytic replication. Thus, multiple BART miRNAs cooperate to repress lytic replication. The BART miRNAs also downregulated pro- and anti-apoptotic mediators such as caspase 3 and LMP1, and their deletion did not sensitize B-cells to apoptosis. To the contrary, the majority of humanized mice infected with the BART miRNA knockout mutant developed tumors more rapidly, probably due to enhanced LMP1 expression, although deletion of the BART miRNAs did not modify the virus transforming abilities in vitro. This ability to slow cell growth could be confirmed in non-humanized immunocompromized mice. Injection of resting B cells exposed to a virus that lacks the BART miRNAs resulted in accelerated tumor growth, relative to wild type controls. Therefore, we found that the M81 BART miRNAs do not enhance B-cell tumorigenesis but rather repress it. The repressive effects of the BART miRNAs on potentially pathogenic viral functions in infected B cells are likely to facilitate long-term persistence of the virus in the infected host.
| The Epstein-Barr virus (EBV) infects more than 90% of the human adult population. Although EBV usually causes an asymptomatic infection, it is oncogenic in a small proportion of infected individuals. EBV produces a large number of microRNAs, a type of RNA that controls the production of their proteins though multiple mechanisms. We addressed the role played by the BART microRNAs, a subgroup of the EBV microRNAs, by generating a virus that lacks them and by comparing the characteristics of this modified virus with those of the unmodified virus. We found that the BART microRNAs cooperate to curb EBV multiplication, both in infected cells and in humanized mice. Furthermore, the BART miRNAs did not potentiate EBV’s ability to form tumors in different types of mice, some of which are unable to mount an immune reaction against the virus, as could have been expected from the literature. This can be explained at the molecular level by the ability of the BART microRNAs to downregulate the synthesis of multiple cellular and viral proteins, among which caspase 3 and LMP1, two essential modulator of cell death and cell proliferation, are likely to play an important role in the outcome of the virus infection. Thus, the BART microRNAs negatively impact on two essential viral functions, probably to maintain a balance between the virus and its host.
| The Epstein-Barr virus (EBV) is a strongly B lymphotropic virus that infects the majority of the world human population and is associated with the development of malignant tumors, mainly lymphomas and carcinomas of the nasopharynx (NPC) and of the stomach [1]. Shortly after infection, B cells start to divide and generate continuously growing cell lines, commonly termed lymphoblastoid cell lines (LCLs) [2]. Infected cells express a set of latent proteins ascribed to subfamilies known as Epstein-Barr virus nuclear antigens (EBNA) and latent membrane proteins (LMP), most of which are essential or strongly potentiate the B cell transformation process [1]. EBV also encodes 44 miRNAs that are divided into two clusters located around the BHRF1 gene (BHRF1 miRNAs) or within the introns of the BART gene (BART miRNAs) [3–5]. Viruses devoid of the BHRF1 miRNA locus are less transforming than their wild type counterparts [6–9]. Indeed, a recombinant virus that lacks this cluster retains only 1/20th of the wild type transforming capacity [8].
The BART miRNAs are present in all EBV-infected cells, but their expression level is up to hundred times higher in epithelial cells than in infected B lymphocytes, suggesting that they exert their main function in the former type of cells [5,10]. EBV-associated carcinomas produce only a restricted number of latent proteins but also the BART miRNAs, making these non-coding RNAs prime suspects in the transformation process [11]. Indeed, miR-BART9 and miR-BART7-3p have been found to promote metastasis of NPC cells [12,13]. Reciprocally, anti-miR-BART7-3p reduced tumor growth in an animal model [13]. In the same vein, miR-BART3* was found to target the tumor suppressor gene DICE1 in NPC cells [14]. In the primary B cell system, Vereide et al. used a B95-8 EBV strain virus with a reconstituted BART locus to show that the BART miRNAs also improve the transforming abilities of the virus [9].
The BART miRNAs have been found to regulate apoptosis by targeting pro-apoptotic proteins such as PUMA and BIM in epithelial cells [15,16]. A photoactivatable ribonucleoside-enhanced crosslinking and immunoprecipitation (PAR-CLIP) screening of the EBV-positive NPC cell line C666 revealed that the antiapoptotic properties of the BART miRNAs can be ascribed to the direct targeting of 10 pro-apoptotic proteins in these cells [17]. The BART miRNAs have also been suggested to act as repressors of EBV lytic replication in B cell or epithelial cell lines induced with drugs such as TPA. MiR-BART6, through its ability to target DICER, and miR-BART20-5p through targeting of BZLF1, control entry into the EBV lytic replication phase [18,19]. MiR-BART18-5p also controls the onset of replication in anti-Ig-treated Akata Burkitt’s lymphoma cell lines and in LCLs induced by TPA through its ability to target the expression of MAP3K2 [20]. This is in agreement with the view that EBV-infected B cells hardly replicate the virus and are mainly latent, whilst infected epithelial cells are the main sites of replication [2]. These data have frequently been collated in tumor cells or in LCLs infected by viruses that carry a partial deletion of the BART miRNAs and might not extend to strains with an intact locus, in particular in vivo. Furthermore, the view that LCLs are primarily latent and must be stimulated to produce virus is restricted to viral strains close to B95-8. We have recently shown that M81, a viral strain that carries a high degree of homology with viruses found in NPC and that infects a substantial proportion of the Chinese population, induces a high degree of spontaneous virus replication upon infection of primary B cells [21]. Importantly, this virus carries an intact BART locus. Both features render M81 a suitable experimental system to study the function of the BART locus in infected primary B cells. Here we report the phenotypic traits of a set of viruses that are devoid of different subsets of the BART miRNA cluster in the fully permissive M81-based replication system.
We set out to determine the role of the BART miRNAs by infecting B cells with a virus devoid of this locus. To this end, we sequentially deleted the BART subcluster 1 (M81/ΔC1), subcluster 2 (M81/ΔC1C2) and miR-BART-2 (M81/ΔAll) (S1 Fig). Importantly, this mutant retains intact RPMS1, LF1 and LF2 exons. We then generated a revertant virus by reintroducing the complete BART locus in M81/ΔAll (M81/ΔAll/rev). We also individually deleted the subcluster 2 or miR-BART2 in the wild type genome to generate M81/ΔC2 and M81/Δb2 (S1 Fig). All these recombinants were stably introduced into 293 cells to generate producer cells lines that carry intact copies of the mutants, and were accordingly termed 293/M81/ΔAll et cetera (S1 Fig). We began by infecting B cells isolated from the peripheral blood with M81, M81/ΔAll and M81/ΔAll/rev to generate a panel of lymphoblastoid cell lines (LCLs). We measured BART miRNA expression in these cells for miR-BART1-3p, miR-BART7* and miR-BART2-5p that are expressed at good levels in these LCLs and are representative of each BART subcluster [22]. This confirmed that the cells infected by the mutants did not express the miRNAs that had been deleted (S2A Fig). Although the construction of the ΔAll mutant left the LF1, LF2 and LF3 genes intact, we also quantified expression of the BART mRNA by RT-PCR in LCLs generated with M81/ΔAll or wild type M81. This assay revealed that the transcript is produced in cells infected by the ΔAll mutant, on average at marginally higher levels relative to wild type (S2B Fig). However, the difference was not statistically significant (S2B Fig). We then assessed lytic replication in the LCL panel as we previously reported that M81 spontaneously replicates in B cells [21]. We first gauged BZLF1 by western blot in 18 donors (Fig 1A and S3 Fig). These assays revealed that expression of this protein is enhanced by an average 3.4 fold in the absence of the BART miRNAs, compared to LCLs infected with wild type EBV or the M81/ΔAll revertant. In two of these cases (samples F and I), the BZLF1 expression was very close or even higher in the LCL infected by the wild type virus (S3A Fig). We then attempted to complement the phenotype of LCLs transformed by M81/ΔAll by transfecting them with a plasmid that encodes all BART miRNAs with the exception of miR-BART-b2, as well as a truncated nerve growth factor receptor (NGFR) that is expressed at the surface of transfected cells, or with a control vector. We first used qPCR to confirm that NGFR antibody-purified cells transfected with the BART miRNAs express them, as shown in Fig 1B, and then performed a western blot with the same cells. Cells transfected with the BART expression vector exhibited lower levels of BZLF1 protein than those transfected with control plasmids, confirming that expression of the BZLF1 protein is modulated by the BART miRNAs.
We then performed immunofluorescence stains (IF) with the same antibody at different time points that showed that the number of BZLF1-positive cells was, on average, 2 to 3 times higher in cells infected with the BART miRNA KO mutant, relative to wild type controls (p<0.02) (Fig 1C). However, the intensity of the staining at the single cell level was not significantly stronger in BZLF1-positive cells. To confirm this observation, we performed a western blot with protein extracts from LCLs infected with the M81/ΔAll mutant or wild type virus, normalized by the number of BZLF1-positive cells in the LCL infected by the wild type control, as determined by IF (Fig 1D). Thus, these samples contained approximately the same number of BZLF1-positive B cells. Both samples generated BZLF1 signals of similar intensity, confirming that the replicating cells from LCLs infected by wild type or ΔAll mutants produce similar amounts of the BZLF1 protein. A RT-qPCR with BZLF1-specific primers performed on the LCL panel revealed that mRNA BZLF1 expression was higher in the LCLs infected with the miRNA mutant, particularly 40 days after infection but did not exceed a factor of 2.5 over time (Fig 1E). Taking into account that excision of the BART miRNAs results in an increase of the number of BZLF1-positive cells but not in an increase in protein expression at the single cell level, the likeliest explanation for the enhanced BZLF1 mRNA production is that an increased number of cell produced it, although a direct effect of the BART miRNAs on RNA stability could have contributed to this process. Altogether, we conclude that the BART miRNAs contribute to the repression of spontaneous reactivation of the lytic cycle in infected cells, but that their absence does not substantially increase BZLF1 protein expression in the individual replicating cells.
We then studied expression of the BZLF1 protein over time. B cells infected by a virus that lacks BZLF1 and BRLF1 (ΔZR) served as a negative control [21]. We performed immunoblots at day 22, 43 and 84 dpi that showed an overall decrease in the expression of this protein. However, the LCLs infected with the M81/ΔAll remained longer strongly BZLF1-positive (Fig 1F). We assessed the consequences of increased BZLF1 expression by staining the infected LCLs with antibodies against gp350. Both the western blot and the IF stains showed a higher number of gp350-positive cells in cells infected with the ΔAll mutant than in those infected with the wild type virus (Fig 2A and 2B). It is important to note that the large majority of the viruses that are produced by the replicating cells bind to their neighbor B cells, some of which become covered with gp350-positive signals, making it difficult to distinguish them from producer cells [21]. Thus, the results presented here include both types of gp350-positive cells. We also measured the viral copy numbers in supernatants from LCLs by qPCR and found them clearly increased in those from LCLs generated with M81/ΔAll (Fig 2C). We then infected primary B cells with these LCL supernatants at low cell density to prove that the increased viral titers were a consequence of enhanced virus production. Indeed, this assay showed an increase in the number of outgrowing clones after treatment with supernatants from the M81/ΔAll LCL, although the efficiency of transformation widely varied between different supernatants (Fig 2D).
MiRNAs typically modulate expression of target genes by forming the RNA-induced silencing complex (RISC) that includes miRNAs, their mRNA targets and a member of the Argonaute family of proteins. Therefore, we tested whether the deletion of the BART miRNAs modifies the recruitment of BZLF1 mRNA to the RISC. We began by measuring the expression levels of DICER in cells lacking the BART miRNAs, as this protein has been suggested as a target of the BART miRNAs [19]. A western blot performed on three pairs of LCLs generated from three different blood donors and infected with either M81 or M81/ΔAll showed an increase in DICER protein expression (Fig 3A). However, we gauged the expression levels of three cellular miRNAs expressed in B cells but could not identify any differences between LCLs transformed with wild type EBV or with the ΔAll mutant (Fig 3B). We conclude that the impact of the BART miRNAs on DICER in infected B cells has no generalized and pronounced functional consequences. We performed RISC immunoprecipitations in couples of LCLs infected with wild type or ΔAll mutant using an antibody directed against Ago2 (S4A Fig) and measured mRNA expression by qPCR in the precipitates. We assessed the efficacy of this protocol by comparing expression of the EBV miR-BHRF1 miRNAs in the Ago2 antibody precipitate with the expression in untreated LCLs. This assay showed more than 10000-fold enrichment of this miRNA after immunoprecipitation (S4B Fig). An immunoprecipitation with an antibody directed against BrdU performed in parallel measured the non-specific background mRNA recovery. We measured the expression levels of GAPDH, HPRT, IPO7, LMP1 and BZLF1 mRNAs in the RISC of LCLs transformed with M81 or M81/ΔAll. GAPDH expression levels were used to normalize for mRNA recovery, HPRT has previously found not to be recruited in the RISC by BARTs, whereas the IPO7 mRNA is a previously validated BART target [9,23–25]. We also quantified expression of these mRNAs in infected cells. Comparison with the expression levels obtained after immunoprecipitation determines the efficacy of recruitment to the RISC. A prerequisite for this analysis is that the number of mRNA molecules than can be recruited to the RISC does not vary too much between the cells, so as to exclude saturation effects. However, we have seen that the increase in BZLF1 protein expression results from an increased number of replicating cells that all produce the protein at approximately the same level. Thus, it is unlikely that the BZLF1 mRNA expression level and recruitment to the RISC will vary significantly between different replicating cells.
The results of these experiments are given in Fig 3C and 3D, respectively. They first show the relative amounts of mRNAs in the RISC after subtraction of the background generated by the BrdU antibody. This experiment shows that BZLF1, LMP1, and to a lesser extent IPO7 mRNAs are more abundant in the RISC of LCLs infected with M81 wild type virus. In contrast, the levels of HPRT mRNAs in the RISC were similar in LCLs infected with either type of virus (Fig 3C). We then calculated the level of enrichment of these mRNAs into the RISC by comparing expression levels after Ago2 immunoprecipitation or in untreated cells (Fig 3D). This figure clearly shows that the relative recruitment in the LCLs infected with wild type viruses was more efficient for BZLF1, LMP1 and IPO7 than for HPRT. We conclude that the BART miRNAs recruit the first three mRNAs to the RISC. However, these genes are still detectable in the RISC of LCLs infected with the ΔAll mutant. Therefore, other miRNAs, presumably miRNAs of cellular origin, recruit these mRNAs to the RISC. We wished to confirm these results by performing luciferase reporter assays. To this end, we constructed a pGL4.5-based luciferase reporter plasmid that carries the BZLF1 3’UTR. We also fused the luciferase gene to the BALF5 3’UTR that has previously been reported to be directly targeted by BART-2 miRNA [26]. The luciferase expression plasmid devoid of 3’UTR was used as a negative control. These plasmids were cotransfected with a rat-CD2 expression plasmid into two independent sets of LCLs generated with either of M81 and M81/ΔAll to determine the transfection efficiency and allow normalization. We measured the luciferase activity in transfected cells and found that the luciferase activity for both plasmids carrying the BZLF1 or the BALF5 3’UTR was reduced by 40 to 60% in the wild type LCL, relative to the LCL generated with the ΔAll mutant (Fig 3E), a result previously reported for BALF5 [26]. The control plasmid pGL4.5 showed no major difference in expression (Fig 3E). These results suggest that M81 LCLs recruit more BZLF1 mRNA to the RISC than M81/ΔAll LCLs by targeting the BZLF1 3’UTR. However, the low transfection rates achieved in LCLs (1 to 2%) indicate that we should exercise caution when interpreting these results
The data gathered so far indicated that the BART miRNAs repress lytic replication in primary B cells. However, B cells infected with wild type EBV undergo lytic replication, although these cells can express the BART miRNAs. This paradox could be resolved if the expression levels of the BART miRNAs differed in replicating and non-replicating cells. To address this question, we constructed a virus that encodes a truncated version of the CD2 molecule, whose expression is driven by the viral EA-D promoter. Thus, infected B cells undergoing lytic replication express CD2 at their cell surface and can be immunocaptured by a specific antibody. We quantified BZLF1 expression in the CD2-positive and CD2-negative populations using quantitative RT-PCR (Fig 4A). As expected, we found that only CD2-positive cells produced BZLF1 at the RNA and protein level. This implies that cells that expressed the BZLF1 mRNA also expressed the BZLF1 protein, ie these mRNAs are not subjected to massive miRNA interference. We then quantified the expression profile of some viral and cellular miRNAs in these 2 cell populations. We found that replicating cells expressed approximately 2 times less BART miRNAs and 3 times less BHRF1 miRNAs than non-replicating cells (Fig 4B). Thus, there is an inverse relationship between EBV miRNA and BZLF1 expression. However, the cellular miRNAs were expressed at the same level irrespective of the replication status of the infected cells, suggesting that these cells did not globally downregulate miRNA synthesis (Fig 4C).
Altogether, the previous results demonstrated that the BART miRNA repress spontaneous expression of BZLF1 in vitro and that their deletion enhances full productive lytic replication with virion production. However, the BART cluster is very large and we wished to learn the respective contribution of its subclusters. Therefore, we quantified BZLF1 expression in LCLs transformed with M81/ΔC1, M81/ΔC2 and M81/Δb2 from 2 independent blood samples at two different time points (Fig 5A, 5B and S5 Fig). These experiments showed that BZLF1 expression is higher in LCLs obtained by infection with M81/ΔC1 than in controls. Although we found no evidence for increased BZLF1 expression in M81/ΔC2, both M81/ΔC1 and M81/ΔC2 LCLs expressed BZLF1 less strongly than M81/ΔAll and M81/ΔC1C2, and this effect remained visible after 101 days of culture. LCLs infected with M81/Δb2 were indistinguishable from the wild type controls in terms of BZLF1 expression, although this miRNA has been suggested to control the onset of lytic replication [26]. However, we measured the expression of BALF5 in LCLs infected with M81/ΔAll or M81/Δb2 or wild type controls and found that the expression of BALF5 is indeed increased in LCLs generated with the mutant, particularly in those that carry M81/ΔAll (Fig 5C and 5D). However, the increased expression of BALF5 in the LCLs infected with the M81/Δb2 virus does not result from an increased replication in these cells as shown by the unchanged BZLF1 expression in LCLs generated by a virus that lacks miR-BART-2 (Fig 5A and 5B).
We wished to complete the phenotypic analysis of the M81/ΔAll mutant by infecting humanized mice. We infected 7 mice intraperitoneally with the mutant and 5 with its wild type counterparts and measured blood viral titers 5 weeks post-infection (Fig 6A). The titers were much higher in 4 out of 7 mice infected with M81/ΔAll than in the positive controls at early time point (Fig 6A). Furthermore, animals infected with the mutant showed signs of wasting (loss of weight, apathy, food refusal, ruffled hair) and four of them had to be euthanized approximately at week 6, 2 weeks before the planned termination of the experiment. This phenomenon was not seen in mice infected with wild type virus and is statistically significant (0/5 in M81-infected versus 4/7 in M81/ΔAll-infected mice; p = 0.019 by one-tailed Chi-square test). To allow comparison with wild type infected mice, we euthanized one of these mice at the same time. Whilst gross examination of the organs showed large neoplastic nodules in the spleen of infected with the M81/ΔAll mutant, there were only a few interspersed EBER-positive cells in the spleen of the animal infected with wild type virus. The remaining mice survived until week 8 without signs of animal suffering at which time they were euthanized. Both mice infected with wild type M81 and those infected with the ΔAll mutant showed tumors in the spleen or tumors in the pancreas for 3 animals (Fig 6B and Table 1). Although the mice were not all investigated at the same time, we found that 3 out of 5 mice infected with wild type M81 and 7 out of 7 mice infected with M81/ΔAll had macroscopic tumors and the difference between the 2 groups of animals is statistically significant (Fig 6H and Table 1). Animals with tumors in the pancreas tended to have lower virus titers than those with tumors in the spleen, possibly because the tumor cells in this case have more restricted access to the blood circulation. Histological examination of the above-described neoplastic infiltrates readily confirmed the presence of activated lymphoid cells that proved to be EBER-positive (Fig 6B). We also found histological evidence of EBV-positive diffuse B cell infiltration of variable intensity after infection with M81/ΔAll and wild type EBV in other organs such as the liver, or the pancreas (Fig 6B and Table 1). The density of EBER-positive cells in these organs was similar in mice infected with wild type or with the mutant, although there was great variation in the concentration of EBV-positive cells in the lymphoid tissues between animals infected with the same virus strain (Fig 6B–6E). We then stained histological sections of the tissues infiltrated with tumor cells for BZLF1 and gp350. All infected tissues contained cells expressing both the early and the late marker of lytic replication but the ratio between EBER and BZLF1 or EBER and gp350 proved to be globally higher in the mice infected with the virus devoid of the BART miRNAs. However, the differences in BZLF1 expression between the two viruses, that proved to be statistically significant, were more pronounced than for gp350 that indeed failed to reach statistical significance (Fig 6D and 6E). One possible explanation for this result is that dead gp350-positive cells are rapidly eliminated in vivo but not in vitro. We also attempted to generate LCLs with the serum from euthanized animals. However, we could generate only one LCL with the serum from a mouse infected with WT virus that did not display the highest viral titers. We conclude that the viral DNA measured in the serum does not derive from infectious virions, but rather probably from decayed infected cells. This observation is concordant with the fact that free virions are captured by B cells [21]. We also stained the tissues for LMP1 and EBNA2 expression and found that, although immunohistochemistry is not an entirely reliable quantitative assay, mice infected with ΔAll express LMP1 at much higher levels than mice infected with wild type viruses (Fig 6B). However, the percentage of infected cells that expressed LMP1 or EBNA2 among the EBV-infected population showed no difference between wild type- and ΔAll-infected mice (Fig 6F and 6G). We then attempted to confirm these observations in another set of non-humanized immuno-compromised mice. The rationale behind this experiment was to evaluate cell growth in the absence of a functional immune system that might reduce cell growth, particularly in mice infected with the ΔAll mutant that produces more lytic antigens. To this end, we injected i.p. two sets of independent peripheral B cells exposed to wild type or ΔAll viruses. In that case, we terminated the experiment at week 5 to exclude tumor overgrowth and allow more direct comparison between tumor burdens. We found that 7 out of 7 animals infected with ΔAll developed macroscopically visible tumors, mainly in the pancreas and to a lesser extent in the kidney and liver, compared to 3 out of 7 in animals infected with wild type virus (S6A Fig and S1 Table). Importantly, the tumor burden was much higher in mice infected with the ΔAll knockout and these differences were statistically significant (S6B Fig). Whilst the tumor burden concentrated in the spleen in humanized mice, this organ was spared in the non-humanized counterparts, as was expected in the absence of human hematopoietic transplantation. There was no difference in the incidence of tumors between the mice groups infected with the 2 different B cell samples. As previously observed, the tumor cells expressed EBER, EBNA2, LMP1, BZLF1 and gp350 (S6C–S6H Fig). Here again there was no statistical difference in the density of EBER-positive cells within the tumors. In the same vein, the percentage of EBNA2-positive cells was similar in either type of virus infection and lytic replication was clearly stronger in mice infected with M81/ΔAll. There was, however, a difference in the proportion of EBER-positive cells expressing LMP1, the latter being higher in mice infected with M81/ΔAll, suggesting that these cells were selected against in immunocompetent mice. As was the case in humanized mice, the proportion of strongly LMP1-positive cells was much higher in mice infected with M81/ΔAll (S6C Fig). In summary, animals infected with a virus that lacks the BART miRNAs showed increased spontaneous lytic replication and frequently accelerated tumor formation in vivo, accompanied by increased LMP1 production.
The accelerated tumor growth in animals infected with the ΔAll mutant could be explained by a shorter doubling time of B cells infected with this virus. Therefore, we assessed the transforming ability of M81/ΔAll at low cell density and low MOI. We infected primary B cells from 5 independent peripheral blood donors and did not find any difference in transforming ability between wild type and mutant viruses (S7 Fig). We also used western blot to quantify expression of latent genes implicated in B cell growth. We found no difference in EBNA2, EBNA3A, or LMP2 expression. However, EBNA3B and EBNA3C were expressed at mildly higher levels in 3 out of 4 LCLs infected with the virus devoid of the BART miRNAs (Fig 7A). Western blots performed with a LMP1-specific antibody showed a clearly increased expression of this protein in 3 out of 4 cell samples infected with ΔAll mutant, relative to wild type controls. We also assessed LMP1 expression in 3 additional LCL triplets that were generated with wild type M81, the ΔAll mutant and the ΔAll revertant and found that LMP1 expression was identical in both wild type and revertant, but strongly increased in the LCL generated with the ΔAll mutant in 2 out of 3 cases, the remaining case showing a minor LMP1 increase in the ΔAll mutant (S8A Fig). It is interesting to note that LMP1 expression varied in LCLs infected with wild type M81 varied markedly between blood samples, with a minority of LCLs showing a much stronger expression. In these latter cases only, the absence of BART did not or only hardly increased LMP1 expression that was already high in the wild type LCL, suggesting that cellular polymorphisms modulate LMP1 expression. Highly variable LMP1 transcription rates in LCLs infected with the same virus were previously reported [27]. We also addressed the relationship between LMP1 expression and lytic replication by infecting 2 additional blood samples with wild type M81, the ΔAll mutant as well as with a mutant that lacks BZLF1 and BRLF1 and is replication-deficient (M81/ΔZR). We confirmed that BZLF1 expression was higher in the LCLs generated with the ΔAll mutant than in the wild type control. As expected the LCLs generated with M81/ΔZR did not replicate at all. In both samples LMP1 expression was identical in the LCLs generated either with wild type virus or with M81/ΔZR and was weaker than in the LCLs infected with ΔAll (S8B Fig). Altogether, we found that LMP1 expression was stronger after infection with ΔAll in 7 out of 9 independent LCLs.
The BART miRNAs were previously found to increase resistance to apoptosis though inhibition of caspase 3 expression [9]. Therefore, we gauged caspase 3 expression at the protein level using a western blot analysis in infected cells from 4 different donors. This assay showed a 2 to 5-fold increase in caspase 3 expression in all cases. We also assessed the ability of LCLs to withstand an apoptotic stress by incubating the cell lines with a panel of apoptosis-inducing drugs including Ionomycin, Staurosporine, Simvastatin and Etoposide [28]. The levels of apoptosis were assessed by TUNEL or caspase 3 cleavage assays. We could not identify significant differences between LCLs infected with the mutants or with the controls after treatment with Etoposide or Simvastatin (Fig 7B and 7C). However, whilst treatment with Ionomycin gave rise to more apoptosis in LCLs infected with wild type viruses in TUNEL assays, staining with caspase 3 revealed increased apoptosis in wild type LCLs treated with Staurosporine.
The study of spontaneous EBV lytic replication has been hampered by the propensity of the virus to enter latency in infected cells. LCLs initiate some degree of lytic replication after treatment with chemicals such as TPA or butyrate [29]. Some non-lymphoid cell lines such as 293 cells can support lytic replication after transfection with BZLF1 [30]. Infection of primary epithelial cells gives rise to spontaneous lytic replication but the efficiency of infection remains low and these cells are difficult to grow in large numbers [31,32]. Thus, tractable experimental systems have not been available for a long time. However, M81, a virus isolated from an NPC patient replicates strongly in primary B cells isolated from any individual tested so far [21]. Furthermore, M81 is amenable to a genetic analysis after its cloning as a bacterial artificial chromosome [21]. We addressed the function of the BART miRNAs by constructing viruses that evince partial or complete deletions of this locus, as well as a revertant thereof. We found that the BART miRNAs negatively regulate spontaneous lytic replication in B cells, as their excision from the M81 viral genome gives rise to an increase in spontaneous lytic replication in vitro and in vivo in humanized mice. This phenotypic trait disappears in the revertant virus or upon complementation. The BART miRNAs seem to target BZLF1 directly as its mRNA is recruited more efficiently to the RISC in cells infected by wild type virus than in LCLs generated with the BART miRNA knockout virus and expression of a luciferase gene fused to BZLF1 3’UTR is lower in LCLs generated with wild type virus relative to LCLs generated with the ΔAll virus. However, the difficulties to transfect primary LCLs with high efficiency somehow qualifies the latter result. Typically, miRNAs and their cognate targets are expressed in the same cells and the miRNA down-regulate protein expression in all cells that express them. In LCLs, the BART miRNAs are expressed in latently infected cells that do not express the BZLF1. Therefore, the BART miRNAs can only exert their function on BZLF1 in the minority of cells that initiate lytic replication in a given LCL. Thus, they do not directly control lytic replication but come into play only in cells that have already initiated BZLF1 synthesis. Such a scenario fits with the observation that the number of spontaneously replicating cells in LCLs infected with M81/ΔAll does not exceed 15%. The remaining 85% are devoid of BART miRNAs but nevertheless remain BZLF1-negative. However, in cells that have already initiated lytic replication through expression of BZLF1, the expression of the BART miRNAs apparently needs to be lower than in non-replicating cells. This, combined with the observation that individual replicating cells in LCLs infected with wild type or with M81/ΔAll express the BZLF1 protein at the same level suggests that the halved level of BART miRNAs present in replicating cells infected with wild type viruses is too low to efficiently down-regulate BZLF1 protein. This would mean that the expression of the BART miRNAs needs be lower than in latent cells, but does not need necessarily need to be completely extinguished.
It remains unclear at this point whether the BART miRNAs are actively downregulated in replicating cells through an unknown active molecular mechanism, or whether the expression of the BART miRNA is stochastically distributed within the different cells of a LCL, with replication taking place in cells that happen to express low levels of these non-coding small RNAs. We previously observed that B cells infected with M81 sustain lytic replication over long periods of time, exceeding 3 months of continuous cell culture growth [21]. However, even in these cells, lytic replication eventually stopped. It is interesting to note that the BART miRNAs, probably independently of their effects on BZLF1, also interfere with the mechanisms that control the long-term ability of a lymphoid cell to support lytic replication, as LCLs generated with M81/ΔAll keep producing virus when LCLs established from the same patient with wild type EBV do not anymore. However, replication in LCLs generated with M81/ΔAll also decreases with time, suggesting that the BART miRNAs accelerate this long-term mode of control of lytic replication but are independent of it. Thus, the BART miRNAs could interfere with lytic replication in several ways, negatively modulating lytic replication at its onset through its effect on BZLF1, but also influencing the long-term ability of the LCL to support lytic replication by interfering with the selection process that favors latency.
We then turned our attention to the miRNAs within the cluster that affect BZLF1 expression. Because the cluster encodes 22 miRNAs, we addressed the role played by the subcluster 1, subcluster 2 and miR-BART2 in this process. We found that viruses devoid of the subcluster 2 or of miR-BART2 do not differ from wild type viruses in their ability to control the expression of BZLF1. However, a virus that lacks both subcluster 1 and subcluster 2 expressed more BZLF1 than those infected with a subcluster 1 deletion mutant. This demonstrates that subcluster 1 plays a predominant role in the control of lytic replication but also that subcluster 2 contributes to this process. We did not identify any difference between the ΔC1C2 and the ΔAll viruses in terms of BZLF1 protein expression. Thus, miR-BART-2 does not seem to be implicated in the onset of lytic replication although it clearly modulates BALF5 expression as previously shown and confirmed in the presented study [26]. Two BART miRNAs that belong to subcluster 2 have been proposed to control lytic replication through modulation of BZLF1 and BRLF1 expression. Jung et al. demonstrated that miR-BART20-5p suppressed lytic replication through direct targeting of BZLF1 and BRLF1 mRNAs in a variety of EBV-positive epithelial cell lines [18]. MiR-BART20-5p is not or barely expressed in EBV-transformed LCLs and is thus unlikely to play a substantial role in spontaneous lytic replication in B cells [10]. MiR-BART18-5p was found to repress lytic replication in anti-Ig-treated Akata Burkitt’s lymphoma cell lines and in LCLs induced by TPA through its ability to target MAP3K2 [20]. These models display obvious differences with the spontaneous replication of LCLs that could explain the relatively minor role that we ascribed to the subcluster 2. Therefore, it is possible that miR-BART18-5p plays an essential role in induced but not in spontaneous lytic replication. Our data point to a control of BZLF1 expression shared by multiple BART miRNAs whose individual contribution might be very limited and undetectable in viruses lacking a single miRNA. In such a case, only deletion of a subset of BART miRNAs could reveal their effect on BZLF1 expression and lytic replication.
We also evaluated the role played by the BART miRNAs in the control of EBV-mediated transformation. We found that the deletion of the BART cluster does not influence the transformation abilities of the virus in vitro. Its impact in humanized mice was more complex to assess. Mice infected with ΔAll or with wild type viruses developed similar tumor burdens. However, this tumor load was already present after 5 weeks of infection in 4 out of 7 mice infected with ΔAll, whilst it took two additional weeks in the remaining animals. Importantly, mice infected with the BART miRNA knockout experienced a higher level of lytic replication that should theoretically allow infection of a larger number of B cells, thereby increasing the ensuing tumor mass. However, higher lytic replication might also boost the immune response against replicating cells. Therefore, we turned to non-humanized immunocompromized mice that cannot mount an immune response and do not have a large reservoir of EBV-negative resting B cells that can be infected by the viruses produced by a replicating B cell. We find that injection of freshly isolated peripheral blood B cells exposed to viruses and injected intraperitoneally gives rise to lymphoid tumors. This experiment allowed direct comparison of the transforming capacities of the virus and of the mutant. It showed that the tumor incidence is more than twice as high in animals treated with B cells infected by ΔAll and that the tumor burden was on average 5 times higher relative to B cells exposed to wild type virus.
We conclude that the absence of BART miRNAs efficiently supported the growth of EBV-transformed cells. This apparently contradicts numerous studies that implicated the BART miRNAs in EBV-induced epithelial carcinogenesis. Here again, the much higher expression levels in EBV-associated carcinomas needs to be considered. Along the same line, a recently established xenograft model in NSG mice that clearly implicates the BART miRNAs in tumorigenesis uses cells that express the BART miRNAs at even higher levels than in NPC [33]. The expression levels of some EBV latent proteins were also moderately increased in some LCLs transformed by the M81/ΔAll mutant. Although this might have contributed to the increased transforming abilities of the mutant, this increase was inconstant and lower in intensity than observed for BZLF1 or LMP1. This suggests that the effect on these latent proteins was indirect. LMP2A was suggested to be a target of the BART miRNAs but we could not confirm this observation in our experimental system [34]. We also evaluated the impact of the BART miRNAs on apoptosis or more generally cell death in infected B cells. We could not observe any increase in apoptosis in LCLs generated with the BART miRNA-negative virus, but to the contrary a moderate protective effect after treatment with some pro-apoptotic drugs such as Staurosporine. Ionomycin induced a higher level of cell death as assessed by an increased number of cells in TUNEL assay but did not modify the level of cleaved caspase 3. Therefore, it is unlikely to reflect an increased level of apoptosis. Reciprocally, Staurosporine increased the percentage of cleaved caspase 3 but not the percentage of positive cells in TUNEL assays, suggesting that the cells entered the apoptosis pathway but could not complete it. We then evaluated the influence of BART miRNAs on some proteins involved in the regulation of apoptosis in EBV-infected cells. We found that the caspase 3 protein is increased after excision of the BART miRNA cluster in LCLs infected with M81/ΔAll. Caspase 3 has been found as a direct target of the BART miRNAs in Burkitt’s lymphoma cells but a more recent study performed on the NPC C666 cell line could not confirm these results [9,17]. This raises the intriguing possibility of a different mode of interaction between the caspase 3 mRNA and the BART miRNAs in different cell lineages. Our own results cannot distinguish between a direct and an indirect effect of the BART miRNAs on caspase 3 expression.
Importantly, the LCLs that showed increased caspase 3 protein production also expressed the LMP1 protein at higher levels than the controls. This confirms previous studies that identify this viral oncoprotein as a direct target of the BART miRNAs [25,35,36]. Importantly, although immunohistochemistry is not an accurate quantitative method, we found that LMP1 is expressed at clearly higher levels in mice infected with ΔAll and this event is likely to have boosted cell growth. LMP1 has been found to facilitate extrinsic apoptosis through its ability to increase the expression of CD95 [37]. Furthermore, the LMP1 transmembrane domain activates apoptosis through activation of the unfolded protein response [38]. However, LMP1 can also protect against apoptosis through induction of BCL2A1, a member of the BCL2 family of proteins [38], or though inactivation of the p53 protein [39]. As EBV-infected cells express LMP1, the anti-apoptotic effect of this viral protein seems to predominate in infected cells [38]. In LCLs that carry the M81/ΔAll mutant, the increase in LMP1 might predominate over the induction of pro-apoptotic proteins caused by the absence of BART miRNAs. Interestingly, the BART miRNAs were found to have anti-apoptotic properties in Burkitt’s lymphoma cells in which apoptosis was induced by the loss of the EBV genome with the help of a dominant negative version of EBNA1. In that case, LMP1 was not present in cells transfected with the BART miRNAs. It is interesting to note that LMP1 is rarely expressed in EBV-associated carcinomas and this might reflect the repressive effects of the BART miRNAs that are produced at much higher levels in this context [2,5]. The BART cluster offers a good example of how the expression level influences the functions of miRNAs. They also show that miRNAs can have a marked influence on cellular functions even if expressed at seemingly low levels. However, it is important to note that the BART miRNAs are expressed on average at slightly higher levels than the BHRF1 miRNAs in LCLs or even than some crucial cellular miRNAs such as those belonging to the let7 family in hematopoietic stem cells in which miRNAs play a crucial role in cell differentiation [10,40].
In conclusion, we used recombinant viruses to reveal functions of the BART miRNA locus that result from multiple, sometimes even contradictory alterations of viral and cellular functions in cells infected in vitro and in vivo. This obviously reflects their high number, but also the fact that they collaborate to downregulate targets such as BZLF1 as shown in the present paper, or NDGR1 as previously reported [41]. Virus knockouts that lack single BART miRNA or a subset of them will provide useful tools to dissect their multiple and intricate molecular functions.
All human primary B cells used in this study were isolated from anonymous buffy-coats purchased from the Blood Bank of the University of Heidelberg. No ethical approval is required. All animal experiments were performed in strict accordance with German animal protection law (TierSchG) and were approved by the federal veterinary office at the Regierungspräsidium Karlsruhe, Germany (Approval number G156-12). The mice were housed in the class II containment laboratories of the German Cancer Research and handled in accordance with good animal practice with the aim of minimizing animal suffering and reducing mice usage as defined by Federation of European Laboratory Animal Science Associations (FELASA) and the Society for Laboratory Animal Science (GV-SOLAS).
HEK293 cell line is a neuro-endocrine cell line obtained by transformation of embryonic epithelial kidney cells with adenovirus (ATCC: CRL-1573) [42,43]. DG-75 is an EBV-negative cell line that was established from a pleural effusion of a patient with a primary abdominal lymphoma that resembled Burkitt’s lymphoma (ATCC: CRL-2625) [44]. Peripheral blood mononuclear cells from buffy coats purchased from the blood bank in Heidelberg were purified on a Ficoll cushion and CD19-positive primary B-lymphocytes were isolated using M-450 CD19 (Pan B) Dynabeads (Dynal) and were detached using Detachabead (Dynal). WI38 are primary human lung embryonic fibroblasts (ATCC: CCL-75). All cells were routinely cultured in RPMI-1640 medium (Invitrogen) supplemented with 10% fetal bovine serum (FBS)(Biochrom), and primary B cells were supplemented with 20% FBS until establishment of LCLs.
All synthesized oligonucleotides used for cloning or PCR are listed in S2 Table.
The wild type EBV strain M81 is available as a recombinant BACMID [21]. The viral genome was cloned onto a prokaryotic F-plasmid that carries the chloramphenicol (Cam) resistance gene, the gene for green fluorescent protein (GFP), and the Hygromycin resistance gene (B240). All PCR primers used for PCR cloning or chromosomal building are listed in S2 Table and are based on the M81 EBV sequence (GenBank accession number KF373730.1). Deletion of the miR-BART subcluster 1 (deletion from nt 139133 to nt 140132) generated ΔC1; deletion of the miR-BART subcluster 2 (deletion from nt 145492 to nt 148777) gave rise to ΔC2. These mutations were achieved by homologous recombination of the recombinant virus with a linear DNA fragment that encodes the kanamycin resistance gene, flanked by Flp recombination sites, and short DNA regions homologous to the regions immediately outside of the deletion to be obtained, as described [45]. The double knockout BART miRNA subcluster 1 plus subcluster 2 (ΔC1C2) was obtained by excising the kanamycin cassette present in ΔC1 with FLP recombinase, followed by a deletion of BART miRNA subcluster 2 via linear targeting with the PCR product that yielded ΔC2. The ΔAll mutant that lacks all BART miRNAs was obtained by exchanging the miR-BART-2’s seed region with an unrelated sequence in the ΔC1C2 recombinant virus using chromosomal building [45]. This mutagenesis generated an additional AclI restriction site and digestion with this enzyme allows distinction between the mutant and the wild-type sequences (S1 Fig). We applied the same strategy to the wild type M81 BAC to obtain a recombinant EBV that lacks mir-BART-2 only (Δb2). We constructed a revertant of ΔAll using chromosomal building on the basis of the original parental M81/ΔAll Bacmid before passaging in 293. Here the complete BART miRNA locus, from the miR-BART subcluster 1 to miR-BART2, was cloned from the M81 BAC and reintroduced into the M81/ΔAll BAC genome.
We introduced the rat CD2 gene under the control of an EA-D promoter into the BXLF1 gene of the M81 genome (nt 131044 to nt 133362) by homologous recombination using a linear vector that included the kanamycin resistance cassette as a selection marker. The disruption of BXLF1 gene does not interfere with the growth of LCLs [46,47]. Upon induction of the lytic replication, CD2 is expressed at the surface of replicating cells. CD2-positive cells can be pulled down with a specific monoclonal antibody (OX34) coupled with anti-mouse IgG Dynabeads and submitted to protein or RNA extraction.
Recombinant EBV plasmids were lipotransfected into HEK293 cells using Metafectene (Metafectene, Biontex) and the selection of stable 293 cell clones carrying the recombinant EBV plasmid was achieved by adding hygromycin to the culture medium (100 μg/ml) as previously described [48]. To assess the genome integrity of recombinant EBV within the stable clones, the circular EBV genomes present in these cells were extracted using a denaturation-renaturation method [49] and transferred into the E.Coli strain DH10B by electroporation (1000V, 25μF, 200 Ohms). The transformed E.Coli clones were further assessed by restriction enzyme analysis of plasmid minipreparations.
293 cells stably transfected with recombinant EBV-BACs were transfected with expression plasmids encoding BZLF1 (p509) and BALF4 (pRA) using the liposome-based transfectant Metafectene (Biontex). Three days after transfection, virus supernatants were collected and filtered through a 0.4 μm filter.
B1124 is a plasmid that contains a bi-directional tetracycline-inducible CMV promoter that encodes a truncated nerve growth factor receptor (NGFR) on one site, and the BART miRNA subcluster 1 and 2 without the PstI repeats and the LF3 gene on the other. B1034 contains only NGFR and was used as a negative control. M81/ΔAll LCLs were electroporated with either B1034 or B1124 and cultured with 1 μg/ml doxycycline for 30 days. NGFR-positive cells were isolated with specific antibodies and used for protein and RNA analyses.
To evaluate EBV genome equivalents per milliliter of supernatant, viral supernatants were treated with DNase I. Following a subsequent treatment with proteinase K, we used quantitative real-time PCR analysis (qPCR) with primers and probe specific for the non-repetitive EBV BALF5 gene sequence to measure the EBV copy numbers in the supernatants [50]. The quantification of EBV DNA genome copies per milliliter in the blood of infected mice was performed from genomic DNA extracted from total blood by using regular phenol extraction.
RNA extracted with Trizol from LCLs was reverse transcribed with AMV-reverse transcriptase (Roche) using a mix of random primers The primers and probes used to detect BZLF1 are listed in the S1 table. The PCR and data analysis was carried out using the universal thermal cycling protocol on an ABI STEP ONE PLUS Sequence Detection System (Applied Biosystems). All samples were run in duplicates, together with primers specific to the human GAPDH gene to normalize for variations in cDNA recovery.
BART miRNAs extracted from cells with Trizol were reverse transcribed using specific stem-loop primers and TaqMan miRNA reverse Transcription Kit (Applied Biosystems), as described elsewhere [8]. The sequences of the stem-loop primers, primers and probes are listed in S2 Table. Reverse transcription and amplification of the cellular snoRNA RNU48 was performed in parallel to normalize for cDNA recovery (Assay ID 001006; Applied Biosystems). Real-time PCR was performed on an ABI STEP ONE PLUS Sequence Detection System (Applied Biosystems).
B cells purified from peripheral blood were exposed to viral supernatant for two hours, then washed once with PBS and cultured with RPMI supplemented with 20% FBS in the absence of immunosuppressive drugs. For transformation assays, the percentages of EBNA2 positive cells in the infected samples were evaluated by immunostaining with a specific antibody at 3 days post-infection (dpi). Cell populations containing 3 or 30 EBNA2-positive cells per well were seeded into 48 wells of U-bottomed 96-well plates that contained 103 gamma-irradiated WI38 feeder cells. Non-infected B cells were used as a negative control. Outgrowth of lymphoblastoid cell clones (LCLs) was monitored at 33 dpi. We also incubated 105 primary B cells with 25ml of cell-free LCL culture supernatants for 2 hours. These cells were plated on 96 well cluster plates at a concentration of 103 cells per well, together with the same number of gamma-irradiated feeder cells.
We stained infected cells with mouse monoclonal antibodies against BZLF1 (Clone BZ.1), gp350/220 (Clone 72A1), EBNA2 (Clone PE2) and a Cy-3-conjugated goat-anti-mouse secondary antibody (Dianova, Invitrogen). We performed western blots with mouse monoclonal antibodies against BZLF1 (Clone BZ.1), gp350 (Clone OT6), DICER (clone F10, Santa Cruz Biotechnology), LMP1 (clone CS1-4), EBNA2 (clone PE2), and Actin (clone ACTN05,C4, Dianova). We also used rabbit polyclonal antibodies against caspase-3 (Cell Signalling Technology) and rat monoclonal antibodies (kindly provided by Dr. E. Kremmer Helmholtz Zentrum Munich and Dr. F. Grässer, University of Homburg) against Ago2 (clone 11A9;), EBNA3A (clone E3AN-4A5), EBNA3B (clone 6C9), EBNA3C (clone A10), LMP2A (clone 4E11), BALF5 (clone 4C12). Mouse monoclonal antibodies specific to LMP1 (clone S12, BD Pharmingen), BZLF1 (Clone BZ.1) and gp350 (Clone OT6) were used for immunohistochemical staining against EBV proteins in infected murine tissues.
Cells were fixed with 4% paraformaldehyde in PBS for 20 min at room temperature and permeabilized in PBS 0.5% Triton X-100 for 2 min except for samples stained for viral glycoproteins (gp350). Cells were incubated with the first antibody for 30 min, washed in PBS three times, and incubated with a secondary antibody conjugated to Cy-3 for 30 min before embedding in 90% glycerol.
Proteins were extracted with a standard lysis buffer (150 mM NaCl, 0.5% NP-40, 1% Sodium deoxycholat, 0.1% SDS, 5 mM EDTA, 20 mM Tris-HCl pH7.5, proteinase inhibitor cocktail (Roche)) for 15 min on ice followed by sonication to shear the genomic DNA. Up to 20μg of proteins denatured in Laemmli buffer for 5 min at 95 degree were separated on SDS-polyacrylamide gels and electroblotted onto a nitrocellulose membrane (Hybond C, Amersham). Proteins extracted to assess gp350 expression were prepared in Laemmli buffer without 2-mercaptoethanol. After pre-incubation of the blot in 3% milk dry powder in PBST (PBS with 0.2% Triton-X100), the antibody against the target protein was added and incubated at room temperature for 1 hr. After extensive washings in PBST, the blot was incubated for 1 hr with suitable secondary antibodies coupled to horseradish peroxidase (goat anti-mouse (Promega), goat anti-rabbit (Life technologies), or rabbit anti-goat (Santa Cruz) IgG). Bound antibodies were revealed using the ECL detection reagent (Pierce).
6 × 108 cells were washed twice in ice-cold PBS and subsequently lysed in 5 ml lysis buffer containing 25 mM Tris HCl (pH 7.5), 150 mM KCl, 2 mM EDTA, 0.5% NP-40, 0.5 mM DTT, 200 u/ml RNAse inhibitor and protease inhibitor cocktail (Roche). Lysates were incubated for 30 min on ice and clarified by centrifugation at 20,000 g for 30 min at 4°C. To estimate the recovery of miRNAs after RISC-IP, total RNA was prepared from 10% of the cell lysates using the TRIzol RNA Isolation Reagents (Life technologies) following the manufacturer's instructions. 6 μg of purified Rat-monoclonal anti-hAgo2 antibody (11A9; Helmholtz Zentrum Munich) or of monoclonal anti-BrdU-antibody (Abcam) was mixed with 20 μl of Dynabeads Protein G (Dynabeads Protein G Immunoprecipitation Kit, Life technologies) and subsequently incubated with 2.5 ml of cell lysates for 4–6 hours under constant rotation at 4°C. The beads were then washed four times with IP wash buffer (300 mM NaCl, 50 mM Tris HCl pH 7.5, 5 mM MgCl2, 0.1% NP-40, 1 mM NaF) and once with PBS to remove residual detergents. The beads were resuspended with 300μl of proteinase K buffer (100 mM Tris-HCl PH 7,4 / 50 mM NaCl /10 mM EDTA) in the presence of proteinase K (0.33 mg/ml) and incubated for exactly 30 min at 37°C with shaking at 600 rpm and immediately transferred onto ice. Total RNA present in complexes after RISC-IP was purified by using TRIzol RNA Isolation Reagents and dissolved in 50 μl RNase-free water.
The 3’UTR regions of the M81 BZLF1 and BALF5 genes were PCR amplified with the pair of primers listed in S2 table. The PCR products were digested with EcoRI and XhoI and ligated into the firefly luciferase expressing vector, pGL4.5 (Promega), which had firstly been modified to insert EcoRI and XhoI cutting sites behind the luciferase coding region. The luciferase reporter assays were performed by electroporation of 10 million LCL cells with 5μg of a pcDNA3.1-CD2 plasmid that encodes a truncated rat CD2 protein and 10μg of the different luciferase expression plasmids. 48 hours post electroporation, cells were washed twice with PBS and the luciferase activity was measured by the Beetle-Juice firefly luciferase assay system (PJK). The transfected cells were also immunostained with an antibody specific to rat CD2 to evaluate the electroporation efficiency.
Total RNA isolated from LCLs or equal volumes of RNA post RISC-IP from different samples were reverse transcribed with AMV-reverse transcriptase (Roche) using a mix of random hexamers. The mRNAs of interest were quantified with the Power SYBR green PCR Mix (Life technologies) using primer pairs specific to the gene of interest. Data analysis was carried out with the universal thermal cycling protocol of the ABI STEP ONE PLUS Sequence Detection System (Applied Biosystems).
We generated humanized mice by intrahepatical injection of human CD34-positive hematopoietic progenitor cells (HPCs) in irradiated (1 Gy) newborn NSG-A2 mice (NOD.Cg-PrkdcscidIl2rgtm1WjlTg (HLA-A2.1) 1Enge/SzJ) (huNSG-A2 mice). CD34-positive HPCs were isolated from human fetal liver tissue (Advanced Bioscience Resources, Alameda, CA, USA) using a human CD34 purifying Microbead kit (Miltenyi Biotec). We used 2 liver samples to generate the 12 humanized mice used in this study. The percentage of human CD45 positive cells was evaluated 12 weeks after HPC transplantation and only mice with more than 40% of human CD45-positive cells were infected. We performed a titration of viral stocks by infecting primary B cells with increasing dilutions. The infected cells were stained at 3 dpi to determine an EBNA2-positive B cells/ml virus titer. In all experiments, we injected intra-peritoneally in each mouse enough viruses to generate 5x106 EBNA2-positive cells as described [21]. Peripheral blood samples were drawn 5 weeks post-infection. Mice were euthanized at week 8 except if signs of animal suffering became apparent and we examined their blood and tissues for signs of viral infection.
We isolated human CD19+ B cells from buffy coats and exposed to virus supernatants at a moi sufficient to generate 20% of EBNA2-positive cells for 2 hours at room temperature under constant agitation. The infected cells were collected by centrifugation and washed twice with PBS for two times. 2*10ˆ6 infected primary B cells, equivalent to 4*10ˆ5 infected cells were injected intraperitoneally into NSG mice. The mice were euthanized at 5 weeks post injection, autopsied and their organs were subjected to macroscopic and microscopic investigation.
The organs from the studied mice were fixed in 10% formalin overnight and embedded in paraffin blocks. 3-μm-thin continuous sections were prepared and submitted to antigen retrieval at 98°C for 40 min in a 10 mM sodium citrate, 0.05% Tween20 pH 6.0 solution. Bound antibodies were visualized with the Envision+ Dual link system-HRP (Dako). In parallel, adjacent sections were stained with hematoxylin and eosin (H&E). The presence of EBV was detected by in situ hybridization with an EBER-specific PNA probe, in conjunction with a PNA detection kit (Dako). Pictures were taken with a camera attached to a light microscope (Axioplan, Zeiss).
We induced apoptosis in LCLs (5*105 cell per well of a 48-well-plate) transformed by ΔAll or wild type viruses at 40–60 dpi by adding Etoposide (4μg/ml, Sigma Aldrich) or Staurosporine (4μg/ml, Sigma-Aldrich) to the culture medium for 20 hrs. Cells were also treated with Ionomycin (4μg/ml, Sigma Aldrich) for 48 hrs or Simvastatin (2μM, Calbiochem) for 5 days. We included DMSO-treated cells and ethanol-treated cells as controls. Cells were then washed twice with ice-cold PBS, dried on glass slides and fixed with 4% paraformaldehyde in PBS to perform a TUNEL assay that labels apoptotic cells with DNA breaks (Cell Death Detection Kit, TMR red, Roche) following the instruction of manufacturer. Cells were also stained with a rabbit antibody specific for cleaved caspase 3 (Cell signal technology).
All results obtained in in vitro studies with LCLs generated by EBV wild type or mutants with B cells from the same blood donors were paired and analyzed by paired student t-test. Unpaired student t-test was applied for analyzing the grouped humanized or non-humanized NSG mice infected by either M81 or M81/ΔAll virus. All p-values were analyzed as 2-tailed and the values equal to 0.05 or less were considered significant unless indicated. We used a Chi square test to analyze the tumor incidence in humanized and non-humanized mice. The statistical analyses were performed with the GraphPad Prism 5 software.
|
10.1371/journal.ppat.1006129 | Respiration of Microbiota-Derived 1,2-propanediol Drives Salmonella Expansion during Colitis | Intestinal inflammation caused by Salmonella enterica serovar Typhimurium increases the availability of electron acceptors that fuel a respiratory growth of the pathogen in the intestinal lumen. Here we show that one of the carbon sources driving this respiratory expansion in the mouse model is 1,2-propanediol, a microbial fermentation product. 1,2-propanediol utilization required intestinal inflammation induced by virulence factors of the pathogen. S. Typhimurium used both aerobic and anaerobic respiration to consume 1,2-propanediol and expand in the murine large intestine. 1,2-propanediol-utilization did not confer a benefit in germ-free mice, but the pdu genes conferred a fitness advantage upon S. Typhimurium in mice mono-associated with Bacteroides fragilis or Bacteroides thetaiotaomicron. Collectively, our data suggest that intestinal inflammation enables S. Typhimurium to sidestep nutritional competition by respiring a microbiota-derived fermentation product.
| Salmonella enterica serovar Typhimurium induces intestinal inflammation to induce the generation of host-derived respiratory electron acceptors, thereby driving a respiratory pathogen expansion, which aids infectious transmission by the fecal oral route. However, the identity of nutrients serving as electron donors to enable S. Typhimurium to edge out competing microbes in the competitive environment of the gut are just beginning to be worked out. Here we demonstrate that aerobic and anaerobic respiratory pathways cooperate to promote growth of Salmonella on the microbial fermentation product 1,2-propanediol. We propose that pathogen-induced intestinal inflammation enables Salmonella to sidestep nutritional competition with the largely anaerobic microbiota by respiring a microbe-derived metabolite that cannot be consumed by fermentation.
| Salmonella enterica serovar Typhimurium (S. Typhimurium) is a common cause of food poisoning. Upon ingestion, the pathogen enters the intestinal epithelium using the invasion-associated type III secretion system (T3SS-1) [1] and deploys a second type III secretion system (T3SS-2) to survive in host tissue [2]. This virulence strategy results in acute intestinal inflammation and diarrhea [3]. Interestingly, gut inflammation increases the abundance of the pathogen within the gut-associated microbial community [4] by generating a respiratory nutrient-niche (reviewed in [5]). One respiratory electron acceptor generated as a byproduct of the inflammatory response is tetrathionate, which confers a luminal growth advantage upon S. Typhimurium in a mouse colitis model [6] by enabling the pathogen to consume ethanolamine in the gut [7].
Genes involved in tetrathionate respiration and ethanolamine-utilization are intact in Salmonella serovars associated with gastroenteritis in humans, but are often disrupted in Salmonella serovars associated with extraintestinal disease [8–10]. Another pathway often disrupted in Salmonella serovars associated with extraintestinal disease is the utilization 1,2-propanediol [8–10], which is produced during the fermentation of rhamnose or fucose [11]. Genes required for the degradation of 1,2-propanediol are encoded by the porR pduF pduABCDEGHJKLMNOPQSTUVWX gene cluster [12], a DNA region conserved among S. enterica serovars, but absent from the closely related species S. bongori [13–15]. Both ethanolamine and 1,2-propanediol-utilization proceeds through a pathway that requires a microcompartment, a respiratory electron acceptor and the cofactor cobalamin [16]. Cobalamin biosynthesis genes are only expressed when S. Typhimurium is cultured under anaerobic or microaerobic conditions [17, 18]. Under anaerobic conditions, tetrathionate can serve as an electron acceptor to support in vitro growth of S. Typhimurium on 1,2-propanediol and ethanolamine by using endogenously synthesized cobalamin [19]. Based on these observations it has been proposed that tetrathionate respiration might enable S. Typhimurium to consume microbiota-derived 1,2-propanediol in the inflamed gut, thereby driving a luminal pathogen expansion [20]. We analyzed the fitness of S. Typhimurium mutants in gnotobiotic or conventional mice to test this prediction.
To determine whether 1,2-propanediol-utilization confers a benefit in the environment of the large intestine, we constructed a S. Typhimurium strain lacking the pduABCDEGHJKLMNOPQSTUVWX gene cluster (pduA-X mutant) (Fig 1A). Compared to the S. Typhimurium wild type strain the pduA-X mutant grew poorly in minimal medium containing 1,2-propanediol as carbon source (Fig 1B), but showed no growth defect in rich medium (S1A Fig). Genetically resistant (CBA) mice were infected intragastrically with a 1:1 mixture of the S. Typhimurium wild type (IR715) and a pduA-X mutant (FF128) to compare the fitness of both strains. By 14 days after infection, the S. Typhimurium wild type was recovered in significantly higher numbers from cecal and colon contents than a pduA-X mutant (P < 0.01), suggesting that 1,2-propanediol utilization conferred a benefit for growth of the pathogen in the large intestine (Figs 1C and S2A). Similar results were obtained with a S. Typhimurium mutant carrying an insertion in the pduC gene (S1B and S2B Figs). After bypassing the gut by the intraperitoneal route of inoculation, the S. Typhimurium wild type was recovered in similar numbers as a pduA-X mutant from organs of mice infected with a 1:1 mixture of both strains (S2C Fig). Growth of the pduA-X mutant in genetically resistant mice could be restored by re-introducing the intact pdu operon through transduction (FF484) (S2D Fig).
Next, we infected mice with either the S. Typhimurium wild type or a pduA-X mutant. By 14 days, mice infected with the S. Typhimurium wild type carried a significantly higher pathogen burden in colon contents than mice infected with a pduA-X mutant (P < 0.05) (Fig 1D), although no significant differences in the severity of intestinal pathology between both groups were noted (S2E Fig).
Since genetically resistant mice develop severe intestinal inflammation by approximately 10 days after infection [21], we wanted to investigate whether the fitness advantage conferred by 1,2-propanediol utilization genes required severe colitis induced by virulence factors. To this end, T3SS-1 and T3SS-2 were inactivated using mutations in the invA and spiB genes, respectively. When mice (CBA) were infected intragastrically with a 1:1 mixture of an invA spiB mutant (FF183) and a invA spiB pduA-X mutant (FF383), both strains were recovered in similar numbers from cecal and colon contents 14 days after infection (Fig 1C). Mice infected with virulent S. Typhimurium strains (i.e. a mixture of wild type and pduA-X mutant) developed severe acute inflammation in the cecal mucosa, while no marked inflammatory changes were observed in mice infected with avirulent S. Typhimurium strains (i.e. a mixture of invA spiB mutant and invA spiB pduA-X mutant) (Fig 1E and 1F). These data suggested that the utilization of 1,2-propanediol conferred a fitness advantage upon S. Typhimurium during growth in the inflamed intestine.
A variety of exogenous electron acceptors can be generated during intestinal inflammation, including tetrathionate, nitrate, organic S-oxides (such as DMSO) or organic N-oxides (such as TMAO), which support growth by anaerobic respiration (summarized in [22]). We next investigated whether exogenous electron acceptors would support anaerobic growth of S. Typhimurium on 1,2-propanediol in vitro. To this end, minimal medium containing 1,2-propanediol as the sole carbon source was supplemented with tetrathionate, nitrate, dimethyl sulfoxide (DMSO) or trimethylamine N-oxide (TMAO) and inoculated with a 1:1 mixture of the S. Typhimurium wild type and a pduA-X mutant. The ability to utilize 1,2-propanediol conferred the largest fitness advantage in media supplemented with tetrathionate, followed by media supplemented with nitrate, DMSO and TMAO (Fig 2A). In contrast, 1,2-propanediol-utilization did not confer a growth advantage in the absence of exogenous electron acceptors, which was consistent with a previous report [19]. Growth of the pduA-X mutant on 1,2-propanediol in media supplemented with tetrathionate could be restored by re-introducing the intact pdu operon through transduction (FF484).
Since tetrathionate becomes available during inflammation [6], it has been proposed that this electron acceptor might enable S. Typhimurium to utilize 1,2-propanediol during colitis [20]. To test this prediction, we inactivated the ttrA gene, encoding tetrathionate reductase and infected genetically resistant (CBA) mice intragastrically with a 1:1 mixture of a S. Typhimurium ttrA mutant (SW661) and a ttrA pduA-X mutant (PT305). Surprisingly, 1,2-propanediol-utilization conferred a fitness advantage even after genetic ablation of tetrathionate respiration by a mutation in ttrA (Fig 2B). These data refuted the hypothesis that tetrathionate respiration was necessary for the growth benefit conferred by 1,2-propanediol-utilization.
The enzymes that enable S. Typhimurium to use tetrathionate, nitrate, organic S-oxides (such as DMSO) or organic N-oxides (such as TMAO) as respiratory electron acceptors under anaerobic conditions contain molybdopterin, a cofactor also required for the use of formate as an electron donor (reviewed in [22]). To determine whether anaerobic respiration contributed to 1,2-propanediol-utilization, we inactivated the S. Typhimurium moaA gene, encoding the enzyme catalyzing the first step in the molybdopterin cofactor biosynthesis. Genetically resistant (CBA) mice were inoculated intragastrically with a 1:1 mixture of the S. Typhimurium wild type and a pduA-X mutant or with a 1:1 mixture of a moaA mutant (FF294) and a moaA pduA-X mutant (FF284). Interestingly, mutational inactivation of moaA did not abrogate the fitness advantage conferred by 1,2-propanediol-utilization (Fig 3A). These data suggested that exogenous electron acceptors for anaerobic respiration were not solely responsible for the ability of S. Typhimurium to utilize 1,2-propanediol during colitis.
S. Typhimurium-induced colitis is accompanied by increased epithelial oxygenation, driving a cytochrome bd-II oxidase-dependent aerobic pathogen expansion by 10 days after infection of genetically resistant (CBA) mice [21]. We therefore deleted cyxA, the gene encoding cytochrome bd-II oxidase, to investigate whether aerobic respiration contributed to 1,2-propanediol-utilization. Genetically resistant (CBA) mice were inoculated intragastrically with a 1:1 mixture of a cyxA mutant (FF286) and a cyxA pduA-X mutant (FF288). Deletion of cyxA did not abrogate the fitness advantage conferred by 1,2-propanediol-utilization (Fig 3A).
To determine whether 1,2-propanediol-utilization involved cooperation between aerobic and anaerobic respiration, genetically resistant (CBA) mice were infected with a 1:1 mixture of a cyxA moaA mutant (FF296) and a cyxA moaA pduA-X mutant (FF292). Remarkably, genetic ablation of both aerobic respiration (through inactivation of cyxA) and anaerobic respiration (through inactivation of moaA) abrogated the fitness advantage conferred by 1,2-propanediol-utilization (P 0.05) (Fig 3A).
Next, we studied the role of 1,2-propanediol-utilization in mice that were genetically susceptible to S. Typhimurium infection (C57BL/6 mice). Genetically susceptible (C57BL/6) mice become moribund before developing severe acute intestinal inflammation during S. Typhimurium infection. However, preconditioning of C57BL/6 mice with streptomycin disrupts the resident microbiota and leads to severe acute cecal inflammation during S. Typhimurium infection [23]. Streptomycin-treated C57BL/6 mice were infected intragastrically with a 1:1 mixture of the S. Typhimurium wild type and a pduA-X mutant or with a 1:1 mixture of an invA spiB mutant and a invA spiB pduA-X mutant. Four days after infection, 1,2-propanediol-utilization conferred a fitness advantage upon S. Typhimurium, as indicated by higher recovery of the wild type than a pduA-X mutant (Fig 3B and S3 Fig). In contrast, similar numbers of the invA spiB mutant and an invA spiB pduA-X mutant were recovered from colon contents. These data provided further support for the idea that benefit conferred by 1,2-propanediol-utilization required the presence of virulence factors.
In streptomycin-treated C57BL/6 mice, inactivation of either moaA or cyxA reduced the fitness advantage conferred by 1,2-propanediol-utilization (Fig 3B). Collectively, these data provided further support for the idea that S. Typhimurium used both aerobic and anaerobic respiration to consume 1,2-propanediol during colitis.
Saccharolytic bacteria in the large intestine break down complex carbohydrates, thereby liberating monosaccharides, such as fucose and rhamnose, which can be fermented to generate 1,2-propanediol [11]. To investigate whether utilization of 1,2-propanediol by S. Typhimurium required members of the gut-associated microbial community, germ-free (Swiss-Webster) mice were infected with a 1:1 mixture of the S. Typhimurium wild type and a pduA-X mutant. Equal recovery of both strains from colon contents suggested that in the absence of the gut microbiota, 1,2-propanediol-utilization did not confer a luminal growth advantage in the mouse intestine (Fig 4A and 4B). When mice mono-associated with Bacteroides fragilis or Bacteroides thetaiotaomicron were infected with a 1:1 mixture of the S. Typhimurium wild type and a pduA-X mutant, 1,2-propanediol-utilization conferred a fitness advantage (Fig 4A and 4B), which was not due to changes in intestinal inflammation between the groups (Fig 4C and 4D), suggesting that S. Typhimurium required the presence of saccharolytic bacteria to utilize 1,2-propanediol in the gut.
Next, we determined whether B. thetaiotaomicron was a source of 1,2-propanediol in the large intestine. To this end, germ-free mice were mock-inoculated or mono-associated with B. thetaiotaomicron and the concentration of 1,2-propanediol in colon contents was determined by gas chromatography/mass spectrometry (GC/MS). In mock-inoculated germ-free mice, the concentration of 1,2-propanediol was close to the limit of quantification. In contrast, 1,2-propanediol was present at a concentration of approximately 40 μM in B. thetaiotaomicron mono-associated mice (Fig 4E). These data suggested that B. thetaiotaomicron is a source of 1,2-propanediol in the large intestine. To investigate whether consumption of 1,2-propanediol by S. Typhimurium would reduce the luminal concentration of this metabolite, mice mono-associated with B. thetaiotaomicron were infected with the S. Typhimurium wild type or a pduA-X mutant. While there was a trend that the luminal 1,2-propanediol concentrations were lower in mice infected with the wild type compared to those infected with the pduA-X mutant, this difference did not reach statistical significance (Fig 4E).
Next, we determined whether the ability of the pathogen to generate 1,2-propanediol through fermentation of rhamnose or fucose was necessary for its ability to utilize 1,2-propanediol. To this end, we generated S. Typhimurium mutants lacking rhaBAD or fucO, respectively. A S. Typhimurium rhaBAD mutant was unable to ferment rhamnose (Fig 4F), but exhibited no growth defect in complex culture medium (S1C Fig). Inactivation of 1,2-propanediol utilization by deleting pduC or pduA-X did not abrogate the ability of S. Typhimurium to ferment rhamnose or fucose (Fig 4F). The fucO gene encodes 1,2-propanediol oxidoreductase, the enzyme catalyzing the interconversion of lactaldehyde to 1,2-propanediol [24]. Deletion of fucO did not abrogate the ability of S. Typhimurium to ferment fucose (Fig 4F) and did not produce a growth defect in complex culture medium (S1C Fig). We next generated spent culture medium by growing a pduC mutant or a fucO mutant anaerobically in minimal medium containing fucose as a sole carbon source. The spent culture medium generated in this fashion was sterilized, supplemented with tetrathionate and inoculated with a 1:1 mixture of the S. Typhimurium wild type and a pduC mutant. 1,2 propanediol-utilization did not confer a fitness advantage for anaerobic growth in spent culture medium from a fucO mutant (Fig 4G), suggesting that the fucO mutant was no longer able to generate 1,2-propanediol as a product of fucose fermentation. We then investigated whether S. Typhimurium strains unable to generate 1,2-propanediol from fucose (fucO mutant), rhamnose (rhaBAD mutant) or either pentose (fucO rhaBAD mutant) were still able to use the pdu operon for expansion in the large intestine of gnotobiotic mice mono-associated with B. thetaiotaomicron. The pdu operon still conferred a significant fitness advantage upon S. Typhimurium strains that could generate 1,2-propanediol neither from fucose nor from rhamnose (Fig 4H), suggesting that 1,2-propanediol generated by B. thetaiotaomicron (Fig 4E) was sufficient to stimulate growth of the pathogen.
Collectively, these data demonstrated that microbe-derived 1,2 propanediol was a critical carbon source for S. Typhimurium, which drove an expansion of the pathogen population during colitis.
Associations revealed by genome and literature mining predict that propanediol utilization pathways are genomic determinants of pathogenicity associated with food poisoning, presumably by promoting growth in the low-oxygen environment of the large intestine [25]. For example, adherent-invasive Escherichia coli (AIEC) isolated from intestines of patients with Crohn's disease (CD) contain genes encoding 1,2-propanediol-utilization, while the gene cluster is commonly absent from commensal E. coli isolates [26]. Comparative genomic analysis of S. enterica serovars suggest that loci involved in 1,2-propanediol utilization are intact in pathogens causing gastrointestinal disease, such as S. enterica serovars Typhimurium or Enteritidis, but are often disrupted in closely related pathogens associated exclusively with extraintestinal disease, such as S. enterica serovars Typhi, Paratyphi A, Gallinarum or Choleraesuis [8–10].
S. Typhimurium growth under conditions mimicking the low-oxygen high-osmolarity environment of the gut induces the synthesis of proteins involved in the utilization of 1,2-propanediol [27]. Transcriptional profiling suggests that 1,2-propanediol utilization genes are expressed during S. Typhimurium growth in the ceca of chickens [28] and in gnotobiotic mice mono-associated with B. thetatiotaomicron [29]. Collectively, studies on gene regulation and the in silico predictions reviewed above imply that 1,2-propanediol utilization genes might contribute to S. Typhimurium growth in the large intestine, but the present report is the first to test this hypothesis using an animal model. Our results provide compelling experimental support for the idea that 1,2-propanediol-utilization contributes to growth of S. Typhimurium during gastroenteritis, suggesting that the pathogen thrives during gut inflammation in part by respiring fermentation products generated by other microbes.
While tetrathionate respiration is necessary for growth on ethanolamine in vivo [7], efficient 1,2-propanediol-utilization in the mouse intestine required both aerobic and anaerobic respiration. Interestingly, 1,2-propanediol represses transcription of ethanolamine utilization genes to avoid detrimental mixing of shell proteins from the corresponding microcompartments [30]. Thus, the finding that different electron acceptors are required to breakdown 1,2-propanediol and ethanolamine in vivo may reflect the need to prevent simultaneous expression of the pdu and eut gene clusters. In turn, this may restrict utilization of 1,2-propanediol and ethanolamine to microenvironments that differ with regard to electron acceptor availability. However, additional work is needed to test this prediction.
1,2-propanediol is generated by the gut microbiota through fermentation of methyl-pentoses, such as fucose or rhamnose [11]. Glycoside hydrolases and polysaccharide lyases expressed by the gut microbiota can liberate fucose from complex carbohydrates [29] (Fig 5). Members of the class Bacteroidia encode the most diverse array of glycoside hydrolases and polysaccharide lyases, suggesting that this group possesses the largest carbohydrate substrate range within the gut-associated microbial community [31]. Consistent with this idea, 1,2-propanediol-utilization conferred a benefit upon S. Typhimurium in mice mono-associated with B. fragilis or B. thetaiotaomicron, but not in germ-free mice.
The picture emerging from this and previous work is that respiratory electron acceptors become available during S. Typhimurium-induced colitis [6, 21, 32, 33], which enables the pathogen to consume 1,2-propanediol, a process that could be viewed as “dumpster diving” for a fermentation product produced by the gut microbiota [34] (Fig 5). Respiration of 1,2-propanediol allows S. Typhimurium to side-step nutritional competition with the fermenting gut microbiota to drive its expansion within the lumen of the large bowel. This outcome is biologically relevant, because an uncontrolled expansion of S. Typhimurium during colitis is important for infectious transmission by the fecal oral route [21, 35].
This study was performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health and approved by the Institutional Animal Care and Use Committee at the University of California at Davis (protocols #15449, 17140, 17939 and 19235).
All strains used in this study are listed in Table 1. S. Typhimurium and E. coli cultures were routinely grown aerobically at 37°C in either Luria-Bertani (LB) broth (10 g/l tryptone, 5 g/l yeast extract, 10 g/l NaCl) or on LB agar plates (1.5% Difco agar) unless indicated otherwise. When necessary, antibiotics were added to the medium at the following concentrations: Nalidixic acid (Nal) 50 mg/l, Kanamycin (Km) 100 mg/l, Chloramphenicol (Cm) 30 mg/l, Carbenicillin (Carb) 100 mg/L. B. thetaiotaomicron and B. fragilis were routinely grown inside an anaerobe chamber (Bactron I Anaerobic Chamber; Sheldon Manufacturing, Cornelius) at 37°C in either Tryptic Soy Broth (TSB) or on Tryptic Soy agar (TSA) plates (1.5% [w/v] Difco agar) supplemented with 5% (v/v) sheep blood.
All strains, plasmids and primers used in this study are listed in Tables 1–3. Standard cloning techniques were used to generate the plasmids used in this study. PCR products were confirmed by sequencing (SeqWright Fisher Scientific, Houston). Suicide plasmids were propagated in E. coli DH5α λpir. Phage P22 HT int-105 was used for generalized transduction. Transductants were cleaned from phage contamination on Evans blue-Uranine (EBU) plates and tested for phage sensitivity by cross-streaking against P22 H5.
To construct the pduA-X mutant (FF128), a region upstream and downstream of pduA and pduX, respectively, were PCR amplified from the S. Typhimurium wild-type strain IR715. Both PCR products (the flanking regions of pduA and pduX genes) were subcloned into pCR2.1 to obtain pPT33 and pPT34, respectively. The inserted PCR fragments were confirmed by sequencing. The pduA fragment from pPT33 was cloned into the suicide plasmid pGP704 by using SalI and XbaI restriction sites to obtain pPT35. The pduX fragment from pPT34 was digested by XbaI and SmaI and then inserted into pPT35 to yield pPT36. Next, the Kanr cassette from pBS34 was inserted into pPT36 to obtain pPT37. Plasmid pPT37 was transformed into E. coli S17-1λpir. Plasmid pPT37 was conjugated into the S. Typhimurium wild type (IR715) and an invA spiB mutant (SPN487) using E. coli S17-1λpir as a donor strain to generate FF128 (IR715 pduA-X::KanR) and FF383 (IR715 invA spiB pduA-X::KanR), respectively. Exconjugants were plated onto LB+Nal+Kan and tested for CarbS to select for clones that carried the deletion.
To construct the moaA mutant, a region upstream and downstream of moaA were PCR amplified from the S. Typhimurium wild-type strain IR715 introducing XbaI and SalI restriction sites into both fragments. Then, both PCR products were digested with XbaI ligated together and cloned into pRDH10 using SalI restriction sites yielding plasmid pCAL25. The inserted PCR fragments were confirmed by sequencing. Plasmid pCAL25 was conjugated into S. Typhimurium IR715 (wild-type) and FF176 (phoN::Tn10d-Cam) using E. coli S17-1λpir as a donor strain to generate FF283 (IR715 moaA) and FF294 (phoN::Tn10d-Cam moaA), respectively. Exconjugants were plated onto LB+Nal+Cm to select for clones that had integrated the suicide plasmid. Sucrose counter-selection was performed as published previously [36]. Strains that were sucrose resistant and CmS were verified by PCR.
The pduC mutant was constructed using the λ Red recombinase method as previously described [37]. PCR products were generated using primers STM20405’+P1 and STM20403’+S2 with pKD4 as the DNA template [37]. The appropriate length of the PCR products, approximately 1.6 kb, was ensured by agarose gel electrophoresis. The PCR product was electroporated into electrocompetent ATCC14028 carrying pKD46, a temperature sensitive plasmid which carries λ Red recombinase under an arabinose inducible promoter [37]. The recipient was prepared as previously described [37]. Following electroporation, transformants were allowed to recover for 1 hour at 37°C and were then plated onto LB+Kan and grown at 37°C to select for clones that had integrated the PCR product. Insertion of the KanR was confirmed by PCR using primers STM2040 5’ with previously described primer k1 and STM2040 3’ with primer k2 [37]. The mutation was also confirmed by Southern blot using a digoxigenin-labeled probe generated by PCR using primers k1 and kt with pKD4 DNA as template. A P22 lysate of the resultant strain was used to transduce the pduC::KanR into IR715 to generate SDL175.
To construct the rhaBAD mutant, regions flanking the rhaBAD operon were PCR amplified from the S. Typhimurium wild-type strain IR715 and cloned into BamHI-digested pRDH10 using Gibson Assembly Master Mix (NEB) yielding plasmid pRhaBAD. Plasmid pRhaBAD was conjugated into S. Typhimurium IR715 (wild-type) and FF128 (pduA-X::KanR) using E. coli S17-1λpir as a donor strain to generate FF489 (IR715 rhaBAD) and FF495 (rhaBAD pduA-X::KanR), respectively. Exconjugants were plated onto LB+Nal+Cm to select for clones that had integrated the suicide plasmid. Sucrose counter-selection was performed as published previously [36]. Strains that were sucrose resistant and CmS were verified by PCR.
To construct the fucO mutant, an internal fragment of the fucO gene was PCR amplified from the S. Typhimurium wild-type strain IR715 and cloned into pCR2.1 to generate pPT20. The fucO fragment from pPT20 was cloned into suicide plasmid pGP704 to obtain pPT21. Using E. coli S17-1λpir as a donor strain, plasmid pPT21 was conjugated into S. Typhimurium wild type (IR715), the phoN::Tn10d-Cam rhaBAD mutant (FF499) and the rhaBAD pduA-X::KanR mutant (FF495) to generate PT206 (IR715 fucO::pPT21), FF505 (IR715 phoN::Tn10d-Cam rhaBAD fucO::pPT21) and FF503 (IR715 rhaBAD fucO::pPT21 pduA-X::KanR), respectively. Exconjugants were plated onto LB+Nal+Carb to select for clones that carried the plasmid insertion.
A P22 lysate of strain FF128 was used to transduce the pduA-X::KanR mutation into SW661 (ttrA), FF283 (moaA), FR102 (cyxA::pPT48) and PT206 (fucO::pPT21) to obtain strains PT305 (ttrA::pSW171 pduA-X::KanR) FF284 (moaA pduA-X::KanR), FF288 (cyxA::pPT48 pduA-X::KanR) and FF497 (fucO::pPT21 pduA-X::KanR), respectively.
A P22 lysate of strain FR102 was used to transduce the cyxA::pPT48 mutation into FF176 (phoN::Tn10d-Cam), FF294 phoN::Tn10d-Cam moaA) and FF284 (moaA pduA-X::KanR) to generate strains FF286 (phoN::Tn10d-Cam cyxA::pPT48), FF296 (phoN::Tn10d-Cam moaA cyxA::pPT48) and FF292 (IR715 moaA cyxA::pPT48 pduA-X::KanR), respectively.
A P22 lysate of strain FF176 was used to transduce the phoN::Tn10d-Cam mutation into FF283 (IR715 moaA), FF489 (rhaBAD) and PT206 (fucO::pPT21) yielding strains FF294 (IR715 phoN::Tn10d-Cam moaA), FF499 (phoN::Tn10d-Cam rhaBAD) and FF509 (phoN::Tn10d-Cam fucO::pPT21), respectively.
To restore the pduA-X::KanR mutant (FF128), an internal fragment of the pduP gene (part of the pduA-X operon) was PCR amplified from the S. Typhimurium wild-type strain IR715 and cloned into pCR2.1 to generate pCR-pduP. The pduP fragment from pCR2.1 was cloned into suicide plasmid pRDH10 to obtain pRDH10-pduP. Using E. coli S17-1λpir as a donor strain, plasmid pRDH10-pduP was conjugated into S. Typhimurium wild type (IR715) yielding strain FF480. A P22 lysate of strain FF480 was used to transduce pduP::pRDH10 into FF128. Sucrose counter-selection was performed as published previously [36]. Strains that were sucrose resistant and CmS were verified by PCR.
All animal experiments were approved by the Institutional Animal Care and Use Committees at the University of California, Davis. Female C57BL/6J and CBA/J wild-type mice aged 8–10 weeks were obtained from The Jackson Laboratory (Bar Harbor). Germ-free Swiss-Webster mice were bred in house.
C57BL/6 Mice were treated with an oral dose of 20 mg streptomycin 24 hours before oral infection with 0.1 ml LB broth (mock-infected) or with 1 x 109 CFU of a 1:1 mixture of the indicated S. Typhimurium strains. Mice were euthanized 4 days after infection, cecal and colon contents were collected for enumeration of bacterial numbers and the cecal tip was collected for histopathology scoring. Bacterial numbers were determined by plating serial ten-fold dilutions onto LB agar containing the appropriate antibiotics.
For intraperitoneal infection, C57BL/6 mice were injected with 105 CFU of a 1:1 mixture of the indicated S. Typhimurium strains. Mice were euthanized 2 days after infection and spleen, liver and mesenteric lymph nodes were collected for enumeration of bacterial numbers. Tissues were homogenized and bacterial numbers were determined by plating serial ten-fold dilutions onto LB agar containing the appropriate antibiotics.
CBA mice were infected with either 0.1 ml of LB broth (mock-infected) or S. Typhimurium in LB broth. For single infections, mice were inoculated with 1 x 109 CFU of the indicated S. Typhimurium strains. For competitive infections, mice were inoculated with 1 x 109 CFU of a 1:1 mixture of the indicated strains. Fecal pellets were collected at the indicated time points to monitor colonization over time. Mice were euthanized at 14 days after infection, cecal and colon contents were collected for enumeration of bacterial numbers and the cecal tip was collected for histopathology scoring. Bacterial numbers were determined by plating serial ten-fold dilutions onto LB agar containing the appropriate antibiotics.
Germ-free Swiss Webster mice were obtained from Taconic Farms. Mice were bred and housed under germ-free conditions inside gnotobiotic isolators (Park Bioservices, LLC). Weekly cultures were performed to monitor the germ-free status of the mice. For experiments, male and female 6–8 weeks old mice were transferred to a biosafety cabinet and maintained in sterile cages for the duration of the experiment. Each recipient germ-free mouse was orally inoculated with TSB (mock-infected) or the indicated Bacteroides strain resuspended in TSB. Mice were mono-colonized for 7 days with either B. thetaiotaomicron or B. fragilis before oral challenge with an equal mixture of 109 CFU of the indicated S. Typhimurium strains. Mice were euthanized 3 days after challenge with S. Typhimurium, cecal and colon contents were collected for enumeration of bacterial numbers and the cecal tip was collected for histopathology scoring. Bacterial numbers were determined by plating serial ten-fold dilutions onto LB agar containing the appropriate antibiotics.
When mice were infected with a 1:1 mixture of bacterial strains, the ratio of recovered bacterial strains was normalized to the ratio present in the inoculum to calculate the competitive index.
Cecal tissue was fixed in 10% phosphate- buffered formalin and 5 μm sections of the tissue were stained with hematoxylin and eosin. Blinded scoring of tissue sections was performed by a veterinary pathologist based on the criteria listed in S4 Fig. Representative images were taken using an Olympus BX41 microscope.
Germ-free mice were orally inoculated with TSB (mock-infected) or B. thetaiotaomicron and colonized for 7 days before oral challenge with LB (mock-infected) or 108 CFU of S. Typhimurium wild-type or the pduA-X mutant. Mice were euthanized 3 days after challenge with S. Typhimurium and colon contents were collected in 250 μL of sterile PBS spiked with 0.1 mM (final concentration) of deuterated 1,2-propanediol (internal standard). The sample weights were measured for later normalizations to determine the correct concentration. Samples were vortexed for 2 minutes at maximum speed to break up the fecal pellet and create a homogenous solution. Samples were then centrifuged at 6,000 x g, 4°C for 15 minutes, the supernatant was transferred into a new tube and stored at -80°C for further processing. Samples, diluted with acetonitrile at a 1:1 ratio, were incubated for 30 min at room temperature. To remove particles, samples were centrifuged for 15 min at 20,000 g at 4°C. The supernatant was transferred to an autosampler vial for gas chromatography-mass spectrometry analysis. A Gas Chromatograph Mass Spectrometer (Shimadzu TQ-8040 GC/MS/MS) was used with an injection temperature of 250°C, injection split ratio of 5 and an injection volume of 1 μl. The GC oven temperature started at 100°C for 1 min, rising to 250°C at 10°C/min with a final hold at this temperature for 4 min. GC flow rate with helium carrier gas was constant at 35 cm/s. The GC column used was a 30 m × 0.25 mm × 0.25 μm Stabilwax-MS (Restek). The interface temperature was 250°C and ion source temperature was 200°C. The mass spectrometer was set to selected-ion monitoring of the three most abundant m/z, m/z 1 at 29, m/z 2 at 43, m/z 3 at 45 for 1,2-propanediol and m/z 1 at 30, m/z 2 at 33 and m/z 3 at 49 for the internal standard 1,2-propanediol-d6. m/z at 45 and m/z at 49 were used to quantify 1-2-propanediol and 1,2-propanediol-d6 respectively, while the other m/z were used as qualifying ions [39]. Recovery was calculated based on the internal standard. An external standard curve was run in triplicate for quantification.
Student’s t-test was performed on logarithmically transformed values for bacterial numbers and competitive indices. A non-parametric test (Man Whitney) was used for comparing histopathology scores.
|
10.1371/journal.pcbi.1004705 | Systems-Wide Prediction of Enzyme Promiscuity Reveals a New Underground Alternative Route for Pyridoxal 5’-Phosphate Production in E. coli | Recent insights suggest that non-specific and/or promiscuous enzymes are common and active across life. Understanding the role of such enzymes is an important open question in biology. Here we develop a genome-wide method, PROPER, that uses a permissive PSI-BLAST approach to predict promiscuous activities of metabolic genes. Enzyme promiscuity is typically studied experimentally using multicopy suppression, in which over-expression of a promiscuous ‘replacer’ gene rescues lethality caused by inactivation of a ‘target’ gene. We use PROPER to predict multicopy suppression in Escherichia coli, achieving highly significant overlap with published cases (hypergeometric p = 4.4e-13). We then validate three novel predicted target-replacer gene pairs in new multicopy suppression experiments. We next go beyond PROPER and develop a network-based approach, GEM-PROPER, that integrates PROPER with genome-scale metabolic modeling to predict promiscuous replacements via alternative metabolic pathways. GEM-PROPER predicts a new indirect replacer (thiG) for an essential enzyme (pdxB) in production of pyridoxal 5’-phosphate (the active form of Vitamin B6), which we validate experimentally via multicopy suppression. We perform a structural analysis of thiG to determine its potential promiscuous active site, which we validate experimentally by inactivating the pertaining residues and showing a loss of replacer activity. Thus, this study is a successful example where a computational investigation leads to a network-based identification of an indirect promiscuous replacement of a key metabolic enzyme, which would have been extremely difficult to identify directly.
| Many enzymes can perform secondary functions at low affinities or rates, but such ‘promiscuous’ functions have never been predicted on a genome-wide scale. Here, we present the first genome-wide method to predict promiscuous functions of metabolic genes, which we apply to E. coli. Notably, we predict and validate several new cases where a ‘replacer’ gene can compensate for the loss of an essential ‘target’ gene through a promiscuous activity. Next, we couple our method with a genome-scale metabolic model, in order to search for ‘replacer’ genes that compensate for essential ‘target’ genes by metabolically bypassing them. We use this network-augmented approach to uncover a novel promiscuous pathway for the production of pyridoxal 5’-phosphate (the active form of Vitamin B6) in E. coli. This study represents an important step in understanding promiscuous functions in bacteria, and is a prime example of a systems-level analysis leading to new biological insight.
| Enzymes have traditionally been associated with discrete activities [1]. Clear-cut annotations populate well-known enzyme databases such as KEGG and Uniprot, and foster an implicit assumption that gene activities are specific. However, recent advancements in our understanding of evolution and enzyme activity have cast this view into question, and suggest instead that non-optimized and/or promiscuously active enzymes are frequent and active across life [2]. ‘Generalist’ enzymes (i.e., those carrying more than one function) have been shown to be abundant, to play different biological roles than ‘specialist’ enzymes, and to behave differently than specialist enzymes during switches in media conditions [3]. Promiscuous enzyme activities (i.e., those that are only active with low affinities/activities) have been shown to play adaptive biological roles (e.g., [4] and [5]), and to often arise through neutral mutations that are not detrimental to the primary enzymatic activity [6]. In particular, antibiotic resistance may emerge initially through the action of over-expressed, promiscuous genes [7, 8].
Because of recent conceptual and modeling advances, there is significant interest in being able to predict and exploit enzyme promiscuity [9–12]. Tools have been developed that mine molecular signatures or three-dimensional catalytic domains of enzymes to predict whether enzymes are promiscuous [13–15], and these methods have been used, e.g., to design likely retrosynthetic pathways [9, 11]. While these methods achieve important goals, they have not been aimed at evaluating the functional implications of existing promiscuous activities on a genome-wide, network scale. A notable recent effort did probe the global relevance of promiscuous enzyme activities, but the study used ‘as is’ data collected from enzyme databases, which, while having the advantage of being fully experimentally verified, include many non-biologically-relevant instances and exclude many biologically relevant ones [16]. Another recent study used knowledge gaps identified in a genome-scale metabolic model of E. coli to identify candidate isozymes, which display substrate promiscuity [17]. In that study, a ‘target’ enzyme that is found to be nonessential in vitro despite being essential in silico is knocked out, after which potential isozymes that are found to be upregulated are cumulatively knocked out until the cell can no longer survive. This method thus identifies enzymes with a secondary isozyme function that, when the cell overexpresses them, can compensate for the loss of the ‘target’. Interestingly, in one case they consider, the target is in fact essential, and adaptive evolution is required for the potential isozyme to be expressed enough to compensate for its loss. Searching for many more such low activity promiscuous functions in an unsupervised, genome-wide way is the focus of this present study.
To accomplish this, we utilize an unsupervised PSI-BLAST based method for assessing potential secondary functions of genes in E. coli. We predict promiscuous ‘replacer’ functions that may compensate for primary ‘target’ functions in other genes if they are altered or lost, and compare these predictions to known cases in which an over-expressed gene can take on a secondary role, as shown in an assay generally referred to as multicopy suppression [18]. Multicopy suppression reveals ‘replacer’ genes whose over-expression suppresses the lethality caused by knocking out conditionally essential ‘target’ genes. It is thus an elegant assay for uncovering promiscuous gene functions. We find that our method for predicting promiscuous functions successfully recapitulates known multicopy suppression events [8, 19] and predicts new ones, several of which we validate in vitro.
We next develop a genome-scale metabolic modeling (GEM) based approach to predict and experimentally test cases of multicopy suppression in which functional promiscuous replacement happens ‘indirectly’ through a bypassing pathway or reaction, rather than by directly replacing the function of the target gene. For this, we focus on the gene target pdxB, for which multiple promiscuous replacers have been reported in the past [8]. PdxB is a key enzyme for producing pyridoxal 5’-phosphate (the active form of Vitamin B6), an essential nutrient in E. coli. Our experiment proceeded in 4 steps: First, we predicted that the gene thiG harbors a promiscuous activity that metabolically bypasses pdxB. Second, we validated this prediction with a new in vitro individual multicopy suppression experiment. Third, we performed a detailed structural analysis of ThiG and hypothesized which residues perform the promiscuous function. And Fourth, we mutated this active site and confirmed loss of pdxB replacement activity. Thus, we begin at the level of a genome-wide screen for promiscuous functions, and proceed to identify and validate a promiscuous pathway for production of an essential nutrient in E. coli, which could have important functional implications for this model organism.
We aimed to predict promiscuous functions of genes in a systematic and unsupervised manner, and specifically to do it on a genome-wide scale. We started by doing permissive PSI-BLAST based searches for distant gene similarities across RAST, a large consistent database of metabolic reactions and compounds that includes thousands of microbes [20]. This search allowed us to assign putative secondary functions for each initial “root” gene in the search (see Methods and Fig 1). We did this across metabolic genes in E. coli, which resulted in a set of phylogenetic trees, one for each root gene, that include up to thousands of genes from other bacteria along with their associated functions.
Using these trees, we searched for instances in which a gene in one of the trees had an assigned function different from the one that the RAST database assigns to the root gene. We restricted our search to genes whose primary functions are metabolic, as this is our focus of interest. Since our tree building method was more permissive than typical BLAST or PSI-BLAST, we considered these predicted functions as potential secondary functionalities that might only be effective at high expression levels or with slight mutations. We entered these cases in a large gene-by-function matrix.
To facilitate testing and validation, we used our promiscuity matrix to predict pairs of genes that correspond to those identified through an in vitro assay called Multicopy suppression [19]. Multicopy suppression tests ‘replacer’ genes for promiscuous activities that either directly (catalyze similar reactions) or indirectly (catalyze distinct reactions) replace the function of a ‘target’ gene and preserve growth. It proceeds in four steps: (1) a ‘target’ gene is identified in the bacterium of interest (target genes must be essential on a certain test medium [usually M9], but not on a rich medium); (2) a strain is created in which the target gene is inactivated (due to the condition in step 1, this strain is not viable on the test medium); (3) a library of native genes (hereafter called ‘replacers’) is individually over-expressed in the organism on high-copy plasmids; and (4) transformed strains are grown on the test medium, and survivors are noted as having identified successful target-replacer gene pairs. Importantly, the replacer genes are native to the genome of the organism, but their natural expression dosage is not sufficient to promiscuously support growth. Hence, replacer genes discovered through multicopy suppression are hypothesized to possess secondary, low-affinity/low-activity functions that compensate for the function of the target. We predict target-replacer pairs in two ways: direct replacements, and indirect replacements.
To predict direct replacements, we search our matrix for instances in which one ‘replacer’ gene in E. coli has a secondary function assigned through our method, which corresponds exactly to the function of some other ‘target’ gene that is also found in E. coli. We call this straightforward method the enzyme PROmiscuity PrEdictoR, or PROPER.
To predict indirect replacements, we use the added insight of a GEM of E. coli (obtained from SEED: [21]). The SEED E. coli GEM was used, as opposed to one of the highly curated E. coli GEMS such as [22], because of our requirement that metabolic reactions from the SEED database can be seamlessly added to the GEM to test their effect on cellular fluxes (which would first require a large reconciliation of nomenclature if using a non-SEED GEM). Specifically, we search for replacers that, through a promiscuous function identified in our matrix, metabolically bypass the need for the target. These predictions were done in three steps: (1) we knocked out the target gene in a GEM (this could only be done for genes that were conditionally lethal on M9 in silico); (2) we searched in our gene similarity trees for any E. coli genes that harbor non-E. coli promiscuous functions; (3) we added each of these promiscuous functions in turn to the GEM, and looked for any that rescued in silico growth. This GEM-based PROmiscuity PrEdictoR (GEM-PROPER) thus identified indirect replacer genes that act through metabolism (see panel 3 of Fig 1).
In all, PROPER predicted 2811 direct target-replacer pairs in E. coli, encompassing 794 metabolic target genes and 753 metabolic replacer genes (see S1 Table). GEM-PROPER predicted a total of 98 indirect target-replacer pairs in E. coli (see full list in S4 Table). Since direct replacers span a large portion of metabolism, we focus on these predictions for our large-scale analyses (i.e., the first several sections of the results). Afterwards, we shift our focus to GEM-PROPER, and present a detailed analysis of a novel indirect predictor that we validated.
The number of direct replacers per target follows an exponentially decaying distribution (see S1 Fig), implying that promiscuous gene functions cluster around key target functions. The target genes with the most putative replacers are involved in fatty acid degradation or synthesis [fadD and fadK (degradation); fabD (synthesis)], or transport processes [malF (maltose/maltodextrin transport); potC, ydcV, potB, and ydcU (spermidine/putrescine transport); and cysW (sulfate transport)]. Fatty acid degradation and synthesis processes are known to involve enzymes with many potential specificities depending on only small alterations [23].
To validate our direct predictions, we collected results from two studies of multicopy suppression in E. coli, which between them had discovered 48 instances of multicopy suppression among 21 target genes [8, 19]. Both of these assays looked for multicopy suppressors of targets that were essential on M9 medium but not in a rich medium, and one used a combinatorial assay that assessed every gene in E. coli as a potential replacer [19]. All but 3 of the target genes for which replacers had been found were metabolic, supporting our focus on metabolism. In all, we predicted a total of 63 replacers for the 21 targets, versus the 48 found previously by multicopy suppression. The replacers we predicted overlapped with those found previously by 8 genes. Notably, for each of the 21 target genes, there were over a thousand metabolic genes that could serve as potential replacers; therefore, the chance of achieving this level of overlap between our predictions and those found experimentally through random guesses is extremely low (p = 4.5e-15 in a hypergeometric test).
As another control test, we performed a BLAST search of each target gene identified in Patrick et. al. against all metabolic genes in E. coli, and kept only those above the default threshold as potential replacers. This yielded 122 predicted replacers for the 21 target genes, among which none matched results from Patrick et. al., and only two overlapped with predictions made by PROPER (neither of those turned out true in our own experiments). This emphasizes the added value of PROPER over simple BLAST searches for finding find such low affinity promiscuous functions.
We also considered it possible that some of our target-replacer predictions are true, but cannot be verified using multicopy suppression because the target gene is not conditionally lethal (which is a necessary condition for doing a multicopy suppression experiment–see Fig 1, panel 4). As a quick test of this, we compared our direct predictions to the isozyme sets aspC/tyrB, argD/astC/gabT/puuE, and gltA/prpC, which were discovered in [17] as described in the introduction. Indeed, all 8 pairwise combinations of genes from within any of these 3 isozyme sets comes up as a reciprocal target-replacer pair in our direct predictions, meaning that either gene in each pair can play the role of the target or the replacer. That all of these pairs would come up in our predictions and, given that, that they would all be reciprocal are both highly unlikely by chance (p = 3e-19 and p = 3e-5 in hypergeometric tests, denoting that all pairs from [17] (a) overlap our predictions vs. a background of all-vs.-all pairings of our predicted targets and replacers, and (b) are reciprocal, assuming the same likelihood for any pair). This suggests that promiscuous functions that have enough activity to be considered isozymes (or, more specifically, to be defined so through the experimental pipeline in [17]) may tend to come up in our predictions as reciprocal, and suggests a complimentary association between our methods and those in the aforementioned work.
The random gene insertion method used by Patrick et al. to screen for potential target-replacer pairs [19] is demonstrably not comprehensive (e.g., 7 replacers were found for pdxB in [8]; none of these were reported in the Patrick study for pdxB, although pdxB was in the set of replacers that Patrick et al. tested). Hence, we next asked how many of our novel predicted target-replacer pairs (those that had not been previously found experimentally) may be true. To test this we developed a single-replacer multicopy suppression assay analogous to that used to validate individual results in the initial multicopy suppression study by Patrick [19]. This assay involves plating target-replacer strains (i.e., strains with the target knocked out, and the replacer over-expressed on a plasmid) on M9 medium, and observing how long it takes for colonies to appear. We judged successful target-replacer pairs as those that grew more robustly than an empty plasmid control (see Methods, as well as Supplementary methods in S1 Text for a fuller explanation of this assay).
Using these criteria, we verified 7 of the 8 target-replacer pairs that were identified in Patrick et. al. and predicted by PROPER (see S2 Table for full list of experimental results). We then assessed each of our 55 novel predicted target-replacer pairs for multicopy suppression, and found three of them to have replacer activity (hisA, cysM, and metB were viable replacers for hisH, ilvA, and metC respectively; see S2 Table). Thus, in all we validated 10 out of 63 direct replacement predictions, which gives a success rate of 20% (after eliminating 14 predictions for ptsI and glyA target genes, which were in practice untestable because the target knockout did not decrease growth from wt in our experiments; see S1 Text). Among the three validated novel target-replacer pairs we validated, the strength of the observed phenotype scaled with the strength of sequence similarity between the replacer gene and its match in the promiscuous gene tree (i.e., Panels 1–2 of Fig 1). Namely, metB fully restored growth in ΔmetC cells, while the less homologous cysM and hisA (in that order) less efficiently supported growth following target knockout (see S1 Text; also compare alignments in S4, S5 and S6 Figs). Thus, these examples suggest that higher similarity to the function-assigning gene corresponds to higher activity or affinity for that enzyme’s substrates, although a more extensive study of multicopy suppression phenotypes would be needed to uphold this observation generally.
Among the 98 indirect target-replacer pairs we predicted, two were for the target pdxB, which is by far the most ubiquitiously replaced target found in vitro across published multicopy suppression datasets (in fact, pdxB is the sole focus of the Kim study: [8]). We confirmed our in silico predictions of these two target-replacer pairs using the manually curated E. coli model iAF1260 [22]. The target, pdxB, is a conditionally essential gene (essential on M9 medium, but not on rich medium) involved in the biosynthesis of pyridoxal 5’-phosphate (P5P), the active form of Vitamin B6. Vitamin biosynthesis reactions are good candidates for multicopy suppression, since vitamins are required in low amounts, so even a moderate flux through an over-expressed promiscuous enzyme could provide enough of the nutrient to enable growth [24]. P5P is an essential cofactor in all known living systems [25].
We tested our two predicted pdxB target-replacer pairs, and found one of them, thiG replacing pdxB, to be a true replacer [∆pdxB/thiG colonies were 1mm diameter, vs. 0.1–0.2mm diameter in ∆pdxB/empty, after 3 days incubation in replicate experiments on M9 medium; see S2 Table]. The observed phenotype was consistent with previously reported pdxB replacers (1mm colonies of a ∆pdxB/replacer strain after 1–2 days (3 cases) or 3–5 days (4 cases) at 37°C in equivalent growth conditions & temperature [8]). We explored the thiG-pdxB target-replacer pair in detail, as follows:
Identifying promiscuous gene functions is a fundamental task in biology. Promiscuous functions have been causally associated with the evolution of new gene functions [24, 31], and may have contributed to the evolution of resistance to antimicrobials and other stresses [7]. Here, we present a new method termed PROPER that uses iterative PSI-BLAST-based phylogenetic trees for predicting potential promiscuous functions based on weaker similarities than are typically considered in assigning gene functions. Based on these predictions, we experimentally validate 4 novel target-replacer pairs, one of which (thiG replacing pdxB), was found by coupling our promiscuity predictor with a GEM. Finally, we predict and then experimentally validate the promiscuous active site of thiG for this replacing activity, revealing a striking new promiscuous route for the production of an essential nutrient, p5p, in E. coli.
While sequence similarity-based methods (e.g., BLAST) are used ubiquitously to determine primary gene functions, their ability to call promiscuous functions is unclear. Our method gives a first estimate, as we were able to experimentally validate around 20% of direct target-replacer pairs we could test in the lab (This is 100x higher than expected if guessing target-replacer pairs randomly). It is likely that the number of target-replacer pairs we validated is an underestimate of true promiscuity, as some true replacers might not show during multicopy suppression due to incorrect expression levels (e.g., due to non-optimal IPTG concentrations), insufficient effects of target knockout, or other confounding factors. Notably, many of our predictions might be correct even though they cannot be directly validated through multicopy suppression (due to the necessity in multicopy suppression that the target gene is conditionally essential). Our comparison to the work of [17] emphasizes this point, as several of our predicted target-replacer pairings came up in that work, regardless of the fact that the target genes examined in that study are not conditionally essential. The promiscuous functions we predict thus might represent a large underground repertoire that could be activated under the right kinds of selective pressure.
Our method for predicting promiscuous gene functions utilizes the RAST database and automatically constructed metabolic models from SEED, resources that include thousands of bacterial and archaeal species. While manually curated GEMS for E. coli are available (e.g., [32]), we chose to use the SEED GEM because it interfaces smoothly with the SEED and RAST databases, which was necessary for implementing GEM-PROPER without needing to reconcile thousands of metabolite and reaction names with an outside model. A second benefit of this choice is that although we focus here on predicting promiscuous gene functions in E. coli, our methods are generic and may easily be extended to any other organism in RAST. This could, for example, aid in determining resistance or adaptation mechanisms across major pathogenic strains, which could then be targeted in new cures.
In all, this work constitutes the first ever genome-wide prediction of metabolic gene multicopy suppression, and the first integration of such predictions with a GEM to achieve network-reliant indirect predictions. We begin with an unsupervised, large-scale method to predict enzyme promiscuity, and finally dial in on and then experimentally verifying a key prediction of biological significance in E. coli. Without such a systems approach, it would have been extremely difficult to identify thiG as a candidate replacer for pdxB. Thus, this study serves as a successful example where systems biology can provide a roadmap for biological investigation, identifying the most promising targets to then follow up on with more costly and time-consuming experiments.
PROPER proceeds in three steps, as outlined in Fig 1: (1) building gene similarity trees for all genes in E. coli; (2) using these trees to build a matrix of promiscuous gene functions; and (3) applying this matrix to determine potential promiscuous gene replacers for target genes of interest. In GEM-PROPER, step (3) is replaced by a GEM-based approach, in which we search for promiscuous functions (from step 2) that can rescue in silico growth of a GEM model after knockout of the target gene. Here, we describe these steps in detail:
Multicopy suppression is an established assay that has been described elsewhere (see panel 4 in Fig 1, and, e.g., [8, 19]). In our experiments, we tested specific target-replacer pairs in the manner used by Kim to validate target-replacer pairs they identified in their large-scale screen. Namely, we inserted a plasmid containing the replacer gene with an IPTG-induced promoter into a KEIO strain with the target gene knocked out, and assessed whether this target-replacer strain grew better than the target knockout strain without the replacer. Specifics of strain construction and the experiments follow:
E. coli strains containing specific gene knockouts were obtained from the KEIO collection [36], which contains single-gene knockout strains for the majority of genes in E. coli. Plasmids for IPTG-inducible expression of predicted replacer genes were obtained from the ASKA collection [37], which contains plasmids that overexpress nearly every individual gene in E. coli. Both collections are obtainable from the National Bioresource Project at the National Institute of Genetics, Japan. KO target strains were made electro-competent (4x washes with ddw/10% glycerol) and transformed with ASKA plasmids over-expressing the corresponding predicted replacer gene.
When we were developing our assay, we found that knockout colonies usually grew after some amount of time on M9 medium regardless of whether the replacer gene had been added to the plasmid, despite the fact that each of the target strains we tested was previously reported to be non-viable on M9 medium. The previous study by Patrick [19] had not plated background target-deleted strains that had empty plasmids inserted. Due to the eventual growth in most strains, this was an important negative control for judging that a replacer was actually compensating for the loss of the target and improving growth (and was done in Kim: [8]). Therefore, for each KO strain we additionally transformed with an empty ASKA plasmid as a negative control. We judged a strain to be a true replacer only if it came up consistently stronger and/or earlier than the strain carrying the empty plasmid.
In the KEIO collection each knocked out gene is replaced by the kanamycin resistance gene (KanR). Thus we verified correct location of the knockout in each target strain by amplifying and sequencing the area directly flanking the kanamycin insertion, using a kanamycin universal primer and a strain-specific reverse primer located downstream of the knocked out gene. Since the orientation of the kanamycin insertion depends on the orientation of the original knocked out gene, the KanR primer 5’-ATATTGCTGAAGAGCTTGG was used when the original gene had been forward coded and the KanR primer 5’-AATGAACTCCAGGACGAG was used if the original gene had been reverse coded. Correct sequence of the replacer gene was verified by amplifying and sequencing the ASKA plasmid insert (forward: 5′-ATC ACC ATC ACC ATA CGG AT; reverse: 5′-CTG AGG TCA TTA CTG GAT CTA).
Starter cultures of each target-replacer (TR) pair were grown overnight in LB+CAP, washed and serially diluted in saline (0.9% NaCl). Aliquots (100 ul) of 10−6 dilution were plated on LB+chloramphenicol and on M9-glucose-kanamycin-chloramphenicol (1× M9 salts, 2 mM MgSO4, 0.1mMCaCl, 0.4% glucose, 34 μg/mL chloramphenicol, 30μg/ml kanamycin) containing IPTG. All TR pairs were plated on both M9 with 50μM and 125μM IPTG; TR pairs of particular interest (hisA/∆hisH, cysM/∆ilvA and thiG/∆pdxB) were plated on 250μM as well. In each TR plating experiment, the empty and frvX negative controls were plated alongside the TR pairs. Plates were wrapped in nylon to avoid dehydration, incubated at 37 and monitored during the next three weeks. Nearly all negative control plates showed some growth phenotype, ranging from very strong (normal sized colonies within 3 days) to very weak (pinpoint colonies after 3 weeks; S3 Table). Thus, a TR pair was considered valid only if colonies consistently formed at a higher rate than in both negative control plates. Each plating experiment was repeated 2–3 times.
Growth of TR pairs that showed exceptionally high growth on the negative control plates (ΔglyA, ΔptsI and ΔpabB) was also monitored in liquid culture. Cultures in LB+CAP were washed once in saline and re-suspended at a dilution of 1:100 into M9-glucose-kanamycin-chloramphenicol minimal media supplemented with IPTG (50 μM). Growth measurements were performed in a 96-well plate incubated for 19–24 h at 37°C in a temperature-controlled plate reader with continuous shaking (ELX808IU-PC; Biotek), and OD595 was monitored every 15 min. Each TR pair was loaded into 2 duplicate wells. Growth of the negative controls (empty and frvX plasmids) for each target strain was likewise monitored during every run.
It was reported in [8] that knockout strains of pdxB sometimes display a heterogeneous phenotype, with some growing on minimal medium and others not. To be sure that this was not a factor in our experiments verifying the thiG/∆pdxB replacer-target pair, we grew individual colonies of ∆pdxB and confirmed that there was no heterogeneity in their growth on M9 medium.
Overnight cultures of each strain were grown in LB with 30 μg/mL kanamycin (ΔpdxB cells) and 25 μg/mL chloramphenicol (ASKA plasmids) selection. Cells were washed in 1x PBS and diluted 1:100 into M9-glucose-kanamycin-chloramphenicol for plate seeding; M9 was prepared without a nitrogen source. A checkerboard matrix was generated in 2mL deep-well, 96-well assay plates by serial dilution of NH4Cl (22.5% w/v maximum concentration) across plate columns and IPTG (500 μM maximum concentration) across plate rows. Wells were uniformly inoculated with cells, and each well contained a final volume of 1.2 mL. Plates were sealed with gas-permeable membranes and grown in a light-protected, microplate incubator shaker at 37°C and 700 RPM; 700 RPM was determined to be equivalent to 300 RPM in a standard incubator shaker. Samples for OD600 measurements were taken at designated timepoints using a SpectraMax M5 microplate multimode plate reader (Molecular Devices), and the gas permeable membrane resealed after each timepoint. 100 μL samples were taken for OD600 measurements to minimize the total volume loss.
For batch culture experiments, cells were prepared from overnight cultures as described above. Cells were diluted 1:100 in 30 mL of M9-glucose-kanamycin-chloramphenicol, containing ~3% NH4Cl (w/v) and 10 μM IPTG, in 250 Erlenmeyer flasks. Cultures were grown at 37°C and 300 RPM in an incubator shaker, with samples for OD600 measurements taken at designated timepoints.
The X-ray structure of pdxS (from B. bacillus) in complex with pdxT, was downloaded from the pdb database, and the residues of the pdxS active site were identified from the publication (active residues are: K149, D102, K81 and D24) [38]. Multiple pdxS units from the multimeric structure were overlaid and were found to be coincident (as can be seen in Figs 2B and S8). No structure of thiG was available from E. coli, so a homology-model was built using the SWISSMODEL pipeline (http://swissmodel.expasy.org/; [39]) based on the thiG template from Thermus thermophilus (51.98% seq identity; PDB ID 2htm [40]). The structures of pdxS were subsequently aligned with that of thiG using pyMol [41].
Low (50μM) IPTG concentration was sufficient to induce all of the replacers. Interestingly, increasing the IPTG concentration affected growth sometimes positively and sometimes negatively, depending on the specific strain. Increasing IPTG concentration up to 250μM caused increases in colony size and number, and decreases in incubation times, for hisA/∆hisH and cysM/∆ilvA. Conversely, growth of purE/∆purK, one of the target-replacer pairs predicted by both Patrick and us, was almost entirely inhibited when IPTG concentration was increased to 125uM. These results illustrate that over-expression can also have deleterious effects [42]. The optimal level of expression (and hence the optimal IPTG concentration) depended on the specific target-replacer combination. In agreement with this, in several target strains, over-expression of the randomly chosen gene frvX caused a decrease of the background seen in the empty plasmid control (see S3 Table).
|
10.1371/journal.ppat.1007672 | Increased mucosal neutrophil survival is associated with altered microbiota in HIV infection | Gastrointestinal (GI) mucosal dysfunction predicts and likely contributes to non-infectious comorbidities and mortality in HIV infection and persists despite antiretroviral therapy. However, the mechanisms underlying this dysfunction remain incompletely understood. Neutrophils are important for containment of pathogens but can also contribute to tissue damage due to their release of reactive oxygen species and other potentially harmful effector molecules. Here we used a flow cytometry approach to investigate increased neutrophil lifespan as a mechanism for GI neutrophil accumulation in chronic, treated HIV infection and a potential role for gastrointestinal dysbiosis. We report that increased neutrophil survival contributes to neutrophil accumulation in colorectal biopsy tissue, thus implicating neutrophil lifespan as a new therapeutic target for mucosal inflammation in HIV infection. Additionally, we characterized the intestinal microbiome of colorectal biopsies using 16S rRNA sequencing. We found that a reduced Lactobacillus: Prevotella ratio associated with neutrophil survival, suggesting that intestinal bacteria may contribute to GI neutrophil accumulation in treated HIV infection. Finally, we provide evidence that Lactobacillus species uniquely decrease neutrophil survival and neutrophil frequency in vitro, which could have important therapeutic implications for reducing neutrophil-driven inflammation in HIV and other chronic inflammatory conditions.
| HIV infection results in chronic immune activation that leads to increased risk of other diseases and premature death, and this has been linked to gastrointestinal tract (GI) damage in infected individuals. In this study, we investigated neutrophils, a cell involved in the immune response to pathogens, in colorectal tissue of HIV-infected individuals receiving treatment. Because neutrophils use methods to contain pathogens that can also damage tissue and have been implicated in tissue damage in other GI diseases, it has been proposed that they contribute to GI damage in HIV infection. However, the role of neutrophils in GI damage in HIV has not been well studied. This study quantifies neutrophils in relation to other white blood cells in the GI tissues in HIV infection and demonstrates that they are increased in the GI in infected individuals. Additionally, we present evidence that neutrophils in the GI in HIV-infected individuals have a longer lifespan, which represents one potential reason for their increased frequency. Finally, we present data that different bacteria that naturally reside in the GI can alter neutrophil lifespan and that changes in the relative abundances of these bacteria in HIV infection may be contributing to increased neutrophil lifespan.
| Gastrointestinal (GI) mucosal damage and immune dysfunction drive chronic inflammation and microbial translocation in HIV infection, which predict and likely contribute to non-infectious comorbidities and mortality[1–5]. Although long-term antiretroviral therapy (ART) partially restores mucosal damage, a degree of mucosal immune dysfunction and inflammation persists and is associated with morbidities and mortality[6–8]. Improving the understanding of this persistent mucosal dysfunction and inflammation during ART is a major hurdle for the development of targeted therapies that may promote health and decrease morbidities and mortality in HIV-infected individuals.
Neutrophils, the most abundant immune cell, are the first responders to most infections and are crucial in the immune response to bacterial and fungal pathogens[9]. However, the role of neutrophils in HIV infection is not well understood. Imaging studies assessing myeloperoxidase (MPO), an enzyme produced and secreted by neutrophils, suggest that neutrophils accumulate in the GI tract in treated and untreated HIV infection, yet the frequency and functionality of accumulated neutrophils has yet to be examined[10]. A better understanding of the mechanisms involved in neutrophil accumulation in the GI in HIV infection is necessary to develop new strategies to alleviate GI inflammation in HIV infection.
While neutrophils are critical in protection from infections, aberrant neutrophil responses can also be harmful. Neutrophil infiltration in the colonic mucosa is one of the distinguishing characteristics of acute inflammation present in inflammatory bowel disease and correlates with disease severity[11–13]. Neutrophil lifespan is tightly regulated in order to limit unintended damage to tissues by secreted reactive oxygen species and granular enzymes meant to degrade extracellular matrix and disrupt tight junctions[14]. Increased neutrophil lifespan is observed under inflammatory conditions and has been attributed to both direct interaction with microbes and the release of cytokines from other immune cells[15] [16,17].
In HIV-infected individuals, studies have reported that the delicate balance of healthy bacterial communities is perturbed, resulting in microbial dysbiosis [18–22]. Importantly, dysbiosis remained evident in individuals on ART and associated with disease progression [19]. However, recent studies demonstrated that these previous results were likely confounded by sexual orientation, for which the study designs did not adequately match or control [23,24]. Taking this into account, another recent study matched controls based on sexual orientation and found both men who have sex with men (MSM)-specific alterations and HIV-specific alterations in MSM and women[25]. Therefore, it is evident that some combination of infection and sexual orientation alters the microbiome in infected individuals. However, to what degree infection itself and lifestyle parameters contribute to microbial alterations in infected individuals requires further study. Additionally, a recent study of experimental dysbiosis induced by antibiotics in rhesus macaques did not lead to increased disease progression in untreated SIV-infected animals, suggesting that dysbiosis may be co-associated with disease progression rather than causative [26]. However, further studies are required in HIV-infected and uninfected human populations to determine the contributions of microbiome dysbiosis in the context of untreated and treated HIV infection.
The effects of microbiome alterations on intestinal neutrophils have not been previously assessed, and given that Toll-like receptor (TLR) activation and cytokine stimulation regulate neutrophil survival, changes in the microbiome composition could alter neutrophil lifespan. In this study, we hypothesized that HIV-infected individuals would have increased neutrophil frequencies in the lower GI with reduced neutrophil apoptosis. We further hypothesized that different bacterial species would differentially contribute to alterations in neutrophil apoptosis.
Previous studies reporting neutrophil infiltration in HIV infection and the nonhuman primate model of SIV have relied solely on MPO staining of neutrophils measured via microscopy[10,27]. Because MPO secretion is increased upon neutrophil activation, it is unclear if increased MPO in the tissues represents increased neutrophil frequency or increased neutrophil activation. In addition, in some cases of inflammation, tissue macrophages also produce MPO and stain positive for the enzyme[28]. For these reasons, we developed a multicolor flow cytometry-based approach that distinguishes neutrophils in the context of other leukocytes in order to obtain a more quantitative measure of neutrophil frequencies in the GI during HIV infection. Using this panel, we were able to identify neutrophils in blood (Supplementary Fig 1A in S1 Supporting Information) and fresh GI tissue (Supplementary Fig 1B in S1 Supporting Information) to assess the frequency of neutrophils as the percentage of total live CD45+ cells.
Colorectal biopsies from a total of 40 HIV-infected, ART-suppressed individuals and 35 HIV-uninfected individuals were collected for this study (Supplementary Fig 2 in S1 Supporting Information). Of these colorectal biopsy samples, neutrophil frequencies were assessed by flow cytometry in real-time in 23 HIV-infected, ART-suppressed and 25 HIV-uninfected participants. Table 1 describes relevant participant demographic information including age, sex, sexual orientation, race/ethnicity, CD4+ T cell count, and time since HIV diagnosis for this subset of individuals. We found increased neutrophil frequencies in the GI of HIV+ individuals compared to uninfected controls (Fig 1A). Importantly, this increase remained when assessed using a multivariate analysis adjusting for age, race, sex, and sexual orientation (Fig 1B). This increase in neutrophils is specific to the GI tract, as we observed no increase in neutrophil frequency in the blood of HIV-infected, ART-suppressed individuals (Fig 1C). These data are the first to demonstrate neutrophils are increased in the GI tract relative to other leukocytes in HIV-infected individuals despite suppressive ART.
One potential mechanism for elevated frequencies of neutrophils in the colon in HIV is prolonged neutrophil survival. Neutrophil clearance is an important mechanism for tissue homeostasis, and neutrophils are generally short-lived, with findings from studies investigating neutrophil lifespan in vivo ranging from 8 hours to 5 days[29]. The least inflammatory mechanism of neutrophil clearance from tissues is caspase-3 mediated apoptosis followed by engulfment by macrophages[30–32]. Neutrophils undergoing apoptosis demonstrate reduced surface CD16 expression[33] and those with reduced CD16 expression also demonstrate reduced functionality[34]. Therefore, in order to evaluate neutrophil survival, we measured non-apoptotic, functional neutrophils as those expressing high levels of CD16 and low levels of active Caspase-3 in leukocytes isolated from colorectal biopsies (Fig 2A). Indeed, we found that the frequency of these surviving, functional neutrophils was increased in biopsies from HIV+ individuals compared to uninfected controls (Fig 2B), which was significant in a multivariate analysis adjusted for age, race, sex, and sexual orientation (Fig 2C). While neutrophil infiltration in inflammatory bowel disease has been attributed in part to delayed neutrophil apoptosis[35], these are the first data demonstrating this as a potential mechanism in ongoing intestinal inflammation in HIV infection. Importantly, we observed no differences in total neutrophils or neutrophil lifespan based on sex or sexual orientation in both an unadjusted analysis and after adjustment for HIV status (Supplementary Fig 3 in S1 Supporting Information).
Alterations to the intestinal microbiome in HIV-infected individuals are well-documented[36]. Given the extensive evidence that microbes and microbial ligands impact neutrophil survival through both direct interactions and by influencing cytokine release by other immune cells[15,16], we sought to determine whether alterations in microbial composition associated with differential neutrophil frequency, function and survival. To do this, we assessed the mucosal microbiome composition by 16S rRNA gene sequencing of colorectal biopsies collected from all 40 HIV-infected, ART-suppressed individuals and all 35 HIV-uninfected individuals. We focused on bacterial composition at both the family and genus taxonomic levels (Fig 3A and 3B) and observed a modest significant association between overall microbial composition at the genus level and HIV status when HIV-infected, ART-suppressed individuals were compared to uninfected controls (p=0.041 as determined by MiRKAT analysis). Importantly, this association remained when adjusted for age, race, sex, and sexual orientation, as can be visualized by principle component analyses of the adjusted relative abundances (p=0.035, Fig 3C). Few individual genera significantly associated with HIV status in either the unadjusted (Supplementary Fig 4A in S1 Supporting Information) or adjusted analyses (Supplementary Fig 4B in S1 Supporting Information) when the false discovery rate was taken into account (q-value<0.05), which may be due to low sample size.
Importantly, recent studies have demonstrated that the dysbiosis previously attributed to HIV infection may actually be a result of sexual risk behaviors, as men who have sex with men (MSM) had an increased abundance of Prevotella, independent of HIV status [23,25]. This highlights the importance of investigating confounding demographic factors in comparisons of HIV infected and uninfected populations. Therefore, we further investigated the microbial composition of the men in this cohort based on sexual orientation at both the family and genus taxonomic levels (Fig 4A and Fig 4B). We observed a significant association between microbial composition at the genus level and sexual orientation as categorized by MSM or non-MSM (p=0.002, Fig 4C as determined by MiRKAT analysis). This association remained when adjusted for age, race, and HIV status (p=0.020). Several alterations in bacterial taxa were associated with sexual orientation in both the adjusted and unadjusted analyses, including a loss of bacteria in the Bacteroides genus and the Barnesiellaceae family and an increase in bacteria of the Streptococcus genus (Supplementary Fig 5 in S1 Supporting Information). The Prevotella genus only significantly associated with sexual orientation prior to adjustment for HIV status.
In order to examine the potential impact of bacterial composition on neutrophils, we assessed associations between all taxa found in at least 25% of individuals and colorectal neutrophils. We found no associations between individual taxa and total neutrophils or neutrophil survival in an unadjusted analysis or following adjustment for age, race, sex, sexual orientation, and HIV status once the false discovery rate threshold was applied (S2 Supporting Data). It is possible that our sample size is too small to detect significant associations across all taxa when adjusting for multiple comparisons. Therefore, we next focused on individual bacteria that we hypothesized could impact neutrophil survival, particularly in HIV-infected populations.
Specific bacteria altered in certain HIV-infected populations have been shown to impact mucosal immune cells [18,37]. Of note, the Prevotella genus had the highest average relative abundance in our study population, and increased Prevotella in one HIV-infected population associated with increased mucosal T cell and dendritic cell activation [18]. In a follow-up study, the authors further reported the ability of Prevotella species to activate leukocytes in vitro by demonstrating that myeloid dendritic cells stimulated with Prevotella stercorea and Prevotella copri produced increased cyctokine[38]. Although the authors did not take into account sexual orientation in these studies, another recent study correlated Prevotella with T cell activation within an MSM population[25]. It remains unclear if the link between Prevotella and immune activation can be applied to other populations of HIV-infected individuals. It has been particularly difficult to assess these associations in non-MSM, HIV-infected men due to the insufficient number of available samples [25], which unfortunately were also unavailable for our study. However, given in vivo and in vitro evidence that Prevotella species are able to activate leukocytes, we hypothesized that Prevotella could impact neutrophil survival. In vitro assessment using a reporter cell line demonstrated that Prevotella copri activated NF-kb through a TLR-4 dependent mechanism [39] and NF-kb activation is known to drive survival factors in neutrophils[40]. Additionally, LPS has been shown to increase neutrophil survival both through direct TLR-4 activation on neutrophils and by inducing the release of TNF-α and IL-1β from monocytes[15]. Contrarily, reductions in Lactobacillus have been shown to impact gut health and immune function [21,41], and Lactobacillus species have been demonstrated to induce apoptosis in epithelial cells and myeloid cells[42]. Therefore, we hypothesized that alterations in these genera may impact neutrophil survival in the GI of HIV-infected, ART-suppressed individuals, and we sought to assess the relationships between these bacteria and neutrophils. While we observed no significant differences in the relative abundances of bacteria in the Prevotella or Lactobacillus genera (Fig 5A), we observed a significant difference in the ratio of Lactobacillus:Prevotella between HIV-infected, ART-suppressed and uninfected individuals, suggesting an altered balance of these genera (Fig 5B). Additionally, the Lactobacillus:Prevotella ratio remained significantly altered following adjustment for age, race, sex, and sexual orientation (Fig 5C). Importantly, the Lactobacillus:Prevotella ratio correlated with neutrophil survival in leukocytes isolated from colorectal biopsies from the same individual (Fig 5D), suggesting that an alteration in bacterial composition may impact neutrophil survival in vivo.
Given the recent interest in balance analyses of microbiome data, we performed a selbal analysis to determine the microbial signature predictive of increased neutrophil lifespan (Supplementary Fig 6 in S1 Supporting Information). The balance we identified as most closely associated with neutrophil lifespan included three genera in the numerator and ten genera in the denominator groups. Interestingly, Prevotella was among the bacteria in the numerator group, which includes bacteria that may be positively contributing to increased neutrophil lifespan. Additionally, Lactobacillus was among the bacteria in the denominator group, or those that may negatively affect neutrophil lifespan. These results suggest that these bacteria may have a role within global microbiome composition changes in altering neutrophil lifespan. However, these analyses should be interpreted with caution given that this method selects bacteria without applying statistical inference. Further study is needed to confirm these relationships in vivo with a larger sample size and in vitro as culturing methods and commercial strains become available.
Given our observation that HIV-infected individuals in our study had significantly altered bacterial community composition in vivo and previously published data suggesting that bacterial ligands can impact neutrophil lifespan, we examined the ability of bacterial ligands and various whole bacteria to impact neutrophil survival in an in vitro culture system with whole blood. These experiments were done using whole blood samples from both HIV-infected and uninfected individuals in order to assess the potential impact of confounding HIV infection on the ability of bacteria to alter neutrophils. In accordance with previous studies, we observed a significantly increased frequency of surviving neutrophils in whole blood after a 20-hour incubation with TLR-4 and TLR-2 agonists compared to unstimulated controls (Supplementary Fig 7A in S1 Supporting Information). Isolated neutrophils incubated with TLR-4 and TLR-2 agonists also demonstrated an increased frequency of surviving neutrophils after a 20-hour incubation that did not reach statistical significance, and to a much lesser extent than that observed in whole blood (Supplementary Fig 7B in S1 Supporting Information). These data suggest that both direct interactions with neutrophils and soluble factors released by other leukocytes impact neutrophil survival in the presence of microbial antigens.
We next examined the ability of various bacteria previously reported to be altered in HIV-infected individuals [37] to impact neutrophil survival in vitro. All bacterial species significantly increased neutrophil survival after incubation with whole blood, with the exception of Lactobacillus species (Fig 6A and Fig 6B). LPS purified from Escherichia coli (E. coli) was used as a positive control due to its ability to significantly impact neutrophil survival in the previous experiment. Importantly, Lactobacillus plantarum and Lactobacillus rhamnosus decreased neutrophil survival. Additionally, we found that both Lactobacillus species decreased total neutrophil frequencies and that the increased neutrophil survival observed upon stimulation by non-Lactobacillus bacteria resulted in sustained neutrophil frequencies (Fig 6C). This suggests that the non-Lactobacillus bacteria shown to reduce active Caspase-3 expression leads to increased neutrophil survival, rather than causing a different form of cell death such as necrosis. In these experiments, the whole blood neutrophils did not respond differently to any stimulation condition based on HIV status. Additionally, given that these were blood neutrophils, we did not expect other demographic factors such as sexual orientation to have an impact and therefore did not collect such data on these individuals. Also, because Ruminococcus bromii and Bacteroides fragilis increased neutrophil survival similarly to the Prevotella species, we assessed the Lactobacillus:Ruminococcus and Lactobacillus:Bacteroides ratios in vivo and found no differences based on HIV status and no association with neutrophil survival (Supplementary Fig 8 in S1 Supporting Information).
Given that bacteria in vivo are unlikely to be spatially separated and would therefore interact with cells simultaneously, we assessed the ability of Lactobacillus to reduce neutrophil survival in the presence of LPS in a subset of individuals. We used the same LPS purified from E. coli as was used in previous experiments to test the ability of Lactobacillus to override strong signals of neutrophil survival. We observed that L. plantarum incubated with whole blood in the presence of LPS reduced neutrophil survival similarly to that of L. plantarum alone (Fig 6D and Fig 6E). L. plantarum is commonly used in therapeutic studies with probiotics and often found in available probiotic supplements. It has also been investigated alone and in supplements with other bacteria for its ability to reduce inflammation in SIV and HIV[43,44]. Therefore, this observation has important implications for the potential use of Lactobacillus to therapeutically target neutrophil survival, as it suggests that Lactobacillus found in common probiotics could potentially override survival signals induced by other microbes or microbial molecules in the environment.
We additionally assessed the effect of different enteric bacteria on the survival of isolated neutrophils and observed much lower survival among isolated neutrophils compared to those in whole blood after 20 hours of incubation in media alone (Supplementary Fig 9 in S1 Supporting Information). No significant effects of bacterial stimulation were observed after correction for multiple comparisons. It is likely that the isolation procedure activated apoptosis pathways that could not be reversed by subsequent stimulation. We therefore cannot conclude to what extent the bacteria may alter homeostatic apoptosis of isolated neutrophils without this confounding activation upon isolation. Further studies will be necessary to better determine the effects of other leukocytes in the alteration of neutrophil survival by different bacteria.
Neutrophils have been suggested to contribute to intestinal inflammation in HIV infection, however the causes and consequences of neutrophil accumulation in the intestines during infection are not well understood. Here we demonstrate that neutrophil lifespan is altered in the GI in treated HIV infection and report data suggesting a potential link between the intestinal microbiome and neutrophil accumulation. While numerous studies have linked reduced neutrophil apoptosis to disease severity in IBD, these are the first data suggesting that increased neutrophil lifespan contributes to GI neutrophil accumulation in HIV Infection. Further, we report that mucosal bacteria have differential effects on neutrophil survival in vivo and in vitro, suggesting that an altered microbiome resulting from a combination of HIV infection and lifestyle, such as sexual orientation, may contribute to neutrophil accumulation and inflammation in the GI through effects on neutrophil apoptosis.
We assessed the alterations to the microbiome in this cohort by 16S rRNA gene sequencing of colon biopsies, and we report that microbial composition was modestly associated with HIV status and robustly associated with sexual orientation in men. Taken together, these data suggest that both HIV infection and sexual orientation may contribute to observed alterations in colon microbial composition in HIV-infected individuals in this study. Interestingly, we observed some alterations in bacterial abundances based on HIV status in the colorectal biopsies that differ from the results of previously published studies in HIV infection. Specifically, decreased Bradyrhizobium associated with HIV status in our cohort (although this was not significant using a multiple comparisons approach), while a previous study reported increased Bradyrhizobium in the duodenum of HIV-infected individuals[45]. However, this previously reported increase was only observed in individuals with abnormal blood CD4+ T cell counts, suggesting that differences in treatment and disease progression likely contribute to differences in observed microbiome alterations. Additionally, we observed that decreased Peptoniphilus associated significantly with HIV status following adjustment for sexual orientation, while a previous study reported an increase in the rectum in ART-treated, HIV-infected individuals[46]. This could be due to differences between colon and rectal microbial composition as well as differences in cohort demographics. For instance, the authors note that Peptoniphilus is reduced in the penile microbiota following circumcision and is found in vaginal and genitourinary tract infections[47,48], suggesting that cohorts may have different abundances of this bacteria in the lower GI depending on circumcision rates and genital tract microbial composition. Additionally, we observed a positive association between Bilophila and HIV status following adjustment for sexual orientation, while a previous study reported reduced Bilophila in the stool in HIV-infected individuals in China[49]. However, the Chinese cohort examined included individuals with prior antibiotic use as well as untreated individuals and they did not control or match for sexual orientation. These differences highlight the variability in reported microbiome alterations that could result from different GI sample types and assessing cohorts from different geographic areas with different antibiotic usage, treatment status, and disease progression.
We further examined Prevotella and Lactobacillus, two genera we hypothesized may impact neutrophil apoptosis based on previously published studies. In this study, although Prevotella was among the top genera that associated with HIV status, this association was not significant by a multiple comparisons approach. Further, after we adjusted for demographics and sexual orientation, Prevotella no longer associated with HIV status, supporting recently published studies linking Prevotella enrichment to MSM sexual orientation rather than HIV status[23]. In accordance with that study, we observed a significant association between Prevotella and MSM in this cohort; however, this association did not remain significant after adjusting for HIV status.
Several studies have indicated that Lactobacillus is depleted in HIV-infected individuals[45,50] and a higher abundance of gut Lactobacillales was associated with reduced microbial translocation, increased CD4+ T cells in the periphery and gut, and less immune activation in HIV-infected, ART-treated individuals[51]. In this study, Lactobacillus did not specifically associate significantly with HIV status after accounting for multiple comparisons. However, we observed a significant difference between the Lactobacillus:Prevotella ratios in HIV-infected and uninfected individuals, suggesting an altered balance of these bacteria. Additionally, we observed an association between this ratio and neutrophil survival in matched colon biopsies, providing evidence that microbial composition may impact neutrophil survival in vivo. Further, Prevotella demonstrated the ability to increase neutrophil survival in vitro, although this ability was not specific to Prevotella, as the Rumincoccus and Bacteroides species similarly increased neutrophil survival. However, in combination with the in vivo ratio and selbal balance analyses indicating that Prevotella abundance associates with neutrophil survival, these data provide evidence of a role for Prevotella in increased GI neutrophil survival within the context of the greater microbial community. Neutrophil survival and Prevotella alterations coexist in several inflammatory conditions, including bacterial vaginosis[52,53], rheumatoid arthritis[54–56], and periodontitis[57,58], suggesting that this link warrants further investigation in larger, matched cohorts of HIV-infected individuals. Finally, the selbal balance analysis indicates that other bacteria, including Roseburia and Treponema, may positively associate with neutrophil lifespan in vivo. Future studies should assess the effects of different bacterial balances and combinations on neutrophils to better understand the role of altered bacterial abundances in neutrophil lifespan in inflammatory conditions.
Importantly, what is considered to be a healthy microbiome varies geographically, and it has been argued that this is likely due to dietary differences[59,60]. We did not assess diet in these individuals, which could have impacted our ability to detect differences in Prevotella and Lactobacillus between the infected and uninfected individuals in this cohort, as both genera have a demonstrated link to diet[61]. Future studies are needed to better understand how such important lifestyle and demographic factors may impact the relationship between the microbiome and neutrophil frequency and lifespan in vivo.
Given our observation that HIV-infected individuals in our study population had altered microbial composition, we performed in vitro assessments of the effects of various enteric bacteria on neutrophils. These experiments revealed the unique ability of Lactobacillus species to increase neutrophil apoptosis, which has not been previously reported and has important implications for potential therapeutic intervention in HIV and other diseases of intestinal inflammation. Importantly, the ability of bacteria to alter neutrophil lifespan in vitro was not affected by the HIV status of the individual, suggesting that these interactions could occur in the context of HIV infection but are not specific to HIV-infected individuals. Anti-inflammatory effects of various Lactobacillus species are well described, and have been attributed to several factors: 1) the ability of superoxide dismutase secreted by Lactobacillus to neutralize reactive oxygen species; 2) the inhibition of the NF-κB pathway leading to a reduction in pro-inflammatory cytokines and chemokines; and 3) the expansion of regulatory T cells [62–64]. Indeed, the increase in neutrophil apoptosis we report in the presence of Lactobacillus may be caused by NF-κB inhibition, which is known to drive the production of survival factors in neutrophils and be an important regulator of apoptosis[40]. These data also suggest that the previously reported ability of Lactobacillus to reduce intestinal inflammation in vivo may be a consequence of increased neutrophil apoptosis and reduced neutrophil accumulation[65] [66,67].
In HIV infection, increased microbial translocation and dysbiosis may also result in increased neutrophil recruitment in addition to increased neutrophil lifespan. As such, IL-8, a potent neutrophil chemokine and a regulator of neutrophil survival, is increased in the colorectal mucosa of HIV infected individuals relative to uninfected controls[68,69]. Additionally, studies investigating neutrophil lifespan have demonstrated that neutrophil apoptosis is regulated by both cytokines released by monocytes and by direct interaction in response to TLR stimulation, and it is likely that both contribute to alterations in neutrophil apoptosis induced by bacteria[15]. The relative contribution of direct interactions with neutrophils and secreted factors from other leukocytes should be further assessed. Finally, bacteria produce molecules in vivo that could additionally impact neutrophil apoptosis. For example, short chain fatty acids have been demonstrated to inhibit NF-κB activation and attenuate antimicrobial and inflammatory neutrophil responses to LPS[61]. Therefore, the ability of microbial products to alter neutrophil apoptosis and neutrophil-driven inflammation in vivo is an important area of future research.
It is important to point out that this study has several limitations. Due to the requirement that samples be processed and analyzed fresh to assess neutrophils, there were geographic and time constraints that inhibited our ability to recruit more individuals for this study or narrow the focus of our recruitment. As such, we have a relatively small sample size and were unable to recruit controls specifically matched for sexual orientation, age, race, and other demographic characteristics. While we applied the appropriate statistical corrections to account for differences in these characteristics between the HIV-infected and control groups, additional studies are necessary to further assess the relationship between neutrophils and the microbiome in HIV infection in vivo, particularly in the context of sexual orientation. Likely due to the relatively small sample size, we were unable to provide evidence that microbiome alterations associate with GI neutrophils within HIV-infected individuals only, and our in vivo analyses associating microbiome alterations and mucosal neutrophil lifespan included both HIV-infected and HIV-uninfected individuals given the skewed distribution of the variables within each population. A larger cohort with a wider distribution of data is necessary in order to fully assess the contribution of microbial changes to neutrophil alterations in the context of HIV infection in vivo. Additionally, we were unable to assess neutrophils or the neutrophil/microbiome relationship in untreated individuals because most individuals that are aware of their status in the geographic areas of this study are on treatment. However, we believe that these data are clinically relevant given that individuals are now treated immediately upon diagnosis. Future studies in areas where there are more untreated individuals could be beneficial to further assess neutrophils in HIV infection and their role in GI inflammation in HIV.
Neutrophil apoptosis has emerged as a therapeutic target for the resolution of acute and chronic inflammation in the lungs, the intestines, and arthritis but no strategies are approved for use in humans to-date[70]. Here, we provide evidence that this may also be a therapeutic target to reduce intestinal inflammation in HIV-infected individuals by demonstrating increased GI neutrophil survival in HIV infection. Further, the ability of Lactobacillus to uniquely reduce neutrophil survival and neutrophil frequency suggests that ongoing studies investigating Lactobacillus-containing probiotics should additionally assess neutrophil accumulation as a potential mechanism for any observed alterations in intestinal inflammation. Finally, these data lead to new avenues of research whereby commensal bacteria, their surface antigens, and their products should be further assessed for their therapeutic ability to reduce neutrophil accumulation and tissue damage in HIV infection and other inflammatory conditions.
HIV+ and HIV- study participants were recruited through either the University of Washington Center for AIDS Research, University of California San Francisco SCOPE cohort, Northwestern University, or the University of Washington AIDS Clinical Trials Unit. Biopsies were obtained by either colonoscopy or rectosigmoidoscopy. Blood samples for bacterial stimulations were collected in EDTA from HIV+ and HIV- participants recruited through the University of Washington Center for AIDS research. All HIV+ participants were on potent combination antiretroviral therapy at time of biopsy or blood draw with no detectable plasma viral load. The appropriate Institutional Review Boards approved all protocols and informed written consent was obtained from all participants. Table 1 describes relevant participant demographics information including age, sex, sexual orientation, ethnicity, CD4+ T cell count, and time since HIV diagnosis for the individuals that had biopsies that could be processed, stained in real-time, and met the threshold criteria (described below) for accurate neutrophil measurements by flow cytometry. Demographic characteristics for additional biopsy samples that did not meet these criteria but were able to be assessed by 16S sequencing are included in Supplementary Table 1 in S1 Supporting Information. Blood samples from additional donors were obtained from the University of Washington AIDS Clinical Trials Unit for in vitro experiments (as depicted in Fig 6 and Supplementary Figs 7 and 9 in S1 Supporting Information), and HIV status was the only demographic data made available for those individuals.
The appropriate Institutional Review Boards approved all protocols and informed written consent was obtained from all participants. Approval numbers at each institution are as follows: Northwestern University: STU00200953: Rectal Biopsies; University of Washington/Harborview Medical Center: STUDY00002763; University of California, San Francisco: 10-01218. All participants were adults >18 years of age.
Gut biopsies were enzymatically digested with media (RPMI 1640 with 2.05mM L-glutamate, 100U/ml Penicillin, 100μg/ml Streptomycin [all from GE Healthcare, Logan, UT]) supplemented with Liberase (40 μg/ml, Sigma-Aldrich, St. Louis, MO) and DNAse (4 μg/ml, Sigma-Aldrich) for 1 hour at 37°C with vigorous stirring, ground through a 70-μm cell strainer into a single cell suspension, and then analyzed by flow cytometry.
Neutrophils were isolated by lysing the red blood cells in whole blood with ACK lysing buffer (ThermoFisher Scientific) and then labeling the leukocytes with CD15 microbeads (Miltenyi Biotec, Bergisch Gladbach, Germany). The labeled cells were then loaded onto a MACS Column, which was placed into a magnetic MACS Separator (both from Miltenyi Biotec). The retained CD15+ cells were then washed with ice-cold buffer (PBS with 0.5% BSA and 2 mM EDTA). The column was then removed from the separator and the labeled cells were eluted in ice-cold buffer. Cells were counted and used immediately in bacterial or TLR stimulation experiments.
Single cell isolations from biopsies were analyzed by flow cytometry immediately after isolation. Biopsies and blood or isolated neutrophils from the bacteria stimulations were stained using the following surface antigen mouse anti-human antibodies with clone denoted in (), from Becton Dickinson, and Co. (BD) Biosciences (Franklin, NJ) unless otherwise stated: CD45 PE-CF594 (HI30),CD11b APC-Cy7 (ICRF44), CD66b PE (Biolegend, G10F5), CD49d PE/Cy5 (Biolegend, 9F10), CD20 Brilliant Violet 570 (Biolegend, 2H7), CD3 Brilliant Violet 570 (Biolegend, UCHT1), CD16 BV605 (3G8), CD15 BV650 (HI98), and CD14 BV786 (M5E2). Following surface staining, cells were permeabilized using Cytofix/Cytoperm (BD Biosciences). Intracellular active Caspase-3 was stained using a v450-conjugated rabbit anti-human antibody (BD Biosciences, C92-605). Stained samples were fixed in 1% paraformaldehyde and collected on an LSR II (BD Biosciences, La Jolla, California). Analysis was performed in FlowJo (version 9.7.6, Treestar Inc., Ashland, Oregon). Samples with less than 100 events in the neutrophil gate were not included in analyses due to an inability to ensure adequate fluorescence separation of populations and therefore accurate gating of the neutrophil cell population.
Genomic DNA was extracted from colon tissue biopsies using the QIAamp PowerFecal DNA Kit (QIAGEN, Valencia, CA). DNA for 16S rRNA sequencing was processed following the Earth Microbiome Project protocols (http://press.igsb.anl.gov/earthmicrobiome/protocols-and-standards/16s/) with the following modifications. During the library preparation, each DNA sample was amplified in triplicate using the FailSafe PCR System (Epicentre, WI) and the 515FB-806RB primer pair to generate a 400 bp amplicon from the V4 variable regions of the 16S rRNA gene. The triplicates reactions were pooled, quantified using Qubit dsDNA High Sensitivity Assay Kit (ThermoFisher Scientific, Waltham, MA), and visualized using a LabChip GX (PerkinElmer, MA). Using the concentration of the 400 bp peak, 0.4 ng of each library was pooled into a single sample. The ~400 bp amplicon from the pooled sample was isolated using a BluePippen System (Sage Science, MA), cleaned using AMPure XP Beads (Beckman Coulter, IN) and quantified using the KAPA Library Quantification Kit (KAPA Biosystems, MA). Sequencing was carried out as detailed in the EMP protocol; specifically, 7 pM of the pooled library with 30% PhiX phage as a control was sequenced using a 300-cycle Illumina MiSeq Kit.
16S rRNA gene sequence data was analyzed using the QIIME 2 software package[71]. Sequences were classified using the Naïve Bayes classifier trained on Greengenes 13_8 and binned into operational taxonomic units (OTUs) at 99% sequence similarity[72]. OTUs were then classified using Greengenes 13_8 and converted to relative abundance at the family and genus taxonomic levels for visualization and statistical analyses. Taxonomy plots were created in RStudio Version 1.1.422 using the phyloseq package[73].
Prevotella stercorea (DSMZ #18206, Braunschweig Germany), Prevotella copri (DSMZ #18205), Bacteroides fragilis (ATCC #25285, Manassas, Virginia), and Ruminicoccus bromii (ATCC #27255) were all grown in anaerobically in chopped meat broth (Hardy Diagnostics, Santa Maria, CA) supplemented with 1% trace minerals (ATCC), 1% vitamin supplements (ATCC), 0.05% Tween 89 (Sigma-Aldrich, Saint Louis, MO), 29.7 mM acetic acid (Sigma-Aldrich), 8.1 mM proprionic acid (Sigma-Aldrich), and 4.4 mM butyric acid (Sigma-Aldrich). Acinetobacter junii (ATCC #17908) was grown aerobically in nutrient agar (BD). Lactobacillus plantarum (ATCC #14917) and Lactobacillus rhamnosus (ATCC #53103) were grown aerobically in MRS broth (BD). All bacteria were counted using CountBright Absolute Counting Beads (Thermo Fisher Scientific, Waltham, MA) and Syto 9 dye (Thermo Fisher Scientific) on the LSR II. Bacteria were frozen as dry cell pellets until reconstituted in PBS for use in stimulations. For stimulations, bacteria were added at 2.5 bacteria per leukocyte to 100 μl of whole blood or 500,000 isolated neutrophils in 1 ml R10 media (RPMI 1640 with 2.05mM L-glutamate, 100U/ml Penicillin, 100μg/ml Streptomycin, and 10% fetal bovine serum [all from GE Healthcare, Logan, UT]) and incubated aerobically for 20 hours at 37°C. Following the incubation, the supernatant was removed and the blood and isolated neutrophils were washed before being analyzed by flow cytometry.
Differences in neutrophil frequencies and active Caspase-3 expression between infected and uninfected individuals were determined by Mann-Whitney test. In an effort to control for imbalances in potential confounders they were included as covariates in subsequent multivariate models. Adjusted multivariate analyses were conducted by regressing log-transformed neutrophil frequencies and active Caspase-3 expression on infection status, age, race (white vs. non-white), sex and sexual orientation. Note that sex and sexual orientation were treated as a single trinary variable, coded using two dummy variables, with female sex as the referent category. A paired one-way ANOVA was used to assess differences between groups stimulated with different bacteria or TLR agonists in the in vitro experiments followed by a Dunnett’s or Tukey’s post-hoc analysis for multiple comparisons. p-values reported are adjusted p-values from the post-hoc analysis. Correlations were assessed using the Spearman’s rank correlation analysis. Taxon counts at the family or genus level were center-log ratio (CLR)[74] transformed to accommodate compositionality and encourage normality. Taxa present in fewer than 25% of samples were omitted. Community level (beta-diversity) analyses were conducted using MiRKAT[75], which is a generalization of PERMANOVA[76], under Euclidean distance. For the adjusted analyses of associations with total neutrophil and neutrophil survival we included age, race, sex/sexual preference, and HIV status as covariates. For the adjusted analyses of associations with HIV status, we included age, race and sex/sexual preference as covariates. For the adjusted analyses based on MSM status we included age, race, and HIV status. Associations with individual taxa were determined by regressing the abundance of each taxon on the variable of interest and additional covariates as described above followed by false discovery rate[77] control for multiple testing.
In addition to community level and individual taxon level analyses, we used the previously published selbal approach [78] to explore whether balances of particular bacterial taxa could predict neutrophil survival. We considered the percentage of active Caspase-3 low, CD16 high neutrophils as the outcome and used the selbal procedure as described to search for a global balance of taxa related to the outcome. No covariates were included.
|
10.1371/journal.pmed.1002101 | Assessment of Adverse Events in Protocols, Clinical Study Reports, and Published Papers of Trials of Orlistat: A Document Analysis | Little is known about how adverse events are summarised and reported in trials, as detailed information is usually considered confidential. We have acquired clinical study reports (CSRs) from the European Medicines Agency through the Freedom of Information Act. The CSRs describe the results of studies conducted as part of the application for marketing authorisation for the slimming pill orlistat. The purpose of this study was to study how adverse events were summarised and reported in study protocols, CSRs, and published papers of orlistat trials.
We received the CSRs from seven randomised placebo controlled orlistat trials (4,225 participants) submitted by Roche. The CSRs consisted of 8,716 pages and included protocols. Two researchers independently extracted data on adverse events from protocols and CSRs. Corresponding published papers were identified on PubMed and adverse event data were extracted from this source as well. All three sources were compared. Individual adverse events from one trial were summed and compared to the totals in the summary report.
None of the protocols or CSRs contained instructions for investigators on how to question participants about adverse events. In CSRs, gastrointestinal adverse events were only coded if the participant reported that they were “bothersome,” a condition that was not specified in the protocol for two of the trials. Serious adverse events were assessed for relationship to the drug by the sponsor, and all adverse events were coded by the sponsor using a glossary that could be updated by the sponsor. The criteria for withdrawal due to adverse events were in one case related to efficacy (high fasting glucose led to withdrawal), which meant that one trial had more withdrawals due to adverse events in the placebo group. Finally, only between 3% and 33% of the total number of investigator-reported adverse events from the trials were reported in the publications because of post hoc filters, though six of seven papers stated that “all adverse events were recorded.”
For one trial, we identified an additional 1,318 adverse events that were not listed or mentioned in the CSR itself but could be identified through manually counting individual adverse events reported in an appendix. We discovered that the majority of patients had multiple episodes of the same adverse event that were only counted once, though this was not described in the CSRs. We also discovered that participants treated with orlistat experienced twice as many days with adverse events as participants treated with placebo (22.7 d versus 14.9 d, p-value < 0.0001, Student’s t test). Furthermore, compared with the placebo group, adverse events in the orlistat group were more severe. None of this was stated in the CSR or in the published paper.
Our analysis was restricted to one drug tested in the mid-1990s; our results might therefore not be applicable for newer drugs.
In the orlistat trials, we identified important disparities in the reporting of adverse events between protocols, clinical study reports, and published papers. Reports of these trials seemed to have systematically understated adverse events. Based on these findings, systematic reviews of drugs might be improved by including protocols and CSRs in addition to published articles.
| Most drugs have adverse effects, or harms, that may become evident in clinical trials.
Pharmaceutical companies seeking to market a new drug must report adverse effects observed in trial participants in the Clinical Study Reports (CSRs), which they provide to regulatory authorities.
Additionally, investigators may report harms in published reports of their trials.
We sought to understand the accuracy, and potential bias, in harms reporting for trials of orlistat, a slimming drug from Roche approved in Europe in 1998 and still marketed in Europe today.
Using a Freedom of Information Act request to the European Medicines Agency (EMA), we obtained CSRs describing seven clinical trials of orlistat. We studied protocol instructions to investigators for reporting harms, the actual reporting of harms in individual CSR records versus summaries, and the final reporting of harms in published papers.
We found that protocol instructions to trial investigators had the potential to dilute the appearance of drug-associated harms.
Between 3% and 33% of the total adverse effects from CSR summaries were described in published papers.
In one trial, we counted adverse events individually and found that both the number of adverse effects and the number of days with adverse effects in participants taking the drug were understated in the corresponding publication.
The reporting of trials of orlistat in the 1990s understated harms in the summarised results submitted to the EMA for drug approval and in the published papers.
Based on the characteristics of harms observed and reported in these trials, we suggest that reports of harms include duration of adverse effects.
We also suggest that systematic reviews of drugs might be improved by including protocols and CSRs in addition to published articles.
| Randomised trials generally underreport harms, which according to Consolidated Standards of Reporting Trials (CONSORT) is the totality of adverse events [1]. In 14% of 185 randomised trials published in major medical journals in 1997, adverse reactions were not mentioned at all, and in 32% they were not shown for each arm, or general statements were used [2]. Only 16% of the trial reports described how adverse events were identified [2], which is also problematic because the way the investigator obtains information impacts greatly on the number [3] and reported characteristics of the events [1]. Another survey found that only 18% of all paediatric randomised trials published between 2006 and 2009 reported harms adequately according to the CONSORT guidelines [4].
Industry-sponsored trials are more likely than other trials to conclude that a drug is safe [5]. A similar bias exists in industry-supported reviews of drugs, which are less transparent, have fewer reservations about methodological limitations of the included trials, and have more favourable conclusions than Cochrane reviews of the same drugs [6].
Selective reporting of harms can have disastrous consequences. Rofecoxib, a selective cox-2 inhibitor, was withdrawn from the market in 2004 due to cardiac adverse events [7]. A study published in 2000 by Merck could have revealed the risk, but due to a nondisclosed cut-off date, not all events were included [8,9]. Pfizer stated that celecoxib does not cause heart attacks at a Federal Drug Administration (FDA) hearing in 2005, despite having evidence to suggest the contrary [10]. In 2009, they called the evidence “inconclusive” in information given to patients invited to participate in a clinical trial [11]. It is estimated that both drugs have caused many deaths due to adverse events [12].
Many steps, decisions, and assumptions precede the reporting or omission of an adverse event. Lack of recorded details has been identified as a problem [13]. Adverse events are coded by the sponsor, which is a potentially bias-prone process. Little is known about whether this process is blinded. In a recent review, we found that reliable interobserver studies of coding have not been conducted, and that modern coding systems might have made statistical detection of adverse events more difficult because of splitting similar events into several categories [14].
When a pharmaceutical company applies for marketing authorisation at a drug agency, they submit an application that includes detailed reports about each of the clinical trials also known as clinical study reports (CSRs). The CSRs are formatted in accordance to a standard developed in 1995 by the International Conference of Harmonisation (ICH) [15].
To better understand these issues of selective reporting, bias, and inadequate recording, we sought to describe how a major pharmaceutical company seeking regulatory approval addressed the issue of collecting and reporting data on harms in its clinical trials. In 2011, we submitted a Freedom of Information request and obtained the CSRs from placebo controlled trials of orlistat—an anti-obesity drug—submitted to the European Medicines Agency (EMA) by Roche for marketing authorisation [16]. Orlistat was approved by the EMA in 1998 but, along with other slimming drugs, has since encountered regulatory barriers. Nearly all slimming pills (but not orlistat) have been withdrawn from European markets because of harms [17–19]. In 2011, the FDA issued a warning regarding orlistat based on 13 cases of liver failure [20]. The CSRs include trials’ protocols and anonymised individual participant data (see Table 1) with narrative descriptions of adverse events. We have used these unique data to study how adverse events and methods for obtaining them were presented in protocols, CSRs, and published papers. The objective was prespecified in our protocol, but we also planned to explore the issues in more detail.
Seven placebo-controlled randomised trials of orlistat were included in the application for marketing authorisation. In total, the CSRs consisted of 8,716 pages and included 4,225 participants. The first couple of pages in the CSRs were a brief synopsis followed by module 1 and 2. Module 1 contained the “core report,” which consisted of around 100 pages and was structured by these sections: methods, results and discussion, and conclusion. The module also contained appendices, which consisted of several hundred pages. Selected appendices were an overview of adverse events by organ system, full anonymised individual participant data of adverse events including the information described in Table 1, and detailed narrative descriptions of serious adverse events and events leading to withdrawal from the study. Module 2 contained the study protocol, a blank case report form, a table comparing investigator adverse event terms with the chosen term from the dictionary, bioanalytical report, an investigator list, and a randomisation list. All data were unredacted. We did not receive module 3, 4, and 5 from the EMA, and we determined that information in modules 1 and 2 was sufficient for our analysis. According to a table of contents these modules included participant listings of efficacy data, adverse event listings by organ system, and a statistical report.
For the protocols, two investigators (EIP and JBS) independently extracted names of authors, withdrawal criteria, coding strategies, and information about how adverse events were planned to be recorded and summarised. We also extracted strategies for handling vitamin deficiency (as orlistat decreases absorption of fat from the gut, it might affect the absorption of fat-soluble vitamins) and measures of quality of life (which can potentially reflect harms).
For the core report of the CSRs, the same investigators noted identifiers such as investigator names, start and end dates, treatment duration and countries, and extracted the following data: from the synopsis, all information about investigator-reported adverse events; from the methods section, information about withdrawals, harms, and quality of life; from the results section, the overview of the adverse events section, number of participants, mean age, mean BMI, gender, participants withdrawn, adverse events, serious adverse events, gastrointestinal adverse events, deaths, quality of life scores, liver function tests, increased heart rate, and number of patients with gallbladder diseases and low vitamin levels; and from the discussion section and the conclusion, all text describing adverse events.
We searched PubMed with “orlistat or Xenical” to find the corresponding publications. The search returned 1,433 items, which were screened by one author. We included studies describing investigator-reported adverse events from each of the seven trials included in the application for marketing authorisation. Based on the abstract, we downloaded 35 articles as full text for further evaluation. We identified nine papers that described the seven trials individually. Each trial had a detailed primary publication that summarised investigator-reported adverse events [22–28] and was included in our study. The two additional publications did not contribute additional data about adverse events [29,30] and were excluded. An additional seven papers with pooled estimates from the trials did not contain investigator-reported adverse events and were excluded [31–37]. One paper explored abnormal liver function test [37], which we had extracted from the CSRs. We had planned to examine how investigator-reported liver safety of orlistat was reported in publications, but since none of the trial reports had investigator-reported adverse events related to the liver, we excluded the study. The remaining studies were excluded because they did not describe one of the seven trials from the application. We extracted information about adverse events from the seven primary publications.
We converted all individual participant adverse event listings from the CSR for one trial (trial 7) by using text recognition software (ABBYY FineReader 10) and transferred the data to Excel. Trial 7 was chosen as an example because it was the newest and also one of the smallest, and it had a relatively simple design.
We studied how adverse events were categorised, recorded, and analysed by comparing protocols, CSR core reports, and publications. We also looked for signals within the CSRs of elevated liver function tests and vitamin deficiency but did not compare this or other abnormal laboratory values to publications, as they are not traditionally considered adverse events and there is less guidance on how they should be reported. Adverse events reported in the CSRs were compared with corresponding publications, and in trial 7 we checked whether adverse events were summarised consistently with the individual adverse events listed in an appendix.
The protocols described seven phase III randomised trials, all with a placebo arm, and with orlistat given as 30 mg, 60 mg, or 120 mg three times a day (Table 2). The duration of the studies was 52 to 104 wk. Trial 2 re-randomised the participants to either placebo or orlistat after 52 wk of treatment and trial 5 changed the dose after 52 wk for half the participants. Participants and treating physicians were blinded to the treatment, but whether the coders of adverse events were blinded was not mentioned in any of the documents.
The trials were conducted between 1992 and 1996 in the United States and Europe. They all had a lead-in period, which mostly lasted for 4–5 wk, when the participants received placebo along with dietary advice. Based on pre-defined criteria some participants were excluded based on their performance in this period. The included participants had a BMI between 28 and 43. Trial 7 only included participants with type II diabetes, whereas the other trials excluded such participants.
All protocols mentioned that vital signs, adverse events, routine laboratory tests, fat-soluble vitamin levels, and ECG should be recorded. All protocols had at least eight withdrawal criteria. Apart from “new smokers,” which was an additional criterion in five protocols, the withdrawal criteria were the same. “Administrative reasons” or “other reasons” were sufficient for withdrawal and were not further specified.
Three protocols (trials 1–3) contained an appendix on how to code adverse changes in defecation patterns; these appendices included guidance and a term list for adverse events, with descriptive definitions for each term (all events were in American English; for consistency, we have used British English). All three protocols read: “In this dictionary the term diarrhoea and constipation has been avoided. In fact, the use of these terms could cause some misunderstandings, as there is no well-accepted definition…” [38]. Instead the following categories were used: “increased defaecation,” “liquid stools,” “soft stools,” “fatty/oily evacuations,” “oily spotting,” “faecal urgency,” “faecal incontinence,” “flatus with discharge,” “decreased defaecation,” “pellets,” and “solid stools.” Even though protocols 4–7 did not contain this appendix, it may have been used because, for all these trials, instances of “diarrhoea” in the CSRs list of adverse events were re-categorised.
The protocols included between 9 and 17 planned visits per participant during the first year, and adverse events were to be recorded at each visit on the case report forms. Only a change from the participants’ pre-treatment condition was to be considered an adverse event, and the protocols provided no guidance on how to question the participants. The investigator was to relate the severity to daily function and also judge the relationship to treatment (two CSRs contained an appendix which offered guidance on this).
For quality of life, six protocols specified that the main outcome was “comparative rates of change” for the subscales “health distress and emotional functioning.” The scales were not specified in any of the protocols; instead, they referred to a questionnaire included in the protocol, which was a 46-item list divided into seven subscales with no information about how the scores from the subscales were to be combined. Secondary quality of life outcomes were simply described as “a variety of scales.”
Fat-soluble vitamin levels were monitored in a blind fashion and vitamin supplements were prescribed if vitamin levels were below a threshold on two consecutive occasions. One study provided multivitamin tablets for all patients.
The only information on the statistical handling of adverse events was that the treatment groups would be compared using “descriptive statistics.”
As in the protocols, it was not specified how the participants had been questioned about adverse events. Coding guidelines for gastrointestinal adverse event were also provided in the CSRs but in contrast to the protocol (trial 1 and 2) some of the terms were marked with an asterisk. The terms without an asterisk should only be considered adverse events when “described as bothersome by the patient” and these included “fatty/oily stool,” “liquid stools” (which term the protocol suggested to be used instead of diarrhoea), “increased defaecation,” “stools soft,” “decreased defaecation,” and “pellets.” “Bothersome” was not a requirement for adverse events outside the gastrointestinal category and was not mentioned in the protocols except for trial 3. Furthermore, according to the narrative descriptions in an appendix to the core report, serious adverse events had been assessed for relationship to the drug by the sponsor, although this was not prespecified in the protocol.
The adverse events were coded according to a Ciba-Geigy modified WHO glossary, which could be updated by the sponsor. For each adverse event described by the investigator, the sponsor would assign a preferred term from the dictionary: “For classification purposes, preferred terms were assigned by the sponsor to the original terms for concomitant medications, diseases, and adverse events entered on the case report form” (trial 2, 3, and 6).
In all CSRs, the methods section described that adverse events would be presented as listings and summary tables by body system, intensity, and relation to drug. For gastrointestinal problems, however, only events more frequent than 1% (four trials) or 3% (three trials) would be summarised.
In all CSRs, the methods section noted that the “primary measure” for quality of life was “overweight distress,” “depression,” and “satisfaction with treatment.” We could not find any explanation, in the CSRs or in the amendments, why the primary outcome for quality of life from the protocol had been changed from “health distress and emotional functioning.”
All CSRs narratively acknowledged that there were many adverse events but also noted that the differences between placebo and active treatment were small, and two CSRs (trial 2 and 3) noted that most adverse events were considered unrelated to the drug by the investigator. Only one CSR (trial 6) mentioned the total number of participants with one or more adverse events in the results section of the core report. None of the core reports mentioned the total number of events for which the difference between placebo and orlistat group was considerably higher, but the information was available in tables.
The increased number of gastrointestinal adverse events observed in the orlistat group was mentioned but it was claimed that this was due to the pharmacological effect of the drug. It was noted that the numbers of gastrointestinal adverse events per participant were often few (1 to 2), and in the core report, there was no information on their duration. All CSRs contained an appendix in module 2 that documented how original investigator terms were coded. In six out of the seven CSRs, investigators used the term “diarrhoea” but in all cases it was re-categorised as “liquid stools” (page 150 [39]). This was not mentioned in the protocol.
Low vitamin levels were common, particularly for vitamin D, for which low levels were found in 19% of the participants receiving orlistat in the largest trial. Low vitamin E and beta carotene levels were the second most common deficiency, which led to additional substitution but rarely withdrawal from the study. The proportion of participants with affected liver function tests was comparable between the two groups. High alanine transaminase and aspartate transaminase were found in 6.2% and 2.1% of the patients in the orlistat group, respectively, and in 6.3% and 2.3% of the placebo patients, respectively. There was no consistent pattern in heart rate changes or in participants with new gallbladder disease.
More participants in the orlistat treatment group were withdrawn due to adverse events (8.1% versus 4.6%, χ2-test p < 0.0001) whereas more participants in the placebo group were withdrawn for “any reason” (28.7% versus 22.0%, χ2-test p < 0.0001). In trial 2, more participants “lost to follow-up” were withdrawn from the placebo group (22 versus 12) and also more participants who “did not cooperate” (26 versus 13). In trial 4, more placebo participants were excluded due to “administrative reasons” (29 versus 10 during the first year).
A brief summary of the papers describing the seven clinical trials are listed in Table 3 [22–28]. There were between 71 and 270 times as many pages in the CSRs as in the corresponding publications. Even though the number of unpublished adverse event is less, the compression factor highlights that a lot of data is being omitted when the study is reported in a publication. Six papers described that “all adverse events were recorded,” (all except trial 5) and one informed that the Ciba-Geigy dictionary was used (trial 3).
Five papers (trials 2, 3, 4, 6, and 7) mentioned that a special dictionary was developed for the expected gastrointestinal adverse events, but none described that only “bothersome” adverse events should be recorded and none described that “diarrhoea” was discouraged as a term.
All papers had severe restrictions on which adverse events were reported, and only four papers presented a table summarising adverse events (trial 1–4). Two papers censored all events that had been considered “unrelated” (trials 1 and 2) by the investigator and only reported events occurring in 3% or 5% of participants. One paper censored both “unrelated” and “remotely related” events (trial 4). Three papers reported only adverse events that were twice as frequent in the orlistat group as in the placebo group (trials 4, 5, and 7), and two of those had the additional criterion that only events occurring in at least 5% of the participants would be reported (trials 5 and 7). These two papers only reported the adverse event rate for the orlistat group.
For four trials, we could extract data on the number of adverse events, and between 3% and 33% of those reported in the CRSs were also reported in the publications (see Table 3). However, the true percentage reported is probably lower, as the grand total of adverse events reported in the CSRs was also lower than in the individual participant data (see trial 7 below).
Only trial 3, which had the greatest difference between placebo and orlistat, reported on quality of life, but there were no data in the paper, only p-values.
All publications mentioned the impact on vitamin measurements. Most reported number of participants who received additional supplements and some reported mean vitamin levels for the entire population.
Trial 1 grouped the gastrointestinal adverse events into two new main categories: “uncontrolled oily discharge,” which included faecal incontinence, flatus with discharge, and oily spotting, and “loose stools,” which included oily evacuation, fatty/oily stool, liquid stools, and soft stools.
In trial 7 almost all participants experienced one or more adverse events (157 participants [96%] in the orlistat groups and 150 [94%] in the placebo group). When we counted the individual participant adverse events we found a total of 3,446 adverse events (2,008 in the orlistat group and 1,438 in the placebo group). These numbers could not be found in the CSR or in the publication, and more events were missing for orlistat than for placebo: in a summary in an appendix in the CSR, the total adverse event count was 1,198 for the orlistat group (60% of our count) and 930 for placebo group (65% of our count). We discovered that multiple events occurring in the same study participant were only counted once; this was not explained in the CSR. We calculated that each participant had 12.8 adverse events, on average, in the orlistat group and 9.6 in the placebo group, corresponding to 3.2 (95% CI: 1.2–5.2, unpaired t test) more adverse events in the orlistat group. This was not mentioned in the report or publication.
The duration of each adverse event was recorded, but was not summarised in the CSR or in the publication. We calculated that the average duration was 22.7 d (95% CI: 20.1–25.2) in the orlistat group and 14.9 d (95% CI: 13.1–16.8) in the placebo group and that the number of days each person was affected by an adverse event was 288 d in the orlistat group and 141 d in the placebo group. Thus, on average, orlistat led to double as many days with adverse events as placebo.
The CSR noted that most adverse events were mild to moderate in intensity. However, we found that the events were more severe in the orlistat group (p < 0.001, χ2 test, not adjusted for dependent observations), which was not mentioned in the CSR or the publication. The relative risk for having a mild adverse event in the orlistat group compared to the placebo group was 0.93 (95% CI: 0.89–0.96); a moderate event, 1.29 (95% CI: 1.13–1.48); and a severe event, 1.39 (95% CI: 0.75–2.59).
More placebo participants were withdrawn due to adverse events (23 versus 12, p = 0.04, χ2 test) which is unusual. However, 14 of the 23 withdrawn participants in the placebo group were discontinued due to abnormal fasting glucose, and this was categorised as withdrawal due to adverse events. The protocol stated that fasting glucose above 220 mg/dl would lead to discontinuation, but only 2 of the 14 withdrawals were listed as an adverse event in the detailed list of adverse events for each participant. Orlistat protects against hyperglycemia, and was also an efficacy outcome in the trial. With the used criteria, the number of withdrawals due to adverse events appears to be more common in the placebo group even though the published paper reported withdrawals due to gastrointestinal adverse events as well. Furthermore, a baseline imbalance in HbA1c could perhaps partly explain the difference (HbA1c was 8.05 in the active group and 8.20 in the placebo group, Student’s t test, p = 0.19). In the first quarter of the trial, 14% of participants on orlistat had a hypoglycaemic episode versus 10% on placebo. In the second quarter, the rates were 12% and 6%, respectively. The CSR referred to an appendix regarding more information, but this was missing. We found 426 hypoglycaemia events in the orlistat group and 300 in the placebo group and an average of 2.7 events per participant in the orlistat group and 2.0 in the placebo group (p = 0.10, unpaired t test).
We found several non-predefined practices in the CSRs and publication that could potentially have resulted in biased reporting of drug-related harms. We had access to protocols, amendments, and content of information of the documents the EMA had not made available to us, and since the filters were not described, they have likely been introduced post hoc. The analysis plan for harms in the protocol consisted of only four lines of text for each CSR. Some gastrointestinal adverse events were only coded if considered “bothersome,” and “diarrhoea” was split into several categories, which can lead to dilution of signals. Only a fraction of adverse events were reported in publications due to various non-predefined censoring filters. Results sections in the core reports of the CSRs often stated that most of the adverse events were considered unrelated to the drug and that they were generally mild to moderate. The many gastrointestinal adverse events were explained as part of the pharmacological effect of orlistat. In one trial we found 11 more participants withdrawn due to adverse events in the placebo group, but this was caused by a high fasting glucose. Fasting glucose and weight loss are correlated, so it was expected that fasting glucose would be higher in the placebo group. With these 11 participants from the placebo arm categorised as withdrawn due to adverse events, adverse events in the orlistat arm may seem less salient. Duration of adverse events was not analysed, even though it was recorded, and no explanation for this was given. In trial 7, including duration of adverse events in the analysis revealed that each treated participant had almost twice as many days with adverse events. Subscales of quality of life were changed without explanation. One aspect of the study design in itself could be a hindrance for identifying adverse events. All trials had a long lead-in period on placebo, from 4 to 24 wk, in which more than 90% of the participants reported at least one adverse event (“complaints”). Should any of these events reoccur during the trial, they would be censored. Since gastrointestinal complaints are normal in healthy people, this type of censoring might have made it more difficult to detect gastrointestinal adverse events caused by orlistat.
Data from observational studies provide an additional perspective on potential harms from orlistat. Slimming pills are often discontinued by the participants [40]. A Canadian study of 16,968 participants on orlistat showed that after 1 y, only 6% of the participants were still taking the drug, and after 2 y, it was only 2% [40]. This suggests that the participants perceive the balance between harms and benefit as unfavourable.
Even though orlistat was approved in 1998, our findings are still relevant. First of all, many drugs approved in this time period are still being sold in large numbers. Secondly, we cannot be sure that analysis plans have improved. Standardised medical dictionaries are now obligatory [14] and the ICH has a guideline on how CSRs should be reported, but analysis of duration of adverse events is still optional and the standard leaves room for interpretation, and therefore, a risk of bias [15]. More supplemental material is available today, but we still need access to protocols to see how authors have arrived at their summary tables. Recent studies have also questioned the quality of the adverse event information in the case report forms [13].
CSRs contain a lot of additional data [41]. Even though it takes more time to use CSRs rather than publications for systematic reviews and meta-analyses, we believe it is worthwhile, as some of the filters used in the case of orlistat would not have been identified in a publication.
Even though publication bias is well covered in the medical literature, few studies have analysed clinical study reports, which in the future could be a very important source of information. Other studies have found that only a fraction of adverse events were reported in published papers compared to the CSRs [42,43]. In one of the studies, Wieseler et al. had access to CSRs of different interventions, and their results might therefore be more generalisable. In our study, we have tried to highlight potential mechanisms of bias that need to be investigated in a confirming study. Our research emphasises the need for detailed analysis plans for harms data.
Our study was explorative and restricted to one drug tested in the mid-1990s; our results might therefore not be applicable to newer drugs. The lack of reporting of important harms could be the consequence of space restrictions in paper journals and could therefore be less of a problem today when electronic appendices are a possibility. Furthermore, standards for reporting CSRs and publications have been developed since the orlistat trials were reported. The CSRs obtained from the EMA and some of the missing modules contained listings related to harms. Based on the table of contents of the missing modules we do not believe access to this data would change our results.
The protocols, CSRs, and publications all reported poorly on how adverse events were planned to be collected, summarised, and analysed. Censoring filters were introduced post hoc, and the guidance on how to code adverse events differed between protocols and CSRs and was absent in publications. The duration of the adverse events was not included in any of the analyses conducted by the sponsor even though the difference between orlistat and placebo was large. Clinical study reports, protocols, and individual participant data should be the primary data sources for systematic reviews of drugs.
|
10.1371/journal.pgen.1007112 | In vivo zebrafish morphogenesis shows Cyp26b1 promotes tendon condensation and musculoskeletal patterning in the embryonic jaw | Integrated development of diverse tissues gives rise to a functional, mobile vertebrate musculoskeletal system. However, the genetics and cellular interactions that drive the integration of muscle, tendon, and skeleton are poorly understood. In the vertebrate head, neural crest cells, from which cranial tendons derive, pattern developing muscles just as tendons have been shown to in limb and trunk tissue, yet the mechanisms of this patterning are unknown. From a forward genetic screen, we determined that cyp26b1 is critical for musculoskeletal integration in the ventral pharyngeal arches, particularly in the mandibulohyoid junction where first and second arch muscles interconnect. Using time-lapse confocal analyses, we detail musculoskeletal integration in wild-type and cyp26b1 mutant zebrafish. In wild-type fish, tenoblasts are present in apposition to elongating muscles and condense in discrete muscle attachment sites. In the absence of cyp26b1, tenoblasts are generated in normal numbers but fail to condense into nascent tendons within the ventral arches and, subsequently, muscles project into ectopic locales. These ectopic muscle fibers eventually associate with ectopic tendon marker expression. Genetic mosaic analysis demonstrates that neural crest cells require Cyp26b1 function for proper musculoskeletal development. Using an inhibitor, we find that Cyp26 function is required in a short time window that overlaps the dynamic window of tenoblast condensation. However, cyp26b1 expression is largely restricted to regions between tenoblast condensations during this time. Our results suggest that degradation of RA by this previously undescribed population of neural crest cells is critical to promote condensation of adjacent scxa-expressing tenoblasts and that these condensations are subsequently required for proper musculoskeletal integration.
| Mobility requires that muscles form appropriate attachments via tendon. The genes regulating this attachment are poorly understood. This is especially true in the head where mesoderm and neural crest cells generate muscle and tendon, respectively. We show that the gene cyp26b1 is critical for muscle tendon attachment in a specific region of the developing head. The movement of tendon generating cells into mature tendons requires cyp26b1. Loss of cyp26b1 disrupts these movements and muscles subsequently extend into inappropriate areas. Cyp26b1 appears to be required in neural crest cells but not the tendon generating cells. Collectively, our work provides insight into the genetic and cellular mechanisms that attach muscle and tendon.
| The movements and functions of the human head depend on 150 individual muscles, and loss of tendon or of tendon-muscle interactions has been implicated in human craniofacial syndromes with muscle defects. Vertebrate craniofacial development is a complex process involving communication between muscles, tendons, cartilages and surrounding tissues. Much of this communication occurs in transient, reiterated, pharyngeal arches. Within each pharyngeal arch are neural crest cells that form a specific set of skeletal elements and mesoderm cells that form a specific set of muscles[1]. The developmental origins of the various cranial tendons remain unclear, save that they derive from the neural crest [2, 3]. Given that neural crest cells control the patterning of cranial muscles [4, 5], it is likely that cranial tendons provide the positional cues needed for this patterning. We know little, however, about the genetics and cellular interactions underlying the mechanisms of development at the myotendinous junction.
We have a limited number of tendon disease models from which to gain an understanding of tendon functions in development. Muscle elements form and elongate after surgical removal of tendon primordia in the developing avian hindlimb, but in these limbs, muscle fibers are ectopically localized [6]. This result suggests that tendons restrict the patterning of limb musculature. Genetic disruption of Scx in mice causes defects in force-transmitting tendons in the trunk, limbs, and tail [7]. However, Scx function is not necessary for tendon to develop and form functional muscle attachments in mutant mice. Scx is also not sufficient for tendon development because tendon progenitors (tenoblasts) are specified in the pharyngeal arches but eventually disappear in muscle-less mutant models [2, 8]. These findings demonstrate that in the head, unlike in the somites [9], specification of tendons is independent of muscle. However, in all tendon populations muscle provides mechanical stimulation that promotes growth factor signaling and Scx expression, and muscle contraction is as necessary as the muscle tissue itself for tendon differentiation [10]. Thus, to study tendon functions it will be essential to discover genetic models in which muscle differentiation is normal but muscle attachments are defective.
Zebrafish is an ideal system for forward genetic analysis and has a complex but well-characterized craniofacial musculoskeletal system (Fig 1A; [1, 2]). In the zebrafish head, the more ventral portion of the first and second pharyngeal arches gives rise to the lower jaw and its supports [1]. Seven muscles extend across the ventral surface of both arches in an hourglass-like shape whose center sits at the anterior tip of the second arch. This central attachment, or mandibulohyoid junction, is comprised of tendon and the ends of four of these muscles, two intermandibularis posterior muscles originating from the first arch and two interhyal muscles from the second arch [2]. Similarly, tendon develops between the two hyohyal muscles at the posterior of the second arch. At the remaining attachments, tendon connects muscle to cartilage. All of these tendons express the zebrafish Scx ortholog, scxa, and xirp2a, whose expression requires muscle [2].
Thus far, there are no in vivo models detailing the morphogenetic dynamics of tendon and muscle. Few genes have yet been determined necessary for tendon development, though tendon function is impaired by genetic knockdown of Scx or tendon extracellular matrix components [7, 11]. In a zebrafish forward genetic screen for craniofacial mutants, we discovered a novel allele of cyp26b1 with a craniofacial skeletal phenotype consistent with previously described loss-of-function alleles [12–14]. Larvae homozygous for this mutation also fail to close their jaws. Our in vivo analysis of wild-type and cyp26b1 mutant embryos revealed essential steps in the morphogenesis of the ventral first- and second-arch muscles responsible for jaw movement. We show that neural crest cells require the retinoic-acid-catabolizing enzyme Cyp26b1 to pattern jaw muscle attachments in the first and second pharyngeal arches. We propose that cyp26b1 expression is required in a mass of non-tendon neural crest that separates adjacent tenoblast populations and promotes tendon condensation necessary for normal muscle patterning.
To investigate the mechanisms responsible for integrating the cranial musculoskeletal system, we analyzed the muscle phenotypes of craniofacial mutants, identified in an ENU-based forward genetic screen in zebrafish. We identified the b1024 allele, whose homozygous mutant offspring had defects in ventral first- and second-arch jaw muscles (see Fig 1A for schematics of these jaw muscles and associated cartilages and tendons). Though these muscles mostly retained their stereotypic pattern in mutants, muscle fibers often split from the main mass to project and terminate ectopically (Fig 1B).
Not all of the jaw muscles were equally defective in b1024 homozygotes (See Table 1 for quantification). Most commonly, the intermandibularis posterior muscles (Fig 1C, arrows, 76% of mutants n = 13/17) and/or interhyal muscles (Fig 1D, white arrowheads, 71% of mutants, n = 12/17) sent ectopic projections into the medial second pharyngeal arch. Intermandibularis posterior muscle fibers overextended toward the midline and terminated at the junction between hyohyal muscles, on the interhyal, or at other positions on the medial second arch surface (Fig 1C and 1D, arrows). Muscle fibers split off from the hyohyal muscles (almost always toward other ectopic muscle fibers) in 24% of mutants (asterisk in Fig 1D, n = 4/17). The intermandibularis anterior muscle fibers split and formed ectopic attachments in 12% of mutants (asterisk in Fig 1C, n = 2/17). The sternohyoideus muscle, which also attaches to this region, projected across the midline in 84% of mutants (arrowhead in Fig 1C, n = 16/19). The disruption to muscle attachment appears highly specific to the ventral interface between the first and second arch as the more dorsal muscles in this region of the head appeared normal in b1024 mutants (S1 Fig). These results indicate craniofacial musculoskeletal development is, at least somewhat, modular and that the b1024 mutation disrupts musculoskeletal connections between the first and second pharyngeal arch midline.
In addition to muscle defects, b1024 mutants had midline skeletal defects (Fig 2A–2D). Homozygous zebrafish had a narrow ethmoid plate and parasphenoid bone (Fig 2C). In the posterior neurocranium, mutants displayed a gap in the parachordal cartilages (arrows in Fig 2A and 2C), medial to the ear, and a gap between the anterior and posterior basicapsular cartilages (arrowheads in Fig 2A and 2C), lateral to the ear (Fig 2C). In the viscerocranium of b1024 mutants, the ventral cartilage elements of the first and second pharyngeal arches, Meckel’s cartilages and ceratohyal cartilages, respectively, were fused in the midline (arrowheads and insets in Fig 2B and 2D). The anterior end of the basihyal cartilage was narrow and hypoplastic compared to wild-type larvae, similar to the ethmoid plate defect (arrows in Fig 2B and 2D). Collectively, these results demonstrated that b1024 mutants have prominent musculoskeletal defects largely localized to the ventral midline.
To identify the genetic lesion in b1024, we performed linkage analysis with simple sequence length polymorphisms. We genetically mapped the b1024 lesion to an interval of chromosome 7 between z8889 and z1239 containing cyp26b1. While muscle phenotypes have not been previously analyzed in zebrafish cyp26b1 mutants, the b1024 skeletal phenotype fit with previous descriptions of cyp26b1 mutants [13]. We discovered a single base substitution (C > T) seven bases into exon 6 (Fig 2E–2G, RefSeq NM_212666). This variant creates an early stop codon (p.Gln384*, RefSeq NP_997831) that is predicted to truncate the last 128, of 511 total, amino acids of Cyp26b1. A single cytochrome P450 superfamily domain comprises most of the protein sequence. This lesion in cyp26b1 truncates 23% of the P450 domain, including the highly conserved heme-binding loop in exon 6, which is required for enzymatic activity [15]. We conclude that b1024 is a recessive loss-of-function, likely null, allele of cyp26b1 like sa2 [14] and ti230g [12].
To characterize the genesis of muscle defects in cyp26b1 mutant zebrafish, we used in vivo time-lapse imaging to track muscle morphogenesis. We used a Tg(-0.5unc45b:mCherry); Tg(fli1:EGFP)y1 double transgenic line for live fluorescence imaging of muscles and neural crest, respectively. By 48 hpf, the interhyal and hyohyal muscle masses emerged on the ventrolateral surface of the second pharyngeal arch and elongated toward the midline (See S1 Movie). Around 51 hpf, two bilateral muscle masses formed on the ventrolateral surface of the first pharyngeal arch. In the next few hours, muscle fibers extended between these masses across the midline to form the intermandibularis anterior muscle. The nascent intermandibularis posterior muscle fibers elongated posteriorly from each mass toward the second pharyngeal arch (See S1 Movie). By 53 hpf, all of these first and second pharyngeal arch jaw muscles were present in both wild-type and cyp26b1 mutant embryos (Fig 3A and 3F). Thus, Cyp26b1 is dispensable for the initial formation of these muscle masses.
As morphogenesis continued the bilateral intermandibularis posterior muscles, from the first pharyngeal arch, integrated in a central four-way junction with the interhyal muscles, of the second pharyngeal arch (S1 Movie, wild-type; Fig 3A–3C). In wild-type embryos, the intermandibularis posterior muscles, upon encountering the anterior edge of the second arch and the medial tips of the elongating interhyal muscles, ceased their rostrocaudal elongation and continued toward the midline with the interhyal muscles (Fig 3B). All four of these muscles arrived in the midline roughly at the same time, around 57 hpf. This “mandibulohyoid junction” (Fig 3C, arrowhead) was situated anteriorly and ventrally in the midline of the second pharyngeal arch, where neural crest cells surrounded the muscle tips (S2 Fig). Lastly, by about 63 hpf, the bilateral hyohyal muscles elongated ventrally past the somite-derived sternohyoideus muscles to reach each other in the midline and form the hyohyal junction (Fig 3D’, arrowhead). Together, these data showed that the attachments for eight muscles, all within the second pharyngeal arch midline, form in a period of roughly ten hours.
Loss of Cyp26b1 function had a profound effect on the connection of these muscles without altering the initial muscle dynamics. As in wild-type embryos, intermandibularis posterior muscles encountered the second arch neural crest and interhyal muscles around 55 hpf in cyp26b1 mutants (Fig 3B and 3G). However, the elongation of interhyal muscles toward the midline was retarded by loss of Cyp26b1 function, (S1 Movie). Their medial tips were further apart at 55 hpf (compare arrows in Fig 3B’ and 3G’), and these and intermandibularis posterior muscles made no connection at the midline as late as 60 hpf. Meanwhile, intermandibularis posterior muscles continued to elongate posteriorly beyond the anterior edge of the second arch in Cyp26b1-deficient embryos (Fig 3I and 3J). Fibers could be observed extending over the neural crest cells of the medial second arch or overlapping with the medial tip of the interhyal muscle. Subsequently, muscle fibers became more disorganized rather than bundling tightly at the mandibulohyoid junction. Later muscle phenotypes indicated that some muscle fibers at the tips of interhyal muscles fan out or split off as well (see Fig 1B–1D). These results demonstrated that Cyp26b1 function is needed for the muscle cell movements necessary to integrate the ventral pharyngeal musculature, particularly at the mandibulohyoid junction, which forms the focus of this manuscript.
To quantify mandibulohyoid junction defects, we measured jaw muscles in 72 hpf embryos (Fig 3E and 3J). By this point the left and right intermandibularis posterior and interhyal muscles connected in the midline in nearly all mutants. However, the width across the mandibulohyoid junction (see arrows in Fig 3C’ and 3J’) was 12.1 μm wider in the average mutant (44.2 μm, n = 11) relative to wild-type fish (32.1 μm, n = 49, Fig 3K). The bundling of muscle fibers at the mandibulohyoid junction in wild-type embryos meant that muscles rarely overlapped or projected ectopically. In the average cyp26b1 mutant, overlapping muscles and ectopic muscle fibers covered an area of 401.9 μm2 (n = 11), compared to 83.7 μm2 in wild-type fish (n = 49, Fig 3L). Together, these findings suggested that Cyp26b1 functions to promote the proper extension of interhyal muscles into the midline and to stop the posterior elongation of intermandibularis posterior muscles at the anterior edge of the second arch.
Because loss of Cyp26b1 disrupts skeletal and muscle morphogenesis, we performed genetic mosaic experiments to test the hypothesis that Cyp26b1 functions in neural crest cells to mediate musculoskeletal development. We transplanted cells from wild-type embryos into cyp26b1 mutant embryos at the onset of gastrulation (Fig 4A). At 24 hours post-fertilization, we selected individuals in which there were large contributions of wild-type crest to the first- and second pharyngeal arches (Fig 4B). We subsequently evaluated skeletal and muscular phenotypes of these individuals at 4 dpf (Fig 4C–4F).
Even though donated neural crest cells contributed to one side of the head, they were able to affect midline skeletal phenotypes in each of 11 mutant hosts. Almost half of our neural crest transplants restored the flared shape of the ethmoid plate (Fig 4D, n = 5/11). Donated crest efficiently rescued the symphysis of Meckel’s (Fig 4E, red arrowhead and inset, n = 7/11) and ceratohyal (Fig 4E, black arrowhead and inset, n = 8/11) cartilages completely or partially. These results indicated that neural crest cells require Cyp26b1 function for ethmoid plate, Meckel’s cartilage, and ceratohyal development.
Consistent with our hypothesis, we found that Cyp26b1 function within the neural crest restored proper jaw muscle patterning. Though donated wild-type neural crest cells could completely rescue mutant muscle phenotypes (n = 1/6), more frequently individuals displayed unilateral improvement (Fig 4F, n = 3/6). In these embryos, jaw muscles looked wild-type on the donor side of the head. These results demonstrated that Cyp26b1 function within the neural crest integrates musculoskeletal development in the face.
The mandibulohyoid junction forms within a population of sox9a-negative neural crest mesenchyme just ventral to the boundary between the ceratohyal cartilage condensations (S2 Fig and S3 Fig). We sought to test the hypothesis that tendons, a neural crest derivative, are defective in cyp26b1 mutants. In wild-type embryos, differentiated Tsp4b-positive tendon tissue was greatly enriched at all cranial muscle attachment points at 4 and 5 dpf (Fig 5A and 5B). Tsp4b was enriched in cyp26b1 mutant embryos at most of the jaw muscle attachment points at 4 dpf (Fig 5C), though the mandibulohyoid junction (arrowheads in Fig 5A and 5C) and sternohyoideus-anchoring tendons (arrows in Fig 5A and 5C) were difficult to detect. In 5-day-old mutant larvae, all visible jaw muscle attachment points, including ectopic attachments, displayed distinct Tsp4b deposition (Fig 5D). These observations suggested that loss of Cyp26b1 function disrupts the morphology of those tendons that associate with mispatterned muscles.
Alterations to Tsp4b deposition in 3-day-old cyp26b1 mutant embryos led us to question whether tenoblasts are disturbed during the period of jaw muscle morphogenesis. Using a fli1:EGFP;scxa:mCherry double transgenic line, we observed that scxa-positive neural crest cells were condensing along the forming mandibulohyoid junction at 60 hpf (Fig 6A, arrow). Four other scxa-positive condensations resided more posteriorly in the medial second pharyngeal arch, two slightly dorsal sternohyoideus tendons which attach to the ceratohyal cartilage at a region of scxa:mCherry;sox9a:EGFP double positive cells (arrows in Fig 6A’, S3 Fig) and two smaller condensations labeling the medial tips of the hyohyal muscles (arrowheads in Fig 6A’). An elongated group of fli1:EGFP expressing cells defined the insertion of the sternohyoideus muscles (S4A–S4D Fig). These brightly fli1:EGFP-positive cells separated the two ceratohyal cartilage condensations, and also separated anterior, ventral tenoblasts from the posterior, dorsal sternohyoideus tendon condensations (arrowheads in S4B and S4C Fig). At 4 dpf, scxa:mCherry expression labeled all cranial tendons and ligaments (Fig 6C). The mandibulohyoid junction (arrow in Fig 6C) and a small population of scxa-positive mesenchyme on the periphery of the intermandibularis posterior muscles displayed the scxa reporter brightly. Expression of the scxa reporter was also strong at the hyohyal junction (Fig 6C’, arrowhead in Fig 6C), and two scxa-positive spurs extended from between the hyohyal muscles (outlines in Fig 6C’) to the anterior edge of each interhyal muscle (outlines in Fig 6E’). Notably, these scxa-positive spurs displayed little to no Tsp4b deposition in wild-type embryos, though Tsp4b was enriched on every other cranial tendon and ligament (compare Fig 5B and Fig 6C). Just dorsal to the hyohyal junction structures were the long, rod-shaped sternohyoideus tendons (Fig 6E). Our findings suggested that tendon components of the mandibulohyoid junction develop from a population of anterior tenoblasts that is segregated, by at least 60 hpf, from those posterior tenoblasts that form the hyohyal junction and the sternohyoideus tendons.
In cyp26b1 mutants, the population of scxa-positive neural crest cells was less organized. In mutants we saw there was less clear separation of anterior and posterior tenoblasts in the second pharyngeal arch (Fig 6B, arrowhead in Fig 6B’). Though visually it appeared that the population of tenoblasts could be expanded in mutants, cell counts indicated that 60 hpf mutants (148.67 tenoblasts, SD = 2.89, n = 3) did not have significantly more or fewer tenoblasts than siblings (159.67 tenoblasts, SD = 16.4, n = 6). Instead, the scxa-positive cells in 60 hpf cyp26b1 mutants were not concentrated at the mandibulohyoid junction, as they were in wild-type embryos (arrow in Fig 6B), though condensations still formed at the sternohyoideus and hyohyal muscle tips. The structure we saw separating the ceratohyal cartilage condensations was also missing in mutants (S4E–S4H Fig). We observed bright, fli1:EGFP expressing cells amidst the tenoblasts, but they formed a round shape ventral to the ceratohyals instead of elongating between them (green arrowhead in S4F Fig). Four-day-old cyp26b1 mutants did not maintain strong scxa expression at the mandibulohyoid and hyohyal junctions (arrow and arrowhead in Fig 6D, respectively). In fact, mutants lacked defined spurs at the hyohyal junction. Instead, a poorly organized population of scxa-positive cells stretched between the tips of the hyohyal muscles and the adjacent interhyal muscles, as well as ectopic muscle projections from these (detail in Fig 6D’). Four day old mutants had developed sternohyoideus tendons, but these were hypoplastic and slightly curved (arrows in Fig 6F’).
We used in situ hybridization to visualize expression of an early muscle-proximal tendon marker, xirp2a, which labels all muscle attachments [2] to further characterize tendons in cyp26b1 mutants. At both 60 hpf and 4 dpf, xirp2a labeled cells at the tips of cranial muscle fibers (Fig 6G and 6I). This remained true in cyp26b1 mutant embryos, in which xirp2a labeled muscle fiber tips even at ectopic points of attachment (Fig 6H and 6J). Notably, in 60 hpf mutants there was no expression of xirp2a in the midline where the mandibulohyoid junction should be forming. Together, these findings suggested that the cyp26b1 mutant muscle phenotype is coincident with a defective pattern of scxa-positive tenoblasts, but mutants showed tendon matrix deposition over time because muscles and xirp2a-positive cells still promote tendon differentiation.
Based on the disruption of mature tendon morphology in cyp26b1 mutants, we hypothesized that loss of Cyp26b1 function disrupts tenoblast behaviors. Live imaging in Tg(-0.5unc45b:EGFP);scxa:mCherry double transgenic control embryos revealed dynamic populations of tenoblasts in the ventral midline of the jaw (S2 Movie, Fig 7A–7F). In control embryos, tenoblasts surrounded muscle fibers of the intermandibularis muscles as they formed near the midline of the first pharyngeal arch around 51 hpf (arrows in Fig 7A’, see S5A Fig for an enlarged image). Tenoblasts appear to migrate toward the midline as the intermandibularis posterior muscles elongated (Fig 7B), populating the midline prior to the muscles masses (S5B Fig, arrowhead). Coincidently, the intermandibularis posterior muscles angled toward the midline by 54 hpf, almost perpendicular to the intermandibularis muscle masses at 51 hpf (compare S6A and S6B Fig). By this time, two other tenoblast masses at the tips of the hyohyal and sternohyoideus muscles became visible (arrowheads in Fig 7B’). By 57 hpf, a bright condensation of tenoblasts localized at the anterior tip of the second pharyngeal arch (outline in Fig 7C’). Over the next few hours, this rudimentary mandibulohyoid junction took shape as tenoblasts continued to condense in the midline, and the intermandibularis posterior and interhyal muscles elongated toward the condensation. Around 60 hpf, two more bright condensations formed from tenoblasts extending from the tips of the sternohyoideus muscles (arrows in Fig 7D’). One final condensation, connecting the hyohyal muscles, formed after tenoblasts extended from each muscle tip to make contact in the midline around 63 hpf (arrowhead in Fig 7E’). By 70 hpf, the sternohyoideus tendons and tendon elements of the mandibulohyoid and hyohyal junctions appeared bright and compact (Fig 7F). Tenoblast condensation thus precedes the morphogenesis of the mandibulohyoid and hyohyal junctions.
In cyp26b1 morpholino-injected embryos, clear tenoblast defects presaged the ectopic elongation of intermandibularis posterior muscles (S2 Movie, Fig 7G–7L). Loss of Cyp26b1 function did not appear to interfere with tenoblast specification, as tenoblasts surrounded the newly forming intermandibularis muscle fibers at 51 hpf (arrows in Fig 7G’). However, these tenoblasts failed to condense at the midline (Fig 7H–7L). At 54 hpf, the intermandibularis posterior muscles were not oriented toward the midline but rather toward the posterior or even slightly laterally (arrows in S6D Fig). Over time, muscle angle did turn toward the midline, but no tenoblast condensation was apparent at the mandibulohyoid junction at 57 hpf (Fig 7I). Instead, a contiguous, diffuse mass of tenoblasts stretched from the oral ectoderm to the sternohyoideus muscles (outline in Fig 7I’). Condensation of the sternohyoideus tendons initiated normally around 60 hpf (arrows in Fig 7J’), and the intermandibularis posterior muscles extended beyond the prospective mandibulohyoid junction to abut these condensations. By 63 hpf, the sternohyoideus tendon condensations were the only tenoblasts touching the tips of the intermandibularis posterior muscles (arrows in Fig 7K’). Loss of Cyp26b1 inhibited tenoblast condensation between the hyohyal muscles, and blocked their connection even out to 3 dpf (arrowhead in Fig 7L’). No tendon condensation was visible between the tips of the intermandibularis posterior and interhyal muscles, but mesenchymal tenoblasts sat in the midline between the intermandibularis posterior muscles (arrow in Fig 7L’). These results indicated that Cyp26b1 function is required for the precise morphogenesis of several cartilage, tendon, and muscle elements that all occupy the ventral midline of the second pharyngeal arch.
Because jaw muscle morphogenesis occurs over several hours, we sought to ascertain when Cyp26b1 function is required for this process. We treated embryos with media containing talarozole, which inhibits Cyp26b1 and its orthologs Cyp26a1 and Cyp26c1, then assessed tendon and muscle morphology at 4 dpf. Compared to controls (Fig 8A), embryos treated with talarozole during the period of tenoblast migration to and condensation at the mandibulohyoid junction (54–60 hpf) had severe phenotypes (Fig 8B). Of 64 individuals, 100% displayed ectopic muscle projections across the second pharyngeal arch, including 58.3% in which the intermandibularis posterior muscles extended past the location of the mandibulohyoid junction. The sternohyoideus muscle was examined in a subset of these, and in 51% of embryos there were fibers extending across the midline in these muscles (n = 27/53). These defects were typically more severe than those observed in cyp26b1 mutants, but were still restricted to the ventral midline (the dorsal adductor mandibulae muscle was normal in 54/55 embryos examined, with a single stray muscle fiber in 1 fish). Overall our results suggest partial redundancy across Cyp26 enzymes in patterning the ventral musculoskeletal attachments.
Treatment during the period of sternohyoideus tendon condensation and hyohyal junction formation (60–72 hpf) affected muscle morphology weakly (Fig 8C). Of 26 individuals, 76.9% displayed mild ectopic muscle projections from the interhyal or hyohyal muscles, but 25 of 26 developed a normal mandibulohyoid junction. These data show that for proper musculoskeletal patterning, Cyp26b1 function is required when mandibulohyoid tenoblasts are migrating and condensing.
We also evaluated the effects of Cyp26 inhibition on the cranial skeleton at 4 dpf. Control larvae had normal phenotypes (Fig 8D). Talarozole treatment from 54–60 hpf recapitulated much of the jaw skeletal phenotype of cyp26b1 mutants. These embryos consistently displayed fused Meckel’s (Fig 8E, red arrowhead and inset) and ceratohyal cartilages (Fig 8E, black arrowhead and inset), and loss or severe hypoplasia of the basihyal cartilage (Fig 8E, arrow). Embryos treated from 60–72 hpf also had midline cartilage fusions and basihyal cartilage hypoplasia (Fig 8F). These findings suggested that Cyp26b1 continues to function in cartilage development after the mandibulohyoid junction forms, and that the mandibulohyoid junction can form normally despite deformity of the basihyal and ceratohyal cartilages.
Cyp26b1 function is required in the neural crest during the process of tenoblast migration and condensation. We turned to in situ hybridization to determine if there were subpopulations of neural crest cells expressing cyp26b1 at these times. Intriguingly, at 54 hpf, a band of cyp26b1-positive cells filled the midline of the posterior second arch between the posterior edge of the mandibulohyoid tenoblast mass (Fig 9B, arrow) and the sternohyoid tenoblast masses (Fig 9B, arrowheads). There were few, if any, cyp26b1-expressing tenoblasts. The mandibulohyoid tenoblasts appeared entirely cyp26b1-negative. At 60 hpf, the gap between condensing mandibulohyoid tenoblasts (Fig 9F, arrow) and sternohyoid tenoblasts (Fig 9F, arrowheads) had grown. Strong cyp26b1 expression labeled bilateral cell masses in the posterior second arch (Fig 9G, arrows). Again, there appeared to be little overlap between the scxa transgenic and cyp26b1 expression. This pattern of expression, with cyp26b1 adjacent to scxa:mCherry expressing cells, is conserved at the tips of other muscles which are not disrupted in cyp26b1 mutants (S7 Fig). Thus, cyp26b1 may be a general marker of mesenchyme adjacent to developing tendon, but does not appear to be required at all of these sites. Together, our results indicated that Cyp26b1 functions in a mesenchymal population of neural crest cells to promote separation and condensation of tendons in the ventral midline of the second pharyngeal arch.
This expression pattern suggests that cyp26b1 is required in a neural crest population outside of tenoblasts for their condensation. If this were the case then wild-type tenoblasts should fail to condense appropriately in an otherwise mutant environment. We performed transplants using small numbers of cells (10–20) from fli1:EGFP;scxa:mCherry hosts into embryos generated from matings of cyp26b1 carriers. In rare instances (n = 3), we found that the ventral midline of the second arch was only populated by fli1:EGFP;scxa:mCherry double-positive tenoblasts (S8 Fig, insets, asterisks mark the oral opening). When these cells were transplanted into wild-type embryos (n = 2), the tenoblasts associated tightly with the muscles and elongated at the mandibulohyoid junction (S8A Fig, arrowhead). In the instance when these wild-type cells were transplanted into a mutant host, most tenoblasts were localized in a region medial to the intermandibularis posterior muscles (S8B Fig, arrow). This region precisely matches the region where tenoblasts are ectopically localized in cyp26b1 mutants (See Fig 7). Additionally, tenoblasts at or near where the mandibulohyoid junction would form are not elongated (S8B Fig, arrowhead). Collectively, our findings support the model in which cyp26b1 is required neural crest autonomously but tenoblast non-autonomously for tendon condensation. Tools to specifically target this previously undescribed cyp26b1-expressing mesenchyme do not currently exist and will be essential to directly determine the autonomy of the mutation.
Here we provide in vivo characterization of developing musculoskeletal attachments in a living embryo and demonstrate the involvement of retinoic acid in this process. We found that Cyp26b1 function is essential for ventral cranial muscle patterning, but not specification. Genetic mosaic analysis demonstrated that neural crest cells required Cyp26b1 function for proper muscle patterning. Time-lapse confocal imaging showed that neural crest derived tenoblasts appear to migrate to and condense in regions that predict muscle attachments. In cyp26b1 mutants, tenoblasts are generated, yet fail to condense appropriately at those muscle attachments that are disrupted. Furthermore, crucial events of tendon and muscle morphogenesis appear to be orchestrated by non-tenoblast neural crest cells expressing cyp26b1.
Our work is consistent with that of others showing that neural crest cells have critical influences on muscle morphology in the face. Transplanted neural crest cells confer their axial [4] and species-specific [5] identities upon cranial muscles. Our findings provide strong evidence for region specific patterning of tendons, with the ventral midline, particularly at the junction of the 1st and 2nd arch, requiring Cyp26 function. What other region specific cues may be present that pattern craniofacial tendon development are currently unknown. However, loss of Edn1 signaling has been shown to dramatically disrupt ventral musculature and skeletal elements [16]. While tendons were not analyzed in this study, it is highly likely that neural-crest-derived tenoblasts use the same patterning information. Edn1 and Bmp act in opposition to Jagged-Notch signaling to pattern the ventral to dorsal axis within the skeletogenic neural crest [17]. It will be of great interest to determine the tendon phenotypes in mutants within these pathways.
In mouse and zebrafish, and presumably all vertebrates, neural crest gives rise to Scx-positive cranial tendons and ligaments [2, 3], unlike other regions of the body where tendon, bone and muscle all derive from mesoderm populations [18, 19]. Our work with that of others demonstrates that, regardless of the source of the progenitors, tendons are critical for muscular patterning. In the chick, ablation of tendon primordia results in mispatterning of the limb musculature [6]. We find in our cyp26b1 mutants similar overextension of specific muscles associated with highly disrupted tendons. Our work demonstrates that the presence of tenoblasts is not sufficient for muscular patterning and that condensation appears to be a critical step in musculoskeletal integration.
In no system are the cell dynamics underlying musculoskeletal integration understood. Our time-lapse imaging suggests that tendon progenitors migrate to muscle attachment points closely followed by the muscle fibers themselves. Because musculoskeletal patterning in the head is neural crest dependent, this behavior suggests that muscle patterning by neural crest [4, 5] could be due to short range cues from tenoblasts. Tendon progenitors form in cyp26b1 mutants but fail to condense appropriately, particularly at the mandibulohyoid junction. Muscle fibers, still in close contact with tenoblasts, fail to elongate toward the midline appropriately without tenoblast condensation. These results could be explained by one of two models. In the first model, the tenoblasts themselves regulate the elongation of muscles, with tenoblast condensation depending upon cyp26b1-expressing cells. In the second model, muscle elongation is independent of tenoblasts, with tenoblast and muscle cell behaviors both being directly or indirectly influenced by cyp26b1-expressing cells. Data from the chick limb would favor the first model, but experiments in which tenoblasts are specifically deleted will be required to answer this question in the head. Given the dearth of knowledge of this process, the cyp26b1 mutant is likely to provide key insights into musculoskeletal integration.
Our results imply a specific role for Cyp26b1 in mandibulohyoid junction formation that is not autonomous to tenoblasts or muscles. Expression of retinoic acid pathway components is highly dynamic, and for much of early pharyngeal arch development cyp26b1 expression is broad [13]. However, at 54–60 hpf, when Cyp26b1 function is required for tenoblast condensation, cyp26b1 expression in the second pharyngeal arch is restricted predominantly to non-tenogenic midline neural crest cells. This differs greatly from RA dynamics reported in chick limbs, where a Cyp26 enzyme is expressed in tendons themselves at the interface with muscle and RA promotes apoptosis of ectopic muscle fibers between adjacent tendons [20]. It remains to be determined if this is a general difference between the role of RA signaling in musculoskeletal development of the head versus the limb.
Our work strongly suggests that a novel population of neural crest modulates musculoskeletal integration in the second pharyngeal arch. We have found that neural crest cells expressing cyp26b1 occupy a region that excludes scxa-positive tenoblasts. Cyp26b1 function in this region of exclusion likely promotes tendon condensation anteriorly and posteriorly. In live imaging of cyp26b1-depleted embryos, tenoblasts are spread widely around the mandibulohyoid junction and across this exclusion region. The intermandibularis posterior and interhyal muscles elongate and connect within this field, but neither these muscles nor their connection drives the tenoblasts to condense. Instead, mandibulohyoid tenoblasts remain a loose mesenchymal population, and the tips of the intermandibularis posterior and interhyal muscles migrate toward the sternohyoideus tendon condensations. Furthermore, the sternohyoideus tendons fail to condense at the same rate as in controls despite contact with supernumerary muscle tips. Thus, cyp26b1 mutants have cranial muscles that maintain tenoblast populations, but these muscles are not sufficient for tendon differentiation in the absence of the cyp26b1-expressing cell population we describe. We propose that signals from these neural crest cells are critical for promoting specific sites of tendon condensations, which then organize the musculature.
Zebrafish stocks were maintained and embryos were raised according to established protocols [31] with approval from the University of Texas at Austin Institutional Animal Care and Use Committee. The approved protocol includes authorization for embryonic zebrafish euthanasia by overdose with MS-222/Tricaine.
The cyp26b1b1024 allele was recovered from a forward genetic screen at the University of Oregon. The Tg(fli1:EGFP)y1 [32], Tg(-0.5unc45b:EGFP) [33] and Tg(foxP2-enhancerA:EGFP) [34] transgenic lines are referred to as fli1:EGFP, 503unc:EGFP and sox9a:EGFP, respectively, throughout the text. The 503unc promoter fragment [33] and Tol2Kit materials and protocols [35] were used to construct the Tg(-0.5unc45b:mCherry) transgenic line, referred to as 503unc:mCherry. The Tg(scxa:mCherry) line is referred to as scxa:mCherry throughout the text. For pharmacological treatments, embryos were bathed in embryo medium with 0.01% DMSO (Thermo Fisher Scientific, Waltham, MA, USA) as vehicle control or 1 μM talarozole (HY-14531, MedChem Express, Monmouth Junction, NJ, USA).
Probes for cyp26b1 [13] and xirp2a (cb1045, [2]) are described elsewhere. Color development for fluorescence imaging was performed with α-Digoxigenin-POD Fab fragments (11207733910, Roche Diagnostics, Indianapolis, IN, USA) and TSA Plus Fluorescein (NEL741001KT) or Cy3 (NEL744001KT, Perkin-Elmer, Inc., Waltham, MA, USA) System. In preparation for immunohistochemistry, embryos were fixed in 95% methanol/5% glacial acetic acid. The protocol for myosin heavy chain/Tsp4b staining is described previously [11]. Primary antibodies utilized include MF 20 (1:100 dilution; Developmental Studies Hybridoma Bank, Iowa City, IA, USA), α-Thbs4b (1:200 dilution; GTX125869, GeneTex, Inc., Irvine, CA, USA), α-GFP (1:200 dilution; sc-9996, Santa Cruz Biotechnology, Inc., Dallas, TX, USA). Alexa Fluor-conjugated secondary antibodies from Thermo Fisher Scientific (Waltham, MA, USA) were used at a 1:1000 dilution.
Confocal z-stacks were collected using a Zeiss LSM 710 and ZEN software. In situ hybridization images were collected using a Zeiss Axio Imager.A1 equipped with an AxioCam HRc, which was operated through AxioVision release 4.9.1 SP1. Images were processed and measured in Fiji [36, 37]. Figures were assembled in the GNU Image Manipulation Program (GIMP).
Fixed 60 hpf embryos from an incross of scxa:mCherry;cyp26b1b1024 were stained with TO-PRO-3 Iodide (642/661) (T3605, Life Technologies, Carlsbad, CA, USA) and imaged as described. Confocal z-stacks were processed and analyzed using Imaris v.8.4.0. Briefly, the Spots module was used to detect nuclei in the TO-PRO channel with a diameter of 3 μm and an automated quality filter and a minimum intensity filter. The Surface module was used to generate a volume rendering of tenoblasts, and a region was selected containing the tenoblasts surrounding the intermandibularis muscles, extending from the oral ectoderm to the point best separating anterior tenoblasts from sternohyoideus tendon condensations. Then, the Imaris XTension “Spots Split to Surface Objects” automatically counted the nuclei located inside the anterior tenoblast Surface object.
Transplantation experiments targeting cranial neural crest were performed as described elsewhere [38, 39].
A previously described morpholino, Cyp26b1-SDEx3 MO [13], was used. Approximately 3 nl of morpholino (working concentration 0.9 mM) were injected into zebrafish embryos between the one-cell and four-cell stages. This concentration of morpholino fully recapitulated the musculoskeletal defects in b1024 (see Figs 3 and 6).
|
10.1371/journal.pcbi.1006025 | Gap junction plasticity as a mechanism to regulate network-wide oscillations | Cortical oscillations are thought to be involved in many cognitive functions and processes. Several mechanisms have been proposed to regulate oscillations. One prominent but understudied mechanism is gap junction coupling. Gap junctions are ubiquitous in cortex between GABAergic interneurons. Moreover, recent experiments indicate their strength can be modified in an activity-dependent manner, similar to chemical synapses. We hypothesized that activity-dependent gap junction plasticity acts as a mechanism to regulate oscillations in the cortex. We developed a computational model of gap junction plasticity in a recurrent cortical network based on recent experimental findings. We showed that gap junction plasticity can serve as a homeostatic mechanism for oscillations by maintaining a tight balance between two network states: asynchronous irregular activity and synchronized oscillations. This homeostatic mechanism allows for robust communication between neuronal assemblies through two different mechanisms: transient oscillations and frequency modulation. This implies a direct functional role for gap junction plasticity in information transmission in cortex.
| Oscillations of neural activity emerge when many neurons repeatedly activate together and are observed in many brain regions, particularly during sleep and attention. Their functional role is still debated, but could be associated with normal cognitive processes such as memory formation or with pathologies such as schizophrenia and autism. Powerful oscillations are also a hallmark of epileptic seizures. Therefore, we wondered what mechanism could regulate oscillations. A type of neuronal coupling, called gap junctions, has been shown to promote synchronization between inhibitory neurons. Computational models show that when gap junctions are strong, neurons synchronize together. Moreover recent investigations show that the gap junction coupling strength is not static but plastic and dependent on the firing properties of the neurons. Thus, we developed a model of gap junction plasticity in a network of inhibitory and excitatory neurons. We show that gap junction plasticity can maintain the right amount of oscillations to prevent pathologies from emerging. Finally, we show that gap junction plasticity serves an additional functional role and allows for efficient and robust information transfer.
| Oscillatory patterns of neuronal activity are reported in many brains regions with frequencies ranging from less than one Hertz to hundreds of Hertz. These oscillations are often associated with cognitive phenomena such as sleep or attention. Local field potential measurements in the neocortex and thalamus show the prevalence of delta oscillations (0.5-4Hz) and spindle oscillations (7-15Hz) during sleep [1]. Theta oscillations (4-10Hz) are also reported in hippocampus and other brain regions [2]. Gamma oscillations (30-100Hz) observed in the cortex are thought to be involved in attention [3–6], perception [7, 8] and coordinated motor output [9, 10]. Thus, at the minimum, oscillations are present during the normal functioning of neural circuits.
However, oscillations are also associated with pathological circuit dynamics, such as hyper-synchronous activity during epileptic seizures [11]. Altered gamma-frequency synchronizations may also be involved in cognitive abnormalities such as autism [12] or schizophrenia [13]. Thus, given both the functional and pathological effects of oscillations, a homeostatic mechanism is necessary to regulate oscillatory behavior.
Several mechanisms can lead to the emergence of oscillations. They can arise in homogeneous population of excitatory neurons, where the positive feedback loop of excitation is only limited by the refractoriness of the neurons [14]. Alternatively, oscillations can also arise in a coupled network of excitatory and inhibitory neurons, where the excitatory and inhibitory neurons burst in opposing phase. [15–19]. Finally, gap junctions between inhibitory neurons promote synchronous oscillatory patterns [20–24].
The inhibitory network oscillations primarily involve fast-spiking interneurons. These neurons represent a large proportion of GABAergic interneurons [25]. They are the main cells targeted by thalamocortical synapses transmitting sensory information to the cortex [26]. They are coupled via chemical synapses and gap junctions. Gap junctions are mostly found between neurons of the same class [26–28] but they can also connect different subtypes, such as fast-spiking and regular spiking cells [26, 29, 30]. Moreover, there is evidence of the critical role of fast-spiking parvalbulmin (FS) interneurons in the emergence of cortical gamma activity in the cortex of rodents in response to sensory stimuli [31–34].
Two main properties of FS interneurons have been found critical in the existence of gamma oscillations. Firstly, FS interneurons selectively amplify gamma frequencies through subthreshold resonance [33]. Secondly, gap junctions between inhibitory interneurons [27] have been shown to enhance synchrony [24, 26, 35–41].
A computational model with both properties, inhibitory neurons with subthreshold resonance, connected by gap junctions, has been shown to support gamma oscillations [24, 42–46].
Recently, gap junction plasticity has been experimentally demonstrated [47–51]. For example, the gap junctions between rod cells in the retina can vary their conductance during day and night cycles [52]. Moreover, they can experience bidirectional long-term plasticity in an activity-dependent manner [49, 53, 54]. High frequency stimulation of a coupled pair of thalamic reticular nucleus (TRN) neurons induces gap junction long-term depression (gLTD) [55]. This occurs only when the TRN neurons burst. There is no data yet on the long-term potentiation of cortical gap junctions. However, [56] show that the pathways leading to gLTD are calcium-dependent which suggest that gap junction long-term potentiation (gLTP) could also be the result of an activity-dependent mechanism. Other passive mechanisms, such as gap junction connexin turnover could compensate for long-term depression as well [57–62].
Given the existence of gap junction plasticity and the omnipresence of oscillations in cortex, we wondered whether gap junction plasticity can regulate network-wide gamma oscillations in cortex. To that end, we developed a computational model of a network of excitatory and FS inhibitory neurons. As demonstrated analytically by [24], we observed two different network behaviors depending on the gap junction strength. For weak gap junction strength, the network exhibits an asynchronous regime, whereas for strong gap junctions, the network synchronizes into coherent gamma oscillations with bursting activity. We then modelled the gap junction plasticity observed by [55] showing that bursting activity leads to gLTD. The plastic network sets itself at the transition between the asynchronous regime, where sparse spiking dominates, and the synchronous regime, where network oscillations dominate and burst firing prevails. Thus, our model shows that gap junction plasticity maintains the balance between the asynchronous and synchronous network states. This is robust to different possible gLTP rules. We then show that the network allows for transient oscillations driven by external drive. This demonstrates that transient, plasticity regulated oscillations can efficiently transfer information to downstream networks. Finally we show that gap junction plasticity mediates cross-network synchronization and allows for robust information transfer trough frequency modulation. Critically, gap junction plasticity allows for the recovery of oscillation mediated information transfer in the event of partial gap junction loss.
To study the effect of gap junction plasticity, we developed a network of coupled inhibitory and excitatory neurons in the fluctuation-driven state (Fig 1A). The Izhikevich model was used for the inhibitory neuron population to fit the fast-spiking inhibitory neuron firing pattern [63]. Excitatory neurons are modelled by leaky integrate-and-fire models. As in [24], the excitatory neurons act as low pass-filters for their inputs while the FS neurons have a sub-threshold resonance in the gamma range [42–46]. To demonstrate this, we injected an oscillatory current of small amplitude in a single cell and recorded the amplitude response for different oscillatory frequencies. Excitatory neurons better respond to low frequency inputs, while FS neurons respond maximally for gamma inputs (Fig 1B). This is in line with the experimental evidence of Cardin et al. showing that FS-specific light stimulation amplifies gamma-frequencies [33].
All neurons have chemical synapses but only inhibitory neurons are also coupled via gap junctions (Fig 1A). The gap junctions are modelled such that a voltage hyperpolarization (depolarization) in one neuron induces a voltage hyperpolarization (depolarization) in the connected neuron. The current contribution of gap junction coupling is proportional to the difference of voltages between the coupled neurons, multiplied by the gap junction strength γ (Fig 1C). Moreover, when one neuron spikes, it emits a spikelet in the coupled neuron. We model this by a positive inhibitory to inhibitory electrical coupling, which we add on top of the negative inhibitory to inhibitory chemical coupling (see Materials and methods).
In order to understand the effects of gap junction plasticity, we initially considered the network without plasticity. We first explored the network behavior for different values of the mean gap junction strengths γ and mean external drive to the inhibitory neurons νI. As demonstrated by [24], our network exhibits two regimes (Fig 1D): an asynchronous irregular (AI) regime and a synchronous regular regime (SR). The AI regime occurs for networks with weak external drive and weak gap junctions. In this regime the network is in the fluctuation driven regime so that the neurons spike due to variations in their input. The SR regime occurs for strong external drive and strong gap junctions. This regime leads to the emergence of gamma oscillations. Mathematically, the network undergoes a Hopf bifurcation [24, 39]. The oscillations arise as the network directly inherits the resonance properties of the individual neurons. This is mediated through the gap junction coupling which effectively allows positive coupling through their spikelets. Moreover, the gap junctions reduce sub-threshold voltage differences between neurons which promotes synchrony. The excitatory neurons are not necessary for the oscillations but they amplify the dynamics (see [24] for mathematical derivations). When placed in the SR regime, the network oscillates in the gamma-range at a frequency near the single neuron resonance frequency (Fig 1E and 1F). In addition, we observe that the spiking activity is characteristic to the network regime, with bursting activity in the synchronous regime and spikes in the asynchronous regime (Fig 1G–1I).
To summarize, increased gap junction coupling and input drive into the network promotes gamma oscillations. To explain the relationship between network activity and gap junction plasticity, we first model the simplest case of plasticity between a pair of electrically coupled neurons. We then apply the plasticity rule to a population of neurons and investigate the effects on the network dynamics.
To determine how gap junction plasticity can alter network dynamics, we developed a model of the plasticity based on experimental observations. [55] have shown that bursts in one or both neurons in an electrically coupled pair lead to long-term depression (gLTD). Therefore, we modeled gLTD as a decrease in the gap junction strength that is proportional to the amount of bursting. The constant of proportionality, αgLTD serves as the learning rate. To infer αgLTD, we reproduced the bursting protocol in Haas et al., where a neuron bursting for a few milliseconds, 600 times for 5 minutes, leads to 13% decrease (Fig 2A).
Activity-dependent gap junction long-term potentiation (gLTP) has not been reported experimentally yet in the mammalian brain. There is evidence for activity dependent short-term potentiation in vertebrates [53, 64]. However, without potentiation, all gap junctions would likely become zero with time. To address this concern, we hypothesize that gap junctions can undergo gLTP and we modeled it such that single spikes induce gLTP by a constant amount given by the potentiation learning rate αgLTP (Fig 2B, first half). Furthermore, we considered activity-independent gLTP rules in the supplementary materials (S1 Fig).
Our plasticity model therefore potentiates gap junctions under spiking activity and depresses under bursting activity. Therefore, we wondered how gap junction plasticity can alter network dynamics. We previously quantified the amount of spiking versus bursting in our network for different levels of fixed gap junction strength and mean drive. For low levels of both, the network is spiking whereas for high levels of both the network is bursting. The spiking to bursting transition (Fig 1G) corresponds to the bifurcation (Fig 1D) from asynchronous irregular to synchronous oscillations at gamma frequency. When inhibitory neurons are oscillating, they fire a burst of spikes at the peak of the oscillations (Fig 1I, γ = 5). Therefore, when gap junctions are plastic, the network steady state can be found on the side of the bifurcation that balances the amount of potentiation due to spiking activity with the amount of depression due to bursting activity. The depression learning rate is inferred from Haas et al., while the potentiation learning rate is left as a free parameter.
We found that a strong relationship exists between gap junction plasticity and network synchrony. When the network is in the AI regime, characterized by low prevalence of bursting activity, gap junction potentiation dominates. However, for a strong mean coupling strength, the emergence of oscillations is associated by high bursting activity which leads to depression of the gap junctions. Therefore gap junction plasticity in our network maintains a tight balance between asynchronous and synchronous activity. Depending on the value of αgLTP, the position of the plasticity fixed point lies either in the asynchronous regime (low αgLTP, Fig 2C) or in the synchronous regime (high αgLTP). For high values of αgLTP, potentiation is fast while for low values, the potentiation is slow.
We wondered how gap junction plasticity would interact with time-varying inputs. For the following experiment we consider slow gLTP. First, we let the network reach its steady state with a low level of drive (Fig 2E, beginning). As previously observed, the mean gap junction strength reaches a value which sets the network near the AI/SR transition. Then, we proceeded by injecting an additional constant current to the network. This new current baseline induces network level oscillations (Fig 2E, transition). However, over time the mean gap junction strength decays due to the gap junction plasticity mechanism. This gap junction depression is followed by a loss of synchrony and the network reaches its new steady state (Fig 2E, end), again near the border of asynchronous and synchronous regimes.
We measured the response of read-out neurons which receive projections from the excitatory and inhibitory neurons in our network (Fig 2D). At the onset of the current step, the network undergoes transient oscillations. When the gap junctions are plastic, the downstream neurons increase their spiking activity only for a few hundred milliseconds during the transient oscillations and then became almost quiescent again (Fig 2F, second panel). This contrasts with the simulation of a static network where the downstream keep a high firing rate (Fig 2F, third panel).
These results suggest that synchronous activity is a powerful signal to provoke spiking in downstream neurons. But oscillations and high firing rates of downstream neurons are also metabolically costly [65]. With transient oscillations however, the downstream neurons only sparsely fire when the stimulus changes but not when it is predictable. Thus, the regulation of oscillations mediated by gap junction plasticity allows for sparse but salient information transfer.
We now sought to study the functional implications of fast gLTP. As stated before, this synchronizes the network into gamma oscillations. Synchronization between networks is considered to be one possible mechanism of information transfer [66–69]. We wondered whether gap junction coupling could mediate cross-network synchronization, and how gap junction plasticity would regulate this synchronization. To test this hypothesis, we considered two subnetworks having different oscillation frequencies and coupled by gap junctions (Fig 3A). A fast network oscillates at a gamma frequency and therefore is called the gamma-network. Then, a slow-network oscillates at a slower frequency as the membrane time constant of its inhibitory neurons is chosen to have a larger value. Indeed, previous analyses show that the network frequency in our model is inherited from the single neuron resonance frequency of inhibitory neurons [24, 70]. As a result, increasing the membrane time constant of the inhibitory neurons results in a decrease of the network oscillation frequency (Fig 3B–3D). Cross-network gap junctions reduce the frequency and phase difference between the gamma- and slow-network (Fig 3E and 3F) and larger differences of subnetwork resonant frequencies require a larger number of cross-network gap junctions for the networks to oscillate in harmony (Fig 3E and 3G). Their common frequency lies between the resonant frequencies of the decoupled networks. Importantly, cross-network synchronization requires the subnetworks to be in phase. If the gamma- and slow-network do not share enough gap junctions, there is little mutual information and no correlation in their population activities (Fig 3H and 3I), despite having a common oscillation frequency in some cases ([Δfres = 0; number of shared GJs = 0] on Fig 3I). However, for small differences in the subnetworks resonant frequency Δfres, increasing the number of shared gap junctions induces the oscillations to lock together. The networks oscillate in phase (Fig 3F, end of first row) as reflected in their mutual information (Fig 3H, dark blue area) and their correlation (Fig 3I, dark red area). In summary, two networks in the SR regime with different resonance frequencies and/or out-of-phase can synchronize if they are coupled by gap junctions. Furthermore, a large number of shared gap junctions is required for large differences of resonant frequency.
As gap junctions can synchronize two oscillating populations of neurons, we wondered whether the same synchronization would occur with one population in the AI regime. First, we initialized the gamma-network in the AI regime while the slow-network was initialized in the SR regime (Fig 4A). After coupling the gamma- and slow-network together, we found that, while the oscillation frequency of the gamma- and slow-network matched (Fig 4B), the two networks could not synchronize. The networks were always out-of-phase with very weak correlation between the population activities (Fig 4C and 4D). The results were similar if the gamma- and the slow-network were initialized in the reverse synchronous and asynchronous parameter regimes, respectively (not shown). Cross-network synchronization is not robust when one network is not oscillatory.
Given these constraints on cross-network synchronization, we wondered if gap junction plasticity could remedy the situation and allow for robust cross-network synchronization. To test this hypothesis, we repeated the simulation protocols with the gamma- and slow-network initialized in the asynchronous and synchronous regimes (respectively) and with plastic gap junctions. Here we considered the case where the gLTP rates were slow. As shown previously, gap junction plasticity regulates oscillations such that the network in the asynchronous irregular regime transitions to the oscillatory regime (Fig 4E). The oscillation frequencies of these two networks match (Fig 4F). Strikingly, even with a large resonant frequency difference, the gamma- and slow-network now synchronize through a small number of shared gap junctions (Fig 4G and 4H). This indicates that gap junction plasticity allows for cross-network synchronization that is robust to the underlying neuronal parameters for small numbers of shared gap junctions.
We hypothesized that cross-network synchronization mediated by plasticity allows information transfer. To investigate this, we considered a similar network architecture as previously studied, with two networks, an input-network and an output-network. The input-network receives an input projected by random weights to its neurons. The output-network is connected to the input-network with a small number of gap junctions and inhibitory chemical synapses.
First, to demonstrate the information transfer capability of the network, we consider static gap junctions with oscillatory inputs to the input-network. The stimulus information is transmitted to the output-network via the frequency modulation of the synchronized oscillations and not by spike transmission nor amplitude modulation (Fig 5A–5D). When sharing gap junctions, the input- and output-network synchronize together (Fig 5A) and their spiking activity is locked (Fig 5B). As the amplitude of the input signal increases, the spiking activity increases in the input-network but not in the output-network (Fig 5C). For a network in the SR, there is a positive correlation between the signal amplitude and the network oscillation frequency (Figs 1E and 5D). This frequency modulation is transferred from the input- to the output-network. Thus, the input amplitude can be estimated from the oscillation frequency of the output-network, despite the absence of chemical synapses between the input-network and the output-network (Fig 5E). However, this synchrony code is only possible for signals below a certain frequency (Fig 5F and 5G). Indeed, the instantaneous oscillation frequency is estimated by measuring the period between consecutive peaks of the population activity. For example, oscillations at 50 Hz have a period of 20 ms. Variations happening within those 20 ms are compressed to a single period value and thus are not transferred via frequency modulation. Mechanisms for estimating the input value from the oscillation frequency of the output-network are discussed further in the methods section. Finally, we tested if this synchrony code was valid for non-oscillatory signals (Fig 5H). We found that non-oscillatory, slowly varying random signals could also be robustly transmitted from the input- to the output-network with gap junction coupling (Fig 5I).
As gap junction plasticity can regulate oscillations, we tested whether the plasticity can make this synchrony code robust to parameter variations or potential gap junction loss. First, as previously shown, gap junction plasticity enhances the ability of networks to synchronize. If initialized in the AI regime and with static gap junctions, there is no information transfer via frequency modulation (Fig 5J, left panel). However, with plasticity and fast gLTP, the oscillations are regulated and the network synchrony is recovered which results in successful information transfer (Fig 5J left panel). A critical amount of oscillation power and a critical number of shared gap junctions are required for information transfer, after which increasing each of them does not yield significant improvement (Fig 5J). Furthermore, we studied whether gap junction plasticity could restore information transfer if gap junctions were deleted. While there is loss in the quality of the transfer when static gap junctions are removed, plastic gap junctions maintain the quality of the transfer by increasing the strength of the remaining gap junctions. This mechanism compensates for the missing gap junctions (Fig 5J and 5K).
To summarize, gap junction plasticity expands the necessary conditions for information transfer. It regulates oscillations, and by promoting phase-locking of oscillations, it contributes to the propagation of information to downstream networks. Finally, if some gap junctions are failing, due to protein turnover perhaps, the remaining ones can increase their strength through plasticity. This helps to maintain accurate information transfer.
Our modelling study tested whether gap junction plasticity can regulate gamma oscillations in cortical network models. Our findings suggest that gap junction plasticity can maintain a balance between synchronous regular and asynchronous irregular regimes. For strong electrical coupling, the network is in the oscillatory regime. The oscillations consist of synchronized bursting mediated by the inhibitory neuron network. These bursts trigger depression of the gap junctions [55] allowing the network to leave the oscillatory regime and spike asynchronously. However, the irregular asynchronous regime is dominated by sparse firing. Either this sparse firing, or constant protein connexin turnover may be a source of gap junction potentiation [48, 56–61]. Thus, the asynchronous irregular regime tends to potentiate gap junctions. Therefore, the network behavior critically depends on the plasticity learning rate. Fast gLTP leads to synchronous activity while slow gLTP leads to asynchronous states. We demonstrate the functional role of plasticity in both cases. In the AI regime, the network can respond to changes in input drives through transient oscillations. Those transient oscillations could serve as an energetically efficient way to transfer information to a downstream neuron. In the SR regime, the network oscillations can serve as the substrate for information routing between networks. These results demonstrate how gap junction plasticity can regulate oscillations to mediate information transfer between cortical populations of neurons.
Despite being less common than chemical synapses, gap junctions are ubiquitous in the central nervous system. Example includes the inferior olivary nucleus [71–73], the thalamic reticular nucleus [74, 75], the hippocampus [36, 76], the retina [52, 77], the olfactory bulb [78], the locus coeruleus [79], or also the neocortex [80, 81]. Moreover, they drastically alter the firing activity of their connecting neurons [82, 83], as well as the network dynamics [20–24]. Furthermore, gap junctions between inhibitory interneurons are reported in many cortical regions where global oscillations of neural activity are observed [21, 27, 84, 85]. These inhibitory neurons exhibit sub-threshold resonance that amplifies a specific frequency range [33]. Therefore, gap junction induced synchrony and inhibitory neurons frequency preference are a possible substrate for global oscillations in these cortical regions. Our work is consistent with recent results showing that together gap junction strength and sub-threshold resonance of inhibitory neuron promote oscillations of neuronal activity [24, 70].
There has been a recent interest in modelling gap junction plasticity. Snipas et al. [86] developed of model of gap junction coupling that would exhibit short-term plasticity. By combining a 36-state model of gap junction channel gating with Hodgkin-Huxley equations [87], they show that gap junction channel gating, induced by bursting activity, could lead to short term depression. In future work, it would be interesting to combine this model of gap junction short-term plasticity with our model. Chakravartula et al. [88] introduced a new type of adaptive diffusive coupling in a network of Hindmarsh-Rose neurons [89, 90]. They assumed that connections between pairs of neurons would follow a Hebb’s law [91], where neurons with simultaneous activity would strengthen their connection, while others with dissimilar activity would weaken their coupling. They observe the emergence of locally synchronized groups of neurons, whose synchronization could be transient or permanent. Their results are consistent with ours showing synchronization of subnetworks coupled with gap junctions.
Recently, Haas et al. [55] reported the first experimental evidence of activity-dependent gLTD of gap junctions of interneurons in the thalamic reticular nucleus, even though the mechanism remains to be investigated [62]. Also Sevetson et al. [56] found that calcium-regulated mechanisms support gap junction gLTD in the thalamic reticular nucleus. The mechanisms are similar to those observed for the plasticity of chemical synapses. We designed a rule for activity-dependent gLTD consistent with those results. We assumed that a cortical fast-spiking interneuron would exhibit the same plasticity properties as a thalamic reticular neuron because gap junctions are mostly made from the connexin Cx36 throughout the central nervous system [74, 92]. To our knowledge, there is no study yet on activity-dependent gLTP of gap junctions. However recent studies suggest that gLTD and gLTP share a common pathway [48, 56]. Therefore, we propose a rule for activity dependent gLTP, assuming that low frequency spiking activity leads to gap junction potentiation. However, our results do not depend on the exact formulation of gLTP. As we have shown, an activity-independent rule yields similar behavior (supplementary material, S1 Fig). Moreover, we did not observe significant changes by modelling asymmetrical gap junctions (supplementary material, S2 and S3 Figs).
Our model demonstrates that the regulation of oscillations is mediated by gap junction plasticity. Fast potentiation leads to bursting activity while slow potentiation leads to asynchronous irregular activity. Our first hypothesis assumed that the potentiation is slow and the network is in the AI regime. Thus, at the steady-state, gamma power is weak or non-existent. Evidence from Tallon-Baudry et al. and Ray et al. [93, 94] is consistent with our results. When no stimulus is provided or task required, electroencephalogram recordings show that power in the gamma-band is weak. After the onset of a sensory stimulus, gamma oscillations can be detected in cortical areas. This has been reported for example with visual stimuli triggering gamma oscillations in the mouse visual cortex [95]. In our model, the neurons oscillate transiently when receiving a constant external stimulation. This mechanism operates by crossing the bifurcation boundary between the AI and SR regime. However, over time the mean gap junction strength decays due to the additional bursting activity. The gap junction depression leads to a loss of synchrony and the network returns to the AI regime. Therefore we predict a loss in gamma power for sustained stimulus. A similar mechanism may be involved in the reduction of gamma oscillation induced by slow smooth movements [96, 97].
We wondered what could be the functional role of this transient oscillatory regime. Projecting the excitatory activity of our network model to downstream neurons revealed that they fire sparsely, for a short duration after stimulus onset, and are quiescent otherwise. Thus, gap junction plasticity could efficiently encode the change in incoming stimuli. This could allow for energy conservation as oscillations are energetically expensive [65]. Moreover, [98] show that cortical circuits near the onset of oscillations could promote flexible information routing by transient synchrony.
The role of gamma oscillations is highly debated [94]. They could play no role and simply be a marker of the excitation-inhibition interaction. However others studies suggest they could be involved in information transfer. It is thought that retinal oscillations carry information to the visual cortex [99]. Moreover they could serve as inter-area communication by promoting coherence in neural assemblies which would align their windows of excitation. This would allow for effective spike transmission [68, 94, 100]. Furthermore, Roberts et al. [101] observed high gamma coherence between layers 1 and 2 of macaque’s visual cortex by dynamic frequency matching. Here, we demonstrate one potential mechanism for information transmission through gamma oscillations. Our networks make use of gamma frequency modulation to transmit information in a robust manner, similar to the principle used for FM radio broadcasting. The amplitude of the input signal modulates the oscillation frequency, which increases almost linearly with the amplitude. Our model demonstrates that gap junction plasticity robustly mediates network oscillations and cross-network synchronization. If some gap junctions are removed, the remaining gap junctions become stronger and compensate for the missing ones. Thus, gap junction plasticity insures the phase-locking of the coupled network and it allows for information routing. In particular, there is evidence suggesting that gap junctions could promote long-distance signaling by implementing frequency modulation of calcium waves in astrocytes [102]. Moreover, correlation was found during gamma activity between amplitude and frequency modulation of local field potential of CA3 pyramidal neurons of anesthetized rats [103]. In addition, our network models could also represent the subnetworks of the TRN, with each connected to a separate excitatory neuron of thalamus [104]. However, TRN inhibitory neurons exhibit longer bursts than those of cortical fast-spiking neurons, due to long lasting T-current (about 50ms) and further work is necessary to make predictions on this brain region behaviour [105].
Failure to regulate oscillations, could be the origin of several cognitive pathologies. Disruption of brain synchrony in the inferior olive is thought to contribute to autism due to the loss of coherence in brain rhythms [106]. Excess of high frequency network wide oscillations in the cortex have been observed to also correlate with autism in young boys [12]. The inferior olive differs for its density of gap junction being the highest in the adult brain [71, 72]. It may be involved in the generation of tremors in Parkinson’s disease, however the severity of induced tremors in Cx36 knockout mice remained the same as in wild-type mice [107, 108]. This could be due to gap junctions made from other connexins (such as Cx43) taking over for the knocked-out ones.
Recent studies highlight the critical role of gap junctions and their plasticity in efficient cognitive processing [109]. As experimental and computational techniques improve, new efforts can further unveil their properties and expand our understanding of cortical functions. Our computational model shows that gap junction activity-dependent plasticity may play an important role in network-wide synchrony regulation.
We consider a network with NI inhibitory neurons (20%) and NE excitatory neurons (80%) with all-to-all connectivity (Fig 1A). Inhibitory neurons are modelled by an Izhikevich model and excitatory neurons by a leaky integrated-and-fire model (LIF) [63, 110]. The simulation time-step is dt = 0.1 ms. Inhibitory neurons are connected by both electrical and chemical synapses, whereas excitatory neurons have only chemical synapses. We designed a novel plasticity model for activity dependent plasticity of gap junctions and we investigated its impact on network dynamics and function. We then investigated the dynamics of two networks coupled by chemical and electrical synapses. We use a decoder to quantify the effects of gap junction plasticity on information transfer. The model is written in Python and takes advantage of the tensorflow library that leverages GPU parallel processing capabilities [111]. It is available on ModelDB (http://modeldb.yale.edu/230324).
We model Fast Spiking (FS) interneurons with Izhikevich type neuron models [63]. This model offers the advantage to reproduce different firing patterns as well as a low computational cost [112]. The voltage v follows
τ v v ˙ = ( v - v r a ) ( v - v r b ) - k u u + R I , (1) τ u u ˙ = a ( v - v r c ) - u , (2)
combined with the spiking conditions,
if v ≥ v t h r e s h F S , then { v ← v r e s e t F S u ← u + b . (3)
where τv is the membrane time constant, vra is the membrane resting potential, vrb is the membrane threshold potential, ku is the coupling parameter to the adaptation variable u, R is the resistance and I is the current. The adaptation variable u represents a membrane recovery variable, accounting for the activation of K+ ionic currents and inactivation of Na+ ionic currents. It increases by a discrete amount b every time the neuron is spiking and its membrane potential crosses the threshold vthreshFS. It provides a negative feedback to the voltage v. The recovery time constant is τu, a is a coupling parameter, vresetFS, and vrc are voltage constants and b is a current constant.
For the FS neurons, we chose the membrane potential reset vresetFS and the spike-triggered adaptation variable b to account for the onset bursting activity observed in vivo. Modifying ku, vra, vrb and vrc was sufficient to observe the emergence of a resonance frequency. We set the time constant τu to obtain a resonance frequency of 45 Hz, which is in the same range as observed in vivo by [33] (Fig 1B). To measure the sub-threshold resonant property (Figs 1B and 3B and 3D), we recorded the amplitude of the neuronal membrane potential VE in response to different oscillation frequencies f of low level sinusoidal currents I(t) = I0 cos(2πft) (with I0 = 0.01 pA). We then normalized the amplitude response as follow
R E ( f ) = | | V E ( I 0 c o s ( 2 π f t ) ) | | m a x f ( | | V E ( I 0 c o s ( 2 π f t ) ) | | ) , (4)
for frequencies between 0 and 1 kHz. The || || denotes the maximum absolute value observed over time.
To model regular spiking excitatory neurons, we chose a leaky integrate-and-fire model,
τ m v ˙ = - v + R m I , (5)
where τm is the membrane time constant, v the membrane potential, I the current and Rm the resistance. Spikes are characterized by a firing time tf which corresponds to the time when v reaches the threshold vthreshRS. Immediately after a spike, the potential is reset to the reset potential vresetRS.
In the single network model (Figs 1 and 2), each neuron is connected to all others by chemical synapses, but in addition, inhibitory neurons are connected via electrical synapses to all other inhibitory neurons, as in [24]. Thus, the current each individual neuron i receives can be decomposed in four components
I i ( t ) = I i s p i k e ( t ) + I i g a p ( t ) + I i n o i s e ( t ) + I i e x t ( t ) , (6)
where I i s p i k e = I i c h e m + I i e l e c is the current coming from the transmission of a spike via electrical (i.e. spikelet) and chemical synapses, I i g a p is the sub-threshold current from electrical synapses (for inhibitory neurons only), I i n o i s e is the noisy background current and I i e x t characterizes the external current. The current due to spiking I i s p i k e on excitatory neurons is given by
I i s p i k e ( t ) = W I E ∑ j = 1 j ≠ i N I ∑ t j k < t exp ( - t - t j k τ I ) + W E E ∑ j = 1 j ≠ i N E ∑ t j k < t exp ( - t - t j k τ E ) . (7)
The current I i s p i k e into inhibitory neurons are
I i s p i k e ( t ) = ∑ j = 1 j ≠ i N I ∑ t j k < t W i j I I exp ( - t - t j k τ I ) + W E I ∑ j = 1 j ≠ i N E ∑ t j k < t exp ( - t - t j k τ E ) , (8)
where Wαβ is the coupling strength from population α to population β with {α, β} = {E, I}. Finally, W i j I I = W I I , c + W i j I I , e is the inhibitory to inhibitory coupling between neuron i and j, consisting of the chemical synaptic strength WII,c and W i j I I , e the electrical coupling for supra-threshold current, also called the spikelet. There is no experimental data yet on the change of the spikelet as function of the strength of the gap junctions. We hypothesize that the contribution of the spikelet is proportional to the gap junction coupling W i j I I , e = k s p i k e l e t * γ i j, where γij is the gap junction coupling between neurons i and j. This spikelet term is necessary due to the fact that our neuron model does not explicitly have a spike kernel in the voltage dynamics [24]. Note that WEE, WEI, WIE, WII,c are identical among neurons, but W i j I I varies as the spikelet contribution depends on the coupling strengths γij, which can be plastic. We also modeled the network with chemical weights following a log-normal distribution, which yielded similar results (data not shown).
We represent the post-synaptic potential response to a chemical or electrical spike with an exponential of the form exp ( - t - t j k τα ) for t > tjk. The excitatory and inhibitory synaptic time constants are τE and τI respectively and tjk represents the kth firing time of neuron j.
In between spikes, for every pair of inhibitory neurons i, j, the gap junction mediated sub-threshold current I i g a p is characterized by
I i g a p ( t ) = ∑ j = 1 j ≠ i N I I i j g a p ( t ) = ∑ j = 1 j ≠ i N I γ i j ( V j ( t ) - V i ( t ) ) , (9)
where γij is the gap junction coupling between inhibitory neurons i and j of respective membrane potential Vi and Vj. In our model, we suppose that gap junctions are symmetric with γij = γji. Gap junctions are initialized following a log-normal distribution with the location parameter μgap = 1 + ln(γ/NI) and the scale parameter σgap = 1.
Neurons also receive the current Inoise which is a colored Gaussian noise with mean ν, standard deviation σ and τnoise the time constant of the low-pass filtering
τ n o i s e s ˙ ( t ) = - s ( t ) + ξ ( t ) (10)
and
I n o i s e ( t ) = 2 τ n o i s e s ( t )σ+ν, (11)
with ξ is drawn from a Gaussian distribution with unit standard deviation and zero mean.
Our plasticity model is decomposed into a depression γ− and a potentiation term γ+.
Haas et al. [55] showed that bursting activity of both neurons or one of the two neurons leads to long-term depression (gLTD) of the electrical synapses. To capture this effect in our model, we first defined a variable bi which is a low-pass filter of the spikes of neuron i τ b b i ˙ ( t ) = - b i ( t ) + τ b ∑ t i k < t δ ( t - t i k ) , (12)
where δ is the Dirac function and τb = 8 ms is the time constant. When bi reaches a value of θburst = 1.3, this indicates that two or more spikes happened within a short time interval. Therefore, the burstiness of neuron i is characterized by H(bi − θburst) where H is the Heaviside function that returns 1 for positive arguments and 0 otherwise.
In our simplified model, we consider that the individual electrical coupling coefficient γ between neurons are non-directional. Every time the interneurons burst, the gap junctions undergo depression,
γ ˙ i j - ( t ) = γ ˙ j i - ( t ) = - α g L T D [ H ( b i ( t ) - θ b u r s t ) + H ( b j ( t ) - θ b u r s t ) ] , (13)
where αgLTD is the depression learning rate.
We fit αgLTD to the data by implementing the stimulation protocol used in [55]. We applied a constant current injection of 300 pA for 50 ms every 0.5 s (2 Hz) and of -80 pA the rest of the time, to maintain the membrane potential at -70 mV. This protocol lasts for 5 minutes. We estimate αgLTD = 15.69 nS ⋅ ms−1 by such that it leads to a depression of 13% of the gap junction strength at the end of the stimulation protocol, as reported by Haas et al.
If gap junctions were only depressed, they would decay to zero after some time. Therefore, there is a need for gap junction potentiation. However, no activity dependent mechanisms was reported yet in the experimental literature, but several studies suggest that the calcium-regulated mechanisms leading to long-term depression could be involved in potentiation as well [48, 53, 56, 113].
We consider two gLTP rules. The first has a soft bound, i.e. the magnitude of modification is proportional to the difference between the gap junction value and a baseline coupling strength γb γ ˙ i j + ( t ) = γ ˙ j i + ( t ) = α g L T P ( γ b - γ i j ( t ) γ b ) [ sp i ( t ) + sp j ( t ) ] . (14)
where αgLTP is the potentiation learning rate and spi(t) = ∑tik<t δ(t − tik).
The second gLTP rule we consider has no maximum bounds
γ ˙ i j + ( t ) = γ ˙ j i + ( t ) = α g L T P [ sp i ( t ) + sp j ( t ) ] . (15)
Moreover, to show that our results do not depend on the specific gLTP rule, we also consider a different gLTP rule where the update is passive and therefore does not depend on neural activity. This alternative rule yields similar results (supplementary material, S1 Fig).
The coupling coefficient is the ratio of voltage deflections when a step current was injected to one neuron of a coupled pair, which were maintained at a baseline voltage of -69 mV. During current injection, the injected neuron is hyperpolarized at -75 mV
c c 12 = Δ V 2 Δ V 1 , (16)
when 1 is the index of the injected neuron. The gap junction conductance used for measuring the coupling coefficient was obtained from the mean value of the gap junction coupling at steady-state. The coupling coefficient is about 5% for a network of 200 inhibitory neurons. Please note that the gap junction conductance and the coupling coefficient scale inversely to the network size in our model. We chose to use the mean value, as there is very little variance (4 orders of magnitude lower that the mean value) in the gap junction coupling strength at steady state. For reference, the coupling coefficient was measure around 12%±8% averaged for 313 pairs in the TRN [55]. Moreover, for adult rats, for 91 paired recordings of adjacent IO neurons, the coupling coefficient varies from 1% to 8% [114], and for 14 pairs of fast-spiking cortical neurons, the coupling coefficient was around 1.5% [115].
To estimate the plasticity direction for different value of external input ν and gap junction strength γ, we observe the activity of the network (without plasticity) in a steady state over a duration T = 6 s. For a chosen tuple (ν;γ), we average over time and over neurons the bursting and spiking activity
A b u r s t i n g = 1 T ∫ 0 T 1 N I ∑ i = 1 N I [ H ( b i ( t ) - θ b u r s t ) ] d t (17)
and
A s p i k i n g = 1 T ∫ 0 T 1 N I ∑ i = 1 N I sp i ( t ) d t . (18)
Then, we explore the values of the ratio of bursting over spiking activity
ratio = A b u r s t i n g A s p i k i n g (19)
as function of the coupling coefficient γ and of the mean external input ν over the parameter space P 1 = [ 0 ; γ m a x ] × [ 0 ; ν m a x ].
To quantify the frequency and the power of the oscillations in the neuronal activity, we perform a Fourier analysis of the population activity r which we define as the sum of neuron spikes within a population, during the time step dt r ( t ) = 1 d t 1 N I ∫ t t + d t ∑ i = 1 N I ∑ t i k < t δ ( u - t i k ) d u . (20)
We compute a Discrete Time Fourier Transform (DFT) and extract the power and the frequency of the most represented frequency in the Fourier domain. The formula defining the DFT is
r ^k = ∑ n = 0 N - 1 r n exp ( - i 2 π k n N ) k = 0 , … , N - 1 . (21)
where the rn sequence represents N uniformly spaced time-samples of the population activities. We measure the amplitude of the Fourier components r ^ k for k = 1..N/2 (because the Fourier signal is symmetric from N/2 to N). We identify the maximal one, its associated frequency f m a x = k N and its power P = ( | r ^ k | / N ) 2.
To simulate the projection of a cortical layer onto another layer, we model downstream read-out neurons with the same regular spiking neuron model as the first cortical layer. The input Ij received by each downstream neuron is the projected activity of all excitatory and inhibitory neurons of the first cortical layer, multiplied by the coefficients WERON and WIRON respectively:
I j ( t ) = W E R O N ∑ i = 1 N E ∑ t i k < t exp ( - t - t i k τ E ) + W I R O N ∑ i = 1 N I ∑ t i k < t exp ( - t - t i k τ E ) . (22)
When delivering the step current Istep to the network (Fig 2D), the time at which the neurons receive Istep follows a normal distribution centered on the transition time, with variance 10 ms. This variability avoids the confound of transient and unstable synchronization of the network due to a strong input delivered to all neurons simultaneously.
We investigate the role of gap junction coupling and its plasticity in synchronizing networks having different oscillation frequency preferences. We design a network consisting of two subnetworks having the same topology as described in Network: Each subnetworks has 800 excitatory neurons and 200 inhibitory neurons. There are all-to-all chemical synapses within each subnetworks (their strengths are reported in Table 1). There are no cross-network chemical synapses. The intra-network gap junctions are all-to-all. In addition, we vary the number of sparse cross-network gap junctions from 0 to 40. The gap junction strengths are initialized following a log-normal distribution as described in Network. We take γ = 3 which yields AI behavior in the network and we take γ = 5.5 which yields bursting behavior in the SR regime.
One of the subnetworks is called the Slow Network (SN) and we change the value of the membrane time constant of its inhibitory neurons τv from 17 ms to 55 ms. This decreases the neuron sub-threshold resonance frequency, which also lowers the frequency of the subnetwork oscillation when it is in the synchronous regime. The second subnetwork has its neuron membrane time constant fixed at 17 ms and is called the gamma-network because it oscillates at gamma frequency. The simulations last 10 seconds, which is long enough for the gap junction coupling to reach its steady state when the gap junctions are plastic.
To quantify the similarity between population activities from both subnetworks, we evaluate the Pearson’s correlation coefficient between their population activities rGN and rSN from the gamma- and slow-network respectively. The firing rates, rGN and rSN are defined as in Eq (20).
We measure the mutual information between the mean currents from SN and GN with
I ( X ; Y ) = ∑ x , y p ( x , y ) log p ( x , y ) p ( x ) p ( y ) , (23)
where p(x, y) is the joint probability function of X and Y, and p(x) and p(y) are the marginal probability distribution functions of X and Y respectively. Time bins of 10 ms are used to estimate the probability functions.
For each subnetwork, we evaluate the frequency and power of their oscillations as described in the section Quantification of oscillation power and frequency. When the difference of oscillation frequency between both subnetworks is less than 1 Hz, we measure the cross-correlation of their population activities rGN and rSN ( r G N ⋆ r S N ) ( τ ) = def ∫ - ∞ ∞ r G N ( t ) r S N ( t + τ ) d t . (24)
The phase difference is measured as the time delay relative to the oscillation period
Δ ϕ = arg max t ( ( r G N ⋆ r S N ) ( τ ) ) T p e r i o d , (25)
where ⋆ is the convolution operator and Tperiod is the oscillation period.
We investigate whether gap junction coupling and its plasticity play a role in routing information between networks. We consider the same system as described in the previous section, with two subnetworks coupled with gap junctions, except here all the inhibitory neurons have the same membrane time constant τv = 17 ms (e.g. corresponding to resonance frequency at gamma). The first network, called the Input Network (IN) receives an input projected to its NIN neurons (NIN = 1000) by NIN weights drawn from a uniform distribution between 0 and 1. The second network is called the Output Network (ON, NON = 1000).
To examine if there is successful transfer of information between both networks, we attempt to reconstruct the input signal from the ON’s population activity rON. First, we obtain the low-pass filtered population activity of ON, rfilt, with
τ r r ˙ f i l t ( t ) = - r f i l t ( t ) + r O N ( t ) , (26)
with τr = 3 ms. Then we detect the rising and falling times of the filtered population activity by detecting when it crosses a threshold θr = 2. This gives us rising times t k *, when it crosses the threshold from below, and falling times, when it crosses the threshold from above. We obtain the peak intervals Tk by measuring the time difference between consecutive rising times T k = t k + 1 * - t k *.
For Fig 5D, we plot x ¯ k, the mean values of the input signal x between the rising times t k * and t k + 1 * as function of their corresponding peak intervals Tk x ¯ k ( t ) = 1 T k ∫ t k * t k + 1 * x ( t ) d t . (27)
We reconstruct the network input (Fig 5E and 5H) by doing a linear interpolation of the inverse of those peak intervals Tk, so that the input signal and reconstructed input have the same length.
x ^ ( t ) = ( 1 / T k + 1 - 1 / T k t k + 1 * - t k * ) ( t - t k * ) + 1 T k , ∀ t ∈ [ t k * ; t k + 1 * ] . (28)
Finally to estimate the quality of the reconstruction, we measure the Pearson’s correlation coefficient (which is invariant by affine transformation) between the input and the reconstructed input.
In order to test the robustness of the system we measure the quality of the reconstruction for an oscillatory input signal of which we vary the frequency f (Fig 5F) and amplitude A (Fig 5G).
x ( t ) = A [ c o s ( 2 π f t ) + 1 ] (29)
Then we measure the routing of random signals x(t) = νIN + σIN ⋅ ηIN(t), where νIN is the signal mean, σIN is the signal standard deviation, ηIN is an Ornstein Uhlenbeck fluctuation with correlation time τx = 100 ms and unit variance. We build a dataset of 10 input signals and then we measure the Pearson’s correlation coefficients between the input x(t) and the reconstructed input x ^ ( t ) for those 10 inputs respectively. For Fig 5I, we scale the log-normal distribution of the gap junction strength (see Network) with γ = 3 to set the network in the asynchronous, with γ = 5.5 to set the gap junction near their plasticity fix point, and γ = 8 for a regime with strong oscillations.
To study the robustness of the information routing to gap junction deletion, we randomly delete an increasing number of gap junctions and measure the evolution of the Pearson’s correlation between x and x ^. We also measure the change in the mean gap junction coupling, if there is plasticity, between the initialization (with γ = 5.5) and the steady-state (after 6 s of simulation).
All parameters are listed in Table 1 unless otherwise specified in a figure.
We list in Table 1 the parameters used for our simulations.
|
10.1371/journal.pcbi.1003975 | Heterogeneity and Convergence of Olfactory First-Order Neurons Account for the High Speed and Sensitivity of Second-Order Neurons | In the olfactory system of male moths, a specialized subset of neurons detects and processes the main component of the sex pheromone emitted by females. It is composed of several thousand first-order olfactory receptor neurons (ORNs), all expressing the same pheromone receptor, that contact synaptically a few tens of second-order projection neurons (PNs) within a single restricted brain area. The functional simplicity of this system makes it a favorable model for studying the factors that contribute to its exquisite sensitivity and speed. Sensory information—primarily the identity and intensity of the stimulus—is encoded as the firing rate of the action potentials, and possibly as the latency of the neuron response. We found that over all their dynamic range, PNs respond with a shorter latency and a higher firing rate than most ORNs. Modelling showed that the increased sensitivity of PNs can be explained by the ORN-to-PN convergent architecture alone, whereas their faster response also requires cell-to-cell heterogeneity of the ORN population. So, far from being detrimental to signal detection, the ORN heterogeneity is exploited by PNs, and results in two different schemes of population coding based either on the response of a few extreme neurons (latency) or on the average response of many (firing rate). Moreover, ORN-to-PN transformations are linear for latency and nonlinear for firing rate, suggesting that latency could be involved in concentration-invariant coding of the pheromone blend and that sensitivity at low concentrations is achieved at the expense of precise encoding at high concentrations.
| Understanding how sensory signals are optimally encoded by nervous systems is of strong interest to neuroscientists, and also to engineers as it may lead to more efficient artificial detection systems. This is particularly relevant to olfaction, because the current electronic noses are far outperformed by their biological counterparts in terms of speed and sensitivity. We here use the moth sex pheromone processing system as a relatively simple model to understand early olfactory coding. We found that performance increases when olfactory information passes from first- to second-order neurons. Second-order neurons respond on average with shorter latency and higher sensitivity than first-order neurons. We show that two critical factors, convergent architecture and neuronal heterogeneity, are needed to account for increased performance.
| In insects and vertebrates the first two neuronal layers of the olfactory system present the same organization where many ORNs in the first layer converge to a small number of output neurons in the second layer – PNs in insects and mitral cells in vertebrates [1], [2]. The ORNs that project onto a single glomerulus express a single type of olfactory receptors. Yet, they present heterogeneous dose–response properties [3].
The functional consequence of this convergence has been the subject of much interest. Theory [4] predicts and experiments [5] confirm that pooling N ORN inputs should increase the firing rate of output neurons by N and improve the signal-to-noise ratio by √N. Experiments in the fruit fly reveal that firing rates rise more rapidly in PNs than in ORNs and that weak odor inputs are more amplified than strong inputs [6]-[8]. Such a non-linear transformation leads to an efficient use of coding capacity [6] and a maximum preservation of information on odor quality [9].
Although previous studies investigated the change in firing rate when sensory information passes from first- to second-order neurons, they did not consider the latency of the response. This is restrictive given that environmental conditions put strong constraints on the behavioral response time. In natural odor plumes, encounters with the stimulus are brief and intermittent, with up to five contacts per 1 s and each contact lasting down to under 20 ms [10]–[12]. Consequently, behavioral responses to odors are fast (<100 ms in rodents [13]–[15] and insects [16], [17]) and the olfactory system, like other sensory systems [18], [19], may rely on response latencies for fast odor discrimination [10], [12], [20].
In this work, we compared the transformations in latency and firing rate from first- to second-order neurons and assessed whether the cell-to-cell heterogeneity contributes to this transformation. We addressed these issues in a favorable model, the male moth subsystem that processes the sex pheromone. Each pheromone component activates with high specificity a single ORN type [5] whose axons project to one of a few glomeruli – the macroglomerular complex (MGC). The second-order neurons (PNs and local neurons LNs) in the largest MGC glomerulus – the cumulus – receive their sole olfactory input (homotypic) from the most abundant ORN type sensitive to the major component, with no lateral olfactory input from other glomeruli (heterotypic) (e.g. [21], [22]). This is a significant advantage for experimental analysis with respect to glomeruli sensitive to general odors that receive both homotypic and heterotypic inputs because of the lack of specificity of generalist (non-pheromonal) ORNs [7]. Apart from this difference, which allowed us to record responses with well-defined input, the isomorphic glomeruli involved in general odor processing and the cumulus share the same functional organization, so that the main properties found here for the pheromonal system should apply also to the general odorant system.
We found that both firing rates and latencies of ORNs and PNs are strongly dose-dependent, that PNs respond with a higher firing rate and a shorter latency than most ORNs, and that the sensitivity and speed of a given neuron are not correlated. We found also that dose-response curves are variable among ORNs (and PNs) and that this variability of essentially biological origin arises more from the variability across neurons (heterogeneity) than within single neurons (irregularity). For firing rate, the ORN-to-PN amplification mechanism is non-linear, which augments sensitivity to weak odor signals in intensity coding and enhances distinction between different general odors in quality coding [6]; it can be explained by the convergence of many ORNs on single PNs. For latencies, on the contrary, the ORN-to-PN transformation is linear, which might favor concentration-invariant coding of odor blends, and requires ORN heterogeneity.
When stimulated with the components of the pheromone, the cumulus of the male moth Agrotis ipsilon was activated only by the main pheromone component, cis-7-dodecenyl acetate (Z7-12:Ac) (Fig. 1A). Conversely, the other glomeruli in the MGC were activated only by the other pheromone components (Fig. 1B–C). In electrophysiological recordings, the Z7-12:Ac-responsive ORNs displayed phasic-tonic responses (Fig. 2A, C) whereas the second-order neurons we studied shared a common multiphasic response pattern with an initial excitation followed by an inhibition (Fig. 2B, D) and frequently a final rebound (Fig. 2E). All stained multiphasic neurons were found to be PNs with dendritic trees in the cumulus and axons in the inner antenno-cerebral tract. The rare stained LNs we found (3 among 67 stained cells) were monophasic. Although these observations do not rule out the existence of LNs with a multiphasic response pattern, they support the contention that multiphasic LNs (if they exist) are rare in our recording conditions, which means that most if not all recorded neurons were PNs. For this reason, in the following, we used the more common term PN.
Even in the absence of pheromone delivery, the Z7-12:Ac-responsive ORNs and PNs spiked tonically. This spontaneous activity is stationary (Fig. 3A) with a median firing rate lower in ORNs than in PNs (Fig. 3A, C). The distributions of spontaneous firing rates Fsp (all symbols are defined in S1 Table) are well fitted to lognormal distributions with a longer tail in PNs than in ORNs (Fig. 3C; S2 Table). To determine whether the PN activity is influenced by ORNs at rest, the antenna was sectioned. The PN firing rate began to decrease ∼10 s after the section and reached a stable regime (∼70% lower, range 58-85%) after less than 5 min (Fig. 3B).
The present study is restricted to two aspects of the olfactory code – the initial firing rate F (as defined in S1B, D Figure) and response latency L (S1A, B Figure) and their dose-dependent transformations – without ignoring that other aspects of the responses, e.g. action potentials after stimulus offset [23] or correlated activity in different neurons [24], may also provide useful information.
A feature of the studied response variables is immediately noticeable. The pairs (F, L) of a given response recorded from ORNs and PNs are quite distinct, especially at low doses. For example at dose C = −1 log ng, PNs fire with higher rates and shorter latencies than ORNs so that most pairs (F, L) from ORNs and PNs do not overlap (Fig. 2F). This may seem paradoxical because one would expect that the shortest latencies be a little longer in PNs than in ORNs on account of axonal conduction and synaptic transmission. This apparent paradox is the main theme of this paper and its resolution required to analyze how the neuron responses depend (or not) on the dose, the ORN and PN variability and the ORN-to-PN convergence ratio.
In order to document this feature and to provide an overview of how the two neuron populations studied respond to a given dose of Z7-12:Ac the firing rates and latencies of all recorded neurons at each applied dose were pooled. It was found in this way that the firing rate F presents four distinct properties (Fig. 4, top row; Fig. 5, left column): (i) The firing rates across neurons stimulated at the same dose follow Gaussian distributions in ORNs (Figs. 4A, 5A) and PNs (Figs. 4C, 5C). (ii) The mean of the distributions increases with the dose (Fig. 5A, C). At the lowest dose applied (−1 log ng for ORNs, −3 for PNs) the distributions are not significantly different from the control stimulations with pure air or hexane (Fig. 4A, C). The frequencies at the two highest doses tested (3 and 4 log ng for ORNs, 0 and 1 log ng for PNs) are also not significantly different (Fig. 4A, C). Therefore, when measured at the population level, the dynamic ranges of ORNs extend from −1 to 3 log ng and for PNs from −3 to 0 log ng. (iii) For doses C≤1, PNs respond more strongly than ORNs (Fig. 3E, G). (iv) The standard deviation of the distributions increases linearly with their mean for F<∼100 AP/s, with the same slope in ORNs and PNs (i.e. same coefficient of variation CV≈0.33), notwithstanding the very different doses evoking the same variability in the two populations (Fig. 5E). Above ∼100 AP/s, before saturation of the mean ORN firing rate, the variability of ORNs and PNs becomes constant (Fig. 5E).
Similarly, the latencies at each dose follow Gaussian distributions (Fig. 5B, D) with standard deviations proportional to their means (Fig. 5F). However, contrary to frequencies, (i) the variability decreases when the dose increases; (ii) the slopes (CVs) for ORNs and PNs are different, indicating a higher variability in PNs (Fig. 5F); (iii) no discontinuity in slope is seen at high doses; and (iv), at the same dose, the PN distribution is always shifted to the left of the ORN distribution (Fig. 4F, H) which shows that, at all doses, the PN latencies are shorter than the ORN latencies, thus confirming that the paradox noted above holds at all doses.
However, the mere statistical pooling of all available responses at a given pheromone dose, as done in the previous section, is not sufficient to describe adequately the neuron properties. A more detailed view of the system requires that responses to stimulations at increasing doses be analyzed on individual cells, not only cell populations. The main features of dose-response curves in single neurons are examined: first their overall “shape” in the present subsection, then their location along the dose axis and their correlations in the next subsections.
Dose-response plots were established for 38 ORNs and 47 PNs. Sigmoid Hill functions (see eqs. 3 and 4 in Methods) were fitted to the dose-firing rate C-F plots (Fig. 6). From the fitted parameters – maximum firing rate FM, efficient dose 50% (ED50) C1/2, and Hill coefficient n – we also derived three characteristics: the doses at threshold C0 and at saturation CS and their difference, the dynamic range ΔC (eqs. 5–7;). The firing rate responses to the lowest doses are not significantly different from controls (Fig. 4A, C) and those to the two highest doses are nearly equal (within 15% in 87% of ORNs and 80% of PNs, showing that the observed maximum firing rates were close to the asymptotic FM) which guarantees that the parameters were correctly estimated. Latencies were analyzed the same way. Decreasing linear functions with a lower bound were fitted to the dose-latency C-L plots (Fig. 7). Each neuron was characterized by its maximum latency LM at threshold C0, its minimum latency Lm and their difference ΔL = LM − Lm.
Complementary aspects of the distributions of these coding properties such as averages, variability and correlations were analyzed. As far as the “shape” parameters are concerned, we found that the typical ORN, reconstructed from the median values of the fitted parameters (S3 Table), has a high maximum firing rate FM (163 AP/s) and a wide dynamic range ΔC (3.6 log units), whereas the typical PN, reconstructed in the same way, has a lower FM (62 AP/s) and a smaller ΔC (2.5 log units). For latencies (S4 Table), the maximum LM (164 vs. 107 ms) and the range ΔL (104 vs 64 ms) are greater for the typical ORN than the typical PN. Except for FM, the variability (SD or interquartile range) of all these properties is slightly higher in PNs than in ORNs (Figs. 6E, F and 7E, F).
Although the dose-response curves of ORNs and PNs are basically similar in shape, the essential difference between them is that they are not located identically on the dose axis, which means that ORNs and PNs respond with similar firing rates and latencies but at very different doses. This calls for a clear distinction of the dose-independent properties (like FM, ΔC, LM, ΔL), analyzed above and the dose-dependent properties (C0, C1/2, Cs) which are examined now.
PNs are clearly more sensitive than ORNs. For example, the recruitment of PNs starts at lower doses as half of the PNs were activated at −3 log ng and half of the ORNs only at −0.5 log ng (not shown). PNs approach saturation at doses 3 orders of magnitude lower than ORNs. These changes testify that major transformations take place in the neural network of the cumulus when the sensory signal passes from ORN to PN. The ORN-to-PN transformations can be represented in two complementary ways: either by pairs of dose-response curves (Fig. 7A, B), or by transfer functions linking the latencies (or frequencies) of the ORNs and PNs at the same doses (Fig. 8C, D). Interestingly these curves (and the transfer functions derived from them) can be determined in two different ways, either directly from the pooled distributions of F and L at each dose (Figs. 4, 5), or from the distributions of the parameters determined on single dose-response plots (Figs. 6, 7). Both methods give practically the same results as shown here by the median and extreme values (quantiles 10% and 90%) of the firing rates (Fig. 8A) and the latencies (Fig. 8B).
So far only univariate distributions have been considered since the 11 properties that describe the C-F curves (fitted parameters FM, C1/2, n and derived characteristics C0, Cs, ΔC) and C-L curves (parameters L0, λ, Lm and characteristics LM, ΔL) were analyzed separately. This analysis must now be completed by examining the links between them. For this purpose bivariate plots and their associated correlations were studied. S5 Table assembles the coefficients of correlations and significance levels of the (112−11)/2 = 55 non-trivial pairs of properties in ORNs (top) and in PNs (bottom). It shows that 19 pairs (35%) are significantly correlated at level 1% in ORNs and also 19 in PNs, indicating that the overall correlative structure is similar in both populations. However, because characteristics are derived from parameters (for example ΔC depends on FM and n, see eq. 7), correlations between characteristics are expected to be more frequent and less informative than the correlations between parameters. Indeed, of the (62−6)/2 = 15 pairs of parameters only 3 (20%) are significantly correlated in ORNs and the same number in PNs. Therefore, in the following, priority is given to the parameters over the characteristics.
Another distinction is between two main types of pairs – those associating properties of the same variable, firing rate or latency, that provide information on the level of redundancy of properties, and those crossing the two variables that provide information on the link between them.
In the first type, of the 3 pairs of F-parameters (FM-C1/2, C1/2-n and n-FM) none is significantly correlated (p>0.01) in ORNs (S2A-C Figure) and a single one in PNs (C1/2-n, S3B Figure), indicating that the most sensitive PNs (with small C1/2) tend to have a steeper slope n. Similarly, of the 3 pairs of L-parameters a single one is uncorrelated in both ORNs and PNs, suggesting that L-parameters are more correlated to one another than F-parameters. The uncorrelated pair (λ-Lm, S2E, S3E Figures) indicates that neurons with longer latencies at C = 0 tend to have steeper slopes λ and longer minimum latencies Lm.
As for the two-variable type, among the 9 pairs between the 3 F- and 3 L- parameters, a single one is significantly correlated but it is not the same in ORNs (C1/2-L0, S2J Figure) and in PNs (C0-Lm, S3P Figure). Although the correlation C1/2-L0 suggests that fast ORNs (small latency L0 at C = 0) have high affinity (small C1/2), it is not confirmed by direct comparison as threshold C0 and minimum latency Lm in each ORN are not correlated (S2P Figure), showing that the ORNs with the lowest thresholds are usually not the fastest. As the reverse situation holds for PNs, this lack of consistency and the low proportion of significant correlations between F- and L-parameters (11%), and still more between F- and L-properties (10% for ORNs, 7% for PNs), support the overall independence of sensitivity and speed in ORNs and PNs.
The firing rates and latencies are variable across ORNs (and PNs) stimulated at the same dose (Figs. 4, 5). As shown below variability plays an essential role in the ORN-to-PN signal transformation which calls for a proper understanding of its sources and structure. First, variability arises from experimental and biological sources. Experimental variability in ORNs results from uncertainties on the dose and the delivery time from the stimulating device, and from the relative geometry of the airflow and the recorded sensillum. It must be reduced but can never be eliminated. Biological variability is more fundamental because it is an intrinsic property of the investigated system. It can be known only by subtracting the experimental variability from the overall observed variability. Second, variability arises from irregularities within single units and heterogeneities across units, where units can be neurons or pheromone stimuli. This distinction is important. The term ‘irregularity’ is used consistently throughout the paper to indicate the variability in firing rate or latency of the same neuron, or the variability on dose or delivery time of the same stimulus, following repeated stimulations with the same cartridge. By extension it indicates also the variability of spontaneous firing of the same neuron over time. Similarly the term ‘heterogeneity’ indicates the variability in response of different neurons, or the variability on dose and delivery time for repeated stimulations with identically prepared cartridges.
Systematic measurements (see Materials and Methods) showed that the biological component accounts for ∼90% of total variability. Biological variability results more from heterogeneity across ORNs (∼95% for F, ∼55% for L) than from irregularity within ORNs. These observations support a relatively simple interpretation, that the observed variability of ORN responses reflects primarily across-neuron heterogeneity. When compared to ORNs, the irregularity within PNs was smaller for firing rate (70%) and latency (80%), and the heterogeneity across PNs was equal for firing rate (Fig. 5E, superimposed blue and red lines) and larger for latency (140%, Fig. 5F, lines with different slopes).
Of special interest from a population coding point of view are the most active and the fastest neurons. To estimate the effect of the heterogeneity of ORNs and PNs and reconstruct their full range of variability, we selected the 10% most extreme values at both ends of the distributions of the parameters describing the dose-response curves (staircases in Figs. 6, 7). Fig. 8A shows that the F/FM curves of the 10% most efficient ORNs (derived from leftmost blue dashed curve) and the typical PN (derived from red solid line) are relatively close. This observation is also true for the C-L curves (Fig. 8B). Thus, the 10% most efficient PNs are likely triggered by a small fraction of ORNs. The transfer functions provide useful complements. The function of the 10% most (respectively least) efficient neurons (Fig. 8C, dashed and dash-dot lines) is less (respectively more) curved than the average function (solid line). The function of the 10% fastest neurons is a steep line that decreases to shorter latencies (Fig. 8D) than the median function.
Not all ORNs in the population contribute equally to the PN response. The major contribution comes from the ORNs whose latency is shorter or equal to the PN latency, since no PN can respond faster than its presynaptic ORNs. In order to determine the fraction of contributing ORNs we relied on a model based on the C-F and C-L curves and the distributions of their parameters established above (see last section “Model of the signal delivered by the ORN population” in Materials and Methods). The model predicts the firing rates and latencies observed in the ORN population (Fig. 9A, B) and allows us to simulate the spike trains fired by this population when stimulated (Fig. 9C) and in the absence of stimulation (Fig. 9D). From these simulations, we calculated at any dose C the proportion of ORNs that respond with a given latency L1 or shorter. This proportion, as shown in Fig. 9E when L1 is the latency of the typical PN reconstructed from the median values of the fitted parameters, decreases with the dose and only 5±2% of the ORNs are enough to activate the typical PN. The proportion is greater (16±7%) for the slow PNs and smaller (2±0.3%) for the fast ones.
The response kinetics were obtained from the model by simulating the total firing of the heterogeneous population of ∼7000 Z7-12:Ac ORNs knowing the statistical distributions of their experimental C-L and C-F curves. At all doses, the simulated PSTH shows that the instantaneous firing rate of the ORN population increases and reaches a maximum ∼200 ms after the stimulation onset (Fig. 9F). The initial growth results from the gradual recruitment of new ORNs. Then, the pheromone dose at threshold can be determined. The ORN population spontaneously fires ∼27 000± SD √27000 AP/s (Fig. 9D). To be detected by PNs, the firing rate must be greater than 27000+r√27000, where r is the signal-to-noise ratio. In another moth species, Bombyx mori, Kaissling [25] showed that, in the range of pheromone loads eliciting a behavioral response in 40% to 80% of male moths (0.01 to 0.1 ng per cartridge in his experimental conditions), r varies from 3 to 31. Assuming that the same range of ratios applies to Agrotis, the ORN population must fire from 27 500 AP/s (for r = 3) to 32 000 AP/s (for r = 31). Fig. 9F shows that the lowest stimulus doses evoking 275–320 AP per 10 ms are in the range −4.6 to −3.1 log ng. At this dose the typical PN fires only 0.3–3.3 AP/s.
In this study, our goal was to interpret the input-output transformations taking place in an identified glomerulus stimulated with a well-defined odorant. We focused on how functionally homogeneous populations of first- (ORNs) and second- (PNs) order neurons with heterogeneous characteristics encode the odorant concentration in their firing rate and in the latency of their first spike. To this end we chose a glomerulus specialized in the processing of the main component of a sex pheromone. This glomerulus with its associated ORNs and PNs is essentially similar to the other glomeruli in the antennal lobe so that the conclusions drawn from its study are of general applicability. However, it presents decisive advantages for the intended investigations because its ORNs are selectively activated by the pheromone component, do not project to other glomeruli, are the most numerous in the antenna and present the largest ORN-to-PN convergence ratio of all glomeruli.
Single ORNs and PNs encode the pheromone concentration in the same way, their dose-firing rate curves being well fitted to Hill functions – a classical fit for ORNs [25], [26]. The maximum firing rate FM of ORNs is usually much larger in insects (100–300 AP/s; e.g. [6], [27] with an exception <60 AP/s, [28]) than in vertebrates (13–50 AP/s, [26]). The same remark holds for PNs in insects (170–250 AP/s; [6], this work) and the analogous mitral cells in vertebrates (∼20 AP/s in frog; [29]). Also, the dynamic range ΔC is much wider in pheromone-sensitive ORNs than in vertebrate generalist ORNs (usually <2 log units, [26]). Because, for a given Hill coefficient, ΔC depends on FM, it is tempting to speculate that the reason why the firing rate of pheromone-responsive neurons is so high, despite its large energetic cost, comes from the importance of detecting pheromones at very low concentration. The high firing rate would then be the price to pay to have a low threshold and a wide dynamic range.
The median latencies LM at threshold in A. ipsilon in ORNs (164 ms) and PNs (107 ms) are consistent with those in non-pheromonal honeybee PNs (∼125–150 ms; [12]), and much shorter than those in frog ORNs (0.7–1.9 s; [25]). A possible interpretation of the faster response of insect ORNs is that insect ORs were proposed to be ionotropic [30], [31] whereas vertebrates ORs are metabotropic [32]. These observations raise the question of whether latency actually contributes to encode pheromone concentration because the stimulus onset defining time zero is not known by the brain. However, it has been shown that relative latencies, i.e. patterns of spikes, can be used as a replacement of stimulus onset [11], [15], [33]. This applies when successive stimuli are well separated (as in pheromone plumes where molecules remain grouped in clumps and filaments separated by clean air [34]) and even when they are not, for example in an ensemble of mitral cells [20]. Moreover, latency coding is consistent with temporal coding mechanisms in which the precise timing of action potentials carries information on the stimulus [35]. It is now recognized that temporal coding is widely used in all sensory systems, whether olfactory [36], tactile [37], [38], auditory [39], or visual [40].
The overall transformations of the olfactory code from ORNs to PNs can be determined in two different ways, either directly at the population level from their pooled responses at a given dose (Figs. 4, 5) or indirectly from responses of individual neurons across doses (Figs. 6, 7). The two approaches are consistent as the same transformations were found in both (Fig. 8A, B) strengthening a posteriori the more complex but more informative single-cell analyses.
Coding properties at the system level depend on the many-to-one ORN-to-PN convergence and individual differences between neurons in the ORN and PN populations.
The ORN-to-PN transformations, which entail stronger and faster responses, considerably differ quantitatively, the transformations at low doses being linear in latency and highly nonlinear in firing rate. These effects may be partly interpreted within the “efficient coding hypothesis” [56], [57] stating that the sensory neurons, including the moth pheromonal ORNs [58], are adapted, through both evolutionary and developmental processes, to the statistics of their natural stimulus.
The effect of the nonlinear ORN-to-PN firing rate transfer function is to give more weight to the low doses than to the high doses. The typical PN reaches saturation at a relatively low dose (CS ≈ 0 log ng) with respect to ORNs (Fig. 8A) and so cannot discriminate doses above 0 log ng. Such a transformation is reasonable from an efficient coding point of view because the high pheromone concentrations found within filaments far from the source [59] are rare and not informative for localizing the source. This nonlinear transformation is reminiscent of logarithmic companders for coding waveforms with a wide dynamic range, such as voice, on a finite number of levels, favoring the most frequent signals at the expense of the least frequent ones [60]. The same interpretation can be applied here to the coding of pheromone concentration (e.g. a lognormal distribution of concentrations in nature will be encoded more uniformly by PNs). In the fly this type of transformation was interpreted as favoring the qualitative discrimination of odorants. At any dose, odorants evoke a wide range of firing rates in ORNs [61] and PNs but their distribution is more uniform in PNs [6]; the hyperbolic transfer function explains this histogram equalization. Thus, the nonlinear ORN-to-PN transformation can be interpreted as favoring qualitative discrimination of odorants in the generalist pathway and detection at low doses in the pheromonal pathway.
In contrast, the linear latency transfer function preserves the type of statistical distribution between ORN and PN latencies (e.g. a Gaussian remains Gaussian) and so cannot perform histogram equalization. A possible advantage of this linearity is to make the coding of pheromone identity concentration-invariant [11]. In A. ipsilon, although the ratio of concentrations of one of the minor components to the major component is 4, linear latency functions with the same slope will introduce the same constant delay between the specialized ORNs responding to the components whatever the blend concentration. Any change of the ratio will produce a detectable change in the delay, signaling an inadequate pheromone blend.
In this view, the ORN-to-PN convergence may be more important for fast detection of the signal than for precise determination of its intensity. The sensitive response of the PNs to the fastest ORNs and their saturation at relatively low doses make the PN output more stable to dose variations than the ORN output and favor information on the temporal structure of the plume over its concentration fluctuations. The cumulus appears to obey the same principles as the ordinary glomeruli with a notable difference. Because of a higher ORN-to-PN convergence ratio, the response of PNs in the cumulus is presumably more sensitive and faster than in ordinary glomeruli. Temporal discrimination, a constraint in ordinary glomeruli, becomes apparently a major issue in the MGC.
Agrotis ipsilon (Lepidoptera: Noctuidae) were reared on an artificial diet until pupation [62]. Adults were fed with a 20% sucrose solution. All experiments were performed on sexually mature virgin males 5 days after emergence.
A constant airflow (24 ml/s) was blown over the antenna through a mixing tube. It resulted from mixing a constant airflow (17 ml/s) with an alternating flow (7 ml/s) of clean air between stimulations or odorized air during stimulations. Stimuli were delivered by means of a device (Syntech, Kirchzarten, Germany) blowing air during 200 ms through a Pasteur pipette containing the pheromone-impregnated filter paper and inserted into the mixing tube. Successive stimulations were separated by intervals of at least 60 s in ORNs and 30 s in PNs; these intervals are sufficient for complete return to spontaneous activity (Fig. 2E), except with the highest doses tested (3 and 4 log ng for ORNs and 1 log ng for PNs) for which longer intervals were used. In electrophysiological experiments, the main pheromone component of A. ipsilon, (Z)-7-dodecen-1-yl acetate (Z7-12:Ac) was used at different loads M (1 pg to 10 µg). In calcium imaging experiments also the two minor pheromone components (Z)-9-tetradecen-1-yl acetate (Z9-14:Ac), and (Z)-11-hexadecen-1-yl acetate (Z11-16:Ac) [63] were used for stimulation at a single load (10 ng). Doses denoted C were expressed as the decimal logarithm of loads in ng.
All 3 components of the pheromonal blend [63] were tested (Fig. 1). Recordings were performed as described in [64]. Briefly, Calcium Green 2-AM was bath-applied for at least 1 hour. A TILL Photonics imaging system (Martinsried, Germany) was coupled to an epifluorescent microscope (Olympus BX-51WI) equipped with a 10x (NA 0.3) water immersion objective. Signals were recorded using a 640×480-pixel, 12-bit monochrome CCD camera (TILL Imago).
Only the major component (Z7-12:Ac) was tested. Recordings from ORNs were performed as described in [62]. Briefly, the glass recording electrode was brought into contact with a cut sensillum and the reference electrode was inserted in an adjacent antennal segment. Great care was taken to cut the sensilla at the same short distance from their tip. Each ORN was recorded for 10 s before and 40 s after the onset of each stimulus at 6 doses from -1 to 4 log ng. Only recordings with spikes clearly attributable to a single ORN were kept for further analysis. PNs were recorded from the cumulus area with two different techniques. Extracellular recordings (EPN) were performed with glass electrodes of 5 MΩ resistance as described previously [65]. The electrical activity of one or several neurons was recorded. Spike-sorting was performed with the R-package SpikeOMatic [66]. All EPNs were stimulated twice at five doses from −3 to 1 log ng. For each stimulus, 12 s of post-stimulus activity were recorded. Intracellular recordings (IPN) were performed with a glass electrode of 150–200 MΩ resistance filled with Lucifer Yellow or neurobiotin as described in [67]. IPNs were tested at four doses, −2, −1, 0 and 1 log ng and recorded for 3 s after the onset of stimulation. The brain was then dissected, histologically treated and scanned in a confocal microscope as a wholemount. All IPNs kept for analysis shared the same physiological and (when available) morphological characteristics of PNs. Only a few LNs were impaled in our recording conditions and they never showed phasic response patterns. The distributions of the response frequencies and latencies of IPN and EPN were compared by Kolmogorov-Smirnov tests. No significant difference was found, so the two samples were pooled.
The time varying spike rate function f(t) was estimated by using the kernel method [68]. The spike trains of ORNs and PNs were convoluted with a Gaussian function of SD 50 ms. The raw response firing rate Fraw was defined as the height of the peak of f(t) (S1B, D Figure). In PNs with triphasic responses (excitation-inhibition-excitation) Fraw was determined on the first peak. Height Fraw was compared to the bumps of f(t) during spontaneous activity in the same neuron. The number of bumps of height f ≥ Fraw was counted; when it was less than 5% of the total number of spontaneous bumps, response Fraw was considered as significant.
Response time T is the time elapsed from the opening of the electric valve to the first spike of the response. For ORNs (S1A Figure), this spike is defined without ambiguity because of their low spontaneous activity. For PNs (S1C Figure), a few ambiguous cases due to spontaneous activity were resolved visually (the correction is conservative as it always increases T). Response latency L was defined as the difference between T and the mean transport time Tt (180 ms), L = T – Tt. As Tt follows a Gaussian distribution with σ≈13 ms (80% values are in the range 170-210 ms); a minor drawback of this definition is that at high doses the shortest latencies of PNs (not ORNs) can become negative if Tt is short.
Control responses (Fig. 4A, B) resulting from puffs of the solvent (hexane) were recorded in ORNs (15± SD 10 AP/s) and from puffs of hexane and non-odorized air in PNs (59± SD 28 AP/s; the distributions for hexane and pure air were pooled as they were not significantly different, p-value 0.72). To obtain the pure olfactory component F the average Fc of the control responses in each neuron was subtracted from the responses Fraw measured in the same neuron, F = Fraw – Fc. Rate Fraw is shown only in Fig. 4A, B; all other figures use the pure component F.
For each set of recordings from a neuron, empirical functions were fitted to the experimental points in the dose-response plots C-F (Fig. 6) and C-L (Fig. 7).
From eq. 4, the firing rate FP of a PN at any dose C can be derived as a function of the ORN firing rate FR at the same dose. Denoting (FRM, CR1/2, nR) the ORN parameters and (FPM, CP1/2, nP) the PN parameters, the transfer function for firing rate is(9)where (10)is a constant. Except for the −1 term in the denominator, function (9) has the same form as Hill function (4).
Eq. 9 can be derived from the ORN and PN Hill functions. However, the reverse is not true because the absolute positions of the firing rate curves along the dose axis (CR1/2 and CP1/2) are lost in the transfer function. Thus, a pair of dose-firing rate curves contains more information (6 parameters) than the corresponding transfer function (5 parameters).
Similarly, the latency LP of a PN at any dose C can be expressed as a function of the ORN latency LR at the same dose. Denoting (LR0, λR) the ORN parameters and (LP0, λP, LPm) the PN parameters, it can be shown from eq. 8 that (11)where (12)
The model is described in [69] with a few changes. Briefly, the response of each ORN stimulated at dose C is a spike train of latency L and interspike interval 1/F (Fig. 9C), where L is given by eq. 8 and F by eq. 4. The distributions of the parameters FM, C1/2, n, L0, λ and Lm (all Gaussian except n and λ, lognormal), their means μ and variances σ2 (see Figs. 6–8 E, F) and the correlations between parameters were determined from experimental data. This knowledge is expressed as the mean vector M and variance-covariance matrix Σ given in S6 Table. Each set of 6 parameter values characterizing an ORN was drawn from the 6-dimensional multinormal distribution defined by M and Σ. The model predicts the observed firing rates and latencies (no difference with observed distributions at doses from −2 to 2 using Kolmogorov-Smirnov test at level 1%, Fig. 9A, B). The spike trains of 7000 simulated ORNs were summated and the number of spikes fired by the population during each bin of 10 ms was counted. The number of Z7-12:Ac-responsive ORNs on the antenna (∼7000) was determined from the number of flagellar segments (∼90), the mean number of long sensilla trichodea per segment (∼80) and the number of Z7-12:Ac-responsive ORNs per sensilla (1).
|
10.1371/journal.ppat.1004310 | Determinants of Influenza Transmission in South East Asia: Insights from a Household Cohort Study in Vietnam | To guide control policies, it is important that the determinants of influenza transmission are fully characterized. Such assessment is complex because the risk of influenza infection is multifaceted and depends both on immunity acquired naturally or via vaccination and on the individual level of exposure to influenza in the community or in the household. Here, we analyse a large household cohort study conducted in 2007–2010 in Vietnam using innovative statistical methods to ascertain in an integrative framework the relative contribution of variables that influence the transmission of seasonal (H1N1, H3N2, B) and pandemic H1N1pdm09 influenza. Influenza infection was diagnosed by haemagglutination-inhibition (HI) antibody assay of paired serum samples. We used a Bayesian data augmentation Markov chain Monte Carlo strategy based on digraphs to reconstruct unobserved chains of transmission in households and estimate transmission parameters. The probability of transmission from an infected individual to another household member was 8% (95% CI, 6%, 10%) on average, and varied with pre-season titers, age and household size. Within households of size 3, the probability of transmission from an infected member to a child with low pre-season HI antibody titers was 27% (95% CI 21%–35%). High pre-season HI titers were protective against infection, with a reduction in the hazard of infection of 59% (95% CI, 44%–71%) and 87% (95% CI, 70%–96%) for intermediate (1∶20–1∶40) and high (≥1∶80) HI titers, respectively. Even after correcting for pre-season HI titers, adults had half the infection risk of children. Twenty six percent (95% CI: 21%, 30%) of infections may be attributed to household transmission. Our results highlight the importance of integrated analysis by influenza sub-type, age and pre-season HI titers in order to infer influenza transmission risks in and outside of the household.
| Influenza causes an estimated three to five million severe illnesses worldwide each year. In order to guide control policies it is important to determine the key risk factors for transmission. This is often done by studying transmission in households but in the past, analysis of such data has suffered from important simplifying assumptions (for example not being able to account for the effect of biological markers of protection like pre-season antibody titers). We have developed new statistical methods that address these limitations and applied them to a large household cohort study conducted in 2007–2010 in Vietnam. By comparing a large range of model variants, we have obtained unique insights into the patterns and determinants of transmission of seasonal (H1N1, H3N2, B) and pandemic H1N1pdm09 influenza in South East Asia. This includes estimating the proportion of cases attributed to household transmission, the average household transmission probability, the protection afforded by pre-season HI titers, and the effect of age on infection risk after correcting for pre-season HI titers.
| Three to five millions severe illnesses and 250,000 to 500,000 deaths worldwide are due to the influenza virus each year [1]. To guide control policies, it is important that the determinants of influenza transmission are fully characterized. Such assessment is complex because the risk of influenza infection is multifaceted. For each individual, it depends on immunity that was acquired naturally or via vaccination; but also on the level of exposure to influenza the individual has in the community or in the household, which may vary by season, household and individual. Here, from the analysis of original data and relying on new and innovative statistical methods, we ascertain in a unifying and integrative framework the relative contribution of variables that influence these different mechanisms.
This task is challenging because both protection and exposure are imperfectly characterized; and uncertainties about one may affect estimates for the other. For example, for haemagglutination-inhibition (HI) assays which are extensively used in the approval process for influenza vaccines [2], [3], it is generally accepted that a HI titer of 1∶40 is associated with a 50% reduction in the risk of infection [4], [5]. However, it has long been acknowledged that HI titers are only an imperfect correlate of protection. For example, in 2009, the proportion of elderly people estimated to be protected against H1N1pdm09 influenza was much higher than had been suggested by pre-pandemic HI titers [6]. In the first study that characterized the protective effect of HI titers, Hobson et al [4] used a challenge design to ensure all subjects in the study had the same level of exposure to influenza; but such approach is expensive and can only be applied to healthy adults. In non-experimental settings, however, it is harder to control for heterogeneity in individual exposures to influenza due to the difficulty to track down all potential sources of infection.
Case-ascertained household transmission studies have been extensively used to quantify exposure in the household setting [7]–[13]. In this design, community-based influenza cases, that are labelled as index cases, are recruited via primary care practices or outpatient clinics. Symptoms of the index case and their household members are then monitored for one to two weeks following symptoms onset in the index case; virological samples may also be collected. However, since the follow-up of each household starts with an influenza case, this approach cannot be used to reliably quantify exposure from the community or to estimate the relative contributions of households and the community in the general epidemic. Furthermore, as index cases must have sufficiently severe symptoms to make contact with a healthcare provider and then have sufficiently high viral loads to be detected by laboratory tests for influenza, there may be a selection bias towards more infectious cases, which may lead the probability of transmission in the household to be overestimated.
An alternative, less common design offers a more representative view of the role of households in influenza transmission. It is based on a cohort of households that are recruited prior to an epidemic and followed up during the epidemic [14], [15]. Although the timing and source of infection is typically unobserved, collection of serum samples at baseline and after the epidemic makes it possible to determine serologically which subjects were infected. Statistical methods exist to estimate from such data the probability of transmission from other household members and from the community [16]–[18]. However, they become cumbersome and numerically intractable as the number of categories of individuals (e.g. child/adult or low/intermediate/high HI titers) or the size of the social unit of interest (e.g. here households) increase [17], [19]. As a consequence, to our knowledge, it has never been possible to evaluate the protective effect associated with HI titers in such a framework, preventing a more integrated analysis of the determinants of influenza transmission.
Here, from the analysis of the large Ha Nam household cohort study [20] conducted from 2007 to 2010 in Vietnam and relying on new and innovative statistical methods [19], we ascertain in a unifying and integrative framework the protective effects associated with HI titers and age, along with the relative contributions of households and the community in influenza transmission. Differences by subtype are also investigated. The analysis makes it possible to ascertain potential biases in case-ascertained household transmission studies which are extensively used for early assessment at the start of influenza pandemics [8], [10]–[12]. The analysis also documents influenza household transmission in South East Asia, which has received somewhat less attention than in Western countries [9], [15], [21]–[23].
Samples were collected from a household-based cohort of 940 participants in 270 households in a single community in semi-rural northern Vietnam as previously described [20]. None of the participants had ever received influenza immunisation. Participants aged 5 years or older were asked to provide serial blood samples at times when national influenza surveillance data indicated that influenza circulation was minimal. The samples described here were collected over a period of three consecutive influenza seasons, from December 2007 through April 2010. Serological samples were collected between 1st–7th December 2007 (bleed 1), 9th–15th December 2008 (bleed 2), 2nd–4th June 2009 (bleed 3), and on the 3rd April 2010 (bleed 4). This provided three sets of paired samples either side of an influenza transmission season: 548 paired samples for season 1 (2008), 501 paired samples for season 2 (Spring 2009), and 540 paired samples for season 3 (Autumn 2009). In season 1, the influenza A virus strains detected in the cohort through ILI surveillance were A/H1N1/Brisbane/59/2007-like and A/H3N2/Brisbane/10/2007-like; in season 2, they were A/H1N1/Brisbane/59/2007-like and A/H3N2/Perth/16/2009-like; and in season 3, it was A/H1N1/California/7/2009-like. There was co-circulation of influenza B Yamagata lineage and Victoria lineage in both season 1 and season 2, with a predominance of Yamagata lineage in season 1 and Victoria lineage in season 2.
For each season and subtype, analysis was restricted to households with at most 1 individual for whom paired serum samples were missing.
Influenza hemagglutination inhibition (HI) assays were performed according to standard protocols [WHO 2011 manual]. The seasonal influenza A viruses used were isolated from participants' swabs or from swabs taken from patients presenting in Ha Noi in the same season and propagated in embryonated hen's eggs or in MDCK cells (ATCC). A reference antigen supplied by WHO (A/H1N1/California/7/2009-like) was used to assess season 3/pandemic sera. A single influenza B virus isolated from a participant during 2008 was used to assess serum for both the first and second seasons. The virus had a titer of 320 with B/Wisconsin/1/2010 (Yamagata) reference antisera and of <10 with B/Brisbane/60/2008 (Victoria) antisera. Each virus was first assessed for haemagglutination of erythrocytes from chickens, guinea pigs and turkeys then titrated with optimal erythrocytes. Serum was treated with receptor destroying enzyme (Denka Seiken, Japan) then heat inactivated and adsorbed against packed erythrocytes. Eight 2-fold dilutions of serum were made starting from 1∶10 and incubated with 4 HA units/25 µl of virus. Appropriate erythrocytes were added and plates read when control cells had settled. Virus, serum and positive controls were included in each assay. Pre- and post-season sera were tested in pairs. Each serum was tested in a single dilution series. The HI titer was read as the reciprocal of the highest serum dilution causing complete inhibition of RBC agglutination, partial agglutination was not scored as inhibition of agglutination. If there was no inhibition of HI at the highest serum concentration (1∶10 dilution) the titer was designated as 5.
Influenza virus infection was defined as a ≥4-fold increase in antibody titer from pre-season to post-season titers, with post season titers ≥40. For the purposes of analysis low, intermediate, and high pre-season HI titers were defined as ≤1∶10, 1∶20–1∶40, and ≥1∶80 respectively.
Data were collected for 3 different seasons s = 1…3: 2008 (s = 1), Spring 2009 (s = 2) and Autumn 2009 (s = 3). We classify the influenza virus into 4 different categories v = 1…4: seasonal A(H1N1) (v = 1); seasonal A(H1N1) (v = 2); seasonal B (v = 3); pandemic A(H1N1) (v = 4). A set of k = 1…K households are under study. Household k ( = 1…K) is of size nk. Individuals are categorized in two types: children i.e. aged ≥5 to ≤15 y.o. and adults.
A subject may be infected by influenza subtype v in the community (i.e. outside the household) or by another household member. Here, we define a generic model for the occurrence of these events.
During season s, the probability that subject i from household k has contacts in the community that would lead to infection by influenza subtype v is defined as . The force of infection from the community is modelled as:where measures the force of infection for subtype v during season s, captures the susceptibility of adults relative to children (i.e children are the reference group) and captures the effect of pre-season titers, with 3 categories low (≤10), intermediate (20–40) and high (≥80) (reference category: ≤10).
The probability that subject i gets infected if household member j is infected is defined as withwhere measures the transmission rate as a function of household size nk (the rate can be inversely proportional to nk [7] or independent of nk, depending on model variant), captures the infectivity of adults relative to children (i.e. children are the reference group).
It is challenging to estimate parameters of the transmission model from final size data because the chains of transmission are not observed. Here, we consider a simplified version of the approach developed by Demiris and O'Neill [19] to tackle the problem. A household of size n is represented by a random directed graph with n vertices (Figure 1). Each vertex represents a household member. Edges are added to represent the unobserved chain of transmission. Two types of edges are possible. If there is an edge between subject j and subject i, it means that subject i is infected if subject j gets infected. If there is an edge between the community and subject i, it means that subject i gets infected.
For a given digraph, it is possible to derive the likelihood function [19]. However, since the chains of transmission are unobserved, different configurations for the edges of the digraph may be consistent with the final size data (Figure 1). The digraph is therefore considered as ‘augmented data’ [24]. The joint posterior distribution of parameters and augmented data is explored by Markov chain Monte Carlo sampling. The algorithm explores the set of digraphs consistent with the data and estimates therefore correctly capture uncertainty about the digraph (see Text S1 for technical details).
We use a Uniform prior U([0; 10,000]) for all parameters except those characterising relative infectivity or relative susceptibility (i.e. to a reference group). For this latter class of parameters, following [8], we choose a log-Normal prior LN(0,1). This prior satisfies the invariance condition that for example the ratio (adult susceptibility/child susceptibility) has the same prior as the ratio (child susceptibility/adult susceptibility). In particular, it gives equal probabilities to the relative susceptibility of children versus adults being larger or smaller than 1.
Since the households under study represent only a fraction of households in the study area [20], we assume here that households are independent of each other. The assumption of independence, which is standard in this type of analysis [8], [14], [16], [25], [26], substantially reduces the computational burden compared with that of the more general model of Demeris et al [19].
A simulation study was carried out to investigate the performances of the statistical approach.
Once the model structure has been defined and methods to estimate the parameters of the model from that data are available, different model variants may be considered. For example, the effect of pre-season HI titers may be the same for all subtypes, may vary by subtype, by age group etc… Here we consider a large number of possible model variants. Each of them is fitted to the data and we determine the model variant that provides the best fit to the data. This model comparison exercise is essential to better characterize key dependencies in household transmission. We use the Deviance Information Criterion (DIC) for model comparison [27]. The smaller the DIC, the better the model. A DIC difference of 5 is considered to be a substantial improvement.
For each variable of interest, we explore the following variants:
In general, no satisfying version of the criterion exists for data augmentation frameworks such as the one used here [28]. This is because the likelihood of the observed data is not available. To solve this problem, we use importance sampling [29] to estimate the likelihood of the observed data and be able to derive the DIC. The likelihood is derived as follows. For each household, we simulate N = 2,000 epidemics in the household. The contribution of a household to the likelihood is then equal to the proportion of simulations where simulated infection statuses in the household perfectly match the observed ones (to avoid computational issues of likelihoods equal to zero, we assume that the sensitivity Se and specificity Sp of the diagnostic is not perfect, i.e. Se = 0.999 and Sp = 0.999).
In order to estimate the proportion of influenza cases that may be attributed to household transmission, we simulate epidemics in households where i) all parameters are drawn from the posterior distribution and ii) all parameters are drawn from the posterior distribution except the within household transmission rate which is assumed to be null. The case counts difference between i) and ii) gives the proportion of cases that may be attributed to household transmission.
For each pair of case-household contact in the dataset, we calculate the associated probability of transmission under the assumption that the case was the first or the only infected in the pair and derive the average household transmission probability across all pairs.
We compare the observed final size distribution with the one simulated with parameters drawn from the posterior distribution.
The research was approved by the institutional review board of the National Institute of Hygiene and Epidemiology, Vietnam; the Oxford Tropical Research Ethics Committee, University of Oxford, UK; and the Ethics Committee of the London School of Hygiene and Tropical Medicine, UK. All participants provided written informed consent.
Between 140 and 155 households (439–502 subjects including 95–121 children and 344–393 adults) were eligible for analysis, depending on the season and subtype. The average household size was 2.9.
Of all the model variants explored in our extensive model comparison exercise, Figure 2 summaries the characteristics of the model that had the best fit based on the DIC. The best fitting model had the following properties. The community risk of infection of children with low pre-season titers varied both with the subtype and the season (Figure 2A). It was minimum for H3N2 in 2008 and maximum for A(H1N1)pdm09 in Autumn 2009. The DIC substantially worsened if the community risk of infection of children varied with the subtype but was assumed to constant from one season to the next (ΔDIC = 24.7).
We found that high pre-season titers were protective against infection, with a reduction in the hazard of infection of 59% (95% CI, 44%–71%) for intermediate titers (20–40) and 87% (95% CI, 70%–96%) for high titers (≥80) (Figure 2B). DIC substantially worsened if the number of titer categories was reduced to 2 (ΔDIC = 20.8) or if pre-season titers were not accounted for (ΔDIC = 44.0).
Even after correcting for pre-season titers, we found that adults had half the risk of acquiring infection in the household compared to children (reduction in the hazard of infection of adults relative to children: 50%; 95% CI 32%–63%) (Figure 2C). Adding an age effect for each subtype did not improve the fit (ΔDIC = 0.2). Distinguishing pandemic versus seasonal influenza only provided a marginal improvement to the DIC (ΔDIC = −4.0) (reduction in the hazard of infection of adults relative to children for seasonal influenza: 41%, 95% CI 15%–58%; reduction in the hazard of infection of adults relative to children for pandemic influenza: 68%, 95% CI 42%–82%). Assuming the effect of age varied by subtype did not improve the fit (Figure S1; ΔDIC = 0.2). Ignoring the effect of the age of the subject on the risk of infection substantially worsened the fit (ΔDIC = 37.7). Assuming that infectivity changed with the age of the case did not improve the fit (ΔDIC = 13.2). Assuming the effect of pre-season HI titers could change with age, we found that a rise in HI titers had a slightly more pronounced effect on children than on adults (Figure S2). However, the fit of this model was not as good as that of our best fitting model (ΔDIC = 6.9).
The probability of transmission from an infected individual to another household member was 8% (95% CI, 6%, 10%) on average, and varied with pre-season titer, age and household size. In a households of size 3, the probability of transmission from an infected individual to a child with low, intermediate and high pre-season titers was estimated to be 27% (95% CI 21%–35%), 12% (95% CI, 8%, 17%) and 4% (95% CI, 1%, 9%), respectively. These probabilities dropped to 15% (95% CI 9%–23%), 6% (95% CI 4%–11%) and 2% (95% CI 0–5%), respectively, if the recipient was an adult. As has been found in studies of households in Western developed countries [7], [8], the best fitting model assumed that household transmission hazard decreased with increasing household size. Ignoring this dependency worsened the fit substantially (ΔDIC = 40.7). After correcting for these variables, estimating an effect of subtype on the probability of transmission in the household did not improve the fit (ΔDIC = 13.1). We estimated that 26% (95% CI: 21%, 30%) of cases may be attributed to household transmission. Figure S3 shows the prevalence of infection along with the estimated contribution of household transmission by season and subtype (NB: Figure 2A captures only partially variations in the prevalence of infection as the distribution of pre-season HI titers vary for each season and subtype and by age group).
The fit of the model to the data was adequate (Table 1).
In a simulation study we found all parameters could be estimated from the data and no important systematic bias was detected (Table S1). Out of 10 simulated datasets and 11 parameters, there was 94% probability that the simulation value was in the 95% CI.
We have characterised the determinants of transmission of seasonal (H1N1, H3N2, B) and pandemic H1N1pdm09 influenza from a household cohort study conducted in 2007–2010 in Vietnam.
We estimated that the household Secondary Infection Risk (proportion of household contacts infected by an index case, SIR) was approximately 8% on average. This is broadly consistent with estimates of SAR derived from case-ascertained studies, when diagnosis of contact cases is based on RT-PCR laboratory confirmation (median SIRPCR: 8%; range: 3%, 38%; n = 12) or on a clinical case definition of Febrile Acute Respiratory Illness (median SARFARI: 11%; range: 3%, 37%; n = 18) [12]. Lau et al [12] also reported two estimates of the proportion of household contacts who seroconverted of 20% [30] and 27% [31]. As expected, these proportions are larger than 8% since they capture transmission from the index case but also from the community for the whole duration of the epidemic. The similarity between our estimates and those derived from case-ascertained studies validates the use of case-ascertained studies as a way to obtain representative estimates of influenza household transmission. Overall, we estimated that 26% (95% CI: 21%, 30%) of influenza infections may be attributed to household transmission. This is consistent with other estimates in the literature [32].
We also estimated the risk factors for household transmission and the risk of infection. Pre-season titer and age had a strong impact on the risk of infection. An HI titer of 40 is generally accepted to give a 50% reduction in the risk of infection [5]. Here we found a slightly more subtle effect of pre-season titer, with the risk of infection decreasing incrementally with HI titer and the reduction being as high as 90% for HI titer ≥80. Even after correcting for pre-season titers, we found that adults had half the risk of acquiring infection compared to children. This supports the idea that HI titer is an imperfect correlate of protection. There is growing evidence that antibodies directed at the stalk domain of the HA protein may be important mediators of protection that accumulates with repeated exposure to influenza viruses but which is not detectable by the HI assay [33]. Consistent with other studies [7], [8], [34], we found that the household person-to-person transmission probability decreased with increasing household size.
Ours is the only contemporary study to prospectively assess the transmission of influenza in a random selection of all households (including those without children) in an unimmunised community over multiple seasons. The use of a final-size model based on serology minimizes the under-ascertainment inherent in studies that detect only symptomatic cases. As such we believe these results are the best available assessment of the risk of acquisition of influenza in the household and the community.
The earlier analysis of this dataset [20] simply reported empirical infection rates by age based on a four-fold or greater increase in HI titers between paired sera, and did not estimate any other transmission parameters nor influences on the probability of transmission. The analysis presented in this manuscript therefore adds substantial new insights including estimates of the probability of transmission from an infected individual to another household member, the proportion of infections acquired in the household and the community, and how the probability of infection is affected by pre-season HI titers, age and household size
This study has some limitations. First, the HI assay has imperfect sensitivity and specificity [35], [36]. As a consequence, the infection status of some individuals may be incorrectly classified. The use of microneutralization assay to detect pH1N1 seroconversions would have increased the sensitivity. The average number of households per season was relatively small (about 150). However, the study was run over 3 seasons and looked at multiple different subtypes (H1N1, H3N2, B, H1N1pdm09), for a total of 6 pairs season/subtype. This means that the amount of information contained in these data is roughly that of a study of 6×150 = 900 households run over 1 season and for 1 subtype. This explains why the credible intervals for most parameters are relatively narrow. We were unable to assess transmission risks in children aged less than 5 years, since serum samples were not obtained from these subjects.
Here, we disentangled the relative contributions of households and the community in the risk of influenza infection. This was made under the assumption that households were independent of each other and that all individuals of an age group were exposed to the same risk of infection in the community. Although standard in such analyses [8], [14], [16], [25], [26], in practice, the risk of infection in the community may have a spatial component, potentially leading to higher transmission rates between households that are close to each other. However, we were unable to test this assumption here since our dataset was not spatially structured. Estimating the effect of space on influenza transmission will be an important step forward. This can for example be done from the analysis of household serological cohort studies in which the spatial location of each household is be documented [37]. Ideally, one would like to integrate such analysis in the framework of Demiris and O'Neill [19], so that the contributions of households and space can be characterized in a single and coherent framework. This is an important subject for future research.
This study considerably extends previously limited evidence on influenza transmission in non-Western countries. It also validates the use of case-ascertained studies as a way to obtain representative estimates of influenza household transmission. This has important implications for early assessment of household transmission in future pandemics, as case-ascertained studies are the only household design that can be used close to real-time.
|
10.1371/journal.ppat.1000611 | A Granulin-Like Growth Factor Secreted by the Carcinogenic Liver Fluke, Opisthorchis viverrini, Promotes Proliferation of Host Cells | The human liver fluke, Opisthorchis viverrini, infects millions of people throughout south-east Asia and is a major cause of cholangiocarcinoma, or cancer of the bile ducts. The mechanisms by which chronic infection with O. viverrini results in cholangiocarcinogenesis are multi-factorial, but one such mechanism is the secretion of parasite proteins with mitogenic properties into the bile ducts, driving cell proliferation and creating a tumorigenic environment. Using a proteomic approach, we identified a homologue of human granulin, a potent growth factor involved in cell proliferation and wound healing, in the excretory/secretory (ES) products of the parasite. O. viverrini granulin, termed Ov-GRN-1, was expressed in most parasite tissues, particularly the gut and tegument. Furthermore, Ov-GRN-1 was detected in situ on the surface of biliary epithelial cells of hamsters experimentally infected with O. viverrini. Recombinant Ov-GRN-1 was expressed in E. coli and refolded from inclusion bodies. Refolded protein stimulated proliferation of murine fibroblasts at nanomolar concentrations, and proliferation was inhibited by the MAPK kinase inhibitor, U0126. Antibodies raised to recombinant Ov-GRN-1 inhibited the ability of O. viverrini ES products to induce proliferation of murine fibroblasts and a human cholangiocarcinoma cell line in vitro, indicating that Ov-GRN-1 is the major growth factor present in O. viverrini ES products. This is the first report of a secreted growth factor from a parasitic worm that induces proliferation of host cells, and supports a role for this fluke protein in establishment of a tumorigenic environment that may ultimately manifest as cholangiocarcinoma.
| The oriental liver fluke is endemic through South-East Asia and is the major cause of cause of liver cancer in north-eastern Thailand. The molecules that are secreted by the parasite cause cells to multiply quicker than they normally would, and excessive cell growth is a key stage in the initiation of many cancers. We identified a secreted protein from the fluke, termed granulin, which has a similar structure to a human growth factor associated with many aggressive cancers. Granulin is secreted by the parasite into the bile ducts where it causes host cells to proliferate. The proliferative activity of fluke secreted proteins was blocked by antibodies against granulin, indicating that it is the major cell growth-inducing molecule released by the parasite. Identifying the function of granulin will enable us to understand how and why this debilitating yet neglected pathogen causes cancer in so many people in South-East Asia. This and future work will contribute towards the development of new strategies to reduce both parasite prevalence and the incidence of the most fatal of liver cancers in Thailand.
| Cholangiocarcinoma (CCA), or cancer of the bile ducts, is prevalent in people from Thailand and Laos whose staple diet includes uncooked fish which harbour the liver fluke, Opisthorchis viverrini, the main risk factor for this cancer in the region [1]. There is no stronger link between a parasite and cancer than that between O. viverrini and CCA - indeed WHO data suggest that as many as one-third of the nine million infected people will contract cancer [2]. This is a striking figure compared to data from other carcinogenic microbes, such as Helicobacter pylori, human papilloma virus and the hepatitis viruses, where less than one percent of infected individuals develop infection-related cancers [2],[3].
For opisthorchiasis, in vivo studies in hamsters and in vitro investigations have indicated that the fluke's excretory/secretory (ES) products, metabolic products excreted and secreted into the external environment from the excretory openings and epithelial surface (tegument), include mitogens that likely play a role in the initiation of CCA in infected humans and experimentally infected hamsters [4],[5]. To gain a better understanding of the host-parasite interactions underlying the molecular pathogenesis of opisthorchiasis, we screened both the transcriptome [6] and the ES proteome (J. Mulvenna et al., unpublished) of the fluke for genes encoding proteins with ontologies that were associated with human cancers. A homologue of human granulin, a secreted growth factor implicated in many aggressive and invasive cancers, was identified.
The granulin domain consists of 12 highly conserved cysteines and is found in diverse phyla from eubacteria to humans, and subsequently has many synonyms [7]. Compounding the confusion, the term granulin can also refer to the small 6–10 kDa granulin domain (also named epithelins or GEM∶granulin/epithelin modules) found in the majority of animals, or the vertebrate protein, progranulin (PGRN), which in mammals is a large 60–90 kDa glycoprotein containing seven tandemly repeated granulin motifs [8]. PGRN protein is also known as PC cell-derived growth factor (PCDGF), proepithelin (PEPI), Granulin/epithelin precursor (GEP), GP88, acrogranin, granulin or epithelin precursor [9]. Herein we will refer to the large multihomodomain form from vertebrates as PGRN, and granulin (GRN) will refer to the individual granulin domains.
There is a broad distribution of PGRN in human organs and tissues, and elevated levels of mRNA are found in organs with neuronal cells (cerebellum), hematopoietic stem cells (spleen) and rapidly dividing epithelium (skin, gastrointestinal tract and wounded epithelia) [10],[11]. Numerous functions for GRNs have been reported but the roles in cell cycle control and wound healing are noteworthy [8]. Numerous mutations have been observed within the human PGRN gene with many linked to psychiatric disorders including Alzheimer's disease and frontotemporal dementia [12],[13].
Over-expression of PGRN is linked to tumorigenesis in numerous human tissues, including liver cancers, and is associated with an aggressive and invasive tumour phenotype [14],[15]. GRN is a potent proliferative agent but has other pro-tumor qualities that are not yet well characterized. It may promote carcinoma progression by promoting angiogenesis, insensitivity to apoptosis, promotion of tumor invasion and anchorage independence which all support tumor expansion in the unfavorable interstitial environment [7],[16],[17]. Preventing over-expression of PGRN in a range of tumor types, either through gene silencing or neutralizing antibodies, reduces or entirely inhibits tumor progression [18]. Over-expression of PGRN is an indicator of poor prognosis for a range of cancer types, and anti-GRN antibodies have been successfully employed in mice as therapy for hepatocellularcarcinoma (HCC) [19].
Large-scale gene sequencing efforts have revealed GRN homologues in the majority of parasitic phyla [20],[21],[22]. Like free-living eukaryotes, parasitic helminths probably utilize GRN to regulate growth and development of their own cells. By contrast, here we describe the detection of GRN in the ES products of O. viverrini and its binding to mammalian biliary epithelial cells in situ. Furthermore, recombinant O. viverrini GRN stimulated proliferation of fibroblasts, whereas antibodies against the recombinant GRN inhibited the ability of ES products to promote proliferation. Together these findings support a role for this fluke protein in establishment of a tumorigenic environment that may ultimately manifest as CCA.
Characterization of the protein profile of O. viverrini adult worm ES products using LC-MS/MS revealed a 19 amino acid peptide, with a MOWSE score of 50 (delta error -0.0464) (Figure 1 inset), that matched to a single contig encoding a protein with sequence similarity to human granulin (not shown). The cDNA was termed Ov-grn-1 and its protein product Ov-GRN-1; the sequences were submitted to GenBank under accession number FJ436341. Verification of the MS/MS identification was then performed using multiple-reaction monitoring (MRM) transitions targeted against the peptide identified in the shotgun proteomics. Firstly, a tryptic digest of O. viverrini ES products, collected after one day of culture, was analyzed and the target peptide identified and fragmented. Next, recombinant Ov-GRN-1 in water was digested and analyzed in the same fashion. The target peptide from the ES sample was identified at the same elution time and with an identical product ion spectrum as that generated for recombinant Ov-GRN-1 (Figure 1). The peptide observed showed very abundant y-ion fragments, as expected, on the C-terminal side of the peptide. The peptide also revealed a missed trypsin cleavage at the Arg residue, due to the Pro residue on its C-terminal side. The combination of the initial MS/MS data, elution time of the target peptide in both samples and the similarity of MRM fragmentation patterns strongly supports the presence of Ov-GRN-1 in the ES products of O. viverrini.
Ov-GRN-1, like homologues from the related liver fluke Clonorchis sinensis and earthworms, has an N-terminal signal peptide followed by a single GRN core domain. Most other proteins containing a GRN domain consist of multiple GRN domains (PGRN) or at least one GRN domain fused to other domains including proteases, protease inhibitors and fibronectin (Figure 2A). Ov-GRN-1 consists of a predicted secretion signal peptide followed by an 84 amino acid GRN domain of 9.04 kDa with twelve conserved cysteines. Whereas no N-linked glycosylation sites were predicted, four putative O-glycosylation sites were identified at Ser-26, Thr-35, Thr-41 and Ser-61. The core GRN domain of Ov-GRN-1 shared 43.6% identity at the amino acid level with granulin F, the closest human homologue, and 85% identity with an EST from the related liver fluke, Clonorchis sinensis (Figure 2B). Data from the few GRN structures available suggest that Ov-GRN-1 adopts the general GRN fold and disulphide bonding pattern akin to carp GRN, the only complete GRN structure available to date [23]. The NMR derived structure of carp (Cyprinus carpio) granulin [23] was used as the template on which to build a molecular model of Ov-GRN-1 (Figure 2C). The two proteins shared 32% identity over their granulin core domains. The lowest energy structure from 50 calculated using MODELLER was selected as an approximation of the structure of Ov-GRN-1. The model contained no violations of distance restraints and the Ramachandran plot, calculated with PRO-CHECK-NMR [24], showed a single residue in the disallowed regions.
Ov-GRN-1 grouped very closely with its orthologue from the related liver fluke, C. sinensis, a parasite that has also been implicated as a cause of human CCA [25], and this clade obtained 100% bootstrap support (Figure 3). GRNs from the placozoan Trichoplax, slime mould, the free-living nematode Caenorhabditis elegans and human granulin B also formed a clade with the liver fluke GRNs, although this did not obtain bootstrap support of greater than 50%. Interestingly, the blood fluke (Schistosoma) GRNs did not group closely with Ov-GRN-1 – most O. viverrini genes share greater sequence identity with other platyhelminth genes [6] than they do with genes from other phyla, suggesting that the phylogeny presented here reflects functional protein relationships rather than taxonomic relationships. It is also noteworthy that the schistosome GRN domains were probably derived from a multi-domain PGRN.
Reverse transcription PCR using RNA from different Opisthorchis life cycle stages amplified a product of the expected size, ∼300 bp, in all developmental stages tested (Figure S1), indicating that Ov-grn-1 was constitutively expressed throughout the developmental cycle of the liver fluke. An amplicon was not detected in the absence of reverse transcriptase enzyme (not shown), confirming the absence of contaminating genomic DNA. The constitutively expressed actin gene served as a control and was expressed in all stages.
Ov-GRN-1 was expressed in E. coli and in Sf9 insect cells. We herein refer to the E. coli-derived recombinant protein as Ov-GRN-1e and Sf9-derived protein as Ov-GRN-1s. Soluble 6×His tagged Ov-GRN-1s protein with a molecular mass of ∼14 kDa was expressed in Sf9 cells at a yield of ≤200 µg purified protein per litre of culture medium. A combination of cation exchange and Ni-NTA affinity chromatographies followed by a second Ni-NTA purification resulted in recombinant granulin that appeared to be greater than 95% pure based on SDS-PAGE gels (not shown). Ov-GRN-1e was highly expressed (up to 60 mg/L) in each of three E. coli cell lines (BL21, Rosetta, Rosetta-gami) but in each case the Ov-GRN-1e was insoluble and required urea (or other chaotropic agents) to solubilise. Different induction times and temperatures were assessed, and none of these variables promoted the solubility of the recombinant Ov-GRN-1e. To obtain denatured recombinant protein, BL21 E. coli cells transformed with pET41a encoding the Ov-GRN-1 ORF were grown at 37°C and induced for 16 h. Recombinant protein was purified on Ni-NTA resin under denaturing conditions to yield >95% pure protein (Figure 4A). Refolding of the purified denatured protein was undertaken, and the best conditions identified yielded ∼15% recovery of soluble protein by refolding in 20 mM Tris pH 7.5, 1 mM CaCl2 for 24 hr.
IgG antibodies raised against recombinant Ov-GRN-1e and GRN-1s proteins in mice were employed to probe the recombinant immunogens and Opisthorchis somatic adult extract (SAE) by Western blotting under both native and denaturing/reducing conditions. Bands were only visible when SAE was probed under native conditions with both anti-GRN-1s and anti-GRN-1e IgGs, with a strong band visible at 38 kDa, higher than the predicted 9 kDa mass of the monomeric protein (Figure 4B). Neither antibody bound to any proteins under reducing/denaturing conditions (not shown), suggesting that conformational epitopes were the target of anti-GRN-1 antibodies, and that the native protein might form homo-multimers or form complexes with other proteins under native conditions. Control antibodies did not produce any bands under native or denaturing/reducing conditions. Anti-GRN-1s IgG was used to localize the sites of expression within adult O. viverrini and in the surrounding bile ducts of an experimentally infected hamster. Ov-GRN-1 exhibited ubiquitous expression through all tissues, particularly in the gut, tegument and tegument extrusions of the adult worm (Figure 4C, right upper panel). Interestingly, the protein was also strongly detected in the bile duct epithelial cells in close proximity to the liver fluke. Control mouse IgG did not bind to any tissue or structures in the fluke or hamster tissues (Figure 4C, left panel). Additionally anti-Ov-GRN-1s IgG showed no affinity for host granulin as indicated by a lack of staining in uninfected hamster liver (Figure 4C, lower right panel)
Others have shown that O. viverrini causes proliferation of fibroblasts and the KKU-100 CCA cell line when these cells are co-cultured in the presence of live adult O. viverrini in a non-contact format [4],[26]. We reproduced these findings (Figures 5A and S2A) and proceeded to show that soluble ES products from O. viverrini stimulated proliferation of NIH-3T3 fibroblasts (Figures 5B, S2B and S3) and the KKU-100 CCA line (not shown). Cell proliferation was measured using two distinct approaches. WST-1 is a measure of metabolic activity of cells, and although it is routinely used to measure cell growth over time, metabolic variations in cells exposed to different conditions (e.g. in the presence of ES products) can mask changes in real cell numbers. We also assessed the growth of cells using a real time index of measurement to corroborate the proliferation quantified by the WST-1 assay using an xCELLigence system. The cell index readout is a real time measure of conductivity which is indicative of cell surface area in contact with the gold electrodes covering the plate surface [27]. We optimised the conditions for ES-induced cell growth to permit a thorough assessment of the effects of ES products (and other treatments) over time. Cells were seeded at an adequate density to determine growth over 3 days in a reliable manner before reaching confluence. Conditions that resulted in a minimum of two-fold growth of NIH-3T3 fibroblasts between samples treated with and without ES over 3 days were determined in the presence of increasing concentrations of bovine calf serum (BCS) in a 96 well plate – final conditions were 0% BCS seeded at 6000 cells/well, 2% BCS at 2000 cells/well, 5% BCS at 700 cells/well and 10% BCS at 300 cells/well. The optimal conditions identified for detection of approximately two-fold proliferation of NIH-3T3 fibroblasts induced by addition of ES compared with an equal volume of PBS (control) were as follows: 2,000 cells seeded per well and cultured for 3–8 h in DMEM containing antibiotic/antimycotic at 37°C in 95% air/5% CO2 and 2% BCS prior to addition of ES products (20 µl) to a final concentration of 20 µg/ml. Cells were cultured in the presence of ES or PBS and cell numbers were determined using the WST-1 dye procedure. Addition of ES to cells grown in all serum concentrations tested resulted in changes in cell growth and morphology; Figure 5C presents the results of cell growth in the presence of 2% BCS. The flattened fibroblastic shape of 3T3 cells changed upon addition of ES products, resulting in a longer, narrower and more refractive spindle shaped cell morphology.
A range of concentrations of recombinant Ov-GRN-1e was included with cells under different culturing conditions, based on information from other investigations with ES, as described above. Nanomolar concentrations (50–200 nM) of Ov-GRN-1e induced significant growth of cells above growth of control cells treated with either PBS or an irrelevant recombinant protein purified under the same conditions as for Ov-GRN-1e (Figures 6 and S4). Four hundred nM Ov-GRN-1e induced significant growth after one day (P<0.05), after which cell growth slowed and cell numbers were equivalent to cells treated with control protein by day 3 (Figure 6A). At higher concentrations (≥800 nM) the cells suffered adverse effects and did not survive beyond 24 hr (not shown). Compared to control protein, 200 nM Ov-GRN-1e promoted significant growth after one day (P<0.05) and 50–200 nM Ov-GRN-1e caused significant growth by day 3 (P<0.05; Figure 6B). Cell proliferation induced by both ES products and recombinant Ov-GRN-1 was completely ablated in the presence of 10 µM U0126, an inhibitor of Erk1/2 signalling, indicating that Ov-GRN-1, like human PGRN, signals via the MAPK pathway.
Intriguingly, refolded Ov-GRN-1e that was concentrated (retentate) using a 3 kDa cut-off centrifugal concentrator membrane did not induce cell growth, however the column flow through (i.e. >3 kDa) did induce proliferation. Indeed, the proliferation was greatly enhanced at lower concentrations (10 nM) compared with refolded Ov-GRN-1e that had not undergone concentration (50–200 nM; measured in real time by xCELLigence - Figures 7 and S5). Using both SDS PAGE and Western blotting with anti-6×His antibody we identified a small amount (∼5–10% of purified and refolded protein) of refolded protein that reproducibly passed through a 3 kDa cut-off membrane.
To determine whether Ov-GRN-1 was responsible for the mitogenic activity of ES, we attempted to neutralize the mitogenic activity of ES with anti-Ov-GRN-1 antibodies. Anti-GRN-1s IgG inhibited ES-induced proliferation of NIH-3T3 fibroblasts (Figures 8A and S6A). After 3 days of cell culture, significant inhibition of proliferation was evident at concentrations of 20 and 40 µg/ml IgG in both 2% and 5% BCS cultures (P<0.01 - <0.001). Similar antibody-induced suppression of proliferation was obtained in the absence of BCS over two days (not shown), but after two days control cells (treated with PBS) began to die.
NIH-3T3 cells were grown under identical conditions as described above −20 µg/ml ES, 20 µg/ml test or control IgGs and 2% BCS over 3 days - and monitored with the xCELLigence system (Figures 8B and S6B). Cells treated with ES alone or ES in the presence of control IgG grew at the same rate (F(2,186) P = 0.65) with a steady increase of 0.2–0.25 units over 24 hours and then a reduced rate of increase of 0.05–0.1 units for the subsequent two days. This was similar to the growth rates measured with WST-1. When ES in the presence of either anti-Ov-GRN-1e or anti-Ov-GRN-1s IgGs were incubated with cells, growth slowed significantly (F(2,186) P<0.0001) compared to cells treated with ES alone or ES plus control IgG and showed only minor variations from the growth profile of NIH-3T3 cells treated with PBS alone (Figure 8B).
The inhibitory effect of anti-Ov-GRN-1 IgG on the growth of the KKU100 CCA cell line induced by ES products was also assessed. Cells were cultured for 3 days in RPMI 1640/2% BCS with a final concentration of 20 µg/ml ES. Cell growth is presented as absorbance at 450 nm rather than as a growth ratio because of the differences in growth characteristics between this cell line and NIH-3T3 fibroblasts. The same general trend was observed, whereby anti-GRN-1e and to a lesser but still significant extent, anti-GRN-1s, IgGs inhibited proliferation induced by ES in an antibody dose-dependent fashion (Figure 8C). Significant inhibition was observed with 16 µg/ml (P<0.05) and 24 µg/ml (P<0.001) anti-GRN-1e IgG. Addition of IgGs to cells in the absence of ES had no effect on cell growth (not shown).
Initiation of CCA in chronic opisthorchiasis in humans [28] and experimentally infected hamsters [29] is thought to be multi-factorial, involving (1) infection-induced inflammation, particularly the release of reactive oxygen and nitrogen species from inflammatory cells, (2) dietary nitrosamines consumed by endemic populations, (3) secretion by the fluke of mitogens into the biliary tree [5]. Co-culture of O. viverrini adult worms with mouse fibroblasts (NIH-3T3) where parasites and cells are separated by a porous membrane results in cell proliferation [4]. Here we show that soluble ES products, in the absence of live O. viverrini parasites, and recombinant Ov-GRN-1e cause proliferation of mouse fibroblasts and a human CCA cell line, and that proliferation caused by ES products can be blocked with anti- Ov-GRN-1e antibodies. This data implies that Ov-GRN-1 is perhaps the major mitogenic factor in ES, and this protein contributes to the development of an environment that is conducive to CCA.
The GRN protein family is found in a diverse range of organisms including bacteria, plants and animals. Our phylogenetic analysis of the protein family suggested that a majority of GRN proteins do not form clades based on taxonomic groupings but rather group according to protein functions. The individual GRN domains from human PGRN form distinct clades with homologues from other species, supporting the notion that these proteins have evolved to perform distinct functions in different organisms, and furthermore, individual GRN domains released after processing of the multi-domain PGRN have also evolved to perform discrete functions. A range of organisational archetypes are seen within the family, ranging from single GRN domains behind a secretory signal peptide, as seen in earthworms and the liver flukes, to multi-homodomain PGRNs and even single GRN domains fused to other protein domains [7].
The structure of GRN is unique, although it can be partially superimposed on the 3-dimensional fold of epidermal growth factor (EGF), despite the absence of primary sequence identity [30]. Furthermore, Ov-GRN-1 and human PGRN, like EGF, triggers similar signalling cascades, including the MAPK pathway [31]. When NIH-3T3 fibroblasts were co-cultured with O. viverrini adult worms without serum, only mRNAs associated with EGF and TGF-beta signalling pathways were significantly upregulated, further supporting a role for Ov-GRN-1 in parasite-induced proliferation and downstream signalling of host cells [32]. Granulin-induced cell proliferation can result in upregulation of EGF family members, such as VEGF, which could account for upregulation of genes involved in the EGF pathway [31].
Ov-GRN-1 was expressed at very low levels in lepidopteran (Sf9) cells, limiting more thorough investigation of this form of the protein. By contrast, Ov-GRN-1 was expressed at high yield in E. coli and could be refolded into an active form that induced proliferation at nanomolar concentrations. This is the first report, to our knowledge, of a secreted growth factor from a parasite that induces proliferation of host cells. This is also the first report of functional recombinant expression of a single domain granulin. Tolkatchev expressed all 7 granulin domains individually from human PGRN but only three, granulins A, C and F, appeared to adopt at least partially correct fold and induced growth [33]. Despite refolding denatured Ov-GRN-1e to generate a functional recombinant protein that induced cell proliferation, the majority of functional recombinant protein passed through a 3 kDa cut-off membrane. If Ov-GRN-1 does indeed adopt a similar fold to carp granulin (Figure 2C), the super helical structure held tightly together by 6 disulphide bonds might well pass through a 3 kDa membrane (Figure 7). Furthermore, the apparent molecular weights of native human granulins purified from leukocytes range from 1.7–3.2 kDa and would likely pass through a 3 kDa membrane, as we observed here with functionally active Ov-GRN-1 [34]. The low nanomolar activity displayed by refolded Ov-GRN-1 is comparable to the activity of purified human granulins [15].
GRN is associated with many aggressive cancers. It is over-expressed in human liver [14],[19], renal [35], breast [31],[36],[37], bladder [16] and brain [15] tumors. It may promote cancer progression by stimulating angiogenesis, suppressing anoikis (a form of apoptosis), promotion of tumor invasion and anchorage independence, all of which support tumor expansion in the unfavourable interstitial environment [7],[16],[17]. Preventing the activity of PGRN in a range of tumor types, either through gene silencing or antibody neutralization, reduces or entirely inhibits tumor progression [18]. Transfection of fibroblasts with PGRN induces serum independent proliferation but does not transform them into neoplastic cells, suggesting that the protein is probably not oncogenic by itself, but over-expression of PGRN in the SW-13 non-malignant adrenal carcinoma cell line made it highly tumorigenic [38]. Indeed, PGRN is a therapeutic target for liver cancer, particularly HCC. An anti-PGRN monoclonal antibody inhibited tumor growth in vivo in nude mice transplanted with human HCC [19]. The anti-PGRN antibody also inhibited growth of hepatoma cells but had no significant effect on normal liver cells, and inhibited the growth and proliferation of established tumors via the p44/42 MAPK and Akt pathways. These findings demonstrate that GRN is an important factor in the initiation of liver cancer and the migration of cancerous cells. We showed here that Ov-GRN-1 signals via the MAPK pathway, further accentuating the potential role of this parasite protein in the initiation of CCA in people with chronic opisthorchiasis.
Increased PGRN expression has not been reported in CCA, however, when gene expression profiles from intrahepatic CCA associated with or without O. viverrini were compared, genes associated with growth factor signalling were the most highly upregulated ontology in the non-fluke associated CCA, whereas genes involved in xenobiotic metabolism were the most highly upregulated genes in fluke associated CCA [39]. It is intriguing that genes involved in growth factor signalling pathways were selectively upregulated in non-fluke associated CCA. This prompts the speculation that Ov-GRN-1 causes excessive proliferation and migration of pre-cancerous and cancerous cells in the bile ducts of infected people, obviating the necessity for local upregulation of the host growth factors and associated signalling molecules during tumorigenesis.
Why O. viverrini secretes such a potent growth factor that acts on host cells is unclear. One potential role for fluke GRN is in the wound repair. Inflammatory cells secrete peptides derived from PGRN [34], and PGRN mRNA is highly induced in dermal fibroblasts and epithelial cells following transcutaneous puncture wounds [8]. Furthermore, recombinant PGRN increased the accumulation of inflammatory cells, blood vessels and fibroblasts at puncture sites, implying a direct role as a wound-healing growth factor [9]. O. viverrini adult worms grasp the bile duct wall with their suckers and feed on the biliary cells, often severely damaging the epithelium. Additional inflammation occurs as a result of the local immune response to resident worms (reviewed in [5]). Ov-GRN-1 might therefore play a role in wound repair at and around the feeding site to minimize the pathology that the parasite causes to the host. Another potential role for Ov-GRN-1 is in the “farming” of host cells for nutritional purposes. By promoting growth of cells at the feeding site, the parasite is ensured of a steady supply of nutrients. Blood-feeding leeches secrete a GRN that inhibits thrombin activity [40], and Ov-GRN-1 might also perform a similar function to interfere with clot formation while feeding.
Like some other O. viverrini proteins, Ov-GRN-1 was identified on the surface of and inside host biliary epithelial cells. O. viverrini ES products adhere to and are internalised by hamster biliary epithelial cells in the first order bile ducts as well as the small extra-hepatic bile ducts where the parasite is too large to reside [41]. Until now, only one ES product that is internalized by host cells had been identified - thioredoxin peroxidase [42]. The mechanism of uptake of O. viverrini ES components by host cells is unknown. With a related fluke, Schistosoma japonicum, fluke glutathione transferase (GST) is translocated from the medium into a variety of mammalian cell types via an endocytotic pathway involving clathrin-coated pits [43]. Like most helminth parasites, O. viverrini secretes a GST (J. Mulvenna et al., unpublished), which along with other ES proteins (such as Ov-GRN-1), might enter host cells via a similar endocytotic mechanism. Translocation of O. viverrini proteins into host biliary epithelial cells is particularly important due to the carcinogenic nature of this parasite and the putative roles that internalised ES play in transforming host cells [5]. Helicobacter pylori delivers the CagA protein into gastric epithelial cells where it interacts with a kinase involved in cell polarity, resulting in disorganization of gastric epithelial architecture, inflammation and carcinogenesis [44]. Translocation of O. viverrini ES products such as Ov-GRN-1, into biliary epithelial cells might likewise interfere with signalling, promoting carcinogenesis. Moreover, liver fluke ES products inhibit apoptosis (B. Sripa, unpublished; [45]), further contributing to a tumorigenic environment.
Despite the deployment of mass drug administration programs throughout Thailand, opisthorchiasis is still a major public health concern, and the prevalence of the infection in some areas is increasing [5],[46]. Like other neglected tropical diseases, an integrated control program is required to have a lasting impact on reducing transmission and disease burden. To this end, a vaccine for opisthorchiasis is desperately needed to reduce worm burdens and minimize pathology. A recombinant vaccine based on Ov-GRN-1 is particularly attractive because of the potential role of this protein in establishing a pro-tumorigenic environment in the bile ducts. Such a vaccine would therefore have a major impact on reducing both parasite burdens and the incidence of CCA, the most prevalent and fatal of the liver cancers in north-east Thailand.
Hamsters used in this study were maintained at the animal research facility of the Khon Kaen University Faculty of Medicine; all work was conducted in accordance with protocols approved by the Khon Kaen University Animal Ethics Committee. Mice used in this study were housed at the Queensland Institute of Medical Research (QIMR) animal facility; all work was conducted in accordance with protocols approved by the QIMR Animal Ethics Committee.
O. viverrini metacercariae were obtained from naturally infected cyprinoid fish in Khon Kaen province, Thailand. The fish were digested with pepsin-HCl, washed and used to infect hamsters (Mesocricetus auratus) by stomach intubation. Adult O. viverrini worms were recovered from bile ducts of euthanized hamsters infected for 3 months. Somatic adult worm extract (SAE) was prepared from frozen and homogenized adult worms resuspended in PBS with a cocktail of protease inhibitors covering serine, aspartic, cysteine and metallo-proteases (protease inhibitor cocktail set #5, Roche). ES products were prepared from live adult worms washed in antibiotics and incubated in modified RPMI-1640 (Invitrogen) at 37°C/5% CO2. Supernatant containing the ES products was harvested daily for 7 days [41]. The supernatant was concentrated 20-fold to 100–300 µg/ml with 3 kDa Jumbosep spin concentrators (Pall) and aliquoted for storage at −80°C.
Ov-GRN-1 was identified in the ES products of adult O. viverrini using liquid chromatography tandem mass spectrometry (LC-MS/MS). The proteins present in ES products were identified by trypsin digestion followed by a combination of off-gel electrophoresis and LC-MS/MS as described by us for the analysis of ES products from the hookworm, Ancylostoma caninum [47]. For multiple reaction monitoring (MRM), recombinant Ov-GRN-1 in water and O. viverrini ES were digested as described [47]. A Dionex 3000 HPLC system (Dionex) was used to perform reversed phase separation of the samples using a C18 300A column (150 mm×2 mm) with a particle size of 5 µm (Phenomenex). Twenty microliter aliquots of samples were dissolved in 5% formic acid (aq) and injected onto the HPLC column. The mobile phase consisted of solvent A (0.1% formic acid (aq)) and solvent B (90/10 acetonitrile/0.1% formic acid (aq)). Tryptic peptides were eluted using a gradient elution programme of 0–40% B in 40 min, 40–80% B in 10 min and finally a 5-min hold at 80% B, followed by a return to 0% B for a 10-min equilibration. The flow rate was 250 µl/min. Eluate from the RP-HPLC column was directly introduced into the TurboIonSpray source. Mass spectrometry experiments were performed on a hybrid quadrupole/linear ion trap 4000 QTRAP MS/MS system (Applied Biosystems). All analyses were performed using MRM, Information Dependant Acquisition Initiation Enhanced Product Ion experiments using both the triple quadrupole and the linear ion trap acquisition modes. Analyst 1.5.1 software was used for data analysis. The acquisition protocol to provide mass spectral data for both identification and characterization involved monitoring the HPLC eluant using MRM scans; ions over the background threshold of 200 counts per second were subjected to examination using the enhanced resolution scan to confirm charge states of the multiply charged molecular ions. The most and next most abundant ions in each of these scans with a charge state of +2 to +4 or with unknown charge were subjected to collision induced disassociation using a rolling collision energy dependent upon the m/z and the charge state of the ion. An enhanced product ion scan was then used to acquire the product ion spectrum. The 4000 QTRAP equipped with a TurboIonSpray Source was operated in the positive electrospray ionization mode.
Sequences were edited and analysed with assistance from the MacVector software package. Homology searches were performed using Blast search at NCBI (http://blast.ncbi.nlm.nih.gov/Blast.cgi). ORFs were analysed for signal peptides/anchors using SignalP-NN prediction and SignalP-HMM prediction at http://www.cbs.dtu.dk/services/SignalP/. The O. viverrini cDNA encoding for GRN was termed Ov-grn-1 and was submitted to GenBank under accession number FJ436341. The structural architecture of GRN family members was obtained from entry IPR000118 at the Interpro 18.0 database [48]. Prediction of potential glycosylation sites was determined using the YingOYang server [49].
Using version 3 of MODELLER, the three-dimensional structure of Ov-GRN-1 was predicted based on comparative modelling to carp granulin [23]. The covalent geometry of the modelled structure was in agreement with the template structures with all but one of the residues occupying the allowed regions of the Ramachandran plot. The quality of the stereo-chemical structures of the models was determined using PRO-CHECK-NMR [24]. Pymol was used to view the homology models (http://www.pymol.org)
The phylogenetic relationship of Ov-GRN-1 with other GRN family members was inferred using a neighbor joining analysis in PAUP beta version 8.0 for Macintosh. Bootstrap values were determined from 1000 replicates. Where bootstrap values were below 50%, clades were collapsed to form polytomies. Where multiple GRN domains were observed within one PGRN protein (e.g., vertebrates and schistosomes), individual GRN domains sharing the greatest identity with Ov-GRN-1 were selected and numbered 1–3 in order of their identities. The individual GRN domains of human PGRN, however, have been alphabetically designated A–G in the order GFABCDE, based on their description in the literature.
Extraction of O. viverrini RNA and subsequent reverse transcription PCR (RT-PCR) was carried out as described [50], with minor modifications. Total RNA from each developmental stage of O. viverrini was extracted with Trizol (Invitrogen) according to the manufacturer's instructions. Contaminating genomic DNA was removed by treatment of RNA with DNase I (Promega). For RT-PCR, first-strand cDNA was synthesized from 1.0 ug of total RNA using avian myeloblastosis virus reverse transcriptase (Promega) and an oligo (dT) primer at 42°C for 60 min. A 1.0 µl aliquot of the cDNA was amplified using primers specific for the control beta-actin mRNA (Forward CGAGGTATCCTCACCCTCAA, Reverse GCGACTCGCAACTCATTGTA) and the target Ov-grn-1 mRNA (Forward CGCGCGCCATGGATACTTTGCAGCCAATT, Reverse GCGCGCCTCGAGTGCGACCTTTCGAGCGTT) based on the following conditions: 30 sec denaturation at 94°C, 30 sec annealing at 55°C, and 30 sec extension at 72°C for 30 cycles. Control RT-PCR reactions were performed without reverse transcriptase to ensure that amplified products were derived from cDNA and not contaminating genomic DNA. PCR products were sized by electrophoresis through 1% agarose and visualized under UV light after staining with ethidium bromide.
Ov-GRN-1 was expressed in both bacterial and insect cell expression systems. The complete ORF minus the predicted signal sequence was amplified using Expand polymerase (Roche) from an adult O. viverrini cDNA library [6] using primers described below and cloned into the plasmid pMIB/V5-His (Invitrogen) for insect (Spodoptera frugiperda) cell expression or cloned into the NdeI and XhoI sites of the pET41a vector (Novagen) for E. coli expression, thereby removing the GST fusion tag but retaining the 6×His tag and allowing for native N-terminal protein expression. Primers for insect cell expression were: pMIB F3 HindIII CGCGCGAAGCTTAATGGATACTTTGCAGCCAATT; pMIB R3 XbaI GCGCGCTCTAGATGCGACCTTTCGAGCGTT. Plasmid preparation, cell transfection, colony selection and growth of Sf9 cells (Invitrogen) was as previously described [51]. Primers for E. coli expression were: pet41 NcoI F7 CGCGCGCCATGGATACTTTGCAGCCAATT; pet41 XhoI R7 GCGCGCCTCGAGTGCGACCTTTCGAGCGTT. Plasmids were prepared using standard techniques and used to transform E. coli cell lines (BL21, rosetta, rosetta-gami cells – Novagen) followed by selection with kanamycin and other appropriate antibiotics according to the manufacturer's instructions. Cells were grown in LB medium at temperatures between 16–37°C in 1 L Schott bottles at 220 rpm. Cultures were induced with 1 mM IPTG upon reaching an OD600 of 0.5 and grown overnight before harvesting the cell pellet.
Since both recombinant proteins (derived from E. coli and Sf9 insect cells) contained C-terminal 6×His tags, the purification procedures included immobilized metal ion affinity chromatography (IMAC). Purification was undertaken using an AKTA basic purification system (GE) at 4°C. Henceforth we refer to the E. coli-derived recombinant protein as Ov-GRN-1e and Sf9-derived protein as Ov-GRN-1s. For Ov-GRN-1s, 3 L of culture supernatant was passed across a 5 ml Hitrap HS HP cation exchange column (GE) with a gradient of 0–1 M NaCl over 10 column volumes. Fractions enriched for recombinant protein were detected using an anti-V5 affinity tag antibody (Invitrogen). Further purification of these fractions was achieved by a subsequent IMAC step; protein was bound to 1 ml Ni-NTA resin (Qiagen) in 10 mM imidazole/sodium phosphate (pH 8) and washed with 10 column volumes of 20, 40 and 60 mM imidazole, followed by elution in 250 mM imidazole. Fractions containing recombinant protein were subjected to a second round of IMAC as above, followed by concentration to 1 mg/ml and buffer exchange into PBS using 3 kDa microsep spin columns (Pall). For purification of Ov-GRN-1e, 3 g of E. coli cell pellet was resuspended in 30 ml binding buffer (50 mM Tris-HCl, 300 mM NaCl, 0.1% Triton X-100) followed by three disruption cycles through a chilled French press at 16–18000 psi. The sample was centrifuged at 4000 g and the pellet was dissolved in 6 M urea in nickel (Ni)-NTA binding buffer with 40 mM imidazole overnight at 4°C with gentle mixing. The supernatant was collected by centrifugation as above and purified by denaturing IMAC, with 6 M urea in all buffers, over a 5 ml His-trap Ni-IDA column (GE). The column was washed with 100 mM imidazole and recombinant protein eluted with 500 mM imidazole/6 M urea. The eluate was concentrated to 20 mg/ml using 3 kDa 15 ml Amicon Ultra centrifuge concentration devices (Millipore) and refolded by drop wise addition to a 20-fold greater volume of a range of refolding buffers [52]. The soluble material was buffer-exchanged into PBS using a PD10 column (GE).
Ov-GRN-1e and GRN-1s were adjusted to 0.4 mg/ml and diluted 1∶1 with Freund's adjuvant and emulsified. Twenty micrograms of protein (100 µl of protein∶adjuvant) was injected subcutaneously into each of four BALB/c female mice every two weeks, using Freund's complete adjuvant for the first immunization and Freund's incomplete adjuvant for the second and third immunizations. Two weeks after the final immunization, mice were euthanized, blood collected by cardiac puncture, and sera recovered from the clotted blood. Control sera were obtained from mice that were either (1) unimmunized or (2) immunized with an irrelevant control protein (recombinant Na-GST-1 glutathione-S-transferase from Necator americanus) [53]. Sera were pooled and diluted 1∶20 with PBS for affinity purification of IgG on Hitrap protein G (GE) at 4°C using an AKTA basic FPLC. Eluted fractions were concentrated with 30 kDa nanosep spin concentrators (Pall), buffer exchanged into PBS and stored at −80°C.
Denaturing and native polyacrylamide gel electrophoresis (PAGE) was performed using 15% gels. Proteins were electro-transferred to Biotrace nitrocellulose membranes (Pall), which were then sliced into 3 mm wide strips. Membranes were blocked in 5% skimmed milk powder (SMP)/PBS with 0.05% Tween 20 (PBST) for 60 min at room temperature (RT) with gentle shaking and probed overnight with 8 µg/ml mouse IgG diluted in 2% SMP/PBST at 4°C. Subsequently, membranes were washed (3×10 min each) in PBST at RT followed by probing with horse radish peroxidase (HRP)-conjugated goat anti-mouse IgG (Zymed Laboratories) diluted 1∶300 in 2% SMP/PBST at RT with gentle rocking for 60 min. After washing to remove the secondary antibody, colorimetric signals were developed in the presence of hydrogen peroxide and diaminobenzidine (DAB) (Pierce).
O. viverrini adult worms or liver tissue from hamsters infected with O. viverrini (weeks 12–16 weeks) were fixed and cut by microtome into sections of 4 µm [41]. The sections were deparaffinised in xylene, hydrated in a series of ethanol and distilled water, respectively. Endogenous peroxidise was eliminated by incubating sectioned tissues in 5% H2O2 in methanol for 30 min, after which the sections were rehydrated in water and PBS. Non-specific staining was blocked by incubation in 5% normal mouse serum in PBS for 30 min. The sections were probed with pooled mouse purified IgG diluted 1∶100–1∶500 (v/v) in PBS and incubated overnight at 4°C. After rinsing 3×5 min with PBS the sections were incubated with horseradish peroxidase-conjugated goat anti-mouse IgG (Zymed Laboratories) for 1 h. Sections were rinsed with PBS 2×10 min, after which the slides were developed with DAB. The sections were counterstained with Mayer's haematoxylin, dehydrated, cleared in xylene and mounted in Permount® (Cen-Med). The sections were examined by light microscopy and images captured with a digital camera.
ES products, recombinant proteins and IgG antibodies that were included in cell culture were filtered under sterile conditions with 0.22 µM syringe filters (Pall). Dilutions were carried out in sterile PBS in Twintec 96 well plates (Eppendorf). Samples were prepared so that 20 µl of ES protein or antibody in sterile PBS was added to 100 µl of cell culture media. Protein concentrations described hereafter refer to the final concentration in cell culture after dilution in media. All assays were conducted either in triplicate or duplicate, as specified in figure legends. Controls were relevant for the sample tested: expression matched recombinant proteins were included to assess effects of recombinant granulin – a hookworm protein, Na-ASP-2, expressed in insect cells [51] served as the control protein for Ov-GRN-1s; Sm-TSP-2 EC2 expressed in E. coli [54] served as the control protein for Ov-GRN-1e. Antibodies against recombinant proteins for use in cell culture were purified as above. The Erk1/2 (MAPK) kinase inhibitor, U0126 (Cell Signalling Technology) was used at a final concentration of 10 µM in some cell cultures to assess its ability to block signalling induced by ES products or recombinant Ov-GRN-1.
NIH 3T3 mouse embryonic fibroblast cells (ATCC) were maintained as specified by ATCC protocols. Briefly, cells were maintained with regular splitting using 0.25% trypsin every 2–5 days in DMEM (Sigma) with 10% bovine calf serum (BCS) and 1×antibiotic/antimycotic (Invitrogen) at 37°C under 5% CO2. To assess the effects of ES on cell growth, multiple culture conditions were investigated in 96 well plates including seed densities (1,000–10,000 cells/well), cell attachment times before sample addition (3 h, 6 h, 16 h), BCS concentration (0, 2, 5 and 10%), ES concentration (10–80 µg/ml) and duration of culture after sample addition (1–3 days). The ability of live O. viverrini flukes to stimulate cell proliferation was assessed using a non-contact co-culture technique as described [4]. KKU-100 is a cell line derived from a human CCA and was maintained as described [55]. Cell proliferation was determined using two different approaches. First, cell numbers were assayed using the WST-1 cell proliferation reagent (Roche) as per product manual. Briefly the procedure involves incubating the cells with 10 ul of WST-1 for up to four hours at 37°C. During this period viable cells, which contain mitochondrial dehydrogenases, convert WST-1 to a water soluble formazan dye which has a peak absorbance at 450 nm which is measured on a Benchmark Plus plate reader (BioRad) and data were converted into cell numbers per well via comparison to standard curves.
Cell proliferation was also assessed by counting cell numbers in real time using a xCELLigence system and E plates (Roche) which monitors cellular events in real time by measuring electrical impedance across interdigitated gold micro-electrodes integrated on the bottom of tissue culture plates. The impedance measurement provides quantitative information about the biological status of the cells, including cell number, viability, and morphology [27]. Cell culture conditions tested were the same as those tested with the WST-1 dye (2% BCS, 20 µg/ml ES, 20 µg/ml IgG).
|
10.1371/journal.pgen.1000049 | Cell-to-Cell Stochastic Variation in Gene Expression Is a Complex Genetic Trait | The genetic control of common traits is rarely deterministic, with many genes contributing only to the chance of developing a given phenotype. This incomplete penetrance is poorly understood and is usually attributed to interactions between genes or interactions between genes and environmental conditions. Because many traits such as cancer can emerge from rare events happening in one or very few cells, we speculate an alternative and complementary possibility where some genotypes could facilitate these events by increasing stochastic cell-to-cell variations (or ‘noise’). As a very first step towards investigating this possibility, we studied how natural genetic variation influences the level of noise in the expression of a single gene using the yeast S. cerevisiae as a model system. Reproducible differences in noise were observed between divergent genetic backgrounds. We found that noise was highly heritable and placed under a complex genetic control. Scanning the genome, we mapped three Quantitative Trait Loci (QTL) of noise, one locus being explained by an increase in noise when transcriptional elongation was impaired. Our results suggest that the level of stochasticity in particular molecular regulations may differ between multicellular individuals depending on their genotypic background. The complex genetic architecture of noise buffering couples genetic to non-genetic robustness and provides a molecular basis to the probabilistic nature of complex traits.
| Although most inter-individual phenotypic variabilities are largely attributable to DNA differences, a wealth of examples illustrate how a single biological system can vary stochastically over time and between individuals. Identical twins are not identical, and similarly, clonal microbial cells differ in many aspects even when grown simultaneously in a common environment. Using yeast as a model system, we show that a population of isogenic cells all carrying genotype A showed higher cell-to-cell heterogeneity in gene expression than a population of isogenic cells of genotype B. We considered this level of intra-clonal heterogeneity as a quantitative trait and performed genetic linkage (on AxB) to search for regulators of it. This led to the demonstration that transcriptional elongation impairment increases stochastic variation in gene expression in vivo. Our results show that the two levels of inter-individual diversity, genetic and stochastic, are connected by a complex control of the former on the latter. We invite the community to revisit the interpretation of incomplete penetrance, which defines cases where a mutation does not cause the associated phenotype in all its carriers. We propose that, in the case of cancer or other diseases triggered by single cells, such mutations might increase stochastic molecular fluctuations and thereby the fraction of deviant cellular phenotypes in a human body.
| Two fascinating area of research on gene expression have been conducted intensively and independently during the past couple of years. A large community of geneticists has contributed to the identification of genetic sources underlying expression differences between individuals. Such expression Quantitative Trait Loci (eQTL) were first mapped in maize[1], yeast[2] and mouse[3] and consecutively identified in many organisms including worms[4], A. thaliana [5] and humans[6],[7]. All these studies shared three important conclusions: gene expression levels differ greatly among individuals of a species, their genetic control is complex, and despite the high number of statistical tests required, genetic mapping of regulators is feasible on a genome×transcriptome scale. In addition, promising methods have emerged to extract causal relationships among molecular regulations[8]–[10], illustrating how expression data can power genetic linkage or association studies. Recently, the genetics of gene expression appeared even more complex when discovering the high degree of variation in human transcript isoforms [11]. This complexity of molecular regulations, which very likely underlies the genetics of complex traits, is now anticipated and integrated in many designs. However, like the large majority of molecular regulations described to date, these observations were made on samples of many (104–109) cells and therefore reflect only averages of cellular states. This limitation can be very frustrating when studying traits such as cancer that can emerge from a single or very few cells.
Simultaneously, another large community of scientists from various disciplines has been investigating the sources and properties of stochastic fluctuations in gene expression. These investigations were powered by the development of single-cell reporter assays. Following previous terminology, we will refer here to noise in gene expression as the stochastic variation of a protein concentration among isogenic cells, grown homogeneously in a common environment. This noise was demonstrated to contribute to non-genetic cellular individuality[12]–[16]. Although non-deterministic fluctuations in gene expression can be detrimental to cellular physiology, they can also provide a mechanism of single-cell memory[17]–[19] and shape differentiation during development[20]. Notably, high noise was observed in old mice hearts suggesting that age-related health decline could result from such stochastic fluctuations[21]. Genetic sources of noise in gene expression were also investigated. So far, the list of factors shown experimentally to contribute to noise includes the SWI/SNF, INO80 and SAGA chromatin modification complexes[22], TATA-box mutations[22],[23], MAP Kinases implicated in the response to yeast pheromones[24], the Swi4 transcriptional activator[25], DNA topology[13] and ribosomal activity in bacteria[26]. This list will very likely increase dramatically in the near future as investigations of single-cell expression levels are becoming more and more popular. In addition, the topology of gene regulatory networks has clearly been shown to drive various levels of instabilities, for example via the presence/absence of functional feedback loops[17].
We present here a study bridging these two fields of investigations, by considering noise in gene expression as a quantitative trait. We quantified noise of a representative reporter system in various strains of S. cerevisiae and found reproducible differences among strains. Genetic segregation of noise values revealed a complex genetic control, and Quantitative Trait Loci mapping allowed the identification of three loci modulating noise levels. One locus led to the identification of transcriptional elongation as an additional source of noise. Based on these observations from a yeast model, we propose a new interpretation of the incomplete penetrance observed for common traits that are triggered by single cells in higher eukaryotes.
To investigate the natural genetic diversity of noise in the expression of a representative gene, we integrated in the genome of five distant S. cerevisiae strains a reporter construct where the green fluorescent protein (GFP) was regulated by the inducible promoter of the MET17 (YLR303W) gene. The strains used were three unrelated laboratory strains (S288c, FL200 and CEN.PK), a wine strain from California (RM11-1a), and a wine strain from Japan (Y9J_1). In each case the construct was integrated at the same HIS3 chromosomal locus. We then quantified the level of expression in individual living cells by flow cytometry. Figure 1A shows representative experiments where 15,000 cells were recorded for each background after two hours of moderate induction. We found that although mean induction was similar between backgrounds, the variance of gene expression level differed. This observation was reproduced when the experiments were repeated at various dates (Figure 1B). This suggested the presence of genetic variation that might control noise without necessarily affecting mean expression of the cell population. To see if the difference in noise between S288c and RM11-1a was specific to the chromosomal environment of the HIS3 locus, we integrated the same reporter system at the LYS2 locus located on another chromosome (Figure S1). Noise and mean expression values were comparable to the results obtained when targeting HIS3, showing that the difference in noise between the two strains could not be accounted for by differences at the integration locus only.
If strain-to-strain difference in noise levels is under genetic control, it should be heritable. To determine if this was the case, we integrated the PMET17-GFP construct at the HIS3 genomic locus of 61 segregants issued from a cross between S288c and RM11-1a, two backgrounds displaying different noise levels. Noise was estimated from triplicate experiments for each segregant. This showed that noise segregated as a quantitative phenotype, with evidence of a polygenic control (Figure 1C–D). Heritability was high (81%) and the continuous, Gaussian-like distribution of noise values among segregants excluded simple Mendelian inheritance. In addition, a few segregants showed noise values outside the range of parental values (transgression), suggesting segregation of alleles with opposite effects. Importantly, mean expression (the average fluorescence of the population of cells) also segregated continuously, and the two traits (noise and mean) were correlated (R2 = 0.51, P = 5×10−11 from linear regression). This scaling between mean expression and noise level is consistent with previous observations[14],[27],[28]. In the case of our genetic design, this scaling of segregant values indicate the presence of genetic loci acting on both mean and noise, although mean values did not differ between the parental backgrounds. This apparent discrepancy can be explained by alleles with opposite effects that compensate mean expression in the parental strains (higher transgression for mean than for noise).
To examine further the natural genetic segregation of noise, we analyzed a cross from another pair of unrelated backgrounds. We crossed GY43 with GY44, two strains carrying the HIS3:PMET17-GFP insertion and derived from FL200 and CEN.PK, respectively. Random spores were generated and were considered further only if they were auxotroph to uracil, to avoid the presence of diploid contaminants. Noise was measured in 55 of these spores, and the distribution obtained also showed high heritability (88%) with a continuous genetic segregation and evidence of transgression (Figure 1E–F). In addition, noise values of GY43xGY44 segregants were enriched for low levels and were not centered at the mid-parental value. This is probably not a bias from our selective choice of ura3 segregants because average noise was also globally low among spores of dissected tetrads (Figure S2). This asymmetry towards low noise is more likely due to the presence of interacting alleles, a particular combination of which being required to confer high noise (epistasis).
We then sought to map genetic variations underlying noise differences between S288c and RM11-1a, which we did by two methods. Firstly, using the noise values of the 61 segregants from S288cxRM11-1a and their genotypes at 3042 marker positions[29], we screened the genome for Quantitative Trait Loci (QTL). Two QTL were found (position 79091 on chromosome III and position 449639 on chromosome XIV) at a genome-wide significance of 1% (Figure 2A). Secondly, we introgressed the high-noise phenotype of RM11-1a into the S288c background, and searched for alleles that had been conserved from RM11-1a in the resulting strains (see Materials and Methods). This approach identified a region on chromosome V (from position 116530 to 207819) as a candidate region for conferring high-noise level (Figure 2B). To validate or refute this locus as a QTL, we backcrossed GY157, the S288c×RM11-1a segregant showing highest noise, with an S288c derivative. Fifty five random spores from this cross were analyzed by flow cytometry to quantify their level of HIS3:PMET17-GFP noise. We took advantage of the presence of the ura3Δ0 auxotrophic marker within the region of interest to genotype the 55 spores by plating them on URA-plates. A significant linkage was found between these genotypes and noise levels (Wilcoxon-Mann Whitney test, P = 3.5×10−3) (Figure 3C), which validated the locus as a third QTL. The three QTL identified showed the following characteristics: Firstly, in all three cases, the molecular control of noise involves trans-regulations (a polymorphism in one gene affecting noise level of another gene) because none of the QTL were located at or near the HIS3 integration site nor the MET17 endogenous regulatory region. Secondly, QTL1 and QTL2 but not QTL3 were also in genetic linkage with the mean expression levels of the samples (Figure 3). Consistently, QTL1 was already detected as an expression QTL (eQTL) locus controlling MET17 mRNA levels in a previous study where only mean expression was measured[29]. This indicated that regulatory variation could scale noise levels by acting on mean expression, raising the possibility that other eQTL identified by ‘genetical genomics’[30] are likely to influence noise as well. Thirdly, and surprisingly, the effects of QTL1 and QTL2 were opposite to the effects expected from the parental difference: alleles from the high-noise background RM11-1a conferred low noise (Figure 3A–B). This was consistent with the transgressive segregation visible on Figure 1C and it supported the presence of additional QTL (such as QTL3) where RM11-1a alleles conferred high noise. Finally, QTL3 effect was extremely low in the panel of S288cxRM11-1a segregants (P = 0.4 from linear regression). From these observations, we conclude that the difference in noise between S288c and RM11-1a backgrounds can not be attributed to one or a few loci but rather results from the cumulative effects of numerous QTL, several of which remain to be identified.
The presence of ura3Δ0 at QTL3 prompted us to test if this mutation was responsible for noise modulation. When introduced in the S288c background, a significant increase in HIS3:PMET17-GFP noise was observed (Figure 4A–B). Consistently, restoring wild-type URA3 in the resulting mutant or in RM11-1a significantly reduced noise (Figure 4A–C), and another null allele (ura3-52) could also increase noise (Figure S3A), as well as treatment with 6-azauracil, a drug inhibitor of the URA3 gene product (Figure 4D). Since random spores of the FL200×CENPK cross mentioned above displayed low noise despite bearing the ura3-52 mutation, we examined additional spores from tetrads and found that, as expected, Ura+ spores from this cross displayed even lower noise (Figure S2). Surprisingly, increasing the concentration of uracil in the culture medium did not reduce noise of a ura3Δ0 strain (Figure S3B). This might be due to limiting steps of the import mechanism. Finally, the ura1Δ and ura2Δ mutations were also found to increase noise levels (Figure S3C). Altogether, these observations validated ura3 as a responsible gene for QTL3 with ura3Δ0 accounting for most (74%) of the locus effect seen in segregants (Figure 3C and 4A). So if additional noise regulators resided at QTL3, we expect their contribution to be minor. The ura3Δ0 allele is not natural but was introduced in RM11-1a for laboratory purposes unrelated to this study[2]. However, null ura3 alleles exist in nature: ura3-52 results from a Ty transposable element insertion[31], and when searching the Saccharomyces Genome Resequencing Project[32] we found three additional severe mutations (G->GA, G->GA, and TTG->TAG(stop) at 183, 219 and 94 nucleotides from ATG, respectively) in two unrelated natural isolates (NCYC361 from an Irish brewery and UWOPS87_2421 from a cladode in Hawaii). Also, ura3 mutations are not the sole source of natural genetic variation in noise, since high noise was found in the Y9J_1 background (a prototrophic strain with functional URA3), and since ura3Δ0 accounted for only 37% of the total noise difference between S288c and RM11-1a (Figure 4C and Materials and Methods).
Inhibition of uracil synthesis is known to reduce the intracellular pool of nucleotides available for RNA synthesis and this shortage is known to affect transcriptional elongation[33]. To directly test if transcriptional elongation was involved in the control of noise, we measured noise levels in a dst1Δ mutant strain lacking TFIIS activity. A dramatic increase of noise was observed, with no detectable difference in mean expression (Figure 4D). This increase was suppressed when the mutation was complemented by integrating the wild-type DST1 (YGL043W) gene at the HO locus (Figure 4E). Even higher noise levels were obtained when dst1Δ cells were treated with 6-azauracil (Figure 4D), highlighting the gradual noise increase with gradual transcriptional elongation defects. To see which of several known partners of elongating RNA PolII were involved in noise modulation, we measured PMET17-GFP noise in strains lacking specific elongation factors (Figure 4E). A pronounced noise increase was observed in spt4Δ mutant, and in mutants lacking the Leo1p or Cdc73p subunits of the Paf1 complex. This suggested that recruitment of Paf1 to elongating RNA PolII (a step requiring Spt4p[34]), was involved in noise control. However, full integrity of the Paf1 complex was not essential since noise remained low in the absence of the Ccr4p subunit. Finally, noise remained low in set2Δ and eaf3Δ mutants, showing that methylation of lysine 36 of histone H3 and recruitment of Rpd3S[35] for histone deacetylation were not involved. Thus, noise appeared to be strongly connected to the facilitation of transcriptional elongation but not to the subsequent resetting of chromatin to an inactive state.
We showed that noise in gene expression can be subjected to natural genetic variation with a complex inheritance pattern in yeast. In agreement with previous studies[27],[28] we observed that natural genetic variation of noise tended to scale with the genetic control of mean expression. However, two divergent backgrounds could differ only in noise while their cross generated segregants varying in both noise and mean. This supports the presence of two classes of alleles: those acting on both traits (such as QTL1 and QTL2) and those acting specifically on noise (such as ura3 and dst1).
We demonstrated that impairing the progression of transcriptional elongation can increase the level of noise in gene expression. When elongating RNA polymerase II is stalled because of such defects, expression of the corresponding messenger in this particular cell is blocked until transcriptional initiation takes place again. It is therefore not surprising that this stalling increases stochasticity, as compared to a wild-type context where elongation can resume rapidly, and our results are consistent with a previous numerical model of elongation defects[36].
The complex genetic control of noise makes it a potentially evolvable trait. Although our study did not address whether this genetic control correlates with any adaptive mechanism, the results can be discussed in the context of selection. Living systems maintain a delicate balance between robustness and flexibility[37]. The former ensures stability of ‘normal’ physiology, and the latter provides adaptability to environmental changes. Thus, fluctuating environments might maintain flexibility. One consequence of the propagation of many alleles contributing to noise is the production of few individuals in which regulations are highly noisy, the term ‘individual’ here referring to a human being, a yeast strain or a congenic animal or plant breed. The individuals displaying high noise are likely to have reduced fitness in ‘standard’ environments but they may be readily adapted to new environmental conditions. One possible advantage provided by genetic complexity is to generate this ‘reservoir’ of individuals without perturbing the bulk of the population, because most individuals harbour only few of the alleles conferring high noise levels. However, whether evolution in fluctuating environments can shape the genetics of noise control remains to be demonstrated.
Finally we propose to revisit the interpretation of incomplete penetrance for traits that arise from one or very few cells in higher eukaryotes. Despite intense investigations on the genetic predisposition to common traits, it remains unclear why the underlying alleles express their effects in only a fraction of carriers[38]. For example, a fortunate ∼20% of women carrying BRCA2 mutations associated with high-risk of breast cancer do not develop the disease[39]. In default of any clear explanation, this incomplete penetrance is usually interpreted as the result of interactions that remain to be discovered. This assumes that causative genes manifest their effect only if the carrier is exposed to specific environmental conditions (gene×environment interactions) or if the carrier possesses particular alleles at additional genes, yet undiscovered, which unbuffer the effect of the causative gene (gene×gene interactions). This explanation probably holds for many cases of incomplete penetrance, but since the underlying interactions are currently extremely difficult to identify, their involvement generally remains hypothetical.
Many common traits such as cancer, developmental defects, autoimmunity, or infection can result from rare cellular events. Considering the huge number of cells constituting a human body, these traits can emerge from a very slight increase in the probability of such events. It is therefore possible that cases of genetic predisposition to these traits are caused by low-penetrance alleles that simply increase the chances of such events, without driving them deterministically, and therefore increase the frequency of peculiar cells. Under such a scenario, incomplete penetrance would naturally result from the probabilistic nature of the traits, without necessarily requiring complex genetic interactions.
One way to increase, even slightly, the probability of rare cellular events is to increase stochastic fluctuations in their underlying molecular mechanism. Our study showed that in yeast, natural allelic differences can influence the level of noise in a particular molecular regulation. It is likely that similar scenarios are present in higher eukaryotes. An exciting area of investigation would be to re-examine disease-predisposing alleles in terms of their probabilistic effects among single cells of the tissue they target.
The NatMX cassette was amplified from the integrative plasmid pFvL99 (kindly provided by F. van Leeuwen and D. Gottschling, FHCRC, Seattle) using primers 5′-GCAAGCGATCCGTCCTAAGAAACCATTATTTAAATGGATGGCGGCGTTAGTATC-3′ and 5′-ATCCGCTTACAGACAAGCTGTGACCGTCTCGACATGGAGGCCCAGAATAC-3′ and cloned by gap-repair recombination into pUG23 (a centromeric plasmid carrying yEGFP3[40] under the control of the MET17 promoter, from J.Hegemann, Düsseldorf, Germany) linearized at BsmBI to generate plasmid pGY6. The ScaI fragment containing replicative and centromeric sequences of pGY6 was replaced by the ScaI fragment of pFvL99 to create pGY8. To generate plasmid pGY12, the HIS3 gene of pGY6 was replaced by LYS2 flanking sequences by transforming strain BY4742[41] with pGY6 linearized at NheI with PCR fragment LYS2-UD and recovering the gap-repaired pGY9 resulting plasmid from HIS-NATR colonies. The LYS2-UD PCR product was obtained by fusing two PCR products, each obtained by amplifying genomic DNA from BY4716[41] with primers 5′-GCATCAGAGCAGATTGTACTGAGAGTGCACCATAAATTCCTAGGAAGCGGTCAGCAAGAAGAAA-3′, 5′-AATATAAGCGGCCGCTCGAGTTTATACAGTACCTTTTTGAACTTCGTC-3′ and primers 5′-TGTATAAACTCGAGCGGCCGCTTATATTCATCATGCTGCGAAGAACTA-3′, 5′-TCCTTACGCATCTGTGCGGTATTTCACACCGCATAGATCCGTCCATGTACAATAATTAAATATGAATTAGG-3′, respectively. The ScaI fragment of pGY9 containing replicative and centromeric sequences was replaced by the ScaI fragment of pFvL99 to create pGY12. For the complementation test of dst1Δ, the DST1 gene of strain BY4716 was amplified using primers 5′-GCGAGCTCTCATTTTATCGTTTTCGT-3′ and 5′-CGGAGCTCTTCTTTAGTTCTGACCGA-3′, the product was digested with SacI and cloned into the SacI site of plasmid pHO-poly-KanMX4-HO[42] to give plasmid pHO::DST1.
The strains used in this study are listed in Table S1. Plasmid pGY8 was linearized at NheI and integrated at the HIS3 locus of FL200, CEN.PK113-5D, BY4716 (isogenic to S288c), YEF1685 (a non-clumpy derivative of RM11-1a), Y9J_1 and in 61 F1 segregants from BY4716×RM11-1a described in Brem et al. 2005 to create GY43, GY44, GY51, GY53, GY445 and the set of S288cxRM11-1a HIS3:PMET17-GFP strains, respectively. At every transformation, cells were separated in three tubes just after heat shock so that recovery in YPD medium and cell division occurred independently before plating each fraction on a separate NAT plate. This way, three independent transformants were obtained each time. Plasmid pGY12 was linearized at XhoI and integrated at the LYS2 locus of BY4709 and YEF1946 to generate GY122 and GY125 strains, respectively. To introgress the RM11-1a alleles conferring high noise into a global S288c background, GY53 and BY4716 were crossed, a resulting spore with high noise but similar mean was selected and crossed with BY4719, a resulting spore with high noise but similar mean was selected and crossed with FYC20-2A, a resulting spore with high noise but similar mean was selected and crossed with BY4713, and a resulting spore with high noise but similar mean was selected and called GY159. To repeat this procedure in a totally independent way, GY51 and YEF1946 were crossed, a spore with high noise but similar mean was selected and crossed with FY67, a resulting spore with high noise but similar mean was selected and crossed with BY4712, a resulting spore with high noise but similar mean was selected and crossed with BY4715, and a resulting spore with high noise but similar mean was selected and called GY174. Thus, GY159 and GY174 theoretically contained only 6.25% of RM11-1a genome but had retained high-noise levels of the PMET17-GFP construct. The 55 spores used to validate QTL3 were obtained by crossing GY157 with BY4714. The strains used to demonstrate the effect of ura3Δ0 on noise were GY244, GY246, GY333 and GY601. GY244 and GY246 were random spores from a cross between GY51 and BY4741. GY333 was obtained by transforming GY246 with a NotI restriction fragment from plasmid HO-hisG-URA3-hisG-poly-HO described in Voth et al. [42]. GY601 was obtained by amplifying the URA3 gene of BY4716 with primers 5′-AGGGAAGACAAGCAACGAAACGT-3′ and 5′-CCAGCCCATATCCAACTTCCAAT-3′ and transforming GY53 with this product. Strain GY321 was obtained by crossing GY172 (which was a spore from GY51×BY4710) with the dst1 strain FY1671 kindly provided by F. Winston. We followed the kinetics of GY321 and GY51 growth in the physiological conditions of PMET17-GFP noise measurements and found identical growth rates (data not shown). For the complementation test of dst1Δ, the 4.6kb NotI fragment of plasmid pHO::DST1 was transformed in strain GY321 to give strain GY361. The corresponding negative control strain GY358 was obtained by transforming GY321 with the NotI fragment of the empty plasmid pHO-poly-KanMX4-HO. To test the effect of the ura3-52 mutation on noise, strains GY51 and FY1679-18D were crossed and two random spores were selected: GY241 and GY243. To test the effect of ura1Δ and ura2Δ mutations, strain GY329 was obtained by amplifying the ura1Δ::KanR mutation from the EUROSCARF strain YKL216W with primers 5′-CGGACGATAAACTTCGAAACAATTC-3′ and 5′-GGCACTTAACAATGTTTCGGAACTC-3′, and transforming strain GY51 with this amplicon; strain GY325 was obtained by amplifying the ura2Δ::KanR mutation from the EUROSCARF strain YJL130C with primers 5′-GCGTATTTTAGTATCTGGGCGTGG-3′ and 5′-CGGACCTGATGTTACCTCCTTACTG-3′ and transforming strain GY51 with this amplicon. Similarly, strains GY602 to GY608 were constructed by amplifying the deletion mutation from the corresponding EUROSCARF strain with about 400bp flanking sequence, transforming GY51 with the amplicon, and checking proper integration by PCR with at least one primer designed outside the mutagenic fragment. We verified that Y9J_1 beard a functional URA3 allele by amplifying it with primers 5′-AGGGAAGACAAGCAACGAAACGT-3′ and 5′-CCAGCCCATATCCAACTTCCAAT-3′ and transforming a ura3Δ0 strain, which led to complementation of the ura-phenotype. We also checked that ura3Δ0 and dst1Δ mutations did not change the fraction of cells in G1 by staining population of cells with propidium iodide as previously described[43], and analyzing distributions of DNA content by flow-cytometry (Figure S5).
4ml of YPD medium was inoculated with an isolated colony, and incubated overnight at 30°C with 220 rpm shaking. This starter culture was used to inoculate at OD600 = 0.1 4ml of autoclaved SD-MET medium [Yeast Nitrogen Base 6.7 g/L, Glucose 2%, Dropout Mix 2 g/L, adjusted to pH = 5.8 with NaOH] supplemented with 1 mM methionine (repressed condition). The Dropout Mix was a powder made of 2 g of uracil , 4 g of leucine, 1g of adenine, and 2 g of each of the following amino-acids: A, R, D, N, C, E, Q, G, H, I, K, F, P, S, T, W, Y, V. The culture was incubated at 30°C for exactly 3 hours with shaking, centrifuged at 1100×g for 5 minutes, and cells were resuspended in 4 ml of SD-MET medium supplemented with 50 µM methionine (moderate induction). Other methionine concentrations were tested in the experiments of Figure 4C–E (0, 20, 50, 100 and 200 µM). In the case of 6-AU treatments, the drug was added at this step to a final concentration of 100 µg/ml. In the case of increased uracil concentrations, uracil was added at both repressed and induced steps from a 2 mg/ml stock solution. The induced culture was incubated at 30°C for exactly 2 hours with shaking and a few micro-liters were analyzed on a FACSCAN (Beckton Dickinson) cytometer to record optical parameters of 15,000 living cells. The parameters were: Forward Scatter (FSC) on a linear scale, Side Scatter (SSC) on a linear scale, and GFP fluorescence (FL1) on a log scale. Raw data were read either directly from the original listmode data files using the RflowCyt package from Bioconductor (www.bioconductor.org), or from ASCII text files obtained after running MFI (Martz, Eric. 1992–2001. MFI: a flow cytometry list mode data analysis program optimized for batch processing under MS-DOS. http://www.umass.edu/microbio/mfi).
All computational analysis was done using the R statistical package (www.r-project.org). Because the distribution of FSC and SSC values differed slightly between the divergent genetic backgrounds, we did not gate the data but applied the following correction for cellular granularity and size: yi→y¯+εi, where yi is the observed FL1value of the ith cell and εi is the ith residual of linear regression FL1 = y¯+b*log(FSC)+c*log(SSC). The conclusions of the study remained if gating was applied instead of this correction (Figure S4). Noise was then defined as the coefficient of variation (standard deviation/mean ratio) of the corrected values.
We searched for QTL by two complementary approaches: genome scanning and introgression. For genome scanning, the three noise values of each S288c×RM11-1a segregant were averaged and genetic linkage was searched at every marker position as follows: segregants were divided in two groups according to the marker genotype, noise difference between the two groups was tested using the Wilcoxon Mann-Whitney test. The genome-wide significance of the corresponding nominal P-values was determined by permuting the segregant indexes, re-scanning the genome and recording the smallest P-value obtained at each run. P<2.7×10−5 was reached in only 5 of the 500 permutation runs, thus defining the 1% genome-wide significance. For introgression, strains GY159 and GY174 were obtained by consecutive backcrosses with S288c derivatives, selecting spores with high-noise levels at each generation. GY159 and GY174 were then genotyped using oligonucleotide microarrays: their genomic DNA was extracted, digested, labeled and hybridized to YGS98 Affymetrix® Yeast Genome microarrays as described previously[44]. The genotype of each strain was obtained at 3015 marker positions by adding the corresponding raw .CEL data file to the dataset of Yvert et al. 2003[45] and by applying the same algorithm as previously described in Brem et al. 2002[2]. We then screened the markers for those harboring the RM11-1a genotype in the two introgressed strains (GY159 and GY174) as well as in the S288c×RM11-1a segregant displaying the highest noise level (GY157). A total of 230 markers were selected this way, 32 of them being clustered at one locus on chromosome V (Figure 2B). To determine if the other 198 markers, which were scattered across the genome, truly reflected RM11-1a genotypes, we directly assessed them by PCR and sequencing or RFLP. We found that most of these markers were in fact of the S288c genotype in at least one of the two introgressed strains and we did not consider them further. The locus on chromosome V was then validated as a QTL of PMET17-GFP noise by analyzing an independent cross as described in text.
Because noise scaled with mean expression, we used various induction levels of the reporter construct by varying the concentration of the repressor (methionine). The data presented on Figure 4C was then treated as follows: a linear model was fitted to S288c values (red), and noise values from the two other strains (blue) were corrected by subtracting the expected noise value from the model. Corrected noise values were then averaged for each strain, estimating at 3.5% the difference between S288c and RM11-1a, and at 2.2% the difference between S288c and the URA3-rescued RM11-1a strain (note that here the phenotype itself is measured as a percentage since it is a coefficient of variation). The ura3Δ0 mutation therefore contributed to (3.5–2.2)/3.5 = 37% of the total difference between the parental backgrounds. |
10.1371/journal.pgen.1003121 | Reciprocal Signaling between the Ectoderm and a Mesendodermal Left-Right Organizer Directs Left-Right Determination in the Sea Urchin Embryo | During echinoderm development, expression of nodal on the right side plays a crucial role in positioning of the rudiment on the left side, but the mechanisms that restrict nodal expression to the right side are not known. Here we show that establishment of left-right asymmetry in the sea urchin embryo relies on reciprocal signaling between the ectoderm and a left-right organizer located in the endomesoderm. FGF/ERK and BMP2/4 signaling are required to initiate nodal expression in this organizer, while Delta/Notch signaling is required to suppress formation of this organizer on the left side of the archenteron. Furthermore, we report that the H+/K+-ATPase is critically required in the Notch signaling pathway upstream of the S3 cleavage of Notch. Our results identify several novel players and key early steps responsible for initiation, restriction, and propagation of left-right asymmetry during embryogenesis of a non-chordate deuterostome and uncover a functional link between the H+/K+-ATPase and the Notch signaling pathway.
| Asymmetries between the left and the right sides of the body are an essential feature of most bilaterian animals, and failure to establish these asymmetries can result in pathological disorders in humans. Left-right asymmetries are established during early development by the asymmetric activity of a signaling pathway in a discrete region of the embryo that plays the role of a left-right axis organizer. Although the role of this signaling pathway appears to be conserved among vertebrates, whether the mechanisms involved in the initial breaking of the symmetry and in the establishment of the left-right organizer are also conserved remains an open question. We report that left-right axis determination in the sea urchin embryo also relies on the activity of a left-right organizer located within the gut of the sea urchin embryo. We also report the unexpected finding that the activity of the H+/K+-ATPase, a previously known but enigmatic player in this pathway, is critically required for activation of the Notch receptor. Our results therefore open the way to analysis of the molecular pathway that regulates establishment of laterality in the sea urchin embryo and uncover a functional link between two essential players of left-right asymmetry.
| Left-right (L/R) asymmetry is an essential feature of development in most bilaterian animals. In vertebrates, the morphology and positioning of many internal organs as well as development of the nervous system is left-right asymmetric and failure to establish these asymmetries can result in pathological disorders [1]–[7]. Left-right asymmetric processes have also been analyzed during development of a number of invertebrates including cephalochordates [8], [9], ascidians [8], sea urchins [10], snails [11] and insects [12], [13]. How left-right asymmetries arise from embryos that are initially bilaterally symmetrical and how the left-right axis aligns consistently with the antero-posterior and dorsal-ventral axes are important questions that have recently become the subject of intensive research in a number of laboratories.
Studies in vertebrates suggest that specification of the left-right axis can be conceptually divided into four distinct steps [1], [5], [14], [15]. The first step involves a directional symmetry-breaking event that allows the L/R axis to be aligned with respect to the A/P and D/V axes. A failure to establish this directional asymmetry results in randomized left-right asymmetries (heterotaxia) characterized, for example by the stochastic positioning of the visceral organs on the left or the right side. In mouse, zebrafish or Xenopus, a leftward flow generated by a ciliated left-right organizer, (the node in the mouse, Küpffer vesicle in zebrafish, and archenteron roof in Xenopus) plays a key role in setting up this initial asymmetry [16]. In contrast, an asymmetrical cell migration at Hensen's node is responsible for establishment of left-right asymmetry in the chick [17]. Furthermore, in both Xenopus and chick, there is evidence for left-right asymmetries being established well before the appearance of cilia in the derivative of the organizer [18]–[20]. It is therefore generally believed that the mechanisms used during the initial symmetry-breaking phase are divergent in different species [2], [21].
The second step in left-right axis determination involves establishment of asymmetric gene expression on the left and/or right side of the embryo in response to the flow of laterality information from the organizer. In contrast to the apparent variety of mechanisms used to break the bilateral symmetry in vertebrates, there is a striking conservation in the role played by the TGF beta Nodal in this process. In all vertebrate and chordate species studied so far, including zebrafish, Xenopus, mouse, rabbit, amphioxus and in the tunicate Ciona, nodal is the earliest known gene expressed in the periphery of the node and in the left lateral plate mesoderm in response to signals from the left-right organizer [2], [8].
During the third step, left-right information is transferred from the organizer to the lateral plate. Elegant genetic experiments in the mouse revealed that during this process, Nodal produced in the node region activates its own expression in the distant lateral plate [22], [23] and that this induction requires the expression of the TGF beta GDF1 in the node [24]. In the lateral plate, Nodal activates the expression of its downstream target pitx2, which by regulating cell proliferation, cell migration and cell adhesion, participates in the fourth and crucial step of left-right axis i.e. the translation of asymmetric gene expression into asymmetric placement and morphogenesis of the organ primordia [4], [25]–[27].
An important and heavily debated question in the field of L/R axis establishment is whether there is a conserved early cascade of laterality upstream of nodal expression [21], [28]–[30]. In several species, the earliest event involved in the establishment of the L/R axis upstream of nodal expression involves the activity of the H+/K+-ATPase. Pharmacological inhibition of the H+/K+-ATPase induces heterotaxia in several vertebrate animal models including zebrafish [31], Xenopus and chick [19], causes random left-right determination in embryos from basal chordates such as tunicates (Ciona intestinalis) and disrupts left-right determination in embryos of basal deuterostomes organisms such as the sea urchin [10], [32]. This strongly suggests that a mechanism involving the activity of the H+/K+-ATPase plays a central and perhaps ancestral role in determination of left-right asymmetry. The exact role played by the H+/K+-ATPase is largely enigmatic. Levin and colleagues suggested that an asymmetric activity of a H+/K+-ATPase may generate gradients of membrane potential that in turn may regulate the directionality of gap junction communication or, alternatively, that the activity of the H+/K+-ATPase may regulate the synthesis or secretion of a right sided determinant [19]. In contrast, Gros et al. reported that chick embryos incubated in the presence of omeprazole, an inhibitor of the H+/K+-ATPase, do not display the asymmetrical cell movements that initiate left-right asymmetry in birds, suggesting that the H+/K+-ATPase may regulate cell movements [17]. Raya et al reported that omeprazole treatment abolishes the Notch-dependent asymmetrical expression of Delta around the Hensen's node and suppressed the expression of nodal in the perinodal region indicating that omeprazole treatments interfere with the transcriptional activation of nodal in the node [33]. More recently, Walentek et al. proposed that the activity of the H+/K+-ATPase is required for canonical and non canonical Wnt signaling and foxJ expression [34]. Therefore, a unifying mechanism for the role of the H+/K+-ATPase is still lacking.
In vertebrates, an early requirement for Notch signaling upstream of nodal expression is another conserved feature of left-right determination. In mouse, chick, and zebrafish, Notch signaling is required to initiate nodal expression around the node and mouse mutant lacking the activity of Delta1, CSL (CBF1/RBPJ/Su(H)/Lag-1)/Suppressor of Hairless or of Notch1 and Notch2, fail to express nodal in the node region and show severe defects of left-right patterning [35], [36], [37]. Work from Izpisua Belmonte and coll. suggested a possible link between the role of ionic flux generated by the H+/K+-ATPase and Notch signaling. These authors proposed that, in addition to promoting the asymmetric expression of Delta1 around the node, an asymmetry in the activity of the H+/K+-ATPase may regulate an accumulation of extracellular calcium on the left side that may in turn promote the activation of the Notch signaling pathway [33]. Clearly, our understanding of the role of proton pumps in determination of L/R asymmetry remains scarce and further studies are required to clarify the links between the activity of the H+/K+-ATPase, extracellular calcium and Notch signaling.
Recently, we started to dissect the process of left-right axis specification in the sea urchin [10]. Sea urchins are invertebrates but, like vertebrates, they belong to deuterostome superclade. This basal evolutionary position, as a sister group of the chordates, makes them an interesting phylum to study the conservation of mechanisms used to build the body plan of deuterostomes. Sea urchin development offers a striking example of left-right asymmetry (Figure 1). Like most echinoderms, sea urchins develop indirectly and their larvae undergo a metamorphosis during which most larval tissues are replaced by adult tissues generated from an imaginal disk called the adult rudiment, that forms exclusively on the left side of an otherwise bilaterally symmetric larva [38], [39]. The rudiment derives from the left coelomic pouch and from a portion of the ectoderm located on the left side of the vestibule, where the mouth is located. Precursors of the coelomic pouches have a double origin: part of these precursors derive from the non-skeletogenic mesoderm that is induced by Delta-Notch signaling at the vegetal pole while another contribution comes from the small micromeres [40]–[43]. Although formation of the rudiment is a textbook example of left-right asymmetry, very little was known until recently on the mechanism that control the asymmetric positioning of this organ [44]–[46]. In particular, rudiment positioning has been shown to depend on a signal released by the micromeres but the identity of this signal is unknown [46].
We showed previously that a Nodal-Lefty-Pitx2 signaling pathway regulates left-right asymmetry during development of the sea urchin embryo [10]. However, intriguingly, nodal in the sea urchin is expressed on the right side of the ectoderm and in the right coelomic pouch at the end of gastrulation and not on the left side as in all vertebrates where its expression has been analyzed. Functional analysis revealed that one function of Nodal signals on the right side is to repress formation of the adult rudiment. Inhibition of Nodal signaling after gastrulation caused formation of an ectopic rudiment while ectopic activation of the pathway after gastrulation prevented formation of the rudiment [10]. Furthermore, we showed that inhibition of the H+/K+-ATPase disrupted the directional left-right asymmetry and randomized both nodal expression and positioning of the rudiment [10].
We now report that establishment of left-right asymmetry in the sea urchin embryo involves reciprocal signaling between the ventral ectoderm that expresses nodal and a left-right organizer of endodermal origin and that this long-range signaling requires Univin/Vg1. We show that in the absence of this organizer or when an organizer forms both on the left and the right sides, nodal expression in the ectoderm is randomized along the left-right axis suggesting that this endomesodermal left-right organizer is only responsible for orienting the symmetry breaking and for making it directional.
We provide evidence that establishment of this organizer requires the activity of several signaling pathways including the Notch, FGF-ERK, BMP2/4 and Univin/Vg1. Finally, we report the unexpected finding that the activity of the H+/K+-ATPase is critically required for Notch signaling and that inhibiting the activity of this ATP driven proton pump phenocopies inhibition of Notch signaling in the early embryo leading to complete suppression of the expression of Notch target genes and to the absence of mesodermal derivatives. Our results therefore open the way to the analysis of the molecular pathway that regulates left-right asymmetry in the sea urchin embryo and uncover a functional link between two essential players of left-right asymmetry i.e. the H+/K+-ATPase and Notch signaling.
Asymmetric expression of nodal along the left-right axis could be detected as early as the mid-gastrula stage (Figure 2A). At this stage (about 22 hpf), while the archenteron had not yet reached the animal pole region, nodal expression was detected in a group of about 2–5 cells embedded into the wall of the archenteron on the right side. Double fluorescent in situ hybridization with the endodermal marker foxA and the mesodermal marker foxF confirmed that these nodal expressing archenteron tip cells are located at or near the boundary between the mesoderm and endoderm, immediately adjacent to the coelomic pouch precursors that express foxF (Figure 2B). During the next 2.5 h period, the territory expressing nodal was progressively displaced towards the animal pole and at 24 hpf, a cluster of about 10–15 cells arranged in a rosette expressed nodal asymmetrically at the tip of the archenteron on the right side (Figure 2A and Figure S1). Based on their position immediately adjacent to the delaminating secondary mesenchymal cells, these nodal expressing cells at 24 h likely correspond to precursors of the right coelomic pouch. Importantly, during this period, nodal expression remained symmetric in the ventral ectoderm. Weak asymmetric expression of nodal was first detected in the ectoderm, on the right side of the presumptive ciliary band territory around 24 hpf. This asymmetry in the distribution of nodal transcripts in the ectoderm further accentuated during the following 3 h period and at 26 hpf, strong asymmetric expression of nodal on the right side was detected both at the tip of the archenteron and on the right side of the ectoderm in most embryos (Figure 2A). Therefore, this analysis revealed that the first asymmetric expression of nodal occurs in the endomesoderm and not in the ectoderm, as previously thought, and that nodal expression subsequently expands from the endomesoderm to the mesoderm.
Similarly, L/R asymmetric expression of univin started to be detected in the right coelomic pouch around 24 hpf, well after asymmetrical nodal expression had been initiated in the endomesoderm, while asymmetric expression of univin in the ectoderm occurred only after 26 hpf, well after nodal expression had switched to the right side of the ventral ectoderm (Figure S1). The finding that the first manifestation of left-right asymmetry determination during sea urchin embryogenesis is asymmetric expression of nodal in the archenteron strongly suggested that during normal development, the first symmetry-breaking event occurs in the endomesoderm. Furthermore, the later shift of nodal and univin expression from a bilaterally symmetric expression in the ectoderm to an asymmetric expression on the right side suggested that the asymmetry initiated in the endomesoderm is subsequently transferred to the ectoderm.
Previous work [10 and unpublished data] as well as unpublished results from our lab indicated that in the sea urchin, like in vertebrates, the H+/K+ ATPase and Delta/Notch are key players required upstream of nodal expression during left-right axis establishment. We therefore first investigated if the activities of Notch and of the H+/K+ ATPase are required for the asymmetric expression of nodal in the endoderm at gastrula stage. Surprisingly, inhibition of Notch signaling by treatment with the γ-secretase inhibitor DAPT ([N-(3,5-Difluorophenacetyl)-L-alanyl]-S-phenylglycine t-butyl ester) or by injection of a morpholino against Delta did not abolish nodal expression in the endoderm but caused instead ectopic expression of this gene on the left side of the archenteron (Figure 3A). Starting at 22 hpf, while in control gastrulae nodal was expressed exclusively on the right side of the archenteron, in DAPT-treated embryos and in Delta morphants, nodal transcripts were expressed bilaterally in two groups of cells in the archenteron. Similarly, blocking the activity of the H+/K+-ATPase by treatment with omeprazole caused nodal to be expressed bilaterally in the endomesoderm at gastrula stage (Figure 3A). These results suggest that Delta/Notch signaling and the activity of the H+/K+-ATPase are required (either directly or indirectly) to repress nodal expression in cells located on the left side of the archenteron.
At pluteus stage, DAPT-treated larvae and Delta morphants expressed nodal and univin asymmetrically in the ectoderm but the expression was detected either on the right side or on the left side (Figure 3B). Consistent with the random expression of nodal at pluteus stage, DAPT treated larvae developed with a rudiment that was randomly positioned on either the right or the left side (Figure 3B). As controls for the effect of DAPT treatment and of the Delta morpholino, we analyzed the expression of marker genes transcribed either asymmetrically (pitx2, sox9) or symmetrically (foxF) in the coelomic pouches precursors or in the muscle cell precursors (tropomyosin) in response to Delta-Notch signaling. Indeed, expression of all four mesodermal marker genes was abolished in most of the DAPT-treated embryos as well as in the Delta morphants consistent with the expected severe reduction of mesodermal derivatives caused by inhibition of Notch signaling (Figure 3C) [41], [42]. Therefore, inhibition of Notch signaling, in addition to preventing specification of the coelomic pouch precursors, caused the early endodermal expression of nodal to become bilateral and randomized nodal and univin expression in the ectoderm at pluteus stage.
To determine when Notch signaling is required for establishment of left-right asymmetry, we treated embryos with DAPT for various time windows and analyzed the expression of nodal (Figure 3D). This analysis revealed that the period during which DAPT is effective at perturbing left-right asymmetry corresponds to early development, with treatments performed during the cleavage/early blastula period being the most effective, the efficiency of the treatment rapidly dropping after early blastula stage, and treatments performed after hatching no longer perturbing left-right asymmetry. The period during which Notch signaling is required to establish left-right asymmetry largely overlaps with the period during which secondary mesodermal precursors are induced by Delta signals expressed in the primary mesenchymal cell precursors [42], [47]. This suggests that Notch signaling regulates nodal expression indirectly, likely through signaling between the mesoderm that is induced by Delta/Notch signaling and the endoderm that expresses nodal. This also suggests that Delta is the signal released by the micromeres that regulates positioning of the rudiment [46].
In the sea urchin embryo like in vertebrates, treatments with the H+/K+ pump inhibitor omeprazole randomize L/R nodal expression [10]. Interestingly, we found a striking similarity between the phenotypes resulting from treatments with H+/K+ ATPase inhibitors, and treatments that interfere with Delta-Notch signaling (Figure 4A). Treatments with omeprazole, like treatments with DAPT or injection of the morpholino against Delta, strongly delayed gastrulation and resulted in development of embryos that largely lacked delaminating secondary mesenchymal cells at the tip of the archenteron and that later were largely albino (Figure 4A). Furthermore, the window during which omeprazole is mostly effective extends from fertilization to the very early blastula stage, i.e. a period very similar to the window of action of the Notch inhibitor DAPT (Figure S2). These observations raised the possibility that omeprazole treatments inhibit Notch-Delta signaling. To test this possibility, embryos were treated with omeprazole during cleavage and blastula stages and the expression of mesodermal marker genes activated in response to Notch activation, such as the immunocyte markers gcm, papss and GATA1/2/3, was analyzed (Figure 4B) [48], [49], [50], [51]. As a control, we analyzed the expression of the Delta ligand and of msp130, two genes that are expressed in the skeletogenic mesoderm territory independently of Delta/Notch signaling [42], [52] as well as the expression of the endodermal marker gene foxA [53]. Strikingly, in most embryos treated with the proton potassium pump blocker, expression of the immunocyte marker genes, which are regulated by Delta signaling, was strongly downregulated or absent. In contrast, foxA was expressed at apparently normal levels in the endoderm precursors. Furthermore, consistent with the previously described expansion of endodermal precursors at the expense of non skeletogenic mesodermal precursors caused by inhibition of Delta-Notch signaling [41], [47], [54]–[56], the vegetal boundary of the territory expressing foxA in the DAPT, Delta-Mo injected embryos or omeprazole treated embryos was shifted towards the vegetal pole (Figure 4B). In contrast, expression of Delta and msp130 in the skeletogenic mesoderm precursors was largely normal in the omeprazole treated embryos. This shows that inhibition of the H+/K+-ATPase does not perturb specification of the skeletogenic mesoderm and endoderm but that it specifically interferes with specification of the non-skeletogenic secondary mesoderm. Since the non-skeletogenic mesoderm is induced by Delta signals emanating from the adjacent skeletogenic mesoderm precursors, this further suggests that omeprazole treatment, may block reception of the Delta signal in the surrounding cells.
We next sought to determine in which step of the Notch pathway, the activity of H+/K+-ATPase may be required by combining Notch gain of function and omeprazole treatments. During secretion in the trans Golgi network, the Notch protein is first processed by proteases of the Furin family that generate a non-covalent heterodimer between the Notch extracellular domain NECD and Notch tethered intracellular domain that interact in a Ca2+ dependent manner [57]. Upon binding of Delta, Notch is cleaved at the S2 site by proteases of the ADAM/TACE family, generating a membrane bound activated form of Notch called NEXT (Notch Extracellular Truncation). NEXT is then the substrate for the gamma secretase, which catalyzes the intramembranous S3 cleavage that releases the Notch intracellular domain NICD [58]. To further define the step in which the activity of H+/K+-ATPase is required for Notch signaling, we used luciferase assays. We overexpressed mRNAs encoding the P. lividus Delta, NEXT or NICD proteins and measured the activity of the Notch reporter gene RBP-JK [59] in the presence or absence of omeprazole (Figure 4C). Omeprazole treatment strongly inhibited the stimulation of Notch signaling induced by overexpression of Delta, consistent with a disruption of Notch signaling caused by the inhibitor. In contrast, omeprazole treatment had no effect on the activation of Notch signaling caused by overexpression of NEXT or NICD. This strongly suggests that the H+/K+-ATPase is required before or at the level of the S2 cleavage that generates NEXT.
In vertebrates, the FGF/MAP kinase pathway is involved in establishment of left-right asymmetry. FGF signaling has been implicated in the symmetry breaking process and in the release of nodal vesicular parcels (NVPs) that carry Sonic Hedgehog and retinoic acid [60]. Furthermore, inhibition of FGF signaling disrupts left-right asymmetry in Xenopus and zebrafish, an effect that has been correlated to a reduction of ciliary length [61]. To investigate if FGF/MAP kinase signaling is required for the early asymmetry of nodal expression in the endomesoderm and for establishment of left-right asymmetry during sea urchin development, we analyzed the expression of nodal following treatments with the FGFR inhibitor SU5402 and with the MEK inhibitor U0126 (Figure 5). As controls for the effects of the inhibitors, we verified that the expression of pax2/5/8 and sprouty, two downstream targets of FGFA in the ectoderm [62], is downregulated in the treated embryos (Figure S3 and data not shown). While treatments with DAPT caused bilateral expression of nodal in the archenteron, in contrast, treatments with U0126 or with SU5402 abolished nodal expression in the endoderm at gastrula stage (Figure 5A). Therefore, a positive input from the FGF/MAP kinase signaling pathway is required for nodal expression in the endomesoderm. To test if FGF/MAP kinase signaling is required for nodal induction through inhibition of Notch signaling, embryos were treated simultaneously with DAPT and U0126. Double inhibition of Notch and MAP kinase signaling prevented nodal expression in the endomesoderm indicating that FGF signaling is likely required downstream or in parallel to Notch signaling for induction of nodal expression (Figure 5A). Interestingly, despite the absence of nodal expression in the endomesoderm at gastrula stage, nodal was expressed asymmetrically in the ectoderm of SU5402 or U0126 treated embryos at pluteus stage, but as in the case of DAPT treated embryos, its expression was randomized along the left-right axis (Figure 5B). Similarly, expression of sox9, pitx2 and univin was randomized following inhibition of ERK signaling (Figure 5B). About one third of the U0126-treated larvae later developed with two rudiments while in the remaining larvae the rudiment was either on the left (31%) or the right side (38%) (Figure 5C). Time-course analysis revealed that the window during which SU5402 and U0126 are effective at perturbing left-right asymmetry extends from early mesenchyme blastula up to the early gastrula stage, i.e., immediately before the initiation of asymmetric nodal expression in the archenteron tip cells (Figure 5D and Figure S3). Collectively, these results demonstrate that FGF/MAP kinase signaling is critically required to initiate asymmetric expression of nodal in the endoderm and that perturbations of nodal expression in this endodermal territory ultimately result in randomized left-right asymmetry in the ectoderm and random positioning of the rudiment (Figure 5E). Furthermore, both the absence of nodal expression in this endomesodermal territory and the bilateral expression of nodal in this region resulted in randomized ectodermal nodal expression along the left-right axis suggesting that this endomesodermal region has the properties of a left-right organizer. Although this organizer does not appear to be necessary for the process of symmetry breaking itself, it is responsible for orienting the symmetry breaking and for making it directional.
Finally, since Nodal and BMP2/4 play antagonistic roles during patterning of the ectoderm in the sea urchin embryo [63], [64] and since BMP signaling is active in the upper part of the archenteron that expresses nodal during gastrulation [65], we investigated if BMP signaling is required for specification of this left-right mesendodermal organizer and for the subsequent establishment of left-right asymmetry (Figure 6). We first tested the effects of perturbations of BMP signaling on nodal expression on the right side at gastrula stage. Treatments with recombinant BMP4 protein very efficiently suppressed nodal expression in the archenteron tip cells and in the ectoderm (Figure 6A, Figure S4) suggesting that elevated BMP signaling can antagonize Nodal signaling in the context of left-right asymmetry. Injection into the egg of morpholino oligonucleotides directed against the bmp2/4 transcript or against the transcript encoding Alk3/6, a type I BMP receptor that is required to transduce BMP2/4 signals [65], also eliminated nodal expression in the left-right organizer indicating that BMP signaling is essential for the early nodal expression in the endomesoderm (Figure 6A). Consistent with the observed loss of nodal expression in the endomesoderm at gastrula stages, nodal expression in the ectoderm was randomized in the bmp2/4 or alk3/6 morphants at pluteus stage. In the absence of BMP signaling, nodal expression in the ectoderm also expanded dorsally suggesting that BMP signaling is required as a dorsal barrier in the ectoderm (Figure 6B). Taken together, these observations suggest that BMP signaling is first required in the endomesoderm to establish nodal expression in the mesendodermal organizer, then, that it is required as a dorsal barrier in the ectoderm to prevent expansion of nodal expression to the dorsal side.
To test if BMP signaling is required in the endomesoderm or in the ectoderm for nodal expression in the left-right organizer, we specifically blocked BMP signaling in the endomesoderm by injecting the alk3/6 morpholino in the four vegetal blastomeres of embryos at the 8-cell stage and analyzed nodal expression at gastrula and pluteus stages. Inhibition of BMP signaling in the endomesoderm prevented nodal expression in the endomesoderm at gastrula stage in 93% of the injected embryos (two experiments n: 30)(Figure 6C, 6D). All the embryos injected with the alk3/6 morpholino in vegetal blastomeres nevertheless developed into pluteus larvae. However, consistent with the absence of nodal expression in the endomesoderm at gastrula stage, ectodermal nodal expression in these larvae was randomized. This result extends the previous observations made after inhibition of BMP signaling at the 1-cell stage and indicates that in the sea urchin embryo, BMP signaling in the endomesoderm plays a positive and essential role in the initiation or maintenance of nodal expression in the mesendodermal organizer.
To better define the role of BMP signaling in the establishment of left-right asymmetry, we injected the bmp2/4 morpholino into one blastomere at the two cell-stage and, at gastrula stage, selected the embryos that inherited the morpholino on either the left or the right side and analyzed nodal expression at gastrula and pluteus stage (Figure 6E–6G). Intriguingly, while targeting the bmp2/4 morpholino to the left side resulted in either the complete suppression (85% n = 20) or strong reduction (15%) of nodal expression in the organizer at gastrula stage, normal nodal expression could be detected in 45% of the embryos that had received the morpholino on the right side (n = 11). The different sensitivities of the left and right sides to the bmp2/4 morpholino raised the possibility that BMP signaling on the left side may be required on the right side for nodal expression in the left-right organizer. Consistent with this idea, injection of the bmp2/4 morpholino into the presumptive right side territory did not perturb the sidedness of nodal at pluteus stage but strikingly, injection of the bmp2/4 morpholino on the presumptive left side randomized nodal expression in the ectoderm. To test if BMP signaling is asymmetric in the archenteron at gastrula stage, we tried to detect endogenous BMP signaling using an antiphosphoSmad1/5/8 antibody. Anti-phospho Smad1/5/8 immunostaining revealed the presence of a domain with strong BMP signaling in the archenteron at gastrula stages (Figure 6H). In most embryos (13/19), nuclear staining in the archenteron was asymmetric, with more intense staining being visible in the dorsal-left quadrant opposite to the region where nodal is expressed (see also Figure S5). These results suggest that in the sea urchin embryo, BMP signaling in the endomesoderm is required to establish nodal expression in the left-right organizer located on the right side. Furthermore, they suggest that at gastrula stage, BMP signaling itself is asymmetric, with stronger signaling occuring on the left side of the archenteron.
As described above, treatments that perturb the early expression of nodal, resulting in either bilateral expression of nodal (inhibition of Delta/Notch signaling) or in the absence of expression of nodal in the endoderm (inhibition of FGF/MAP kinase or of BMP signaling), ultimately randomize the expression of nodal in the ectoderm at later stages. This suggested that during sea urchin development, the first left-right asymmetry appears in the endomesoderm and that this asymmetry is subsequently transmitted to the ectoderm in the form of an asymmetric expression of nodal and univin on the right side of the ciliary band region. Consistent with this idea, previous experiments had shown that inhibition of nodal mRNA translation at the egg stage followed by local injection of nodal mRNA into one animal blastomere (belonging to the presumptive ectoderm), efficiently rescued dorsal-ventral polarity, but failed to rescue left-right polarity in the endomesoderm and did not restore ectodermal expression of nodal and pitx2 on either side of the larva [10]. However, paradoxically, previous results from our laboratory also showed that inhibition of Nodal function in the ectoderm abolished the asymmetric expression of pitx2 in the endomesoderm suggesting that ectodermal Nodal signals were required upstream of endomesodermal Nodal expression [10]. One scenario that may reconcile these observations is that Nodal signals coming from the ectoderm may first be required for the asymmetric expression of nodal and pitx2 in the endomesoderm, then this asymmetry may be subsequently transmitted through Nodal signaling from the endomesoderm to the right ectoderm. To test this idea, we blocked Nodal signaling in either the ectoderm or the endomesoderm and analyzed nodal expression in the endomesoderm at gastrula stages as well as nodal and pitx2 expression in the ectoderm and coelomic pouches at pluteus stages (Figure 7). Injection of Nodal morpholino into the four animal blastomeres at the 8-cell stage abolished nodal expression in the endomesoderm at gastrula stage and produced radialized embryos consistent with previous results (Figure 7A, 7B) and Table 1 [10]. Therefore ectodermal Nodal signals are required upstream of endomesodermal nodal expression. In embryos radialized by treatments with recombinant Nodal or nickel chloride, however, nodal was expressed radially in the ectoderm but expression in the endomesoderm was abolished (Figure S6). Therefore, normal dorsal-ventral patterning of the ectoderm is required for nodal expression in the endomesoderm. Consistent with the idea that endomesodermal nodal expression requires ectodermal Nodal signals, blocking translation of nodal mRNA or blocking reception of Nodal signals in the endomesoderm by injection of alk4/5/7 morpholinos into the four vegetal blastomeres prevented nodal expression in the endomesoderm at gastrula stage (Figure 7C, 7D and Table 2). Injection of alk4/5/7 morpholinos into the four vegetal blastomeres did not affect establishment of dorsal-ventral polarity but it randomized nodal expression in the ectoderm at pluteus stage and eliminated pitx2 expression in the right coelomic pouch. Therefore, endomesodermal Nodal signals are indeed required to establish the directional asymmetry of nodal expression in the ectoderm.
We also investigated if interfering with Nodal function in the endoderm perturbs establishment of left-right asymmetry in the ectoderm by using chimeras (Figure 7E, 7F). Eggs were injected with the Nodal morpholino together with a lineage tracer and allowed to develop up to the 16/32-cell stage, then, the animal and vegetal regions were separated and recombined with their complementary halves derived from wild type embryos. When the function of Nodal was inhibited in the animal hemisphere, the resulting chimeras displayed a phenotype very similar to that observed following injection of the morpholino into the egg: the embryos lacked both dorsal-ventral and left-right polarity, consistent with the essential role of nodal in establishment of these embryonic axes (not shown) [10], [66]. In contrast, chimeras in which the Nodal morpholino was present in the vegetal hemisphere developed into morphologically normal pluteus larvae (Figure 7E, 7F) (100% n = 12). However, in these embryos, nodal expression in the ectoderm was randomized (Figure 7F). This shows that, while Nodal function in the ectoderm is clearly important for establishment of left-right asymmetry in the endomesoderm, Nodal signaling in the endomesoderm is in turn essential for transmission of left-right asymmetry to the ectoderm. Therefore determination of left-right asymmetry in the sea urchin embryo most likely requires reciprocal signaling between the ectoderm and endomesoderm.
If Nodal signals emitted from the ventral ectoderm drive nodal expression in the endomesoderm, why, in the dorsal-ventral axis rescue experiments mentioned above, local expression of nodal into one animal blastomere at the 8-cell stage is not able to rescue the expression of L/R markers in the endomesoderm of nodal morphants? We reasoned that in the rescue experiment, the size of clone expressing nodal is much smaller, than the presumptive ventral ectoderm that normally expresses nodal. Furthermore, in these rescue experiments, the progeny of the nodal expressing blastomere typically occupies the center of the ventral ectoderm that gives rise to the region surrounding the stomodeum and, importantly, it does not overlap with the more lateral ectoderm that normally expresses univin at gastrula stage. Univin is a Vg1/GDF1 related factor that is very important during dorsal-ventral axis formation and Nodal/GDF1 heterodimers have been shown to be much more potent and to signal over a longer range compared to Nodal homodimers in other systems [24]. This raised the possibility that the failure of ectopic nodal to rescue left-right patterning in the endomesoderm might be due to the absence of overlap between the nodal expressing clone and the univin expressing territory and to the failure to form Nodal-Univin heterodimers at gastrula stages. To test this possibility, we analyzed pitx2 and sox9 expression following injection of nodal mRNA alone or of a mixture of nodal and univin mRNAs into one blastomere at the 8-cell stage of nodal morphants (Figure 8). While injection of nodal mRNA alone into an ectodermal precursor was unable to induce expression of pitx2 in either the endomesoderm or in the ciliary band, strikingly, co-injection of nodal and univin rescued expression of pitx2 in the right coelomic pouch and induced a massive expression of pitx2 throughout the right and left portions of the distant ciliary band (Figure 8B). This shows that local and symmetric expression of nodal and univin in the ectoderm of nodal morphants is sufficient to rescue asymmetric expression of pitx2 in the endomesoderm, consistent with previous data showing that Nodal signaling in the ectoderm is essential for driving asymmetric nodal/pitx2 expression in the endomesoderm. Conversely, targeting the Univin morpholino to the right side of the embryo completely blocked the asymmetric expression of nodal in the ectoderm on the right side (Figure S7). Taken together, these results strongly suggest that Nodal and Univin synergize to signal both locally and over a long range during left-right patterning in the sea urchin embryo.
In conclusion, these results (summarized in Figure 9B) strongly suggest that determination of left-right asymmetry in the sea urchin embryo involves two successive reciprocal long-range signaling events between the ectoderm and the endomesoderm mediated by Nodal-Univin heterodimers (Figure 9A). First, during gastrulation, a Nodal/Univin signal emitted by the ventral ectoderm cooperates with an FGF signal of unknown origin and with a BMP signal coming from the left side of the archenteron to initiate nodal expression in cells on the right side of the tip of the archenteron (Figure 10). On the left side, an unidentified signal, likely emitted by the mesoderm induced by Delta/Notch signaling is required to repress nodal expression. Together, these positive and negative signals are responsible for establishment of a left-right mesendodermal organizer on the right side of the tip of the archenteron, which starts to express nodal then univin. At late gastrula/prism stage, Nodal/Univin signals emitted from this organizer are responsible for transferring left-right asymmetry from this mesendodermal organizer to the lateral ectoderm by inducing nodal and univin expression in cells located on the right side of the ventral ectoderm and ciliary band.
Although the function of Nodal in establishment of left-right asymmetry is highly conserved from mollusks to vertebrates, the existence of a conserved molecular pathway initiating left-right asymmetry upstream of Nodal is still questioned [2], [6], [21]. A cilia-based flow has been shown to be necessary and sufficient for establishment of left-right asymmetry in mammals [67], [68] and is essential for this process in teleost fish [68]–[70] and amphibians [71]. However, whether this cilia based mechanism is the first symmetry-breaking event in all these species is strongly debated. In the chick, asymmetric cell movements at the Hensen's node and not flux across the node, determine left-right asymmetry [17]. Furthermore, in Xenopus, the activity of the H+/K+-ATPase is required very early for establishment of left-right asymmetry [19]. Remarkably, a similar early requirement for a H+/K+-ATPase has been described in the chick [19] and in zebrafish [31] Since an early requirement for a H+/K+ ion exchanger has also been described in the sea urchin [10], [32], this early requirement for a proton/potassium exchanger upstream of nodal expression appears as a highly conserved mechanism upstream of nodal during specification of the left-right axis in deuterostomes. The Notch pathway is another conserved pathway that acts upstream of nodal in vertebrates. Notch signaling is required for nodal expression in the Hensen's node region in chick [33], zebrafish [31], [36] and mouse [35]–[37]. However, whether Notch signaling is required upstream of nodal for L/R asymmetry outside vertebrates was unknown. Finally, Vg1/GDF1 signaling has been implicated in the transfer of L/R laterality from the node to the lateral plate mesoderm in Xenopus and mouse [24], [72]–[74] but whether the function of this TGF beta in the regulation of left-right asymmetry is conserved outside vertebrates had not been investigated.
In this study, we showed that several of the signaling pathways that regulate left-right asymmetry in vertebrates, also regulate establishment of left-right asymmetries in the sea urchin embryo. First, we uncovered an essential and early role for the Notch signaling pathway in directing the unilateral expression of nodal in the endomesoderm of the sea urchin embryo, providing evidence that in addition to ion flux and Nodal signaling, the role of the Notch pathway is also primordial during establishment of left-right asymmetry in the embryo of a non-chordate deuterostome animal. However, unlike in vertebrates, where Notch signaling is directly required to promote nodal expression, in the sea urchin, Notch signaling is required indirectly to restrict nodal expression to the right side. Furthermore, we showed that in the sea urchin, the activity of the H+/K+-ATPase is essential for the induction of several endogenous Delta/Notch target genes and for the expression of a Notch activity reporter gene, suggesting that this proton pump is directly required for transduction of the Notch signal. Therefore, these results uncover a functional link between two major players of L/R determination: the H+/K+-ATPase and the Notch pathway. Finally, we showed that in the sea urchin as in vertebrates, FGF and BMP signaling as well as signaling by Univin, a TGF beta related to Vg1 and GDF1, are essential for specification of left-right asymmetry.
The Notch signaling pathway plays a key role in establishment of left right asymmetry in vertebrates. However, the mechanisms by which Notch acts in this pathway differ significantly between the mouse and the zebrafish. Genetic analysis in the mouse showed that expression of nodal in the node is crucial for subsequent propagation of nodal expression to the lateral plate [22], [23]. Several studies have demonstrated that perturbations of the Notch pathway strongly affect this early expression of nodal in the node and disrupt establishment of left-right asymmetry. Embryos mutant for Delta1, or double mutant for Notch1 and Notch2 or lacking the function of CSL, (the main transcriptional effector of the Notch pathway), fail to express nodal in the node and subsequently are unable to establish the left-sided expression of nodal in the lateral plate [35]–[37]. Indeed, expression of nodal in the node is directed by a cis-regulatory module that contains binding sites for CSL and mutations of these sites abolish the activity of this enhancer. In zebrafish, however, nodal expression in the node is not eliminated by disruption of Delta/Notch signaling. In this case, Notch signaling appears to control cilia length in the Kupffer's vesicle by regulating the expression of the master cilia regulator foxJ1 [75]. Another primary target of Notch signaling in the zebrafish appears to be the gene encoding the Nodal antagonist of molecule Charon [75], [76]. Charon is first expressed symmetrically in the node region, then asymmetrically with a stronger expression on the right side of the node where Charon antagonizes Nodal signaling. The finding that in Delta mutants or following DAPT treatments, expression of charon, but not nodal expression, is strongly reduced and the presence of several CSL binding sites in the charon promoter strongly suggest that Notch signaling regulates charon expression in the zebrafish. While in the mouse inhibition of Notch signaling prevents nodal expression, in zebrafish, inhibition of Notch signaling causes instead nodal to be expressed bilaterally in the lateral plate [76].
Our results clearly showed that the Notch pathway also plays a crucial role during establishment of left-right asymmetry in the sea urchin embryo. Inhibition of Notch signaling by injection of morpholino directed against Delta or treatment of embryos with a γ-secretase inhibitor caused bilateral expression of nodal in the endoderm at gastrula stage and randomized nodal expression in the ciliary band at later stage. The function of Notch signaling in the sea urchin embryo therefore does not appear to be in the activation of nodal expression like in the mouse, but instead in the repression of nodal expression on the left side, like in the zebrafish, since inhibition of Notch signaling caused bilateral expression of nodal in the endomesoderm. How Notch signaling promotes unilateral expression of nodal on the right side in the sea urchin is presently unclear. Since the mesodermal precursors lie immediately on the top of the invaginated archenteron, and since Notch signaling is primarily required for specification of these mesodermal precursors, one possibility is that Notch signaling is required early for specification of mesodermal cells, which in turn send an inhibitory signal during gastrulation that prevents nodal expression on the left side of the underlying endoderm (Figure 10). Alternatively, Notch may be required for the correct positioning of a signal that induces Nodal expression on the right side. A third possibility is that, by analogy to the role of Notch signaling in the chick, Notch signaling may regulate cell rearrangements that would be required for establishment of left-right asymmetry. In line with this idea, previous studies reported that the progeny of the small micromeres partition asymmetrically into the two coelomic pouches with the left coelomic pouch inheriting a larger fraction than the right coelomic pouch [43]. It is important, however, to keep in mind that the period during which Notch signaling is required for correct right sided expression of nodal is separated from the onset of nodal expression in the archenteron by 15 h and therefore that the effect of Notch signaling on nodal is most likely very indirect. Future studies are required to understand how Notch signaling regulates left-right asymmetry in the sea urchin embryo. In particular, the identity of the inhibitory signal X remains to be established. It is interesting to draw a parallel between the repressive effect of the non-skeletogenic mesoderm on endodermal precursors of the left-right organizer and the repressive effects that the PMCs exert on the non skeletogenic mesodermal precursors. When the skeletogenic precursors (micromeres or PMCs) are removed, non skeletogenic precursors transfate to replace the missing skeletal precursors [77]. It will interesting to determine if the repressive effects of the PMCs on SMC conversion to a skeletogenic fate and the repressive effects of the non skeletogenic precursors on endodermal precursors conversion into a nodal expressing left-right organizer rely on similar molecular mechanisms.
Finally, ablation of micromeres at the 16-cell stage has been reported to perturb left-right asymmetry and to randomize positioning of the rudiment suggesting that micromeres release a signal that regulates left-right asymmetry [46]. Our results strongly suggest that this signal is Delta, which is expressed in the progeny of the large micromeres where it induces non skeletogenic mesoderm precursors from surrounding endomesodermal precursors [42], [78], [79].
One of the most striking results of our study is that treatments with the H+/K+-ATPase inhibitor omeprazole mimicked inhibition of Notch signaling in the early embryo. Treatments with omeprazole, like injection of the Delta morpholino or treatments with DAPT, abolished formation of non-skeletogenic mesodermal precursors causing a strong delay in gastrulation [80], [81] and resulting in gastrulae with a smooth archenteron, devoid of secondary mesenchymal cells, and later, in larvae lacking pigment cells [41], [78]. At the molecular level, expression of several marker genes expressed in the secondary mesodermal precursors (gcm, papss and GATA1/2/3) was abolished following inhibition of the H+/K+-ATPase. Therefore, both in the context of establishment of left-right asymmetry and in the context of induction of the germ layers, omeprazole treatments mimicked inhibition of Notch signaling. One study had implicated the activity of the H+/K+-ATPase in the modulation of Notch signaling at the extracellular level. In the chick, the activity of H+/K+-ATPase has been associated with a transient left-right accumulation of extracellular calcium and this transient rise in extracellular calcium has been proposed to promote Notch signaling partly by promoting asymmetrical expression of Delta [33]. It is very unlikely that Notch activity is regulated by an increase in extracellular calcium in the sea urchin since this organism develops in an environment that already contains an extremely high (10 mM) concentration of extracellular calcium. More recent studies have implicated Wnt signaling in the regulation of foxJ1 [82], and the activity of the H+/K+-ATPase in canonical Wnt signaling [34]. In the sea urchin embryo, the phenotypes caused by omeprazole treatment are very different from those resulting from inhibition of Wnt signaling [83], [84]. Furthermore, we showed that omeprazole treatment did not interfere with the Wnt dependent expression of endodermal marker genes such as foxA, ruling out a role for the H+/K+-ATPase in the Wnt pathway. Instead, omeprazole specifically interfered with expression of mesodermal markers, indicating a more direct role in the Notch pathway. To our knowledge, this is the first report that the activity of H+/K+-ATPase is fundamental for Notch signaling. So how may the activity of the H+/K+-ATPase regulate Notch signaling? Two recent studies reported that Delta-Notch signaling is highly pH dependent and that the activity of the V-ATPase, a proton pump that controls the acidity of lysosomes, plays a central role in Notch signaling. In one study Vaccari and coll. showed that cells mutants for the V-ATPase accumulate an uncleaved form Notch in the endosomes and lysosomes and fail to activate Notch signaling [85]. Similarly, in a screen for mutations that disrupt the Notch pathway, Yan et al found that mutations that inactivate genes encoding either Rabconnecting 3 (Rbcn3), a known regulator of V-ATPase in yeast, or VhaC39, a gene encoding a subunit of the V-ATPase, recapitulate a number of phenotypes caused by inactivation of the Notch pathway including defective oogenesis and abnormal patterning of imaginal discs [86]. Cells lacking Rbcn3 or VhaC39 function fail to acidify intracellular compartments and accumulate Notch in late endosomes. How the function of V-ATPases regulates Notch signaling is presently unknown but a number of studies have implicated V-ATPases in the regulation of a number of essential cellular processes such as endocytosis, lysosomal degradation or secretion. It is therefore possible that V-ATPase is required for trafficking of Notch or Delta. Another possibility is that the V-ATPase mediated acidification is required for generation of NICD, the intracellular and active form of Notch. The active form of Notch requires two successive proteolysis events mediated by ADAM metalloprotease and γ-secretase [58]. Interestingly, in Drosophila, expression of NICD, the form of Notch generated by γ-secretase cleavage, but neither expression of full length Notch nor expression of NEXT, can rescue the defects caused by inactivation of Rbcn3 or V-ATPase function, strongly suggesting that V-ATPase is required at or downstream of γ-secretase-mediated S3 cleavage of NEXT [86].
In the sea urchin embryo, omeprazole treatment inhibited the stimulation of Notch signaling induced by overexpression of Delta but had no effect on overexpression of NEXT or NICD. Therefore, omeprazole treatments appear to affect a step located at or upstream of the S2 mediated cleavage of Notch. Since S2 cleavage is mediated by secreted metalloproteases of the ADAM/TACE/Kuzbanian family, one possibility is that the activity of the H+/K+-ATPase is required for the activity of these enzymes. Alternatively, the activity of the H+/K+-ATPase may be required in the signal sending cells through processes such as trafficking or endocytosis of Delta. The localization of the H+/K+-ATPase on the apical surface of epithelial cells is consistent with these proposed roles [87]. The activity of the H+/K+-ATPase was previously shown to be essential for establishment of left-right asymmetry in zebrafish and Xenopus. However, to our knowledge, its role in the regulation of Notch signaling had never been investigated. We showed that the function of the H+/K+-ATPase is mandatory for Notch signaling in the sea urchin embryo and that embryos treated with omeprazole fail to express Notch target genes and later display randomized expression of nodal along the left-right axis. Therefore our results tie together and extend different observations on the roles of proton pumps on establishment of left-right asymmetry and Notch signaling.
Studies in vertebrates suggested the existence of three distinct steps in the establishment of left-right asymmetry: symmetry breaking, initiation of asymmetric expression of nodal in a left-right organizing center and propagation of left-right asymmetry to more distant tissues. These three steps can be identified during establishment of L/R asymmetry in the sea urchin embryo and although the picture is still incomplete, what emerges from this study is that there are both conserved as well as to notably divergent features in the strategies and mechanisms used in echinoderms and vertebrates to establish left-right asymmetry. The role of a discrete mesendodermal region playing the role of a left-right organizer emerges as a conserved feature. Similarly, the implication of BMP signaling in the regulation of nodal expression is another feature that appears to be conserved between sea urchin and vertebrates. Finally, the role of Vg1/GDF1 in promoting propagation of left-right asymmetry to more distant regions is a third feature that appears to be common to sea urchin and vertebrate embryos. In contrast, the role of Notch may not be conserved since in the sea urchin, unlike in vertebrates, the role of Notch signaling appears to be very indirect and temporally separated from nodal expression.
In vertebrates, the node plays the function of a left-right organizer. Left-right asymmetry first becomes apparent in and around the node and subsequently propagates to the rest of the embryo. In the sea urchin embryo, the first manifestation of left-right asymmetry is expression of nodal on the right side of the tip of the archenteron. Several lines of evidence strongly suggest that this asymmetry of mesendodermal precursors is crucial for establishment of left-right asymmetry in the ectoderm and that this asymmetry is transmitted to the ectoderm at later stages resulting in right-sided expression of nodal in the ciliary band (Figure 9). First, both the absence of nodal expression and bilateral expression of nodal in the mesoderm result in random expression of nodal in the ectoderm. Second, inhibition of Univin function on the right side forced nodal to be expressed on the left side of the ciliary band. Finally and importantly, using chimeras, we showed that inhibition of nodal function in the endomesoderm randomizes nodal expression in the ectoderm. Our results are largely consistent with results of Amemiya et al. who showed that ablation at gastrula stage of the tip of the archenteron together with part of the ectoderm on the right side reversed positioning of the rudiment in 70% of the embryos while excision that removed the right ectoderm but left the archenteron intact had a more much more modest effect on left-right asymmetry, reversing positioning of the rudiment in only 30% of the embryos [45]. We therefore propose that the nodal expressing mesodermal cells located at the tip of the archenteron may therefore play the role of a left-right organizer similar to the node of vertebrates (Figure 11). However, this organizer is only responsible for orienting the symmetry breaking and for making it directional. Left right asymmetry can be established in the absence of this organizer but it is not directional.
There is accumulating evidence that the BMP pathway plays a dual and crucial role in vertebrates both in promoting expression of nodal on the left side and in preventing nodal activation on the right side. Nearly as many studies have implicated BMP signaling in the repression of nodal expression on the right side [88]–[93] as in the positive regulation of nodal expression on the left side [94]–[97]. For example in the mouse embryo, a reduction of BMP signaling causes nodal to be expressed bilaterally in the lateral plate. In the sea urchin, inhibition of BMP signaling by injection of a bmp2/4 or alk3/6 morpholino into the egg or blocking BMP signaling specifically in the endomesoderm prevented nodal expression in the organizer at gastrula stage and randomized nodal expression at pluteus stage. Intriguingly, targeting of the BMP2/4 morpholino to either the left or the right side revealed that BMP signaling on the left side is required for nodal expression on the right side. Consistent with this idea, we found that BMP signaling is stronger on the left side of the archenteron at gastrula stage and that the sector in which pSmad1/5/8 is detected and the region where the nodal expressing left-right organizer is formed are complementary. Furthermore, the asymmetry of nodal expression in the left-right organizer was detected slightly before the asymmetry of BMP signaling. It is therefore likely that an initially symmetric BMP signaling participates in the induction of nodal expression on the right side and that asymmetric Nodal signaling is in turn responsible for the asymmetry of BMP signaling possibly by antagonizing BMP signaling in the dorsal-right sector of the endomesoderm. The fact that all the genes encoding BMP ligands and BMP antagonists are expressed symmetrically along the left-right axis (our unpublished data) is consistent with this idea. Taken together, these observations suggest that formation of the left-right organizer is regulated by a combination of both positive and negative regulatory interactions (Figure 10). On the left side of the archenteron, a repressive signal produced by the secondary mesoderm prevents nodal expression. On the right side of the archenteron, three signals cooperate to induce nodal in the left-right organizer. The first signal is Nodal/Univin produced from the ventral ectoderm, the second signal is a member of the FGF family of growth factors (the tissue that produces it is presently not identified), and the third signal is likely produced in the dorsal part of the archenteron downstream of BMP signaling.
In vertebrates, left-right asymmetry propagates from the node to the lateral plate. Elegant rescue experiments using transgenic lines driving expression of GDF1 in the node or in the lateral plate demonstrated that the activity of GDF1 in the node is required for expression of nodal in the lateral plate [23], [24]. In addition communication between the node and the lateral plate has been recently shown to require functional gap junctions in the adjacent endodermal cells [98]. It is unlikely that gap junctions are involved in long range communication between the ectoderm and the endomesoderm since genes encoding gap junction proteins (connexins, innexins) are absent from the sea urchin genome [99]. In contrast, we showed that Univin, a TGF beta related to Vg1 and GDF1, is critically required for long range signaling between the ectoderm and the endomesoderm and for propagation of the left-right asymmetry signal. This suggests that the role of Univin as a TGF beta critically required for long range signaling by Nodal during left-right patterning is an evolutionary conserved and probably ancient feature in the left-right determination pathway (Figure 11).
In vertebrates the expression of lefty1 in the midline is thought to play a crucial role in the initiation and maintenance of unilateral expression of nodal [100], [101]. In the sea urchin embryo, there is presently no argument to suggest that there is a midline similar to the lefty expressing midline of vertebrate embryos that would act as a barrier to prevent propagation of nodal expression to the right side. Consistent with this idea, lefty in the sea urchin embryo is not expressed in the midline. Despite the absence of expression of lefty in the midline, a robust expression of nodal on the right side of sea urchin embryos is established at the end of gastrulation. How is this asymmetric expression established? There is strong evidence that in the sea urchin like in vertebrates [102], the epigenetic system constituted by short range Nodal autoregulation and long range inhibition by Lefty plays a crucial role in restricting nodal expression [103]. Lefty is both a very potent and highly diffusible inhibitor of Nodal signaling in the sea urchin embryo and lefty expression shifts to the right side at the end of gastrulation. Any small bias of nodal expression towards the right side will therefore be amplified and maintained by the self enhancement and lateral inhibition mechanism resulting in a robust expression of nodal and lefty on the right side in the absence of any midline barrier. We propose that the function of the left-right mesendodermal organizer on the right side of the archenteron is to provide this initial bias of nodal expression and that the reaction-diffusion mechanism between Nodal and Lefty further amplifies this bias, establishing a stable nodal expression on the right side.
Of the three steps involved in establishment of left-right asymmetry, the first i.e. symmetry breaking, remains the most enigmatic. Our data in the sea urchin embryo, point to the endomesoderm as the site where the symmetry is first broken and identify the Notch, FGF and BMP signaling pathways as critical early actors in the molecular cascade leading to determination of laterality. However, many questions remain on the mechanism by which Notch signaling represses nodal expression on the left side. Does Notch signaling regulate nodal expression by promoting asymmetrical cell movements, as proposed in the chick or does Notch signaling regulate nodal expression by promoting the local production by mesodermal cells of a factor that inhibits nodal expression? To answer these questions, future experiments should attempt to identify the inhibitory signal X that prevents nodal expression on the left side and should define the identity of the cells that send it. Similarly, the identity of the FGF ligand that promotes nodal expression on the right side is presently unknown and whether there is any connection between these inhibitory (Notch/factor X) or activating (FGF, BMP) signals remains to be explored. Future experiments should also examine the mechanisms responsible for asymmetrical BMP signaling in the archenteron and clarify the mechanisms by which BMP signals promote nodal expression. Finally, two important questions that future experiments should address are: to what extent is the left-right organizer of the sea urchin embryo homologous to the left-right organizer of vertebrates and do the archenteron tip cells require cilia to fulfill their role of left-right organizing cells?
In conclusion, our results provide a framework for the future dissection of the molecular pathway that regulates establishment of left-right asymmetry in the sea urchin. Furthermore, they demonstrate a strong connection between two players of the left-right determination pathway that were previously thought to be largely independent: the H+/K+-ATPase and Notch signaling. Finally, in addition to regulating left-right asymmetry, Notch signaling plays multiple and crucial roles in the etiology of various cancers [104] and particularly in acute T cell leukemia (T-ALL). Our finding that omeprazole, an extremely well tolerated and world-wide standard drug used to treat gastritis and ulcers, inhibits Notch signaling in the sea urchin embryo may be of clinical interest. In line with this idea, previous studies reported that omeprazole has an antiproliferative effect on pancreatic or colon cancer cells leading to the suggestion that omeprazole treatments could be used to develop new therapeutic strategies [105], [106]. Our finding that omeprazole inhibits Notch signaling in echinoderm embryos raises the possibility that the effect of omeprazole on tumor reversion may be linked to inhibition of Notch signaling, an hypothesis that should be investigated in future studies.
Adult sea urchins (Paracentrotus lividus) were collected in the bay of Villefranche-sur-Mer. Embryos were cultured at 18°C in Millipore-filtered sea water and at a density of 5000 per ml. Fertilization envelopes were removed by adding 1 mM 3-amino-1,2,4 triazole (ATA) 1 min before insemination to prevent hardening of this envelope followed by filtration through a 75 µm nylon net [107].
Treatments with the γ-secretase inhibitor DAPT (10–30 µM in sea water, Calbiochem), omeprazole (150–200 µM in sea water, Sigma), U0126 (5–10 µM in sea water, Calbiochem), SU5402 (30–50 µM in sea water, Calbiochem) were performed by adding the chemical diluted from stocks in Dimethylsulfoxyde (DMSO) in 24-well plates protected from light at the desired time. As controls, DMSO was added alone at 0.1% final concentration. Treatments by these inhibitors were performed continuously starting after fertilization. Treatments with recombinant BMP4 protein (0.5 µg/ml) were started at the 16-cell stage.
Experiments involving treatments with pharmacological inhibitors (DAPT, omeprazole, U0126, SU5402) were repeated multiple times with the same results.
Larvae were reared in 2-liter beakers with constant stirring at a density of one larva per 5 ml. They were fed every day with a freshly grown culture of the unicellular alga Isocrysis thaliana at a density of about 1000–5000 cells per ml. The presence and position of the rudiment was scored with a dissecting microscope after 3–4 weeks of culturing, and the larvae were photographed with a Zeiss Axiophot with dark-field and DIC illumination. To observe metamorphosis, single larvae competent to metamorphose were transferred to a Petri dish and observed at regular intervals. Metamorphosis was usually completed in 1–3 h.
Embryos devoid of fertilization envelopes were operated in Ca2+-free artificial sea water. Embryos microinjected with the nodal-Morpholino and a fluorescein-lysine dextran (FLDX) at the 16–32-cell stage were placed in a Kiehardt chamber on a dissecting microscope and vegetal halves were recombined to animal halves of unlabeled control embryos at the same stage in Ca2+-free seawater. Thirty-six hours post-fertilization, the embryos were imaged using a fluorescent microscope to record morphology and the presence of the dyes. The embryos were then fixed individually and analyzed by in situ hybridization with a nodal probe.
Immunostaining with the phosphoSmad1/5/8 antibody was performed as described by Lapraz et al. 2009 [65].
Morpholino antisense oligonucleotides were obtained from Gene Tools LLC (Eugene, OR). Characterization of the nodal, BMP2/4, univin, alk4/5/7 and alk3/6 morpholinos has been described in [65], [108], [109]. The specificity of the alk4/5/7, alk3/6 and nodal morpholinos has been demonstrated by rescue experiments. In the case of Delta, we designed and tested two morpholinos. The phenotypes observed with the Delta morpholino were considered specific since this morpholino caused a phenotype identical to the phenotype caused by DAPT treatment or by injection of a dominant negative form of Delta (truncation of the cytoplasmic domain). This phenotype is characterized by development of embryos lacking secondary mesenchymal cells at the tip of the archenteron during gastrulation [42] and lacking pigment cells and blastocoelar cells at later stages [48]. The phenotypes observed were therefore very consistent with the zygotic expression pattern and with previous well-established functional data. These phenotypes are very similar to those caused by inhibition of Notch signaling in other species [42], [79]. The sequences of all the morpholino oligomers used in this study are listed below. The most efficient morpholino of each pair is labeled with a star. Delta morpholinos are both directed against the 5′ UTR of the Delta transcript.
Delta Mo1*: 5′-GTGCAGCCGATAGCCTGATCCGTTA-3′.
Delta Mo2: 5′-CTTTTCTTATCAGTCCAAACCAGTC-3′.
univin Mo1*: 5′-ACGTCCATATTTAGCTCGTGTTTGT-3′.
univin Mo2: 5′-GTTAAACTCACCTTTCTAAACTCAC-3′.
nodal Mo1*: 5′-ACTTTGCGACTTTAGCTAATGATGC-3′.
nodal Mo2: 5′-ATGAGAAGAGTTGCTCCGATGGTTG-3′.
alk4/5/7 Mo 1: 5′-TAAGTATAGCACGTTCCAATGCCAT-3′.
alk3/6: Mo1: 5′-TAGTGTTACATCTGTCGCCATATTC-3′.
bmp2/4 Mo1*: 5′-GACCCCAGTTTGAGGTGGTAACCAT-3′.
bmp2/4 Mo2: 5′-CATGATGGGTGGGATAACACAATGT-3′.
Morpholino oligonucleotides were dissolved in sterile water and injected at the one-cell stage together with Tetramethyl Rhodamine Lysine Dextran (RLDX) (10000 MW) at 5 mg/ml or Fluoresceinated Dextran (FLDX) (70000 MW) at 5 mg/ml. Fluoresceinated Dextran is used as a lineage tracer of the injected cell. For each morpholino a dose-response curve was obtained and a concentration at which the oligomer did not elicit non-specific defect was chosen. Approximately 2–4 pl of oligonucleotide solution at 0.5 mM were used in most of the experiments described here. For morphological observations, about 150–200 eggs were injected in each experiment. To analyze gene expression in the morphants a minimum of 50–75 injected embryos were hybridized with a given probe. All the experiments were repeated at least twice and only representative phenotypes observed in more than 80% of embryos are presented.
Synthesis of capped mRNA coding for Nodal and Univin are respectively described in [66] and [109]. The pCS2 Delta construct is described in [48]. The Notch NICD and NEXT constructs were derived from a full length Paracentrotus lividus cDNA clone. For the NEXT construct the coding sequence of Notch corresponding to the aminoacids 1570–2528 of Notch (from the lin12 repeats up to the end of the protein) was amplified and cloned in pCS2. For the NICD construct, a region corresponding to aminoacids 1728–2528 of Notch (starting immediately after the transmembrane domain and extending to the end of the protein) was similarly cloned into pCS2. Delta induced overproduction of pigment cells when injected at 500 µg/ml while mRNA encoding NEXT caused the same effect when injected at 1 mg/ml and mRNA encoding NICD when injected at 200 µg/ml.
The Genebank accession numbers for the sequences discussed in this paper are: Notch (JQ861276), Nodal (AAS00534), BMP2/4 (DQ536194), Alk3/6 (FJ976181), FoxA (ABX71819), Univin (ABG00200), Pitx2 (AAW51825), Sox9 (AAW51826), Delta (ABG00198), Gcm (ABG66953),PAPSS (DQ531774), GATA1/2/3 (ABX71821).
Gene regulatory network diagrams were constructed using the biotapestry program available at http://www.biotapestry.org/ [110].
Dual luciferase assays were performed with the Promega Dual Luciferase Reporter system (Promega). Microinjection of purified and linearized plasmids was carried out by established protocols [111]. In the case of RBPJ-K luciferase reporter, the linearized plasmid was injected at 3.5 µg/ml, together with Endo 16-Renilla DNA at 1 ng/µl and carrier DNA (Hind III digested sea urchin DNA) at 17 µg/ml. For induction of Delta/Notch signaling, Delta mRNA was used at 500 µg/ml, NICD (Notch Intracellular Domain) mRNA at 200 µg/ml and NEXT (Notch extracellular truncation) RNA at 1000 µg/ml. For each measurement, 200 embryos were injected, collected at hatching blastula stage then lyzed following the manufacturer's instructions. The level of RBP1 derived Firefly Luciferase was detected according to the manufacturer's instructions using a GloMax luminometer with an integration of 10 s. The level of luciferase activity was normalized to the level of Renilla activity for each experiment. All the experiments were repeated two to three times using separate batches of embryos.
In situ hybridization was performed following a protocol adapted from Harland et al. 1991 [112] with antisense RNA probes and staged embryos. Probes derived from pBluescript vectors were synthesized with T7 RNA polymerase after linearization of the plasmids by NotI, while probes derived from pSport were synthesized with SP6 polymerase after linearization with SfiI. Control and experimental embryos were developed for the same time in the same experiments. The nodal, univin, pitx2, sox9 probes have been described already respectively in [10], [66], [109]. For double fluorescent in situ hybridizations, embryos were incubated overnight in hybridization buffer with the two probes. The nodal probe was labeled with Digoxigenin (DIG mix from Roche- Ref: 11277073910); The foxA and foxF probes were labeled with fluorescein (Fluo Mix frome Roche- Ref: 11427857910). After washing of the probes, embryos were incubated with an Anti-Digoxygenin Antibody coupled to HRP (Roche-ref: 11 207 733 910), diluted at 1/2000 overnight at 4°C, and staining was developed with the Cy3-Tyramide Signal Amplification System (TSA-Plus Kit-Perkin Elmer-Ref: NEL753). Embryos were rinsed with TBST until disappearance of background. The anti-digoxygenin-HRP antibodies were removed by treatment with Glycine, 0,1 M pH: 2.2, H2O2 1%, Tween 0.1% in TBST, and embryos were incubated with the Anti-Fluorescein Antibody coupled with HRP (Roche- Ref: 11 426 346 910), diluted 1/2000 during two hours at room temperature, and revealed with Cy2-tyramide signal amplification. Embryos were rinsed with TBST then mounted with City fluor and observed with a DIC and fluorescence microscope Axioimager.
To visualize the clones of injected cells after in situ hybridization, we used an antibody against fluorescein coupled to alkaline phosphatase. At the end of the in situ hybridization protocol, embryos were rinsed with PBST+EDTA 5 mM then incubated in a buffer containing glycine 0.2 M pH: 2.2, Tween 0.1% to inactivate the anti-digoxigenin antibody. Embryos were then washed six times in PBST, incubated in blocking solution (1% BSA, 2% Sheep serum inactivated in TBST) then with the anti-Fluorescein antibody coupled to Alkaline phosphatase (1/4000) at 4°C overnight. For Alkaline phosphatase staining, embryos were washed six times with TBST and briefly rinsed in Tris 100 mM pH: 8.2 and stained using FastRed as substrate in Tris 100 mM pH: 8.2. Staining was stopped by four rinses with PBST+EDTA 5 mM, then two rinses with PBST 25% Glycerol and 50% Glycerol. Embryos were then mounted and observed with a DIC microscope.
|
10.1371/journal.pntd.0004144 | Copulation Activity, Sperm Production and Conidia Transfer in Aedes aegypti Males Contaminated by Metarhizium anisopliae: A Biological Control Prospect | Dengue is the most prevalent arboviral disease transmitted by Aedes aegypti worldwide, whose chemical control is difficult, expensive, and of inconsistent efficacy. Releases of Metarhizium anisopliae—exposed Ae. aegypti males to disseminate conidia among female mosquitoes by mating represents a promising biological control approach against this important vector. A better understanding of fungus virulence and impact on reproductive parameters of Ae. aegypti, is need before testing auto-dissemination strategies.
Mortality, mating competitiveness, sperm production, and the capacity to auto-disseminate the fungus to females up to the 5thcopulation, were compared between Aedes aegypti males exposed to 5.96 x 107 conidia per cm2 of M. anisopliae and uninfected males. Half (50%) of fungus-exposed males (FEMs) died within the first 4 days post-exposure (PE). FEMs required 34% more time to successively copulate with 5 females (165 ± 3 minutes) than uninfected males (109 ± 3 minutes). Additionally, fungus infection reduced the sperm production by 87% at 5 days PE. Some beneficial impacts were observed, FEMs were able to successfully compete with uninfected males in cages, inseminating an equivalent number of females (about 25%). Under semi-field conditions, the ability of FEMs to search for and inseminate females was also equivalent to uninfected males (both inseminating about 40% females); but for the remaining females that were not inseminated, evidence of tarsal contact (transfer of fluorescent dust) was significantly greater in FEMs compared to controls. The estimated conidia load of a female exposed on the 5th copulation was 5,200 mL-1 which was sufficient to cause mortality.
Our study is the first to demonstrate auto-dissemination of M. anisopliae through transfer of fungus from males to female Ae. aegypti during mating under semi-field conditions. Our results suggest that auto-dissemination studies using releases of FEMs inside households could successfully infect wild Ae. aegypti females, providing another viable biological control tool for this important the dengue vector.
| Dengue virus (four serotypes) is transmitted primarily by the mosquitoes Aedes aegypti and currently 2.5 billion people are in risk of being infected. The incidence of this neglected disease is increasing in developing countries where communities have not been able to effectively remove mosquito sources and their economies cannot afford the vector chemical control. Our study is collecting some of the baseline information necessary to evaluate a novel biological control strategy that would release Ae. aegypti males mosquitoes infected with the fungus Metarhizium anisopliae. These fungus-infected males transfer the fungus to female mosquitoes through leg contact during the mating process (This is called auto-dissemination). The fungus killed 50% of males within 4 days of being exposed. The fungus infection also increase the time that males need to mate with 5 females mosquitoes and reduced the sperm production by 87% 5 days after being infected. The ability of fungus-exposed males (FEMs) to find and mate with females in the laboratory or a small greenhouse was the same as for uninfected males. In particular, in the small greenhouse FEMs made more mating attempts without insemination than the uninfected males (more than twice the attempts). During both attempts and successful matings, the FEMS were able to transfer fungus to females confirming that auto-dissemination does occur. We also showed that the amount of fungus transferred to female, even after the 5thmating (about 10% of male’s conidia load) was sufficient to kill 50% of females within 3 days. These results indicate that there is potential for auto-dissemination of M. anisopliae from males to females as a dengue control tool.
| Aedes aegypti is the principal vector of the four dengue (DENV) virus serotypes [1]. Although its control through larval source removal is effective, the only rapid but inconsistent way to interrupt epidemic transmission is by chemical insecticides [2,3]. The scarcity of natural enemies of Ae. aegypti [4,5] has led to promising research into biocontrol with entomopathogenic fungus. Metarhizium anisopliae and Beauveria bassiana have been examined by direct exposure of larvae to conidia (asexual, non-motile fungus spores) in water/oil, and through contact of resting adults on fungus-impregnated black clothes/nets [6–11]. Furthermore, M. anisopliae also reduces Ae. aegypti vectorial capacity by interfering with dengue virus replication; females co-infected with M. anisopliae and DENV-2 had lower viral loads in heads compared to females infected only with DENV-2 [12]. Metarhizium anisopliae is a hyphomycetous insect-pathogenic fungus of which the conidia infect insects by penetrating the cuticle. Metarhizium spp. are endemic worldwide and are not harmful to birds, fish, or mammals including humans [13]. Their pathogenicity/toxicity/allergenicity has been studied intensively representing only minimal risk to vertebrates, the environment and public health [14].
Auto-dissemination—the transfer of agents from infected organisms to others in a population—has been shown to occur through tarsal contact during copulation for entomopathogenic fungi associated with agricultural pests [15–17], but information for human disease vectors is far more limited. They include the transfer of M. anisopliae and B. bassiana between female and male Glossina morsitans morsitans that has been reported [18], and successful auto-dissemination of M. anisopliae from female to male Anopheles gambiae under laboratory conditions [19]. M. anisopliae conidia transfer from fungus-exposed males (FEMs) to female Ae. aegypti [20] could provide an additional tool for integrated dengue vector control programs through intradomicile releases of FEMs. The mating behavior of Ae. aegypti might favor auto-dissemination through the release of FEMs, because the polygamous males do not discriminate between virgin or mated female mosquitoes [21], although the insemination occurs only in the first 5 to 7 female mosquitoes [22]. Additionally, a small but potentially significant portion of females have been observed to mate multiple times within a 48-hour period under semi-field conditions [23]. In fact initial studies looking at conidia-transfer from FEMs to females in captivity have been promising [24]; a single M. anisopliae-exposed male after 48-h confinement with 30 female mosquitoes, infected the 85% of the females and killed the 50% in 7 days with a 90% sporulation and 99% fecundity reduction [20]. Prior to conducting a small-scale field trial of M. anisopliae auto-dissemination, it is critical to investigate numerous parameters; this paper contributes by extending baseline semi-field experiments [23] and evaluating the fungal effect on male sexual performance and the conidia transfer through successive copulations. Understanding any impact on the quantity of sexual encounters is relevant because the fungus is transferred from males to females by tarsal contact, which is the first step in the Ae. aegypti copulation process [25].
This study examined baseline parameters necessary to evaluate the potential of fungus auto-dissemination via the release of FEMs as a biocontrol tool, including the direct impact of fungal infection on male mosquito survival, mating success both under laboratory and semi-field conditions, copulation speed, sperm production, and finally the impact of conidia load on the FEMs to transfer lethal doses to exposed females.
Survival and conidia-transfer experiments described below used 4–7 day-old, sugar-fed virgin male and female mosquitoes from an Ae. aegypti colony established in 2006 in Monterrey, NL, Mexico and maintained as described in Garcia-Munguía et. al. (2011). Experiments to evaluate sperm production in Ae. aegypti males exposed to M. anisopliae compared to controls were carried out in 3 day-old, sugar-fed virgin males.
All reported experiments utilized the Ma-CBG-2 strain of M. anisopliae at an experimental conidial dose (ED) of 5.96 x 107conidia cm-2 prepared as previously reported [20]. Briefly, the fungus was cultured in potato-dextrose-agar plates incubated at 25 ± 2°C for 20 days in the dark. The conidia yield was estimated by using a mixture of 0.5% Tween-20 and 0.5% Triton-X in 0.85% saline solution. Spore suspension was centrifuged at 3,500 rpm for 10 minutes diluted up to 1.6 x 108 conidia mL-1 with a hemocytometer, and 5 ml were applied to a 2.5 μm pore, 8 cm diameter filter paper to set up the ED. Both treated and untreated filter papers were placed in chambers described previously [12], where mosquitoes were confined for 24-h. Insectary conditions were maintained at 25 ± 1°C, relative humidity (RH) of 80 ± 5%, and a photoperiod of 14:10-h L: D.
Three replicate experiments were conducted, each compared 25 males mosquitoes exposed to ED treated filters with 25 male mosquitoes exposed to untreated control filters for 24-h. After exposure mosquitoes were transferred to separate 1-liter translucent plastic flasks covered with a mesh and a cotton pad soaked with 5% sucrose. The flasks were monitored daily for mortality. Cadavers were removed and twice submerged in 1% sodium hypochlorite, washed in distilled water, and placed in humid chambers for fungus sporulation.
The ability of FEMs to compete with uninfected males and successfully copulate with females was examined with the aid of fluorescent powders used to mark the male mosquitoes. Before application of the fluorescent powder, 10 male Ae. aegypti were transferred to a 180 ml paper cup covered with mesh and anaesthetized by exposure to 4°C for 5 minutes. The cup was placed in a plastic bag and each powder was applied by filling a syringe (5 ml with 0.6 × 25 mm needle) with fluorescent powder up to 0.5 ml. The syringe was inserted through the mesh at the top of the cup and with one gentle push the powder was blown out of the syringe marking the mosquitoes inside the cup [26]. The FEMs were marked with red powder and uninfected males were marked with yellow powder. Next, the mosquitoes were anaesthetized again by exposure to 4°C for 5 minutes for use in the following experiments.
For each replicate, 20 male Ae. aegypti mosquitoes were exposed 24 hrs, either to fungus (3 replicates) or a clean control filter (3 replicates). Immediately after exposure to treated or clean filters, the 20 males per replicate were individually isolated in 1-liter translucent plastic bottles covered with a mesh and a pad soaked in 5% sucrose. Then, after 30 minutes of rest, 5 virgin female mosquitoes were confined successively with the same male. For each male, observations began at 9:00 am and concluded when he copulated with the 5th female (a process completed in less than 4 hours in the experiment). Here, “copulation” was the direct genital contact of a couple that lasted at least for 10 seconds [21]. Immediately after copulation the female was removed and replaced by a new one with a mouth aspirator. For all replicates the same 2 people measured the time to each copulation using chronometers. The total time (minutes) for each Ae. aegypti male to copulate successively with 5 different female mosquitoes was the average of two observations. So, the RV registered here was total time as the minutes elapsed since the introduction of the 1st female up to the copulation with the 5th one.
The sperm production in 3-day old male Ae. aegypti after exposure to M. anisopliae was examined. Two treatments were tested: males exposed for 24-h to the ED, plus a control where the males were exposed to clean filters for 24-h. Each treatment had 3 replicates. Initially for each replicate (FEM, control), approximately 600 eggs (taken from different cages) were placed in 3 enamel trays with 1 liter of distilled water. After 24-h, 300 first instar larvae were transferred to a tray (50 x 50 x 10 cm) with 5 liters of distilled water (60 larvae/liter) and fed daily with 1.5 g of dog food. Then, to acquire mainly male pupae, 100 small pupae within the first 24-h of pupation per tray were transferred to 1-liter plastic cup, which was placed in a cage for adult emergence with a pad soaked in 5% sucrose. Any females that emerged were removed, and when the males were 2-day old, they were exposed to the fungus or control filters. Starting the following day (3-d old), 10 males mosquitoes were removed for dissection each day for 5 consecutive days per day (n = 50). The males were immobilized by exposure to 4°C for 15 minutes and the spermatozoa were counted [27]. Testis and seminal vesicle were dissected gently, placed in a culture multi-well plate with 50 μL of phosphate-buffered saline (PBS), and added 150 μL of PBS to have 200 μL of stock solution per male per well. From each well, 5 μL was extracted and deposited in a concave microscope slide. Slides were dried at laboratory conditions (23 ± 2°C, 60 ± 10% RH) for 1-h, washed in 70% ethanol, stained with Giemsa for 1-h, washed 5 times in 1 ml of distilled water and again dried at laboratory conditions. Spermatozoa heads stained red were counted under a 40x microscope and multiplied by 40 to convert the spermatozoa number observed in 5 μL to the total number in 200 μL of stock solution per male.
To measure the conidial load attached to the body of male mosquitoes, 200 males earlier exposed to the ED for 24-h were placed in a 5-liter plastic flask covered with a mesh and then killed immediately by exposure to -20°C for 3 minutes. Next, 20 pools of 3 males (n = 180) each placed in 1.5 mL centrifuge tubes containing 500 μL of the same mixture used for conidia harvest. Three replicates were carried out for a total of 60 pools of 3 males each. Conidia were removed from the cuticle of the 3 males per tube by vortexing 3 times for 1 second, the mosquitoes were discarded, and each tube containing the conidia was centrifuged at 5,000 rpm for 5 minutes. The pellet with conidia was re-suspended in 20% lactophenol-blue solution [28] and 3 conidia counts by hemocytometer were used to determine the mean ± SE of conidia mL-1 per pool. Likewise, a scanning electron microscope (SEM) was used to snap the conidia attached to the cuticle of front tarsi of 10 fungus-contaminated males prepared as described previously [29]. Males were killed by freezing, dissected gently, and mounted on metal stubs, then glued with copper paint, gold-coated, and examined through the SEM.
The conidial load delivered to female mosquitoes during the 1st and 5th successive copulation was compared between females copulating with FEMs (males exposed to fungus for 24h) or uninfected males exposed to clean filters. Three replicates of 20 males each were carried out. Each replicate was contained in a 1-liter transparent plastic bottle covered with a mesh and a pad soaked in 5% sucrose. Individual males were introduced into the experimental chamber, and then after 30 minutes of rest, 5 virgin female mosquitoes per treatment were confined successively with the same male. After copulation, each female was removed promptly with a mouth aspirator, and a 2ndone was introduced into the flask to induce copulation, and the process repeated until the 5th female was introduced and copulated by the same male. Afterwards, the conidial load adhered to the bodies of the females that participated in the 1st and 5th copulations was estimated on the same day as described for males above.
The survival of females after exposure to 5,000 conidia mL-1 by topical application of 300 μL per insect was assessed. This dosage was chosen to represent the quantity of conidia that FEMs are able to transfer to females during the 5th copulation (see Results). Two treatments were tested: females that received on the thorax 300 μL of a stock of 5000 conidia mL-1 and those females that received the same volume of the mixture used to yield conidia but free of fungus. Each treatment had 3 replicates and 10 females per replicate; mortality was registered daily up to the death of the last female while cadavers and sporulation were processed as described above.
For survival, the median lethal time (LT50) in each treatment was computed by the Kaplan-Meier model and compared by a χ2 log rank test; the survival of treated and uninfected individuals was compared separately for males and females. To compare copulation activity between FEMs and control males in the laboratory, the following model was constructed: Females = treatment (inseminated by FEM, inseminated by male, inseminated by both, not inseminated) + day + treatment*day + error, which measured the variation of the least square means (LSMs) of female mosquitoes among treatments for the entire set of 10 days, within each day, and the interaction between treatment and day. For the green-house experiments, the variation of the LSMs of the RV was analyzed among treatments and copulation status by the model: Females = treatment + copulation status + error, where “copulation status” was defined as a group of females with presence or absence of insemination/M. anisopliae-infection. To assess the effect of M. anisopliae on the time (minutes) required by a male to copulate 5 successive female mosquitoes, a dataset was constructed for total time for the 2 treatments (24-h and control). Sixty data points were gathered per treatment and 120 data points for the full dataset. The LSM of “minutes” per male was examined among treatments and replicates with the model: Minutes = treatment + replicate + error. For the sperm production, the LSMs for treatments (24-h and control) were computed from the counts of spermatozoa per male, and compared between treatments for the 5 days, among days within the same treatment and for the interaction treatment*day by the model: Females = treatment + day + treatment*day + error. To examine the conidial loads, the arithmetic means (rounded to the nearest digit) from 3-mosquito pools of the load of males, the load of females of the 1st and the 5th copulation, were used as data to estimate and compare the LSMs by the model: Load = treatment + copulation + pool + error.
All the models were negative binomial (NB) regressions ran by the procedure glimmix with SAS 9.4, a method that computed the LSMs and ran F and t tests by Tukey-Cramer multiple comparisons to measure the variation of LSMs among the explanatory variables (qualitative and quantitative). The procedure also estimated the robustness of each model by the goodness of fit of the RV to the NB distribution by the ratio Pearson χ2 / freedom degrees (total observations), which should be ≤ 1 [30–31].
All Ae. aegypti male mosquitoes died within 6 days post-exposure (PE), whereas survival in the control extended beyond 30 days. The LT50 for FEMs was 3.69 ± 0.16 days compared to 23.62 ± 0.58 days for uninfected males (χ2 = 168.96, df = 1, p < 0.0001). The sporulation rate in cadavers of FEMs was 100% indicating that all Ae. aegypti males deaths were indeed caused by fungal infection. The survival of female mosquitoes exposed to 5,000 conidia mL-1 by topical application was significantly shorter (LT50 = 3.36 ± 0.25 days) than for uninfected females (LT50 = 25.80 ± 0.60 days; χ2 = 65.06, df = 1, p < 0.0001).
In the laboratory, there was no difference in the ability of FEMs to find and copulate with confined females with that of uninfected males. The LSM number of inseminated females was 5.47 ± 0.75 for FEMs, and 5.59 ± 0.76 for uninfected males. Interestingly, 3.21 ± 0.61 females had and just one female had mixed inseminations whereas only 1.22 ± 0.17 females had no evidence of copulation (Fig 3). The LSMs of female mosquitoes were different only for the explanatory variable “treatment” in the model (F = 24.93, df = 3, p< 0.001) meanwhile day and interaction treatment*day were not significant. The model was reliable; the NB goodness of fit test showed a ratio Pearson χ2/freedom degrees of 0.58.
In the small greenhouse, there was no difference in the ability of FEMs and uninfected males to search for and contact females. The LSMs number of females were not affected by treatment and day, but only by copulation status (F = 9.31, df = 4, p<0.0001) in the model, which was robust with a ratio Pearson χ2/ freedom degrees of 0.69; that is there was no statistical difference between the average number females inseminated by FEMs (9.86 ± 1.44) or controls (8.17 ± 1.17). Furthermore the number of females that were grasped (marked with red powder) but not inseminated (copulation attempts) by FEMs (7.48 ± 1.18) was greater than controls (with yellow powder) captured by uninfected males (3.25 ± 0.62); though both groups did not differ from total females (with no powder marking) that were not contacted by any male (5.10 ± 0.50) (Fig 4). During this experiment the daily temperature varied between 28°C and 35°C, and RH between 68 and 88%.
The total time (minutes) required by males to copulate with 5 successive female mosquitoes varied between treatments (F = 97.36, df = 1, p< 0.0001). The ratio Pearson χ2/freedom degrees of 0.98 indicated an acceptable NB goodness of fit. Therefore, the LSMs for 24-h exposure and control were 165.61 ± 3.58 and 109.23 ± 2.80, respectively. The total time invested by FEMs was 34% longer than that of uninfected males. For FEMs the total time to attempt or to successfully copulate ranged between 110 and 220 minutes PE, whereas for uninfected males the range was 50 to 150 minutes PE (Fig 5).
The LSM of spermatozoa per male differed between treatments (F = 1015.31, df = 1, p< 0.0001), among days (F = 12.42, df = 4, p< 0.0001), and in the interaction treatment*day (F = 167.85, df = 8, p< 0.0001). The ratio Pearson χ2/ degrees of freedom of the NB goodness of fit test was 0.97. On day 1 PE, the LSM of spermatozoa per male after 24-h fungus exposure was 1,029 ± 49, which was 46% less than the 1,913 ± 36 spermatozoa in uninfected males. At day 5 PE, the LSM was 550 ± 30 spermatozoa per FEM, which was 92% less than the spermatozoa of uninfected males that had increased up to 6,601 ± 16 per male. Across the entire 5 day evaluation, sperm production in FEMs was reduced by 47% but was augmented by 71% in uninfected males.
The LSM of conidia per pool varied significantly between treatments (F = 63.51, df = 2, p< 0.0001) in the model. The ratio Pearson χ2/freedom degrees of the NB goodness of fit test was 0.29. The LSM estimated per pool of three FEMs was 147,866± 21,064 mL-1, equivalent to 49,288 ±7,021 mL-1 per individual FEM. The SEM photograph indicated that conidia layers and clumps of polyhedronic shape of conidia remained attached on cuticle of front tarsi (Fig 6). The LSM conidial load per pool of 3 females from those that participated in the 1st copulation was of 31,348 ± 4,507 spores mL-1(10,449 ± 1,502 mL-1 per individual female), whilst the same for those involved in the 5th copulation was 15,811 ± 2,285 conidia mL-1(5,270 ± 761 mL-1 per individual female). Therefore, the conidial load of females of the 5th copulation was 50% lower than for females from the 1st copulation, and only 10% of the conidial load of males.
Regarding direct impacts on survival, the M. anisopliae strain (CBG-Ma2) had a LT50 of 4 days, which was similar to that reported for other strains [7–8, 10]. Although it is difficult to compare results across different studies, here the highest copulation rate (inseminated and not) of 75% recorded for FEMs is comparable to the 65%–85% range reported when 5 males were confined for 24 hours with 20 females in large field cages [32]. Concerning the parameters pertaining to mating activity, both negative and beneficial results were recorded. The negative impacts were that a longer time was invested by FEMs to successively copulate with 5 females than uninfected males. Also fungus infection reduced the sperm production by 87% by day 5 PE.
It is well established that following six inseminations in rapid succession, sperm becomes depleted in Ae. aegypti males, and they require 3 days of “sexual resting” to replenish the seminal vesicles [22–33]. Therefore our results suggest that if FEMs are released in field, they will only spread the pathogen during their first swarming event, but this should subsequently result in decreased oviposition by exposed females.
Fungus infection, however did have impacts that were beneficial for the potential use of auto-dissemination as a biocontrol tool. There was no difference in the ability of FEMs to find and copulate with females in the laboratory or small greenhouse when compared with uninfected males. More interestingly, in the small greenhouse the FEMs made more mating attempts without insemination than the uninfected males. This difference indicates that the fungus increases male copulations by 27%, which would facilitate auto-dissemination control strategies. The reason for this observed difference however, is difficult to explain. One hypothesis is this increase in mating attempts is an indirect effect associated with the innate immune response of male Ae. aegypti. It is possible that the fungus modulates male sexual behavior to increase its dispersal and propagation, a phenomenon noted in other fungus/insect interactions [34].
The tarsal contact occurring during attempted and successful copulations clearly transfers conidia from the FEMs to females. When an Ae. aegypti male copulates, he uses his front tarsi to hold the female’s hind femora and other legs [21, 25]; and with our ED and exposure methodology we observed the M. anisopliae conidial loads primarily on the legs of the FEMs (Fig 6). The 5th female which was copulated received a load of 5,200 conidia mL-1 which was 10% of that on the males (49,000 mL-1). Further, through direct topical application, we observed that this quantity of conidia was sufficient (LT50 = 3 days) to cause significant mortality of the females. This is supported by a recent study in which Ae. aegypti females infected with Indian strains of M. anisopliae, there was a mean lethal concentration (LC50) of 5,920conidia mL-1[35].
In conclusion, our study is the first to measure the effect of M. anisopliae on copulation behavior of male Ae. aegypti. Although the fungus killed the 50% of males in 4 days, the FEMs in semi-field conditions captured up to 15 females in successful copulations (8 with insemination) or mating attempts (7 with no insemination) during the first 3-h of confinement/release, and a single FEM transferred significant and lethal amounts of conidia to the first 5 females copulated successively. Our baseline results suggest that biological control of Ae. aegypti by releasing M. anisopliae-contaminated males to spread the pathogen by mating with wild females is feasible.
|
10.1371/journal.pcbi.1005178 | Individual Differences in Dynamic Functional Brain Connectivity across the Human Lifespan | Individual differences in brain functional networks may be related to complex personal identifiers, including health, age, and ability. Dynamic network theory has been used to identify properties of dynamic brain function from fMRI data, but the majority of analyses and findings remain at the level of the group. Here, we apply hypergraph analysis, a method from dynamic network theory, to quantify individual differences in brain functional dynamics. Using a summary metric derived from the hypergraph formalism—hypergraph cardinality—we investigate individual variations in two separate, complementary data sets. The first data set (“multi-task”) consists of 77 individuals engaging in four consecutive cognitive tasks. We observe that hypergraph cardinality exhibits variation across individuals while remaining consistent within individuals between tasks; moreover, the analysis of one of the memory tasks revealed a marginally significant correspondence between hypergraph cardinality and age. This finding motivated a similar analysis of the second data set (“age-memory”), in which 95 individuals, aged 18–75, performed a memory task with a similar structure to the multi-task memory task. With the increased age range in the age-memory data set, the correlation between hypergraph cardinality and age correspondence becomes significant. We discuss these results in the context of the well-known finding linking age with network structure, and suggest that hypergraph analysis should serve as a useful tool in furthering our understanding of the dynamic network structure of the brain.
| Complex patterns of activity in each individual human brain generate the unique range of thoughts and behaviors that person experiences. Individual differences in ability, age, state of mind, and other characteristics are tied to differences in brain activity, but determination of the exact nature of these relationships has been limited by the intrinsic complexity of the brain. Here, we apply dynamic network theory to quantify fundamental features of individual neural activity. We represent functional connections between brain regions as a time varying network, and then identify groups of these interactions that exhibit similar behavior over time. The result of this construction is referred to as a hypergraph, and each grouping within the hypergraph is called a hyperedge. We find that the number of these hyperedges in an individual’s hypergraph is a trait-like metric, with considerable variation across the population of subjects, but remarkable consistency within each subject as they perform different tasks. We find a significant correspondence between this metric and the subject’s age, indicating that the dynamics of functional brain activity in older individuals tends to be more dynamically segregated. This new insight into age-related changes in the dynamics of cognitive processing expands our knowledge of the effects of age on brain function and confirms our methods as promising for quantifying and examining individual differences.
| Functional connectivity (FC) analyses based on fMRI data are effective tools for quantifying and characterizing interactions between brain regions. Many approaches borrow methods from the field of graph theory, in which FC is used to build graphs that model the brain as a complex network, treating brain regions as nodes and using functional connections (pairs of nodes with significantly related BOLD signal dynamics) to determine the edge structure of the network [1, 2]. Individual differences in both underlying FC and the complex network structure resulting from graph theory approaches have been investigated for a variety of task states, developmental stages, and clinical diagnoses [3–5].
Certain characteristics of FC have been found to vary consistently over the course of normal human aging. The loss of clear segmentation between neural systems is widely reported: many intrinsic functional connectivity networks in the brain tend to become less internally coherent with age, and the functional differences between these intrinsic networks generally become less pronounced [6–8]. These changes are most commonly reported in the default mode network (DMN) [9–15], although they have also been observed in other networks, including those associated with higher cognitive functions [9, 11, 14–16]. In addition, inter-network connectivity between the DMN and other regions of the brain has been found to increase, diminishing the ability to discriminate between networks based on FC [13, 15]. There are some intrinsic functional networks, however, that show no changes or even increased intra-network connectivity with age, such as sensory networks [10, 12, 14].
The bulk of studies on age-related changes and other individual differences in FC, including those that use methods from complex networks and graph theory to represent FC patterns, are performed using static FC analysis, which represents the similarities of brain region activity (or some other measure of concordance) aggregated across an entire data set. In the present investigation, we build upon recent advances in network science to study individual differences in human brain activity and behavior from a dynamic network science perspective [17]. Dynamic functional connectivity (DFC) extends FC to examine how functional organization evolves over time [18, 19], allowing investigation of the changes in FC during the course of a cognitive task or scanning session. Efforts to probe the dynamics of functional brain networks have revealed that functional structure reconfigures over time in response to task demands [20–24] and spontaneously at rest [18, 25]. DFC methods have also been used to inform understanding of individual differences related to aging. In particular, dynamic community structure was found to vary significantly with age [26] and amplitude of low-frequency fluctuations of FC (ALFF-FC) was used to show age-dependent changes in the dynamics of interactions between networks [27]. Both studies imply that functional dynamics should be considered when investigating how aging affects brain network organization.
To address this, we use hypergraph analysis, a method from dynamic graph theory, to examine individual differences in DFC network structure in fMRI data acquired as subjects perform cognitively demanding tasks. The method is based on a generalization of standard graph theoretical techniques. In particular, by defining the standard node-node FC graph in successive temporal epochs, we construct a set of edge timeseries—that is, a vector of how the edge changes over time. The edge-edge DFC graph is constructed by treating these edge timeseries analogously to the node timeseries in the first step, and computing the relationship between every edge pair. Finally, we focus on “hyperedges,” which are connected components of the absolute valued edge-edge DFC graph (described in more detail in Methods) [28]. To contextualize hypergraph analysis, we define the graph theoretic elements used to construct hypergraphs as follows:
Node: As in the FC literature, nodes denote brain regions, or groups of voxels.
Edge: Also corresponding to the FC literature, edges denote correlations in activity between pairs of nodes over time. A node-node graph G = {V, E} on N nodes will have ( N 2 ) edges, because each pair is considered. Unlike the majority of FC analyses, the edges are not thresholded for significance in the hypergraph analysis.
Links: Links denote significant correlations in activity between pairs of edges over time. An edge-edge graph G′ = {V′, E′} on ( N 2 ) edges will have ( ( N 2 ) 2 ) possible links, but tends to be sparse in practice.
Hyperedge: A hyperedge denotes a group of links connecting two or more edges with significantly correlated temporal profiles. Hyperedges are the simplest form of link community, since they are simply the connected components of the edge-edge graph G′ = {V′, E′}, where V is the set of edges and E is the set of links.
Hypergraph: A hypergraph is a set of hyperedges.
The hypergraph analysis is a simple first step toward understanding the structure of functional dynamics. Hyperedges are the connected components of the edge-edge graph, and so avoid the introduction of additional unconstrained parameters, unlike many common FC and DFC methods such as community detection.
The groups of brain regions that comprise hyperedges are not necessarily strongly active or strongly interconnected brain regions. Rather, correlations in the dynamic connectivity of these regions are the defining characteristics that determine hyperedge structure. As a result, hyperedge analysis is able to identify groups of dynamic connections that change from strong to weak (or vice versa) cohesively together over time, providing complementary information to other DFC methods that focus on only the strongest node-node correlations, such as dynamic community detection [26, 29, 30]. Note that our choice of hyperedge metrics, as opposed to any other graph theoretic measure, is due to the simplicity of the hyperedge. Although it is beyond the scope of the present investigation, other graph properties of the edge-edge graph are likely to provide insight into dynamic brain network structure along other relevant dimensions. Nonetheless, hyperedges have some appealing intuitive validity in terms of the neural properties they might uncover—that is, in defining collections of nodes (or more technically, edges) on the basis of their similar dynamics.
In previous work, we demonstrated that hyperedges discriminate between diverse task states in a group-level analysis of an fMRI data set spanning four tasks, which we refer to as the “multi-task” data set [24]. We also observed notable variation in descriptive hypergraph measures across individuals. However, given the level of abstraction involved in the construction of the hypergraph, an important first question is whether the method is able to capture well-known phenomena. In this paper, we investigate the relationship between the variability in hypergraph cardinality and other individual difference measures. We develop and employ hypergraph measures that capture individual differences in functional brain dynamics to determine correspondences between dynamics and specific demographic and behavioral measures. In the multi-task data set, we find that hypergraph cardinality—the number of distinct hyperedges within a subject’s hypergraph—exhibits marked variation across individuals. At the same time, we find this measure is consistent within individuals, across overall hypergraphs and those associated with specific tasks.
To elucidate the drivers of this striking variation in hypergraph metrics observed across subjects, we explore systematic relationships between hypergraph cardinality and individual difference measures spanning distinct domains such as demographics, cognitive strategy, and personality. In the multi-task data set, we find a suggestive relationship between hypergraph cardinality and participant age. This relationship is confirmed with an independent analysis of a data set with participants who range in age from 18 to 75, which we refer to as the “age-memory” data set. We report a strong positive relationship between age and hypergraph cardinality: older participants are significantly more likely to have a larger number of distinct hyperedges in their hypergraph. This agrees with the widely reported phenomenon of the loss of cohesion within intrinsic functional brain systems, because an increase in the number of distinct hyperedges linking various brain regions points to interconnections between functional groups evolving in time [13, 15]. Thus, the hypergraph method agrees with previous descriptions of age-related brain changes, while capturing information about dynamics that adds a novel dimension to previous studies. This work further recommends the hypergraph as a useful tool in studying structure in dynamic functional connectivity.
Informed written consent was obtained from each participant prior to experimental sessions for the multi-task and age-memory experiments. All procedures were approved by the University of California, Santa Barbara Human Participants Committee.
The majority of the methods are identical to those discussed for the multi-task data set. Below, we point out aspects that differ between the two analyses.
As mentioned above, the hyperedge method has been applied to the multi-task data set in a previous study [24]. Here, we first recapitulate the key findings from that investigation and provide results of exploratory analyses that motivate the followup analyses on the age-memory data set. We then present results from the age-memory analysis.
A previous study of the multi-task data identified measures that capture significant differences in population-level hypergraph structure across tasks [24]. Furthermore, extensive variation was observed in several hypergraph measures, including hypergraph cardinality, across individuals. These results emphasize that hypergraph structure can be used to differentiate between task states and motivates our investigation of the correspondence between hypergraph structure and individual difference measures.
Fig 3 depicts the empirical cumulative hyperedge size distributions for all hyperedges found across all subjects in the multi-task data set. As a null test, we shuffle the data over time and find no hyperedges of size greater than one. There is a rough power law for the smaller sizes (s < 100), followed by a gap in the distribution from about 100 to 1000 and a sharp drop at the system size (s = ( 194 2 ) = 18721). The shape of the distribution is due to the consistent hypergraph structure across individuals; the majority of subjects in this study have a hypergraph composed of one large hyperedge and many small hyperedges. While this characteristic structure is common to most subjects in the study, the size of the largest hyperedge varies across individuals. This size is closely related to the hypergraph cardinality, defined as the number of hyperedges in a hypergraph, a measure which also exhibits large variation.
Fig 3 also depicts task-dependent differences in the cumulative size distributions of task-specific hyperedges. Memory-specific hyperedges tend to be more numerous than those specific to the rest and attention tasks. However, the total number of task-specific hyperedges for any task is at least ten times fewer than the total number of hyperedges. Our strict definition of task specificity includes only hyperedges specific to a single task and discards those associated with more than one task. This approach is conservative, and likely leaves some meaningfully task-related hyperedges unclassified. However, it reduces the dimension of the task-specific results, and provides greater confidence that any hyperedges classified as task-specific are indeed providing truly task-driven information due to coherence within that task alone, rather than coherence due to an unrelated driver that is common to several tasks.
There are significant differences in the spatial organization of task-specific hyperedges over all individuals that are visualized in Fig 4. The plots depict task-specific hyperedge degree across the brain for each of the four tasks. In addition to the differences in magnitude between word memory and the other tasks, the locations of high hyperedge concentration vary with task.
These significant differences in hypergraph structure between the tasks confirm that hypergraph structure varies between task states. However, persistent variability in hypergraph measures across individuals indicates that the hypergraph method reflects innate differences beyond the current task state. The work presented here follows this line of inquiry, beginning with an analysis of individual differences in the multi-task data set.
Here, we illustrate and quantify the wide variation in hypergraph measures across individuals in the multi-task data. In brief, we identify a particular measure, hypergraph cardinality, that demonstrates large variance across all individuals but is consistent within individuals. Following this, we investigate relationships between the variation in individual difference measures and the variation in hypergraph cardinality. The results from this study are not statistically significant due to the limited variation in individual difference measures and strict corrections for multiple comparisons. However, we report a marginally significant result relating demographics and word-memory hyperedge cardinality that motivates further analyses on the age-memory data set.
To supplement the findings from the multi-task data set, we perform a parallel set of analyses on the age-memory data set. The data set includes participants with ages ranging from 18 to 75, a range three times larger than the range of ages in the multi-task study. Furthermore, the age-memory study uses an almost identical task to the multi-task word-memory task. In this section, we combine hypergraph results for all participants in the age-memory data set and obtain a distribution of hyperedge size over all participants with similar features to the hyperedge size distribution from the word-memory task of the multi-task data. We then identify and test specific drivers of individual variation in hypergraph cardinality for the age-memory study participants. We find a strong correspondence between age and hypergraph cardinality that confirms the preliminary result from the multi-task study.
Improving our understanding of the drivers of individual differences in functional brain imaging data can give insight into the dynamic mechanisms that lead to individual behavior. Dynamic FC has been used over groups to explain changes in the brain attributed to individual differences in learning [30, 49, 63]. Hypergraphs in particular have been used to analyze how long-term learning impacts the functional network structure [30] and how the brain switches between cognitive states [24]. A previous DFC study found task-dependent hypergraph properties at the level of the group, indicating that hypergraphs can be used to describe how functional dynamics differ between tasks [24]. Here, we develop new hypergraph metrics to investigate individual differences in hypergraph structure and possible drivers of these variations. Our primary goal in the present investigation is to continue validating the hypergraph approach by demonstrating its ability to reproduce a well-known phenomenon in the FC literature.
Hypergraphs are constructed from correlations between edges, providing a method of analysis complementary to static and dynamic graph theoretic methods including dynamic community detection and ALFF-FC [26, 27]. In this method, hyperedges are defined as connected components of the edge-edge graph. A natural extension of the hyperedge formalism would be to perform edge-based community detection on the edge-edge adjacency matrix, which would further partition the connected components of the edge-edge graph [64, 65]. Similarly, any graph theoretic measure that can be computed on the standard node-node graph—clustering coefficient, assortativity, global efficiency, et cetera—can be computed with respect to the edge-edge graph, although the interpretation would of course be quite different.
The hypergraph method provides a rigourous graph theoretical formalism to study network dynamics. Throughout this study, we investigate hypergraph cardinality as a dependent variable. However, future investigations should be performed to determine whether hypergraph cardinality is a useful independent variable with predictive power.
As we showed in the Multi-Task Analysis, the hypergraph cardinality varies widely across individuals, but is consistent between task states. Previous work on the multi-task data set found that the probability for hypergraphs to appear in a particular network configuration over individuals was significantly different depending on task state [24]. Consistent spatial organization rules for each task existed at the level of the group. There were similarities in the spatial arrangement of hyperedges in the brain for differing tasks, but certain properties were found to vary significantly between tasks. Brain areas in the occipital lobe in particular were highly likely to participate in the hypergraph across individuals and across tasks, likely due to the visual nature of most of the cognitive tasks studied.
Here, we study hypergraph cardinality, which displays high variability across individuals and consistency across tasks within individuals (Fig 5). This indicates that hypergraph cardinality serves as an individual signature of a subject’s brain dynamics. The similarities across subjects in the spatial distributions of hypergraphs described in [24] capture information orthogonal to the information summarized by hypergraph cardinality. For example, there are some individuals for whom the visual brain regions are linked by many hyperedges, and some for whom those same regions are linked by relatively few hyperedges, but these regions are more likely than others to be included in hypergraphs in the majority of subjects. This suggests that, for some subjects, brain regions tend to be more dynamically integrated in general, with co-varying functional relationships across many brain circuits; in other subjects, connectivity dynamics are more fragmented across the brain.
The high degree of variability in hypergraph cardinality across subjects and consistency within subjects, combined with the significant differences in spatial hyperedge arrangement across tasks, indicate that hypergraphs are a useful analysis tool for investigating both individual and task-based differences in brain function in a variety of settings. At the same time, hypergraphs can provide a view of dynamic patterns that complements other commonly used DFC methods. For example, many FC methods exclusively investigate the structure of strong correlations in functional data [29, 66–68]; hypergraph analysis captures information about both strongly and weakly correlated dynamics and how sets of brain regions transition between them [28].
Although they are highly informative, many of the hypergraph metrics we study here are representative measures that greatly reduce the dimension of the hypergraph and only reveal a small part of the information contained in its structure. Further development of methods to utilize more of the information that hypergraphs provide will allow characterization of the consistency of particular hyperedges and dynamic modes, an understanding of which are important for behavior, or influenced by demographics or disease. Future work is also needed to further quantify the spatial differences in hypergraph arrangement across both individuals and tasks, to clarify the extent of overlap between the two types of information, and to determine whether the individual variability in cardinality can be mapped to individual spatial differences in hypergraph structure.
FC studies have established clear trends associated with aging, including a decrease in connectivity within functional networks and an increase in connectivity across different functional networks in resting and task states [15, 69–72]. Many of these studies have considered resting-state FC, because the absence of task stimulus provides a simple and reliable setting for comparison between subjects [73], although recent studies have successfully used FC networks to study various cognitive proceses [74]. The default mode network (DMN) and similar resting-state analyses may miss functional changes evoked by task states; while the DMN FC decreases with age, task-related sensorimotor network FC has been shown to increase with age [12, 14]. Similarly, FC in memory tasks shows increased segmentation with age [75]. Extending these analyses to incorporate the dynamics of functional interactions is a necessary step towards quantifying individual changes in functional brain dynamics associated with age.
Several efforts have been made to capture individual age-related differences with methods from dynamic FC. Dynamic community structure and amplitude of low-frequency fluctuation of FC were both found to be strongly correlated with age, illustrating that functional dynamics are closely linked with aging [26, 27]. In the dynamic community detection analysis, functional communities were found to be more fragmented with age, which agrees with the hypergraph cardinality result presented here [26]. A multi-scale community detection analysis uncovered similar fragmentation with age for small scales [76]. Our finding that hypergraph cardinality also increases with age aligns with this result and provides further information based upon its ability to capture higher-order dynamic patterns across larger ensembles of brain regions. Not only do the functional similarities of communities of brain regions themselves become less distinct as humans age, but the temporal profiles of these functional similarities also become less integrated across brain regions. The agreement of this result with known age-related changes in FC [6–8, 13, 15] demonstrates the ability of hypergraph methods to capture and quantify major brain changes. Moreover, since the hypergraph analysis is not limited to strong correlations, our analysis further suggests that age is related not only to the organization of functional activity in groups of brain regions with strongly coherent activity, but also to the coordination between groups of regions that transition from being strongly to weakly correlated over time (or vice versa).
The reported correspondence between age and hypergraph cardinality is significant in the age-memory data set, but our analysis did not include data that could verify this relationship for cognitive tasks other than the word memory task. Although memory is a cognitive ability known to decline with age in many individuals, it is unlikely that the specific task studied in the age-memory data set drives this result. Rather, the consistency of hypergraph cardinality across tasks seen in the multi-task data set in Fig 5(B) suggests that similar hypergraph cardinalities may be found during other tasks in data sets with higher age variability, and that the relationship between age and cardinality is unlikely to depend primarily on the behavioral task. Further investigation is needed to determine whether individual differences in hyperedge structure have any significant relationship to behavioral or cognitive performance on any particular task.
Here, we have shown that the considerable differences in functional connectivity dynamics across individuals are closely linked with age. The hypergraph method is presented as an analysis tool that captures information about group-level similarities that differ between task states as well as individual differences that are consistent within individuals, across tasks. Further investigation into a single hypergraph metric (hypergraph cardinality) that varies across individuals uncovers a significant relationship between hypergraph cardinality and age. Specifically, there are a greater number of hyperedges in older individuals’ hypergraphs, suggesting that there are more small groups of regions with cohesively evolving dynamics and indicating a loss of coherence across larger, spatially distributed intrinsic functional connectivity networks. This complements widely reported relationships between FC and human aging by providing new insight into how FC activity and the co-evolution of FC activity are altered with increasing age, including the loss of large groups of co-evolving brain regions in older individuals. The correspondence with and extension of classic FC results to new dynamic regimes, along with the unique capacity of hypergraphs to probe multiple dimensions of both strong and weak dynamic variability, show that hypergraph analysis is a valuable tool for understanding age-related changes and other individual differences in dynamic brain function.
|
10.1371/journal.pbio.1001255 | Development and Function of Invariant Natural Killer T Cells Producing TH2- and TH17-Cytokines | There is heterogeneity in invariant natural killer T (iNKT) cells based on the expression of CD4 and the IL-17 receptor B (IL-17RB), a receptor for IL-25 which is a key factor in TH2 immunity. However, the development pathway and precise function of these iNKT cell subtypes remain unknown. IL-17RB+ iNKT cells are present in the thymic CD44+/− NK1.1− population and develop normally even in the absence of IL-15, which is required for maturation and homeostasis of IL-17RB− iNKT cells producing IFN-γ. These results suggest that iNKT cells contain at least two subtypes, IL-17RB+ and IL-17RB− subsets. The IL-17RB+ iNKT subtypes can be further divided into two subtypes on the basis of CD4 expression both in the thymus and in the periphery. CD4+ IL-17RB+ iNKT cells produce TH2 (IL-13), TH9 (IL-9 and IL-10), and TH17 (IL-17A and IL-22) cytokines in response to IL-25 in an E4BP4-dependent fashion, whereas CD4− IL-17RB+ iNKT cells are a retinoic acid receptor-related orphan receptor (ROR)γt+ subset producing TH17 cytokines upon stimulation with IL-23 in an E4BP4-independent fashion. These IL-17RB+ iNKT cell subtypes are abundantly present in the lung in the steady state and mediate the pathogenesis in virus-induced airway hyperreactivity (AHR). In this study we demonstrated that the IL-17RB+ iNKT cell subsets develop distinct from classical iNKT cell developmental stages in the thymus and play important roles in the pathogenesis of airway diseases.
| T cells are a diverse group of immune cells involved in cell-mediated acquired immunity. One subset of T cells is the innate-like invariant natural killer T (iNKT) cells that recognize glycolipid ligands on target cells instead of peptides. We know that functionally distinct subtypes of iNKT cells are involved in specific pathologies, yet their development, phenotypes, and functions are not well understood. Here, we determine the relationship between various mouse iNKT cell subsets, identify reliable molecular markers for these subsets, and show that these contribute to their functional differences. We identify four iNKT cell subsets that we show arise via different developmental pathways and exhibit different cytokine profiles. Importantly, we show that these subsets can be isolated from the thymus (the organ of all T cells), as well as from peripheral tissues such as spleen, liver, lung, and lymph nodes. Contrary to the general understanding that iNKT cells mature after their exit from the thymus and their migration into peripheral tissues, we conclude that distinct phenotypic and functional iNKT cell subsets can be distinguished in the thymus by virtue of the presence or absence of the cytokine receptor IL-17RB and another cell surface molecule called CD4, and these subsets then migrate to peripheral tissues where they retain their phenotypic and functional characteristics. Regarding functional significance, we show that those iNKT cell subsets that lead to airway hyper-responsiveness to respiratory viruses are different to those that lead to allergen-induced airway hyperreactivity, which will enable researchers to focus on specific subsets as potential targets for therapeutic intervention.
| Natural killer T (NKT) cells, unlike conventional T cells bearing diverse antigen receptors, are characterized by the expression of an invariant T cell receptor (TCR), Vα14Jα18 paired with Vβ8, Vβ7, or Vβ2 in mice [1] and the Vα24Jα18/Vβ11 pair in humans [2],[3], that recognizes glycolipid antigens in conjunction with the monomorphic MHC class I-like CD1d molecule [4],[5]. Therefore, these cells are termed invariant NKT (iNKT) cells. Another characteristic feature of iNKT cells is their rapid and massive production of a range of cytokines, such as those typically produced by T helper cell (TH) 1, TH2, and TH17 cells [6]–[8], upon stimulation with their ligand, α-Galactosylceramide (α-GalCer) [9],.
It is speculated that the ability of iNKT cells to produce these various cytokines is due either to the microenvironment in which they undergo priming or to the existence of functionally distinct subtypes of iNKT cells producing different cytokines; however, there is no clear-cut evidence to support the latter notion. It has been reported that iNKT cells include both CD4+ and CD4− subtypes [6],[11], each of which produces different cytokines. Human CD4+ iNKT cells produce both TH1 and TH2 cytokines, whereas CD4− iNKT cells produce mainly TH1 cytokines [12],[13]. Although such functional differences were originally less apparent in mouse CD4+ and CD4− iNKT cells, two functionally distinct subtypes of iNKT cells in the mouse thymus have since been identified based on NK1.1 expression; NK1.1− iNKT cells produce a large amount of IL-4 and little IFN-γ, whereas NK1.1+ iNKT cells produce less IL-4 and more IFN-γ [14],[15]. Furthermore, CD4− iNKT cells in the liver have been found to be more effective in mediating tumor rejection than CD4+ iNKT cells in the liver or any other tissues [16].
There is also further heterogeneity of CD4+ iNKT cells in terms of expression of the IL-17 receptor B (IL-17RB), a receptor for IL-25 [17]. IL-25 is a key factor in TH2 immunity, including allergic reactions and airway hyperreactivity (AHR). The CD4+ IL-17RB+ iNKT cells produce large amounts of IL-13 and IL-4 but little IFN-γ in response to IL-25, mediating a key role in IL-25-driven AHR [17],[18]. Another subset of newly identified iNKT cells within the NK1.1− CD4− subset is the retinoic acid receptor-related orphan receptor (ROR)γt+ iNKT cells. These cells can induce autoimmune disorders by their production of IL-17A and IL-22 [8],[19], even though IL-17A-producing iNKT cells are not restricted to a particular iNKT cell subset [20].
The emergence of functionally distinct subpopulations of iNKT cells is reminiscent of how iNKT subtypes develop in the thymus and expand in the periphery. Here, we demonstrate IL-17RB+ iNKT cells are a subtype distinct from the CD122+ iNKT cells producing IFN-γ. Moreover, there are two subtypes of IL-17RB+ iNKT cells; CD4− produces TH17 cytokines in an IL-23-dependent fashion, whereas the other CD4+ produces TH2 and TH17 cytokines in an IL-25 dependent manner. In addition, these IL-17RB+ iNKT cells contribute to the induction of virus and viral antigen-induced chronic AHR.
We previously identified a fraction of splenic CD4+ iNKT cells that expresses IL-17RB and produces TH2 cytokines after treatment with IL-25 [17]. In order to directly analyze the function of IL-17RB on iNKT cells, we generated IL-17RB-deficient mice by the disruption of exon 1 and exon 2 of the Il17rb gene (Figure S1). We then compared the number and function of iNKT cells in the spleen and the liver from Il17rb−/− mice on a C57BL/6 (B6) background to those of wild type (WT) B6 mice. We also included in our comparison Il15L117P mice, in which leucine (CTT) at amino acid position 117 of IL-15 was substituted with proline (CCT), because IL-15 is reported to be critical for the development and homeostatic maintenance of iNKT cells [21],[22], as well as other cell types such as NK and CD8+ memory T cells [23],[24]. As shown in Figure 1A, the number of iNKT cells in Il17rb−/− mice was only slightly decreased in the spleen, and was almost comparable in the liver, compared with B6 mice, findings consistent with our previous analysis on the distribution of IL-17RB+ iNKT cells (3%–5% in the spleen and almost none in the liver) as detected by a specific monoclonal antibody [17]. Similarly, Il15L117P mice appear to recapitulate the previously reported phenotype of Il15−/− mice [21],[22] because iNKT and NK cells were decreased by 50% in the spleen and by 90% in the liver, indicating that the L117P mutation resulted in the loss of IL-15 function.
We then analyzed the frequency of IL-17RB+ subtypes among α-GalCer/CD1d dimer+ iNKT cells in the spleen and liver of WT, Il17rb−/−, and Il15L117P mice (Figure 1B). The percentage of IL-17RB+ iNKT cells was increased more than 4 times in the spleen and 10 times in liver of the Il15L117P mice.
By using Il17rb−/− and Il15L117P mice, we further analyzed the iNKT cell subtypes in terms of their ability to produce cytokines (Figure 1C and 1D). α-GalCer/CD1d dimer+ TCRβ+ iNKT cells from the spleen of WT, Il17rb−/−, and Il15L117P mice (Figure 1C) and those from the liver of B6 and Il17rb−/− mice (Figure 1D) were sorted and co-cultured with GM-CSF-induced bone marrow derived dendritic cells (BM-DCs) in the presence of α-GalCer. The Il17rb−/− iNKT cells produced normal levels of IFN-γ, but this was significantly decreased in Il15L117P iNKT cells. Intriguingly, there was impaired production of not only TH2 cytokines such as IL-9, IL-10, and IL-13, but also of TH17 cytokines IL-17A and IL-22 in Il17rb−/− iNKT cells, but not in Il15L117P iNKT cells in the spleen (Figure 1C and 1D), even though the number of iNKT cells were only slightly decreased in Il17rb−/− (see Figure 1A). The iNKT cells derived from WT, Il17rb−/−, or Il15L117P failed to produce any indicated cytokines when co-cultured with BM-DCs from Cd1d1−/− mice (unpublished data), indicating the cytokine production from iNKT cells are absolutely CD1d/α-GalCer dependent.
To examine the functional activity of Il17rb−/− iNKT cells in vivo, we administered α-GalCer (2 µg) intravenously (i.v.) and monitored serum cytokine levels (Figure 1E). The production of IFN-γ peaked normally at 12 to 24 h after stimulation in the Il17rb−/− mice. On the other hand, the production (around 1–6 h) of other cytokines, such as IL-9, IL-10, IL-13, IL-17A, and IL-22, was severely impaired in the Il17rb−/− mice. The results suggest that IL-17RB+ iNKT cells are distinct from IL-17RB− iNKT cells, which mainly produce IFN-γ, and also that IL-17RB+ iNKT cells produce IL-9, IL-10, and IL-13 among TH2 cytokines and IL-17A and IL-22 TH17-type cytokines.
iNKT cells in the spleen and liver from il17rb−/− mice are defective in the production of IL-9, IL-10, IL-13, IL-17A, and IL-22, while IFN-γ production is diminished in Il15L117P iNKT cells (Figure 1). We therefore attempted to identify the origin of IL-17RB+ iNKT cells in the thymus by comparing α-GalCer/CD1d dimer+ iNKT cells in B6 with those in Il17rb−/− and in Il15L117P mice on a B6 background (Figure 2A). The percentage and number of iNKT cells in the thymus were severely decreased in Il15L117P mice to a similar extent as previously reported in Il15−/− mice [21]. By contrast, the percentage and number of iNKT cells in Il17rb−/− mice was only slightly decreased, to a similar extent to that seen in the spleen and liver (Figure 1C and 1D). In order to analyze their phenotype precisely, enriched α-GalCer/CD1d dimer+ iNKT cells were further divided based on the expression of CD44 and NK1.1 (Figure 2B), because iNKT cells can be classified into developmental stages based on the cell surface expression of these molecules, i.e., CD44lo NK1.1− (Stage 1), CD44hi NK1.1− (Stage 2), and CD44hi NK1.1+ (Stage 3) [14],[25]. In agreement with earlier results [21], there was a decrease in the CD44hi NK1.1+ (Stage 3) population of α-GalCer/CD1d dimer+ iNKT cells in the thymus of Il15L117P mice. By contrast, the percentage and number of iNKT cells in Il17rb−/− mice were reduced, especially in the CD44lo NK1.1− (Stage 1) and CD44hi NK1.1− (Stage 2) populations, although the CD44hi NK1.1+ (Stage 3) population was unchanged (Figure 2B). To determine whether the reduction in absolute numbers of developmental Stages 1 and 2 iNKT cell populations in Il17rb−/− mice is due to a developmental defect or to bypassing of these developmental stages, we analyzed surface expression of IL-17RB and CD122, a receptor for IL-15 (Figure 2C). Consistent with the observation shown in Figure 2B, IL-17RB expression was detected mainly in the Stage 1 and Stage 2 populations in both CD4− and CD4+ fractions (Figure 2C), whereas CD122 expression was mainly in the Stage 3 population as previously reported [21], and is inversely correlated with the expression of IL-17RB (Figure 2C). In order to investigate whether IL-17RB+ iNKT cells are distinct from IL-15-dependent iNKT cells, thymic iNKT cells from B6 and Il15L117P mice were divided based on the expression of IL-17RB and CD4, and were further analyzed in the expression of CD44 and NK1.1 (Figure 2D and 2E). The percentage of CD4− and CD4+, IL-17RB+ iNKT cells was higher in Il15L117P mice (Figure 2D), due to the reduction in the numbers of IL-17RB− iNKT cells. Concerning the distribution of the expression of CD44 and NK1.1 in iNKT cell subtypes, even though IL-17RB+ iNKT cells comprised only ∼10% of the thymic iNKT cells, more than half of them were Stage 2, while almost all (>97%) of the CD4− and CD4+, IL-17RB− iNKT cells were Stage 3 (Figure 2E). Furthermore, more than 80% of Stage 1/2 iNKT cells were IL-17RB+ iNKT cells, while only ∼2% of the Stage 3 iNKT cells were IL-17RB+ (Figure 2D). The percentage (Figure 2F) and absolute number (Figure 2G) of IL-17RB+ iNKT cells among the total iNKT cells and in developmental Stages 1 and 2 were similar to those of Il15L117P mice, while those of IL-17RB− iNKT cells (i.e., CD122+ iNKT cells) among the total and in developmental Stage 3 were also comparable to those in Il17rb−/− mice, indicating that two distinct iNKT cell subsets are present in the different stages of iNKT cell development, i.e., the IL-17RB+ subtype in Stages 1 and 2 and the CD122+ subtype in Stage 3.
In order to determine if iNKT cell subtypes arise as a distinct population in the thymus of each other, each subtype in Stage 1 or Stage 2 was sorted and co-cultured with a fetal thymus (FT) lobe from Jα18−/− mice (Figure 2H and 2I). IL-17RB− subtype in Stage 1 gave rise to cells in Stage 2 and Stage 3 with IL-17RB− phenotype (Figure 2H and 2I, lower left), whereas IL-17RB+ subtype in Stage 1 gave rise to cells in Stage 2 but not to Stage 3 with IL-17RB+ phenotype (Figure 2H and 2I, upper left). Furthermore, IL-17RB− subtype in Stage 2 gave rise to cells in Stage 3 with IL-17RB− phenotype (Figure 2H and 2I, lower left), whereas IL-17RB+ subtype in Stage 2 kept in Stage 2 with IL-17RB+ phenotype (Figure 2H and 2I, upper left), indicating that IL-17RB+ iNKT cells arise in the thymus as distinct phenotypic subtypes from IL-17RB− iNKT cells, which undergo a series of developmental stages (i.e. Stages 1–3) previously characterized [14],[15].
To confirm the differences among subtypes of iNKT cells, we compared global gene expression profiles in WT B6 CD4− or CD4+, IL-17RB+ or IL-17RB−, iNKT cells to each other (Figure S2A), and also WT B6 CD4− or CD4+, IL-17RB+ iNKT cells to the same cell types from Il15L117P mice (Figure S2B). The genome-wide expression profile of the CD4− and CD4+, IL-17RB+ iNKT cells were similar to each other but different from those of CD4− and CD4+, IL-17RB− iNKT cells (Figure S2A). Moreover, the gene expression profiles of CD4− or CD4+, IL-17RB+ WT iNKT cells were similar to those in Il15L117P mice (Figure S2B). Therefore, it is likely that IL-17RB+ iNKT cell development in the thymus is distinct from the IL-17RB− (i.e. CD122+) iNKT cells. The gene expression profiles of the CD4+ IL-17RB+ iNKT cells were quite similar to those of the CD4− IL-17RB+ cells rather than the CD4− or CD4+, IL-17RB− iNKT cells (Figure S2A), suggesting that these two subtypes, CD4− and CD4+, IL-17RB+ iNKT cells, develop from the same precursors, whereas the precursors for IL-17RB− iNKT cells are distinct.
In order to investigate functional differences in the IL-17RB+ and IL-17RB− subsets of iNKT cells, we analyzed the ability of thymic iNKT cells in B6, Il17rb−/− and Il15L117P mice to produce cytokines in response to α-GalCer (Figure S3). IFN-γ was produced at similar levels by Il17rb−/− and WT iNKT cells, but was greatly reduced in the Il15L117P iNKT cells, while the production of IL-9, IL-10, IL-13, IL-17A, and IL-22 was impaired in the Il17rb−/− but not in the Il15L117P iNKT cells, similar to what we had observed in the spleen and liver (Figure S3).
In a previous study, IL-17RB+ iNKT cells were fairly abundant in the spleen of TH2-prone mice, but were barely detectable in TH1-prone mice [17]. Thus, we examined whether the frequency of IL-17RB+ iNKT cells in the thymus of BALB/c mice is different from that of B6 mice. Intriguingly, more than one-third of thymic iNKT cells were IL-17RB+ in TH2-prone BALB/c mice, four times higher than in TH1-prone B6 mice (Figure S4A). The genome-wide expression profiles of CD4− or CD4+, IL-17RB+ iNKT cells in BALB/c were similar to each other, but different from those of CD4− or CD4+, IL-17RB− iNKT cells (Figure S4B). Cluster analysis also showed that CD4− or CD4+, IL-17RB+ or IL-17RB− iNKT cells in B6 and BALB/c mice were essentially equivalent (Figure S4C).
iNKT cells in the thymus can be divided into four populations based on their expression of CD4 and IL-17RB (Figures 2D and S4A), and thymic Il17rb−/− iNKT cells had a decreased ability to produce TH2 and TH17 cytokines (Figure S3). Therefore, we analyzed the function of iNKT cell subtypes in the thymus of B6 (Figure 3) and BALB/c mice (Figure S5). We first used quantitative real-time PCR to investigate TH1/TH2/TH17-related gene expression patterns in FACS sorted thymic iNKT subtypes. The levels of Cd4 and Il17rb transcripts were correlated with the surface expression of these molecules (Figures 3A and S5A). Il2rb ( = Cd122) expression was restricted to CD4− and CD4+, IL-17RB− iNKT cell subtypes (Figures 3A and S5A) in correlation with their surface protein expression (Figure 2C). The expression levels of TH1-related transcripts, such as Ifng, Tbx21, and Stat4, were more than 10 times higher in those of CD4− and CD4+ IL-17RB− iNKT cells. Higher levels of TH2-related transcripts, such as Il4, were detected in CD4+ IL-17RB+ iNKT cells, even though Gata3, a transcription factor essential for TH2 cytokine production, was expressed at a similar level in all subtypes (Figures 3B and S5B). On the other hand, the expression of TH17-related transcripts, such as Il17a, Il22, and Rorc, were restricted to the CD4− IL-17RB+ iNKT cells (Figures 3B and S5B).
We then investigated the gene expression level in cells derived from Stages 1 and 2 by FT organ culture (Figure 2H and 2I). Consistent with the findings above, Ifng expression was restricted to the cells derived from CD4− and CD4+, IL-17RB− iNKT precursors (Figure S6A and S6B). Higher levels of Il4 were detected in CD4+ IL-17RB+ derived cells and restricted expression of Il17a in cells derived from CD4− IL-17RB+ precursors (Figure S6A and S6B), supporting each subtype arise from Stage 1 as a functionally distinct subtype.
Based on these findings, we analyzed potential production of cytokines from these thymic iNKT cell subtypes. Sorted iNKT cell subtypes were stimulated with PMA plus ionomycin (Figure S7A). Similar to the cytokine expression, IFN-γ was exclusively produced by the CD4− and CD4+, IL-17RB− subtypes, while IL-10 and IL-13 were mainly produced by the CD4+ IL-17RB+ iNKT cells and IL-17A was produced predominantly by CD4− IL-17RB+ iNKT cells. It should be noted that all four subtypes have a potential to produce IL-4, in correlation with their mRNA expression of Il4 and Gata3 (Figures 3B and S5B).
We then further analyzed cytokine production after α-GalCer activation. Sorted iNKT cell subtypes were co-cultured with BM-DCs in the presence of α-GalCer (Figures 3C and S5C). IFN-γ was exclusively produced by the CD4− and CD4+, IL-17RB− subtypes, while IL-4, IL-9, IL-10, and IL-13 were mainly produced by the CD4+ IL-17RB+ iNKT cells. Similarly, IL-17A and IL-22 were produced predominantly by CD4− IL-17RB+ iNKT cells. These cytokine production patterns correlated with their differential expression of TH1/TH2/TH17-related genes in the different iNKT subtypes.
We also analyzed the expression profiles of cytokine receptor genes. Il12rb2 transcript was expressed in CD4− and CD4+, IL-17RB− iNKT cells, and Il23r expression was restricted to CD4− IL-17RB+ iNKT cells (Figures 3D and S5D), suggesting that CD4− and CD4+, IL-17RB− iNKT cells respond to IL-12 through IL-12Rβ2/IL-12Rβ1, while CD4− IL-17RB+ iNKT cells respond to IL-23 through IL-23R/IL-12Rβ1. In fact, CD4− and CD4+, IL-17RB− iNKT cells produced large amounts of IFN-γ but not TH2 and TH17 cytokines in response to IL-12 (Figures 3E and S5E), while CD4− IL17RB+ iNKT cells produced large amounts of TH17 cytokines, IL-17A and IL-22, but not IFN-γ and TH2 cytokines in response to IL-23 (Figures 3F and S5F). IL-25-mediated activity requires not only IL-17RB but also IL-17RA expression [26], which is expressed on all iNKT cell subtypes (Figures 3D and S5D). IL-25 acts on thymic CD4+ IL-17RB+ iNKT cells to induce a large amount of TH2 cytokines, along with moderate amounts of TH17 cytokines (Figures 3G and S5G) similar to previous observations in the CD4+ IL-17RB+ iNKT cell subtype in the spleen [17]. Interestingly, however, IL-25 does not stimulate CD4− IL-17RB+ iNKT cells, despite their expression of IL-17RB (Figures 3G and S5G). We also found that cytokine production from iNKT cells in these experimental settings was hardly observed when BM-DCs derived from Cd1d1−/− mice (unpublished data), indicating signals from TCR are also required for cytokine production from iNKT cells. These results suggest that three types of iNKT cells, i.e. CD4− IL-17RB+ (iNKT-TH17, IL-23 reactive), CD4+ IL-17RB+ (iNKT-TH2/17, IL-25 reactive), and CD4− and CD4+, IL-17RB− (iNKT-TH1, IL-12 reactive), exist as distinct subpopulations in the thymus.
The chemokine receptor expression patterns are also distinct among thymic iNKT cell subtypes. Ccr4 and Ccr7 expression was restricted to both CD4− and CD4+, IL-17RB+ iNKT cells, and Ccr6 expression was only observed on CD4− IL-17RB+ iNKT cells (Figures 3H and S5H). Cxcr3 expression was several times higher on IL-17RB− iNKT cells than on the other subtypes. Surprisingly, the expression of Cxcr6, which has been reported to be abundantly expressed by all iNKT cells [27],[28], was also restricted to the IL-17RB− iNKT cells (Figures 3H and S5H). Note that the expression patterns and levels of all of the genes tested were almost equivalent between B6 and BALB/c mice, consistent with our finding that all iNKT subtypes are present in these strains.
Distinct expression of chemokine receptors among thymic iNKT cell subtypes (Figures 3H and S5H) may reflect the differential distribution of iNKT cell subtypes in the periphery. We thus investigated the frequency of total iNKT cells and subtypes in the spleen, liver, BM, lung, inguinal lymph node (LN), and mesenteric LN in WT B6, BALB/c, and Il17rb−/− mice (Figures 4 and S8). The absolute number and percentage of iNKT cells were slightly decreased in the spleen, lung, inguinal LN, and mesenteric LN of Il17rb−/− mice, but were unchanged compared to WT mice in liver and BMs (Figures 4A and S8A). We then gated on α-GalCer/CD1d dimer+ TCRβ+ iNKT cells and further analyzed them for the expression of CD44 and NK1.1 in B6 background mice (Figure 4B). The percentage of NK1.1− subtype cells was higher in the spleen, lung, inguinal LN, and mesenteric LN, but lower in the liver and BM, and was decreased in Il17rb−/− mice, suggesting that the majority of iNKT cell subtypes maintain surface expression of NK1.1− after emigration from the thymus (Figure 2B). Similarly, we examined the expression of CD4 and IL-17RB on the iNKT subtypes (Figures 4C and S8B). Interestingly, IL-17RB+ iNKT cells were abundant in the lung, inguinal LN, and mesenteric LN, but barely detectable in the liver and BM of both B6 and BALB/c mice. More than 40% of iNKT cells were IL-17RB+ in the lung, inguinal LN, and mesenteric LN, whereas more than 90% were IL-17RB− in the liver and BM (Figures 4C and S8B). Therefore, the distribution patterns of the iNKT cell subtypes are distinct in the tissues. In agreement with a previous study [21], we found that the number of iNKT cells was decreased in the spleen (∼1/3) and liver (∼1/30) in Il15L117P mice (Figure S9A). Reduction of iNKT cell number was also observed in BM (∼1/8) in these mice (Figure S9A), probably due to the selective reduction of the IL-17RB− subtypes (Figure S9B). We finally compared iNKT cell subtypes in the thymus and periphery of B6 and BALB/c mice (Figure 4D). The total iNKT cell number was almost equivalent between these two strains, but BALB/c had ∼4 times more CD4+ IL-17RB+ subtype cells, but lower (∼1/3) numbers of CD4− IL-17RB− cells, resulting in a higher number of CD4+ IL-17RB+ cells in the spleen (∼5 times), lung (∼2 times), inguinal LN (∼1.5 times), mesenteric LN (∼4 times), and lower numbers of CD4− IL-17RB− cells, especially in liver (∼1/6) and BM (∼2/5) of BALB/c mice (Figure 4D).
To confirm the distribution profiles of each subtype in the periphery, we performed intracellular cytokine staining after PMA plus ionomycin stimulation (Figure S7B) and quantitative real-time PCR analysis (Figure S10A–D) on these iNKT cells that were tested in the thymic iNKT cell subtypes (Figures 3A, 3B, 3D, 3H, S7A). The gene expression profiles and potential cytokine production in the iNKT cell subtypes were almost equivalent among those in the different peripheral tissues, but higher than those in the thymus, strongly suggesting that each iNKT subtype in the periphery is derived from the same iNKT subtypes in the thymus.
We next compared global gene expression profiles of CD4− or CD4+, IL-17RB+ or IL-17RB− iNKT subtypes in the thymus and spleen in order to test whether each subtype is functionally and phenotypically stable or plastic. Each of the four subtypes in spleen was highly correlated with the corresponding subtype in the thymus (Figure 5A), suggesting that iNKT subtypes can be divided by CD4 and IL-17RB expression both in the thymus and the periphery. Furthermore, iNKT cell subtypes in the periphery (Figure S10) showed similar quantitative gene expression profiles as in the thymus (Figure 3).
In order to confirm the stability and plasticity of iNKT cell subtypes, we sorted thymic iNKT cell subtypes based on the expression of CD4 and IL-17RB from WT B6 and transferred them into iNKT cell-deficient Jα18−/− mice. Ten days after transfer, we analyzed the IL-17RB expression by iNKT cell subtypes in the spleen. The results clearly showed that the majority of transferred cells maintained their surface IL-17RB expression (Figure 5B), suggesting that IL-17RB expression is stable as the cells migrate from the thymus to the periphery.
We further analyzed in cytokine production of splenic iNKT cell subtypes from B6 and BALB/c mice. The cytokine production profiles of splenic iNKT cell subtypes in response to α-GalCer (Figures 5C and S11A), IL-12 (Figures 5D and S11B), IL-23 (Figures 5E and S11C), and IL-25 (Figures 5F and S11D) were quite similar to those of the thymic iNKT cell subtypes (Figures 3C, 3E, 3F, 3G, S5C, S5E, S5F, S5G). Taken together, all of the iNKT subtypes detected in the thymus also exist as phenotypically and functionally distinct subtypes in the peripheral tissues.
Both thymic and peripheral iNKT cells in the steady state contain Ifng mRNA in the CD4− and CD4+, IL-17RB− cells (Tbx21 expressed, iNKT-TH1, IL-12 reactive), and Il17a and Il22 mRNA in the CD4− IL-17RB+ cells (Rorc expressed, iNKT-TH17, IL-23 reactive). The expression of these cytokine transcripts is thought to result from the fact that peripheral iNKT cells are not truly quiescent, but instead appear to be continuously activated at a low level due to their recognition of endogenous self-glycolipid ligand(s) in vivo. However, the CD4+ IL-17RB+ iNKT cells do not contain Il9, Il10, Il13 (unpublished data), Il17a, or Il22 mRNA (Figures 3B, S5B, S10B) in the steady state, even though these cytokines are immediately produced after activation by α-GalCer, similar to cases of IFN-γ from IL-17RB− iNKT cells. These results suggest differences in the transcriptional regulation of cytokine genes in the different iNKT cell subtypes.
One of the candidate genes is E4BP4, a mammalian basic leucine zipper transcription factor that regulates IL-10 and IL-13 production not only by CD4+ T cells and regulatory T cells but also by iNKT cells [29]. E4BP4 expression was markedly induced in IL-25-treated iNKT cells, and its expression level correlated with Il10 and Il13 expression [29]. Furthermore, iNKT cells lacking E4bp4 had reduced expression of IL-10 and IL-13 in response to either IL-25 or α-GalCer stimulation, but the IFN-γ and IL-4 production were unaffected [29], indicating that E4bp4 controls the TH2 cytokine production in a particular iNKT cell subtype.
Therefore, we analyzed the role of E4bp4 in iNKT cell subtypes. The expression of E4bp4 was selectively and strongly induced by IL-25 treatment in CD4+ IL-17RB+ iNKT cells both from thymus and spleen (Figure 6B). However, CD4− IL-17RB+ iNKT cells failed to induce E4bp4 expression even after treatment with IL-23 (Figure 6B), suggesting the cell type-specific function of E4bp4 and its possible role not only in Il10 and Il13 expression but also in Il9, Il17a, and Il22 expression by IL-25-treated CD4+ IL-17RB+ iNKT cells. To test this hypothesis, we analyzed cytokine production by CD4+ IL-17RB+ iNKT cells lacking E4bp4 after treatment with IL-25 in the presence of BM-DCs (Figure 6C). The production of IL-9, IL-10, IL-13, IL-17A, and IL-22 cytokines by both thymic or splenic CD4+ IL-17RB+ iNKT cells in response to IL-25 was completely abrogated, indicating E4BP4 turned out to be an intrinsic regulator of IL-25-mediated production, not only of IL-10 and IL-13 but also of IL-9, IL-17A, and IL-22.
We then investigated the role of IL-17RB+ iNKT cells in the pathogenesis of virus-induced AHR, which is known to be different from allergen-induced AHR [30]. Certain viruses, such as respiratory syncytial virus (RSV), Sendai virus, metapneumovirus, and parainfluenza virus, cause childhood asthma and COPD-like symptoms, which include AHR, airway inflammation, and mucus hypersecretion [31]–[33]. However, it has been very difficult to understand how such symptoms develop, even long after the apparent clearance of viruses. It has been reported that, in mouse models of infection with parainfluenza virus or Sendai virus, virus-induced chronic inflammation leads to asthma that resembles human asthma and COPD [34]. The chronic pulmonary symptoms evolved independently of CD4+ T cells but required CD4− iNKT cells and did not occur in Cd1d−/− and Jα18−/− mice [34].
Therefore, we attempted to determine whether or not the CD4− IL-17RB+ iNKT cells are responsible for chronic inflammatory lung disease induced by RSV infection. We used the secreted form of recombinant G protein of RSV (rec Gs) (Figure S12) as an immunogen because priming with a recombinant vaccinia virus (rVV) expressing rec Gs induced a more TH2-biased response and enhanced pulmonary eosinophil and macrophage infiltration following RSV challenge than did priming with rVV expressing either wild-type G or membrane anchored G (Gm) proteins [35],[36]. Mice were inoculated i.n. with RSV (106 pfu/100 µl) or PBS as a control four times at 10-d intervals and were intraperitoneally (i.p.) immunized with rec Gs/alum (50 µg/2 mg) 4 d after the first RSV infection. Three days after the last RSV administration, mice were exposed i.n. to 50 µg rec Gs and then, 24 h later, measured for AHR (Figure 7A). In this experimental setting, RSV/rec Gs-induced AHR was observed in WT BALB/c but not in Jα18−/− or Il17rb−/− mice, which had a similar response level as PBS/rec Gs-induced WT controls, indicating that IL-17RB+ iNKT cells contribute to the development of RSV plus viral antigen-induced AHR (Figure 7B). Airway macrophage and lymphocyte numbers, which were relatively higher than eosinophils and neutrophils, were recruited into the bronchoalveolar lavage (BAL) fluid of RSV/rec Gs-induced WT mice but not the other mice (Figure 7C). These results suggest that IL-17RB+ iNKT cells are required for the development of RSV-induced AHR. Low level of cytokines (IL-4, IL-9, IL-10, IL-13, IL-17A, and IL-22) in the BAL fluid was detected in this experiment (Figure 7D). The production of IL-13 and IL-22, which plays a crucial role in the activation of macrophages and neutrophils, respectively, was detected higher in RSV/rec Gs-induced WT mice. Hematoxylin and eosin (H&E) staining of the lung tissue revealed that a large number of inflammatory mononuclear cells had infiltrated into the peribronchiolar region, a response that was higher in RSV/rec Gs-induced WT mice compared to RSV/rec Gs-induced Jα18−/− or Il17rb−/−, mice (Figure 7E, upper panel). By periodic acid-Schiff (PAS) staining, mucus-producing cells were abundant only in RSV/rec Gs-induced BALB/c mice but not in Jα18−/− or Il17rb−/− mice (Figure 7E, lower panel). To confirm the findings that IL-17RB+ iNKT cells are essential for the development of RSV/rec Gs-induced AHR, we transferred enriched splenic IL-17RB+ iNKT cells into Jα18−/− mice and tested their ability to develop AHR (Figure 7F). The cell transfer of IL-17RB+ iNKT cells, but not IL-17RB− iNKT cells nor PBS alone, restored AHR induced by RSV plus rec Gs, dependent of cell number transferred, demonstrating the important contribution of IL-17RB+ iNKT cells in the pathogenesis of development in virus plus viral antigen-induced AHR.
In the present study, we identified IL-17RB− and IL-17RB+ subtypes of iNKT cells both in the thymus and the periphery. The IL-17RB− iNKT cells express CD122 (IL-15Rβ chain), expand in an IL-15-dependent manner, and produce IFN-γ in response to IL-12. On the other hand, the IL-17RB+ iNKT cells do not express CD122 or respond to IL-15. The IL-17RB+ iNKT cells can be further divided into at least two subtypes: (1) CD4+ IL-17RB+ iNKT cells produce TH2, TH9, and TH17 cytokines in an E4BP4-dependent fashion in response to IL-25, and (2) CD4− IL-17RB+ iNKT cells are RORγt+ and produce TH17 cytokines in response to IL-23, but independently of E4BP4. In the thymus, the IL-17RB+ iNKT cells have a developmental pathway distinct from the IL-17RB− iNKT cells.
It has been proposed that iNKT cell differentiation stages can be categorized based on the expression patterns of CD44 and NK1.1, for example CD44lo NK1.1− for Stage 1, CD44hi NK1.1− for Stage 2, and CD44hi NK1.1+ for Stage 3 [14],[25]. However, the majority (>80%) of IL-17RB+ iNKT cells was present in both the Stage 1 and Stage 2 subsets, while IL-17RB− iNKT cells were enriched in Il17rb−/− mice and were mainly detected in Stage 3, suggesting that a certain but not all of the Stage 1 and Stage 2 IL-17RB+ iNKT cells are not precursors for the Stage 3 cells. It is believed that iNKT cells acquire their ability to produce IL-4 and IL-10, but make little IFN-γ in Stages 1/2 populations, whereas iNKT cells in Stage 3 produce abundant IFN-γ but less if any IL-10 [14],[15],[25],[37]. These finding are in agreement with the present results that the Stage 1/2 populations mainly contain IL-17RB+ iNKT cells that can produce IL-4 and IL-10, but not IFN-γ, whereas the majority of the Stage 3 iNKT cells are IL-17RB− iNKT cells producing IFN-γ but not TH2 cytokines. The results shown here also indicated that all of the four iNKT subtypes already existed in Stage 1 and developed into phenotypically and functionally distinct iNKT cells as CD4− or CD4+, IL-17RB+ in Stage 2 and CD4− or CD4+, IL-17RB− through Stage 2 to Stage 3.
It has reported that IL-15 plays an important role in the expansion of iNKT cells [21]. Our present data showed that IL-15 requires only for the expansion of IL-17RB− iNKT cell subtypes but not for IL-17RB+ iNKT cells, even though it has still been unclear that the cytokine(s) are required for the development and expansion of IL-17RB+ subtypes. In fact, IL-17RB− iNKT cell subtypes were greatly reduced in number among iNKT cell subtypes but already had an ability to produce IFN-γ in Il15L117P mice, resulting in the reduced IFN-γ production after iNKT cell activation due to the reduced number of these subtypes.
In the previous reports, the IL-17A-producing subtypes were proposed to be contained within the CD44hi NK1.1− CD4− RORγt+ subpopulation [8],[19]. In the present studies, we found that the CD4− IL-17RB+ iNKT cell subtype is CD44hi NK1.1− CD4− (about 50%–70% of the cells are IL-17RB+) and has a restricted expression of Il17a, Rorc, Ccr6, and Il23r genes, for a phenotype similar to the previously reported CD44+ NK1.1− CD4− RORγt+ population that produces IL-17A [19],[38]. These results indicate that IL-17RB (and CD4−) is a reliable and specific phenotypic marker for RORγt+ IL-17A-producing iNKT cells in the thymus.
In the periphery, the tissue distribution of the iNKT cell subtypes seems to largely depend on the expression of chemokine receptors: CCR6+ CCR4+ CCR7+ expression by CD4− IL-17RB+ iNKT cells, CCR4+ CCR7+ expression by CD4+ IL-17RB+ iNKT cells, and CXCR3+ CXCR6+ by CD4− and CD4+, IL-17RB− iNKT cells. Indeed, the number of liver iNKT cells, the majority of which are the CD4− and CD4+, IL-17RB− iNKT cells identified here, depends on the chemokine receptor CXCR6, whereas iNKT cells in other tissues are less dependent as reported [27],[28]. In Ccr4−/− mice, the lung has fewer iNKT cells and a corresponding reduction in iNKT cell-mediated AHR [39], implicating the reduction of pulmonary localization of IL-17RB+ iNKT cells.
IL-17A-producing iNKT cells have been described in other studies in the thymus, liver, spleen, lung, LNs, and skin [8],[19],[20],[38],[40]. In these studies, it was suggested that all NK1.1− iNKT cells have the potential to secrete IL-17A. However, in the present study, we show heterogeneity among NK1.1− iNKT cells. Accordingly, CD4− but not CD4+, IL-17RB+ iNKT cells correspond to the IL-17A-producing iNKT cells previously reported, as does the exclusive expression of Ccr6 along with Itgae ( = Cd103) and Il1r1 ( = Cd121a) in CD4− IL-17RB+ iNKT cells (unpublished data) [40].
CD4+ IL-17RB+ iNKT cells produce not only the previously described IL-13 and IL-4 [17],[18] but also IL-9 and IL-10 along with IL-17A and IL-22 in response to IL-25 in an E4BP4-dependent fashion. Even though it is still unclear whether IL-25-reactive CD4+ IL-17RB+ iNKT cells can be further divided into differentially functional subsets (i.e. iNKT-TH2, iNKT-TH9, iNKT-TH17), it is noteworthy that a recently described subset of differentiated T cells [41], termed TH9, which can be induced by IL-4 plus TGF-β, produces IL-9 and IL-10 in response to IL-25. This IL-9 production is IL-4 independent, highlighting the role of IL-25 in the regulation of both TH2 and TH9 cells [42]. We demonstrated here that IL-25 induces not only IL-13 and IL-4 but also IL-9 and IL-10 from CD4+ IL-17RB+ iNKT cells, which can thus be characterized as iNKT-TH2 and iNKT-TH9 cells. Concerning the cytokine production by CD4+ IL-17RB+ iNKT cells in response to IL-25, not only IL-10 and IL-13 but also IL-9, IL-17A, and IL-22 were attenuated in the absence of E4bp4, recently defined as a transcription factor that regulates IL-10 and IL-13 production by CD4+ T cells and iNKT cells [29], suggesting that E4BP4 also controls IL-25-mediated production of IL-9, IL-17A, and IL-22. Although the precise mechanisms by which IL-25 mediates cytokine expression still remains unclear, E4BP4 itself directly or indirectly controls IL-9, IL-10, IL-13, IL-17A, and IL-22 expression by genetic/epigenetic regulation in CD4+ IL-17RB+ iNKT subtypes. It will be of interest to determine if E4BP4 regulates IL-9, IL-17A, and IL-22 production by CD4+ TH cells. Taken collectively, our studies indicate that CD4− or CD4+, IL-17RB+ iNKT cells become functionally stable iNKT-TH17 or iNKT-TH2/9/17, respectively, during their development.
The study described here indicates that iNKT cell-mediated AHR was not induced by viral infections in Jα18−/− or Il17rb−/− mice, suggesting that IL-17RB+ iNKT cells are responsible for the pathogenesis of many different forms of airway inflammation. Although distinct subsets of iNKT cells have been reported to be involved in different forms of asthma [17],[18],[34],[43], they are now consolidated into CD4− and/or CD4+ IL-17RB+ iNKT cell subsets.
iNKT cells are also known to mediate regulatory functions controlling various pathological conditions, such as infectious diseases caused by microbes [44], autoimmune diseases (colitis, lupus, diabetes) [45],[46], atherosclerosis [47], and malignancy [48]. It will be interesting to elucidate whether subsets of iNKT cells play differential roles in mediating and controlling these diverse pathological conditions.
B6 and BALB/c mice were purchased from Charles River Laboratories or Clea Japan, Inc. Il17rb-deficient mice were generated as shown in Figure S1 and were backcrossed >8 times to B6 or BALB/c mice. Il15L117P mutant mice were produced by N-Ethyl-N-nitrosourea (ENU) mutagenesis by ENU administration to male C57BL/6J mice, and their sperm was mated to wild-type eggs and preserved as founder embryos [49],[50]. Jα18-deficient mice were generated as previously described [51] and were backcrossed >10 times to B6 or BALB/c mice. Cd1d1-deficient mice [52] were provided by Dr. Luc van Kaer (Nashville, TN). E4bp4-deficient mice were generated as previously described and were backcrossed 8 times to B6 mice [29]. All mice were kept under specific pathogen-free conditions and were used at 8–16 wk of age. All experiments were in accordance with protocols approved by the RIKEN Animal Care and Use Committee.
Cytokines except IL-22 in culture supernatants and BAL fluids were analyzed by cytometric bead array (BD Biosciences) according to the manufacturer's protocol. IL-22 was quantified by an ELISA reagent set (eBioscience) according to the manufacturer's protocol.
Cells were analyzed by FACS Calibur (BD Biosciences) or FACS Canto II (BD Biosciences) and sorted by FACS Aria (BD Biosciences). Antibodies (BD Biosciences or eBioscience) used for staining mouse cells were as follows: FITC or APC-Cy7 anti-TCRβ (H57-597), Pacific blue anti-CD4 (RM4-5), FITC anti-CD44 (IM7), PE-Cy7 anti-NK1.1 (PK136), PE anti-CD122 (TM-β1), FITC anti-CD8α (53-6.7), PerCP-Cy5.5 anti-CD25 (PC61), PE anti-IFN-γ (XMG1.2), PE anti-IL-4 (11B11), PE anti-IL-10 (JES5-16E3), PE anti-IL-13 (eBio13A), PE anti-IL-17A (TC11-18H10), and PE rat IgG1 (A110-1). Biotinylated anti-mouse IL-17RB (B5F6) was generated previously [17] and detected by staining with PE or PE-Cy7 Avidin (BD Biosciences). APC α-GalCer loaded CD1d dimer (BD Biosciences) for iNKT cell enrichment and detection was prepared as previously described [53].
The procedures for the coculture with a deoxyguanosine (dGuo)-treated FT lobe under high oxygen submersion conditions have been described in detail previously [54],[55]. Basically, single dGuo-treated FT lobes from Jα18−/− of B6 background were placed into wells of a 96-well V-bottom plate, to which cells from B6 mice to be examined were added. Culture medium was supplemented with IL-7 (1 ng/ml), IL-15 (10 ng/ml), and soluble IL-15Rα (10 ng/ml). The plates were centrifuged at 150× g for 5 min at room temperature, placed into a plastic bag (Ohmi Odor Air Service), the air inside was replaced by a gas mixture (70% O2, 25% N2, and 5% CO2), and incubated at 37°C. After 10 d of culture, cells were harvested from each well and analyzed by FACS and quantitative real-time PCR.
Intracellular cytokine staining was performed as described previously [53]. For cytokine production from sorted iNKT cell subtypes, Brefeldin A (Sigma-Aldrich) was added for the last 4 to 5 h of culture to accumulate intracellular cytokines after PMA (25 ng/ml, Sigma) with ionomycin (1 µg/ml, Sigma) treatment. Following fixation with Cytofix/Cytoperm plus (BD Biosciences), cells were stained for indicated intracellular cytokines for 15 min at room temperature.
PCR primers and probes were designed with Universal ProbeLibrary Assay (Roche) or with TaqMan Gene Expression Assays (Applied Biosystems). Sequence of primers and probes in the latter case are shown in Table S1. PCR was performed with the TaqMan universal master mix with ROX (Applied Biosystems) according to the protocol provided. ABI PRISM7900HT Fast system (Applied Biosystems) or Biomark system (Fludigm) was used for quantitative real-time PCR according to the manufacturer's instructions. To ensure the specificity of the amplification products, a melting curve analysis was performed. Results were normalized and analyzed by ΔCt or ΔΔCt methods using the internal control gene Hprt1.
Gene expression detected using microarrays was normalized by the quantile normalization method [56]. Pearson's correlation values of logarithms of all signal intensities from 45,101 probes were calculated, and we performed hierarchal clustering of correlation matrices to indicate the degree of similarity between cell types. Scatter diagrams were drawn to display how similarly or differently genes were expressed in two samples. These diagrams contain only probes whose signals were present and coefficient values were shown in the figures.
Strain A2 of human RSV was used in this study. The general protocol for analyzing airway remodeling during RSV infection in mice is as follows: AHR was induced by sensitizing and challenging with OVA/alum (3–4 times) and/or infection with RSV (3–4 times), and then challenging with OVA, resulting in the examination of various pathological endpoints as previously described [57]–[59]. In the present study, we modified these protocols in order to analyze the physiological role of iNKT cells in the development of AHR mediated by RSV. In brief, mice were i.n. administered with RSV (106 pfu) or PBS as a control 4 times at 10-d intervals. Mice were i.p. immunized with rec Gs/alum (50 µg/2 mg) 4 d after first RSV infections. Three days after the last RSV administration, mice were exposed i.n. to rec Gs recombinant protein and AHR responses were measured 1 d later.
Airway function was measured for changes in lung resistance (RL) and dynamic compliance in response to increasing doses of inhaled methacholine (1.25, 2.5, 5, 10, and 20 mg/ml) by using an invasive FlexiVent (SCIREQ Scientific Respiratory Equipment Inc.).
After measurement of AHR and sacrifice, the mouse trachea was cannulated, the lungs were lavaged twice with 1 ml PBS (10-fold PBS dilution), and the BAL fluid was pooled as previously described [30]. Lymphocytes from thymus, spleen, liver, lung, BM, inguinal LN, and mesenteric LN were isolated as described previously [53].
The statistical significance of differences was determined by t test, analysis of variance (ANOVA), or the Kruskal-Wallis test. The values were expressed as means ± SEM from independent experiments. Any differences with a p value of <0.05 were considered significant (* p<0.05; ** p<0.01).
|
10.1371/journal.pcbi.1000006 | Multi-Scale Simulations Provide Supporting Evidence for the Hypothesis of Intramolecular Protein Translocation in GroEL/GroES Complexes | The biological function of chaperone complexes is to assist the folding of non-native proteins. The widely studied GroEL chaperonin is a double-barreled complex that can trap non-native proteins in one of its two barrels. The ATP-driven binding of a GroES cap then results in a major structural change of the chamber where the substrate is trapped and initiates a refolding attempt. The two barrels operate anti-synchronously. The central region between the two barrels contains a high concentration of disordered protein chains, the role of which was thus far unclear. In this work we report a combination of atomistic and coarse-grained simulations that probe the structure and dynamics of the equatorial region of the GroEL/GroES chaperonin complex. Surprisingly, our simulations show that the equatorial region provides a translocation channel that will block the passage of folded proteins but allows the passage of secondary units with the diameter of an alpha-helix. We compute the free-energy barrier that has to be overcome during translocation and find that it can easily be crossed under the influence of thermal fluctuations. Hence, strongly non-native proteins can be squeezed like toothpaste from one barrel to the next where they will refold. Proteins that are already fairly close to the native state will not translocate but can refold in the chamber where they were trapped. Several experimental results are compatible with this scenario, and in the case of the experiments of Martin and Hartl, intra chaperonin translocation could explain why under physiological crowding conditions the chaperonin does not release the substrate protein.
| Chaperonin complexes capture proteins that have not yet reached their functional (“native”) state. Non-native proteins cannot perform their function correctly and threaten the survival of the cell. The chaperonins help these proteins to reach their native state. The prokaryotic GroEL-GroES chaperonin is an ellipsoidal protein complex that is approximately 16 nm long. It consists of two chambers that are joined at the bottom. Interestingly, protein repair by this chaperonin is not a one-step process. Typically, several capture and release steps are needed before the target protein reaches its native state. It is commonly assumed that substrate proteins cannot translocate, i.e., move inside the complex from one chamber to the other. In the absence of translocation, proteins that have not yet reached their functional conformation have to be released into the cytosol before being recaptured by a chaperonin. We present multi-scale simulations that show that it is, in fact, surprisingly easy for substrate proteins to translocate between the two chambers via an axial pore that is filled with disordered protein filaments. This finding suggests that non-native proteins can be squeezed like toothpaste from one chamber to the other: the incorrect structure of the protein is broken up during translocation and the protein has an increased probability to find its native state when it reaches the other chamber. The possibility for intra-chaperonin translocation obviates the need for a potentially dangerous release of non-native proteins.
| Proteins that have not yet folded to their native state may interfere with the machinery of the cell. For this reason, prokaryotic and eukaryotic cells have evolved special macro-molecular “chaperone” complexes that capture and refold partially folded proteins, thereby preventing them from indulging in cellular mischief [1],[2],[3,]. An important class of chaperone complexes are the cage chaperones or chaperonins. These complexes can efficiently trap partially folded proteins in a cavity that is barely larger than the target protein, and assist in the folding of an entire class of proteins with different amino acid sequences. Hence, the chaperonin is able to distinguish partly folded states from the native state, independently of the specific amino-acid sequence. It is important to stress that in the presence of molecular crowding (similar to the one present in a cell) the chaperonin complex has been demonstrated to not release the substrate protein before it reaches the native state [4]. Below, we report a detailed numerical study of protein dynamics inside the so-called GroEL-GroES chaperone complex. The GroEL complex consists of two barrel-shaped protein complexes joined at the bottom (see Figure 1). Non-native proteins can be captured in an open GroEL “barrel”. The GroES “lid” can then cap a protein-containing barrel, thereby initiating the refolding process. After about 15 seconds and several refolding cycles, the GroES cap is released and the other barrel is capped (if it contains a protein). A single “cycle” of the GroEL-GroES chaperone hydrolyses seven ATPs [5]. This energy is presumably used to compress the protein in a smaller, more hydrophilic GroEL cavity, thus increasing the thermodynamic driving force to expel this protein. Recently we reported simulations of the kinetics of chaperone-induced protein refolding, using a lattice model for the GroEL-GroES complex [6]. This study suggested that proteins may refold either inside the cavity in which it has been captured or, surprisingly, by translocating from one barrel of the GroEL dimer to the other (see Figure 2). This second route is unexpected because it is generally believed that proteins cannot cross the equatorial plane that separates the joined GroEL barrels [7],[8],[9].
In the present paper we use atomistic and mesoscopic simulations to test whether such a translocation scenario is compatible with the available structural information on the GroEL complex. Our simulation studies focus on the equatorial regime of the GroEL complex that might be expected to act as a barrier against translocation. Crystallographic studies indicate that most protein units in the chaperonin complex have a fairly rigid structure both in the open and closed configurations [5]. However, low-resolution small-angle neutron scattering experiments [7] and cryo-electron microscopy [8],[9] indicate the presence of disordered residues in a central cavity of the equatorial region. These chains do not show up in the X-ray crystallographic structure of the GroEL complex.
The presence of disordered protein chains in the pore that joins the two GroEL chambers will certainly affect the permeability of the equatorial plane, but they need not block translocation. There are, in fact, examples [10] where disordered protein chains near a pore act to enhance the selectivity of the translocation process. Interestingly, the chemical composition of the disordered chains in the GroEL complex is similar to that of chains in known translocation channels in the nuclear pore complex.
We have performed fully atomistic and coarse-grained simulations that do reproduce the structural data of [7], and allowed us to bridge the computational cost of computing the translocation free energy barrier of a short alpha helix. For the fully atomistic simulations in explicit water we used the GROMACS Molecular Dynamics (MD) simulation package [11]. MD simulations of 10 ns were performed on the structure of the central region at which time the system had equilibrated (Figure S1). In order to compute the scattering profile we used the program CRYSON from Svergun et al. [12]. Figure 3 shows that the neutron-scattering form factor computed on the basis of equilibrated structure of the trans ring agrees well with the experimental data of Krueger et al. [13]. Interestingly, the simulations show that chains on the cis ring do not obstruct the passage between the two GroEL chambers (see Figure 4 and Figure S2). The chains in the trans ring fluctuate in a region between 5 and 15 Å from the center, in agreement with hollow-cylinder model proposed by Krueger et al. on the basis of their experimental data [13].
To compute the free-energy barrier for protein translocation, the MD approach described above would have been prohibitively expensive. We therefore performed Monte Carlo simulations on a suitably coarse-grained model for the GroEL complex. We focused on the structural fluctuations within a spherical region (diameter 40 Å) around the trans side of the equatorial cavity (Figure 1), because the cis chains did not appear to represent an obstacle to translocation. The disordered chains in the cavity (22 monomeric units long) were rigidly anchored on a circular rim around the trans hole of ∼30 Å radius (Figure 1 inset). To this end, we represented all peptide backbones using a model that keeps track of the positions of 5 distinct types of backbone atoms (H, N, Cα, C, and O). Side chains are represented as hard spheres with a radius of 2.5 Å, centred on the Cα atoms. Neighbouring spheres along the chain are allowed to overlap (see Figure S3). We used this coarse-grained model to estimate the free-energy cost associated with the insertion of a short and rigid helix, 21 monomeric units long, in the region of disordered protein chains. We sampled the free energy as a function of a reaction coordinate Qs that measures the progress of the translocation process. Qs is defined as the total number of Cα atoms that have passed the entrance of the trans ring. We define the entrance as a plane through the average position of the hydrogen atoms in the anchoring amino acid of the chains. In order to translocate, a protein must first “find” the translocation hole. From our study of a lattice model GroEL [6], we know that this first step is relatively easy. The key question is therefore whether or not the free-energy cost for the subsequent translocation is prohibitive. The present calculations address this issue by computing the free energy difference involved in moving an α-helix from the entrance of the pore region to the inside. Of course, the free-energy barrier depends on the interaction between the α-helix and the disordered chains that consist mainly of Gly and Met.
We start by considering a very naive estimate that has the advantage that it is based on the fully atomistic simulations. From these simulations, we know the density profile of Cα atoms in the trans ring (see Figure 4). If, in the spirit of the Flory model, we assume that the density fluctuations of independent polymer Kuhn segments are Poisson distributed, we can estimate the probability P0 that a tube with the diameter of an α-helix contains no Cα atoms at all. This would lead us to an estimate of the free energy barrier that is equal to −kTlnP0. Using the density profile of Figure 4 and an estimate [14] for the persistence length of a protein filament, we obtain a translocation barrier of approximately 4 kBT. If we make the (unrealistic) assumption that all Cα's in a single chain are fully correlated, then we estimate the barrier height to be only 1 kBT, which should be a significant underestimate. To see whether such a rough estimate is at all reasonable, we can repeat the same procedure for the coarse-grained model where we can also perform direct free-energy calculations. To be consistent with the previous case, we assume that the there are only excluded-volume interactions between the (mainly Gly) chains and the helix residues. In terms of the interaction matrix of [15] this is equivalent to assuming that the helix consist entirely of Thr residues. Assuming all Kuhn segments fluctuate independently, we estimate the barrier to be 4 kBT, and the assumption of fully correlated fluctuations will again yield an estimate of order 1 kBT. The good agreement between the fully atomistic and coarse grained estimates is, of course, somewhat fortuitous, in view of the fact that the two density distributions are not identical. However, it suggests that the coarse-grained model may be of practical use.
Next, we compute the free energy barrier for the coarse-grained model system using the MC method described in the Methods.
First we considered the case of pure steric interactions between both the chains and the helix. In Figure 5 we plot the free energy F(QS) as a function of the reaction coordinate QS that measures the number of Cα's that have entered the pore region. The plot shows a symmetric barrier with a height of approximately 2 kBT, which is surprisingly close to the estimate obtained assuming fully correlated fluctuations of protein segments. In other words: the chains tend to move as a whole in an out of the central area of the pore. This picture is supported by the snapshot of the pore region (Figure S4). The main conclusion that we can draw from the coarse-grained free-energy calculations is that the presence of seven protein chains in the central core region of the trans ring is not enough to obstruct translocation on steric grounds alone.
Of course, the interaction between a typical translocation protein segment and the ring chains is not purely steric. To consider the effect of both attractive and repulsive interactions, we consider the two cases separately. As the chains consist predominantly of Gly, we consider the scenarios that the interactions between the filament residues and the Cα atoms of the helix are all equal to the twice the average of all attractive (resp. repulsive) interaction energies of Gly in the Betancourt-Thirumalai interaction matrix [15] (−0.1kBT and +0.1kBT, respectively). The strength of attractive/repulsive interactions between the Cα's of the helix and the filament is therefore −0.2kBT (resp +0.2kBT). By taking an interaction that is double the average attractive/repulsive interaction strength, we are presumably modeling rather extreme cases that should put bounds on the actual translocation barrier.
Figure 5 shows the computed free-energy barriers for translocation in the case of attractive (resp. repulsive) interactions. The translocation barrier is appreciably lower when the chains attract the α helix (2 kBT) than in the opposite limit (4.5 kBT). However, the most striking observation is that the barrier is quite small in either case - a barrier of 4.5 kBT can easily be crossed due to the action of thermal fluctuations.
In fact, in the case of attractive interactions, there is virtually no barrier for translocation. This absence of a barrier may provide a rationale for the experimental observation that Krueger et al. observed in their SANS experiments [13] that a non-native protein (DPJ-9) was partially sucked into isolated trans rings. If proteins can indeed translocate through the GroEL equatorial plane then this may also be relevant for the mechanism by which the GroEL/GroES chaperonin can help the refolding of proteins that are too big to be encapsulated. In such cases, portions of the protein could be attracted to the inside of the pore and perform either a complete or a partial translocation (Figure S5). According to [6] either process can enhance the refolding efficiency.
The translocation of encapsulated non-native proteins is most likely in cases where the initial structure is far native. The reason is two fold: first of all, for such conformation there should be a low free-energy cost associated with partial unfolding—a necessary first step in translocation. Secondly, non-native chains that are trapped in a hydrophilic cage tend to be compressed. They can lower their free energy by translocating out of the cage. The simulations of [6] suggest that the driving force for such translocation can be as much as 0.5 kBT per amino-acid residue. Such a free-energy gradient is enough to completely remove a small free-energy barrier that might oppose translocation (Figure S6).
In conclusion, our simulation results are not compatible with the assumption that the disordered protein chains in the cis or trans rings provide an effective barrier against translocation. The present findings may help explain a puzzling experimental finding concerning refolding experiments in the presence of crowding agents [4]. The experiments of [4] demonstrated that, under physiological crowding conditions, the substrate protein does not escape from the chaperonin until it has reached its native state. This phenomenon is difficult to reconcile with the standard scenario where a protein (folded or not) is expelled from the cis-chamber as another non-native protein binds to the ATP-trans chamber. However, if it is not another protein that binds to the hydrophobic rim of the trans chamber, but the original protein that has translocated from the cis-chamber (see Figure 2), then it becomes clear why non-native proteins are unlikely to escape. We stress that the present findings do not rule out the possibility that non-native proteins fold into the native state without translocation [16]—translocation is simply an added route for protein folding. Such a route maybe very important for proteins that folds co-translationally, where confinement in a optimal size tunnel is crucial for efficiently reaching the native state [17].Our simulations suggest that it would be interesting to carry out refolding experiments on GroEL with mutated chains that would strongly stick to each other (or that could be cross-linked). Such mutation would impede the translocation and should thereby reduce the efficiency of the GroEL/GroES complex.
The flexible nature of this region prevented accurate X-ray determination of the chains filling the interconnecting pore. To obtain a full-atomistic model, the program MODELLER [18] has been used to generate a starting configuration of the chains missing in the X-ray structure (PDB code: 1AON) of the GroEL/GroES complex loaded with ADP. The reconstructed fragments (sequence KNDAADLGAAGGMGGMGGMGGM) are added at the C-term extremity of each monomeric building block of the chambers. In order to avoid steric clashes between the chains, the procedure has taken into account of the quaternary assembly of the chains. After the generation of the chains structures, three steepest-descent minimisations were performed, using the program GROMACS [11] (energy minimisation tolerance: 0.1, 0.05 and 0.01 kJ/mol−1nm−1). Molecular Dynamics (MD) simulations were subsequently performed with the GROMACS [11] package by using GROMOS96 force field with an integration time step of 2 fs. Non-bonded interactions were accounted for by using the particle-mesh Ewald method (grid spacing 0.12 nm) [19] for the electrostatic contribution and cut-off distances of 1.4 nm for Van der Waals terms. Bonds were constrained by LINCS [20] algorithm. The system was simulated in the NPT ensemble by keeping constant the temperature (300 K) and pressure (1 atm); a weak coupling [21] to external heat and pressure baths was applied with relaxation times of 0.1 ps and 0.5 ps, respectively. As we intended to simulate a solution at a pH-value of 7 the protonation states of pH sensitive residues were assigned as follow: Arg and Lys were positively charged, Asp and Glu were negatively charged and His was neutral. The protein's net charge was neutralised by the addition of Cl− and Na+ ions. It would have been prohibitively expensive to simulate the entire chaperonin plus surrounding water. However, this was not necessary, as our aim was to study the structure and dynamics of the strongly fluctuating the equatorial rings, rather than the relatively rigid remainder of the GroEL “chamber”. We therefore immobilised the chamber atoms that are not directly connected to the pore chains. Of course, the equatorial chains were free to move and relax in the pore. In order to further reduce the number of degrees of freedom treated, we only considered water molecules (SPCE [22]) inside the GroEL chamber. We achieved this by imposing a strong repulsive external potential outside the GroEL chamber. Ignoring the water outside the cage is not an unreasonable simplification, as we found that the disordered chains were completely solvated by water molecules and never moved outside the atoms of the internal surface of the chamber. We assumed periodic boundary conditions only along the symmetry axis of the GroEL complex (“z-axis”).
The Caterpillar model is a modification of the tube model of Maritan and co-workers [14],[23],[24]. The main differences are that we treat the structure of the backbone in more detail and that our scheme to account for self avoidance by means of bulky side groups is computationally cheaper than the approach of Maritan et al. who introduced a three-body interaction to achieve the same. The interaction between amino acids with different side chain ECA is given by the following expression(1)where is the distance between nonadjacent Cα atoms in the protein and rmax is the distance at which the potential has reaches half ε . For ε we use the 20×20 matrix derived with the method of Betancourt and Thirumalai [15].Although these interaction energies are strictly speaking neither energies nor free energies, they do provide a reasonable representation of the heterogeneity in the interactions between different amino acids. We modeled the hydrogen bonds between the hydrogen and the oxygen of the backbone with a 10-12 Lennard-Jones potential:(2)where the minimum is at σ = 2.0 Å and ELJ = 3.1 kBT. The directionality of the hydrogen bond was taken into account by multiplying the Lennard-Jones potential by a pre-factor(3)where θ1 and θ2 are the angles between the atoms COH and OHN respectively. The large hard spheres centered on the Cα atoms ensure that the orientation factor is maximum only for angles close to π. Apart from rotations around the dihedral angles φ1 and φ2 (Figure S3), the backbone is rigid. We have verified that this model can indeed reproduce typical protein motifs such as alpha helices and beta sheets, depending on the amino-acid sequence.
To sample the conformations of the protein chains anchored on the trans ring, we use two basic Monte-Carlo moves: branch rotation and an improved version of the biased Gaussian step [25], while for the translocating alpha helix we allow only translation moves and rotation around the center of mass. |
10.1371/journal.pbio.1000518 | Non-Canonical NF-κB Activation and Abnormal B Cell Accumulation in Mice Expressing Ubiquitin Protein Ligase-Inactive c-IAP2 | Chromosomal translocations between loci encoding MALT1 and c-IAP2 are common in MALT lymphomas. The resulting fusion proteins lack the c-IAP2 RING domain, the region responsible for its ubiquitin protein ligase (E3) activity. Ectopic expression of the fusion protein activates the canonical NF-κB signaling cascade, but how it does so is controversial and how it promotes MALT lymphoma is unknown. Considering recent reports implicating c-IAP1 and c-IAP2 E3 activity in repression of non-canonical NF-κB signaling, we asked if the c-IAP2/MALT fusion protein can initiate non-canonical NF-κB activation. Here we show that in addition to canonical activation, the fusion protein stabilizes NIK and activates non-canonical NF-κB. Canonical but not non-canonical activation depended on MALT1 paracaspase activity, and expression of E3-inactive c-IAP2 activated non-canonical NF-κB. Mice in which endogenous c-IAP2 was replaced with an E3-inactive mutant accumulated abnormal B cells with elevated non-canonical NF-κB and had increased numbers of B cells with a marginal zone phenotype, gut-associated lymphoid hyperplasia, and other features of MALT lymphoma. Thus, the c-IAP2/MALT1 fusion protein activates NF-κB by two distinct mechanisms, and loss of c-IAP2 E3 activity in vivo is sufficient to induce abnormalities common to MALT lymphoma.
| MALT (mucosal associated lymphoid tissue) lymphomas commonly express a mutant protein that contains a portion of the ubiquitin protein ligase cellular Inhibitor of Apoptosis 2 (c-IAP2) and a portion of the paracaspase MALT1. Expression of this fusion protein activates the anti-apoptotic transcription factor NF-κB, but how it does so and whether or not this activity contributes to lymphomagenesis is not known. Here we identify the mechanisms by which the fusion protein activates NF-κB and show that absence of c-IAP2 ubiquitin protein ligase activity in mice, as is the case in patients that express the fusion protein, results in spontaneous activation of NF-κB and many of the phenotypic cellular features of MALT lymphoma. Our findings demonstrate that c-IAP2 ubiquitin protein ligase activity dampens constitutive NF-κB activity and maintains B cell homeostasis, and provide genetic evidence that the loss of this enzymatic activity in the fusion protein has a major contributing role in MALT lymphomagenesis.
| The defining characteristic of the IAP (Inhibitor of Apoptosis) gene family is the presence of one or more baculovirus IAP repeats (BIRs) (reviewed in [1]). These ∼70 amino acid regions mediate protein-protein interactions, and in the context of adjacent sequences are responsible for the association of certain IAP family members with caspases. There are eight mammalian IAPs. Some IAPs also contain a RING motif that confers ubiquitin protein ligase (E3) activity. c-IAP1 and c-IAP2 are such RING-containing proteins that bind caspase-7 or -9 but, unlike XIAP, do not inhibit their enzymatic activities [2]. c-IAP1 and c-IAP2 also bind the adaptor protein TNF Receptor Associated Factor 2 (TRAF2) and are components of the Tumor Necrosis Factor Receptor 1 (TNF-R1) and 2 signaling complexes [3]. Upon TNF-R2 occupancy, c-IAP1, but not c-IAP2, ubiquitinates TRAF2 and the mitogen activated protein (MAP) kinase kinase kinase ASK1, resulting in the proteasomal degradation of all three proteins, cessation of MAPK signaling, and an increased susceptibility to cell death [4]–[6].
An emerging body of evidence has implicated the c-IAPs in regulating the activation of the transcription factor NF-κB. NF-κB can be activated by two distinct signaling mechanisms (reviewed in [7],[8]). The most common (the canonical pathway) depends on IκB kinase (IKK) β-mediated phosphorylation of inhibitory IκB proteins, leading to their ubiquitination and degradation. This frees cytosolic NF-κB heterodimers to translocate to the nucleus and regulate gene transcription. The second activating mechanism (the non-canonical pathway) is downstream of a limited number of receptors, including CD40, lymphotoxin β receptor, and BAFF receptors, and involves the proteolytic removal of carboxy-terminal ankyrin motifs in the NF-κB protein p100 to yield p52 [9],[10]. p52/Rel B-dimers translocate to the nucleus and regulate gene transcription [11]. Processing of p100 to p52 is dependent on the sequential activation of the upstream kinases NIK (NF-κB-inducing kinase) and IKKα [12]–[14]. Transient overexpression of c-IAP2 in cells has been shown to induce the ubiquitination and degradation of IκB, the essential antigen receptor NF-κB signaling intermediate Bcl-10, and NIK [15]–[18]. Overexpression of c-IAP1 induced the ubiquitination and degradation of TRAF2 and NIK, and its knockdown with silencing RNA impaired TNFα-induced NF-κB activation [4],[18]–[20].
Despite the (mostly in vitro) evidence for c-IAP regulation of NF-κB, primary cells from c-IAP1 and c-IAP2 knockout mice showed no obvious abnormalities in NF-κB activation [21],[22]. However, studies using synthetic “Smac mimetics” that induce the proteasomal degradation of c-IAP1 and c-IAP2, or siRNA to knock down the remaining c-IAP molecule expressed in cells from c-IAP1- and c-IAP2-deficient mice, have suggested that these two proteins may work redundantly to promote TNF-α-induced NF-κB activation and inhibit spontaneous non-canonical NF-κB activation [18],[20],[23]–[25]. Binding of c-IAPs to TRAF2 brings them into proximity with TRAF3-associated NIK. The result is a repressive complex that causes ubiquitination and degradation of NIK and maintains non-canonical NF-κB signaling in a basal state [3],[26],[27]. Consistent with this, tandem deletions of the c-IAPs have been associated with increased non-canonical NF-κB signaling and the development of multiple myeloma [28],[29], and conditional deletion of TRAF2 and TRAF3 results in stabilization of NIK, increased non-canonical signaling, and B cell hyperplasia [30]–[33].
MALT (mucosal associated lymphoid tissue) lymphomas are indolent neoplasms that have cytological features and bear cell surface markers of marginal zone B cells, and typically invade epithelial organs such as the gut and lung [34]–[36]. The molecular events that contribute to MALT lymphomagenesis are not well understood, but it is thought to involve the constitutive activation of NF-κB [37]. A variety of chromosomal abnormalities are associated with this disease; the most prevalent is a translocation, t(11;18)(q21;q21), that results in the production of a fusion protein containing the NH2-terminal (BIR-containing) fragment of c-IAP2 and the COOH-terminal portion of MALT1, a paracaspase involved in antigen receptor signaling [38]–[40]. Ectopic expression of this fusion protein in cell lines activates NF-κB [41], and transgenic overexpression in mice results in an increase in marginal zone B cells [42]. It is thought that the fusion protein activates NF-κB via the canonical signaling pathway [43]–[46]. The relevance of the different domains in the c-IAP2/MALT1 fusion protein to the development of MALT lymphoma has not been addressed.
Here we investigate the mechanism by which the c-IAP2/MALT1 fusion protein contributes to the development of MALT lymphoma. Ectopic expression of the fusion protein in cell lines activated both the canonical and non-canonical NF-κB signaling pathways, the former but not the latter being dependent on the MALT1 paracaspase activity. Expression of a mutant c-IAP2 that, like the c-IAP2 portion of the fusion protein, lacks E3 activity activated non-canonical but not canonical NF-κB. Knockin mice expressing this same c-IAP2 mutant in lieu of the wild type gene accumulated abnormal B-cells that had elevated non-canonical but not canonical NF-κB signaling, a cell-autonomous survival advantage in vivo, and other features of MALT lymphomas. The many points of similarity between mice expressing a c-IAP2 E3-inactive mutant and patients expressing a c-IAP2 E3-inactive MALT1 fusion protein suggests that the loss of this activity activates non-canonical NF-κB and predisposes to malignancy.
Ectopic expression of the c-IAP2/MALT1 fusion protein causes p65 to translocate to the nucleus, evidence of canonical NF-κB activation [43]. We assessed the mechanism of NF-κB induction in 293T cells transfected with the fusion protein (Figure 1A and 1B). Expression of the c-IAP2/MALT1 fusion protein induced IκB phosphorylation, as did a constitutively active form of IKKβ (IKKβ-CA). Unlike IKKβ-CA, however, c-IAP2/MALT1 resulted in little if any IκB degradation, suggesting that it is a much less potent activator of canonical signaling. Notably, the c-IAP2/MALT1 fusion protein, but not IKKβ-CA, also increased the levels of NIK and p52, hallmarks of non-canonical signaling. Expression of MALT1 did not induce IκB phosphorylation or degradation, or increase NIK or p52. Therefore, the fusion protein can trigger both arms of the NF-κB signaling cascade.
The MALT1 portion of the c-IAP2/MALT1 fusion protein has paracaspase activity, and it has been shown that expression of an inactivating mutation resulted in approximately 2-fold less NF-κB reporter activity than the paracaspase-active form [47]. We compared NF-κB activation in 293T cells transfected with the native sequence or paracaspase-inactive (c-IAP2/MALT1C464A) c-IAP2/MALT1 cDNA (Figure 1C). The canonical pathway, as judged by IκB phosphorylation, was markedly reduced by the mutation, but increases in the non-canonical pathway components NIK and p52 were unaffected. The fusion protein lacks the c-IAP2 RING domain and therefore its E3 activity, and c-IAPs have been shown to ubiquitinate NIK and repress non-canonical NF-κB signaling [18],[26],[27]. In fact, expression of c-IAP2 lacking its c-terminal half, as occurs in c-IAP2/MALT1 fusion proteins, increased both NIK and p52 levels (Figure S1). To ask if this was due specifically to the loss of E3 activity, we expressed c-IAP2 in which a RING histidine that is critical for E3 activity was replaced by alanine (c-IAP2H574A), but the protein was otherwise intact (Figure 1A) [48],[49]. Expression of c-IAP2H574A induced little if any IκB phosphorylation but increased NIK and p52 levels (Figures 1D and S1). These results show that the c-IAP2/MALT1 fusion protein activates both the canonical and non-canonical NF-κB signaling cascades and that there are two distinct mechanisms. The finding that expression of E3-defective c-IAP2 (Figure 1D) but not MALT1 (Figure 1B) activated non-canonical NF-κB raised the possibility that a similar mechanism might account for non-canonical NF-κB activation by the c-IAP2/MALT1 fusion protein.
To investigate the consequences of expressing c-IAP2 lacking E3 activity in vivo, we generated gene-targeted knockin mice that express an E3-inactive mutant of c-IAP2 (c-IAP2H570A) under the control of the native regulatory regions (Figure 2A). ES cells that had integrated the targeting vector were used to generate chimeric mice that were crossed to the C57BL/6 background. The presence of the H570A substitution in F1 offspring and subsequent generations was assessed by long-template PCR followed by Spe 1 restriction endonuclease digestion. The expected fragment sizes generated from the wild type allele are 4.9 and 0.7 kb, and those from the c-IAP2H570A allele are 4.3 and 0.6 kb (Figure S2 and Figure 2B). Acquisition of the mutant allele in c-IAP2+/H570A and c-IAP2H570A/H570A mice caused the appearance of shorter fragments in a gene dose-dependent manner.
Mutation of the Zn2+-coordinating histidine in the c-IAP1, c-IAP2, and XIAP RING domains [48],[49] prevents autoubiquitination and results in increased protein levels in cells transiently expressing the corresponding cDNAs. Furthermore, under physiologic conditions c-IAP1 downregulates c-IAP2 protein levels by trans-ubiquitination and proteasomal degradation [21]; there does not seem to be a reciprocal regulation of c-IAP1 by c-IAP2 [22]. To determine how c-IAP2 E3 activity might affect c-IAP levels, splenocyte lysates were immunoblotted with an antiserum that recognizes both c-IAP2 and c-IAP1 (Figure 2C) [50]. The antibody detected a doublet in wild type cells, the upper and fainter band being c-IAP2 and the lower and more prominent being c-IAP1 [21]. There was a marked increase in c-IAP2 expression in c-IAP2+/H570A cells and an even greater increase in c-IAP2H570A/H570A cells. In contrast, there was only a small increase in the level of c-IAP1. We compared the susceptibility of wild type c-IAP2 and the RING-less c-IAP2/MALT1 fusion protein to ubiquitination-dependent degradation. Consistent with a previous report [51], only levels of c-IAP2 increased in response to proteasome inhibition, indicating that the lack of E3 activity also stabilizes the fusion protein (Figure S3). Given that c-IAP2 expression is also regulated by c-IAP1-mediated ubiquitination [21], these results indicate that the combined activity of the c-IAPs is required to maintain c-IAP2 at physiologic levels.
Homozygous c-IAP2 knockin mice were viable, fertile, and displayed no obvious phenotypic abnormalities. Analysis of peripheral lymphoid organs in 6–7-month-old c-IAP2H570A/H570A mice, however, revealed a number of abnormalities. Unlike the spleen, cell numbers of pooled peripheral lymph nodes (axial, brachial, superficial cervical, and inguinal) as well as mesenteric lymph nodes were markedly increased (Figure 3G, A, and D). There was a reduction in the percentage of T cells with a corresponding increase in the percentage of B (B220+) cells (Figure 3B, E, and H). The result was approximately a 5-fold and 4-fold increase in the absolute number of pooled and mesenteric lymph node B cells, respectively, and a smaller (2-fold) increase in T cell number (Figure 3C and F). The CD4+∶CD8+ T cell ratio in c-IAP2H570A/H570A mice was normal (unpublished data). c-IAP2H570A/H570A lymphocytes had an unactivated phenotype, with normal levels of B7.1 and I-Ab (B cells) and CD25 and CD69 (T cells) (unpublished data). Two- to three-month-old c-IAP2H570A/H570A mice also had increases in lymph node B cells, although to a lesser extent than older animals (Figure S4). Analysis of B and T cell precursors in bone marrow and thymus, respectively, revealed no abnormalities. Among splenic B cells there was reproducibly an approximately 3-fold increase in the percentage of cells with a marginal zone phenotype (CD21hiCD23−), with a compensatory decrease in the percentage of follicular (CD21intCD23hi) and immature (CD21−CD23−) B cells (Figure 3J). Although lymph nodes normally have few B cells with a marginal zone phenotype [52], there was a small increase in these cells in c-IAP2H570A/H570A lymph nodes. Circulating IgA was increased approximately 3-fold in c-IAP2H570A/H570A mice, and there were highly statistically significant increases in IgM and IgG3, and a reduction in IgG1 as well (Figure 4). No statistically significant changes were found in IgG2b and IgE levels.
B cell hyperplasia, particularly of marginal zone B cells, in gut-associated lymphoid tissue (GALT) and lung is a feature of MALT lymphomas [34],[36]. Gross examination revealed that c-IAP2H570A/H570A mice had enlarged GALT and mesenteric lymph nodes, which was confirmed by histological evaluation (Figure 5A). There were also mild to moderate lymphocytic infiltrates in the lung (Figure 5B), with no evidence of neoplasia in either organ. Despite the increased size of the GALT in c-IAP2H570A/H570A mice, immunohistochemistry and flow cytometric analysis of both wild type and c-IAP2H570A/H570A GALT revealed primarily B cells with a follicular phenotype (Figure 5C and 5D), organized T-cell-enriched areas (compare Figure 5C with Figure S5), and no evidence of cellular activation (unpublished data). The lymphocytic infiltrates in the lungs of the c-IAP2 knockin mice also consisted of B cells and T cells (unpublished data). Taken together, these results demonstrate that mice with catalytically inactive c-IAP2 acquire a lymphoid phenotype that shares many features with MALT lymphomas.
The increase in B cell numbers in vivo could be due to decreased death, increased expansion, or a combination. Susceptibility to cell death was determined by culturing splenocytes in the absence of growth or survival factors and quantifying cell viability of B220+ and TCRβ+ cells by measuring 7-AAD incorporation (Figure 6A). c-IAP2 knockin B cells died more slowly than wild type cells, with 10%–15% still viable even after 64 h, compared to 3% for wild type cells. Addition of BAFF or agonistic anti-CD40 partially rescued the survival of B cells of both genotypes with similar dose-response curves (Figure 6B and unpublished data). There was no difference between the genotypes with regard to T cell survival (Figure 6A). Proliferative ability was addressed by stimulating purified B cells with anti-μ F(ab′)2 or lipopolysaccharide (LPS) and measuring 3H-thymidine incorporation (Figure 6C). c-IAP2H570A/H570A B cells had enhanced responses to both stimuli, with approximately a 3-fold shift in the dose response curve toward lesser concentrations of stimulus compared to wild type cells. During the course of the proliferation assays there were no differences between the two genotypes with regard to cell death (unpublished data). To determine if these in vitro observations correspond to B cell behavior in vivo, experiments were performed in which a mixture of wild type and c-IAP2 knockin splenic B cells was adoptively transferred into RAG2-deficient mice (Figure 6D). Although equal numbers of cells of each genotype were injected, after 45 d a 3-fold (lymph node) to 5-fold (spleen) preponderance of c-IAP2 knockin B cells was observed. These results show that the absence of c-IAP2 E3 activity in B cells results in a cell-intrinsic abnormality that increases their capacity to survive and/or proliferate in vitro and in vivo.
Ectopic expression of a c-IAP2/MALT1 fusion protein spontaneously activates NF-κB, as does depletion of c-IAPs with Smac mimetics or silencing siRNAs [18],[25],[26],[38],[41],[53],[54]. We therefore asked if selective loss of c-IAP2 E3 activity, in an otherwise physiological setting, affects NF-κB. Quantitative RT-PCR found that transcripts for NF-κB-responsive genes encoding GADD45β, IκB, c-IAP2, and ferritin heavy chain were elevated in c-IAP2H570A/H570A B cells (Figure 7A) [15],[55]–[57]. There was no increase, however, in the expression of Bcl-2, a gene product that has been reported to increase in response to canonical but not non-canonical NF-κB activation [58],[59], raising the possibility that NF-κB activation in c-IAP2H570A/H570A B cells was pathway-specific. Activation of the canonical pathway was assessed by measuring IκB levels and its state of phosphorylation. IκB levels were similar to or perhaps slightly increased in c-IAP2H570A/H570A B cells (Figure 7B) and murine embryonic fibroblasts (MEFs) (Figure 7C) compared to wild type cells. More importantly, there was no increase in spontaneously phosphorylated IκB in c-IAP2H570A/H570 cells, arguing against spontaneous canonical NF-κB activation. In contrast, the levels of both NIK and p52 were elevated in knockin B cells (Figure 7D) and MEFs (Figure 7E). The levels of TRAF2 and TRAF3, two components of a c-IAP-containing inhibitory complex thought to degrade NIK [26],[27], were unaffected by the loss of c-IAP2 E3 activity (unpublished data). In T cells, the amount of NIK was lower in wild type T than wild type B cells, and there was little increase in T cells expressing E3-inactive c-IAP2 (Figure 7D). There was correspondingly little increase in p52, although a small amount was detected in c-IAP2H570A/H570A T cells. Together, these results indicate the E3 activity of c-IAP2 is required to inhibit constitutive non-canonical NF-κB activation in B cells, MEFs, and to a much lesser degree, T cells.
Although the E3 activity of c-IAP2 is absent in both c-IAP2−/− and c-IAP2H570A/H570A cells, only the latter has increased spontaneous NF-κB activation. Because c-IAP1 also binds TRAF2, which is essential for c-IAP-mediated repression of the non-canonical signaling cascade [26], the results are consistent with the possibility that the E3-defective c-IAP2 competes with endogenous c-IAP1. In fact, c-IAP2H570A/H570A is able to bind TRAF2 at least as well as the wild type protein (Figure S6). To ask if the c-IAP2 RING mutant interfered with endogenous c-IAP1, c-IAP2-specific siRNA was used to knock down c-IAP2 in wild type and c-IAP2 knockin MEFs (Figure 7F). As seen in splenocytes (Figure 2C), there was a large increase in c-IAP2 and a small increase in c-IAP1 levels in c-IAP2 knockin MEFs (Lanes 1 and 3). Transfection of wild type MEFs with c-IAP2 siRNA specifically reduced c-IAP2 but had little if any effect on the levels of p52. In contrast, knockdown of c-IAP2 in c-IAP2H570A/H570A MEFs resulted in a substantial reduction of p52 levels. To determine if c-IAP2H570A interferes with c-IAP1-mediated ubiquitination/degradation of NIK, 293T cells were co-transfected with NIK and c-IAP1, with or without c-IAP2H570A (Figure 7G). Consistent with previous reports [18],[26], expression of c-IAP1 reduced NIK to undetectable levels; this was prevented by co-expression of E3-inactive c-IAP2. Thus, the E3-defective c-IAP2H570A can inhibit constitutive c-IAP1-mediated ubiquitination/degradation of NIK and de-repress the non-canonical signaling cascade.
Unmanipulated mice deficient for c-IAP1 and c-IAP2 have no obvious phenotypic abnormalities [21],[22], which has made it difficult to ascribe a physiologic role to these proteins in vivo. Recent studies have suggested that redundancy among the c-IAPs, at least with regard to NF-κB activation, could account for the lack of apparent abnormalities [18],[25],[26]. If so, this could be an even bigger factor in c-IAP1 knockout mice, in which c-IAP2 levels are elevated because it is no longer ubiquitinated by c-IAP1 and targeted for degradation [21]. c-IAP1 is not elevated in cells from c-IAP2-deficient mice [22], suggesting that even normal c-IAP1 levels are sufficient to compensate for the loss of c-IAP2. In contrast to the c-IAP2 knockout animals, we have found that substitution of wild type c-IAP2 with an E3-defective point mutation does result in constitutive NF-κB activation and abnormal B cell accumulation. This is likely because the endogenous c-IAP1 is unable to compensate for the lack of c-IAP2 E3 activity. The N-terminal BIR-containing region of both proteins binds to TRAF2, a prerequisite for c-IAP-mediated NIK ubiquitination [18],[26]. Furthermore, only one c-IAP molecule can bind one TRAF2 trimer at a time [60]. We found that overexpressed c-IAP2H570A/H570A interferes with c-IAP1-mediated degradation of NIK and that knockdown of the c-IAP2 mutant restored repression of non-canonical NF-κB. These data argue that the mutant c-IAP2 prevented c-IAP1 from associating with the repressive complex. The c-IAP2H570A/H570A mice therefore represent an example in which replacement of the endogenous gene with an inactive form, but not a complete knockout, can reveal normal function.
Abnormal B cell expansion has been observed in a number of animal models in which NF-κB activity is chronically elevated. For example, overexpression of B cell activating factor (BAFF) or NIK, both of which lead to non-canonical NF-κB activation, results in B cell hyperplasia with increased numbers of CD23loCD21hi B cells [59],[61]. Similarly, mice lacking either TRAF2 or TRAF3 in B cells have elevated non-canonical NF-κB, an expanded B cell compartment, increased numbers of cells with a marginal zone phenotype, and elevated serum immunoglobulins [31]–[33]. TRAF2 and TRAF3 are adaptor molecules downstream of BAFF receptors that constitutively form a complex with c-IAP1, c-IAP2, and NIK [26],[27]. These associations result in c-IAP-dependent ubiquitination of NIK and its proteasomal degradation, which is thought to maintain the non-canonical NF-κB activation pathway in a basal state. Although we found no alterations in expression of TRAF2 and TRAF3 in c-IAP2H570A/H570A mice, NIK levels and NF-κB activity were increased, and the mice developed age-dependent B cell hyperplasia in a manner similar to BAFF and NIK transgenic mice, or TRAF2 and TRAF3 knockout mice [31]–[33],[59],[61]. The data are all consistent with the notion that basal ubiquitination of NIK by c-IAP2 is an important mechanism for regulating constitutive NF-κB activity and B cell homeostasis.
It is widely believed that the c-IAP2/MALT1 protein is pathogenic because it activates the canonical NF-κB signaling pathway [37]. A variety of mechanisms have been suggested, including proteolytic cleavage of A20, a negative regulator of NF-κB activation, ubiquitination of NEMO, binding of the fusion protein to lysine 63-linked polyubiquitinated NEMO, and the failure of the fusion protein to degrade Bcl-10 [37]. However, a potential role for non-canonical NF-κB activation has not been explored. We have found that the c-IAP2/MALT1 fusion protein activates both canonical and non-canonical signaling pathways, and activation of the latter in mice is sufficient to promote the development of features common to MALT lymphoma. Our results are in agreement with a report that overexpression of the fusion protein in 3T3 cells resulted in an NF-κB complex that was supershifted with antibodies to RelB [62]. Interestingly, introduction of a Bcl-10 transgene, which mimics the MALT lymphoma-associated t(1;12)(p22;q32) chromosomal translocation that deregulates Bcl-10, results in marginal zone B cell hyperplasia and elevated non-canonical as well as canonical NF-κB signaling [63]. It is noteworthy that mice lacking the COOH-terminal ankyrin domain of p100, which results in constitutive activation of p52, develop B cell hyperplasia and enlarged GALT. Thus, activation of the non-canonical pathway may be a major contributor to the development of MALT lymphoma. The development of MALT lymphoma-like abnormalities in the c-IAP2 E3-defective mice raises a cautionary note that drugs that reduce c-IAP levels, such as SMAC mimetics, may have unintended side effects due to activation of non-canonical NF-κB signaling, especially if administered chronically.
RAG2-deficient and CD45.1 congenic mice were obtained from the Jackson Laboratory. All restriction endonucleases were obtained from New England Biolabs. pCMV9 containing carboxy-terminal myc-tagged human NIK cDNA was obtained from Nobuhiko Kayagaki and Vishva Dixit (Genentech) and pRK5 containing Flag-tagged human c-IAP2 and c-IAP2/MALT1 was obtained from Xiaolu Yang (University of Pennsylvania). pRK5-Flag-tagged human c-IAP2H574A and pRK5-Flag-tagged human c-IAP2/MALT1C464A were generated by site directed mutagenesis using the primers 5′-GTCCATAGTGTTTATTCCTTGTGGTCATCTAGTAGTATGCAAAGATTGTGC-3′, 5′-GCACAATCTTTGCATACTACTAGATGACCACAAGGAATAAACACTATGGAC-3′, 5′-GACTTAATGTGTTCTTATTGGATATGGCTAGGAAAAGAAATGACTACGATGATAC-3′, 5′-GTATCATCGTAGTCATTTCTTTTCCTAGCCATATCCAATAAGAACACATTAAGTC-3′, respectively, and the QuickChange mutagenesis system from Stratagene. pRK5-Flag-tagged human c-IAP2ΔCARD-RING was generated by cloning a PCR product amplified from human c-IAP2 cDNA into pRK5 that already contained cDNA encoding the Flag-tag using the primers 5′-GCTCGTGAATGCGGGATCCTCTAGAAACATAGTAGAAAACAGC-3′ and 5-GCTGCAACGTAAGCTTTCATTCATTTGATTCTTTTTCCTCAGTTGC-3′, BamH1 and HindIII. Presence of the mutations was confirmed by direct sequencing. pCMV-Tag2 murine c-IAP2 has been described [21]. GST-tagged murine c-IAP2 was obtained by subcloning into pGEX-6P-1 (Amersham). GST-tagged murine c-IAP2H570A was generated by site directed mutagenesis using primers that have been described [21]. Myc-tagged murine c-IAP1 was obtained by subcloning into pCMV-Tag5 (Clontech). IKKβ-CA has been described [64]. pCMV4 containing Flag-tagged IκB cDNA was obtained from Dean Ballard (Vanderbilt University). The anti-c-IAP antibody was obtained from Herman Chung and Bob Korneluk (Apoptosis Research Center, Children's Hospital of Eastern Ontario), anti-NIK and anti-p52 from Cell Signaling Technologies, anti-phospho-IκB and anti-IκB from Santa Cruz, and anti-FLAG and anti-β-actin from Sigma. Anti-CD40 (HM40-3) was obtained from BD Biosciences. BAFF was obtained from Peprotech. The fluorescently labeled antibodies used for analysis of lymphoid populations in the thymus, bone marrow, lymph node, and spleen by flow cytometry were obtained from BD Biosciences. The Mouse Immunoglobulin Isotype Panel (Southern Biotech) was used to quantify the serum immunoglobulin titers for IgM, IgG1, IgG2b, IgG3, and IgA. The OptEIA Set Mouse IgE (BD Biosciences) was used to quantify the amount of serum IgE. The anti-B220/CD45R and anti-CD3 used from B and T cell immunohistochemistry were purchased from BD Biosciences and Serotec, respectively. The B cell and T cell enrichment kits were obtained from Stemcell Technologies. For some experiments B cells were purified using the Mouse B cell Recovery Column Kit from Cedarlane Laboratories Ltd. The primer sequences used in the quantitative PCR are as follows, GADD45β 5′ (5′-CTGCCTCCTGGTCACGAA-3′), GADD45β 3′ (5′-TTGCCTCTGCTCTCTTCACA-3′), IκB 5′ (5′-TCACGGAGGACGGAGACTCG-3′), IκB 3′ (TGGAGATGCTGGGGTGTGC), ferritin heavy chain 5′ (5′-GGAGTTGTATGCCTCCTACGTCT-3′), ferritin heavy chain 3′ (5′-TGGAGAAAGTATTTGGCAAAGTT-3′), c-IAP2 5′ (5′-TATTTGTGCAACAGGACATTAGGAGT-3′), c-IAP2 3′ (TCTTTCCTCCTGGAGTTTCCG), Bcl-2 5′ (5′-GTACCTGAACCGGCATCTG-3′), and Bcl-2 3′ (5′-GGGGCCATATAGTTCCACAA-3′). The HPRT primers have been described [65]. c-IAP2 siRNA has been described [26] and was modified to Stealth RNAi siRNA. The sequences of the oligonucleotides are 5′-AAGUGGUAGGGACUUGUGCUCAAAG-3′ and 5′-CUUUGAGCACAAGUCCCUACCACUU-3′.
The BamH1-EcoR1 and EcoR1-EcoR1 recombination arms used to generate the c-IAP2H570A targeting construct were obtained from BAC-DNA (clone 239-13P; Research Genetics) using the respective endonucleases and subcloned into shuttle vectors. To insert the silent mutation in the neighboring leucine codon introducing a novel Spe1 restriction endonuclease site and then replace the histidine codon with an alanine codon, the BamH1-Ecor1 arm was sequentially mutagenized using mutagenic primers 5′-CATCGTGTTCATTCCCTGTGGCGCACTAGTCGTGTGCAAAGACTGCG-3′ and 5′-CGCAGTCTTTGCACACGACTAGTGCGCCACAGGGAATGAACACGATG-3′, and then 5′-CATTCCCTGTGGCCATCTAGTCGTGTGCAAAGACTGC-3′ and 5′-GCAGTCTTTGCACACGACTAGATGGCCACAGGGAATG-3′ using the QuickChange mutagenesis system from Stratagene. The presence of H570A in exon 9 and absence of other spontaneous mutations in the other exons were confirmed by direct sequencing. After subcloning both recombination arms into a vector containing a neomycin cassette flanked by two loxP recombination sites, the resultant targeting vector was linearized with Not1 and transfected into ES cells. Stable transfectants were screened by southern blotting and long-range polymerase chain reaction (LR-PCR) coupled with Spe1 restriction endonuclease digestion. The primers used to screen the c-IAP2H570A/H570A mice were obtained from Invitrogen and their sequence was 5′ CGAAAAAGATGCCCATCTACTCAG-3′ and 5′-TATCCCTAAAATGTCATCCAATAAATAACAG-3′. The clone that had correctly integrated the targeting construct at the c-IAP2 locus was injected into blastocytes to generate chimeric mice. F1 offspring of the chimeric mice were backcrossed 6 additional times to the C57BL/6 (B6) background and then c-IAP2+/H570A were interbred to obtain c-IAP2H570A/H570A mice. B6 mice bred in the CRC Vivarium (NIH) were used as controls for all experiments.
All animal experimental procedures were approved by the Animal Care and Use Committee of the National Cancer Institute.
The fragment spanning the recombination arm containing c-IAP2H570A was amplified from tail DNA using buffer 3 from the Expand Long Template PCR System (Roche) and the c-IAP2 locus 5′ and c-IAP2 locus 3′ primers, digested with Spe1, and resolved by agarose gel electrophoresis. Total RNA was isolated from purified B cells using the Utraspec RNA isolation reagent (Biotecx laboratory) and reverse transcribed using Superscript II Reverse Transcriptase kit (Invitrogen) following the manufacturers' protocol. The amount of ferritin heavy chain, IκB, c-IAP2, GADD45β, Bcl-2, and hypoxanthine phosphoribosyltransferase (HPRT) mRNA was quantified using the respective primers, SYBR Green PCR Master Mix (Applied Biosystems), and the 7500 Real Time PCR System (Applied Biosystems). The values were normalized to HPRT and the percent increase relative to wild type was calculated by dividing the c-IAP2 knockin values by the wild type values.
Bone marrow, thymus, spleen, lymph nodes (superficial cervical, axillary, brachial, inguinal, and mesenteric) and GALT were harvested from wild type and c-IAP2H570A/H570A mice, disrupted by teasing, and total cell suspensions made by gently mashing the debris through 40 µM nylon mesh (BD Biosciences). The cells were counted and the distribution of lymphoid populations in each organ was determined by cell surface staining and flow cytometry. B and T cells were purified from spleen and lymph nodes from wild type and c-IAP2H570A/H570A mice using B and T cell enrichment kits following the manufacturer's protocol. The purity was determined by cell surface staining and flow cytometry, and for all experiments, greater than 90%. In some experiments B cells and splenocytes were cultured in RPMI supplemented with 10% fetal calf serum, 100 U/ml penicillin, 100 µg/ml streptomycin, 2 mM L-glutamine, and 50 µM-β-mercaptoethanol. MEFs were prepared from day 13.5 embryos as described [66] and maintained in Dulbecco's Modified Eagle's medium supplemented with 10% fetal calf serum, 100 U/ml penicillin, 100 µg/ml streptomycin, 2 mM L-glutamine, and 50 µM-β-mercaptoethanol.
For quantifying cell death, splenocytes (7.5×105 cells/ml) were incubated in vitro, stained with fluorescently labeled anti-B220/CD45R and anti-TCRβ, and incubated with 7-amino-actinomycin D (7AAD; 1 µg/ml). Uptake of 7AAD by dying B (B220+) and T (TCRβ+) cells was quantified by flow cytometry. The percentage of viable cells was calculated by dividing B220+7AAD− or TCRβ+7AAD− by the total B220+ or TCRβ+ cells at each time point. For BAFF- and anti-CD40-induced survival, purified B cells were incubated at (7.5×105 cells/ml) with the indicated concentrations of BAFF or agonistic anti-CD40 (100 ng/ml) for 66 h, stained with fluorescently labeled anti-B220 and 7AAD, and analyzed by flow cytometry. To assess proliferation, purified B cells (2.5×105 cells/ml) were stimulated with anti-μ F(ab′)2 (Jackson ImmunoResearch Laboratories, Inc.) or LPS (Sigma), and during the final 18 h of the 66 h period, DNA synthesis was measured by adding 1 µCi 3H-thymidine to the culture. The cells were then harvested and lysed, and the DNA was transferred to a filtermat. The amount of incorporated 3H-thymidine was quantified using a scintillation counter.
B cells, T cells, and MEFs were lysed in a buffer containing 20 mM Tris pH 7.5, 150 mM NaCl, 1% Triton X-100, 1 mM EDTA, 30 mM NaF, 2 mM sodium pyrophosphate supplemented with Complete (Roche) protease inhibitor cocktail, and the detergent-soluble lysate was collected after centrifugation. Lysates were normalized to protein concentration, denatured in sample buffer (50 mM Tris pH 6.8, 10% glycerol, 2% SDS, 2% β-mercaptoethanol, and 0.04% bromophenol blue), resolved by SDS-PAGE, and immunoblotted with the appropriate antibodies. For knockdown studies, 3.0×105 MEFs were plated in 60 mm cell culture dishes and 16 h later transfected with 30 nM of Universal Lo GC content non-targeting or c-IAP2 Stealth iRNA siRNA using Lipofectamine RNAiMAX (Invitrogen) following the manufacturer's protocol. After 24 h the cells were washed twice with phosphate-buffered saline (PBS) and lysed. For ectopic expression studies, 293T cells were transfected with the indicated plasmids using Lipofectamine 2000 (Invitrogen) following the manufacturer's protocol. Twenty-four hours later the cells were harvested, washed with PBS, counted, and lysed in sample buffer.
Glutathione S-transferase (GST)-tagged proteins were expressed in DH5α cells with 0.05 mM isopropyl-β-thiogalactopyranoside at 16°C for 20 h and lysed in 20 mM HEPES pH 7.5, 100 mM NaCl, 1.5 mM MgCl2, 1% Triton X-100. The recombinant proteins were purified from clarified lysates using glutathione Sepharose 4B beads (Amersham Biosciences). The beads were washed extensively and incubated with 35S-labeled TRAF2 that had been translated in vitro using the TNT Quick Coupled Transcription/Translation System (Promega) for 3 h at 4°C in binding buffer containing 120 mM NaCl, 10% glycerol, 1% Triton X-100, and 50 mM Tris pH 7.5. The bead-bound complexes were washed with the binding buffer, eluted with sample buffer, and resolved by SDS-PAGE.
Equal number of splenic B cells purified from wild type (CD45.1+) and c-IAP2H570A/H570A (CD45.2+) knockin mice were mixed and 107 cells were injected into the tail veins of RAG2-deficient (CD45.2+) mice. Forty-five days later the percentage of wild type and c-IAP2H570A/H570A B cells in the spleens and lymph nodes was determined by staining cell suspensions with B220, CD45.1, and CD45.2 and analyzed by flow cytometry. The ratio was generated by dividing the percentage of c-IAP2H570A/H570A B cells by the percentage of wild type B cells.
Mice were euthanized using CO2 inhalation and necropsies were performed. A comprehensive set of organs and tissues were collected and fixed in 10% neutral buffered formalin. Tissues were paraffin-embedded, sectioned at 5 µm, and stained with hematoxylin and eosin. For lymphocytes, slides were stained with biotin-conjugated anti-B220/CD45R or anti-CD3. The antigens were retrieved by microwaving in EDTA (B220) or citrate buffer (CD3). Detection of B220 was performed using the avidin-biotinylated enzyme complex (Vector Laboratories) with 3,3′-diaminobenzidine (Sigma) as chromagen. Detection of CD3 was accomplished using the Rabbit Elite kit (Vector Laboratories) using 3,3′-diaminobenzidine as chromagen. Slides were counterstained with hematoxylin. Stained sections were evaluated by a boarded veterinary pathologist. Serum immunoglobulin isotypes were quantified by ELISA following the manufacturer's protocol. p values were calculated using GraphPad Prism and a two-tailed t test.
|
10.1371/journal.pbio.2006357 | Evolutionary diversification of the HAP2 membrane insertion motifs to drive gamete fusion across eukaryotes | HAPLESS2 (HAP2) is a broadly conserved, gamete-expressed transmembrane protein that was shown recently to be structurally homologous to viral class II fusion proteins, which initiate fusion with host cells via insertion of fusion loops into the host membrane. However, the functional conformation of the HAP2 fusion loops has remained unknown, as the reported X-ray structure of Chlamydomonas reinhardtii HAP2 lacked this critical region. Here, we report a structure-guided alignment that reveals diversification of the proposed HAP2 fusion loops. Representative crystal structures show that in flowering plants, HAP2 has a single prominent fusion loop projecting an amphipathic helix at its apex, while in trypanosomes, three small nonpolar loops of HAP2 are poised to interact with the target membrane. A detailed structure-function analysis of the Arabidopsis HAP2 amphipathic fusion helix defines key residues that are essential for membrane insertion and for gamete fusion. Our study suggests that HAP2 may have evolved multiple modes of membrane insertion to accommodate the diversity of membrane environments it has encountered during eukaryotic evolution.
| The fusion of gamete plasma membranes is the fundamental cellular event that brings two parental cells together to form a new individual, yet we know surprisingly little about this process at the molecular level. HAPLESS 2 (HAP2) is a conserved sperm plasma membrane protein that is essential for gamete fusion in a diverse array of eukaryotes. It was recently shown to share a common ancestor with viral proteins that drive fusion of the viral envelope with host membranes, but its mechanism of action remained elusive, since the reported structure did not resolve the proposed membrane interaction surface. Here, we report two new HAP2 structures revealing that HAP2 has evolved diverse membrane interaction surfaces. In the flowering plants, HAP2 uses an amphipathic helix that presents nonpolar residues to the target membrane; in trypanosomes, the membrane interaction surface comprises three shallow nonpolar loops.
| The fusion of gamete plasma membranes to form a zygote is central to sexual reproduction, yet a molecular mechanism for this fundamental process has only very recently been proposed. The crystal structure of the C. reinhardtii HAPLESS 2 (HAP2) ectodomain (CrHAP2e) [1] revealed that this broadly conserved, gamete-expressed transmembrane protein [2–4] has the same three-dimensional fold as class II viral fusion proteins. It was proposed that, like its viral counterparts, HAP2 initiates gamete fusion by insertion of fusion loops into the opposing gamete plasma membrane [5]. Consistent with this idea, mutations in the proposed fusion loops of CrHAP2e disrupted its ability to insert into artificial membranes in vitro and mediate gamete fusion in vivo [1]. Because the reported CrHAP2e structure did not include the critical region of the fusion loops, the question of how HAP2 inserts into the opposing gamete plasma membrane was left unanswered.
Viral fusion proteins have been divided into several structural classes that use either fusion peptides (class I) or fusion loops (classes II and III) to insert into host cell membranes [6]. HAP2 is homologous to the viral class II fusion proteins, which fold into an articulated rod made of three β-sheet-rich domains, termed I, II, and III, with the central domain I flanked by domains II and III in the prefusion form [7–9]. Domain III connects to the C-terminal transmembrane domain via a flexible linker, while domain II bears the fusion loops at the opposite end.
Stable insertion into the host membrane is required to withstand pulling forces that occur as the fusion protein bridges the intermembrane space via a transient extended intermediate state that later collapses into a trimeric “hairpin” conformation, which drives the merger of the viral envelope and host membrane [10]. The interaction with membranes was best studied for the class I proteins, in which the fusion peptide is an N-terminal extension that folds independently of the rest of the fusion protein upon its insertion into a lipid bilayer, exposing a relatively extensive nonpolar platform to the outer leaflet [11]. In the case of the influenza virus hemagglutinin, the prototypic class I fusion protein, it was shown that the fusion peptide forms an α-helical hairpin [12], exposing nine bulky nonpolar side chains to the external lipid layer of the host membrane. In contrast, class II proteins from arthropod-borne viruses (arboviruses) were proposed to insert internal fusion loops located at the domain II “tip,” i.e., at the end of the protein opposite the C-terminal transmembrane domain integral to the viral envelope. In contrast to the class I proteins, the fusion loops do not change conformation upon interaction with lipids, and the fusion protein of the Rift Valley fever virus (RVFV) was shown to feature an internal pocket that specifically accommodates the head group of glycerophospholipids from the outer leaflet of the target membrane [13]. The resulting polar interaction network allows stable insertion of the protein with only one or two bulky aromatic side chains of the fusion loops exposed to the aliphatic moiety of the membrane. Understanding how HAP2 achieves stable insertion into the target membrane is key for understanding the molecular mechanism driving gamete fusion. The diversity of reproductive systems in which HAP2 has been implicated across eukaryotic organisms likely represents a large variety of fusion environments (e.g., lipid compositions of gamete membranes), and it remains elusive whether HAP2 has evolved different membrane interaction surfaces to secure the initial interaction with the target membrane.
Here, we carried out a comparative structure/function study of HAP2 from distantly related eukaryotes and found striking sequence variability at the domain II tip, which features multiple insertions and deletions, contrasting with the relative conservation of the rest of the protein. We then obtained structural data for HAP2 from two organisms that displayed the most contrasting diversity in this region: the flowering plant Arabidopsis thaliana (AtHAP2) and the protozoan Trypanosoma cruzi (TcHAP2). The X-ray structures confirmed that the membrane interaction surfaces were totally different: while in AtHAP2 there is a single fusion loop that projects an amphipathic helix (termed αF) toward the membrane (an unprecedented feature in viral class II fusion proteins), in TcHAP2, the membrane interaction surface is composed of three short loops. Biochemical and genetic experiments focusing on αF confirmed that the nonpolar residues are required for membrane insertion in vitro and for gamete fusion in vivo. Bioinformatic analyses show that αF is likely to be conserved across flowering plants, and functional studies revealed that it is interchangeable between rice and Arabidopsis, which are among the most distantly related flowering plants.
The domain II tip is the region of class II fusion proteins that must firmly insert into target membranes to initiate the fusion process. In order to probe the potential structural diversity of HAP2 in this region, we used the structure of CrHAP2e to guide the alignment of 38 HAP2 amino acid (aa) sequences representing organisms from four eukaryotic kingdoms (Fig 1, S1 Table). We focused on the region corresponding to the b, c, and d β-strands and their interstrand connections at the domain II tip (Fig 1), which contains the fusion loops [1] (Fig 1B). Nine invariant residues anchored the alignment: seven cysteines (in green background) participating in highly conserved disulfide bonds; a conserved glutamate in β-strand b; and a conserved arginine residue in the cd connection, making a salt-bridge with the conserved glutamate, both residues shown in red background in Fig 1B (see Materials and methods). The resulting alignment showed, as expected, that sequences from relatively closely related groups (e.g., flowering plants) were similar to each other. But the comparison between different phylogenetic groups revealed three regions (vertical blue frames, Fig 1B) with high diversity in length and primary sequence right at the tip of domain II: regions 2 and 3 correspond to the two AtHAP2 predicted fusion loops [1] in the connection between strands c and d. Variable region 1 is at the beginning of the long bc connection, immediately downstream of strand b and preceding disulfide 2 at the membrane-proximal end of the bdc β-sheet (Fig 1C). Residues within this region of the bc connection were indeed shown to be part of the membrane-interacting region of other class II fusion proteins, e.g., bunyavirus Gc [14, 15]. Although variable region 1 is very short and retracted in CrHAP2 [1], and residues from this region are not predicted to reach the target membrane, the sequence alignment indicates that this is not the case in HAP2 from other organisms. For example, several orthologs have an insertion (left-most vertical blue box, variable region 1, Fig 1B)—the most prominent in the apicomplexan parasite Toxoplasma gondii—potentially making a loop projecting apically to make contact with the target membrane.
The HAP2 segment predicted to form fusion loop 1 in Chlamydomonas is highly variable (blue frame 2 in Fig 1B). This region is absent in flowering plants, displays an additional and unique pair of cysteine residues expected to make an extra disulfide bond in kinetoplastids, and is much larger in insects (66-residue insertion in Tribolium castaneum). The alignment also shows that most orthologs have significant deletions in the variable region 3 (third blue frame, Fig 1B), as observed in protozoan and animal sequences from the Porifera, Cnidaria, and Annelida. The short segment corresponding to the CrHAP2 α0 helix varies in length from four to seven residues and features α-helix-breaking residues (glycine or proline) in several orthologs (Fig 1B, in between the second and third blue boxes), suggesting α0 may not be a conserved structural feature of HAP2. The alignment of the predicted HAP2 membrane interaction domain from species representing the eukaryotic diversity therefore suggests that HAP2 has evolved multiple structural motifs for insertion into the target membrane.
To understand the organization of the divergent structural motifs used by HAP2 for membrane insertion, we selected representative sequences for further study from the flowering plants and the kinetoplastids, which were among those with the most contrasting features in variable regions 2 and 3 (Fig 1B). We expressed the recombinant ectodomains of A. thaliana HAP2 (AtHAP2e; aa 25–494) and T. cruzi HAP2 (TcHAP2e, aa 26–516) in Drosophila Schneider 2 (S2) cells for structural studies (see Materials and methods). Size exclusion chromatography (SEC) and multiangle static light scattering (MALS) revealed that both proteins behaved as monomers in solution (S1A–S1D Fig). AtHAP2e crystallized in the P63 hexagonal space group and diffracted anisotropically to a mean nominal resolution of 2.75 Å. TcHAP2e only crystallized upon limited proteolysis using subtilisin (see Materials and methods). The resulting purified protease-resistant 40-kD fragment (termed TcHAP2esub, S1E Fig) crystallized in the hexagonal space group P6122 (S1F and S1G Fig), and the best crystals diffracted to 3.1-Å resolution (see Materials and methods; crystallographic statistics are listed in S2 Table). We determined both structures by the molecular replacement method using a search model derived from the structure of CrHAP2e (Protein Data Bank [PDB]: 5MF1)
The experimental electron density map of AtHAP2 allowed us to trace 446 out of 469 ectodomain residues, including the region at the tip of domain II (S1H Fig, right panel) that was disordered in CrHAP2. The resulting atomic model of AtHAP2 revealed a trimer in unequivocal postfusion hairpin conformation (Figs 2A and 3A), as observed previously for CrHAP2. The crystals of TcHAP2esub consisted of domain II with a short extension into domain I (Fig 2A). There was continuous electron density for the loops at the tip of domain II (S1H Fig, left panel), and we could build the polypeptide chain unambiguously in this region, albeit for a 2-residue break immediately after strand b (i.e., in the bc loop, Fig 2A). Monomeric TcHAP2esub domain II displayed the same conformation adopted by domain II in the trimers of both AtHAP2e and CrHAP2e (Fig 2A). We were thus able to superpose the TcHAP2 domain II on its counterparts of either the CrHAP2 or the AtHAP2 trimer with reasonable subunit packing, suggesting that the postfusion TcHAP2 trimer would look very similar (Fig 3A).
The prominent α2 helix in domain II is nearly identical in length and orientation between AtHAP2, TcHAP2, and CrHAP2 (Fig 2A and Fig 3) and thus appears as a HAP2 structural hallmark. Pairwise comparisons of domain II from the three available HAP2 structures indeed showed very high DALI scores [21], relative to a similar comparison between viral class II proteins (S3 Table). To further compare HAP2 to its viral counterparts, we superposed the bdc β-sheet of domain II. The root-mean-square deviation (rmsd) obtained when superimposing the core 19 Cα atoms of the HAP2 bdc β-sheet onto the corresponding element in the flavivirus, alphavirus, and bunyavirus class II proteins ranged between 2.3 and 2.7 Å. Pairwise comparisons of the corresponding 19 Cα atoms from the three HAP2 structures discussed revealed an rmsd of <0.73 Å, highlighting the conservation of the structural core of the HAP2 domain II (Fig 2A and 2B). The conserved HAP2 salt bridge (E126–R185 in CrHAP2 [1], E117–R163 in AtHAP2, and E121–R176 in TcHAP2) superposes such that the side chains fall on top of each other when the superposition is based on the bdc β-sheet Cα atoms (Fig 2B). Furthermore, alignment of the bdc β-sheets also brings into superposition the ij loop (which is included in the pfam10699 segment, conserved among identified HAP2 sequences, [1]) together with the α2 helix. The high structural conservation of the central core of domain II thus contrasts with the high variability of its membrane insertion region.
The AtHAP2 fusion loop features an amphipathic helix (αF, Figs 2, 3 and 4) at its apex, positioned to interact parallel to the target membrane by insertion of its nonpolar surface (I171, F172, M175, I176) (Fig 4). Heliquest [22] predicted amphipathic helices in this segment for all plant and also for algal HAP2 sequences, including CrHAP2 (S2 Fig, sequences in Fig 1B). This server also predicted an amphipathic helix in variable region 2 of the cnidarian animal Nematostella vectensis (S2 Fig), suggesting that this motif may be of widespread use for HAP2 target membrane insertion.
In contrast to AtHAP2—in which the bc strand connection is basal to αF (Fig 2C) and away from the lipid contact area—in TcHAP2, both cd and bc connections are located roughly at the same level at the tip of domain II, suggesting that, together, they may comprise the membrane insertion element (Fig 1C, Fig 4D, right panel). A surface representation of the modeled TcHAP2 trimer indeed suggests a relatively flat membrane interaction surface with a tripartite nonpolar patch comprising V129 (in the N-terminal side of the bc connection) and L167, L168 and I183, F184 (in the variable regions 2 and 3, respectively, of the cd connection, S3A Fig). Compared to CrHAP2, in TcHAP2, the α0 helix features an additional helical turn after disulfide 5 (Fig 2B, compare left and middle panels). As anticipated from the aa sequence, the loop preceding α0 is stabilized by an extra disulfide bond, numbered 5b (Figs 1B and 2C, middle inset). The second loop (variable region 3) is just a turn connecting α0 to the d strand. Residues I183/F184 are located in the extra turn of α0 and are pointing inward (toward the bdc β-sheet) in the structure (S3 Fig), raising the possibility that the local conformation may be different in the aqueous solution used for crystallization than when bound to a lipid bilayer. A similar situation was observed for the fusion loops of the rubella virus fusion protein E1 [19].
The structural results, when combined with the sequence alignment, support the notion that HAP2 has evolved multiple modes for membrane insertion, with the algae having two fusion loops also predicted to bear amphipathic α-helices, only one loop in the flowering plants, and three small loops in the kinetoplastids (Figs 1B and 2C). The insertion mode of the RVFV Gc was shown to involve a lipid head group binding pocket located between the fusion loops and the end of the bdc β-sheet most proximal to the membrane (Fig 4D, left panel). In the case of AtHAP2, the structure suggests a different insertion mode, as the bdc β-sheet is retracted from the membrane (blue arrow in Fig 4D) at a distance incompatible with interactions with lipid heads. In contrast, AtHAP2 αF is poised to make multiple interactions within the hydrophilic region of the membrane (indicated roughly by the dotted lines in Fig 4D), while its nonpolar side can embed in the aliphatic moiety.
To test that AtHAP2 can indeed interact with membranes as predicted by the structures, we analyzed the behavior of AtHAP2e by mixing the monomeric protein (S1B Fig) with liposomes of varying composition (see Materials and methods), followed by gradient ultracentrifugation. We designed the gradients so that proteins bound to liposomes would float to the top fractions, while the unbound protein would sediment. We found that the soluble, wild-type (WT) AtHAP2e migrated to the top of the gradient, in contrast to a triple alanine substitution mutant in αF, which remained in the bottom of the tube (Fig 5A, S1 Data), indicating that HAP2 binds membranes and that the bulky nonpolar side chains of αF are required. Of note, we found that AtHAP2 association with liposomes was enhanced when artificial membranes included the negatively charged phospholipid 1,2-dioleoyl-sn-glycero-3-phospho-L-serine (DOPS; see Materials and methods). Electron microscopy further showed that both AtHAP2e (Fig 5B–5D) and TcHAP2e (S3B and S3C Fig) decorated the liposome surface, as had been observed with CrHAP2e. Side views of the proteoliposome edge showed approximately 12-nm-long projections with a tapered end toward the membrane (Fig 5B and 5D, S3A and S3B Fig), and top views showed a typical pattern of hexagonal packing of postfusion trimeric ectodomains at the liposome surface (Fig 5C), consistent with the overall shape, orientation, and lateral packing of postfusion trimers inserted into liposomes observed earlier for viral class II fusion proteins [19, 23, 24]. The size and shape of the projections observed on the HAP2e proteoliposomes are also compatible with a trimeric postfusion form and not with the monomers that were used to initiate the experiment (Fig 5B–5D), indicating that trimerization takes place upon interaction with lipids. Taken together, these results confirm that, as observed for CrHAP2 and a number of viral class II fusion proteins, the interaction with membranes leads to trimerization and that the trimers interact with the membrane via the variable, nonpolar surface of the domain II tip. A similar trimerization process occurred in the crystallization drops in the case of AtHAP2 and CrHAP2, which were also set to crystallize as monomers. Trimerization in the crystallization drops was also described for the flavivirus class II protein, albeit at acidic pH [25]. In the HAP2 case, the trigger for trimerization—which is irreversible for class II fusion proteins [26]—is likely to have been the very high protein concentration used in the crystallization trials. The physiological trigger of HAP2 trimerization to induce gamete fusion in vivo remains to be understood.
Flowering plant sperm are nonmotile and are delivered to female gametes in the cytoplasm of a pollen tube ([27], Fig 6A). Rupture of the pollen tube releases a pair of isogenic sperm cells; one fuses with the egg to produce a zygote, the other with the central cell to initiate endosperm development (Fig 6B and 6C). These two gamete fusion events are the defining feature of the flowering plants and are essential for the production of grain crops. To assess the role of αF in flowering plant gamete fusion, we utilized a genetic transmission assay (Fig 6D), which allowed us to determine whether HAP2 variants were able to restore gamete fusion to Arabidopsis hap2-2 null mutant sperm. In crosses between WT (HAP2/HAP2) females and hap2-2/HAP2 heterozygous males, the hap2-2 allele is not inherited by progeny (Fig 6D, [28, 29]). However, if a transgene carrying fully functional HAP2 is introduced into hap2-2/HAP2 plants, 33% of progeny will inherit hap2-2 (Fig 6D, AtHAP2 WT). This assay provided a quantitative readout for HAP2 function in vivo and facilitated dissection of the critical residues of the domain II tip (Fig 6D, S4 Table). We first tested whether AtHAP2 αF is essential for function by deleting the helix along with two N- and C-terminal residues (box in Fig 6E, AtΔαF). Deletion of αF rendered HAP2 nonfunctional; hap2-2 transmission rates were zero (Fig 6F, S2 Data, AtΔαF) even though expression of HAP2ΔαF:yellow fluorescent protein (YFP) in sperm was similar to the WT control (Fig 6G, S4 Fig).
We predicted that the nonpolar residues (I171, F172, M175, I176) on the nonpolar face of αF (Fig 4A–4C) would be critical for insertion into female gamete plasma membranes. To address this hypothesis, we tested whether these residues were important for association of AtHAP2e with liposomes (Fig 5A). When we mutated F172 to alanine, membrane association was decreased, and when I171, F172, and I176 were mutated to alanine (IFI>AAA triple mutant), membrane association was almost completely abrogated; the majority of AtHAP2e was found at the bottom of the sucrose gradient (Fig 5A) and was not associated with liposomes (Fig 5E and 5F). In our genetic analysis in vivo, mutating either I171 or I176 to alanine led to mild reductions in function (Fig 6F), while F172A more strongly reduced HAP2-induced gamete fusion (Fig 6F). The I171F172I176>AAA triple mutant completely eliminated HAP2 function in vivo (Fig 6F), in line with the abrogated association with liposomes observed in vitro (Fig 5A). We conclude that these hydrophobic αF residues, which are not individually essential for function, provide a hydrophobic surface at the domain II tip required for insertion into the egg and central cell plasma membrane to initiate HAP2-driven gamete fusion.
It was proposed that the salt bridge between the invariant R185 and E126 in CrHAP2 (Figs 1B, 2B and 2C) constrains the domain II tip and helps present the predicted fusion loops to the target membrane [1]. However, Tetrahymena thermophila (Tt) mating was not eliminated when the corresponding arginine was mutated [5], raising the question of whether the proposed function is maintained across species. We found that A. thaliana gamete fusion was completely abolished upon mutation of the corresponding arginine residue to alanine (R163A, Fig 6F). The aa sequence of TtHAP2 shows that there is a lysine two residues downstream from the conserved arginine (Fig 1B, 21st line, Ciliophora in the “Chromista” block). It is possible that given the plasticity of this region, a salt bridge between this lysine and the conserved glutamate in β-strand b can rescue the fusion activity, but further experiments with TtHAP2 would be required to test this option.
To determine whether the helical nature of αF is critical for AtHAP2 function, we mutated D173—which is in the middle of the hydrophilic face of αF—to proline, a mutation predicted to introduce a kink in the helix. We found that D173P, but not D173A (which is expected to maintain helical character), reduced HAP2 function in vivo (Fig 6F) and reduced the ability of AtHAP2e to associate with liposomes (Fig 5A), indicating that altering the helical conformation in this region affects membrane insertion. Both the length and amphipathicity of αF appear to be conserved among flowering plants (Fig 1B). In addition, Lysine 179 is invariant among plant HAP2 sequences (Fig 1B) and is situated just below αF (Fig 4A and 4B). We found that mutating K179 to alanine strongly reduced the function of HAP2 in the gamete fusion assay (Fig 6F) but did not affect the ability of HAP2e to insert into liposomes in vitro (Fig 5A). This result suggests that K179 is not required for membrane insertion but could be critical for a different stage of the gamete fusion reaction in plants.
To further address the sequence requirements for αF, we tested whether AtαF could be functionally replaced with the corresponding amphipathic helix of rice (Oryza sativa [Os], S2 Fig), a distantly related flowering plant species. This replacement resulted in a chimeric HAP2 variant that was functional (OsαF, Fig 6E and 6F). OsαF maintains the amphipathic nature of the helix and shares an isoleucine at position 171 but substitutes a tryptophan at position 173 and a threonine at position 176 (S2 Fig). The interchangeability of αF suggests plasticity in the mode of interaction between sperm-expressed HAP2 and its target membranes, at least within the flowering plant lineage and in spite of around 150 million years of divergence [30]. In contrast, when the entire rice ectodomain was used to replace the Arabidopsis ectodomain, this chimeric protein was unable to restore function to hap2 mutant sperm [29], indicating that other aspects of the gamete fusion mechanism (e.g., regulation of HAP2 activity by specific interaction with additional, lineage-specific proteins) have diverged within flowering plants.
In this study, we provide evidence at the primary sequence (Fig 1) and at the structural (Fig 2) levels that HAP2, while maintaining its overall structure (Fig 3), has evolved highly divergent membrane interaction motifs by way of focal diversification across the eukaryotes in which it was positively identified (S5 Fig). Analysis of the HAP2 gene structure in intron-rich genomes (e.g., flowering plants, algae, protozoa, cnidarians, and insects) revealed the consistent presence of an intron between the regions encoding the domain II β-strands b, c, and d (S6 Fig), providing a potential route for focused diversification of the intervening functional loops via alteration of splicing patterns and/or exon shuffling during evolution. The X-ray structures of HAP2 from two representative organisms exhibiting a contrasting pattern of insertions and deletions in the loops at the domain II tip suggest very different modes of membrane insertion. In Arabidopsis, this motif consists of a single loop that projects the amphipathic αF helix for insertion into the lipid bilayer (Fig 4), while in T. cruzi, the membrane interaction surface comprises three small loops (Figs 2 and S3). Although the domain II tip of TcHAP2 lacks an amphipathic helix and in this respect appears similar to viral class II proteins, such as that of RVFV Gc, the elements required for binding a glycerophospholipid head group in the latter (Figs 2B and 3D, [13]) are absent, suggesting that TcHAP2 uses yet another way of interacting with the lipid head groups. Moreover, the amphipathic helix αF observed in the flowering plants (Fig 4) is unlike the membrane interaction surface described for any of the three structural classes of viral fusion proteins (Fig 2B, Fig 3). It will be interesting to determine the forces that have driven diversification of the HAP2 fusion loops. One possibility that needs to be explored is that female gametes have evolved specific plasma membrane compositions important for gamete fusion. We found that AtHAP2 inserted more efficiently into artificial liposomes containing DOPS (see Materials and methods), a phosphatidylserine mimic. Understanding of female gamete lipid composition is currently limited to bulk membrane analysis in species with large and easily accessible oocytes [31, 32]. However, the use of genetically encoded phospholipid sensors is leading to increased awareness of important functions for even low-abundance anionic phospholipids (e.g., phosphatidylserine) and can now be used to define the membrane composition of small and inaccessible female gametes like those of the flowering plants [33].
We have shown that Tc and AtHAP2e insert into artificial membranes (Fig 5, S3 Fig) and have identified key nonpolar residues on αF that are essential for AtHAP2e membrane insertion in vitro (Fig 5A) and for gamete fusion in vivo (Fig 6). These data are consistent with previous genetic and biochemical analyses in Chlamydomonas [1] and with the recent finding that antibodies against the predicted fusion loops of Plasmodium HAP2 block parasitic gamete fusion and transmission [34]. The hypothesis that HAP2 functions by direct membrane insertion is also supported by experiments in which a synthetic peptide corresponding to the predicted fusion loop of TtHAP2, associated with membranes [5].
Class II viral fusion proteins are activated by exposure to acidic pH upon entry into the host endosome and are positively and negatively regulated by partner viral proteins [26, 35]. Whether and how HAP2 is triggered and the nature of the potential partner proteins involved in negative or positive regulation of HAP2 function will be active areas of future research; it will be interesting to determine whether these mechanisms are lineage specific or broadly conserved. For example, in Arabidopsis, EGG CELL 1 (EC1) has been shown to activate HAP2 for fertilization only after the sperm cells are released from the pollen tube to female gametes [36].
Our results show a very strong correlation between the ability of AtHAP2 to insert into membranes in vitro and its functionality in vivo, indicating that membrane insertion is an essential step in the fusion process, as demonstrated for the viral fusion proteins. Nevertheless, a recent report [37] proposed a HAP2 fusion mechanism similar to that of the C. elegans somatic fusion factor proteins (epithelial fusion failure 1 [EFF-1] and anchor-cell fusion failure 1 [AFF-1]; [38]), which do not function via target membrane insertion. These are the only members of the class II fusion protein structural family that lack a target membrane insertion surface at the domain II tip and were proposed to function instead by trans-oligomerization of proteins resident in the membranes of adjacent cells destined to fuse. Our findings do not support such a mechanism for HAP2, in line with the observation that in Arabidopsis, HAP2 is essential for sperm fertility but is not required for female fertility [28, 39]. A virus-like fusion mechanism is also in agreement with the observation that HAP2 is expressed in only one of the two gamete types to be fused in multiple plant, protozoan, and animal species in which it was studied [2, 3, 28, 39–41]. Indeed, expression from only one gamete is sufficient for gamete fusion in all species tested thus far [2, 28, 39, 40, 42], making the requirement for a second HAP2-like membrane fusion protein in the opposite gamete for fusion unlikely.
HAP2 is the only gamete plasma membrane fusion protein to be identified thus far [43]. aa sequence analyses have identified a HAP2-specific motif (pfam10699, [44]) that has detected orthologs in four out of the five eukaryotic kingdoms ([1, 2, 4], Figs 1 and S5), suggesting that this protein was present in the last common ancestor to all eukaryotes and was a seminal innovation in the evolution of sexual reproduction. But these analyses have not identified orthologs in some well-studied clades like nematodes, vertebrates, and fungi (S5 Fig). While we cannot exclude the possibility that HAP2 was replaced in some lineages by a fusion protein of a different origin, its widespread but sporadic identification in eukaryotic genomes (S5 Fig) suggests the more likely scenario that for many organisms, its sequence has diverged enough to escape detection by traditional sequence-based searches. A possible evolutionary force that may have driven HAP2 divergence in some lineages is positive selection for sequence diversity to maintain barriers between species. Proteins mediating cellular interactions critical for fertilization are well known to diversify rapidly and reinforce interspecific fertility barriers [45]. Such interactions between HAP2 and partner proteins that regulate its fusion activity may have driven further divergence of the HAP2 sequence in organisms like yeast, C. elegans, mice, or humans, which currently lack candidate gamete fusion proteins.
The two HAP2 structures provided here (Fig 2A and Fig 3A), together with the previously reported CrHAP2e [1], could be useful in defining a structural signature to identify additional orthologs. X-ray structures were required to identify viral class II fusion proteins because they lack any detectable sequence similarity across viral genera [26, 35]. A comparison of the structural conservation of the HAP2 domain II structure across eukaryotes (Fig 3A) with that of the class II fusion proteins from different viral families (Fig 3B) shows that diverse HAP2 molecules conserve the relative orientation and length of most of its secondary structure elements, while the viral class II proteins show higher variation. Indeed, the HAP2 structures share an α2 helix of identical length and orientation, the same organization of the ij loop, and the core bdc β-sheet, including an invariant salt bridge anchoring the variable fusion loops to the central core of the molecule. In comparison, the viral class II proteins (Fig 3B), in spite of sharing the same elements within domain II (α2 helix, ij hairpin, core bdc β-sheet), differ in their relative orientations and positioning and are thus more structurally diverse. This is clearly seen in the table of DALI scores (S3 Table), which reveals a large gap in between the very similar HAP2 domain II structures and the more divergent domain II structures from the other known class II proteins. This gap suggests the possibility that HAP2 orthologs with much lower sequence identity may be found that display the same three-dimensional fold. HAP2 orthologs identified thus far share approximately 30% sequence identity between the aligned residues across domain II, whereas viral fusion proteins are only about 10% identical in the aligned residues. It is thus plausible that additional HAP2 orthologs with lower sequence identity exist at a level insufficient for sequence-based identification. Irrespective of whether class II fusion proteins have a viral or cellular origin [46], the fact that the viral evolutionary clock is several orders of magnitude faster than its cellular counterpart suggests that HAP2 may have maintained an ancestral organization relative to the class II proteins of present-day viruses. The high conservation of the HAP2 core structure further suggests that bioinformatics approaches should be able to translate the observed structural similarity into signatures detectable in more distant HAP2 sequences, for instance, by analyzing the covariance of interacting residues distant in the sequence.
Seeds were stratified (at least 2 days at 4°C) on solid Murashige and Skoog (MS) medium (Sigma Aldrich, St. Louis, MO) supplemented with appropriate antibiotics and germinated at 22°C under constant light (Percival incubator). After 7–14 days, seedlings were transferred to sterile #2MIX potting media (www.fafard.com) with fertilizer (N:P:K, 15:5:15) and were grown at 20°C, 50%–60% humidity, on a 16-hour-light / 8-hour-dark light cycle in growth chambers (Environmental growth chambers, Chagrin Falls, OH, United States of America).
Drosophila S2 cells (ATCC CRL-1963) were cultured in Schneider’s complete media (Thermo Scientific) before transfection and in Insect Xpress media after transfection (Lonza, Basel, Switzerland). Culturing and transfection of S2 cells has been described previously [47].
Codon-optimized synthetic cDNA corresponding to a soluble C-terminally truncated version of the HAP2e comprising residues 25–494 from A. thaliana and 26–516 from T. cruzi were cloned into a modified Drosophila S2 expression vector described previously [48], and transfection was performed as reported earlier [47]. For large-scale productions, cells were induced with 4 μM CdCl2 at a density of approximately 7 × 106 cells/ml for 8 days and pelleted, and the soluble ectodomains were purified by affinity chromatography from the supernatant using a StrepTactin Superflow column followed by SEC using a Superdex200 column in 10 mM TRIS pH8 100 mM NaCl. Pure proteins were concentrated to approximately 8 and 4 mg/ml, respectively.
Purified HAP2e were subjected to SEC using a Superdex 200 column (GE HealthCare) equilibrated with 10 mM TRIS pH8 100 mM NaCl. Separation was performed at 20°C with a flow rate of 0.5 ml min−1. Online MALS detection was performed with a DAWN-HELEOS II detector (Wyatt Technology, Santa Barbara, CA, USA) using a laser emitting at 690 nm. Online differential refractive index measurement was performed with an Optilab T-rEX detector (Wyatt Technology). Data were analyzed, and weight-averaged molecular masses (Mw) and mass distributions (polydispersity) for each sample were calculated using the ASTRA software (Wyatt Technology).
DOPS, 1,2-dioleoyl-sn-glycero-3-phosphoethanolamine (DOPE), 1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC), cholesterol, and sphingomyelin were purchased from Avanti Polar Lipids. Liposomes were freshly prepared in PBS by the freeze-thaw and extrusion method using molar ratios of 17% DOPE, 16% DOPC, 50% cholesterol, 17% sphingomyelin. Alternatively, for coflotation assays on WT and mutant AtHAP2e proteins, the following alternative composition was used: 50% cholesterol, 30% DOPC, 20% DOPS. Then, 0.07 μM purified HAP2e was mixed with 8 mM liposomes and incubated for 1 hour at 21°C in 100 μL PBS. Samples were then adjusted to a final concentration of 20% sucrose, overlaid with a 5%–60% sucrose gradient (in PBS), and centrifuged for 1 hour at 4°C at approximately 150,000 × g. Fractions from the top, middle, and bottom of the gradient were analyzed by immunoblotting using a monoclonal anti-strep tag antibody, and the bands quantified using the GeneTools Syngene software. The percentage of HAP2e in either fraction was calculated as the ratio between HAP2e in individual fractions and total HAP2e (sum of HAP2e in top and bottom fractions).
Purified HAP2e (A. thaliana or T. cruzi) mixed with liposomes was spotted on glow-discharged carbon grids (CF300, EMS, USA), negatively stained with 2% phosphotungstic acid (PTA) PH 7.4, analyzed with a Tecnai G2 Bio-Twin electron microscope (FEI, USA), and imaged with an Eagle camera (FEI, USA). For cryo-electron microscopy, liposomes mixed with purified HAP2e were applied on a glow-discharged Lacey Carbon grid (Agar Scientific, UK). Samples were plunge-frozen in liquid ethane using an automated system (Leica EMGP, Austria) and visualized on a Tecnai F20 electron microscope operating at a voltage of 200 kV. Image frames were recorded in low-dose mode on a Falcon II direct electron detector (FEI, USA).
Crystals of HAP2e from A. thaliana were grown at 293 K using the hanging-drop vapor-diffusion method in drops containing 1 μL protein solution mixed with 1 μL reservoir solution containing 100 mM Sodium Citrate pH 4.5 and 200 mM zinc acetate. Diffraction-quality rod-like crystals appeared after 1 week and were flash-frozen in mother liquor containing 30% (v/v) glycerol. The crystals diffracted anisotropically to 2.24 Å along the c* axis but only to 3.7 Å in orthogonal directions. Therefore, an ellipsoidal cut off was applied using the StarAniso server (STARANISO version 1.7.2 18-Apr-17 Ian J. Tickle, Global Phasing, Cambridge, UK; http://staraniso.globalphasing.org/cgi-bin/staraniso.cgi) in order to remove noise, and refinement was carried out to a nominal resolution of 2.75 Å.
Limited proteolysis of HAP2e from T. cruzi was carried out by adding subtilisin (dissolved in 10 mM Tris pH8, 30 mM NaCl at 10 mg/mL) to a solution containing T. cruzi HAP2e at 16 mg/mL in 10 mM Tris pH8, 100 mM NaCl at a 1:300 w:w ratio. After 1 hour of incubation at room temperature, the protease was inactivated by addition of 1 mM PMSF, and the protease-resistant fragment accounting to approximately 60% of the digested protein was purified by SEC using a Superdex200 column (TcHAP2esub). Crystals of TcHAP2esub were grown at 293 K using the hanging-drop vapor-diffusion method in drops containing 1 μL protein solution mixed with 1 μL reservoir solution containing 100 mM CHES pH 9.0, 200 mM NaCl, and 10% w/v PEG 8k. Diffraction-quality rodlike crystals appeared after 7–10 days and were flash-frozen in mother liquor containing 25% (v/v) glycerol.
Data collection was carried out at the ESRF (ID30A-3) and the Synchrotron Soleil (Proxima-1). Data were processed, scaled, and reduced with XDS [49], Pointless [50], and programs from the CCP4 suite [51]. For the structure of A. thaliana HAP2e, the full C. reinhardtii HAP2 monomer (PDB 5MF1) was used to create a search model with Sculptor [52]. The so-called tip region was deleted (bc and cd connections). The structure was determined by the molecular replacement method using Phaser [53]. Zinc atoms were localized in the density by calculating an anomalous difference map using ANODE [54]. To determine the structure of the protease-resistant fragment of T. cruzi HAP2, we initially used Sculptor [52] to create a search model for molecular replacement based on the structure of a protomer of the C. reinhardtii HAP2 trimer (PDB 5MF1). The full monomer model was divided into 3 individual domains (domains I, II, and III), and loops and exposed side chains were trimmed off. The structure was finally determined by the molecular replacement method using Phaser [53] and an isolated domain II as search model. Phases were refined using the anomalous signal of a highly redundant Sulfur-SAD data set collected at a wavelength of 2.06641 A on crystals of the native protein.
For both orthologs, model building was performed using Coot [55], and refinement was done using AutoBuster [56].
Two independent TcHAP2 trimer models were generated from superposition on either the AtHAP2 or CrHAP2 using the secondary structure matching function in Coot [55], and the resulting two TcHAP2 trimers were structurally very similar, with rmsd = 2.6 Å for 711 common Cα atoms (237 per subunit).
Arabidopsis HAP2 coding sequence variants were generated using a transfer-DNA plasmid containing the native HAP2 promoter and coding sequence with a C-terminal YFP fusion (pHAP2:HAP2cds:YFP; PGL290; CDS, GenBank AAY51999.1; pGreen [57] backbone; Basta-resistant seedlings). Mutations were made using the NEB Q5 Site-Directed Mutagenesis Kit (NEB #E0554S); primer sequences are provided in S5 Table. Mutations were confirmed by sequencing (Genewiz, South Plainfield, NJ). Agrobacteria tumefaciens–mediated transformation (strain GV3101 [58] with pSoup helper plasmid [57]) of hap2-2/HAP2 (SALK_152706; [28, 59]) was performed by floral dip [60]. hap2-2/HAP2 seedlings were selected on MS plates supplemented with sucrose (5 g/L) and kanamycin sulfate (50 mg/L; Life Technologies/Thermo Fisher Scientific, Waltham, MA).
After seeds were collected from transformed plants, they were sterilized using a 50% bleach solution containing 0.02% Triton X-100 for 7 minutes, followed by washing 4 times with sterilized water. Seeds were resuspended in sterilized 0.1% agarose and were plated on 15-cm dishes containing MS media supplemented with Basta (25 mg/L; Chem Service, West Chester, PA; Oakwood Chemical, Estill, SC). After at least 2 days at 4°C, plates were moved to a 22°C incubator with constant light. After 7–14 days, Basta-resistant seedlings were transferred to soil made up with 1X fertilizer (4 plants / pot). For primary transformants (T1), DNA was isolated using the leaf boil method [61]. T1 plants were genotyped for the hap2-2 allele using primer sets to detect both the Salk tDNA insertion (LbaI, [59]; hap2seqTR3 [28]) and WT genomic HAP2 (hap2c2; hap2seqTR3; see S5 Table).
Pollen from T1 plants that were heterozygous for hap2-2/+ were hand-pollinated onto male sterile 1 (ms1-1) pistils [62]. Seeds from individual ms1 crosses were gas-sterilized and plated on MS Kanamycin (50 mg/L) plates supplemented with sucrose (5 g/L). Seeds were cold-treated at 4°C for at least 2 days and then transferred to a 22-°C incubator with constant light. hap2-2/+ seeds were simultaneously plated on MS Kanamycin plates to ensure the drug resistance was working correctly. For analysis of hap2-2 transmission, resistant versus sensitive seedlings were counted manually after 6–10 days at 22°C.
Mutations were confirmed in at least one T1 line per transgene by sequencing of a PCR product. PCR products were generated using a EcoRI-HAP2-CVFP-F1 forward primer and the HAP2Ex7R reverse primer. For HAP2 variants that had 0% transmission of hap2-2, seeds were collected from the T1 plant and plated on MS Kanamycin and MS Basta separately to confirm segregation of the hap2-2 allele and the transgene, respectively, in the next generation.
Expression of HAP2:YFP in pHAP2:HAP2variant:YFP transgenic plants was initially screened using fluorescence microscopy. Pollen from T1 plants was hydrated on slides in pollen growth media [63] and imaged using a Zeiss Axiovert 200M microscope (images not shown). Each transgenic line was analyzed using confocal microscopy (Fig 5G, S4 Fig) of the T1 and/or T2 generation. Pollen were hydrated in pollen growth media and imaged using a Zeiss LSM 800 confocal microscope. All images were taken using excitation with the 488 laser at 10% (488 settings kept constant) and using the 40× water objective with a 2× zoom. Single optical slices were collected (974 × 974 pixels, 16-bit). All images were imported to ImageJ (FIJI version) [64] and cropped to 300 × 300 pixels.
Data are presented as mean ± SD unless otherwise indicated in figure legends, and experimental repeats are indicated in figure legends.
The following criteria were used to analyze HAP2:YFP variant transgenic plants: at least three independent transgenic lines were analyzed, at least three crosses were performed for each line, and only crosses with at least 10 seeds were analyzed (S4 Table). Transmission data did not follow a normal distribution for each transgene (confirmed by Anderson-Darling test for normality in R studio [65], using R version 3.2.3 [66]); therefore, a nonparametric statistical test was used. Each HAP2 mutant was compared to the WT transgene (AtHAP2) using a Wilcoxon rank sum test with continuity correction in R studio.
The atomic coordinates and structure factors for two structures have been deposited in the PDB under the accession numbers 5OW3 (AtHAP2) and 5OW4 (TcHAP2).
|
10.1371/journal.ppat.1006707 | Structural basis of glycan specificity of P[19] VP8*: Implications for rotavirus zoonosis and evolution | Recognition of specific cell surface glycans, mediated by the VP8* domain of the spike protein VP4, is the essential first step in rotavirus (RV) infection. Due to lack of direct structural information of virus-ligand interactions, the molecular basis of ligand-controlled host ranges of the major human RVs (P[8] and P[4]) in P[II] genogroup remains unknown. Here, through characterization of a minor P[II] RV (P[19]) that can infect both animals (pigs) and humans, we made an important advance to fill this knowledge gap by solving the crystal structures of the P[19] VP8* in complex with its ligands. Our data showed that P[19] RVs use a novel binding site that differs from the known ones of other genotypes/genogroups. This binding site is capable of interacting with two types of glycans, the mucin core and type 1 histo-blood group antigens (HBGAs) with a common GlcNAc as the central binding saccharide. The binding site is apparently shared by other P[II] RVs and possibly two genotypes (P[10] and P[12]) in P[I] as shown by their highly conserved GlcNAc-interacting residues. These data provide strong evidence of evolutionary connections among these human and animal RVs, pointing to a common ancestor in P[I] with a possible animal host origin. While the binding properties to GlcNAc-containing saccharides are maintained, changes in binding to additional residues, such as those in the polymorphic type 1 HBGAs may occur in the course of RV evolution, explaining the complex P[II] genogroup that mainly causes diseases in humans but also in some animals.
| Rotaviruses (RVs) are diverse, infecting humans and/or animals. Significant advances in understanding ligand-associated RV host ranges have been made but how such host ligands drive RV evolution leading to the diverse genotypes/genogroups already identified remains unclear. In this study, through solving crystal structures of P[19] VP8*-ligand complexes with two different ligands, we demonstrated how genetic variations could configure a totally new ligand binding site leading to a distinct new evolutionary lineage. Sequence comparison also identified further changes of the binding site which may occur over the course of RV evolution leading to different P[II] genotypes infecting different populations, including some animal species widely seen everywhere around the world. The elucidation of the genetic and evolutionary relationships among all members of the P[II] lineage including the two genotypes in P[I] is highly significant for advancing our understanding of RV host ranges, disease burden and zoonosis of human diseases.
| Rotaviruses (RVs) are a major cause of severe gastroenteritis in children under the age of 5, causing 200,000 deaths[1–3]and account for 2 million childhood hospital admissions with an estimated cost of over 1 billion US dollars per year[4, 5]. It has been shown that RV attachment to cell surface carbohydrates, mediated by the VP8* domain of the spike protein VP4, is a required first step of an effective infection [6–9]. RVs are genetically diverse. Based on the VP4 sequences, the group A RVs have been grouped into 40 genotypes (P[1]-P[40])[10, 11]. In a study based on the VP8* sequence, the different RV genotypes have been grouped into five genogroups (P[I]-P[V])[12]. Different genotypes/genogroups cause diseases in different populations and/or various animal species and each genotype and genogroup may have distinct glycan binding specificities responsible for their host ranges or tropism.
The previously identified “sialidase-sensitive” animal genotypes in the P[I] genogroup are the first examples exhibiting such genotype specificities that require interaction with terminal sialoglycans for cell entry [13–15]. However, most other genotypes in the five genogroups are found to be sialidase-insensitive and many of them have been found to recognize HBGAs [12, 16–19]. Human HBGAs are highly polymorphic containing the ABO, secretor (H) and Lewis families with wide distributions in the world populations and therefore may significantly affect RV epidemiology and disease burden. For example, the P[III] RVs that infect humans and many animal species have been found to recognize the type A HBGAs that are shared between humans and many animal species [12]. The P[11] genotype in P[IV] has been found to recognize the type 2 HBGAs precursors, which may be responsible for their age-specific host ranges in neonates and young infants due to the stepwise synthesis of HBGAs that is developmentally regulated [18, 19]. The P[11] RVs also commonly infect bovines which is believed to be due the occurrence of the specific HBGA precursors in these animals [20].
Significant advances of the molecular basis of ligand-controlled host ranges of RVs have been made following crystallographic studies of the RV VP8*s in complex with their genotype-specific ligands, including the sialic acid dependent animal RVs P[3] and P[7], the A antigen binding P[14] that infect both humans and animals, and P[11] that recognizes the type 2 precursor and commonly infects neonates, young infants and some animals[16, 20–22]. Interestingly, the VP8*s from all these animal and human RVs adopt a similar galectin-like fold, and interact with a specific glycan in the cleft region although the P[11] cleft is wider than in the other three RVs[20]. In addition, while the glycan binding site of P[11] VP8* has moved to span almost the entire length of the cleft, the P[14] and P[3]/P[7] RVs shared a common binding site located in one corner of the cleft region. These data are valuable for our understanding of RV host ranges, RV evolution and particularly zoonosis as many RVs infect both humans and animals.
Despite recent progress in elucidating VP8*/receptor interactions, the molecular basis of host ranges of the major human RVs (P[4], P[6] and P[8]) in the P[II] genogroup that are responsible for over 90% of human infections remains unknown due to lack of a conclusive data detailing the precise interactions of VP8* domains with host ligands as receptors for these human RVs[17, 23, 24]. Crystal structures of the native VP8* of the P[4] and P[8] RVs also showed a galectin-like fold with a similar wider cleft between the two β-shifts as that of P[11] RVs, which has led to a deduction of a similar cleft-glycan-binding site for P[4] and P[8] RVs[16, 20, 22, 25]. However, in our recent study of another P[II] RV (P[19]) that is genetically closely related with the other P[II] RVs, we found that P[19] VP8*s have a unique property of binding two types of glycans, the mucin core and type 1 histo-blood group antigens (HBGAs) and may use a binding site different from those described above[26], indicating that the P[II] RVs may be evolutionarily distantly related with other genogroups. To seek direct evidence on the possible shifted ligand binding site of P[19] VP8* and to explore the molecular basis how could a single binding site accommodate two structurally related but distinct glycan ligands, in this current study we resolved the crystal structures of the P[19] VP8* in complex with the two glycans. The structural data showed that the P[19] VP8* uses a completely new binding site that is different from the one of other RVs and this single binding site is able to interact with either type I HBGA pentasaccharide Lacto-N-fucopentaose I (LNFP I) or mucin core 2 glycan with a similar binding mode. Structure based-sequence comparisons also confirmed the conserved binding site of P[19] with other P[II] RVs. Unlike the other three P[II] RVs that mainly infect humans, the P[19] RVs are rarely found in humans but commonly infect animals (porcine). Thus, our study helps establish the genetic and evolutionary relationships among these human and animal RVs, which advanced our understanding of RV host ranges, disease burden, epidemiology, and zoonosis of human diseases.
Previous glycan array studies have shown that the P[19] VP8* recognizes the mucin core glycans with a key GlcNAcβ1-6GalNAc motif and the type 1 HBGA precursor with inclusion of the internal Gal (Fucα1-2Galβ1-3GlcNAcβ1-3Gal)[26]. Prior to crystallographic studies, the binding specificity of P[19] VP8*was validated by ELISA using recombinant GST-tagged VP8* with the type 1 HBGA penta-saccharide LNFP I and mucin core 2 glycan (Fig 1). The binding of P[14] VP8* to A-type HBGAs was included as a positive control. Our results showed that the binding signals of P[19] VP8*exhibited typical dose-responses to both LNFP I and mucin core 2 glycans. As expected in the controls, the VP8* of a P[14] RV (P[III]) only recognized the A antigen but neither the LNFP I nor mucin core 2.
The crystal structures of P[19] VP8* in complex with LNFP I and mucin core 2 were solved at 1.94 Å and 1.90 Å, respectively. The electron density was clear for both bound ligands, which allowed unambiguous assignment of the two ligands (S1 Fig). The secondary structural elements from N- to C- termini are designated as: βA (73–74), βB (80–85), βC (90–96), βD (102–108), βE (115–121), βF (124–130), βG (137–144), βH (152–159), βI (163–169), βJ (172–177), βK (184–189), βL (197–200), βM (204–208), and αA (212–221) (S2 Fig). In comparison with the previously solved P[19] VP8* native structure [27], ligand binding did not cause any significant conformational changes, with the root mean squared deviation (RMSD) for alpha carbons of the backbone atoms between the bound P[19] VP8* and the free VP8* being 0.531 Å (mucin core 2) and 0.505 Å (LNFP I), respectively. Interestingly, a structural rearrangement was noted among residues 87–90 in the B-C loop after P[19] VP8* bound to either of the glycans (S3 Fig), with the RMSD for alpha carbons of these residues between bound and free P[19] VP8* being 1.070 Å (mucin core 2)and 0.992 Å (LNFP I), respectively.
Similar to other known VP8* structures, P[19]VP8*s adopted a classical galectin-like fold with two twisted antiparallel β-sheets consisting of strands A, L, C, D, G, H and M, B, I, J, K, respectively (S2 Fig). The common shallow cleft between the two β-sheets where the glycan-binding sites of other known RVs are located (Fig 2a–2d) is wide in P[19] (Fig 2a and 2e) similar to the P[11] and P[8]/P[4] VP8*s but wider than the P[3]/P[7] and P[14] VP8*s. It was noted that the P[19] glycan binding site is located away from the cleft and thus represents a completely new glycan binding site, consistent with our previous observations based on NMR and mutagenesis studies [26]. This new glycan-binding site is composed of the carboxyl-terminal α-helix and the β sheet that composed of the βB, βI, βJ, and βK strands (Figs 2e and 3).The residues involved in the LNFP I interaction include W81, L167, H169, G170, G171, R172, W174, T184, T185, R209, E212, and T216, while those involved in mucin core 2 interactions are W81, L167, H169, G170, G171, R172, W174, T185, R209, and E212 (Fig 4).
LNFP I binds P[19] VP8* with a network of hydrogen-bonding interactions and hydrophobic interactions (Fig 3). All four residues in the type 1 HBGA chain backbone, Galβ1-3GlcNAcβ1-3Galβ1-4Glc participated in the interaction, with the motif Galβ1-3GlcNAcβ1-3Gal playing a central role. The fucose in the penta-saccharide LNFP I, referred as Fuc-I, lies almost 90 degree relative to the plane formed by the rest of LNFP I residues and thus projects out from the VP8* binding surface without making direct contacts with VP8*. The galactose next to Fuc-I, referred as Gal-II, interacts with residues T184 and T216 via hydrophobic interactions, which is further stabilized by forming hydrogen-bonds with T185 and two water molecules. The GlcNAc at the third position (GlcNAc-III) inserts into the deep binding pocket formed by W81, L167, W174, T185, R209, and E212, forming hydrophobic interactions between the LNFP I backbone and residues W81, L167, W174, and E212, and forming hydrogen bonds between the LNFP I and the side chains of residues T185, R209, and E212 (Fig 3b). Two other water molecules help stabilize GlcNAc-III through hydrogen bonds with its acetyl moiety. The fourth saccharide (Gal-IV) makes contacts with G170, G171, and W174 through hydrophobic interactions and forms hydrogen bonds with H169, G170 and two water molecules. The fifth saccharide (Glc-5) interacts with R172 through hydrophobic effects.
P[19] VP8* binds mucin core 2 using the same binding site (Fig 4) through an almost identical conformation as that of bind LNFP I, with a RMSD 0.118 Å for the alpha carbon backbone atoms. Two amino acids, T184 and T216, which are involved in binding to LNFP I, did not participate in binding to the shorter mucin core 2 trisaccharide (Fig 5). The GalNAc (GalNAc-I) forms hydrogen bonds with G171 and R172, while the Gal (Gal-II) makes hydrophobic interactions with G170 and G171. The amino acid residue of threonine that links to GalNAc-I pointed into the solvent. Interestingly, the GlcNAc (GlcNAc-III) inserted into the same deep binding pocket that binds the GlcNAc-III of LNFP I (Fig 3). It was noted that H169 contributed to the mucin core 2 binding interaction with GlcNAc-III, but this interaction was not seen with GlcNAc of LNFP I (Fig 5). The GlcNAc-III was further stabilized by two amino acids (T185, R209) through hydrogen bonding interactions. The superimposition of LNFP I and mucin core 2 in interaction with P[19] VP8* and the schematic interacting diagram was shown in Fig 5.
To examine the biological significance of reactive glycan ligands to P[19] RV function, viral replication inhibition assays were performed on a human P[19] RV. P[19] RV titers were significantly reduced following incubation of the viruses with combination of both LNFP I and mucin core 2 (Fig 6).
Structural-based sequence alignment of P[4], P[6], P[8] and P[19] among P[II] RVs showed significant amino acid conservation on the newly identified ligand binding interface (Fig 7). Residues W81, L167, W174, T184, T185, R209, and E212 are identical among all four P[II] genotypes based on representative sequences from each genotype, while there are slight difference at residues 169, 170, 171, 172, and 216. For example, the H169 in P[6] and P[19] is changed to Y169 in P[4] and P[8], and the R172 in P[4], P[8] and P[19] is changed to S172 in P[6]. As LNFP I and mucin core 2 share the same binding interface in P[19] VP8*, we use LNFP I as a representative to investigate whether P[4] and P[8] could accommodate the ligand in the newly found glycan-binding site as their native crystal structures are available.
The overall structures of the LNFP I-bound P[19] VP8* domain and the apo P[4] VP8* domain (PDB: 2AEN) were very similar based on the backbone alpha carbon RMSD of 0.569 Å(Fig 8a). Superimposition of the P[4] VP8* apo structure onto the P[19] VP8* structure in its complex with LNFP I showed that the LNFP I fit almost perfectly onto the surface of the apo P[4] VP8* defined by the same residues observed in the binding interface of LNFP 1 to the P[19] VP8* (Fig 8a, inset). The only notable differences were that the sidechains of Y169 and R209 in the P[4] VP8* tilted away from the LNFP 1 binding pocket, which may destabilize the interaction between LNFP I and P[4] VP8*. The LNFP I-bound P[19] VP8* also had a similar overall structure as the apo P[8] VP8* domain (PDB: 2DWR), with the RMSD of the alpha carbons of the backbone being equal to 0.493 Å (Fig 8b). Again, superimposition of the two structures indicated that the side chains of almost all of the residues involved in binding to LNFP I in the P[19] VP8* domain were also in position to interact with the LNFP 1 except that the side chain of R172 in the P[8] VP8* domain sterically clashed with the oxygen atom of the Gal-IV moiety of LNFP I in the superimposed structure, which likely is responsible for preventing P[8] VP8* from binding to LNFP I. These structural analyses suggested that P[4] and P[8] as well as P[6] (the crystal structure of P[6] remains unavailable) may all utilize the same ligand binding site identified in P[19], however, slight sequence and structural variations among the VP8* domains encoded within the different genotypes may be responsible for genotype-specific ligand-binding patterns and specificities of different genotypes.
Our crystallography studies clearly demonstrated that the P[19] RVs use a novel glycan binding site that is different compared to known binding sites of other RV genotypes and genogroups. This new binding site is capable of interacting with two structurally related but distinct glycans, the mucin core 2 and type 1 HBGA of LNFP I, using a common binding pocket and similar binding mode. Sequence comparisons showed high levels of amino acid conservation of the P[19] binding site with other P[II] genotypes (P[4], P[6] and P[8]) and two genotypes in P[I]. These data confirmed the new binding site of P[19] previously deduced from NMR and mutagenesis analyses[26], supporting the hypothesis that P[II] RVs are under strong selection of the host mucin core and/or type 1 HBGAs as common traits. However, given the complicated P[II] genogroup with multiple genotypes and variable host ranges among different human populations and/or some animal species, further defining the molecular basis of how these two structurally related host ligands drive RV evolution leading to such diverse P[II] genogroup is important.
Both mucin cores and HBGAs are O-linked glycans commonly seen in nature on the mucosal surfaces and cellular membranes of many mammals. These O-linked glycans are synthesized step-wisely by a group of glycosyltransferases, in which the GlcNAc-containing oligosaccharide motifs that are recognized by P[II] RVs serve as the starters or precursors (S5 Fig). For example, the type 1 HBGA precursor Galβ1-3GlcNAc can be extended to different A, B, H and Lewis HBGA products by adding one specific saccharide in each step. This process is developmentally regulated in the early lives of children [18, 20, 26], which may also occur in animals, leading to shared precursors, intermediates and/or full HBGA products in some animal species. Given the high sequence conservation constituting the GlcNAc-binding pocket among all P[II] RVs, we deduced that the P[II] genogroup may originate from an ancestor with a simple binding site recognizing these GlcNAc-containing motifs (GlcNAc-Gal) and circulating in one or a group of species that share such glycans. Such a binding site could further expand its receptor binding repertoire through genetic variations in the course of RV evolution by adapting to additional residues when encountering new hosts producing longer and more complicated HBGA products or along the course of RV-host co-evolution.
The above deduction assumes that, while the binding ability to the GlcNAc-containing moiety is maintained, extended interactions with additional saccharides could affect receptor ligand binding affinity and/or specificity, therefore potentially changing the binding outcomes. This assumption is supported by the observed variable binding patterns of the four P[II] RVs (P[4], P[6], P[8], and P[19]) to the tetra- (LNT), penta- (LNFP I) and hexa-saccharide (LNDFH I) of the type 1 HBGAs that either contain (LNDFH I) or do not contain (LNT and LNFP I) the Lewis epitopes[26]. The observed inability of P[4] and P[8] binding to the penta-saccharide LNFP I without the Lewis epitope is further supported by homology modeling of the native P[4] and P[8] VP8* structures in comparison with the LNFP I bound P[19] VP8* with an identification of binding clash for both P[4] and P[8] VP8* (Fig 8). In addition, the amino acids H169 and T216 of P[19] that are involved in interactions with the residue Gals next to GlcNAc have changed to Y169 and N216, which may also lead to different binding outcomes in the two genotypes. Finally, it was noticed that there is a slightly shifted orientation of the GlcNAc-containing motif inside the binding pocket between mucin core 2 and the type 1 HBGAs, leading to a significant orientation shift of the backbones between mucin core 2 and type 1 HBGA within the binding cleft (Fig 5a). This indicated a mechanism for how P[19] VP8* is able to achieve a broader binding specificity to the extended molecules of the two glycan types.
While the majority of P[I] RVs infect animals, most of the P[II] RVs infect humans. Thus, we deduced that the P[II] RV may come from P[I] with an animal host origin and were introduced to humans by adapting to the polymorphic human HBGAs, leading to different P[II] genotypes infecting different human populations and/or some animal species depending on their evolutionary stages. For example, the P[19] genotype may represent an early evolutionary stage after the ancestors of P[II] genogroup started adapting to human receptors but they may still retain the binding specificities to the backbones of the mucin core 2 and type 1 HBGAs. This may explain why the P[19] RVs are commonly found in animals (porcine) but rarely in humans. On the other hand, the P[4] and P[8] RVs are genetically closely related and both genotypes are well developed that recognize the much more matured HBGA product, the Lewis b (Leb) antigen that is widely distributed in humans, and together these two genotypes are responsible for ~90% of human RV infections worldwide. Furthermore, the P[6] RVs have been found to recognize the much less matured type 1 HBGA precursors, which is consistent with the fact that the P[6] RVs commonly infect neonates and young infants through recognizing the age-specific precursor glycans that commonly occur in the early lives of children [28, 29]. The P[6] RVs are also commonly found to infect porcine, likely through the type 1 HBGA precursor glycans that are share between humans and pigs. Thus, the elucidation of such genotype-specific host ranges controlled by the host HBGA makeups is important for understanding the disease burden and epidemiology therefore vaccine strategy against RVs based on the P type vaccine approach [30].
The findings of sequence conservation of P[10]/P[12] binding sites with that of the P[19] and similar glycan binding profiles between P[10] and P[19] [26] extends our understanding of P[II] evolution, in which these two P[I] RVs may represent even earlier ancestors than P[19] of the P[II] lineage. Both P[10] and P[12] are minor genotypes occasionally found in humans and bats and horses, respectively [31, 32], while the majority of other P[I] RVs were more commonly found to cause diseases in different animal species, indicating that the P[10]/P[12] RVs are unique in P[I] and should be evolutionarily grouped to the P[II] lineage. In fact, P[10]/P[12] were genetically closer to P[19], P[8]/P[4] and P[6] than the rest genotypes based on the full VP4 sequence phylogeny analyses [33, 34]. Thus, the P[10]/P[12] RVs are considered to be the earliest traceable ancestor of P[II] lineage. The reason of low abundance of these two genotypes in any species remains unknown.
In conclusion, P[II] RVs represent a unique evolutionary lineage starting from an ancestor in P[I] with a possible animal host origin. While the original binding specify to the mucin core and type 1 HBGA precursors is maintained, additional interactions with adjacent residues may have occurred during the evolution of the ancestor genotype as it adapted to human receptors. This led to the diversity of RV strains seen today, with some mainly infecting animals with others mainly infecting humans [26]. Since viruses in all group A RV genotypes and genogroups must be from a single ultimate common ancestor, the deduced evolutionary path of animal-to-human transition from P[I] to P[II], but not the other way around, may apply to other genotypes and genogroups. This deduction is important for RV classification and epidemiology, which may impact prevention and control strategies, such as vaccine design against RVs. For example, since the majority of the genotypes in P[I] exclusively infect animals, they may not be suitable for developing live human vaccine, because they may not be able to replicate in human guts due to the lack of proper receptor. This issue is urgent as the Jennerian approach is still widely used to develop live animal reassortant RV vaccines in many countries.
Our research still has certain limitations. For example, the deduced binding sites for the major human RVs P[4], P[6] and P[8] remain to be verified by co-crystallization studies of the VP8* in complex with their ligands. This task is challenging as our homology model data indicated that the free VP8* does not accept Lewis b, consistent with previous observations from several other groups [23, 24]. In addition, the precise saccharide sequences and structures recognized by the major P[II] RVs remain unknown. As the cleft where the P[19] binding site is located is long and fairly deep, it is likely to accommodate a long glycan extending to both sides of the GlcNAc-containing oligosaccharide motifs, and future studies to explore this issue by glycan arrays with more representative human glycan pools are necessary. Finally, it is noted that the new binding site of P[19] shifts toward the C terminus of VP8* and is located in the bottom of VP8*, which may need the support of VP5* to exhibit its true binding characteristics. This issue also needs to be studied.
The VP8* core fragments (amino acids 64 to 223) of the human RV P[19] with an N-terminal glutathione S-transferase (GST) tag was overexpressed in Escherichia coli BL21 (DE3) cells as previously described [12, 17]. Cells were grown in 1L Luria broth (LB) medium supplemented with 100 μg ml-1 ampicillin at 310 K. When the OD600 reached 0.8, 0.5 mM isopropyl-β-D-thiogalactopyranoside was added to the medium to induce protein expression. The cell pellet was harvested within 12 h after induction and re-suspended in the 30 ml phosphate-buffered saline (PBS) buffer (140 mM NaCl, 2.7 mM KCl, 10 mM Na2HPO4, 1.8 mM KH2PO4, pH 7.3). The cells were lysed by French press (Thermo Fisher Scientific, Waltham, MA), then the cell debris was removed by centrifugation at 12,000×g for 30 min. The supernatant of the bacterial lysate was loaded to a disposable column (Qiagen, Hilden, German) pre-packed with glutathione agarose (Thermo Fisher Scientific). After three washes with PBS buffer, the GST fusion protein of interest was eluted with elution buffer (10 mM reduced glutathione, 50 mM Tris-HCl, pH 8.0). The GST tag of the VP8* protein was removed using the thrombin (Thermo Fisher Scientific) after dialysis into the buffer (20 mM Tris-HCl, 50 mM NaCl, pH 8.0). The flow-through was collected after passing the mixture through glutathione agarose, and the purified protein was concentrated to about 10 mg ml-1 with an Amicon Ultra-10 (Millipore, Billerica, MA). The purity of the protein was judged using 15% SDS-PAGE stained with Coomassie Brilliant Blue.
The binding of purified VP8* to different glycans was confirmed by ELISA. Synthetic polyacrylamide polymer (PAA) conjugated oligomers were used to study their specificity as a ligand for P[19] RVs. Briefly, microtiter plates (Thermo Fisher Scientific) were coated with recombinant VP8* proteins (10 μg/ml) at 4°C overnight. After blocking with 5% nonfat cow milk, synthetic polyacrylamide polymer (PAA)-biotin conjugated oligomers (Glycotech, Gaithersburg, MD) were added at serial dilutions and incubated at 4°C overnight. The bound oligosaccharides were then detected using HRP-conjugated-streptavidin (Jackson Immuno Research Laboratories, West Grove, PA) and displayed using the TMB kit (Kierkegaard and Perry Laboratory, Gaithersburg, MD).
The hanging-drop vapor-diffusion method was used for crystallizing human RV P[19] VP8* protein and co-crystallizing VP8* complexed with LNFP I and mucin core 2. Crystals were obtained from drops where 1 μl purified P[19] VP8* was mixed with 1 μl of the reservoir buffer: 0.5 M ammonium sulfate, 0.1 M sodium citrate tribasic dihydrate pH 5.6, 1.0 M lithium sulfate monohydrate. The sugar LNFP I (Dextra, Reading, UK) and threonine linked mucin core 2 (kindly provided by James C. Paulson at the Scripps Research Institute) were prepared at 60 mM and 100 mM, respectively, in PBS with 10% glycerol and soaked into the crystals. Diffraction data were collected at the Advanced Photon Source (APS) beamline 31-ID-D, Argonne National Laboratory, in Chicago, Illinois. A total of 360 images were collected using 0.5° oscillation during 20 s exposures. The images were integrated with MOSFLM [35], and scaled with SCALA [36]. Molecular replacement was performed with PHASER [37] using the coordinates of chain A from 5GJ6 [27] as the search model. Iterative model building was manually carried out in COOT [38], and refinements using 5% of reflection in Free-R set were carried out in REFMAC [39] implemented in the CCP4 suite [40]. The structure quality was assessed using Mol Probity [41]. Final model and scaled reflection data were deposited at the Protein Data Bank (PDB ID 5VKS and 5VKI). Processing and refinement statistics for the final model are presented in Table 1. The visualization and investigation of the final model was analyzed using Chimera [42].
The P[19] RVs (strain 210) which was cell-culture adapted by multiple blind passages on the MA104 cells (passage 4–6) were then used for inhibition assays with the oligosaccharides LNFP I and mucin core 2 using procedures described previously[26]. The P[19] RVs at 300 fluorescent forming units (FFU)/10 μl) were pre-incubated with different inhibition reagents for 30 min. After rinsing twice with serum-free DMEM and chilling of all reagents and the 24-well plates on ice, duplicated wells of confluent MA104 monolayers were inoculated with the virus-oligosaccharide on ice with continuous rocker platform agitation for 1 h. The inoculum was then removed and the cells were washed twice with ice-cold serum-free DMEM. The plates were then placed back in the 37°C incubator for 18 to 20 h prior to quantification of infected cells by immunofluorescence with a rabbit anti-rotavirus antibody followed by a FITC-labeled goat anti-rabbit secondary antibody.
|
10.1371/journal.pbio.2005069 | The zinc transporter ZIPT-7.1 regulates sperm activation in nematodes | Sperm activation is a fascinating example of cell differentiation, in which immotile spermatids undergo a rapid and dramatic transition to become mature, motile sperm. Because the sperm nucleus is transcriptionally silent, this transition does not involve transcriptional changes. Although Caenorhabditis elegans is a leading model for studies of sperm activation, the mechanisms by which signaling pathways induce this transformation remain poorly characterized. Here we show that a conserved transmembrane zinc transporter, ZIPT-7.1, regulates the induction of sperm activation in Caenorhabditis nematodes. The zipt-7.1 mutant hermaphrodites cannot self-fertilize, and males reproduce poorly, because mutant spermatids are defective in responding to activating signals. The zipt-7.1 gene is expressed in the germ line and functions in germ cells to promote sperm activation. When expressed in mammalian cells, ZIPT-7.1 mediates zinc transport with high specificity and is predominantly located on internal membranes. Finally, genetic epistasis places zipt-7.1 at the end of the spe-8 sperm activation pathway, and ZIPT-7.1 binds SPE-4, a presenilin that regulates sperm activation. Based on these results, we propose a new model for sperm activation. In spermatids, inactive ZIPT-7.1 is localized to the membranous organelles, which contain higher levels of zinc than the cytoplasm. When sperm activation is triggered, ZIPT-7.1 activity increases, releasing zinc from internal stores. The resulting increase in cytoplasmic zinc promotes the phenotypic changes characteristic of activation. Thus, zinc signaling is a key step in the signal transduction process that mediates sperm activation, and we have identified a zinc transporter that is central to this activation process.
| Sperm are specialized cells with transcriptionally silent DNA that has been packaged for delivery into the egg. In their final step of development, immature sperm undergo a rapid transition from nonmotile cells to mature, motile sperm capable of fertilization. The signals that trigger this change are not clearly understood. By identifying mutants in the roundworm Caenorhabditis elegans that are defective in sperm activation, we discovered a conserved transmembrane protein, ZIPT-7.1, that transports zinc and promotes sperm activation in both sexes. ZIPT-7.1 is expressed in the germ line and functions there to control sperm activation. When expressed ectopically in mammalian cells, the protein specifically transports zinc across membranes and localizes primarily to membranes within the cell. Previous genetic studies had identified two pathways that mediate sperm activation in C. elegans, and our results suggest that zipt-7.1 acts at the end of one of these two, the spe-8 pathway. We propose that when this pathway triggers sperm activation, it acts through ZIPT-7.1, which mediates the release of zinc from internal stores in the immature sperm. This released zinc functions as a second messenger to promote the differentiation of mature, motile sperm.
| In animals, each differentiated cell type expresses a unique set of genes, resulting in the characteristic array of proteins that together confer its identity. These proteins equip the cell for the specific functions it needs to perform its role within a complex animal. Sperm provide a fascinating example of how multiple specializations combine to promote one distinct function—delivering the male pronucleus to the egg to produce a fertilized zygote. One specialization is that sperm chromatin has been compacted to prepare for its delivery into the egg, resulting in transcriptional silencing [1, 2]. As a result, changes in sperm phenotype or behavior must be accomplished without new gene transcription. A second specialization is that inactive sperm from many species are stored until ejaculation, when they undergo dramatic postmeiotic transitions called sperm activation, capacitation, or spermiogenesis [3, 4]. In this process, the immotile spermatid becomes a mature, motile sperm. Precisely regulated activation conserves energy until it is needed to drive motility and fertilization.
Only the early stages of spermatogenesis involve changes in gene expression; thus, sperm activation occurs in cells with transcriptionally silent nuclei. Because well-established pathways such as wnt, hedgehog, receptor tyrosine kinase, TGFβ, and Notch do not appear to mediate activation, sperm are likely to use novel signaling pathways to induce this phenotype. And because sperm activation is widespread in animals, the mechanisms that control it might be ancient and conserved. To date, these mechanisms remain poorly understood.
The nematode Caenorhabditis elegans is ideal for studies of sperm differentiation and activation. In this species, inactive spermatids are round and immotile, with membranous organelles located just beneath the plasma membrane [5]. During activation, these organelles fuse with the plasma membrane, and major sperm protein (MSP) is reorganized into filaments that generate a pseudopod for crawling. This amoeboid movement is common in nematode sperm and may be adapted to the folded surfaces of the female or hermaphrodite reproductive tracts.
Forward genetic screens for sterile animals have resulted in the identification of many genes that are essential for one or more steps of sperm development in C. elegans [6, 7]. The molecular analysis of these genes has identified proteins that participate in two distinct signal transduction pathways that regulate sperm activation [4]. One of these pathways is triggered by a secreted trypsin protease [8], but the signal for the other pathway remains elusive. Although several chemicals that can activate sperm in vitro have been identified [9, 10], they have not resolved this problem. Here we demonstrate that the zinc transporter ZIPT-7.1 plays a critical role in this second sperm activation pathway.
To identify new genes involved in sperm biology, we analyzed mutations that cause hermaphrodites to become self-sterile. The new mutations hc130 and as42 cause a loss-of-function in zipt-7.1, resulting in partial or complete sterility in both sexes. This defect is due to a failure of sperm activation, a function conserved in the related nematode C. tropicalis. The ZIPT-7.1 protein is homologous to mammalian ZIP7, a zinc transporter from the ZIP family. We confirmed that C. elegans ZIPT-7.1 can act as a zinc-selective importer in cultured cells. Moreover, ZIPT-7.1 is expressed in the nematode germ line, consistent with a function in sperm. Genetic studies indicate that it interacts with other genes involved in sperm activation and functions downstream of spe-6. Finally, ZIPT-7.1 can bind the presenilin SPE-4.
Thus, we propose a new model for sperm activation. In spermatids, inactive ZIPT-7.1 is localized to the membranous organelles, which contain much higher levels of zinc than the cytoplasm. When activation is triggered, a signal transduced by the SPE-8 group of proteins opposes SPE-4 and SPE-6. As their function decreases, ZIPT-7.1 becomes active and transports zinc into the cytoplasm. The resulting increase in cytoplasmic zinc promotes the phenotypic changes that are characteristic of activation, including motility. Thus, the release of zinc from internal stores is a key part of the signal transduction process that mediates sperm activation.
These discoveries have important implications for the fields of zinc biology and sperm activation. Zinc is essential for all life and has well-established functions as a cofactor for numerous proteins. Zinc binding is necessary for the tertiary structure of many of these proteins, such as zinc finger transcription factors, and zinc binding to many enzymes is critical for catalysis. Although it has been suggested that changes in zinc concentration in specific compartments might have second messenger effects, it has been difficult to demonstrate this type of signaling. The best-established setting is the extracellular release of zinc during synaptic transmission, which changes the concentration of zinc in the synaptic cleft [11]. By contrast, examples of zinc signaling that control cell fate and development are lacking. Our demonstration that a zinc signal controls sperm activation places zinc signaling in a specific biological context, in which changes in cell identity cannot be mediated by changes in gene expression. Furthermore, we have identified a zinc transporter that is central to this activation process. Finally, our discoveries show that signal transduction using zinc can control how cells differentiate during development.
Two independent lines of investigation converged on zipt-7.1 as a critical regulator of fertility. The first approach was based on a forward genetic screen for sterile C. elegans hermaphrodites, which led to the identification of the recessive mutation hc130. Genetic mapping experiments were used to position this mutation to the right of dpy-4, near the end of chromosome IV (Fig 1A), and whole genome sequencing revealed a missense mutation that eliminated the ATG start codon of T28F3.3, a gene located in this region (Fig 1B).
The second approach was designed to elucidate mechanisms of zinc biology by conducting a reverse genetic study of C. elegans genes encoding ZIP proteins. Homology searches identified 14 such genes, and phylogenetic analyses revealed that many are closely related to human proteins. Thus, we named these genes ZRT- and IRT-like protein transporters (zipt) and assigned numbers corresponding to the most similar human genes (Fig 1C, S1 Table). By analyzing deletion alleles, we discovered that zipt-7.1(ok971), which deletes T28F3.3, caused hermaphrodite sterility. Complementation tests showed that hc130/ok971 heterozygotes were sterile, confirming that the missense mutation identified in T28F3.3 causes the hc130 phenotype.
Finally, we used a screening procedure in which sterile mutants were identified by their failure to form “bags-of-worms” when prevented from laying eggs [12] to identify another mutation that causes this phenotype. This allele, as42, has a G797A mutation in T28F3.3, which changes a glycine to glutamic acid within a predicted transmembrane domain. Taken together, these three alleles identify a previously uncharacterized zipt gene required for nematode fertility.
To analyze zipt-7.1 function, we studied the null allele ok971, which deletes the entire coding region (Fig 1B). Whereas wild-type hermaphrodites had an average brood size of 225 self progeny, and individuals were invariably fertile, zipt-7.1 mutants had significantly smaller broods, and most individuals were completely sterile (Fig 2A, S1A Fig). Thus, zipt-7.1 loss-of-function causes a fully penetrant reduction in the number of self progeny and partially penetrant sterility. Furthermore, these mutant hermaphrodites laid large numbers of unfertilized oocytes (Fig 2B, S1A Fig), which implies that the MSP signal that stimulates ovulation is intact [13]. Because both of these defects were corrected by crossing zipt-7.1(ok971) hermaphrodites with wild-type males (Fig 2A and 2B), we infer that the mutant hermaphrodites make defective sperm but functional oocytes.
To characterize this fertility defect, we used differential interference contrast (DIC) optics to view live animals. In wild-type hermaphrodites, sperm actively moved into the two spermathecae. As a result, each ovulation resulted in fertilization and the release of a new embryo into the uterus (Fig 2C). By contrast, in zipt-7.1 mutant hermaphrodites the spermathecae were empty and scattered spermatids and unfertilized oocytes were visible in the uterus (Fig 2D). We infer that the mutant sperm retained the ability to stimulate ovulation but were unable to migrate back to the spermathecae after being pushed into the uterus during ovulation [6].
To study male sperm, we used crosses with self-sterile hermaphrodites or females. We first tested the ability of male sperm to compete with spe-8 hermaphrodite sperm, which fail to activate unless stimulated by male seminal fluid [14]. The wild-type male sperm competed efficiently with Trans-activated spe-8 hermaphrodite sperm, fertilizing all of the oocytes and yielding only cross progeny. By contrast, the zipt-7.1 male sperm competed poorly, fertilizing only a minority of the oocytes and resulting in numerous self progeny (Fig 2E).
To learn if the zipt-7.1 male sperm were defective in competition or had an absolute decline in function, we measured the ability of zipt-7.1 males to fertilize fog-2 females, which make no self-sperm. Even though the zipt-7.1 mutant sperm had no competition, we observed a dramatic decrease in successful fertilizations compared to wild-type males (Fig 2F). Thus, zipt-7.1 activity promotes the function of sperm in both hermaphrodites and males, and it appears to regulate either activation or motility.
Nematode spermatids remain round and immotile until they receive an activating signal, which causes them to extend a pseudopod and begin to crawl (Fig 3A) [4]. Sperm isolated by dissection from wild-type hermaphrodites displayed the extended pseudopods characteristic of in vivo activation. By contrast, those isolated from zipt-7.1 hermaphrodites lacked pseudopods and appeared to be round, immotile spermatids, suggesting activation had not occurred (S1 Fig). Hence, we began studying the ability of zipt-7.1 spermatids to activate.
The trypsin protease TRY-5 is an endogenous activating signal [8], and zinc [10], trypsin, and the protease mixture Pronase [9] can stimulate activation in vitro. To measure the response of spermatids to these signals, we dissected adult animals to release spermatids into sperm medium. Because wild-type hermaphrodites produce only small numbers of sperm prior to oogenesis, we analyzed fem-3(q96) mutants, which produce numerous sperm throughout their lives [15]. Spermatids dissected from fem-3 hermaphrodites displayed robust activation in response to all three signals in vitro; by contrast, those dissected from fem-3 zipt-7.1 double mutants displayed significantly lower levels of activation (Fig 3B). We repeated these experiments with spermatids isolated from males and obtained similar results (Fig 3C). Thus, zipt-7.1 regulates activation.
Two points were notable. First, zipt-7.1 mutant spermatids occasionally activated, indicating that this defect is not completely penetrant. This result might explain our observation that zipt-7.1 sterility is also partially penetrant. Second, the least effective activator of zipt-7.1 mutant spermatids was zinc, which is consistent with the model that zipt-7.1 functions in zinc biology.
To determine the expression pattern of zipt-7.1, we used Reverse transcription polymerase chain reaction (RT-PCR) to analyze transcript levels in mutant strains that had altered germ cell fates. The zipt-7.1 transcripts were readily detectable in animals containing only sperm or only oocytes, but almost undetectable in animals that lacked most germ cells (Fig 4A). These results suggest that zipt-7.1 is predominantly expressed in the germ line or that its expression in other tissues depends on germ cells. By contrast, transcripts of the related gene zipt-7.2 were readily detectable in animals that lacked most germ cells, indicating expression in somatic tissues.
It was technically challenging to visualize the ZIPT-7.1 protein in situ. Two different polyclonal antibodies raised against ZIPT-7.1 peptides were unable to detect ZIPT-7.1 expression in situ, although they could detect ZIPT-7.1 expressed in human cells (S3 Fig); thus, in vivo expression levels might be low. Multiple attempts to insert epitope tags into the endogenous locus were unsuccessful. Ultimately, we used gene editing to insert sequences encoding green fluorescent protein (GFP) into the endogenous zipt-7.1 locus to encode a fusion protein with GFP inserted in the first predicted cytoplasmic loop, between amino acids 25 and 26 (Fig 4B). Animals homozygous for the modified allele developed normally, demonstrating that this GFP::ZIPT-7.1 fusion protein is functional. Protein expression could only be detected with anti-GFP antibodies, because the expression level was too low to observe GFP fluorescence. We found that GFP::ZIPT-7.1 expression levels were highest in developing spermatocytes, consistent with the analysis of transcript expression (Fig 4C). Furthermore, it appeared to be excluded from the nucleus and concentrated in puncta in the cytoplasm, suggesting that it localized to subcellular organelles. GFP::ZIPT-7.1 could not be visualized in spermatids or mature sperm with this protocol, which could be due to its low level of expression.
Based on these data, we hypothesized that zipt-7.1 functions cell autonomously to promote sperm activation. To test this model, we used RNA interference (RNAi) in rrf-1 mutants. Wild-type animals are susceptible to RNAi in the germ line and most somatic cells. By contrast, rrf-1 function is required for RNAi to work in most of the somatic tissues, so rrf-1(−) mutants are mainly susceptible to RNAi in the germ line [17–19]. RNAi directed against zipt-7.1 caused similar sterility in both wild-type and rrf-1 mutant hermaphrodites (Fig 4D). Thus, zipt-7.1 is necessary in the germ line to promote fertility.
Taking all of these results together, we conclude that ZIPT-7.1 functions in developing sperm to regulate activation, rather than in the soma to control an activating signal.
Many ZIP family proteins specifically transport zinc, but some transport iron or other metals. Thus, we investigated the role of ZIPT-7.1 in zinc biology. To analyze levels of labile zinc in vivo, we isolated male spermatids and stained them with Zinpyr-1, a dye that fluoresces when it binds zinc [20]. Wild-type spermatids displayed punctate fluorescence (Fig 5A) [10]. Some puncta colocalized with the dye LysoTracker, suggesting they were membranous organelles, and others colocalized with the dye MitoTracker, suggesting they were mitochondria (S2 Fig). Although zipt-7.1 mutant spermatids displayed a similar pattern of fluorescence, the intensity of the fluorescence was significantly lower (Fig 5A and 5B). Because this difference is detectable after spermatogenesis is complete but before activation has occurred, we infer that zipt-7.1 promotes the accumulation of intracellular zinc during spermatogenesis. Following activation, the mature sperm extend a pseudopod that does not display labile zinc, membranous organelles, nor mitochondria (S2 Fig).
If zipt-7.1 functions in zinc biology, then its expression might be regulated by the level of available zinc. To test this prediction, we cultured wild-type animals with the zinc-specific chelator N,N,N',N'-tetrakis(2-pyridinylmethyl)-1,2-ethanediamine (TPEN) to induce zinc deficiency and analyzed zipt-7.1 transcript levels. Although the expression of control genes did not change, levels of zipt-7.1 mRNA increased 4-fold (Fig 5C), consistent with the model that zipt-7.1 plays an in vivo role in zinc biology. This finding is consistent with a recent report by Dietrich and colleagues [21] that the zipt-7.1 locus contains a low zinc activation (LZA) enhancer motif that mediates transcriptional activation in response to low dietary zinc.
To test the prediction that ZIPT-7.1 transports zinc, we expressed it in mammalian cells and measured its subcellular localization and zinc uptake using radioactive 65Zn. We used human HeLa cells for localization studies because of their well-defined cellular architecture. Antibody staining revealed that ZIPT-7.1 was expressed in a punctate pattern, and some puncta appeared to colocalize with the Cis-Golgi marker GM130 and the lysosomal marker LMP2. In addition, some ZIPT-7.1 appeared to be localized on the nuclear envelope (Fig 5D–5I, S3 Fig). We used human HEK293T cells for uptake studies because of their high efficiency of transfection. The expression of ZIPT-7.1 significantly increased the uptake of radioactive 65Zn from the medium (Fig 5J–5L). Furthermore, this transport activity was zinc specific, because it was effectively competed by nonradioactive zinc but not by several other metal ions (Fig 5K and 5L).
Thus, three lines of evidence suggest that ZIPT-7.1 is a zinc-specific transporter. (1) The expression of zipt-7.1 transcripts in nematodes is regulated by zinc, (2) it controls intracellular zinc levels in developing spermatids, and (3) it is capable of specifically transporting zinc across membranes when assayed in mammalian cells.
C. elegans ZIPT-7.1 and ZIPT-7.2 are both closely related to a single human protein, ZIP7, indicating that a primordial gene duplicated and diverged in the nematode lineage. To investigate this gene duplication, we analyzed the genomes of related nematodes. Orthologs of both genes are present throughout the elegans group of the genus Caenorhabditis, but the nematode Onchocerca volvulus contained only a single ZIP7 homolog (Fig 6A) [22]. We conclude that the duplication and functional divergence of ZIPT-7.1 and ZIPT-7.2 occurred relatively recently during nematode evolution.
To see if the function of zipt-7.1 is conserved, we used gene editing to alter exon #1 of the zipt-7.1 gene in the related hermaphroditic species C. tropicalis [23–25]. We recovered seven mutant alleles, including two frameshift mutations predicted to eliminate gene function (Fig 6B). Both of these frameshift mutations caused hermaphrodite sterility (Fig 6C). We tested one allele in males, for whom it also caused sterility (Fig 6D). Finally, Zinpyr-1 staining showed that this mutation resulted in reduced zinc accumulation in spermatids (Fig 6E and 6F). Thus, all three zipt-7.1(−) phenotypes observed in C. tropicalis are similar to those already described for C. elegans. These results suggest that zipt-7.1 has a conserved function in Caenorhabditis—it promotes sperm activation by regulating zinc.
In C. elegans, multiple genes can be mutated to block sperm activation or cause constitutive activation [4]. These genes have been organized into two genetic pathways, and the subcellular localizations of several proteins have been identified. In hermaphrodites, sperm activation is controlled by an unknown signal that acts through five proteins located at the cell membrane of the spermatid [4]. Mutations in any of these five genes (spe-8, spe-12, spe-19, spe-27, or spe-29) prevent activation. By contrast, mutations in spe-6 and spe-4 suppress the defective activation phenotype produced by these spe-8 group genes and also cause male sperm to activate prematurely, prior to ejaculation [26, 27]. These results suggest that spe-4 and spe-6 act downstream of spe-8 and its partners, and spe-4 and spe-6 currently define the downstream end points of this sperm activation pathway.
To position zipt-7.1 in this pathway, we generated double mutants with spe-4(hc196) and spe-6(hc163). Germ cells in the spe-4; zipt-7.1 double mutant arrested as abnormal primary spermatocytes that failed to divide. Because they made no sperm that could be tested for activation, this approach was not informative. By contrast, spe-6 mutant males displayed prematurely active sperm in their spermathecae [26], but spe-6(−); zipt-7.1(−) males did not (Fig 7A). Furthermore, spe-6(−); zipt-7.1(−) hermaphrodites were self-sterile, suggesting that zipt-7.1 functions downstream of spe-6 in both sexes, or that these two genes act in parallel (Fig 7A). By contrast, spe-8(−); spe-6(−) hermaphrodites are self-fertile [26]. These results distinguish zipt-7.1 from the spe-8 group and suggest that zipt-7.1 functions downstream of spe-6, at the end of the sperm activation pathway (Fig 8A).
In males, sperm can also be activated by the extracellular protease TRY-5, which is likely to act through the membrane protein SNF-10 [8, 29]. Prior to ejaculation, TRY-5 is inhibited by the SWM-1 protease inhibitor, which prevents premature activation [30]. Thus, swm-1 mutant males have abnormally active sperm crawling inside the reproductive tract, similar to spe-6 or spe-4 mutant males. The phenotype of zipt-7.1(−); swm-1(−) double mutant males was intermediate between that of each single mutant (Fig 7C), so zipt-7.1 might function in parallel to the try-5 pathway (Fig 8A).
To complement these genetic experiments, we performed biochemical studies using the split-ubiquitin two-hybrid system (Fig 7B, S4 Fig). ZIPT-7.1 interacted robustly with SPE-4, a presenilin localized to the membrane of the membranous organelles [27], but not with SPE-6, SPE-8, SPE-19, SPE-27, or SPE-43. Thus, SPE-4 might directly inhibit ZIPT-7.1 function in spermatids to prevent premature sperm activation, and relief of this inhibition by the sperm activation pathway might allow ZIPT-7.1 to transport zinc, elevating the zinc concentration in the cytoplasm and promoting sperm activation.
The analysis of three mutations demonstrates that zipt-7.1 promotes sperm activation. Two are molecular null alleles—hc130 eliminates the start codon and ok971 deletes the entire coding region—whereas as42 changes a glycine to glutamic acid in a predicted transmembrane domain. All three mutations severely reduced production of hermaphrodite self progeny, and rescue by crossing with wild-type males indicates a defect in hermaphrodite sperm. There is also a defect in male sperm, because zipt-7.1 mutant males were impaired in fertilizing hermaphrodites. Spermatids dissected from zipt-7.1 hermaphrodites had failed to activate, and mutant spermatids from either sex responded poorly to triggering agents like zinc, Pronase, or trypsin, pinpointing the problem to sperm activation.
Whereas other animals such as fruit flies and humans have a single ZIP7 gene, C. elegans has two—zipt-7.1 and zipt-7.2—suggesting a gene duplication occurred in nematodes. These two genes have separate functions, because the zipt-7.2(ok960) mutation causes lethality (wormbase.org), indicating that somatic functions of ZIP7 are necessary for survival and are provided by zipt-7.2. To determine if these specializations have been conserved in Caenorhabditis, we obtained mutations of zipt-7.1 in the related nematode C. tropicalis. As observed with C. elegans, the C. tropicalis zipt-7.1 mutations caused a decrease in hermaphrodite and male fertility, indicating a sperm defect in both sexes, and abnormal zinc staining of spermatids, indicating a defect in zinc biology. Thus, the duplication of the ZIP7 genes during Caenorhabditis evolution appears to have been accompanied by the specialization of zipt-7.1 to control sperm development. As a result, C. elegans zipt-7.1 is a powerful model for studying ZIP7 genes during reproduction, because the essential functions of ZIP7 transporters are covered by its sister, zipt-7.2, thereby avoiding the complexities of pleiotropic phenotypes and inviable mutants.
zipt-7.1 is the first zinc transporter to be implicated in sperm activation, and this discovery opens up exciting new models for the role of zinc in this process of rapid cell differentiation. We hypothesize that ZIPT-7.1 mediates the regulated release of zinc from the lumen of intracellular vesicles, and that released zinc functions as a second messenger to promote sperm activation (Fig 8B). A variety of approaches have led to the identification of multiple genes that are involved in sperm development in worms, but zipt-7.1 is the first of these genes to have a clear role in zinc biology. In other animals, such as Drosophila and vertebrates, zinc transporters have not been demonstrated to function in sperm activation, highlighting the novelty of this discovery.
Sperm activation is a unique cell differentiation process that is rapidly induced in response to extracellular cues but does not involve transcriptional changes. Despite extensive investigations in multiple species, the signaling cascade that initiates sperm activation remains poorly defined. Here we propose that ZIPT-7.1 mediates the release of zinc from intracellular organelles in spermatids and that zinc may function as a second messenger that promotes sperm activation (Fig 8B).
C. elegans ZIPT-7.1 is similar to human ZIP7 and Drosophila Catsup, which implies that it transports a divalent cation. When expressed in mammalian cells, ZIPT-7.1 robustly increased zinc uptake, and competition experiments demonstrated that this transport was relatively specific for zinc. Thus, the zinc transport activity of ZIPT-7.1 was predicted by homology and confirmed by biochemical studies. Furthermore, ZIPT-7.1 protein localized primarily to internal membranes of mammalian cells, including the lysosome and Golgi, so it is likely to control release of zinc from internal stores. This localization pattern is similar to that of vertebrate ZIP7 [31], which implies that it is an evolutionarily conserved property of ZIP7 family members.
In nematodes, the expression and function of zipt-7.1 are restricted to germ cells. First, zipt-7.1 mRNA was predominantly expressed in the germ line. Second, a GFP::ZIPT-1 fusion protein expressed from the endogenous locus was localized in developing spermatocytes. Third, zipt-7.1(RNAi) caused sterility in an rrf-1 mutant background, in which sensitivity is restricted primarily to the germ line [17–19], indicating that zipt-7.1 functions in germ cells. These results suggest that zipt-7.1 does not act in the soma to produce an activation signal but in sperm to mediate their response.
Two additional observations link zipt-7.1 to zinc biology. First, zipt-7.1 mRNA levels increased in response to zinc deficiency. Dietrich and colleagues [21] discovered an LZA enhancer element in the zipt-7.1 gene that likely mediates this regulatory response. Second, the levels of labile zinc were reduced in spermatids from mutant animals, showing that zipt-7.1 is required for the uptake and storage of wild-type levels of zinc in developing spermatocytes. Taken together, these observations pinpoint the expression pattern of ZIPT-7.1 (gonad), its site of action (gonad), its biochemical activity (zinc transporter increasing cytoplasmic zinc), and its subcellular localization (primarily internal membranes).
Extensive genetic and molecular studies have identified two pathways required for nematode sperm activation, referred to as the SPE-8 and TRY-5 pathways [4]. Both pathways contain negative regulatory genes that, when mutated, cause spermatids to activate constitutively and prematurely: spe-6 and spe-4 in the SPE-8 pathway and swm-1 in the TRY-5 pathway. We took advantage of these alleles to perform genetic epistasis studies. A zipt-7.1 mutation strongly suppressed the spe-6 phenotype, suggesting that zipt-7.1 acts downstream of spe-6 if these genes act in a linear pathway. It remains possible that they function in parallel.
To characterize ZIPT-7.1 interactions with the SPE-8 pathway further, we used the yeast two-hybrid system to investigate protein–protein interactions. ZIPT-7.1 specifically bound SPE-4, a nematode presenilin that localizes to internal membranes in spermatids [32]. The role of presenilins in the regulation of zinc has formed an important area of research in Alzheimer disease for many years [33]. The fact that SPE-4 also acts late in the sperm activation process [27] and can bind ZIPT-7.1 supports the model that ZIPT-7.1 functions downstream of other known proteins in the SPE-8 pathway. Taken together, these results are consistent with the model that ZIPT-7.1 acts inside spermatids at the end of the known SPE-8 pathway to transmit an extracellular signal that triggers sperm activation (Fig 8A).
The effects of zinc on sperm have been investigated in several animals, including vertebrates (human, mouse, hamster), sea urchin, and C. elegans. Our results extend previous studies by (1) identifying the relevant zinc transporter and (2) providing a unified model for the function of zinc during sperm activation.
Sea urchin spermatids are normally ejaculated into sea water, and diluting them into sea water triggers activation and sperm motility [34]. Further studies demonstrated that zinc is the active ingredient in sea water and that zinc causes an increase in intracellular pH and intracellular calcium levels, triggering the acrosome reaction, a membrane fusion event [34, 35]. However, the mechanism by which zinc activates sea urchin sperm has not been defined, and no zinc transporters have thus far been implicated. The role of zinc in vertebrate sperm activation has been controversial; experiments with zinc chelators and the addition of supplemental zinc have suggested a variety of roles for zinc, indicating its importance, but specific functions have remained elusive [36].
High levels of extracellular zinc can activate C. elegans sperm in vitro [10], but the direct mechanism was not previously defined. We propose that physiological sperm activation involves zinc release from intracellular stores, which increases the cytoplasmic concentration of zinc and causes activation. High extracellular zinc leads to zinc entry into spermatids, thereby mimicking this physiological signal. The C. elegans model is similar to sea urchins in that rising concentrations of zinc stimulate sperm activation, with the difference that extracellular zinc is the physiological source in sea urchins, whereas intracellular stores are likely to be the physiological source in C. elegans. Although high levels of extracellular zinc can activate C. elegans sperm in a zipt-7.1–dependent manner in vitro, our results suggest this is unlikely to be the physiological trigger, because zipt-7.1 acts downstream of SPE-8 pathway proteins located at the spermatid membrane.
Well-established functions for zinc involve stable binding to proteins to influence tertiary structure or facilitate catalysis. In addition, zinc has been proposed to act as a second messenger, like calcium [37], but this function is just beginning to be explored and many questions remain. Proposed examples of zinc signaling can be divided into extracellular and intracellular. In the vertebrate nervous system, zinc is concentrated in synaptic vesicles with neurotransmitters and released into the synaptic cleft upon nerve stimulation, where it may modulate the activity of neurotransmitter receptors [11]. The second example of extracellular release is the “zinc spark” that has been visualized during mammalian oocyte fertilization [38]. This spark is caused by the synchronous fusion of many zinc-containing vesicles. In both cases, zinc is released to the extracellular space by vesicle fusion. By contrast, Yamasaki and colleagues visualized an intracellular “zinc wave” in vertebrate mast cells that had been stimulated to undergo degranulation by an extracellular ligand [39]. This zinc wave appeared to originate from the endoplasmic reticulum. In addition, a rapid increase in cytoplasmic zinc has been observed in T lymphocytes and leukocytes responding to extracellular signals [40, 41]. Finally, Hogstrand and colleagues proposed that the vertebrate ZIP7 is localized to intracellular membranes and mediates a zinc signal in breast epithelial cells [42].
The results presented here identify a new biological system for zinc signaling—sperm activation. It is intriguing that the “zinc wave” in mast cells mediates degranulation, a vesicle fusion event, and the zinc signal in sperm also mediates vesicle fusion. These results raise the possibility that intracellular zinc signals have a conserved function in promoting vesicle fusion. Our results are consistent with the model of vertebrate ZIP7 mediating an intracellular zinc signal, although the cell type and biological consequences are entirely different. Because nematode sperm activation can be dissected by forward and reverse genetics and manipulated both in vivo and in vitro, it offers a powerful model to define mechanisms of zinc signaling.
A critical question in sperm biology is how the rapid and dramatic changes that define activation are controlled without transcriptional regulation. The answer may involve converging lines of research on ion fluxes. First, it is well established for many animals, including nematodes, that intracellular pH increases during sperm activation [9]. Lishko and Kirichok identified a voltage-gated proton channel that mediates this pH rise in human sperm [43]. Second, calcium signaling has been repeatedly implicated in sperm activation and plays an important role in nematodes [44, 45]. The CatSper channel mediates this increase in intracellular calcium in mammalian sperm [46]. C. elegans does not have an obvious homolog of CatSper, but they do contain multiple predicted calcium channels; thus, calcium signaling in worms could be mediated by an alternative channel [47, 48]. Third, we propose here that zinc is a second messenger that activates sperm, and ZIPT-7.1 is the relevant transporter.
Considering our results in light of the documented roles of other ions, we speculate that sperm activation might be caused by a coordinated change in the cytoplasmic levels of three different ions: H+, Ca++, and Zn++. According to this model, a sperm activation signal results in the activation of an H+ exporter, a calcium importer, and the zinc importer ZIPT-7.1. The activity of these transporters changes the cytoplasm from a high concentration of H+ and low concentrations of Ca++ and Zn++ (the spermatid state) to a low concentration of H+ and high concentrations of Ca++ and Zn++ (the activated state). Changes in protein activity could result from alterations of pH and interactions with zinc and calcium cofactors without the need for altered transcription. Individual proteins might be influenced by one or more of these ion changes, and the combination of all three ion changes might rapidly alter the activity of a large number of proteins. Future work will be required to identify specific target proteins and determine whether pH, calcium, zinc, or a combination causes changes of protein activity.
C. elegans strains were derived from Bristol N2 [49]. They include fog-1(q253) I, glp-4(bn2) I, rrf-1(pk1417) I, Dsp2/spe-8(hc53) dpy-5 I, spe-6(hc163) dpy-18(e364) III, dpy-13(e184) IV, zipt-7.1(hc130) IV [informally referred to as spe-24(hc130)], zipt-7.1(ok971) IV [informally referred to as hke-4.1(ok971)], zipt-7.1(as42) IV, fem-3(q96) IV, fem-1(hc17) IV, swm-1(ok1193) V, him-5(e1490) V, and fog-2 (q71) V. Double or triple mutants made in this study include (1) zipt-7.1(ok971) IV; him-5(e1490) V, (2) swm-1(ok1193) him-5(e1490) V, (3) zipt-7.1(ok971) fem-3(q96) IV, (4) zipt-7.1(ok971) IV; swm-1(ok1193) him-5(e1490) V, and (5) spe-6(hc163) dpy-18(e364) III; zipt-7.1(ok971) IV. The zipt-7.1(ibp18 gfp insertion) strain was produced using CRISPR genome editing technology [50].
C. tropicalis mutants were derived from the wild isolate JU1373 and include him-8(v287), unc-23(v277), and try-5(v275) [25, 51]. The zipt-7.1 mutations v332, v334, and v335 described here were made using TALENs [25].
Dpy hermaphrodites from the strain hc130 dpy-13(e184)/nT1 were crossed with N2 males to separate the hc130 allele from dpy-13. Next, sterile, non-Dpy hermaphrodites isolated from the F2 were crossed with males from the polymorphic strain CB4856 [52, 53]. Finally, 480 F2 hermaphrodites from this cross were picked at the L4 stage to individual wells of 24-well plates. Each worm was scored for sterility the following day, and about 80 sterile F2 hermaphrodites were combined for isolation of sheared genomic DNA, as describe by Smith and colleagues [54].
We used 10 ng of fragmented DNA to prepare a library using the Hyper Prep Kit (KAPA Biosystems), as instructed by the manufacturer. Next, fragments sized 300–500 bp were selected using the QIAQuick Gel Extraction Kit (Qiagen) and sequenced on an Illumina HiSeq2000 instrument using the 50-cycles, single-end mode. We obtained 8,598,343 reads.
Data were analyzed at usegalaxy.org using the CloudMap Hawaiian Variant Mapping with WGS Data tool [55, 56]. The top candidate for the hc130 mutation was a G to A transition in the start ATG codon of zipt-7.1. Sanger DNA sequencing confirmed the presence of this mutation in an hc130 dpy-13(e184) strain but not in our N2 strain.
The following protein sequences are aligned in Fig 1D, identified by species, isoform, and NCBI reference sequence code: CeZIPT-7.1 (C. elegans, NP_503070.2); CeZIPT-7.2 (C. elegans, isoform 2, NP_510563.2); DmCATSUP (D. melanogaster, NP_524931.1); and HsZIP7 (H. sapiens, isoform 1 precursor, NP_008910.2). The alignment was performed with ClustalX.
To measure hermaphrodite fertility, we placed individual L4 animals on freshly seeded dishes, transferred them to new dishes every 8–16 hours for 5 days, and scored the number of fertilized eggs and unfertilized oocytes on each dish. To measure C. elegans male fertility, we used two assays. First, individual males were crossed to individual spe-8(hc53); dpy-5 L4 hermaphrodites for 24 hours. The male was removed, the hermaphrodite was transferred to a new dish every 24 hours, and the offspring were scored as Dpy (self) or non-Dpy (cross) progeny upon reaching adulthood. Second, males were crossed to fog-2(q71) L4 females using similar procedures. To measure C. tropicalis male fertility, we crossed males with unc-23(v277) hermaphrodites and scored progeny as either Unc (self) or non-Unc (cross).
L4 males were placed onto new NGM dishes without hermaphrodites for 48–60 hours. The males were then dissected and sperm released into a droplet of SM buffer (50 mM HEPES, 45 mM NaCl, 25 mM KCl, 1 mM MgCl2, 5 mM CaCl2, 10 mg/mL PVP, pH 7.0). These sperm were maintained in a chamber constructed by mounting a 22×30 mm glass coverslip onto a glass slide over parallel strips of two-sided sticky tape. Then, 1 mM zinc, 200 μg/mL Pronase, or 1 mg/mL trypsin was poured into the chamber and incubated for 10–15 minutes to activate the sperm. After the treatment, sperm were observed using an Axio Imager M2 microscope (Carl Zeiss) with DIC optics.
For whole worm observations, young adult worms were placed on a 5% agarose pad with a droplet of 1 mM levamisole and viewed with DIC optics. Images were captured using a Zeiss Axiocam digital camera with Zeiss AxioVision software and assembled using Adobe Photoshop.
Groups of five adult worms were collected and processed as described [57]; two independent samples were used to confirm reproducibility. RNA was extracted as described, and the reverse transcription reaction was performed with MMLV Reverse Transcriptase (ThermoFisher Scientific). The PCR was conducted using HotMaster Taq DNA polymerase (5PRIME); PCR reactions were run for 30 cycles for the zipt-7.1 and zipt-7.2 genes or 24 cycles for the act-1 gene. Primers are listed in S2 Table.
We used the qRT-PCR method described by Davis and colleagues [58], with minor modifications. Mixed-stage populations of animals were cultured for 16 hours on NAMM dishes supplemented with 0 or 40 μM TPEN, seeded with concentrated Escherichia coli OP50, and collected by washing. RNA was isolated using TRIzol (Invitrogen), treated with Dnase I, and reverse transcribed with the High Capacity cDNA Reverse Transcription kit (Applied Biosystems). PCR was performed using an Applied Biosystems 7900 thermocycler and iTaq Universal SYBR Green Supermix (Bio-Rad). Primers used to detect zipt-7.1 are in S2 Table.
Double-stranded RNA (dsRNA) was synthesized using T7 RNA polymerase. The templates were amplified from nematode cDNA with PCR primers that contained the T7 promoter (S2 Table), purified with a PCR Purification kit (Qiagen), and transcribed using MegaScript (Ambion). After annealing overnight at 37°C, dsRNA was purified with MegaClear (Ambion). RNAi was performed by injection into hermaphrodites, and the progeny of injected animals were analyzed at the young adult stage, 24 hours after being picked as L4 larvae.
To visualize the distribution of zinc, we stained isolated spermatids with 10 μM Zinpyr-1 (Sigma-Aldrich) in divalent cation-free SM for 10 minutes at room temperature. LysoTracker Red and MitoTracker Red (ThermoFisher Scientific) were used at a final concentration of 1 μM to label membranous organelles [59] and mitochondria, respectively. Some spermatids were activated by Pronase after staining with dyes to identify differences in the distribution of zinc between spermatids and spermatozoa. Fluorescent images were obtained using an Olympus Fluoview 1200 confocal microscope.
To investigate GFP::ZIPT-7.1 expression and localization, we isolated the male gonad and fixed with 4% paraformaldehyde in SM buffer at 4°C overnight. Fixed samples were permeabilized with 0.5% Triton X-100 in PBS for 1 hour and then blocked with 2% BSA in PBS at room temperature for 6 hours. The primary anti-GFP Mab (Roche, 1:100 dilution) was incubated with samples at 4°C overnight. The secondary antibody was a goat anti-mouse IgG conjugated to horseradish peroxidase. Visualization used 5 minutes of Tyramide signal amplification with Alexa Fluor 488 conjugated to tyramine (ThermoFisher Scientific). Slides were mounted in solution (83 mg/mL mowiol, 25 mg/mL DABCo, 1 mg/mL DAPI in 100 mM Tris buffer, pH 8.0) and images captured using an Olympus Fluoview 1200 confocal microscope.
Rabbit polyclonal anti-ZIPT-7.1 antibodies were generated and purified by YenZyme. Two peptides based on the predicted ZIPT-7.1 protein sequence were used as antigens (#68–82, TSHREIQHSRLSTLK, and #138–150, SLSPHDHSHDHHD).
Plasmid pCT15, which was used to express ZIPT-7.1 in cultured cells, was constructed by inserting a 1,182-nucleotide zipt-7.1 cDNA encoding the full-length predicted protein and Kozak sequence into pcDNA3.1(+) (Invitrogen), using restriction sites HindIII and XbaI. The zipt-7.1 cDNA was produced by performing RT-PCR on a sample of C. elegans RNA. The plasmid was verified by DNA sequencing.
HeLa cells were grown on coverslips, transfected, fixed in 4% formaldehyde for 30 minutes at room temperature, and permeabilized in 0.1% Saponin/2% IgG-free BSA/PBS for 30 minutes. The primary antibodies were polyclonal rabbit antibodies against ZIPT-7.1 used in combination with mouse monoclonal antibodies against Golgi GM130 (1:1,000) (Abcam), Calnexin MAB3126 (1:250) (Millipore), or Lamp2 CD107b (1:200) (Pharmingen). The secondary antibodies were goat anti-rabbit antibodies conjugated with Alexa488 and goat anti-mouse antibodies labelled with Alexa594 (Invitrogen). Cells were mounted in Vectashield/DAPI (Vector Laboratories). Confocal fluorescent images were taken with a 63×/1.4NA planapo objective (Zeiss) on a spinning disc confocal microscope UltraVIEW VoX/Observer.Z1 (PerkinElmer; Zeiss) using Velocity 6.3. Images were exported as 16-bit TIFs and converted into 8-bit RGB TIFs using Fiji ImageJ, and the figures were assembled in Photoshop (Adobe).
HEK293T cells were cultured in Dulbecco’s Modified Eagle Medium (DMEM) (ThermoFisher) supplemented with 10% fetal bovine serum (FBS) (Sigma) and 1× Penicillin/Streptomycin Solution (Corning). HeLa cells were cultured in DMEM supplemented with 2 mM glutamine, 10% FBS, and Penicillin/Streptomycin. All cells were grown in a humidified incubator with 5% CO2 at 37°C.
Zinc uptake assays were performed as described, with minor modifications [60, 61]. Briefly, HEK293T cells were seeded on Poly-D-lysine–coated 24-well plates (Corning). The next day, the cells were transfected with a plasmid encoding ZIPT-7.1 or pcDNA-3.1(+) (a vector only control) using Lipofectamine 2000 (Invitrogen). After 48 hours, cells were washed once with prewarmed uptake buffer (15 mM HEPES, 100 mM glucose, 150 mM KCl, pH 7.0) and incubated for 15 minutes in prewarmed uptake buffer that contained the radioactive tracer 65ZnCl2 (PerkinElmer) and varying concentrations of nonradioactive metal ions. Uptake was halted by applying the same volume of ice-cold stop buffer (15 mM HEPES, 100 mM glucose, 150 mM KCl, 1 mM EDTA, pH 7.0). Cells were gently washed with ice-cold stop buffer twice and disassociated with trypsin. The radioactivity that had been incorporated into the cells was then measured with a Beckman LS 6000 Scintillation Counter. In parallel experiments conducted without adding metals, the cells were lysed with lysis buffer (2 mM Tris-HCl, 150 mM NaCl, 1% Triton X-100), and protein levels were measured with the Bio-Rad DC protein assay. 65Zn uptake was normalized to total protein measured in this parallel assay. These metals were obtained from Sigma: ZnCl2, MgCl2, CdCl2, FeCl3 (Fe3+ was reduced to Fe2+ with 1 mM sodium ascorbate shortly before the experiment), MnCl2, and CaCl2. The data shown in Fig 5J–5L are typical of multiple independent experiments.
TALEN mRNAs were designed and produced as described [25] (S3 Table). To create mutants, we injected pairs of TALEN mRNAs into the gonads of young adult hermaphrodites. The F1 progeny from a 6- to 30-hour time window were singled to new dishes and raised at 20°C. F1 worms that carried new mutations were identified by PCR analysis of the target site (primers in S3 Table), and mutants were confirmed by DNA sequencing.
To generate the GFP::ZIPT-7.1 expression strain, we used CRISPR technology [50] to modify the endogenous locus, so as to encode GFP between amino acids 25 and 26 of ZIPT-7.1. We designed the sgRNA sequence (TCCTTCATGGTGATGCTCGTGG) using the CRISPR design tool (http://crispr.mit.edu). The target site conformed to the sequence G(N19)NGG; the initial G optimizes transcription driven by the U6 promoter, and the NGG (PAM) motif is required for Cas9 activity.
To insert the sgRNA sequence into the CRISPR-Cas9 vector pDD162 (Addgene, #47549), the vector was amplified using primers that had a 15-bp overlap sequence (primers in S2 Table) and Phusion high-fidelity DNA polymerase (New England Biolabs). Next, the PCR products were treated with DpnI to remove the vector template and transformed into TOP10 competent cells. The plasmid was confirmed by DNA sequencing.
To insert gfp into the endogenous zipt-7.1 locus, we built a homologous recombination template comprised of left-arm sequence (1,031 bp), GFP deleted-stop-codon sequence with a 6 amino acid linker, and right-arm sequence (1,316 bp). The left- and right-arm sequences were amplified from genomic DNA using Phusion high-fidelity DNA polymerase and primers P1 and P5 (S2 Table) and inserted into the pPD95.77 backbone by In-fusion cloning (ClonTech). Next, the GFP sequence was amplified and inserted between the left and right arms. Positive clones were identified by the PCR, and the final plasmid was confirmed by DNA sequencing.
We injected Cas-9-sgRNA, homologous recombination template, and the selection marker pRF4 (rol-6) into N2 hermaphrodites at 50 ng/μL concentration. F1 Rol progeny were placed on individual dishes and harvested for PCR analysis after laying eggs for one to two days. We used primers P3 and P6 to identify worms that contained the gfp insert, and primers P2, P3, and P4 to screen their progeny for homozygotes (S2 Table). The gfp knock-in worms were confirmed by DNA sequencing the PCR product of primers P2 and P4.
The split-ubiquitin yeast two-hybrid assay used the DUALmembrane kit (Dualsystems Biotech) per manufacturer's instructions. The zipt-7.1 gene was amplified from him-5 cDNA using primers Sp24-1Forward and Sp24-1179Reverse (S2 Table). Next, we amplified a truncated portion of zipt-7.1 using primers Sp24-1Forward and C-Sfi-Sp24-1116Reverse, and cloned this fragment into the bait plasmid pBT3-STE. This construct expresses a form of ZIPT-7.1 lacking the last 21 C-terminal amino acids, which includes the final predicted transmembrane domain (S3 Fig). Based on the predicted topology of ZIPT-7.1, this alteration is necessary to localize the ubiquitin tag (which is fused to the C-terminal end of ZIPT-7.1) in the cytoplasm. Details describing the constructs for each prey protein are described elsewhere [62].
The yeast strain NMY51 was transformed with ZIPT-7.1 bait plasmid and selected on synthetic medium lacking leucine. These yeasts were then transformed with prey plasmid and selected on synthetic medium lacking both leucine and tryptophan. Each resulting strain harbored the bait plasmid and one type of prey plasmid; the ability to turn on reporter genes was tested by plating on synthetic medium lacking Trp, Leu, His, and Ade. Yeast cells transformed with bait plasmid and pAI-Alg5, which expresses wild-type Nub (N-terminal ubiquitin), served as a positive control. Yeast cells transformed with bait plasmid and empty prey vectors pPR3-STE or pPR3-N were used as negative controls.
|
10.1371/journal.pcbi.1000585 | Predicting Protein Ligand Binding Sites by Combining Evolutionary Sequence Conservation and 3D Structure | Identifying a protein's functional sites is an important step towards characterizing its molecular function. Numerous structure- and sequence-based methods have been developed for this problem. Here we introduce ConCavity, a small molecule binding site prediction algorithm that integrates evolutionary sequence conservation estimates with structure-based methods for identifying protein surface cavities. In large-scale testing on a diverse set of single- and multi-chain protein structures, we show that ConCavity substantially outperforms existing methods for identifying both 3D ligand binding pockets and individual ligand binding residues. As part of our testing, we perform one of the first direct comparisons of conservation-based and structure-based methods. We find that the two approaches provide largely complementary information, which can be combined to improve upon either approach alone. We also demonstrate that ConCavity has state-of-the-art performance in predicting catalytic sites and drug binding pockets. Overall, the algorithms and analysis presented here significantly improve our ability to identify ligand binding sites and further advance our understanding of the relationship between evolutionary sequence conservation and structural and functional attributes of proteins. Data, source code, and prediction visualizations are available on the ConCavity web site (http://compbio.cs.princeton.edu/concavity/).
| Protein molecules are ubiquitous in the cell; they perform thousands of functions crucial for life. Proteins accomplish nearly all of these functions by interacting with other molecules. These interactions are mediated by specific amino acid positions in the proteins. Knowledge of these “functional sites” is crucial for understanding the molecular mechanisms by which proteins carry out their functions; however, functional sites have not been identified in the vast majority of proteins. Here, we present ConCavity, a computational method that predicts small molecule binding sites in proteins by combining analysis of evolutionary sequence conservation and protein 3D structure. ConCavity provides significant improvement over previous approaches, especially on large, multi-chain proteins. In contrast to earlier methods which only predict entire binding sites, ConCavity makes specific predictions of positions in space that are likely to overlap ligand atoms and of residues that are likely to contact bound ligands. These predictions can be used to aid computational function prediction, to guide experimental protein analysis, and to focus computationally intensive techniques used in drug discovery.
| Proteins' functions are determined to a large degree by their interactions with other molecules. Identifying which residues participate in these interactions is an important component of functionally characterizing a protein. Many computational approaches based on analysis of protein sequences or structures have been developed to predict a variety of protein functional sites, including ligand binding sites [1]–[3], DNA-binding sites [4], catalytic sites [2],[5], protein-protein interaction interfaces (PPIs) [6],[7] and specificity determining positions [8]–[12]. In this paper, we focus on the task of predicting small molecule binding sites from protein sequences and structures. In addition to aiding in the functional characterization of proteins, knowledge of these binding sites can guide the design of inhibitors and antagonists and provide a scaffold for targeted mutations. Over the past 15 years, a large number of methods for predicting small molecule binding sites have been developed. Structural approaches have used geometric and energetic criteria to find concave regions on the protein surface that likely bind ligands [1], [13]–[21]. Sequence-based approaches, on the other hand, have largely exploited sequence conservation, or the tendency of functionally or structurally important sites to accept fewer mutations relative to the rest of the protein [22].
We introduce ConCavity, a new approach for predicting 3D ligand binding pockets and individual ligand binding residues. The ConCavity algorithm directly integrates evolutionary sequence conservation estimates with structure-based surface pocket prediction in a modular three step pipeline. In the first step, we score a grid of points surrounding the protein surface by combining the output of a structure-based pocket finding algorithm (e.g., Ligsite [16], Surfnet [14], or PocketFinder [23]) with the sequence conservation values of nearby residues. In the second step, we extract coherent pockets from the grid using 3D shape analysis algorithms to ensure that the predicted pockets have biologically reasonable shapes and volumes. In the final step, we map from the predicted pockets to nearby residues by assigning high scores to residues near high scoring pocket grid points. Using this pipeline, ConCavity is able to make predictions of both regions in space that are likely to contain ligand atoms as well as protein residues likely to contact bound ligands.
We demonstrate ConCavity's excellent performance via extensive testing and analysis. First, we show that ConCavity, by integrating conservation and structure, provides significant improvement in identifying ligand binding pockets and residues over approaches that use either conservation alone or structure alone; this testing is performed on the diverse, non-redundant LigASite database of biologically relevant binding sites [24]. We find that ConCavity's top predicted residue is in contact with a ligand nearly 80% of the time, while the top prediction of the tested structure-alone and conservation-alone methods is correct in 67% and 57% of proteins respectively. The notable improvement of ConCavity over the conservation-alone approach demonstrates that there is significant added benefit to considering structural information when it is available. Second, we demonstrate that ConCavity significantly outperforms current publicly available methods [1],[19],[25] that identify ligand binding sites based on pocket finding. Third, we show that ConCavity performs similarly when using a variety of pocket detection algorithms [14],[16],[23] or sequence conservation measures [2],[26]. Fourth, we characterize ConCavity in a range of situations, and compare its performance in identifying ligand binding sites from apo vs. holo structures as well as in enzymes vs. non-enzymes. Fifth, we test how well ConCavity can identify catalytic sites and drug binding sites. Sixth, we examine problematic cases for our approaches, and highlight the difficulty that multi-chain proteins pose for structure-based methods for identifying ligand binding sites. Finally, we demonstrate that our methodological improvements in pocket extraction and residue mapping give our implementations of existing methods a significant gain in performance over the previous versions. In fact, without these improvements, the previous structural approaches do not outperform a simple sequence conservation approach when identifying ligand binding residues. Overall, ConCavity significantly advances the state-of-the-art in uncovering ligand binding sites. Our detailed analysis reveals much about the relationship between sequence conservation, structure, and function, and shows that sequence conservation and structure-based attributes provide complementary information about functional importance.
Sequence-based functional site prediction has been dominated by the search for residue positions that show evidence of evolutionary constraint. Amino acid conservation in the columns of a multiple sequence alignment of homologs is the most common source of such estimates (see [22] for a review). Recent approaches that compare alignment column amino acid distributions to a background amino acid distribution outperform many existing conservation measures [2],[27]. However, the success of conservation-based prediction varies based on the type of functional residue sought; sequence conservation has been shown to be strongly correlated with ligand binding and catalytic sites, but less so with residues in protein-protein interfaces (PPIs) [2]. A variety of techniques have been used to incorporate phylogenetic information into sequence-based functional site prediction, e.g., traversing phylogenetic trees [28],[29], statistical rate inference [26], analysis of functional subfamilies [9],[12], and phylogenetic motifs [30]. Recently, evolutionary conservation has been combined with other properties predicted from sequence, e.g., secondary structure and relative solvent accessibility, to identify functional sites [31].
Structure-based methods for functional site prediction seek to identify protein surface regions favorable for interactions. Ligand binding pockets and residues have been a major focus of these methods [1], [13]–[21]. Ligsite [16] and Surfnet [14] identify pockets by seeking points near the protein surface that are surrounded in most directions by the protein. CASTp [17],[19] applies alpha shape theory from computational geometry to detect and measure cavities. In contrast to these geometric approaches, other methods use models of energetics to identify potential binding sites [23], [25], [32]–[34]. Recent algorithms have focused on van der Waals energetics to create grid potential maps around the surface of the protein. PocketFinder [23] uses an aliphatic carbon as the probe, and Q-SiteFinder [25] uses a methyl group. Our work builds upon geometry and energetics based approaches to ligand binding pocket prediction, but it should be noted that there are other structure-based approaches that do no fit in these categories (e.g., Theoretical Microscopic Titration Curves (THEMATICS) [35], binding site similarity [36], phage display libraries [37], and residue interaction graphs [38]). In contrast to sequence-based predictions, structure-based methods often can make predictions both at the level of residues and regions in space that are likely to contain ligands.
Several previous binding site prediction algorithms have considered both sequence and structure. ConSurf [39] provides a visualization of sequence conservation values on the surface of a protein structure, and the recent PatchFinder [40] method automates the prediction of functional surface patches from ConSurf. Spatially clustered residues with high Evolutionary Trace values were found to overlap with functional sites [41], and Panchenko et al. [42] found that averaging sequence conservation across spatially clustered positions provides improvement in functional site identification in certain settings. Several groups have attempted to identify and separate structural and functional constraints on residues [43],[44]. Wang et al. [45] perform logistic regression on three sequence-based properties and predict functional sites by estimating the effect on structural stability of mutations at each position. Though these approaches make use of protein structures, they do not explicitly consider the surface geometry of the protein in prediction. Geometric, chemical, and evolutionary criteria have been used together to define motifs that represent known binding sites for use in protein function prediction [46]. Machine learning algorithms have been applied to features based on sequence and structure [47],[48] to predict catalytic sites [5], [49]–[51] and recently to predict drug targets [52] and a limited set of ligand and ion binding sites [53]–[55]. Sequence conservation has been found to be a dominant predictor in these contexts.
Most similar to ConCavity are two recent approaches to ligand binding site identification that have used evolutionary conservation in a post-processing step to rerank [1] or refine [56] geometry based pocket predictions. In contrast, ConCavity integrates conservation directly into the search for pockets. This allows it to identify pockets that are not found when considering structure alone, and enables straightforward analysis of the relationship between sequence conservation, structural patterns, and functional importance.
For simplicity of exposition, we begin by comparing ConCavity's performance to a representative structural method and a representative conservation method. We use Ligsite+ as the representative structure-based method, and refer to it as “Structure”. Ligsite+ is our implementation (as indicated by superscript “+”) of a popular geometry based surface pocket identification algorithm. We demonstrate in the Methods section that Ligsite+ provides a fair representation of these methods. We choose Jensen-Shannon divergence (JSD) to represent conservation methods and refer to it as “Conservation.” JSD has been previously shown to provide state-of-the-art performance in identifying catalytic sites and ligand binding sites [2]. We have developed three versions of ConCavity that integrate evolutionary conservation into different surface pocket prediction algorithms (Ligsite [16], Surfnet [14], or PocketFinder [23]). When the underlying algorithm is relevant, we refer to these versions as ConCavityL, ConCavityS, and ConCavityP. However, for simplicity, we will use ConCavityL as representative of these approaches and call it “ConCavity.”
ConCavity and Structure produce predictions of ligand binding pockets and residues. The pocket predictions are given as non-zero values on a regular 3D grid that surrounds the protein; the score associated with each grid point represents an estimated likelihood that it overlaps a bound ligand atom. Similarly, each residue in the protein sequence is assigned a score that represents its likelihood of contacting a bound ligand. Conservation only makes residue-level predictions, because it does not consider protein structure. All methods are evaluated on 332 proteins from the non-redundant LigASite 7.0 dataset [24]. To evaluate pocket identification performance, we predict ligand locations on the the holo version of the dataset, in order to use the bound ligands' locations as positives. When evaluating residue predictions, we predict ligand binding residues on the apo structures, and the residues annotated as ligand binding (as derived from the holo structures) are used as positives.
We quantify the overall performance of each method's predictions in two ways. First, for both pocket and residue prediction, we generate precision-recall (PR) curves that reflect the ability of each method's grid and residue scores to identify ligand atoms and ligand binding residues, respectively. (Just as residues are assigned a range of ligand binding scores, grid points in predicted pockets get a range of scores, since there may be more evidence that a ligand is bound in one part of a pocket than another.) Second, for each set of predicted pockets (corresponding to groups of non-zero values in the 3D grid), we consider how well they overlap known ligands via the Jaccard coefficient. The Jaccard coefficient captures the tradeoff between precision and recall by taking the ratio of the intersection of the predicted pocket and the actual ligand over their union. The Jaccard coefficient ranges between zero and one, and a high value implies that the prediction covers the ligand well and has a similar volume. We assess the significance of the difference in performance of methods on the dataset with respect to a given statistic via the Wilcoxon rank-sum test.
Figure 1 compares ConCavity with its constituent structure and conservation based components. Figure 1A shows that, within predicted pockets, grid points with higher scores are more likely to overlap the ligand, and that the significant improvement of ConCavity over Structure (p<2.2e−16) exists across the range of score thresholds. Figure 1B demonstrates that the superior performance of ConCavity holds when predicting ligand binding residues as well (p = 6.80e−13). ConCavity's ability to identify ligand binding residues is striking: across this diverse dataset, the first residue prediction of ConCavity will be in contact with a ligand in nearly 80% of proteins. ConCavity also maintains high precision across the full recall range: precision of 65% at 50% recall and better than 30% when all ligand-binding residues have been identified. As mentioned above, this large improvement exists when predicting ligand locations as well; however, the PR curves illustrate that fully identifying a ligand's position is more difficult for each of the methods than finding all contacting residues.
The ligand overlap statistics presented in Table 1 also demonstrate the superior performance of ConCavity. In nearly 95% of structures, ConCavity's predictions overlap with a bound ligand. Structure's predictions overlap ligands in nearly 92% of the proteins considered. The differences between the methods become more stark when we examine the magnitude of these overlaps. Both ConCavity and Structure predict pockets with total volume (Prediction Vol.) similar to that of all relevant ligands (Ligand Vol.), but ConCavity's pockets overlap a larger fraction of the ligand volume. Thus ConCavity has a significantly higher Jaccard coefficient (p<2.2e−16). This suggests that the integration of sequence conservation with structural pocket identification results in more accurate pockets than when using structural features alone.
Figure 1B also provides a direct comparison of ligand binding site prediction methods based on sequence conservation with those based on structural features. Structure outperforms Conservation, a state-of-the-art method for estimating sequence conservation. Protein residues can be evolutionarily conserved for a number of reasons, so it is not surprising that Conservation identifies many non-ligand-binding residues, and thus, does not perform as well as Structure.
Figures 2 and 3 present pocket and residue predictions of Conservation, Structure, and ConCavity on three example proteins. In general, different types of positions are predicted by Conservation and Structure. If we consider the number of known ligand binding residues for each protein in the dataset, and take this number of top predictions for the Structure and Conservation methods, the overlap is only 26%. The residues predicted by sequence conservation are spread throughout the protein (Figure 2); ligand-binding residues are often very conserved, but many other positions are highly conserved as well due to other functional constraints. In contrast, the structure-based predictions are strongly clustered around surface pockets (Figure 3, left column); many of these residues near pockets are not evolutionarily conserved. However, these features provide largely complementary information about importance for ligand binding. Over the entire dataset, 68% of residues predicted by both Conservation and Structure are in contact with ligands, while only 16% and 43% of those predicted by only conservation or structure respectively are ligand binding. ConCavity takes advantage of this complementarity to achieve its dramatic improvement; it gives high scores to positions that show evidence of both being in a well-formed pocket and being evolutionarily conserved.
The examples of Figures 2 and 3 illustrate this and highlight several common patterns in ConCavity's improved predictions. For 3CWK, a cellular retinoic acid-binding protein, Structure and ConCavity's residue predictions center on the main ligand binding pocket (Figure 3A), while Conservation gives high scores to some positions in the binding site, but also to some unrelated residues (Figure 2A). Looking at the ligand location predictions (green meshes in Figure 3A), Structure and ConCavity both find the pocket, but the signal from conservation enables ConCavity to more accurately trace the ligand's location. This illustrates how the pattern of functional conservation observed at the protein surface influences the shape of the predicted pocket. Ligands often do not completely fill surface pockets; if the contacting residues are conserved, our approach can suggest a more accurate shape.
The results for 2CWH (Figure 3B) and 1G6C (Figure 3C) demonstrate that ConCavity can predict dramatically different sets of pockets than are obtained when considering structure alone. In 2CWH, both methods identify the ligands, but Structure over-predicts the bottom left binding pocket and predicts an additional pocket that does not have a ligand bound. ConCavity traces the ligands more closely and does not predict any additional pockets. Structure performs quite poorly on the tetramer 1G6C: it predicts several pockets that do not bind ligands; it fails to completely identify several ligands; and it misses one ligand entirely. In stark contrast, ConCavity's four predicted pockets each accurately trace a ligand. The incorporation of conservation resulted in the accurate prediction of a pocket in a region where no pocket was predicted using structure alone. Images of predictions for all methods on all proteins in the dataset are available in the Text S1 file, and ConCavity's predictions for all structures in the Protein Quaternary Structure (PQS) database are available online.
We now compare the performance of ConCavity to several existing ligand binding site identification methods with publicly available web servers. LigsiteCS [1] is an updated version of geometry-based Ligsite, and LigsiteCSC [1] is a similar structural method that considers evolutionary conservation information. Q-SiteFinder [25] estimates van der Waals interactions between the protein and a probe in a fashion similar to PocketFinder. CASTp [19] is a geometry-based algorithm for finding pockets based on analysis of the protein's alpha shape. Each of the servers produces a list of predicted pockets represented by sets of residues; however, none of them provide a full 3D representation of a predicted pocket. As a result, we assess their ability to predict ligand binding residues. See the Methods section for more information on the generation and processing of the servers' predictions. In brief, the residues predicted by each server are ranked according to the highest ranking pocket to which they are assigned, i.e., all residues from the first predicted pocket are given a higher score than those from the second and so on. We re-implemented the conservation component of LigsiteCSC, because the conservation-based re-ranking option on the web server did not work for many of the proteins in our dataset. We used JSD as the conservation scoring method.
Figure 4 presents the ligand binding residue PR-curves for each of these methods. ConCavity significantly outperforms LigsiteCS, LigsiteCSC+, Q-SiteFinder, and CASTp (p<2.2e−16 for each). Surprisingly, Conservation is competitive with these structure-based approaches. Several of the servers did not produce predictions for a small subset of the proteins in the database, e.g., the Q-SiteFinder server does not accept proteins with more than 10,000 atoms. Figure 4 is based on 234 proteins from the LigASite dataset for which were able to obtain and evaluate predictions for all methods. Thus the curve for ConCavity is slightly different than those found in the other figures, but its performance does not change significantly.
LigsiteCSC+ is the previous method most similar to ConCavity; it uses sequence conservation to rerank the pockets predicted by LigsiteCS. LigsiteCSC+ provides slight improvement over LigsiteCS, but the improvement is dwarfed by that of ConCavity over Structure (Figure 1). This illustrates the benefit of incorporating conservation information directly into the search for pockets in contrast to using conservation information to post-process predicted pockets.
The poor performance of these previous methods at identifying ligand binding residues is due in part to the fact that they do not distinguish among the residues near a predicted binding pocket. The entire pocket is a useful starting place for analysis, but many residues in a binding pocket will not actually contact the ligand. Knowledge of the specific ligand binding residues is of most interest to researchers. The predictions of our methods reflect this---residues within the same pocket can receive different ligand binding scores. The inability of previous methods to differentiate residues in a pocket from one another is one reason why we elect to use our own implementations of previous structure-based methods as representatives of these approaches in all other comparisons. See the Methods section for more details.
We tested an additional approach for combining sequence conservation with structural information that was suggested by the observation that clusters of conserved residues in 3D often overlap with binding sites [41],[42]. Briefly, the method performs a 3D Gaussian blur of the conservation scores of each residue, and assigns each residue the maximum overlapping value. Thus residues nearby in space to other conserved residues get high scores. This approach improved on considering conservation alone, but was not competitive with ConCavity (Text S1). We also considered the clusters of conserved residues generated by the Evolutionary Trace (ET) Viewer [57]. The clusters defined at 25% protein coverage were ranked by size, and residues within the clusters were ranked by their raw ET score. This approach did not perform as well as the above clustering algorithm (data not shown), and was limited to single chain proteins, because ET returns predictions for only one chain of multi-chain proteins.
In the previous sections, we used ConCavityL, which integrates evolutionary sequence conservation estimates from the Jensen-Shannon divergence (JSD) into Ligsite+, to represent the performance of the ConCavity approach. However, our strategy for combining sequence conservation with structural predictions is general; it can be used with a variety of grid-based surface pocket identification algorithms and conservation estimation methods.
Figure 5 gives PR-curves that demonstrate that ConCavity provides excellent performance whether the structural approaches are based on geometric properties (Ligsite+, Surfnet+) or energetics (PocketFinder+). The significant improvement holds for predicting both ligand locations in space (p<2.2e−16 for each pair) (Figure 5A) and ligand binding residues (p = 6.802e−13 for Ligsite+, p<2.2e−16 for PocketFinder+, p<2.2e−16 for Surfnet+) (Figure 5B). The three ConCavity versions perform similarly despite the variation in performance between Ligiste+, Surfnet+, and Pocketfinder+. In the following sections we will include performance statistics for all three methods when space and clarity allow. When not presented here, results for all methods are available in the supplementary file Text S1.
We have also found that ConCavity achieves similar performance when a different state-of-the-art method [26] is used to score evolutionary sequence conservation (Text S1).
Proteins consisting of multiple subunits generally have more pockets than single-chain proteins due to the gaps that often form between chains. To investigate the effect of structural complexity on performance, we partitioned the dataset according to the number of chains present in the structure predicted by the Protein Quaternary Structure (PQS) server [58] and performed our previous evaluations on the partitioned sets. Figure 6 gives these statistics for ConCavity, Structure, and Conservation. To enable side-by-side comparison, we report the area under the PR curves (PR-AUC) rather than giving the full curves.
As the number of chains in the structure increases, there is a substantial decrease in the performance of Structure. The pattern is seen both when predicting ligand binding residues (Figure 6A) and pockets (Figure 6B, C). This effect is so large that, for proteins with five or more chains, Conservation outperforms Structure. The number of chains in the protein has little effect on Conservation's performance. The performance of Random on proteins with a small number of chains is slightly worse than on proteins with many chains (e.g., Residue PR-AUC for 1 chain: 0.097, 2 chains: 0.110, 3 chains: 0.127, 4 chains: 0.119, 5+ chains: 0.142), so the drop in Structure's performance is not the result of the proportion of positives in each set. These observations emphasize the importance of including multi-chain proteins in the evaluation.
The homo-tetramer 1G6C in Figure 3C provides an illustrative example of the failure of Structure on multi-chain proteins. There is a large gap between the chains in the center of the structure, and several additional pockets are formed at the interface of pairs of contacting chains. As seen in the figure, the large central cavity does not bind a ligand; however, it is the largest pocket predicted by Structure. This is frequently observed among the predictions. While some pockets between protein chains are involved in ligand binding, many of them are not. As the number of chains increases, so does the number of such potentially misleading pockets.
By incorporating sequence conservation information, ConCavity accurately identifies ligand binding pockets in multi-chain proteins. The conservation profile on the surface of 1G6C provides a clear example of this; the pockets that exhibit sequence conservation are those that bind ligands (Figure 2C). 1G6C is not an exception. ConCavity provides significant performance improvement for each partition of the dataset in all three evaluations, and greatly reduces the effect of the large number of non-ligand-binding pockets in multi-chain proteins on performance. ConCavity also provides improvement over Structure on the set of one chain proteins. This is notable because these proteins do not have between-chain gaps, so the improvement comes from tracing ligands and selecting among intra-chain pockets more accurately than using structural information alone (as in Figure 3A).
The binding of a ligand induces conformational changes to a protein [59]. As a result, the 3D structure of the binding site can differ between structures of the same protein with a ligand bound (holo) and not bound (apo). In the holo structures, the relevant side-chains are in conformations that contact the ligand, and this often defines the binding pocket more clearly than in apo structures. To investigate the effect of the additional information provided in holo structures on performance, we evaluated the methods on both sets (Table 2).
As expected, all methods performed better on the holo (bound) structures than the corresponding apo (unbound) structures. However, all previous conclusions hold whether considering apo structures or holo structures; the ranking of the methods is consistent, and the improvement provided by considering conservation is similarly large. PR curves for this comparison are given in the supplementary file Text S1. We will continue to report residue prediction results computed using the apo structures when possible in order to accurately assess the performance of the algorithms in the situation faced by ligand binding site prediction methods in the real world.
The LigASite apo dataset contains protein molecules that carry out a range of different functions. Enzymes are by far the most common; they make up 254 of the 332 proteins in the dataset. The remaining 78 non-enzyme ligand binding proteins are involved in a wide variety of functions, e.g., transport, signaling, nucleic acid binding, and immune system response.
Table 3 compares the performance of the ligand binding site prediction methods on enzymes and non-enzymes. There is more variation within each method's performance on non-enzyme proteins, and all methods perform significantly better on the enzymes (e.g., p = 3.336e−4 for ConCavityL ). Active sites in enzymes are usually found in large clefts on the protein surface and consistently exhibit evolutionary sequence conservation [60],[61], so even though enzymes bind a wide array of substrates, these common features may simplify prediction when compared to the variety of binding mechanisms found in other proteins.
Despite the drop in performance on non-enzyme proteins, the main conclusions from the earlier sections still hold. However, the improvement provided by ConCavity is not as great on the non-enzymes. This could be the result of the more complex patterns of conservation found in non-enzyme proteins, and the comparatively poor performance of Conservation in this setting. It is also possible that Ligsite+'s approach is particularly well suited to identifying binding sites in non-enzymes. Overall, these results highlight the importance of using a diverse dataset to evaluate functional site predictions.
Knowledge of small molecule binding sites is of considerable use in drug discovery and design. Many of the techniques used to screen potential targets, e.g., docking and virtual screening, are computationally intensive and feasible only when focused on a specific region of the protein surface. Structure based surface cavity identification algorithms can guide analysis in such situations [52].
To test ConCavity's ability to identify drug binding sites, we evaluated it on a set of 98 protein-drug complexes [62]. The superior performance provided by ConCavity over Structure on the diverse set of proteins considered above suggests that ConCavity would likely be useful in the drug screening pipeline. Table 4 compares the ligand overlap PR-AUC and Jaccard coefficient for the three versions of ConCavity and their structure-based analogs. Each ConCavity method significantly improves on the methods that only consider structural features (e.g., p = 1.25e−6 on overlap PR-AUC and p = 2.06e−6 on Jaccard for ConCavityL). While the improvement is not quite as large on this dataset as that seen on the more diverse LigASite dataset, it is still significant. It is possible that this is due to the fact that drug compounds are not the proteins' natural ligands; the evolutionary conservation of the residues in binding pockets may reflect the pressures related to binding the actual ligands rather than the drugs.
While ConCavity signficantly outperforms previous approaches, its performance is not flawless. In Figure 7, we give three example structures that illustrate patterns observed when ConCavity performs poorly. Handling these cases is likely to be important for further improvements in ligand binding site prediction.
The first pattern common among these difficult cases is evolutionary sequence conservation information leading predictions away from actual ligand binding sites. Figure 7A provides an example in which the ligand binding site is less conserved than other parts of the protein. The ActR protein from Streptomyces coelicolor (PDB: 3B6A) contains both a small molecule ligand-binding and a DNA-binding domain [63]. The ligand-binding domain is in the bottom, less-conserved half of the structure. The DNA-binding domain is found in the more conserved top half of the given structure. The greater conservation of this domain causes ConCavity to focus on the DNA-binding site over the ligand binding site. In other cases, conservation information is uninformative due to a lack of homologous sequences. Conservation estimates based on low quality sequence alignments may harm performance for some structures, but we have found that they still provide a net performance gain overall (Text S1).
Figure 7 also provides two examples of another difficult case: ligands bound outside of clearly defined, concave surface pockets. In Figure 7B, ConCavity identifies the center of the ring-shaped structure of the pentameric B-subunit of a shiga-like toxin (PDB: 1CQF) as the binding site. This protein binds to glycolipids, like the globotriaosylceramide (Gb3) shown, via a relatively flat interface that surrounds the center of the ring [64]. The center cavity (ConCavity's prediction) is filled by a portion of the A-subunit of the toxin (not included in the structure) which after binding breaks off and enters the host cell. Figure 7C shows the structure of a dimeric noncatalytic carbohydrate binding module (CBM29) from Piromyces equi complexed with mannohexaose (PDB: 1GWL). The carbohydrate ligands bind in long flat clefts on the protein surface [65]. Even though these sites exhibit significant evolutionary conservation, their geometry prevents them from being predicted. Instead, a less conserved pocket formed between the chains is highlighted by ConCavity.
Overall, cases such as these are rare; ConCavity's predictions fail to overlap a ligand in only 5% of structures. In addition, some of these “incorrect” predictions are actually functionally relevant binding sites for other types of interactions as illustrated in Figure 7.
Ligand-binding sites are not the only type of functional site of interest to biologists. A large amount of attention has been given to the problem of identifying catalytic sites. As noted above, the majority of enzyme active sites are found in large clefts on the protein surface, so even though the structural methods considered in this paper were not intended to identify catalytic sites, they could perform well at this task.
Table 5 gives the results of an evaluation of the methods' ability to predict catalytic sites (defined by the Catalytic Site Atlas [66]) in the LigASite apo dataset. Compared to ligand binding site prediction, the relative performance of the methods is different in this context. The ConCavity approach still significantly outperforms the others (p<2.2e−16 for Structure, p = 8.223e−4 for Conservation). Most surprisingly, Conservation significantly outperforms methods based on structure alone (p = 9.863e−3 Ligsite+, p = 4.694e−6 Pocketfinder+, p = 1.171e−6 Surfnet+). All the methods have lower PR-AUC when predicting catalytic sites than predicting ligand-binding residues (e.g., ConCavityL has PR-AUC of 0.315 versus 0.608); this is due in large part to the considerably smaller number of catalytic residues than ligand-binding residues per protein sequences.
These results imply that being very evolutionarily conserved is more indicative of a role in catalysis than being found in a surface pocket. Though catalytic sites are usually found in pockets near bound ligands, there are many fewer catalytic sites per protein than ligand-binding residues. As a result simply searching for residues in pockets identifies many non-catalytic residues. This is consistent with earlier machine learning studies that found conservation to be a dominant predictive feature [5],[49],[50], and it suggests that new structural patterns should be sought to improve the identification of catalytic sites.
Several previous methods have combined sequence conservation and structural properties in machine learning frameworks to predict catalytic sites [5],[50],[51]. Direct comparison with these methods is difficult because most datasets and algorithms are not readily available. Tong et al. [51] compared the precision and recall of several machine learning methods on different datasets in an attempt to develop a qualitative understanding of their relative performance. While it is not prudent to draw conclusions based on cross-dataset comparisons, we note for completeness that ConCavity's catalytic site predictions the diverse LigASite dataset achieve higher precision (23.8%) at full recall than the maximum precision (over all recall levels) reported for methods in their comparisons.
Evolutionary sequence conservation and protein 3D structures have commonly been used to identify functionally important sites; here, we integrate these two approaches in ConCavity, a new algorithm for ligand binding site prediction. By evaluating a range of conservation and structure-based prediction strategies on a large, diverse dataset of ligand binding sites, we establish that structural approaches generally outperform sequence conservation, and that by combining the two, ConCavity outperforms conservation-alone and structure-alone on about 95% and 70% of structures respectively. Overall, ConCavity's first predicted residue contacts a ligand in nearly 80% of the apo structures examined, and it maintains high precision across all recall levels. These results hold for the three variants of ConCavity we considered, each of which uses a different underlying structure-based component. In addition, ConCavity's integrated approach provides significant improvement over conservation and structure-based approaches on the common task of identifying drug binding sites.
Combining sequence conservation-based methods with structure information is especially powerful in the case of multi-meric proteins. Our analysis has shown that the performance of structural approaches for identifying ligand binding sites dramatically decreases as the number of chains in the structure increases; conservation alone outperforms structure-based approaches on proteins with five or more chains. It is difficult to determine from structural attributes alone if a pocket formed at a chain interface binds a ligand or not. However, ligand binding pockets usually exhibit high evolutionary sequence conservation. ConCavity, which takes advantage of this complementary information, performs very well on multi-chain proteins; the presence of many non-ligand binding pockets between chains has little effect on its performance.
While ConCavity outperforms previous approaches, we have found two main causes of poor results: misleading evolutionary sequence conservation information and ligands that bind partially or entirely outside of well-defined concave surface pockets. Ligand binding sites may lack strong conservation for a number of reasons: the underlying sequence alignment may be of low quality, there may be other more conserved functional regions in the protein, and some sites are hypervariable for functional reasons [67]. The alignment quality issue will become less relevant as sequence data coverage and conservation estimation methods improve. The second two cases may require the integration of additional features to better distinguish different types of functional sites. Similarly, finding biologically relevant ligands that bind outside of concave surface pockets will likely require the development of additional structural descriptors. Missing or incomplete ligands also affect the apparent performance of the methods, but such issues are unavoidable due to the nature of the structural data.
In implementing and evaluating previous 3D grid-based ligand binding site prediction approaches, we have found that the methods used both to aggregate grid values into coherent pockets as well as to map these pockets onto surface residues can have a large effect on performance. In order to focus on the improvement provided by considering evolutionary sequence conservation, the results for previous structure-based methods presented above use our new algorithms for these steps. We describe the details of our approaches in the Methods section. On a high level, the new methodologies we propose provide significant improvement by predicting a flexible number of well-formed pockets for each structure and by assigning each residue a likelihood of binding a ligand based on its local environment rather than on the rank of the entire pocket. We have used morphological properties of ligands to guide pocket creation, but the most appropriate algorithms for these steps depend strongly on the nature of the prediction task. These steps have received considerably less attention than computing grid values; our results suggest that they should be given careful consideration in the future.
We have focused on the prediction of ligand binding sites, but the direct synthesis of conservation and structure information is likely to be beneficial for predicting other types of functionally important sites. Our application of ConCavity to catalytic site prediction illustrates the promise and challenges of such an approach. Catalytic sites are usually found in surface pockets, but considering structural evidence alone performs quite poorly---worse than sequence conservation. Combining structure with evolutionary conservation provides a modest gain in performance over conservation alone. Protein-protein interface residues are another appealing target for prediction; much can be learned about a protein by characterizing its interactions with other proteins. However, protein-protein interaction sites provide additional challenges; they are usually large, flat, and often poorly conserved [68]. ConCavity is not appropriate for this task. Other types of functional sites also lack simple attributes that correlate strongly with functional importance. Analysis of these sites' geometries, physical properties, and functional roles will produce more accurate predictors, and may also lead to new insights about the general mechanisms by which proteins accomplish their molecular functions.
In summary, this article significantly advances the state-of-the-art in ligand binding site identification by improving the philosophy, methodology, and evaluation of prediction methods. It also increases our understanding of the relationship between evolutionary sequence conservation, structural attributes of proteins, and functional importance. By making our source code and predictions available online, we hope to establish a platform from which the prediction of functional sites and the integration of sequence and structure data can be investigated further.
This section describes the components of the ConCavity algorithm for predicting ligand binding residues from protein 3D structures and evolutionary sequence conservation.
ConCavity proceeds in three conceptual steps: grid creation, pocket extraction, and residue mapping (Figure 8). First, the structural and evolutionary properties of a given protein are used to create a regular 3D grid surrounding the protein in which the score associated with each grid point represents an estimated likelihood that it overlaps a bound ligand atom (Figure 8A). Second, groups of contiguous, high-scoring grid points are clustered to extract pockets that adhere to given shape and size constraints (Figure 8B). Finally, every protein residue is scored with an estimate of how likely it is to bind to a ligand based on its proximity to extracted pockets (Figure 8C).
Grid-based strategies have been employed by several previous systems for ligand binding site prediction (e.g., [14],[16],[23]). However, our adaptations to the three steps significantly affect the quality of predictions. First, we demonstrate how to integrate evolutionary information directly into the grid creation step for three different grid-based pocket prediction algorithms. Second, we introduce a method that employs mathematical morphology operators to extract well-shaped pockets from a grid. Third, we provide a robust method for mapping grid-based ligand binding predictions to protein residues based on Gaussian blurring. The details of these three methods and an evaluation of their impacts on ligand-binding predictions are described in the following subsections.
We have compared ConCavity to several methods for ligand binding site prediction. Many of these methods lack publicly accessible implementations, and those that are available output different representations of predicted pockets and residues. In this section, we describe of how we generate predictions for all previous methods considered in our evaluations. In some cases we have completely reimplemented strategies and in others we have post-processed the output of existing implementations. Table 8 provides a summary of these details. As mentioned earlier, a “+” appended to the method name indicates that it is (at least in part) our implementation, e.g., Ligsite+.
The prediction methods described in this paper take protein 3D structures and/or multiple sequence alignments as input. Protein structures were downloaded from the Protein Quaternary Structure (PQS) server [58]. Predicted quaternary structures were used (rather than the tertiary structures provided in PDB files) so as to consider pockets and protein-ligand contacts for proteins in their biologically active states. All alignments come from the Homology-derived Secondary Structure of Proteins (HSSP) database [70]. All images of 3D structures were rendered with PyMol [71].
Ligand binding sites as defined by the non-redundant version of the LigASite dataset (v7.0) [24] were used to evaluate method predictions. This set consists of 337 proteins with apo (unbound) structures, each having less than 25% sequence identity with any other protein in the set. Five of the 337 structures were left out of the evaluation: 1P5T, 1YJG, and 3DL3 lacked holo ligand information in the database, and 2PCY and 3EZM, because their corresponding holo structures are not in PQS or HSSP. Each apo structure has at least one associated holo (bound) structure in which biologically relevant ligands are identified in order to define ligand binding residues and map them to the apo structure. If multiple holo structures are available for the protein, the sets of contacting residues are combined to define the binding residues for the apo structure. We select the structures for our LigASite holo evaluation set by taking the holo structure with the most ligand contacting residues for each apo structure. The average number of holo structures for each apo structure is 2.58, and the maximum for any single structure is 32. The average chain length is 276 residues with a minimum of 59 and a maximum of 1023. The average number of positives---sites contacting a biologically relevant ligand---per chain is 25 residues (about 11% of the chain). The apo dataset includes many proteins with multiple chains; the average number of chains per protein is 2.22. The chain distribution is: 1 chain: 143, 2 chains: 112, 3 chains: 18, 4 chains: 35, 5 or more chains: 24.
The drug dataset comes from a set of 100 non-redundant 3D structures selected by [62]. This set contains a diverse set of high-quality structures (resolution <3 Å) with drug or drug-like molecules (molecular weight between 200 and 600, and 1−12 rotatable bonds) bound. Structure 1LY7 has been removed from the PDB, and 1R09 could not be parsed. We consider the 98 remaining structures.
The catalytic site annotations were taken from version 2.2.9 of the Catalytic Site Atlas [66]. There are 153 proteins in the LigASite apo dataset with entries in the Catalytic Site Atlas. These proteins have an average of 3.2 catalytic sites per chain (just over 1% of all residues in the chain).
Predictions of ligand binding pockets are represented by non-zero values in a regular 3D grid around the protein. These represent regions in space thought to contain ligands. These predictions are evaluated in two ways: on the pocket level by computing their overlap with known ligands, and on the grid level by analyzing how well the grid scores rank grid points that overlap ligand atoms. We use a grid with rasterized van der Waals spheres for ligand atoms from the PQS structure as the “positive” set of grid points. From this, we calculate the intersection and union of the actual ligand atoms and the predictions. We compare methods using the over-prediction factor (Prediction Volume/Ligand Volume), precision (Intersection Volume/Prediction Volume), recall (Intersection Volume/Ligand Volume), and Jaccard coefficient (Intersection Volume/Union Volume).
We also create precision-recall (PR) curves, which compare precision (TP/(TP + FP)) on the y-axis with recall (TP/(TP + FN)) on the x-axis, to evaluate the ability of each method to predict whether a ligand atom is present at a grid point. We consider grid points that overlap a ligand atom as positives. To construct the PR curve, we calculate the precision and recall at each cutoff of the grid values in the pocket prediction grid. To summarize the performance of each method, we construct a composite PR curve [72] by averaging the precision at each recall level for each structure in the dataset. As a reference point, we include the performance of a random classifier averaged over all the structures as well. The expected performance of a random method is the number of positives over the number of all grid points. The method and code of Davis and Goadrich [73] is used to calculate the area under the PR curve (PR-AUC). The significance of the difference between methods is assessed using the Wilcoxon signed-rank test over paired performance statistics for all structures in the dataset. The significance of the difference in performance of a single method on different datasets is calculated with the Wilcoxon rank-sum test.
For the residue-based evaluation, we consider how well each method's residue scores identify ligand binding residues. Positives are those residues in contact with a ligand as defined by LigASite database. PR curves were made by calculating, for each chain, the precision and recall at each position on the ranked list of residue scores. Composite PR curves were computed as described for the grid point evaluation, but curves were first averaged over the chains in a structure and then over structures. PR curves were constructed similarly for the catalytic site analysis, but positives were defined as those residues listed in the Catalytic Site Atlas.
|
10.1371/journal.pgen.1006696 | Lethality of mice bearing a knockout of the Ngly1-gene is partially rescued by the additional deletion of the Engase gene | The cytoplasmic peptide:N-glycanase (Ngly1 in mammals) is a de-N-glycosylating enzyme that is highly conserved among eukaryotes. It was recently reported that subjects harboring mutations in the NGLY1 gene exhibited severe systemic symptoms (NGLY1-deficiency). While the enzyme obviously has a critical role in mammals, its precise function remains unclear. In this study, we analyzed Ngly1-deficient mice and found that they are embryonic lethal in C57BL/6 background. Surprisingly, the additional deletion of the gene encoding endo-β-N-acetylglucosaminidase (Engase), which is another de-N-glycosylating enzyme but leaves a single GlcNAc at glycosylated Asn residues, resulted in the partial rescue of the lethality of the Ngly1-deficient mice. Additionally, we also found that a change in the genetic background of C57BL/6 mice, produced by crossing the mice with an outbred mouse strain (ICR) could partially rescue the embryonic lethality of Ngly1-deficient mice. Viable Ngly1-deficient mice in a C57BL/6 and ICR mixed background, however, showed a very severe phenotype reminiscent of the symptoms of NGLY1-deficiency subjects. Again, many of those defects were strongly suppressed by the additional deletion of Engase in the C57BL/6 and ICR mixed background. The defects observed in Ngly1/Engase-deficient mice (C57BL/6 background) and Ngly1-deficient mice (C57BL/6 and ICR mixed background) closely resembled some of the symptoms of patients with an NGLY1-deficiency. These observations strongly suggest that the Ngly1- or Ngly1/Engase-deficient mice could serve as a valuable animal model for studies related to the pathogenesis of the NGLY1-deficiency, and that cytoplasmic ENGase represents one of the potential therapeutic targets for this genetic disorder.
| Ngly1 is a cytoplasmic de-N-glycosylating enzyme that is ubiquitously found in eukaryotes. This enzyme is involved in a process referred to as endoplasmic reticulum-associated degradation (ERAD), one of the quality control mechanisms for newly synthesized proteins. A genetic disorder, NGLY1-deficiency, caused by mutations in the NGLY1 gene has recently been discovered. However, the precise mechanism for the pathogenesis of this devastating disease continues to remain unclear. We report herein that Ngly1-deficient mice are embryonically lethal in a C57BL/6 background. Surprisingly, the lethality was suppressed by crossing the mice with an outbred mouse strain (ICR), suggesting that the phenotypic consequence of Ngly1 is greatly influenced by their genetic background. In both cases, the additional deletion of Engase in Ngly1-deficient mice could strongly mitigate the phenotypes. Interestingly, the remaining defects in Ngly1-deficient or Ngly1/Engase-deficient mice were reminiscent of the symptoms of subjects with an NGLY1-deficiency. Our results clearly point to the importance of Ngly1 in mammals and show that the inhibition of ENGase represents an effective therapy for treating an NGLY1-deficiency. Most importantly, the mice described herein could serve as valuable viable model mice for studies related to the pathophysiology of an NGLY1-deficiency.
| The endoplasmic reticulum (ER) is the organelle responsible for the biosynthesis of proteins that pass through the secretory pathway. This organelle has an efficient protein quality control or homeostasis system, in which abnormal or excess proteins are targeted for destruction [1–3]. In such systems, cytoplasmic proteasomes play a major role in degrading proteins, and the degradation system is often referred to as ER-associated degradation, or ERAD in short. Ngly1 is a highly conserved deglycosylating enzyme that is involved in the ERAD process [4–6]. Ngly1 cleaves N-glycans from misfolded glycoproteins during the ERAD process, and is thought to play an important role in the efficient degradation of misfolded glycoproteins [7–11]. N-glycans released by Ngly1 are known to be processed by cytoplasmic glycosidases such as ENGase and the α-mannosidase, Man2C1 (Fig 1A, upper scheme) [12–14]. Cytoplasmic ENGase is another deglycosylating enzyme that acts directly on N-glycans that are attached to glycoproteins but leaves a single N-acetylglucosamine (GlcNAc) residue attached to the protein (Fig 1A, lower scheme). This enzyme belongs to the Glycosyl Hydrolase (GH) family 85, and while conserved widely in eukaryotes, some yeast, including S. cerevisiae and S. pombe, do not produce this enzyme.
Several studies dealing with the biological functions of Ngly1 and/or ENGase have been reported. Mutants of Ngly1 and its orthologues have been analyzed in various organisms [7, 15–19]. Interestingly, the phenotypic consequences were found to be quite distinct between species; for example, budding yeast show no significant viability/growth defects [7], while mutant flies exhibited severe growth delay/arrest [15]. The molecular details behind the phenotypes, however, remain poorly understood. As of this writing, mutants of Engase have been analyzed in A. thaliana [20] and C. elegans [21] but, in both cases, no significant phenotypes were observed.
In 2012, the first patient harboring mutations in NGLY1 alleles (NGLY1-deficiency) was identified through an exome analysis [22] and, since then, clinical cases involving humans have been reported [23–26]. These patients show very severe systemic symptoms such as delayed global development, movement disorders, hypotonia and hypo/alacrima [24, 26]. These reports point to the biological significance of Ngly1 in the normal development of mammals. More recently, using a model ERAD substrate, we reported that the ablation of Ngly1 causes a disruption in the ERAD process in mouse embryonic fibroblast (MEF) cells [27]. Interestingly, this ERAD disruption was found to be caused by an unexpected deglycosylating activity of ENGase, and the direct action of this enzyme towards the model substrate was shown to result in the formation of aggregation-prone N-GlcNAc proteins [27]. Moreover, the disruption of ERAD in Ngly1-deficient cells was restored by the additional deletion of the Engase gene. While this result using a model substrate suggests that an ENGase inhibitor could be a potential therapeutic target for treating an NGLY1-deficiency, the issue of how Engase-deletion affects mice phenotypes lacking Ngly1 remains unknown.
The goal of this study was to clarify the details of the biological function of the deglycosylating enzymes, Ngly1 and ENGase, at the individual level in mice. To this end, we used mice (C57BL/6 background mice) that had been generated in a previous study, in which the Ngly1 and Engase genes had been knocked out [27]. The knockout constructs of Ngly1 and Engase are shown in Fig 1B and 1C, respectively. While the generation of these KO mice has been described previously [27], a detailed phenotypic analysis of those mice has not been reported. The targeted genomic disruption of Ngly1 and ENGase was confirmed by PCR using 8 sets of primers (Fig 1B and 1C; S1 Table for primer sequences). The loss of Ngly1 activity was confirmed by an activity assay [27]. We further examined the expression of the Ngly1 protein by western blot analysis using cytoplasmic fractions from MEF cells. As shown in Fig 1D, the loss of Ngly1 in MEF cells from Ngly1−/− mice was confirmed. In the case of ENGase, we confirmed the loss of the ENGase activity through an activity assay in a previous study [27] as well as by conducting a detailed structural analysis of the free oligosaccharides (fOSs) in the cytoplasm of MEF cells [28]. The Ngly1 heterozygous (Ngly1−/+) mice were fertile and did not show any obviously recognizable phenotypes. However, viable homozygous Ngly1−/− pups were not produced, despite the repeated crossing of the Ngly1−/+ mice (Table 1), suggesting that the deletion of the Ngly1 allele results in a lethal condition in C57BL/6 mice. To delineate the timing of the lethality, Ngly1−/+ mice were crossed and embryos were collected at several stages of gestation. The viability of collected embryos was confirmed by checking their heart beat and their genotypes were analyzed using genomic DNA extracted from the amnion. The results of this genotyping are summarized in Table 1. The Ngly1−/− embryos were viable, even at embryonic day 18.5 (E18.5), one day prior to birth. At the same time, however, about 30% of Ngly1−/− embryos were inviable at later stages of development (E17.5–18.5). When embryos were collected early in the morning of the day of birth, only the Ngly1−/+ and Ngly1+/+ embryos were alive when we revived them by gentle massaging, but no Ngly1−/− mice could be revived. We also analyzed the genotype of pups within a few hours after their birth (P0) and confirmed the absence of Ngly1−/− pups (Table 1). At E14.5 and E16.5, however, only viable Ngly1−/− embryos were observed. Therefore, the lethality caused by the Ngly1 deficiency appears to occur between E16.5 and before birth.
To investigate the defects in Ngly1−/− embryos in more detail, X-ray micro-computed tomography (μ-CT) analyses were carried out on Ngly1−/− or Ngly1+/+ embryos at E16.5 (Fig 2A and 2B). As shown in Fig 2B, the Ngly1−/− embryos showed a ventricular septal defect (VSD) (5 out of 5 embryos (5/5)). Histological analyses also confirmed the occurrence of a VSD in Ngly1−/− embryos (3/3) (Fig 2C and 2D). VSD is one of the most frequently-observed cardiovascular phenotypes in embryonic/perinatal lethal mice [29]. We also found that some Ngly1−/− embryos showed anemia (12/28 [42.86%], Fig 2E, left panel) or edema (4/28 [14.29%], Fig 2F, left panel), which were not observed in Ngly1+/+ embryos (0/44, Fig 2E and 2F, right panel).
In sharp contrast to the case of Ngly1−/− mice, the Engase−/− mice showed normal behavior/values in several tests as follows: behavior test (open field), morphology/behavioral/sensory test (RIKEN modified-SHIRPA), hematology/clinical chemistry test (hematology, urinalysis, clinical blood chemistry), pathology test (body weight), cardiovascular test (blood pressure, electrocardiogram) and neurology/psychiatry test (light/dark transition, home-cage activity, tail suspension, hot plate, tail flick) (the number of tested mice are described in Materials and Methods). According to our previous cell-based study using an ERAD model substrate [27], we hypothesized that ENGase could have the ability to function as a deglycosylating enzyme and that its deletion could rescue the defects caused by the lack of Ngly1. In this study, we attempted to verify the effect of the additional Engase deletion of Ngly1−/− in mice at an individual level. To this end, we crossed Ngly1−/+;Engase−/+ or Ngly1−/+;Engase−/− mice and found that surviving Ngly1−/− mice were produced upon the crossing (Table 2). Upon further examination, all of the surviving mice were found to be Ngly1−/−;Engase−/− double-knockout mice, strongly indicating that the additional deletion of Engase partially rescued the lethality caused by the defect of Ngly1 at an individual level.
We next investigated the issue of whether VSD was present in Ngly1−/−;Engase−/− embryos at E16.5. If VSD was to be critical for the lethality of Ngly1−/− embryos, then this phenotype should be suppressed by the additional deletion of Engase. As we expected, VSD was not observed in the Ngly1−/−;Engase−/− embryos, as evidenced by μ-CT (0/2) and histological analyses (0/2) (Fig 2G and 2H). It is therefore possible that VSD is at least one of the critical phenotypes that cause the lethality of Ngly1−/− mice/embryos and the deletion of the Engase gene, for unknown reasons, rescues this phenotype. We also examined the gross morphology of embryos at E16.5 and found that some Ngly1−/−;Engase−/− embryos showed anemia (5/16 [31.25%], Fig 2I, left panel) while no Ngly1+/+;Engase−/− embryo showed this phenotype (0/20, Fig 2I, right panel). Additionally, Ngly1−/−;Engase−/− embryos did not show edema (0/16). Unexpectedly, there is no significant difference of the appearance ratio between Ngly1−/− embryos and Ngly1−/−;Engase−/− embryos at E16.5 (Table 1). These results suggest that Engase deletion resulted in the rescue of not all but several phenotypes caused by Ngly1 deletion such as VSD and edema. These results also indicated that multiple pathways appear to be involved in the pathophysiology of Ngly1−/− mice.
We next analyzed the phenotypes of the viable Ngly1−/−;Engase−/− mice in C57BL/6 background. When the mice were tested for viability, an apparent decrease in the viability of the Ngly1−/−;Engase−/− mice was observed after 45 weeks (about 10 months), suggesting that they show age-related defects (Fig 3A). We then attempted to characterize the major defects that were observed in Ngly1−/−;Engase−/− mice. When they were young (about 4~5 weeks of age), the differences between the Ngly1−/−;Engase−/− mice and their littermates were negligible. However, when they became older (about 6~7 months of age), the Ngly1−/−;Engase−/− mice showed several defects/abnormal behavior, such as bent spines (Fig 3B, upper panel, 15/15), trembling (S1 and S2 Movies, 15/15), hind-limb clasping (Fig 3C, S3 Movie, 15/15), and front-limb shaking (S3 Movie, 15/15). Moreover, the Ngly1−/−;Engase−/− mice showed a significant reduction in body weight compared to their littermates (Fig 3D and 3E). We currently have no explanation for why only male Ngly1−/+;Engase−/− mice show a slow weight gain relative to Ngly1+/+;Engase−/− mice, but it may, in part, be due to the fact that male mice are larger than female mice and therefore make a change in weight gain could be detected more easily. We observed no other haploinsufficient effects in male/female Ngly1−/+;Engase−/− mice.
Ngly1−/−;Engase−/− mice also had coarse fur (12/15 [80%]) which may be due to either intrinsic skin problems or, alternatively, impaired grooming behavior (Fig 3F). Female Ngly1−/−;Engase−/− mice exhibited eye opacity much more frequently than their littermates after 20~30 weeks of age (8/10 [80%], Fig 3G). Those results clearly indicate that, while the deletion of the Engase gene results in suppressing the critical phenotypes, including the VSD caused by deletion of the Ngly1 gene, not all of the defects were fully rescued.
It was previously shown that a change in the genetic background of Fut8−/− mice from an inbred C57BL/6 strain to an outbred ICR strain, that maintain a certain genetic variation within the colonies, postnatal lethality was rescued [30]. It is also noteworthy that, when we examined the symptoms of NGLY1-deficiency patients, there were a wide spectrum of symptoms and a genotype-phenotype correlation was not so obvious [23–25]. These results imply that the genetic background may influence the severity of the phenotypes caused by the defects of Ngly1. With that in mind, we crossed Ngly1−/+ mice in a C57BL/6 background with ICR mice. Among the F1 mice that were produced, the Ngly1−/+ mice were further crossed to produce F2 mice. As shown in Table 3, we found that viable F2 Ngly1−/− mice were obtained from these crosses. The phenotypes of the Ngly1−/− mice in the C57BL/6 and ICR mixed background were found to be much more severe compared with the Ngly1−/−;Engase−/− mice in the C57BL/6 background, and about 70% of them died within 3 weeks (Fig 4A). The Ngly1−/− mice in the C57BL/6 and ICR mixed background also showed a significant reduction in body weight (Fig 4B and 4C) and hind-limb clasping (Fig 4D, 9/9), which were also observed in aged Ngly1−/−;Engase−/− mice in the C57BL/6 background (Fig 3C–3E). It should be noted that the deletion of the Engase gene resulted in a slight reduction in the body weight in mice of the C57BL/6 and ICR mixed background (ex. compare Fig 3B, 3C, 3D and 3E for Ngly1−/+ mice). This observation suggests that the loss of the Engase gene affects weight gain in mice. We suspect, however, that this fluctuation in the data could be simply due to the diverse genetic background of the mice that were analyzed, especially since ICR mice are, in general, much larger than C57BL/6 mice. Consistent with this assumption, pathological tests at the Japan mice clinic showed that the average weight of 26-week old Engase−/− mice (average 26.64/26.61 g for male/female mice (n = 5)) was not affected when compared with that of wild-type mice (26.79/26.64 g for male/female mice (n = 1)) with the C57BL/6 background. However, we currently cannot exclude the possibility that the loss of the Engase gene may become problematic when Ngly1 levels are compromised.
In the case of the Ngly1−/− mice in the C57BL/6 and ICR mixed background, however, they began to show these abnormalities when they were still young (4 weeks old). Therefore, the phenotype of the Ngly1−/− mice in the C57BL/6 and ICR mixed background turned out to be much more severe than that of Ngly1−/−;Engase−/− mice in the C57BL/6 background, even when they partially evaded embryonic lethality. These results also indicate the suppressive effects of the deletion of Engase for phenotypes of Ngly1−/− mice, since mice with the C57BL/6 and ICR mixed background are generally thought to be more tolerant to genetic mutations when compared with the inbred C57BL/6 background.
To determine if the partial suppression of lethality in the ICR and C57BL/6 mixed background may have been caused by the low ENGase activity in ICR mice, we carried out ENGase enzyme assays using PA-labeled sugar derivatives on liver tissue obtained from ICR and C57BL/6 mice. The findings indicated that the samples from the ICR and C57BL/6 mice had equivalent ENGase activity (0.39±0.11 pmol/min/mg protein for C57BL/6 and 0.42±0.07pmol/min/mg protein for ICR), and no ENGase activity was detected in the liver of Engase−/− mouse. These results indicate that the rescue of the lethality of Ngly1−/− mice resulting from a change in their genetic background does not appear to be correlated with ENGase activity.
To test the effect of Engase deletion in the viable Ngly1−/− mice, we next generated Ngly1−/−;Engase−/− mice in the C57BL/6 and ICR mixed background. As shown in Table 3, we were able to obtain Ngly1−/−;Engase−/− mice in the F3 mice by crossing the F2 Ngly1−/+;Engase−/− mice in the C57BL/6 and ICR mixed background. All of the F3 Ngly1−/−;Engase−/− mice were alive at 30 weeks (n = 9), in sharp contrast to the case with the F2 Ngly1−/− mice in the C57BL/6 and ICR mixed background, where none of them had survived at 30 weeks (n = 29) (Fig 4A). Moreover, the Ngly1−/−;Engase−/− mice showed an attenuated body weight reduction compared to the Ngly1−/− mice in the C57BL/6 and ICR mixed background (Fig 4E and 4F). These results clearly indicate that the disruption of ENGase can alleviate, at least partially, the defects caused by Ngly1 deletion in the viable mice (e.g. reduction of % survival and body weight). The phenotypes of the Ngly1−/−;Engase−/− mice from this cross were found to be even more mild than the Ngly1−/−;Engase−/− mice in the C57BL/6 background, in that they did not show any overt defects, other than hind-limb clasping (9/9), suggesting that the deletion of the Engase gene and mixing with ICR-background have an additive effect in suppressing the defect of the Ngly1−/− mice.
Regarding the potential mechanism for how ENGase can rescue the defects due to the loss of Ngly1, we previously proposed an “N-GlcNAc hypothesis” [27]. In this proposed process, ENGase acts on cytoplasmic misfolded glycoproteins in a stochastic fashion, especially when Ngly1 activity is compromised. As a result, N-GlcNAc modified proteins (proteins that are modified with one GlcNAc unit) would be generated in the cytoplasm. The formation of the excessive levels of N-GlcNAc proteins could somehow lead to detrimental effects (Fig 5A). We also previously showed that there appears to be higher levels of cytoplasmic N-glycoproteins in Ngly1−/−;Engase−/− MEF cells compared with wild-type MEF cells, as judged by lectin blot experiments [27]. Such cytoplasmic N-glycoproteins may serve as precursors for endogenous N-GlcNAc proteins, some of which may be linked to the pathogenesis of an NGLY1-deficiency, in Ngly1−/− cells.
To identify specific N-glycoproteins that accumulate in Ngly1−/−;Engase−/− MEF cells, a glycoproteomics analysis was carried out on the cytoplasm of Ngly1−/−;Engase−/− MEF cells. To minimize contamination from the membrane fraction, we utilized a mild digitonin-treatment to recover the cytoplasmic fraction [31]. The cytoplasmic fractions thus prepared were subjected to a glycoproteomics analysis. After trypsinization, the N-glycopeptides were concentrated by acetone precipitation [32], and the N-glycans were released by PNGase F digestion, followed by the detection of deglycosylated peptide in which N-glycosylated asparagines are converted into aspartic acids. Among the 79 potential N-glycopeptides that were detected (S2 and S3 Tables), 13 N-glycopeptides were detected as deamidated peptides, even in the absence of PNGase F-treatment (S3 Table), indicating that those 13 glycopeptides are formed by deamidation of the asparagine residues in a non-enzymatic fashion. The presence of N-glycans on the remaining 66 peptides (S2 Table) was further confirmed by the detection of signals for N-GlcNAc peptide ions on LC/MS/MS analysis of PNGase F-untreated sample [32]. Through this analysis, 18 N-glycopeptides derived from 12 glycoproteins were unequivocally identified in the Ngly1−/−;Engase−/− cytoplasm (Table 4, S1 Fig for product ion spectra). A comparison of the MS peak areas of these glycopeptides between Ngly1−/−;Engase−/− and wild-type revealed that larger signals were detected for several N-glycopeptides in the Ngly1−/−;Engase−/− cytoplasm samples (Table 4, Fig 5B), while our analysis is by no means quantitative and the difference in the amount of N-glycoproteins in the two different cells may also be due to the differential recovery of the N-glycoproteins during the concentration of the N-glycoproteins. Nevertheless, our data clearly indicates that, indeed, the cytoplasm did contain N-glycoproteins, which may become potential substrates for Ngly1/ENGase in the cytoplasm and therefore can serve as precursors for N-GlcNAc proteins.
In this study, the functional significance of the Ngly1 gene in mice were characterized. The results are consistent with the recent reports on NGLY1-deficient human subjects bearing mutations in the NGLY1 allele [22–26]. Surprisingly, the lethality of the Ngly1−/− mice was partially suppressed by the additional knockout of the Engase gene. This result is consistent with findings reported in an earlier study [27], which showed that the disruption of ERAD in Ngly1-KO cells was restored by the additional deletion of Engase, at an individual level. Through analyses of various mutants for Ngly1 gene orthologues, an enzyme-independent function of the Ngly1-orthologue was proposed for some organisms [15, 18, 33]. The results reported herein clearly suggest, however, that the embryonic lethal phenotype of Ngly1−/− mice is mainly related to its deglycosylation-dependent function, given the fact that the lethality was partially suppressed by the additional deletion of Engase, a gene encoding another cytoplasmic deglycosylating enzyme.
Regarding the ventricular septum defect observed for Ngly1−/− embryos, it should be noted that no significant heart problems have been reported for human patients [22–26], suggesting that the functional importance of Ngly1 for the heart may be limited to its embryonic developmental stage. In this connection, it is also noteworthy that one of the patients was diagnosed as “small drop out in interventricular septum” by a fetal echo analysis (personal communication from Mr. Andreas Thermann).
While the lethality of the Ngly1−/− mice was partially rescued by the additional deletion of Engase or by changing the genetic background, recognizable phenotypes were still observed in the Ngly1−/−;Engase−/− mice, as well as the Ngly1−/− mice in the C57BL/6 and ICR mixed background. It should be noted here that most of the phenotypes observed for those mice are reminiscent of the symptoms observed for an NGLY1-deficiency in human subjects (S4 Table). Since the Ngly1−/− mice are embryonic lethal in the C57BL/6 background, the Ngly1−/−;Engase−/− mice as well as the Ngly1−/− mice in the C57BL/6 and ICR mixed background could serve as viable model animals for studying the functions of Ngly1, which could be of great value, especially for evaluating the efficacy of potential drugs for the treatment of NGLY1-deficiency subjects in the future.
With respect to the potential mechanism responsible for the detrimental effect of ENGase in Ngly1−/− mice, we previously proposed an N-GlcNAc hypothesis as described above [27]. Such N-GlcNAc proteins were recently reported to be present in murine synaptosome proteins [34]. In addition, in plants, at least one of the N-GlcNAc proteins was shown to be generated by the action of the cytoplasmic ENGase [35]. There was also a report suggesting that the proximal N-GlcNAc on N-glycosylation sites has a critical role in stabilizing carrier proteins [36]. We therefore hypothesize that ENGase might cleave glycans on misfolded glycoproteins in the cytoplasm to generate N-GlcNAc proteins, and that the presence of an excess of these N-GlcNAc proteins somehow contributes to the detrimental effects in mice, such as N-GlcNAc proteins forming aggregates, or impairment of cytoplasmic O-GlcNAc signaling (N-GlcNAc hypothesis) (Fig 5A). Indeed, we found that a model ERAD substrate was converted into an N-GlcNAc-modified protein in Ngly1−/− MEF cells [27]. In this connection, it should be noted that a GalNAc-GalNAc (GalNAc: N-acetylgalactosamine) interaction was recently reported to occur in the binding of mucins with Tn (GalNAc α1-Ser/Thr) or sialyl Tn cancer antigens [37]. It would be interesting to see if similar interactions, at least in part, contribute to the aggregation of N-GlcNAc proteins. On the other hand, we have not been able to detect Ngly1-deficient specific change in the O-GlcNAc modification among MEF cells by western blotting (S2 Fig). However, the possibility that changes in specific O-GlcNAc proteins had occurred cannot be excluded. Clearly, a more comprehensive analysis will be required. Moreover, Ngly1-deficient specific protein aggregation was not observed by histological analyses (e.g. Congo red staining, PAS staining) of tissue sections from mice embryos. Further studies will be needed to verify our N-GlcNAc hypothesis and efforts are currently underway to identify endogenous N-GlcNAc proteins in MEF cells derived from various genotypes (wild-type, Ngly1−/−, Engase−/− and Ngly1−/−;Engase−/−). It should be noted that Ngly1-KO does not lead to a general ERAD-defect [8, 9, 38]. These observations strongly indicate that the pathogenesis of Ngly1−/− mice cannot be attributed to general ER stress, but rather specific substrates likely appear to be involved.
We previously reported that larger amounts of N-glycoproteins appeared to be accumulated in the cytoplasm of Ngly1−/−;Engase−/− MEF cells compared with those of wild-type, Ngly1−/− or Engase−/− MEF cells [27]. To extend our analysis, in this study we identified several N-glycoproteins that had accumulated in the cytoplasm of Ngly1−/−;Engase−/− cells. Such glycoproteins can serve as precursors for endogenous N-GlcNAc proteins that are potentially formed in Ngly1−/− cells. While we used sequential detergent extraction for the cytoplasmic fraction to minimize contamination of other organelles, traces of lysosomal components could still be detected in the cytoplasmic fraction (S3 Fig). It is therefore possible that some glycoproteins of lysosomal origin may be present due to contamination by lysosomes in the cytoplasmic fraction.
In any event, the phenotypes observed for Ngly1−/− and Ngly1−/−;Engase−/− mice were found to be very analogous to the symptoms observed in patients with an NGLY1-deficiency (S4 Table), which make those mice valuable animal models for this genetic disorder. It should also be noted that the degree of suppression by Egnase-deletion are distinct among phenotypes (ex Fig 2). These observations clearly suggest that multiple pathways, i.e. ENGase-dependent and–independent pathways, are involved in the pathogenesis of Ngly1-KO mice or NGLY1-deficiency subjects.
Since many of the Ngly1−/−;Engase−/− mice in both the C57BL/6 background and the C57BL/6 and ICR mixed background survive for more than 1 year after birth (one lived for more than 2 years), this either suggests that (1) once they escape the developmental defect during embryogenesis, the Ngly1−/− mice could have a relatively normal life span or (2) the effect of suppression by Engase-deletion is quite effective for adult mice. However, the Ngly1−/− mice in the C57BL/6 and ICR mixed background showed quite severe phenotypes, favoring the latter possibility. Moreover, the additional deletion of Engase in the C57BL/6 and ICR background also resulted in a significant improvement of the phenotypes (% survival, body weight gain), further supporting the conclusion that a defect in the Engase-gene has strong positive effects for viable mice. In this connection, it should be noted that, while the birth rate of Ngly1−/−;Engase−/− mice from Ngly1−/+;Engase−/− cross was found to be better in the C57BL/6 and ICR mixed background (12.33%) than the C57BL/6 homogeneous background (8.98%) (Tables 2 and 3), it was still significantly lower than the expected rate (25%). These results clearly suggest that the deletion of Engase is not sufficient to fully suppress the Ngly1 defect during embryogenesis. Further studies will be required to understand the molecular details on the functional importance of Ngly1 in embryogenesis.
Given the dramatic rescue of the mice phenotypes by the additional Engase deletion, one could easily envision that an in vivo inhibitor of cytoplasmic ENGase would be a promising therapeutic approach for the treatment of an NGLY1-deficiency in humans (Fig 5A). It has previously been shown that the thiazoline modified oligosaccharide functions as an effective inhibitor for the cytoplasmic ENGase [39], and it may be possible to design cell permeable inhibitors for ENGase derived from this compound.
All animal experiments were approved by the institutional committee of RIKEN (approved number is: H28-2-003(2)). All mice in this study were euthanized by cervical dislocation.
We generated Ngly1- or Engase-deficient mice as described previously [27]. Mice were housed in a facility with access to food and water and were maintained under 12-hour light/12-hour dark cycle.
Engase−/− mice analyses were performed at Japan Mouse Clinic (http://ja.brc.riken.jp/lab/jmc/mouse_clinic/en/index.html). For open field test, 18 Engase+/+ mice, 21 Engase−/+ mice and 20 Engase−/− mice were used. For RIKEN modified-SHIRPA, hematology, urinalysis, clinical blood chemistry, pathology, blood pressure, and electrocardiogram tests, 11 Engase+/+ mice, 14 Engase−/+ mice and 14 Engase−/− mice were used. For light/dark transition, home-cage activity, tail suspension, hot plate, and tail flick tests, 7 Engase+/+ mice, 7 Engase−/+ mice and 6 Engase−/− mice were used.
MEF cells were isolated from E13.5 or E14.5 embryos by trypsinization [40]. The isolated MEF cells were cultivated in DMEM (4.5 g/L glucose, 10% FBS, 100 unit/mL penicillin and 100 μg/mL streptomycin) at 37°C in a 5% CO2 atmosphere.
Genotyping to confirm the targeted disruption of Ngly1 and ENGase was performed by PCR analysis using genomic DNA extracted from the tail or amnion and 8 sets of primers, as shown in S1 Table. PCR products were separated by 2(w/v)% agarose gel electrophoresis, and the band size was analyzed. Primer set 1–2, 3–4, 5–6, 7–8 can detect Ngly1+/+, Ngly1−/−, Engase+/+, Engase−/− allele respectively. PCR product size (band size) of each primer sets are shown in S1 Table.
Embryos were fixed in Bouin’s solution. Fixed embryos were washed with 70% ethanol until the yellow color of picric acid was no longer visible. The washed embryos were dehydrated through a series of ethanol solutions (70%, 80%, 90%, 95%, 99% and 100%), cleared with xylene, and embedded in paraffin. Samples were sectioned at 5 μm and stained with hematoxylin and eosin (HE staining) using the general protocol.
The collected embryos were fixed with a 10% neutral buffered formalin solution (SIGMA). Fixed embryos were soaked in contrast agent, a 1:3 mixture of Lugol solution and deionized distilled water and then analyzed by μ-CT. μ-CT analyses were performed as previously reported [41]. Briefly, mouse embryos were scanned using a SCANXMATE-E090 scanner (Comscantechno) at a tube voltage peak of 40 kVp and a tube current of 100 μA. Samples were rotated 360° in steps of 0.24°, generating 1500 projection images. The μ-CT data were reconstructed at an isotropic resolution of 10.5 μm. Three-dimentional, tomographic images were obtained using the OsiriX software (www.osirix-viewer.com).
An anti-mouse Ngly1 antibody was raised against the E. coli expressed N-terminus fragment of Ngly1 using pET-mPng1N [42], which express the (His)6-tagged N-terminal fragment of Ngly1 (aa1-181). In brief, to a 500 mL BL21(DE3)pLysS suspension of growing-phase cells carrying pET-mPng1N, was added isopropyl-β-thiogalactoside (final conc. 1 mM) at 37°C for 3 hours. The cells were then collected, suspended in 25 mL of lysis buffer (50 mM Tris-HCl buffer, pH 8.0, with 1 mM Pefabloc (Roche) and 0.15 M NaCl) with 200 μg/mL lysosyme and incubated on ice for 30 min. Cells were lysed by sonication, and a crude cell extract was obtained by removing the debris. A 12 ml portion of the protein extract (equivalent to 240 ml E. coli culture) was applied to a Profinia protein purification system with a Bio-Scale Mini Ni-IMAC Profinity column (Bio-Rad) according to the manufacturer’s protocol. The purified (His)6-tagged protein thus obtained (2.5 mL) was desalted by passage through a PD-10 column (GE), and the purified protein fraction thus obtained (~4 mg/mL) was used as an immunogen. The antiserum raised against the purified Ngly1 (aa1-181), as well as the affinity purified IgG fraction using the antigen-conjugated resin, was prepared by Biogate Co., Ltd (Gifu, Japan).
Cultured MEF cells were washed with PBS 3 times. The following steps were carried out either on ice or at 4°C. Cells collected were suspended with 10 mM Tris-HCl buffer (pH7.5) containing 1mM ethylenediaminetetraacetic acid (EDTA), 250 mM sucrose, 1 mM dithiothreitol (DTT), 1 mM AEBSF (Pefabloc SC, Roche) and various protease inhibitors (1 × complete protease inhibitor cocktail (Roche)) at a density of 5 × 107 cell/mL, and were homogenized using a Potter-Elvehjem homogenizer. The extracts thus obtained were cleared at 15,000 rpm for 10 min at 4°C and the supernatant was transferred to a new tube. The sample was ultracentrifuged at 100,000×g for 1 h at 4°C, and the supernatant was collected and used as the cytoplasmic fraction.
2 × 106 MEF cells were seeded on a 150 cm2 flask 48 h before extraction. Cultured MEF cells were washed 3 times with 10 mL of PBS. The flask was gently coated with 1 mL of permeabilization buffer (25 mM Hepes (pH 7.2), 110 mM potassium acetate, 2.5 mM magnesium acetate, 1 mM ethyleneglycoltetraacetic acid (EGTA), 1 mM DTT, 1 mM phenylmethylsulfonyl fluoride (PMSF), 0.015(wt/vol)%, digitonin) and slowly rocked at 4°C for 5 min. The buffer was collected and centrifuged at 7,500×g for 10 min at 4°C. The supernatant was transferred to a new tube and used as cytoplasmic fraction for glycoproteomics analyses. The flask was then washed with 5 mL of wash buffer (25 mM Hepes (pH 7.2), 110 mM potassium acetate, 2.5 mM magnesium acetate, 1 mM EGTA, 1 mM DTT, 1 mM PMSF, 0.004(wt/vol)%, digitonin). The flask was coated 1 mL of lysis buffer (25 mM Hepes (pH 7.2), 400 mM potassium acetate, 15 mM magnesium acetate, 1 mM DTT, 1 mM PMSF, 1(wt/vol)% NP-40, 0.5(wt/vol)% sodium deoxycholate) and rocked at 4 for 5 min. The buffer was collected and centrifuged at 7,500×g for 10 min at 4°C. The supernatant was transferred to a new tube and used as membrane fraction.
The sample preparation were performed as previously described [32]. Briefly, following the addition of 4 μL of 1 M DTT, the detergent extracted cytoplasmic fractions (100 μL, 60 μg) were incubated at 65°C for 30 min. 5.6 μL of 1 M iodoacetoamide was added and the mixtures were incubated at room temperature for 40 min in the dark. The reaction was stopped by adding 2.4 μL of 1 M DTT, and then desalted by acetone precipitation prior to tryptic digestion. The precipitated samples were dissolved in 30 μL of 50 mM Tris-HCl (pH 8.5) containing 0.1(w/v)% sodium deoxycholate. Modified trypsin (3 μg) (Promega) was added to the samples, and the mixtures were incubated at 37°C for 16 h. The peptide mixtures were diluted with Milli-Q water to a final volume of 50 μL. A five-fold volume of ice cold acetone was then added to the peptides and the glycopeptide mixtures were incubated at −25°C for at least 16 h to precipitate the glycopeptides (glycopeptides enrichment). The precipitated glycopeptide fractions were collected by centrifugation at 12,000×g for 10 min at 20°C, and the precipitated glycopeptides fractions were dissolved in Milli-Q water. A part of the glycopeptides fractions were desalted using a Pierce C-18 Spin Column (Thermo Scientific) following the manufacturer’s instructions and used for glycopeptide analyses. A second portion of the glycopeptide fraction was dissolved in 50 μL of 50 mM phosphate buffer (pH 7.2) containing 10 mM EDTA and used for analyses of deaminated peptides. To release the N-glycans, the sample was incubated with 1 U of PNGase F with 50% glycerol at 37°C overnight. After drying with a SpeedVac concentrator, the sample was desalted as described above. PNGase F treated samples were used for analyses of deamidated peptides to estimate the N-glycosylation sites on peptides.
The samples prepared were dissolved in 20 μL of 0.1% formic acid, and a 1 μL aliquot of the sample was injected into an EASY-nLC 1000 (Thermo Scientific) with an Acclaim PepMap 100 trapping column (75 μm×2 cm, nanoViper; Thermo Scientific) and a Nano HPLC Capillary Column (75 μm×120 mm, 3 μm, C18; Nikkyo Technos). The mobile phases were 0.1% formic acid (A buffer) and 0.1% formic acid in acetonitrile (B buffer). The glycopeptides were eluted at a flow rate of 0.3 μL/min with a linear gradient from 0 to 45% B over 55 min. Mass spectra were acquired on a Q Exactive mass spectrometer (Thermo Scientific) equipped with Nanospray Flex Ion Source (Thermo Scientific). The electrospray voltage was 2.0 kV, and the resolution was 70,000. Full mass scans were performed at a range of m/z 700–2,000, and product ion spectra were acquired data-dependently in the positive ion mode. The deglycosylated peptides were analyzed in a similar manner with the exception of full mass scans at m/z 350–2000.
The deglycosylated (PNGase F treated) peptides were identified by a database search analysis using the SEQUEST search engine (Thermo Scientific) with applying the following parameters: a specified trypsin enzymatic cleavage with 2 possible missed cleavages, precursor mass tolerance of 6 ppm, fragment mass tolerance of 0.02 Da, static modification of cysteine (carbamidomethylation), and dynamic modification of methionine (oxidation) and asparagine (deamidation). All of the identified peptides were then filtered at a false discovery rate (FDR) threshold of 1%. The amino acid sequences of glycopeptides were identified by matching between the product ion of peptide carrying a single N-acetylglucosamine residue and the identified deglycosylated peptide.
Cytoplasmic fractions of MEF cells derived from Ngly1−/−, or wild type mice were separated by sodium dodecylsulfate (SDS)-polyacrylamide gel electrophoresis (PAGE) and electroblotted onto polyvinylidene fluoride (PVDF) membrane. The membrane was blocked for 1 h with 4(w/v)% skim milk in PBS and incubated with anti-mouse Ngly1 antibody in PBS supplemented with 0.05% (wt/vol) Tween20 (PBST) (1:3000) at 4°C for 16 h, followed by HRP goat anti-rabbit IgG (1:3000, GE Healthcare) in PBST at room temperature for 2 h. The membranes were monitored by an Immobilon Western Chemiluminescent HRP Substrate (Millipore). The chemiluminescence was visualized using a LAS-3000 Imager (Fujifilm).
Livers obtained from 7-week-old female mice of ICR background (n = 3), and C57BL/6 background (n = 3), as well as a 16-week-old female C57BL/6 mouse of Engase−/− (n = 1) were suspended in 2 volumes of cytoplasm extraction buffer containing 10 mM Hepes-NaOH (pH7.4), 250 mM sucrose, 1 mM DTT, 1 mM ethylenediaminetetraacetic acid (EDTA), 1×complete protease inhibitior cocktail (EDTA-free; Roche) and 1 mM Pefabloc (Roche), and homogenized on ice using a Potter-Elvehjem homogenizer. The homogenate was centrifuged twice at 10,000xg for 5 min at 4°C to remove debris, and then further centrifuged at 100,000xg for 60 min at 4°C. The supernatant was collected, not including the floating lipid layer, as the cytosolic extract for the ENGase activity assay. The amount of protein was determined using a Bradford assay (Bio-Rad Protein Assay, Bio-Rad) with bovine serum albumin as the standard.
The ENGase activity assay reaction (total 10 μL) included 2 μL of cytosolic fraction in which the protein concentration was adjusted to 30 mg/mL), 4 pmol of Man9GlcNAc2-PA (TaKaRa Bio Inc.) in a final 50 mM of Mes-NaOH buffer (pH6.0), and the reaction was carried at 30°C for 1 hour. The reaction was terminated by adding 100 μL 75% (vol/vol) ethanol and the resulting solution was incubated on ice for 15 min. After centrifugation at 15,000×g for 10 min at 4°C, the supernatant was evaporated to dryness and dissolved in 40 μL of water, and was subjected to HPLC analysis using a Shodex NH2P-50 4E column as described previously [43].
|
10.1371/journal.ppat.1006878 | The fungal pathogen Magnaporthe oryzae suppresses innate immunity by modulating a host potassium channel | Potassium (K+) is required by plants for growth and development, and also contributes to immunity against pathogens. However, it has not been established whether pathogens modulate host K+ signaling pathways to enhance virulence and subvert host immunity. Here, we show that the effector protein AvrPiz-t from the rice blast pathogen Magnaporthe oryzae targets a K+ channel to subvert plant immunity. AvrPiz-t interacts with the rice plasma-membrane-localized K+ channel protein OsAKT1 and specifically suppresses the OsAKT1-mediated K+ currents. Genetic and phenotypic analyses show that loss of OsAKT1 leads to decreased K+ content and reduced resistance against M. oryzae. Strikingly, AvrPiz-t interferes with the association of OsAKT1 with its upstream regulator, the cytoplasmic kinase OsCIPK23, which also plays a positive role in K+ absorption and resistance to M. oryzae. Furthermore, we show a direct correlation between blast disease resistance and external K+ status in rice plants. Together, our data present a novel mechanism by which a pathogen suppresses plant host immunity by modulating a host K+ channel.
| Plant nutritional status can greatly influence plant immunity in response to pathogen invasion. Rice blast, a devastating rice disease caused by the hemibiotrophic fungus Magnaporthe oryzae, causes a significant reduction in yield and affects food security. In this study, we demonstrate that the M. oryzae secreted protein AvrPiz-t interacts with rice OsAKT1, a potassium (K+) channel protein, and suppresses OsAKT1-mediated inward K+ currents, possibly by competing with the OsAKT1 upstream regulator, OsCIPK23. We also show that both OsAKT1 and OsCIPK23 are required for K+ uptake and resistance against M. oryzae infection in rice. This study provides new insights into the molecular basis of pathogen-mediated perturbation of a plant nutrition pathway.
| Potassium (K+) plays important roles in many fundamental processes in plants, including enzyme activation, cellular homeostasis, membrane transport, osmoregulation and immunoreaction [1, 2]. The uptake and translocation of K+ in plants relies on a number of K+ channels and transporters. In the model plant Arabidopsis (Arabidopsis thaliana), the K+ channel AKT1 and the K+ transporter HAK5 have been reported to mediate most of the K+ absorption [3]. Similar to the regulation of AKT1 in Arabidopsis, the activity of its homolog in rice (Oryza sativa), OsAKT1, is regulated by a protein complex comprising OsCBL1 (calcineurin B-like protein 1) and the cytosolic protein kinase OsCIPK23 (CBL-interacting protein kinase 23) [4], while OsHAK5 is mainly regulated at the transcription level [5].
In agricultural production, the application of potassium fertilizer has been reported to decrease the incidence of plant diseases [2, 6]. For example, the foliar application of potassium chloride (KCl) reduces damage caused by Septoria tritici on wheat (Triticum aestivum) in field experiments [7], and the application of KCl to K-deficient soils increases rice resistance to stem rot and aggregate sheath spot [8]. Although previous studies have investigated the effect of K+ nutrition on disease development in plants [2], it is not known how host K+ nutrition reduces pathogen virulence and enhances host immunity.
Adapted plant pathogens secrete effectors into the apoplast of their host or deliver them inside the host cells to promote infection [9, 10]. These molecules can alter plant processes and target a wide range of host proteins, including important components in pathogen-associated molecular pattern (PAMP)-triggered immunity (PTI), effector-triggered immunity (ETI), vesicle trafficking, autophagy, chloroplast and mitochondrial functions, sugar transport, phytoalexin production, etc. [11–16]. The mechanisms underlying the effector-mediated suppression of host immunity have been extensively studied over the last two decades; however, to date no effector has been reported to manipulate K+ transport pathways in plants.
The hemibiotrophic fungus, Magnaporthe oryzae, causes rice blast disease in all rice-growing countries [17–19] and also causes wheat blast in South America and Bangladesh [20–22]. We previously identified an M. oryzae effector protein, AvrPiz-t, which functions as a virulence factor and increases blast susceptibility of rice in the absence of the blast resistance (R) protein Piz-t [23, 24]. Yeast two-hybrid (Y2H) screening of a rice cDNA library revealed that AvrPiz-t interacted with 12 APIPs (AvrPiz-t interacting proteins) in rice [24]. Of these, APIP6 and APIP10, two RING-type E3 ligase proteins, have been shown to ubiquitinate and degrade AvrPiz-t, accompanied by the degradation of these two E3 ligases [24, 25]. APIP6 and APIP10 are positive regulators of PTI, while APIP10 acts as a negative regulator of Piz-t accumulation, thereby modulating ETI [24, 25]. Thus, the targeting of different rice E3 ligases by AvrPiz-t to suppress rice immunity involves a multilayered strategy [26]. Recently, we reported that the bZIP-type transcription factor, APIP5, can also be bound by AvrPiz-t [27]. AvrPiz-t suppresses the transcription and protein accumulation of APIP5 at the necrotrophic stage; however, APIP5 also interacts with the R protein Piz-t, thereby stabilizing APIP5 and preventing effector-triggered necrosis [27].
Here, we report that AvrPiz-t interacts with the rice K+ channel protein OsAKT1 and suppresses OsAKT1-mediated inward K+ current. AvrPiz-t competes with a cytoplasmic kinase, OsCIPK23, for binding to OsAKT1. Both OsAKT1 and OsCIPK23 serve dual functions on K+ absorption and blast disease resistance. In addition, we demonstrate the positive effect of K+ on rice blast resistance. Based on these results, we propose a working model wherein the M. oryzae effector AvrPiz-t suppresses rice immunity by interfering with K+ signaling components that are important for both K+ absorption and host resistance.
The rice OsAKT1 (APIP7) was one of the 12 APIPs identified in our previous study [24]. OsAKT1 has been identified as an inward-rectifying K+ channel and play important roles in K+ uptake [4, 28]. OsAKT1 is a typical Shaker family K+ channel protein with 6 transmembrane domains (TMs), a putative cyclic nucleotide binding domain (cNMP) and 5 ankyrin repeats (ANKs) in the intracellular domain (S1A Fig). The P-loop between the fifth and sixth TM domains contains a TxxTxGYG motif which is the hall-mark of K+-selective channels [29]. OsAKT1 also shares high similarities with its orthologs from other plant species [4].
In this study, we first confirmed the interaction between AvrPiz-t and OsAKT1 in yeast. In a Y2H assay, AvrPiz-t (without the signal peptide) interacted with a truncated OsAKT1-C1 fragment (corresponding to amino acids (aa) 607–935 of the full-length protein) derived from a rice cDNA library (Fig 1A). Although we did not detect an interaction between OsAKT1-C (aa 341–935, containing the full intracellular domain) and AvrPiz-t in yeast (S1A and S1B Fig), in vitro GST-pull down and in vivo luciferase complementation assays demonstrated that both OsAKT1-C1 and OsAKT1-C can bind to AvrPiz-t (Fig 1B and 1C). We then performed a co-immunoprecipitation (Co-IP) assay by transiently expressing OsAKT1-C-HA or OsAKT1-C1-HA with AvrPiz-t-DsRed in Nicotiana benthamiana expression system, and confirmed that OsAKT1-C and -C1 fragments interact with AvrPiz-t in planta (Fig 1D).
To identify the region of OsAKT1 responsible for the interaction, we performed a Y2H analysis with AvrPiz-t and various truncated versions of an OsAKT1 C-terminal protein fragment. This revealed that the third ANK domain was required for the interaction (S1A and S1B Fig). In addition, we tested the interaction between OsAKT1-C1 and a set of AvrPiz-t point mutants and found that the C70A mutant protein showed a weakened interaction with OsAKT1-C1 (S1C Fig). This point mutation is also important for the binding of AvrPiz-t to APIP5 [27], indicating the important role of this residue in AvrPiz-t’s interactions with rice targets. Collectively, these results demonstrate that AvrPiz-t directly interacts with the OsAKT1 intracellular domain both in vitro and in vivo.
OsAKT1 can mediate inward K+ currents in HEK293 cells [4, 28]. To determine whether AvrPiz-t can affect OsAKT1-mediated inward K+ currents, we co-expressed OsAKT1 as a fusion protein with green fluorescent protein (OsAKT1-GFP), together with AvrPiz-t in HEK293 cells. The full-length OsAKT1 CDS was fused to the N terminus of GFP under the control of the human cytomegalovirus (CMV) immediate early promoter (S2A Fig). Because AvrPiz-t is a small effector protein (91aa, without the signal peptide) [23], we expressed it under the CMV promoter without adding a tag that might affect the biochemical function of AvrPiz-t. To monitor the transfection efficiency, we cloned the fragment of the red fluorescent protein (DsRed) in the same vector under the control of the eukaryotic translation elongation factor 1α (EF-1α) promoter (S2A Fig). The red fluorescent signals were used as a transfection marker in the electrophysiology assays. The protein expression of OsAKT1-GFP and DsRed in the HEK293 cells was determined with immunoblot analysis (S2B Fig).
The above assays showed that OsAKT1-GFP alone, or OsAKT1-GFP expressed with the DsRed mediated strong inward K+ currents (Fig 2A, left panel); however, the K+ currents were substantially suppressed by co-expression with AvrPiz-t or the full-length AvrPiz-t (FLAvrPiz-t) (Fig 2A, top two images of the right panel). In contrast, the combination of AvrPiz-t and DsRed, or FLAvrPiz-t and DsRed, did not mediate any K+ currents (Fig 2A, third and fourth images of the right panel). After co-transfection with AvrPiz-t or FLAvrPiz-t, the currents mediated by OsAKT1 decreased from -698±68 pA/pF to -146±14 pA/pF and -111±9 pA/pF, respectively, at -200 mV (Fig 2B). However, the AvrPiz-t and FLAvrPiz-t proteins only inhibited the conductance of OsAKT1, but not the voltage dependence of OsAKT1 (Fig 2C).
To confirm that AvrPiz-t specifically suppresses the OsAKT1-mediated inward K+ currents, we used AvrPii, another M. oryzae effector protein [30], which did not interact with OsAKT1-C1 or OsAKT1-C in Y2H analysis (S3A Fig), in the electrophysiology experiment. We co-expressed AvrPii (without the signal peptide) and the full-length AvrPii (FLAvrPii) with OsAKT1 in HEK293 cells (S2A and S2B Fig), and found neither AvrPii nor FLAvrPii significantly affected the OsAKT1-mediated inward K+ currents (Fig 2D, S3B and S3C Fig). These data suggest that AvrPiz-t specifically suppresses OsAKT1-mediated inward K+ currents and that this inhibition is dependent on the direct interaction between the two proteins.
A previous study showed that ectopic expression of AvrPiz-t in transgenic rice impairs blast resistance and PAMP induced production of reactive oxygen species (ROS) [24]. To investigate whether AvrPiz-t affects the K+ absorption in rice, we performed a K+-depletion assay and found that AvrPiz-t transgenic plants exhibited weaker K+ uptake than segregated wild-type (sWT) plants (S4A Fig). K+ content analysis showed that AvrPiz-t transgenic plants displayed reduced K+ level in 0.1 mM K+ solution compared to the sWT, while this difference was disappeared when the external K+ concentration was elevated to 1.0 mM, suggesting that AvrPiz-t inhibition effect could be restored by the increase of K+ supply (S4B Fig). We then measured the plant net K+ fluxes using non-invasive micro-test technology (NMT) in the primary root meristems. Compared with the sWT plants, the net K+ influx of AvrPiz-t transgenic seedlings was clearly lower through out of the 10-min interval measurement when supplied with 0.1 mM K+ (S4C Fig, upper panel). However, when the external K+ concentration was increased to 1.0 mM, no significant difference was detected (S4C Fig, bottom panel), which is consistent with the K+ content analysis. Taken together, these data indicate that AvrPiz-t can partially inhibit the K+ uptake in rice plants.
OsAKT1 has been reported to play central roles in K+ uptake in rice [4]. To determine whether OsAKT1 contributes resistance to M. oryzae, we first analyzed its expression pattern in Nipponbare (NPB) plants that were infected by a compatible M. oryzae isolate. qRT-PCR analysis showed OsAKT1 expression was highly induced at 24-hour post inoculation (hpi) and continued to increase until at least 120 hpi (S5A Fig). To assess the potential role of OsAKT1 in rice blast resistance, we identified a T-DNA insertion knock-out mutant of OsAKT1 in the Dongjin (DJ) background (S5B and S5C Fig). K+ content analysis showed that the osakt1 mutant accumulated lower levels of K+ in roots and shoots compared with DJ plants (Fig 3A). We then punch-inoculated osakt1 and DJ plants with a compatible M. oryzae isolate. Ten days after inoculation, the osakt1 mutant plants showed larger disease lesions and more fungal biomass than DJ (Fig 3B and 3C).
To confirm the increased susceptibility phenotype of the osakt1 mutant, we generated OsAKT1 RNA interference (RNAi) plants in NPB background. T1 plants with a single T-DNA insertion and significantly reduced OsAKT1 expression (Fig 3D) were selected for blast inoculation. Because OsAKT1 has a homologue (LOC_Os07g07910) in the rice genome that shares 70% identity with OsAKT1 [28], we tested the expression of this gene in the OsAKT1 RNAi lines to determine the silencing specificity. The qRT-PCR results indicated that the expression of LOC_Os07g07910 was not significantly affected in the OsAKT1 RNAi lines, indicating the specificity of the RNAi fragment (S6 Fig). Similar to the osakt1 mutant, the OsAKT1 RNAi lines showed a significant decrease in K+ content in roots and shoots (Fig 3E), as well as enhanced susceptibility to M. oryzae, with increased fungal biomass in the lesion area compared with the sWT plants (Fig 3F and 3G). Together, these results demonstrate that OsAKT1 plays a positive role in rice immunity to M. oryzae, likely by modulating K+ absorption.
AvrPiz-t can promote the degradation of its target proteins APIP6 and APIP10 [24, 25]. To investigate whether AvrPiz-t affects OsAKT1 protein stability, we co-expressed AvrPiz-t with OsAKT1 intracellular fragments or the full-length OsAKT1 in rice protoplasts. However, immunoblot analysis showed that AvrPiz-t did not obviously affect the stability of OsAKT1-C, OsAKT1-C1 and the full-length OsAKT1 proteins (S7A–S7C Fig), indicating a different regulation mechanism from APIP6 and APIP10.
OsAKT1-mediated K+ uptake is regulated by the OsCIPK23 complex [4]. Since a previous study showed that CBL10 can compete with CIPK23 for binding to AKT1 and thus negatively regulates AKT1 activity in Arabidopsis [31], we attempted to determine whether AvrPiz-t interferes with the association of OsAKT1 with OsCIPK23 in a similar manner. We first confirmed the interaction between OsAKT1 and OsCIPK23 by luciferase complementation and Co-IP assays in N. benthamiana. The assay showed that both OsAKT1-C and OsAKT1-C1 interacted with OsCIPK23 while AvrPiz-t did not (S8A and S8B Fig). To ascertain whether AvrPiz-t interferes with the OsAKT1-OsCIPK23 interaction, we included AvrPiz-t-DsRed in the luciferase complementation assay. Compared with the DsRed control combination, the relative luciferase activity significantly decreased when AvrPiz-t-DsRed was co-expressed with OsAKT1-C or -C1 and OsCIPK23 (Fig 4A, lanes 2 to 5). This result was further confirmed in a competitive Co-IP assay. We transiently co-expressed a same amount of OsAKT1-C-HA and CLuc-OsCIPK23 in N. benthamiana leaves with an increasing amount of AvrPiz-t-DsRed by adding different concentrations of Agrobacteria carrying AvrPiz-t-DsRed. Immunoblot analysis showed that, as AvrPiz-t levels were increased (Fig 4B, second panel from the bottom), the immunoprecipitated OsCIPK23 protein levels were decreased significantly (Fig 4B, third panel from the bottom). Next, we performed a pull-down assay to establish whether AvrPiz-t interferes with the OsCIPK23-OsAKT1 association in vitro. The results showed that GST-OsAKT1-C bound to both MBP-AvrPiz-t-HA and MBP-OsCIPK23-cMyc at the same time; however, with increasing amounts of MBP-AvrPiz-t-HA (Fig 4C, panels 2 and 5), the enrichment of MBP-OsCIPK23-cMyc gradually decreased to 20% compared with the control (Fig 4C, panel 1, lanes 2 to 6). As the controls, GST-OsAKT1-C did not pull down MBP-AvrPii-HA and increasing the amount of MBP-AvrPii-HA protein did not obviously affect the protein levels of retrieved MBP-OsCIPK23-cMyc compared with MBP-AvrPiz-t-HA (S9 Fig). These results demonstrate that AvrPiz-t specifically interferes with the OsAKT1-OsCIPK23 association both in vivo and in vitro.
The interference of AvrPiz-t with the OsAKT1-OsCIPK23 interaction prompted us to investigate the function of OsCIPK23 in rice immunity. We measured the expression of OsCIPK23 during M. oryzae infection and found that its transcription increased at 72 hpi (S10A Fig). Then we identified a T-DNA insertion mutant of OsCIPK23 in the DJ background (S10B and S10C Fig). The K+ content of oscipk23 mutant roots and shoots was much lower than that in DJ plants (Fig 5A), which is similar to OsCIPK23 RNAi plants [4]. A punch inoculation experiment showed that oscipk23 mutant plants were more susceptible to a compatible M. oryzae isolate (Fig 5B), showing increased fungal growth in the lesion area compared with the DJ plants (Fig 5C).
To confirm the phenotype of oscipk23, we generated gene-editing plants of OsCIPK23 in the NPB background with clustered regularly interspaced short palindromic repeat (CRISPR) and CRISPR associated protein 9 (Cas9) system [32, 33]. We designed two single-guide RNAs (sgRNAs) in one construct which targets two different sites followed with protospacer adjacent motifs (PAMs) in the first exon of OsCIPK23 (S11A Fig). Based on PCR and sequencing analysis, we selected three homozygous lines with different mutation types to test their resistance to M. oryzae (S11B Fig). Consistent with the T-DNA insertion mutant, the punch inoculation showed all these three lines displayed reduced resistance to the compatible M. oryzae isolate (S11C and S11D Fig).
Since AvrPiz-t weakens both K+ uptake (S4A Fig) and ROS accumulation in rice [24], we performed similar K+ depletion and PAMP-induced ROS detection assays using osakt1 and oscipk23 mutant plants. The results indicated that both osakt1 and oscipk23 mutant plants showed reduced K+ uptake and ROS burst after treatment with the PAMP chitin compared with DJ plants (Fig 5D and 5E). Taken together, these results suggest that, similar to OsAKT1, OsCIPK23 positively regulates immunity to M. oryzae and K+ absorption in rice.
As shown above, loss function of OsAKT1 and OsCIPK23 has decreased rice resistance to M. oryzae, and ectopic expression of AvrPiz-t has impaired the K+ uptake and reduced blast resistance, indicating that K+ has a positive role in rice immunity against M. oryzae. To further investigate the effect of K+ on rice immunity, we used a compatible M. oryzae isolate to inoculate NPB seedlings which were grown in 1 mM or 5 mM K+ solution to mimic normal and high K+ levels [5, 34], respectively. Five days after inoculation, we observed a significant decrease in disease symptoms on the rice plants cultivated in 5 mM K+ compared with those grown in 1 mM K+ (Fig 6A and 6B). We measured the K+ content in the whole plants and found that the K+ content is ~ 20% higher in the 5 mM K+ plants than in the 1 mM K+ plants (Fig 6C). In addition, the basal H2O2 levels as well as two defense-related genes, OsPR1a and WRKY45, were elevated in the plants cultivated in the higher K+ conditions (Fig 6D and 6E). Furthermore, we observed increased dry weight and shoot lengths in the plants cultivated in the 5 mM K+ solution (Fig 6F and 6G), suggesting a positive function of K+ in regulating plant development.
To test the direct effect of K+ on M. oryzae growth, we cultured two isolates on the complete medium that contained additional 40, 80, 120, 150 and 200 mM K+ according to the estimated K+ concentration in plant cytosol (ranged from 100–200 mM [35]). The colony diameters were measured 14 days after inoculation. The results showed that with increasing of K+ concentration, the fungus growth was gradually inhibited (S12A and S12B Fig). Taken together, these results indicate that relative high external K+ concentration promotes plant growth and blast resistance; however, the enriched K+ in plant cells may has a negative effect on fungal growth.
AKT1 is one of the major K+ uptake components in Arabidopsis [36]. The electrophysiological activity of AKT1 is positively regulated by CBL1/9-CIPK23 complexes [36], but suppressed by the 2C-type protein phosphatase AIP1 [37], the Shaker family α-subunit AtKC1[38], and CBL10 [31]. The Arabidopsis AKT1 homolog in rice, OsAKT1, has also been characterized as an important potassium channel for K+ absorption [4]. However, the AKT1 gene in both plants has not previously been associated with immunity to pathogens. In this study, we show that the M. oryzae effector AvrPiz-t interacts with OsAKT1 and suppresses OsAKT1-mediated inward K+ currents (Figs 1 and 2), thereby likely interfering with the K+ signaling pathway. Genetic analysis showed that loss of OsAKT1 function led to reduced K+ content, K+ uptake and enhanced susceptibility to M. oryzae (Figs 3 and 5D). Consistently, we observed a decrease of K+ uptake and K+ content in the AvrPiz-t transgenic plants at lower K+ levels (S4 Fig). We also found that elevated K+ concentrations in rice tissues can enhance resistance against M. oryzae (Fig 6A–6C) and high K+ concentrations in culture medium can inhibit M. oryzae growth (S12 Fig), suggesting that rice may absorb more K+ for immune activations and fungal growth inhibition during M. oryzae infection. It has been reported that the bacterial pathogen Xanthomonas oryzae overcomes rice defenses by regulating the redistribution of the micromutrient copper through activation of XA13, COPT1 and COPT5 [39]. Our study has revealed a new strategy by M. oryzae to interfere the OsAKT1 complex to modulate the uptake of the macronutrient K+ in favor of fungal pathogenesis in rice. However, it is still not fully clear how high K+ level in the host enhances immunity against M. oryzae.
In Arabidopsis, CBL1/9 interacts with CIPK23 and recruits it to the plasma membrane, where CIPK23 phosphorylates and activates AKT1 to mediate K+ absorption [36, 40]. As CBL proteins act as Ca2+ sensors in plant cells, the activation of AKT1 by CBL1/CBL9-CIPK23 is reported to be in a Ca2+-dependent manner [40]. As a second messenger, the cytosolic free Ca2+ concentration ([Ca2+]cyt) is elevated upon PAMP (e.g. flg22, elf18 and chitin) treatments [3, 41, 42]. Although the elevated [Ca2+]cyt is accompanied with rapid membrane depolarization and K+ efflux at the earlier stage [43, 44], the activation of CBL proteins by accumulated [Ca2+]cyt may induce K+ acquisition at the late stage. The observation of induction of OsAKT1 and OsCIPK23 at the later stage of M. oryzae infection may further strengthen this hypothesis (S5A and S10A Figs). CIPK23 is not only a master regulator of AKT1, but also regulates the activity of the K+ transporter HAK5, the NO3- sensor CHL1 and the NH4+ transporters AMT1;1 and AMT1;2 [45–47]. Thus, CBL1/9-CIPK23 cascade plays key roles in regulating the K+, NO3- and NH4+ homeostasis in Arabidopsis [47]. The activation of CIPKs relies on the direct interaction of the self-inhibitory NAF motif with a particular CBL protein [48–51]. The rice OsAKT1 protein acting alone can mediate inward K+ currents in HEK293 cells, while OsCBL1 and OsCIPK23 further enhance the strength of the K+ currents [4]. Physiological analyses have demonstrated that both OsAKT1 and OsCIPK23 positively regulate K+ absorption [4], and here we show that AvrPiz-t competes with OsCIPK23 for binding to OsAKT1 (Fig 4). This may affect the stability of the OsAKT1-OsCIPK23 complex and decrease OsAKT1 activity. Although it has been reported that activation of AKT1 in Arabidopsis depends on the phosphorylation of CIPK23 [36, 40], we were not able to demonstrate that OsCIPK23 has kinase activity in vitro, even when incubated with OsCBL1 or deleted the self-inhibition NAF motif as indicated in Arabidopsis CIPK23 [36]. We speculate that OsCIPK23 kinase activity may require another unknown protein except OsCBL1. Consistent with the AvrPiz-t competition effect, the oscipk23 mutant showed significantly reduced K+ levels and enhanced susceptibility to M. oryzae (Fig 5A–5C). Moreover, the osakt1 and oscipk23 mutants showed impaired K+ uptake and ROS accumulation (Fig 5D and 5E), which provides further evidence for the importance of K+ in enhancing resistance to M. oryzae in rice. PexRD54, an effector protein secreted by the oomycete pathogen Phytophthora infestans, also utilizes a competition mechanism to subvert host defense [12]. PexRD54 directly binds to ATG8CL and competes with the host cargo receptor Joka2 that causes Joka2 out of the ATG8CL autophagosomes, thereby promoting disease susceptibility [12]. PsAvh23, another effector from the soybean pathogen Phytophthora sojae, attacks ADA2 subunit of histone acetyltransferase SAGA to block the association of ADA2 with the catalytic subunit of GCN5, and thus to suppress H3K9 acetylation and increase plant susceptibility [52]. Those examples suggest that pathogen effectors may interfere the function of host protein complex by competitive binding.
In conclusion, our study reveals a novel function of AvrPiz-t, as well as the role of OsAKT1-OsCIPK23-mediated K+ signaling in rice innate immunity. We provide evidence of an intimate connection between plant nutrition status and disease resistance and propose a working model depicting the mechanism by which AvrPiz-t promotes the infection of the blast fungus via perturbing the function of the K+-associated OsAKT1-OsCIPK23 complex in rice (Fig 6H).
For hydroponic cultivation, the rice seeds were sown on wet filter paper and pre-germinated in an incubator for 3 days. The germinated seeds were transferred to nutrient solution containing 1 mM or 5 mM K+. The solution was replaced every 3 days. For the transgenic plants and T-DNA insertion mutants, seeds were germinated on half-strength Murashige and Skoog (MS) medium containing 50 mg/L hygromycin for 10 days and then transferred to soil. Seedlings were kept in a growth chamber at 26°C and 70% relative humidity with a 12-h light/dark photoperiod.
The M. oryzae isolates RO1-1 and RB22 were cultivated on oat meal medium under weak light for 2 weeks to generate spores. For spray inoculation, three-week-old seedlings were sprayed with 1–1.5×105/mL spores of M. oryzae as previously described [53]. 5–7 days after inoculation, the typical susceptible lesion numbers in each seedling were counted to evaluate the infection level.
Four-six weeks old rice plants were tested with the punch inoculation method as previously described [24]. Different from spay inoculation, the concentration of spores is higher for punch inoculation (about 5×105/mL). The rice leaves were lightly punched with a mouse ear clip and a 10 μL volume of spore suspension was dropped on the punched sites of leaves. Then the spore suspension was hold by sealing with Scotch tape on both sides. Twelve days after inoculation, the inoculated leaves were photographed and the relative fungal biomass was calculated to determine the fungal growth in leaves. The calculation of the relative fungal biomass was performed as described before [24]. Briefly, 4 cm of rice leaf with lesion was cut for DNA extraction with classical cetyltrimethyl ammonium bromide (CTAB) extraction protocol. Relative fungal biomass was measured with DNA-based quantitative PCR (qPCR) using the threshold cycle value (CT) of M. oryzae MoPot2 gene against the CT value of rice genomic Ubiquitin (OsUbq) gene according to the formula 2[CT(OsUbq)-CT(MoPot2)]. The qPCR was performed with 2×SYBR Green Mix (GeneStar) on ABI Prism 7500 PCR instrument. Primers used for analysis were listed in S1 Table.
The Y2H and GST pull-down assays were performed as previously described [27] with slight modifications. Briefly, the ProQuest Two-Hybrid System (Invitrogen) was used for the Y2H experiments. The AvrPiz-t coding region without the predicted signal peptide sequence [24] was cloned in-frame into the pDBleu vector as the bait. The sequences encoding the OsAKT1 intracellular C terminal fragments were cloned into the pPC86 vector as the prey. After co-transformation of yeast (Saccharomyces cerevisiae strain Mav203) and screening on SD/-Leu-Trp plates, positive clones were selected to grow on SD/-Leu-Trp-His medium with different concentration of 3-Amino-1,2,4-Triazole (3AT), or tested with β-galactosidase (X-gal).
For the GST pull-down assay, ~10 μg GST fusion proteins were mixed with ~10 μg or the indicated amount of MBP fusion proteins. The mixtures were incubated at room temperature for 1 h with gentle shaking, and then 20 μL pre-rinsed glutathione sepharose beads (GE Healthcare) was added, followed by incubation at room temperature for another 1 h. The beads were then washed 5–7 times with 1× TBST buffer. Finally, 1× SDS sample-loading buffer was added to the beads, and the mixture boiled for 5 min, prior to SDS-PAGE analysis.
The luciferase complementation assay was conducted as previously described [54]. Agrobacterium tumefaciens (strain EHA105) containing the desired constructs was used to infiltrate N. benthamiana leaves after adjusting the concentration of bacterial solution with MES buffer (10 mM MgCl2, 10 mM MES, pH 5.6) to optical density 600 (OD600) of 0.5. At 36 h after infiltration, leaf discs were taken and incubated with 150 ng/ml D-luciferin potassium in a 96-well plate, and the relative LUC activity was detected with a GLOMAX 96 microplate luminometer (Promega).
For the Co-IP assays, proteins were extracted from infiltrated leaf tissues with native buffer (50 mM Tris-MES, pH 8.0, 0.5 M sucrose, 1 mM MgCl2, 10 mM EDTA, 5 mM DTT, and protease inhibitor cocktail) and were subjected to anti-HA immunoprecipitation. After incubation with HA antibody (Roche) for 4 h at 4°C, 20 μL pre-rinsed Protein G beads (Millipore) was added to the protein-antibody mixtures and incubated for another 3 h. The beads were then washed 3–5 times with 1× TBST buffer. 1× SDS loading buffer was added to the samples and boiled for 5 min to elute the proteins prior to SDS-PAGE and immunoblot analysis.
The patch-clamping experiments were performed as described previously [4]. Briefly, HEK293 cells (human embryonic kidney cell line 293) were purchased from ATCC (American Type Culture Collection) and pre-cultured in Dulbecco’s modified eagle medium with 4.5 g/L glucose and 10% fetal calf serum for 24 h at 37°C with 5% CO2. OsAKT1 full length CDS was cloned in-frame in the pEGFP-N1 vector (Clontech). DsRed was used to monitor the transfection efficiency and was added to the pBudCE4.1 vector (Invitrogen) to allow expression driven by the EF-1α promoter, while AvrPiz-t or AvrPii variants were cloned into another site in the vector under control of the CMV promoter. The constructs pEGFP-N1-OsAKT1 and pBudCE4.1-DsRed-AvrPiz-t/AvrPii were co-transfected into HEK293 cells and the transfected cells were collected by centrifugation at 160 g for 5 min. The cells with both GFP and RFP fluorescence were selected for whole-cell recording, which was conducted at 20°C in dim light with an Axopatch 200B amplifier (Axon Instruments). The pipette solution and batch solution were as previously described [4]. Primers used for vector constructions were listed in S1 Table.
The shoots and roots of three-week-old plants were collected, rinsed with deionized water and dried at 65°C to a constant weight (at least 3 days). The dry samples were weighed and then incinerated in a muffle furnace at 300°C for 1 h and 575°C for 7 h, as previously described [4]. After incineration, the ashes were dissolved in 0.1 N hydrochloric acid and diluted with water to a suitable K+ concentration (~100 mM) based on the K+ content of rice (~3%). The K+ concentrations were determined by microwave plasma emission spectrometry (Agilent 4100 MP-AES).
The K+ depletion assay was performed as previously described [4]. Rice seeds were germinated on half-strength MS medium at 28°C under full light. Seven days after germination, 7 seedlings (fresh weight ~0.8g) with the endosperm removed were pretreated with starvation solution (0.2 mM CaSO4, 5 mM MES, pH 5.75) for 18 h then transferred to depletion solution (0.25mM KNO3, 0.2 mM CaSO4, 5 mM MES, pH 5.75). The treatments were conducted at 28°C under full light on a shaking table and samples were collected at different time points.
Rice leaf samples were collected from three-week-old seedlings for DNA or RNA isolation. DNA was isolated with CTAB buffer (2% CTAB, 100 mM Tris-HCl, 20 mM EDTA, 1.4 M NaCl, 0.1% 2-mercaptoethanol). Total RNA was isolated with Trizol regent (Invitrogen) according to the manufacturer’s instruction. First strand cDNA was synthesized with reverse transcriptase (Promega) after digestion of total RNA with DNase (TransGen). Genomic PCR and semi-quantitative PCR were performed with 2×TSINGKE Master Mix (TSINGKE). qPCR was performed with 2×SYBR Green Mix (GeneStar) on ABI Prism 7500 PCR instrument. Gene expression levels were calculated with the data from three technical repeats. Primers used for analysis were listed in S1 Table.
Rice seeds were germinated on half-strength MS medium at 28°C under 12 h dark and 12 h light for 10 days. Seedlings were removed endosperm and placed in starvation solution for 2 days. The net K+ flux was measured at the Xuyue (Beijing) Biofunction Institute by using NMT (NMT100 series; YoungerUSA LLC, Amherst, MA01002, USA) and imFluxes V2.0 software (YoungerUSA LLC, Amherst, MA01002, USA). The roots of seedlings were first equilibrated in measuring buffer (0.1 mM CaCl2 and 0.2 mM MES, PH 6.0) and then transferred to a measuring chamber with measuring buffer containing either 0.1 mM or 1 mM K+. The ion-selective electrodes were calibrated with measuring buffer containing 0.05, 0.1, and 0.5 mM K+ before measurement. Net K+ fluxes were measured for 10 min under experimental conditions. At least 6 individual plants were measured in an independent experiment.
To measure H2O2 levels in seedlings, we used a H2O2 detection ELISA kit (Kmaels DRE-P9104c). Seedlings (fresh weight ~200mg) were homogenized using a blender in 2 mL acetone and centrifuged at 10,000 g for 10 min. The supernatant was assayed for H2O2 concentration. The standard samples and testing samples were added to the ELISA plate according to the manufacturer’s instructions. After the reaction, the optical density was detected at 450 nm (OD450) using a microtiter plate reader within 15 min. A standard curve was generated by plotting the average OD450 of the standard samples.
For the ROS kinetic analysis, leaf disks were cut from 6-week-old plants and preincubated in sterile distilled water for about 10 h. Three leaf disks were transferred to a 1.5 mL microcentrifuge tube (Axygen) containing 100 μL of luminol (Bio-Rad Immun-Star horseradish peroxidase substrate), 1 μL of horseradish peroxidase and also 8 nM chitin (hexa-N-acetyl-chitohexaose) or water. Luminescence was then immediately measured using a Glomax 20/20 luminometer. Three biological replicates were assayed for each sample.
Statistical analyses were performed using GraphPad Prism 7.0 software. Descriptions of the tests used are introduced in figure legends.
Sequence data of rice genes can be found in the Rice Genome Annotation Project under following accession numbers: LOC_Os01g45990 for OsAKT1 and LOC_Os07g05620 for OsCIPK23. The gene sequences of M. oryzae can be found in the Gene Bank database with the following accession codes: EU837058 for AvrPiz-t and AB498874 for AvrPii.
|
10.1371/journal.pgen.1003445 | Smaug/SAMD4A Restores Translational Activity of CUGBP1 and Suppresses CUG-Induced Myopathy | We report the identification and characterization of a previously unknown suppressor of myopathy caused by expansion of CUG repeats, the mutation that triggers Myotonic Dystrophy Type 1 (DM1). We screened a collection of genes encoding RNA–binding proteins as candidates to modify DM1 pathogenesis using a well established Drosophila model of the disease. The screen revealed smaug as a powerful modulator of CUG-induced toxicity. Increasing smaug levels prevents muscle wasting and restores muscle function, while reducing its function exacerbates CUG-induced phenotypes. Using human myoblasts, we show physical interactions between human Smaug (SMAUG1/SMAD4A) and CUGBP1. Increased levels of SMAUG1 correct the abnormally high nuclear accumulation of CUGBP1 in myoblasts from DM1 patients. In addition, augmenting SMAUG1 levels leads to a reduction of inactive CUGBP1-eIF2α translational complexes and to a correction of translation of MRG15, a downstream target of CUGBP1. Therefore, Smaug suppresses CUG-mediated muscle wasting at least in part via restoration of translational activity of CUGBP1.
| Myotonic dystrophy type 1 (DM1) is the most common among the muscular dystrophies causing muscle weakness and wasting in adults, and it is triggered by expansion of an untranslated CUG repeat. To identify potential therapeutic approaches, we used a Drosophila DM1 model to screen for genes capable of suppressing CUG-induced toxicity. Here we report that increased levels of the smaug gene prevent muscle wasting and, perhaps more impressively, also prevent muscle dysfunction caused by the DM1 mutation. Smaug interacts genetically and physically with CUGBP1, an RNA–binding protein previously implicated in DM1. We used myoblasts from DM1 patients and control individuals to investigate how Smaug suppresses CUG-induced myopathy. We found that increased human SMAUG1 (a.k.a. SMAD4A) levels revert the abnormal accumulation of CUGBP1 in myoblasts nuclei and restore normal translation of at least one mRNA regulated by CUGBP1 in the cytoplasm. These findings demonstrate that manipulating Smaug activity protects against the effects of the DM1 mutation, and they also support the idea that restoring normal CUGBP1 function is a potential therapeutic approach.
| Myotonic Dystrophy type 1 (DM1) is a multisystemic neuromuscular disorder that has become a paradigm of a class of diseases caused by RNA toxicity. DM1 arises from expansion of a CTG triplet repeat in the 3′ untranslated region of the DMPK gene, and it accounts for the majority of adult cases of muscular dystrophy [1]–[5].
In DM1 the CUG-expanded mRNA is trapped in the nuclei where it forms nuclear foci and sequesters MBNL1 protein leading to loss of its activity [6], [7]. In addition, the mutant mRNA leads to increased steady-state levels of CUGBP1 (a.k.a CELF1) [8], [9] through its stabilization as a result of PKC phosphorylation [10]. Both MBNL1 and CUGBP1 are RNA-binding proteins involved in regulation of splicing [11]–[14], and aberrant splicing of the insulin receptor [12], muscle-specific chloride channel [13], [15] and many other genes [16], [17] occur in DM1.
The critical significance of MBNL1 sequestration for DM1 pathogenesis is eloquently demonstrated in loss of function and overexpression experiments. MBNL1 mutant mice show cataracts, myotonia, and other muscle abnormalities [7] that closely resemble a number of DM1 pathological features, and they also share many of the splicing aberrations observed in transgenic mice expressing CUG repeats [16], [17]. Importantly, MBNL1 overexpression ameliorates, muscle wasting in a Drosophila DM1 model [18], and myotonia and splicing aberrations in mouse models [19].
Evidence of the relevance of increased steady-state levels of CUGBP1 in DM1 pathogenesis comes from overexpression experiments. Transgenic mice expressing CUGBP1 show delays in muscle development and differentiation [20], muscle wasting [21], splicing misregulation [22] and DM1-like cardiac abnormalities [23]. Besides its nuclear role in splicing, CUGBP1 also has other functions in the cytoplasm including regulation of mRNA translation and stability [24]–[26]. Alterations of protein [25] and mRNA [16] levels occur in DM1 consistent with the idea that perturbation of CUGBP1 cytoplasmic functions contribute to DM1 pathogenesis. CUGBP1 cellular localization depends on its phosphorylation status [25]. Several kinases phosphorylate CUGBP1 at different residues and affect its localization within the cell. Activation of the Akt pathway increases CUGBP1 phosphorylation at Ser-28 altering the transition from proliferating myoblasts to differentiated myotubes in DM1 [27]. On the other hand, DM1 cells show decreased activity of cyclin D3-cdk4, another kinase that phosphorylates CUGBP1. This renders higher levels of unphosphorylated CUGBP1, which forms inactive complexes with eIF2α (CUGBP1-eIF2α) affecting translation of mRNAs required for myoblast differentiation. These inactive complexes containing CUGBP1 accumulate in the cytoplasm of DM1 cells in stress granules (SG) [25].
The richness of evidence implicating CUGBP1 in DM1 pathogenesis suggests the possibility that correcting the abnormal levels and activity of CUGBP1 may be a therapeutic approach to ameliorate DM1 pathogenesis. In support of this idea, Wang and colleagues used a pharmacological approach to inhibit PKC in mice expressing (CUG)960 in the heart; this treatment ameliorates the mortality rates and cardiac conduction as well as contractile abnormalities in this heart-specific DM1 mouse model [28]. Additional evidence comes from the observation that overexpression of a nuclear dominant negative CUGBP1 protein reverses dysregulation of a splicing minigene reporter in cultured cells, and of the CUGBP1 target Nrap exon 12 in DM1 mice [29].
Here we report that smaug, a gene not previously known to be implicated in DM1, is a powerful suppressor of CUG-induced myopathy when overexpressed in Drosophila. We show that human SMAUG1 protein (a.k.a SAMD4A) interacts with CUGBP1 and decreases its abnormally high steady-state levels in DM1 nuclei. Furthermore, increasing the levels of SMAUG1 in myoblasts of DM1 patients decreases the amount of inactive CUGBP1-eIF2α translational complexes. This suggests that SMAUG1 improves the activity of the CUGBP1-containing translational complexes that are dysfunctional in DM1, a hypothesis that is supported by data showing SMAUG1-mediated increased translation of the CUGBP1 translational target MRG15 in DM1 myoblasts.
To identify previously unknown genes implicated in DM1 pathogenesis, we used a well characterized Drosophila DM1 model [18]. Since DM1 is caused by expansion of an untranslated transcript, and MBNL1 and CUGBP1 are themselves RNA-binding proteins, we hypothesized that DM1 modifier genes may be enriched among genes encoding RNA binding proteins (RNA-BPs). Thus, we screened a collection of 93 loss of function and 17 overexpression alleles in 73 RNA-BP genes for their ability to modulate pathogenesis caused by expanded CUG repeats.
First, we used an external eye phenotype induced by expression of (CUG)480 as a primary screen to identify genes able to ameliorate or enhance CUG-induced toxicity. To validate the identified modifiers we tested the ability of the primary screen hits to modify CUG-induced muscle wasting.
Among the RNA-BPs tested, we uncovered the Drosophila gene smaug as a strong modifier of both the eye and muscle degeneration. As shown in Figure 1 increased levels of Smaug rescue the eye disorganization and loss of bristle phenotypes induced by (CUG)480 expression (compare Figure 1C with 1B). Consistent with this result, partial loss of function of smaug caused by a heterozygous mutation enhances the (CUG)480-induced eye phenotype (compare Figure 1D with 1B). As shown in Figure S1 these overexpression and loss-of-function alleles do not induce any abnormal phenotypes in control animals that do not express expanded CUG repeats.
The DM1 Drosophila model shows progressive muscle wasting, which is easily studied in the indirect flight muscles of the thorax. While 1-day-old (CUG)480 flies have muscles that appear wild type, animals that are 20 days old show muscle disorganization, vacuolization and loss of muscle fibers [18]. We investigated the effect of increasing the levels of smaug on (CUG)480-induced muscle wasting. As shown in Figure 1E overexpression of smaug dramatically suppresses CUG-induced myopathy.
Next we investigated whether increased smaug levels could restore muscle function in addition to muscle integrity. Animals expressing (CUG)480 show a severe impairment in flying ability prior to showing any signs of muscle wasting by histological analysis (see green bars in Figure 1F). Increased levels of smaug sharply improve flying ability in animals expressing (CUG)480. (compare orange and green bars in Figure 1F). These muscle histology and behavioral data further support the idea that smaug is a suppressor of expanded-CUG toxicity in a variety of cellular contexts.
In addition, we investigated whether Drosophila Smaug protein and the expanded-CUG RNA co-localize in nuclear foci. Previous studies have shown that Smaug accumulates in cytoplasmic foci similar to stress granules, and that it can shuttle between the nucleus and the cytoplasm [30]. To determine whether Smaug protein localization is altered due to expression of (CUG)480, we performed in situ and immunofluorescense analysis of larval muscles of animals expressing (CUG)480. As shown in Figure 1G Smaug accumulates mainly in the cytoplasm in the form of granules (Figure 1G, green), and it does not co-localize with the nuclear CUG-containing foci (Figure 1G, red, NF). This observation suggests that the mechanism by which Smaug modulates expanded-CUG toxicity does not involve direct interaction with the nuclear foci.
The data described above and shown in Figure 1G does not suggest sequestration of Smaug in nuclear foci as a mechanism for Smaug modification of expanded-CUG toxicity. Consequently, we investigated possible interactions between smaug and the known key players in DM1 pathogenesis: MBNL1 and CUGBP1. Overexpression of human MBNL1 or CUGBP1 in the Drosophila eye leads to a mild disorganization phenotype [18], and Figure 2A and 2D. We used these phenotypes as assays to test potential genetic interactions with smaug. We found that smaug overexpression suppresses the phenotype induced by CUGBP1 expression (compare Figure 2B with Figure 2A). In addition, smaug partial loss of function enhances this phenotype (compare Figure 2C with Figure 2A). In contrast, altering smaug levels does not have an effect on the MBNL1-induced eye phenotype (Figure 2D–2F).
In summary, we find that CUGBP1 and Smaug interact genetically in Drosophila.
To further investigate the interaction between SMAUG1 and CUGBP1, we performed immunofluorescense on COSM6 cells transfected with SMAUG1 and (CUG)960. We found that CUGBP1 localizes predominantly in the nucleus in cells transfected only with (CUG)960 (see arrowhead in Figure 3A), an observation that is consistent with previous reports [8], [9], [31], [32]. We found, however, that nuclear CUGBP1 steady-state levels are significantly decreased in cells transfected with both (CUG)960 and SMAUG1 (Figure 3B, arrowhead). CUGBP1 can be seen in these cells both diffuse in the cytoplasm as well as co-localizing with SMAUG1 in cytoplasmic granules (Figure 3B, arrow). As control we transfected with GFP and we could not observe differences in CUGBP1 signal between GFP-transfected (Figure 3C, arrow) and GFP-untransfected (Figure 3C, arrowhead) cells. A similar experiment was performed with MBNL1 and SMAUG1, but we found no evidence of changes in the accumulation of MBNL1 in nuclear foci following expression of SMAUG1 (Figure S2).
To validate these data on a cell type more relevant to DM1, we investigated whether CUGBP1 distribution is altered by SMAUG1 expression in myoblasts from DM1 patients. DM1 myoblasts transfected with GFP show predominantly nuclear CUGBP1 signal (Figure 4A). In contrast, DM1 myoblasts transfected with SMAUG1 show significantly decreased levels of nuclear CUGBP1 (Figure 4B, compare intensity of CUGBP1 staining in SMAUG1-transfected (arrowhead) vs. untransfected myoblasts). We quantified the intensity of the signal of nuclear CUGBP1 staining in DM1 myoblasts transfected with SMAUG1 versus controls transfected with GFP, and we found that SMAUG1-transfected DM1 myoblasts show a significant decrease in the nuclear signal intensity compared to controls transfected with GFP (Figure 4D, p<0.0001).
In addition we find that in SMAUG1-transfected DM1 myoblasts CUGBP1 and SMAUG1 co-localize in cytoplasmic granules (Figure 4C, arrows, and Figure S3). Cytoplasmic co-localization of both proteins was also observed in normal myoblast (Figure 5B). In spite of the observation that SMAUG1-expressing DM1 myoblasts show reduced nuclear CUGBP1, we did not detect an increase on cytoplasmic CUGBP1 in DM1 myoblasts transfected with SMAUG1 when compared to GFP-transfected controls (Figure S3, see also western blot of cytoplasmic fraction in Figure 6A). In control non-DM1 myoblasts transfected with SMAUG1 nuclear CUGBP1 signal remains the same (Figure 5).
Prompted by the genetic interaction between Drosophila smaug and human CUGBP1 and the co-localization of SMAUG1 and CUGBP1 in cells, we also investigated whether human SMAUG1/SAMD4A and CUGBP1 proteins physically interact. To do so, we performed co-immunoprecipitation experiments with cellular extracts from human myoblasts expressing SMAUG1. As shown in Figure 4E, SMAUG1 signal is detected after pull-down with anti-CUGBP1 antibody both in normal and DM1 myoblasts.
The intriguing finding that increased levels of SMAUG1 leads to decreased nuclear accumulation of CUGBP1 suggests that restoration of normal alternative splicing patterns may explain SMAUG1-mediated suppression of CUG-induced myopathy. To test this potential mechanism of SMAUG1 suppression, we examined the alternative splicing changes induced by either expanded CUG repeats or CUGBP1 overexpression.
Using a cTNT minigene reporter we found no evidence that overexpression of SMAUG1 restores normal alternative splicing changes caused by either expanded CUG repeats or CUGBP1 overexpression (Figure S4).
Since we did not find evidence that increased SMAUG1 restore alternative splicing patterns, we investigated whether they restore CUGBP1 normal function in the cytoplasm. CUGPB1 regulates the translation and stability of mRNAs, and these activities are impaired in DM1 [24]–[26]; thus, we asked if the translational activity of CUGPB1 is influenced by SMAUG1.
In the cytoplasm, CUGBP1 interacts with eukaryotic translation initiation factor eIF2α (eIF2α), and its translational activity is mediated by CUGBP1-eIF2α complexes [33]. CUGBP1 phosphorylated at S302 binds to unphosphorylated eIF2α (non-pS51-eIF2α) making active CUGBP1-eIF2α translational complexes, whereas CUGBP1 unphosphorylated at S302 binds with higher affinity to inactive pS51-eIF2α forming CUGBP1-eIF2α inactive translational complexes [25]. In DM1 cells, the levels of inactive eIF2α (pS51-eIF2α) are increased, and formation of inactive CUGBP1-eIF2α complexes inhibits translation of certain mRNAs in DM1 myoblasts [25].
Therefore, we examined the effects of SMAUG1 on the abundance of inactive CUGBP1-eIF2α complexes. Western blot analysis of cytoplasmic extracts from transfected normal and DM1 myoblasts and fibroblasts show that the total cytoplasmic levels of CUGBP1 are increased in DM1 myoblasts and fibroblasts, and are not affected significantly by SMAUG1 (Figure 6A–6B and Figure S5). Additionally, we investigated if levels of total eIF2α are altered by SMAUG1 expression in cytoplasm. As shown in Figure 6A and 6B, total levels of eIF2α remain unchanged upon SMAUG1 transfection in both normal and DM1 myoblasts and fibroblasts. We then performed co-IP experiments on cytoplasmic protein extracts from normal and DM1 myoblasts and fibroblasts to test whether SMAUG1 expression alters the levels of inactive CUGBP1-eIF2α complexes. We found that in control GFP-transfected DM1 myoblasts/fibroblasts pS51-eIF2α-CUGBP1 inactive complexes are abundant. This is in striking contrast to SMAUG1-transfected DM1 myoblasts (Figure 6A, see CUGBP1-IP) and Figure S6) and fibroblasts (Figure 6B, see CUGBP1-IP) and Figure S6) where these complexes are undetectable. Thus, increasing SMAUG1 levels decreases the steady-state levels of CUGBP1-eIF2α inactive translational complexes.
Previous reports have shown that MRG15 mRNA translation is controlled by CUGBP1-eIF2α complexes. Particularly, inactive CUGBP1-eIF2α complexes trap MRG15 mRNA in stress granules and reduces protein levels of MRG15 in DM1 compared to normal myoblasts [25]. Since we found that expression of SMAUG1 reduces amounts of inactive CUGBP-eIF2α complexes, we investigated if this reduction corrects translation of MRG15, a target of the CUGBP-eIF2α complex. Western blot analysis of nuclear protein extracts shows that DM1 cells contain reduced amounts of MRG15; however, SMAUG1 restores translation of MRG15 in DM1 cells to near normal levels in both myoblasts and fibroblasts (Figure 6C, Figure S7). In summary, these data indicate that expression of SMAUG1 significantly reduces the amounts of inactive CUGBP1-eIF2α complexes and enhances translation of MRG15.
Here we show that increased expression levels of smaug, a conserved gene involved in translational regulation, suppresses CUG-induced muscle wasting and, notably, it also restores normal muscle function in a Drosophila model of DM1. Experiments in DM1 myoblasts indicate that the human homolog SMAUG1/SAMD4A suppresses the toxic effects of expanded CUG repeats at least in part by restoring impaired CUGBP1 translational functions.
Early during DM1 pathogenesis CUGBP1 steady-state levels increase as a consequence of PKC-mediated phosphorylation [10]. This leads to disrupted regulation of alternative splicing, as well as impairments in mRNA stability and mRNA translation, all of which contribute to the multiple features of the disease (reviewed in [34], [35]). In addition, CUGBP1 overexpression in wild-type mice mimics some of the functional, histopathological and molecular features of DM1 [22], [36], [37], while CUGBP1 overexpression in Drosophila enhances expanded-CUG induced pathology [18]. Together these observations suggest that restoring normal CUGBP1 levels and activities may reverse DM1 pathology. This approach however may prove difficult to execute. First, there are several CUGBP1-like proteins in mammals and in Drosophila making proof-of-principle experiments using loss-of-function alleles complicated. To circumvent the problem of functional redundancy, a dominant-negative CUGBP1 construct was expressed in culture cells and mice, and this resulted in the reversion of abnormal alternative splicing [29]. Expression of dominant-negative CUGBP1, however, also leads to cardiac and skeletal muscle pathology [38], [39]. A second and perhaps more important caveat is that CUG expansion leads to increased nuclear levels of CUGBP1 [8], [40] (i.e., a gain of function), while in the cytoplasm the same mutation leads to the inactivation of CUGBP1 translational complexes [25] (i.e., loss of function). Hence, restoring normal CUGBP1 activities in both nucleus and cytoplasm by modulating CUGBP1 itself seems challenging. An alternative approach is to target other factors modulating CUGBP1 function. One such factor is PKC, a kinase that phosphorylates and stabilizes CUGBP1 [10]. Indeed, PKC pharmacological inhibition ameliorates the cardiac phenotypes in a heart-specific DM1 mouse model [28].
The data presented here reveals that SMAUG1/SAMD4 is able to restore CUGBP1 normal levels and activities. We found that increasing the levels of SMAUG1 leads to decreased levels of nuclear CUGBP1 in DM1 myoblasts. This intriguing observation suggested that rescue of alternative splicing alterations may be a possible mechanism to explain the observed suppression of CUG-induced myopathy. This is an open possibility because even though we did not find evidence of SMAUG1 modulating splicing on a cTNT minigene, we cannot rule out its effects on other unknown splicing targets.
We showed that SMAUG1 can re-activate impaired CUGBP1 translational activities in the cytoplasm. smaug was first discovered in Drosophila as a translational regulator of nanos mRNA in the posterior pole of the embryo [41]. In this context it functions as a translational repressor by capturing transcripts containing Smaug recognition elements, forming stable ribonucleoprotein particles, and displacing the eIF4G initiation factor [42]. Smaug also promotes the destabilization and degradation of nanos mRNA by recruiting a deadenylation factor [43]–[45]. There are two smaug homologous genes in mammals [30]. One of them, SMAUG1/SAMD4A, forms mRNA-silencing foci at postsynapses of hippocampal neurons that respond to NMDA and modulate synapse formation [46]. We find that SMAUG1 has a positive function in the context of CUGBP1-dependent translation in myoblasts suggesting that SMAUG1 is not a dedicated repressor of translation, but rather a translational regulator whose function is context dependent.
In DM1, high levels of CUGBP1 unphosphorylated at S302 form inactive translational complexes with pS51-eIF2α. We found that increased levels of SMAUG1 lead to a dramatic reduction of CUGBP1-eIF2α inactive complexes. It is unlikely that this is a result of nuclear CUGBP1 being exported to the cytoplasm because we did not detect an increase of CUGBP1 in western blots of cytoplasmic extracts from SMAUG1-transfected DM1 myoblasts or fibroblasts. This observation was confirmed by immnofluorescence experiments showing similar levels of cytoplasmic CUGBP1 between SMAUG1-transfected and GFP-transfected DM1 myoblasts. An attractive possibility is that the interaction between SMAUG1 and CUGBP1 promotes repair of defective initiation complexes. In support of this hypothesis we observe an increase in translation of MRG15. Translation of MEF2A, C/EBPbeta, p21, and other translational CUGBP1 targets such as cyclin D1 and HDAC1 are promoted by active CUGBP1/elF2 complexes (i.e., formed by p-S302-CUGBP1 and elF2 not phosphorylated at S51). However, we only know of one target, MRG15, whose translation is inhibited by inactive CUGBP1/elF2 complexes (i.e., formed by CUGBP1 not phosphorylated at S302, and p-S51-elF2) [25]. Thus, we expect that other mRNA targets of CUGBP1 whose translation is impaired in DM1 may be corrected as well; however, these other mRNAs have not been identified yet.
The only therapy available for DM1 patients is used to treat the symptoms rather than the cause of the disease. Efforts to develop therapeutic avenues for DM1 pathogenesis include: 1) to revert the instability of the expansion, 2) to target the toxic RNA with ribozymes or antisense oligonucleotides [47]–[51], 3) to target the CUG RNA hairpins with siRNA [52]. Potential alternatives are to develop therapeutic approaches to restore CUGBP1 and MBNL1 protein levels and activities [18], [19], [28] (reviewed in [53]). The data reported here suggests that therapeutics designed to increase SMAUG1 protein levels could be useful to ameliorate the toxicity of the mutant RNA in DM1.
The transgenic lines UAS-(CTG)480, UAS-MBNL1, and UAS-CUGBP1 have been previously described [18]. Mhc-GAL4 was obtained from G. Davis (UCSF). gmr-GAL4, smg1 and smgEP3556 were obtained from Bloomington Stock Center (Indiana) and Szeged Stock Center (Hungary).
Processing of flies for SEM and image acquisition were performed following previously published procedures [54].
For paraffin sections, adult thoraxes were dissected out, fixed overnight in 4% formaldehyde in PBS, washed in PBS and dehydrated in increasing concentrations of ethanol. Thoraxes were embedded in paraffin. Serial sections of 10 µm were obtained and rehydrated to water. Sections were stained with eosin (Sigma) and the fluorescent images were captured using an AxioCam MRc camera (Zeiss) attached to a Microphot-FXA microscope (Nikon).
Individual adult flies were dropped one at a time from the top of a 12-inch cylinder and the landing position in the cylinder was recorded. One hundred flies per genotype were scored and each fly was tested three times.
The protocols were previously described in [18]. Anti-Smaug antibody (provided by C.A. Smibert) was used at a concentration of 1∶50.
Constructs used for transfection were (CUG)960 (T.A. Cooper) and SMAUG1-ECFP (G.L. Boccaccio). COSM6 cells were transfected with (CUG)960 alone or together with SMAUG1-ECFP using Amaxa Cell Line Nucleofector Line R (Lonza). Two days after transfection cells were fixed in 4% formaldehyde for one hour, washed and hybridized with a Cy3-labelled 5′-CAG-3′ LNA probe for one hour, followed by incubation with mouse anti-CUGBP1 3B1 antibody (1∶120) overnight at 4C. Secondary anti-mouse antibody labelled with Cy5 was used to visualize CUGBP1.
Human primary myoblasts derived from control individuals or from DM1 patients with 300 CTG repeats were grown for no more than 12 passages and transfected with SMAUG1-ECFP or control pmaxGFP using Amaxa Cell Line Nucleofector Line NHDF (Lonza). Two days after transfection in situ and immunofluorescense was performed as described in the above paragraph.
For quantification of CUGBP1 nuclear signal, pictures taken at the confocal microscope under the same conditions were analyzed using ImageJ software. Pictures of at least 50 different cells were taken. Data sets were compared using ANOVA followed by Student's t analysis.
Transfection of myoblasts and fibroblasts with SMAUG1-V5 (G.L. Boccaccio) was performed using Amaxa Nucleofector Line NHDF (Lonza). For co-immunoprecipitation of myoblasts in Figure 4E, anti-CUGBP1 3B1 antibody (Novus Biologicals) was coupled to Dynabeads M-270 Epoxy (Invitrogen), and co-IP was performed with Dynabeads Co-Immunoprecipitation kit (Invitrogen) using anti-V5 (Invitrogen) antibodies.
For western blot analysis, control and DM1 myoblasts (300 CTG repeats) and fibroblasts (2000 and 1600 CTGs) were transfected as described above. Two days after transfection nuclear and cytoplasmic protein fractions were extracted [31]. Twenty five µg of cytoplasmic proteins were separated by gel electrophoresis, transferred on membrane and incubated with anti-CUGBP 3B1 and anti-eIF2α (FL-315, Santa Cruz, CA, USA).
Co-IP with 100 µg of cytoplasmic protein was performed using the protocol associated with Trueblot Antibodies from eBioscience. Antibody for pS51-eIF2α was S51-sc12412-R from Santa Cruz.
For detection of MRG15, nuclear protein fractions of cells transfected with SMAUG1 or GFP were separated by gel electrophoresis, transferred on membrane and incubated with anti-MRG15 (F-19) and anti-β-actin (AC-15) from Santa Cruz.
|
10.1371/journal.pgen.1003706 | The Architecture of a Prototypical Bacterial Signaling Circuit Enables a Single Point Mutation to Confer Novel Network Properties | Even a single mutation can cause a marked change in a protein's properties. When the mutant protein functions within a network, complex phenotypes may emerge that are not intrinsic properties of the protein itself. Network architectures that enable such dramatic changes in function from a few mutations remain relatively uncharacterized. We describe a remarkable example of this versatility in the well-studied PhoQ/PhoP bacterial signaling network, which has an architecture found in many two-component systems. We found that a single point mutation that abolishes the phosphatase activity of the sensor kinase PhoQ results in a striking change in phenotype. The mutant responds to stimulus in a bistable manner, as opposed to the wild-type, which has a graded response. Mutant cells in on and off states have different morphologies, and their state is inherited over many generations. Interestingly, external conditions that repress signaling in the wild-type drive the mutant to the on state. Mathematical modeling and experiments suggest that the bistability depends on positive autoregulation of the two key proteins in the circuit, PhoP and PhoQ. The qualitatively different characteristics of the mutant come at a substantial fitness cost. Relative to the off state, the on state has a lower fitness in stationary phase cultures in rich medium (LB). However, due to the high inheritance of the on state, a population of on cells can be epigenetically trapped in a low-fitness state. Our results demonstrate the remarkable versatility of the prototypical two-component signaling architecture and highlight the tradeoffs in the particular case of the PhoQ/PhoP system.
| A mutation can cause significant changes to a protein's function. Since proteins often act together in genetic circuits to control various cellular processes, mutant proteins can lead to unexpected consequences for system-level behavior. In this study, we describe a remarkable example of this phenomenon in a mutant of a well-studied bacterial circuit. PhoQ and PhoP are the primary regulatory proteins in a circuit that responds to low magnesium. The wild-type (unmutated) network responds to environmental signals in an analog or graded manner. In contrast, the mutant responds to signals in an OFF-or-ON or digital fashion. Moreover, the distribution of OFF and ON cells is strongly influenced by how cells were cultured in the past. These remarkable changes can be traced to features of the wiring diagram of the PhoQ/PhoP circuit. Since these features are shared among a broad class of bacterial signaling circuits, we suggest that other circuits may show similar remarkable properties when mutated.
| A few mutations can lead to significant changes in a protein's functional properties. Examples include mutations that change the absorption and emission spectra of a fluorescent protein [1], the substrate specificity of an enzyme [2], or the allosteric control of a transcription factor [3]. In all of these examples, the change in phenotype can be directly traced to modifications in intrinsic properties of the protein. However, networks of interacting proteins can have system-level characteristics that bear a complex relationship to the intrinsic properties of the component molecules [4]. This complexity makes some network architectures inherently versatile, with different networks that share the same architecture exhibiting qualitatively different system-level behavior [5]. It remains a challenge to identify aspects of network architectures that promote versatility and permit novel properties to emerge by a few mutations to network components.
In this study, we demonstrate the versatility of the E. coli PhoQ/PhoP system. We show that a single point mutation in the histidine kinase PhoQ produces a striking change in the properties of the circuit. The PhoQ/PhoP system, which has an architecture found in many bacterial two-component signaling systems [6], responds to a variety of environmental conditions such as low Mg2+ [7], low pH [8], and the presence of cationic antimicrobial peptides [9], and controls transcription of a large set of genes [10]. The histidine kinase PhoQ senses these signals and modulates the phosphorylation level of the response regulator PhoP (PhoP-P), which functions as a transcription factor. PhoQ autophosphorylates and then transfers the phosphoryl group to PhoP, but also acts as a phosphatase, catalyzing PhoP-P dephosphorylation [11]. This bifunctional design, which is shared among many two-component systems, affects various properties of the system, including buffering the input-output relationship of the system to changes in histidine kinase and response regulator concentrations, and suppression of cross-talk [12]–[15]. The PhoQ/PhoP system is also autoregulated, that is, transcription of the phoPphoQ operon is activated by PhoP-P. Autoregulation is another common feature of many two-component systems [6] and is a mechanism for ultrasensitive response to stimulus without the need for cooperativity [16] as well as “learning” behaviors where prior exposure to stimulus improves response times to subsequent stimulating conditions [17], [18]. In the case of the PhoQ/PhoP system, autoregulation improves the dynamic range of the network output at high stimulus [13] and also gives rise to a surge in transcription upon activation [19].
The wild-type PhoQ/PhoP system responds to external stimulus in a graded rather than an all-or-none manner, with increasing stimulus (lower Mg2+ concentration) resulting in a higher mean response [13]. Moreover, the response of a population of cells is unimodal, with the PhoP-P levels of individual cells (inferred from transcriptional reporters) clustered around the population mean [13]. In this study, we show that a single point mutation that abolishes phosphatase activity in PhoQ produces a dramatic change in phenotype. The mutation, which produces a T281R substitution in PhoQ, results in bistability with mutants persisting in morphologically distinct OFF and ON states for many generations. We find that the architectural features of the network that allow the mutant to exhibit bistability are shared among many two-component systems. For PhoQ (T281R) mutants, however, the bistable phenotype comes at a fitness cost. We find that a population of cells can be epigenetically trapped in a low-fitness state.
A previous study reported that the phoQ (T281R) mutation results in a broad distribution of PhoP-regulated transcription in a population of E. coli cells [13]. To explore the origin of this heterogeneity, we engineered a phoQ (T281R) strain with the mutation at the native phoPphoQ locus. The strain also contained a PhoP-P responsive promoter controlling yfp transcription and a constitutive promoter controlling cfp, allowing the use of YFP fluorescence to infer PhoP-P levels (Figure 1A and Methods). Two fluorescent colony phenotypes could be discerned on agar plates: YFP-dim, which we designated phoQ (T281R) OFF, and YFP-bright, which we designated phoQ (T281R) ON (Figure 1B). Considering that each colony is composed of hundreds of millions of cells that originated from a single cell, the appearance of two distinct colony phenotypes suggests that individual cells have two phenotypic states and that these states are heritable.
To determine whether single cells showed a bimodal response, overnight cultures in LB medium were inoculated with OFF and ON colonies, and individual cells were imaged after dilution and growth to mid-exponential phase in minimal medium with 100 µM Mg2+ (Figure 1C–D). As with colonies, individual cells could be classified as OFF (low PhoP-P, YFP-dim) or ON (high PhoP-P, YFP-bright) with a ∼60-fold difference in YFP fluorescence between the two states (Figure 1D). The wild-type phoQ strain (denoted phoQ (WT)) cultured in the same manner showed an intermediate fluorescence, roughly 12-fold higher than the OFF state of the mutant. In addition to the bimodal distribution of phenotypes for the phoQ (T281R) strain, we also observed phenotypic hysteresis, i.e., an OFF colony yielded mostly OFF cells, and an ON colony gave rise to mostly ON cells, even though both populations of cells were cultured under the same conditions (Figure 1D). In these experiments, it is likely that the mgrB promoter driving yfp is near saturation in the ON state and the true change in PhoP-P levels between the OFF and ON states may be higher than 60-fold. Furthermore, since the YFP protein used in this study is stable, the switching of YFP-state from ON to OFF is limited by dilution of YFP due to growth. Consequently, one can see cells with an intermediate YFP-state occasionally even though the PhoP-P levels may actually be low.
Interestingly, the single-cell experiments also revealed a morphological difference between OFF and ON state cells (Figure 1C–D). ON cells have, on average, lower cell-widths (as quantified by the minor axis of the best-fit ellipse) than OFF cells, and the latter are similar to wild-type cells. This is likely an indirect effect of high PhoP-P levels, but we do not know the mechanism.
Why is the phosphatase-deficient mutant not constitutively ON? To gain insights into this question, we formulated a simple mathematical model of the PhoQ/PhoP network that consisted of only three species, viz., PhoP, PhoQ, and PhoP-P and ignored PhoQ-P and intermediate complexes (Text S1). In this model, the kinase rate is a proxy for any factor that can influence the production of PhoP-P from PhoP. Analysis of this model revealed that at high kinase rates, a phosphatase-deficient mutant would indeed be constitutively ON and the network would be monostable (Figure S1A). However, at low kinase rates, the phosphatase-deficient PhoQ/PhoP network could exhibit bistability (exist in OFF and ON states). The bistability results from positive feedback (transcriptional autoregulation, see Figure 1A) and the presence of two non-linearities: (a) the kinase and phosphatase reactions each depend on the product of two concentrations, and (b) the non-linear dependence of phoPQ operon transcription on PhoP-P concentration (Text S1). In vitro experiments have demonstrated that PhoQ (T281R) is a poorer kinase compared to PhoQ (WT) [13]. The model thus suggests it is both the low kinase activity and the absence of phosphatase activity of the PhoQ (T281R) mutant that facilitates the emergence of bistability. One of the assumptions in our model is that growth-mediated dilution is the dominant mechanism for reduction in concentrations of the stable proteins PhoP, PhoQ and PhoP-P (in the absence of a specific phosphatase). Consequently, growth rate is another parameter that influences the response of the network, with low growth rates leading to slower dilution and a constitutively ON phenotype.
To experimentally validate the insights from the simple model, we reasoned that changing [Mg2+] could be used to modulate the PhoP→PhoP-P flux (the kinase rate equivalent). Accordingly, we grew cultures of mostly OFF cells in minimal media with different [Mg2+] and maintained them exclusively in the exponential phase by serial dilution (growth rates were similar over the [Mg2+] range tested). Surprisingly, we found that high [Mg2+] (10 mM) drove all OFF cells to ON, whereas low [Mg2+] (100 µM or 1 mM) preserved the OFF state (EXP lineages, Figure 2A–B). This, in effect, represents a reversal of sensitivity to Mg2+ in the phoQ (T281R) strain as the wild-type strain is repressed (produces lower PhoP-P) by high Mg2+. If a portion of the same batch of OFF cells that was maintained in the exponential phase in the EXP lineages was instead passaged through stationary phase (effectively, a slow growth phase) and subsequently diluted and grown to mid-exponential phase, then even the low [Mg2+] lineages turned mostly ON (STA lineages, Figure 2A–B). Note that the different outcomes of STA and EXP lineages at low [Mg2+] represent yet another manifestation of hysteresis, since both lineages have identical starting points (same culture of OFF cells) and similar end points (mid-exponential cultures with the same [Mg2+]), but different history between start and end.
In the present context, the phenotypic hysteresis of phoQ (T281R) OFF shown in Figure 1D (i.e. the persistence of these cells in the OFF state) seems anomalous since the OFF cultures were passaged through stationary phase in LB. LB is a rich medium with low (but undetermined) [Mg2+]. Consistent with the reversal of Mg2+ sensitivity in phoQ (T281R), we find that supplementing LB with high [Mg2+] increases conversion of OFF cells to the ON state substantially (Figure 2C). In addition, when we grew LB overnight cultures at 30°C instead of 37°C, we were able to observe OFF→ON conversion at levels comparable to minimal medium with low [Mg2+] at 37°C, suggesting that the higher growth rate in LB plays a major role in the absence of OFF→ON conversion at 37°C (Figure 2D). However, we also note that the apparent low OFF→ON conversion in LB at 37°C (Figure 2C) may reflect the competitive advantage of OFF cells over ON ones in LB stationary phase cultures (see below), which would suppress our ability to detect the number of cells that had turned ON.
Taken together, the above results indicate that slow growth histories or high [Mg2+] are sufficient for en masse conversion of OFF cells to ON (Figure 3A). We call this deterministic conversion from OFF to ON state “priming”.
To further explore the mechanisms responsible for the phenomena described above, we developed a more detailed mathematical model of the PhoQ/PhoP network consisting of six species (PhoP, PhoP-P, PhoQ, PhoQ-P, and two intermediate protein complexes). While bistability provides an explanation for the observed all-or-none behavior based on deterministic steady-state analysis, it is also possible for a monostable network to exhibit bimodal behavior because of slow kinetics and positive feedback, as has been seen in stochastic models [16], [20], [21]. To compare and contrast these mechanisms, we examined the detailed model using both deterministic steady-state analysis and stochastic simulations (Text S1).
Qualitatively, the steady state behavior of the detailed model is similar to the simpler 3-species model indicating that the simpler model captures the essential features of the network required for bistability. In the model, there are three important kinetic parameters influencing bistability – the kinase rate of PhoQ (T281R), the maximal expression rate of the phoPQ operon, and the growth rate of the organism (Text S1). When one of these parameters is varied while keeping all others constant, the system can transition from a monostable OFF regime to a bistable regime and thence to a monostable ON regime (Figure 3B). Note that our model does not incorporate the role of [Mg2+] explicitly, but we posit that for PhoQ (T281R), raising [Mg2+] leads to an increase in the kinase rate or in the maximal operon expression rate.
Stochastic simulations also largely agree with the deterministic analysis, except that noise-induced bimodality can be observed with parameters in the monostable ON regime close to the bistable regime (Text S1). While noise-induced bimodality and deterministic bistability are different mechanisms, operationally it is difficult to distinguish them without detailed measurement of kinetic parameters in vivo. In either case, one would see inheritance of OFF and ON states over several generations.
Within our modeling framework, exponential phase in minimal medium with low [Mg2+] can be considered a condition in, or close to, the bistable regime. In this regime, cells retain their state except for rare, stochastic switching events (Figure 3B, Regime II). Slow growth rates or high [Mg2+] can independently drive the system towards the monostable ON regime, whereupon OFF cells proceed deterministically towards the ON state (Figure 3B, Regime III). Note that the rate of priming may be kinetically limited, but eventually all cells will turn ON. When these ON cells are sub-cultured in low [Mg2+] minimal medium, a bistable (or near-bistable) regime is established again, but hysteresis ensures that cells remain in the ON state as seen in STA lineages in Figure 2B.
Given that a majority of OFF cells can be turned ON by increasing [Mg2+] or by passaging through stationary phase, we asked whether there were culture conditions in which an ON population could be deterministically transformed to the OFF state. In other words, we were interested in establishing a monostable OFF regime (Figure 3B, Regime I). According to our model, this could in principle be achieved with low [Mg2+] and high growth rates. However, we were unable to observe a monostable OFF regime for exponential growth in minimal medium with 100 µM Mg2+ (data not shown) or for growth in LB (Figure 1D). We could not use significantly lower [Mg2+] levels without affecting growth rate. These results suggest that it may not be possible to experimentally realize the monostable OFF regime, i.e., the ON state may always be stable. A bistable system with this property is termed irreversible [22]. Note that irreversibility of the system only means that a population of ON cells cannot be deterministically turned OFF. Individual ON cells can still transition to the OFF state in the bistable regime because of stochastic fluctuations (Figure 3B, Regime II). Stochastic switching from the ON to OFF state and its implications are examined later in this study.
The high PhoP-P level in the ON state has pleiotropic effects on the cell. As presented in Figure 1C–D, cell morphology is affected in the ON state. We also determined that the ON state has a lower fitness in stationary phase in LB. To explore fitness differences between ON and OFF cells, we performed competitions using chloramphenicol resistant (CmR) and sensitive (CmS) phoQ(T281R) strains prepared in the ON and OFF states (Figure 4, S3 and Methods). When ON and OFF cells were competed for 10 hours in stationary phase in LB, the ON fraction in the population showed a significant decrease (competitive ratio, Figure 4, columns 2 and 4). In contrast, competitions between CmS and CmR OFF cells or between CmS and CmR ON cells yielded a near-neutral competitive ratio (Figure 4, columns 1 and 3 respectively).
To examine stochastic ON→OFF transitions more closely, we performed long-term culture experiments in LB. As mentioned above, a low [Mg2+], high growth rate medium such as LB is ideal for observing ON→OFF switching since that growth condition is likely to be close to the theoretical monostable OFF regime. Furthermore, we reasoned that the competitive advantage of OFF cells in stationary phase in LB could be used to amplify the effects of switching and enhance our ability to detect switching events. We established independent lineages by inoculating LB cultures with either phoQ (T281R) OFF or phoQ (T281R) ON colonies and maintained them through million-fold dilution once per day. The state of the population was assayed by spreading overnight cultures on minimal media plates on alternate days and measuring the fraction of ON (YFP-bright) colonies (Figure S4). This protocol subjected populations to ∼20 generations of exponential growth per day but the majority of time (>14 hours/day) was spent in stationary phase. Furthermore, at least ∼2000 cells were transferred from one day to the next, which meant that even if the fraction of OFF cells in the saturated previous day culture of an ON lineage was as low as 0.1%, there was a ∼90% chance that an OFF cell would be present in the inoculum for the next day culture based on the statistics of binomial sampling.
As expected, lineages inoculated with phoQ (T281R) OFF yielded mostly OFF colonies on the assay plates (Figure 5). Lineages inoculated with ON colonies, however, showed a different pattern. These remained close to 100% ON for several days (Figure 5, see Day 3 time point), but on Day 5, an appreciable fraction of YFP-dim colonies could be seen in 6 out of 7 lineages. Furthermore, the fraction of YFP-dim colonies obtained in the different lineages was not the same. This divergence highlights both the stochastic nature and the low probability of ON→OFF switching. Since all the ON lineages are likely to converge to a mostly OFF state eventually, the ON lineage in LB is metastable – a long-lived, but not truly stable state due to the competitive advantage of the OFF state. We note that the ON→OFF transitions seen in the ON lineages are unlikely to be the result of mutational events since the YFP-dim colonies obtained in these lineages could be primed ON by overnight growth in minimal medium with 10 mM Mg2+ (1 dim colony was tested per lineage, data not shown).
These results show that strong epigenetic inheritance can effectively trap a population in a low-fitness state under some circumstances. Irreversibility enhances this phenomenon by rendering rare, stochastic transitions between states as the fastest means for “escape” from the epigenetic trap.
Having examined the unique characteristics of the phoQ (T281R) strain, we asked whether there were any essential features in the PhoQ/PhoP network architecture that enabled a single point mutation to produce such remarkable behavior. Our modeling suggested that bistability (or near-bistability for the stochastic model) in the network was strongly dependent on PhoP-P regulation of both phoP and phoQ transcription (Text S1). We found that an altered PhoQ/PhoP network where phoP transcription is autoregulated but phoQ (T281R) transcription is independent of PhoP-P was likely to be monostable, especially if the phosphatase defect in PhoQ (T281R) stemmed from a high dissociation constant of the complex between PhoP-P and PhoQ (an intermediate in the phosphatase reaction) (Text S1 and Figure S1D).
To test whether elimination of PhoP-P dependent transcription of phoQ (T281R) abrogated bistability, we constructed a strain in which phoQ was deleted from its native locus and phoQ (T281R) was inserted at a phage attachment site under the control of the IPTG-inducible (and PhoP-P insensitive) Ptrc promoter (Figure 6A). We measured YFP/CFP in individual cells after 10 hours of growth (∼15 generations) at various IPTG concentrations starting from uninduced and fully induced populations of cells (Figure 6B). We did not observe any evidence of hysteresis in this strain: cultures started with both fully induced and uninduced cells converge to similar distributions for all IPTG concentrations tested (Figure 6B). Furthermore, distributions spanning intermediate values of YFP/CFP could be observed even after 15 generations of culture (i.e. populations did not converge to distributions with low and high modes). Taken together, these observations suggest that there is no bistability in the decoupled strain. The wide distributions seen at 12.5 µM and 25 µM IPTG can be attributed to a combination of noise in IPTG induction [23], [24] and the sensitivity of the system to induction level in this range (Figure 6B inset) and are also seen in our stochastic simulations (Figure S5D). The slightly lower median of samples derived from the uninduced culture compared with the induced culture could be a result of a stochastic effect that slows down induction kinetics in autoregulated systems [16]. We also determined that a strain similar to the one depicted in Figure 6A, but with phoQ (T281R) under the control of its native phoPQ promoter instead of Ptrc exhibited bistability (data not shown).
As documented in Figure 2, one of the defects in the phoQ (T281R) strain is that OFF cells transition to the ON state both in response to specific signals such as high [Mg2+] and to extraneous factors such as growth rate reduction in stationary phase. We therefore explored whether the phoQ (T281R) network could be modified to reduce priming in stationary phase whilst minimally affecting other characteristics of the strain. Based on our model, we predicted that a reduction in the basal level of phoPQ transcription (PhoP-P independent transcription) would extend the range of growth rates for which the OFF state was stable, thereby reducing conversion to the ON state at lower growth rates (Text S1 and Figure S1C). To test this prediction, we took advantage of the fact that phoPphoQ transcription is driven by two promoters: one is activated by PhoP-P and the other is constitutive [25]. We constructed a variant of the phoQ (T281R) strain with the −35 and upstream region of the constitutive P2 promoter deleted (Figure 6C). As with the parent strain, we obtained YFP-dim and YFP-bright colonies and designated these ΔP2 OFF and ΔP2 ON respectively. Consistent with the model predictions, only half of ΔP2 OFF cells prime in minimal medium with low [Mg2+] compared to ∼90% priming in OFF cells of the parent strain in these conditions (Figure 6D, compare maroon and white bars). The primed fraction in ΔP2 OFF also decreases in minimal medium with high [Mg2+], but only to ∼80% indicating that sensitivity to high [Mg2+] is mostly retained (Figure 6D, last column). As expected, neither ΔP2 OFF cells nor OFF cells of the parent strain show any priming in LB overnight cultures (Figure 6D, first column). In contrast to the OFF cells, ΔP2 ON cells remain ON in all four culture media indicating that the irreversibility of the system is not affected (Figure 6D, blue bars). We also verified that the deletion of the P2 promoter did not adversely affect the PhoP-P responsiveness of the P1 promoter by comparing YFP reporters driven by the native phoPQ promoter and P1 promoter alone and determining that the two reporters behave similarly at high PhoP-P levels (Figure S6).
We have demonstrated that a single point mutation, causing a T281R substitution in the histidine kinase PhoQ, results in a remarkable change in phenotype, with cells exhibiting bistability/bimodality and an inverted response to [Mg2+]. Previous modeling work indicates that two-component systems have the potential to exhibit bimodal behavior as a result of stochastic fluctuations in molecular components [16], [20], [21] as well as bistability [26], [27]. Bistability has been observed experimentally in the MprB/MprA two-component system [28], although additional feedback loops not present in simple two-component signaling architectures may be responsible for the bistability in that system [29]. Beyond two-component systems, the response of the lac operon to certain gratuitous inducers is another setting where bistability and strong epigenetic inheritance has been observed in bacteria [30], [31]. Like the lac operon, we find the OFF and ON states in PhoQ (T281R) can be inherited over many generations. However, we have not been able to find conditions where ON cells can be deterministically converted to an OFF state. Such irreversibility has been seen in the development of Xenopus oocytes [32] and is more akin to transient or terminal differentiation seen during bacterial competence or sporulation. Both competence and sporulation are highly regulated phenomena [33] and neither is, strictly speaking, a manifestation of a simple positive-feedback based bistable switch [34], [35]. In this sense, irreversibile bistability is a distinctive feature of the PhoQ (T281R) network among characterized two-component signaling systems.
While bistability and noise-induced bimodality are alternate explanations for the observed behavior of the PhoQ (T281R) network, it may be difficult to differentiate between these mechanisms experimentally. First, in our simulations, we observe noise-induced bimodality only for monostable parameters close to the true bistable regime (Text S1). Thus, distinguishing between these mechanisms would likely require careful measurement of kinetic parameters in vivo, which can be challenging even for simple networks. Second, for the mechanism based on noise-induced biomodality, if the timescale for conversion from OFF to ON state in the bimodal regime is tens of generations, then any small fitness difference between the two states can become relevant and even lead to the emergence of bistability, as has been shown in a synthetic circuit [36].
What are the architectural features of the PhoQ/PhoP circuit that enable the T281R mutation to produce such unusual properties? The modeling presented here reveals that the mutant network can readily show bistable behavior when both the kinase and phosphatase activities of the histidine kinase are low (Text S1). A point mutation that simultaneously affects both functions of the histidine kinase can produce a bistable response, as is the case with the T281R mutation, where the kinase activity is low and the phosphatase activity is not detectable. Modeling also indicated that autoregulation of both the response regulator and histidine kinase is required for bistability (Figure 5B). Thus, since phoP and phoQ are organized as an operon that is driven by a PhoP-P responsive promoter, the architecture of the PhoQ/PhoP network (Figure 1A) is poised to show bistability upon introduction of the T281R mutation in PhoQ. Given that bifunctional histidine kinases and autoregulation are common themes in two-component signaling [6], mutations decreasing phosphatase activity in other histidine kinases may similarly lead to bistable behavior.
Since our model does not explicitly incorporate the role of [Mg2+] as an input signal, we can only speculate on the potential mechanism for the inversion of [Mg2+]-sensitivity in the mutant. In terms of our model, the inversion can be explained if higher [Mg2+] increases the kinase rate of PhoQ (T281R) or increases the maximal expression rate of the phoPphoQ operon (potentially through a PhoP-P independent mechanism). Both of these effects could be masked in the wild-type network. For example, an increase in kinase rate at high [Mg2+] could be accompanied by a much larger increase in phosphatase rate in PhoQ (WT) resulting in the observed repression of PhoP-P levels under these conditions. Any increase in transcriptional activity is likely to be unnoticed in the wildtype as the output for the wild-type circuit is relatively insensitive to PhoP and PhoQ protein levels for most ranges of magnesium concentration [13]. We should also note that the reversal of input sensitivity may be an idiosyncrasy of the PhoQ/PhoP system and not a general feature of phosphatase mutants of bifunctional histidine kinases.
The phoQ (T281R) mutant strain exhibits phenotypic hysteresis, that is, the state of a population depends strongly on the culture history (Figure 1D and 2B). OFF cells retain their state in exponential phase, but prime to the ON state upon passage through stationary phase in minimal medium with low [Mg2+] (Figure 2B). In the lac operon, slow growth rate has been shown to increase the transition probability from the uninduced to induced state through the accumulation of inducer and the permease protein LacY [37]. A reduction in growth rate of the phoQ (T281R) strain can similarly cause PhoP and PhoQ to accumulate because of residual expression of the phoPphoQ operon. In addition, PhoP-P would be expected to accumulate from PhoQ (T281R) kinase activity and from the lack of a specific phosphatase. Either mechanism can account for the phenomenon of priming.
Is it possible for strains with wild-type phoQ to show bistability? The wild-type PhoQ/PhoP system responds to lowering Mg2+ concentrations in a unimodal, graded manner [13]. As long as the loss of PhoP-P due to the phosphatase activity of PhoQ (WT) dominates over the reduction in PhoP-P concentration due to growth-mediated dilution, the network is monostable (Text S1 and ref. [13]). It is conceivable that under highly activating conditions, the phosphatase activity of PhoQ (WT) is sufficiently low, in effect, phenocopying the phosphatase defect in phoQ (T281R). Under these circumstances, our analysis (Text S1) suggests that bistability is possible provided that (1) enough PhoP-P can be produced to begin to saturate the PhoP-P mediated feedback expression of the phoPQ operon and (2) the constitutive expression of the operon is sufficiently low (Figure S1B). However, at a Mg2+ concentration of 100 µM, which is an activating stimulus for PhoQ (WT), condition (1) is not satisfied in the wild-type network [38]. It is, therefore, possible that the wild-type circuit has evolved to always maintain a significant level of phosphatase activity to avoid bistability.
The phenotypic heterogeneity associated with bistability can improve fitness in fluctuating or unpredictable environments [39], [40]. Bistability also provides a mechanism for storing information about past environmental conditions, and the persistence of this information can be influenced by the network architecture [41]. In the case of phoQ (T281R), however, bistability comes at a fitness cost because of its irreversible nature. Long-term culture experiments in LB showed that a population of ON cells could be trapped in this low-fitness state. In principle, epigenetic trapping can also occur in a reversible system, but in that case there would be environmental conditions in which the population can deterministically transition to the high-fitness state. In an irreversible system, rare stochastic transitions followed by enrichment due to selection are the only means of escaping from the trap.
Interactions between multiple levels of organization in biological systems can result in unexpected or counter-intuitive phenomena. For example, noise at the molecular level can produce bimodal outcomes in a genetic network that is monostable when analyzed as a deterministic system [42]. Likewise, bistability can emerge in a monostable network because of the influence of the output of the network on the growth rate of the organism [36]. In the case of phoQ (T281R), we find that a bistable genetic network gives rise to a stable and a metastable mode since one of the stable states of the network is inherited efficiently, but has a fitness disadvantage. The phoQ (T281R) mutation also perfectly illustrates a recent suggestion by Kitano that biological networks may be more sensitive to ‘fail-on’ failures where components function in unexpected ways than to ‘fail-off’ failures where components do not function at all or are removed [43]. While a phoQ deletion would be unresponsive to Mg2+ levels, it would not show bistability or epigenetic trapping. In contrast, a single point mutation in the robust PhoQ/PhoP signaling module results in a phosphatase-deficient PhoQ protein that gives rise to bistability and an ON state with pleiotropic alterations in cell physiology.
Strains and plasmids used in this study are listed in Tables S1 and S2 respectively, and details of their construction are included in Text S2. Table S3 lists primers used for strain construction. A figure-wise list of strains used to collect data is included as Table S4. Freshly streaked E. coli strains grown on Miller LB Agar (Fisher BioReagents, Pittsburgh, PA) plates at 37°C were inoculated either in Miller LB broth (Difco - BD, Franklin Lakes, NJ) or in Minimal A medium (MinA, [44]) supplemented with 0.2% glucose, 0.1% casamino acids (Difco - BD, Franklin Lakes, NJ) and with 100 µM, 1 mM or 10 mM MgSO4 as indicated. The phenotype of the colony (YFP-dim or YFP-bright) was always verified prior to inoculation. Overnight cultures were grown at 37°C with aeration in a roller drum and for ∼22.5 hours unless otherwise indicated. These were typically diluted 1000-fold into the indicated medium and grown until most of the cultures reached an OD600 of 0.2–0.3 for microscopy. Some ON state cultures did not reach an OD600 of 0.2 within 4.5 hours of dilution and these were concentrated by gentle centrifugation and resuspension in a smaller volume prior to imaging.
IPTG-induction experiments with the decoupled phoQ (T281R) strain (Figure 6A–B) were performed by overnight culture of the strain in MinA/100 µM Mg2+ for 12 hours at 37°C and diluting the overnight 1000-fold into two MinA/100 µM Mg2+ cultures. After 2.2 hours, one of these cultures was fully induced by adding isopropyl β-D-1-thiogalactopyranoside (IPTG) to a final concentration of 1 mM. Both the induced and uninduced cultures were grown for an additional 2 hours and then washed twice in MinA/100 µM Mg2+ by gentle centrifugation and resuspension. Resuspended cultures were then diluted 105-fold into MinA/100 µM Mg2+ with indicated IPTG concentrations, grown at 37°C for 10 hours and imaged as described below. For comparing the phoPQ reporter and ΔP2 reporter strains (Figure S6), overnight cultures of these strains were grown in MinA/1 mM Mg2+ or MinA/10 mM Mg2+ for ∼22.5 hours at 37°C, diluted 1000-fold into MinA/1 mM Mg2+ or MinA/10 mM Mg2+ with indicated IPTG concentrations and imaged as described below.
Mid-exponential cultures were cooled rapidly in ice-water slurry and streptomycin was added to a final concentration of 250 µg/ml to inhibit further protein synthesis. Imaging was performed on a motorized inverted microscope (Olympus IX81) essentially as described previously [13], [45]. CFP and YFP fluorescence were quantified in single-cells using custom macros written for ImageJ [46]. At least 150 cells were imaged per condition. For quantifying cell-widths, phase images were thresholded and used to segment individual cells. The minor axis of the best-fit ellipse for each identified cell was used as a measure of cell-width.
Overnight cultures were diluted 106-fold in MinA supplemented with 0.2% glucose and 0.1% casamino acids. 100 µl of the diluted culture was spread on 1.5% agar plates composed of MinA with 0.2% glucose, 0.1% casamino acids, and no added MgSO4. These plates were incubated at 37°C for 20 hours and imaged with a home-built fluorescence illuminator as described previously [15].
Freshly streaked colonies of indicated chloramphenicol-sensitive (CmS) and chloramphenicol-resistant (CmR) strains were inoculated in LB medium and grown at 37°C for 20 hours with aeration in a roller drum. At the beginning of the competition, these saturated cultures were mixed 1∶1 by volume and combined cultures were grown for another 10 hours at 37°C. Total and CmR colony forming units (cfus) at the beginning and end of the competition were determined by spreading dilutions on LB Agar plates without antibiotic and with 15 µg/ml chloramphenicol respectively.
The phoQ (T281R) network was mathematically modeled as described in Text S1. Symbolic manipulations and numerical computations were performed using custom code written in MATLAB (Mathworks, Natick, MA) or C programming language. The region of bistability seen in deterministic models is shown in Figures S7, S8, S9, and representative results from stochastic simulations are shown in Figure S5. Parameters and reactions used in stochastic simulations are presented in Tables S5 and S6.
|
10.1371/journal.pgen.1005698 | Developmental Dynamics of X-Chromosome Dosage Compensation by the DCC and H4K20me1 in C. elegans | In Caenorhabditis elegans, the dosage compensation complex (DCC) specifically binds to and represses transcription from both X chromosomes in hermaphrodites. The DCC is composed of an X-specific condensin complex that interacts with several proteins. During embryogenesis, DCC starts localizing to the X chromosomes around the 40-cell stage, and is followed by X-enrichment of H4K20me1 between 100-cell to comma stage. Here, we analyzed dosage compensation of the X chromosome between sexes, and the roles of dpy-27 (condensin subunit), dpy-21 (non-condensin DCC member), set-1 (H4K20 monomethylase) and set-4 (H4K20 di-/tri-methylase) in X chromosome repression using mRNA-seq and ChIP-seq analyses across several developmental time points. We found that the DCC starts repressing the X chromosomes by the 40-cell stage, but X-linked transcript levels remain significantly higher in hermaphrodites compared to males through the comma stage of embryogenesis. Dpy-27 and dpy-21 are required for X chromosome repression throughout development, but particularly in early embryos dpy-27 and dpy-21 mutations produced distinct expression changes, suggesting a DCC independent role for dpy-21. We previously hypothesized that the DCC increases H4K20me1 by reducing set-4 activity on the X chromosomes. Accordingly, in the set-4 mutant, H4K20me1 increased more from the autosomes compared to the X, equalizing H4K20me1 level between X and autosomes. H4K20me1 increase on the autosomes led to a slight repression, resulting in a relative effect of X derepression. H4K20me1 depletion in the set-1 mutant showed greater X derepression compared to equalization of H4K20me1 levels between X and autosomes in the set-4 mutant, indicating that H4K20me1 level is important, but X to autosomal balance of H4K20me1 contributes only slightly to X-repression. Thus H4K20me1 by itself is not a downstream effector of the DCC. In summary, X chromosome dosage compensation starts in early embryos as the DCC localizes to the X, and is strengthened in later embryogenesis by H4K20me1.
| In many animals, males have a single X and females have two X chromosomes, creating an X chromosomal gene dosage imbalance between sexes. This imbalance is corrected by dosage compensation mechanisms, which are essential for development in mammals, flies and worms. The timing and the molecular mechanisms of dosage compensation during development remain unclear. In Caenorhabditis elegans, a complex of proteins called the dosage compensation complex (DCC) binds to and represses transcription of both X chromosomes in hermaphrodites (XX) by half to match X expression to that of males (XO). We found that the DCC starts repressing X chromosomes in early embryogenesis, but average X-linked transcript levels remain higher in hermaphrodite embryos compared to males until the larval stage. Later in embryogenesis, the DCC increases H4K20 monomethylation on the X chromosomes, which is important for dosage compensation. Our results suggest that H4K20 monomethylation does not carry out the transcriptional repressive function by itself but strengthens the DCC’s effect. In summary, the DCC localization starts X chromosome repression in early embryogenesis, and in later embryogenesis sets up a chromatin environment that aids dosage compensation that continues through development.
| Dosage compensation equalizes X chromosome gene expression between sexes. Different animals use different strategies of dosage compensation by co-opting diverse mechanisms of gene regulation to the X chromosome [1]. In mammals, dosage compensation transcriptionally inactivates one of the two X chromosomes in XX females to equalize overall X expression to that of XY males. In Drosophila melanogaster, the X chromosome is transcribed two-fold higher in XY males. In Caenorhabditis elegans, dosage compensation represses both X chromosomes by half in XX hermaphrodites, equalizing overall X-chromosomal transcript levels to that of XO males.
X chromosome dosage compensation is established during, and is essential for development in mammals, D. melanogaster and C. elegans. In mice, failure to inactive the X results in continual deterioration of the embryo and death around 10 days post coitum [2–4]. In D. melanogaster, the dosage compensation complex (male specific lethal (MSL) complex) localizes to the X chromosome at the late blastoderm/early gastrula stage [5,6]. Mutations in any of the four MSL complex members slow development and lead to lethality at the late larval and early pupal stages [7–9]. In C. elegans, mutations in several dosage compensation complex (DCC) subunits are maternal effect lethal, where the progeny of homozygous null mutant worms die at early larval stages [10–12].
Mammalian X inactivation, D. melanogaster MSL complex, and the C. elegans DCC all regulate X chromosome chromatin structure [1,13]. In mammals, X inactivation leads to enrichment of various heterochromatic histone modifications on the inactive X [14,15]. In D. melanogaster, MSL complex increases H4K16 acetylation on the male X chromosome via its histone acetyl transferase subunit MOF [16,17]. In C. elegans, DCC binding is required for H4K20me1 enrichment and H4K16ac depletion on X chromosomes in hermaphrodites [18,19]. Although several histone modifications are associated with dosage compensation, it is unclear if these histone modifications act as downstream effectors of the dosage compensation complexes.
In C. elegans, the five-subunit core of the DCC is a condensin complex, called condensin IDC (hereafter condensin DC) [20]. Condensin DC interacts with additional proteins that have roles in hermaphrodite and X-specific recruitment of the DCC to the X chromosomes [21–24]. Condensins are evolutionarily conserved protein complexes that are essential for chromosome condensation and segregation during cell division (reviewed in [25]). In metazoans, there are two types of condensins, named condensin I and II. In C. elegans, condensin DC is distinguished from condensin I by a single subunit, DPY-27 [20]. Unlike condensin DC, condensin I and II bind to all chromosomes [20,26]. In C. elegans, knockdown of condensin II specific subunit KLE-2 suggested that condensin II is also repressive [26]. Condensins have been implicated in transcriptional regulation in other organisms, but the molecular mechanisms by which condensins regulate transcription remain unknown (reviewed in [27]).
C. elegans DCC is as a clear paradigm for studying the mechanisms of transcription regulation by condensins. The DCC begins localizing to the hermaphrodite X chromosomes at around 40-cell stage (Fig 1A) [22,28]. DCC binding leads to enrichment of H4K20me1 on the X that begins around the 100-cell stage [18,29]. Here, we analyzed X chromosome dosage compensation during C. elegans development by performing mRNA-seq comparison of gene expression between sexes, and in dpy-27 (condensin DC subunit) and dpy-21 (non-condensin DCC member) mutants. Our results suggest that condensin DC starts repression as it localizes to the X chromosomes in early embryos (4–40 cell stage). Gene expression differences between dpy-21 and dpy-27 mutants suggest an additional DCC independent role for dpy-21, emphasizing that condensin DC core and the non-condensin DCC members functionally differ. To clarify the role of H4K20me1 in X chromosome dosage compensation, we performed mRNA-seq and H4K20me1 ChIP-seq analyses in set-1 (H4K20 monomethylase) and set-4 (H4K20 di-/-tri methylase) mutant worms. Our results suggest that H4K20me1 levels are important for X chromosome dosage compensation, but H4K20me1 by itself does not act as a downstream effector of the DCC.
We measured X chromosome dosage compensation in early embryos, when zygotic transcription is active but the DCC is not yet fully localized to the X chromosome. Here, we measured dosage compensation of a gene as the ratio of its mRNA levels between hermaphrodites and males. We generated mRNA-seq datasets from 2–5 biological replicates (S1 File) that highly correlated (S1 Fig). Because of the difficulty in obtaining pure male populations prior to the onset of sex specific gene expression, we compared hermaphrodite embryos (XX) to mixed sex (XX and XO) embryos. Mixed sex embryos were isolated by crossing males and hermaphrodites of obligate out-crossing strain (fog-2(oz40)) to ensure a 50% male population. “Early embryos” are between 4-cell and 40-cell stages (S2 Fig). At this stage, zygotic transcription is activated and the DCC starts localizing to but not fully enriched on the X chromosome in all cells (Fig 1A) [22,29].
In early embryos, X chromosome expression was higher in hermaphrodites compared to mixed sex (Fig 1B). The median log2 ratio of expression between hermaphrodite and mixed sex embryos was significantly higher on the X (0.182) compared to autosomes (ranges from 0.034 to 0.052) (Fig 1C, one sided Wilcoxon rank-sum test p < 1.77 x 10−9). Additionally, genes with significantly higher expression in hermaphrodites were enriched on the X (Fig 1D, Fisher’s exact test p = 3.12x10-6). Thus, X chromosome is not fully dosage compensated between sexes in early embryos, but X chromosomal transcripts are fairly similar between sexes, as the observed median log2 expression ratio of 0.182 is less than expected if X expression were two-fold higher in XX hermaphrodite embryos in the XX/mixed sex comparison (0.411).
One mechanism that might equalize X chromosomal transcript levels between sexes is the maternal loading of mRNAs into oocytes. To test this, we asked if similarly expressed genes are maternally loaded. While most methods identify genes that are differentially expressed between sexes, a lack of differential expression does not show that genes are statistically similarly expressed. We identified 780 genes that are statistically similarly expressed between hermaphrodite and mixed sex early embryos by taking genes that show less than 30% difference in expression within a 95% confidence interval (see Materials and Methods for full explanation). Among the 36 similarly expressed X-chromosomal genes, 35 were maternally loaded (FPKM>1 in oocytes [30]). Approximately 95% of similarly expressed genes on autosomes were also maternally loaded, thus maternal loading of transcripts into oocytes helps equalize gene dosage between hermaphrodite and male embryos (Fig 1E).
To specifically study dosage compensation of zygotic gene expression, we identified genes that are not expressed in 2-cell stage embryos but are newly expressed in early embryos. K-means clustering of expression between previously published expression data in 2-cell hermaphrodite embryos [30] and our data in early embryos resulted in five distinct clusters (Fig 2A). Genes in cluster 3 are zygotic because they showed little or no expression in 2-cell embryos and increased expression in early embryos (Fig 2B). Since we cannot distinguish between stable and newly transcribed mRNAs, genes in the remaining clusters may also be zygotically expressed. Thus, to study zygotic expression, we focused on the newly expressed genes (cluster 3). Hermaphrodite expression from the X chromosome was significantly higher compared to the autosomes (one sided Wilcoxon rank sum test, p = 0.019), but the difference was less than expected if there were no dosage compensation. Therefore, early zygotic expression from the X is dosage compensated, but not completely (Fig 2C).
In C. elegans, incomplete dosage compensation in early embryos is logical, as sex is determined by the ratio of X and autosomal sex elements (XSEs and ASEs) [31–33]. In early embryos, we observed that the male-specific her-1 gene is expressed 11.2 fold higher in mixed-sex samples compared to hermaphrodites (Fig 2D). This corresponds to an estimated 17 fold higher expression in males, consistent with previous measurements of ~20-fold difference in her-1 expression between males and hermaphrodites [34]. Interestingly, male promoting gene xol-1 transcript was only slightly higher in mixed sex samples (Fig 2D). It is possible that the difference in endogenous xol-1 transcript level between sexes is smaller than seen for transgenes [35]. It is also possible that the X-linked xol-1 is being repressed by the DCC in early embryos, as dpy-27 starts to repress the X chromosomes in early embryos (Fig 2E), and xol-1 was shown to be repressed by the DCC [36]. We also found that mRNA levels of both XSEs and ASEs were higher in hermaphrodites, suggesting that XSE and ASE dosage does not function through a simple ratio of mRNA levels between sexes (S3 Fig). Instead, small and transient differences in the transcription of XSEs, ASEs and xol-1 may be amplified post-transcriptionally to ultimately regulate her-1 expression, which is clearly higher in males (Fig 2D).
The DCC is coupled to sex determination through sdc-2, which is specifically expressed in hermaphrodite embryos [22]. SDC-2 is required for DCC localization to the X, and for repressing male-specific her-1 gene [37]. A recent study showed that in 50-cell stage embryos, half the cells contain DPY-27 specifically on the X chromosomes [29]. In our early embryo collections (4–40 cell stage), sdc-2 is more highly expressed in hermaphrodites and her-1 is expressed in males, therefore the DCC is activated (Fig 2D). To test if the DCC represses X-chromosome expression shortly after localizing to the X, we collected dpy-27 null mutant embryos using a strain with genetically balanced (y56) allele (Materials and Methods). In early embryos, dpy-27 null mutation caused significant derepression of newly transcribed zygotic genes on the X (Fig 2E), suggesting that dpy-27 represses X chromosomes in early embryos. Dpy-27 is not detectable in the XX hermaphrodite germline by immunostaining, but it is maternally deposited and required for development [38–40]. Therefore, we interpret the observed dpy-27 null mutation effect coming from the embryos that loaded the DCC onto the X chromosomes in some cells by the 40-cell stage. This implies that the DCC is able to repress X chromosomes quickly after loading.
To determine if and when average X chromosome expression is equalized between sexes, we analyzed further time points including comma stage embryos, L1 and L3 larval stages, and young adults. At the comma stage, X chromosome expression remained slightly but significantly higher in hermaphrodites compared to mixed sex worms (Fig 3A). In, L1, L3 larvae and young adults, average X-chromosome expression was no longer higher in hermaphrodites. In addition to measuring dosage compensation at each time point, for each gene we also calculated change in dosage compensation between two consecutive time points. From early to comma stage embryos, the X chromosome was slightly repressed, but X chromosome gene expression became significantly more balanced between sexes after the comma stage (Fig 3B, Wilcoxon rank sum p = 1.13 x 10−63). To address if more genes are dosage compensated after the comma stage, we identified the statistically similarly expressed genes at each developmental stage. The percentage of X chromosomal genes that were statistically similarly expressed between sexes increased from early to comma stage embryos (4.6% to 12.8%, respectively), and stayed at similar levels in L1, L3 and young adults (10.3%, 10.4%, 13.5%, respectively). Thus, on average more X chromosomal genes are repressed after the comma stage (Fig 3B), but the proportion of equivalently expressed X chromosomal genes increases after early embryogenesis.
In comma stage embryos, average X expression may be higher in hermaphrodites due to stable transcripts retained from early embryos. To test this, we used K means cluster analysis and further filtering to identify those genes that are newly expressed in comma stage embryos and in L1 larvae (Fig 3C). The newly expressed X chromosomal genes in L1 larvae were dosage compensated and repressed by the DCC (Figs 3C and S5). In comma stage embryos, the newly expressed X chromosomal genes showed significantly higher hermaphrodite expression compared to autosomal genes, thus transcript stability does not fully explain significant hermaphrodite-biased expression from the X chromosomes. It is possible that in comma embryos, X is expressed higher in hermaphrodites because DCC-mediated dosage compensation is incomplete or more genes on the X chromosome have hermaphrodite-biased expression.
Comparison of gene expression between sexes is a measure of dosage compensation, but sex-biased gene expression confounds the interpretation of the data (see discussion). To specifically study DCC mediated X repression, we analyzed gene expression changes in hermaphrodites mutant for dpy-27 or upon dpy-27 RNAi knockdown. As mutants and RNAi treated worms showed more variability in staging, we mainly used “mixed stage embryos” isolated by bleaching gravid adults. Mixed stage embryos contained 100–300 cells (S4A Fig). We used dpy-27 RNAi in mixed embryos and L3, because of the difficulty collecting dpy-27(y56) null mutant. Western blot analyses showed ~70% and ~40% knockdown of DPY-27 in embryos and L3s, respectively (S4B Fig). Mutation (Fig 3D) or depletion (Fig 3E) of dpy-27 caused significant X chromosome derepression in early and mixed-stage embryos, L1 and L3 worms. Analysis of newly expressed genes in each stage also indicated that condensin DC represses X chromosomes throughout these developmental stages (S5 Fig). Therefore, although previous studies that used temperature sensitive mutants of various DCC subunits found that the critical time period for DCC activity is centered around the comma stage of embryogenesis, our results suggest that the DCC acts throughout development [12].
We also analyzed dpy-21, a non-condensin subunit of the DCC, whose null mutation (e428) is not lethal, but leads to X chromosome dosage compensation defects including a dumpy phenotype [10,11,41]. In mixed embryos and L3, dpy-21(e428) mutation also caused X chromosome derepression (Fig 3F). In early embryos, dpy-21(e428) had a relatively less X-specific effect compared to dpy-27(y56). In dpy-21(e428) early embryos, 72% of X and 58% of autosomal genes were significantly upregulated. In dpy-27(y56) early embryos, DESeq analysis did not identify any genes that met the significance cutoff (adjusted p value < 0.05). But, among the top five percent of dpy-27 regulated genes, 81% of X and 31% autosomal genes were upregulated. The effect of dpy-21(e428) on X chromosome expression was more specific in the later stage worms. Clustering and heatmap analysis of gene expression changes showed that dpy-21(e428) caused both increased and decreased expression from the X chromosomes in early embryos, whereas majority of the X chromosomal genes were derepressed in comma stage embryos and L3s (S4C Fig). A recent study noted that regulation of H4K16 acetylation on the X chromosomes differs between dpy-21 and other DCC subunits in early embryos [29]. Since regulation of X chromosome expression also differed between dpy-27(y56) and dpy-21(e428) mutants, dpy-21 may have a DCC-independent role in early embryos.
An earlier microarray study suggested that approximately half of X-chromosomal genes are dosage compensated. These genes were identified by the criteria that they were upregulated in the DCC mutants and were equally expressed between XX and XO hermaphrodite embryos [42]. Subsequently, global run on analysis (GRO-seq) of active transcription in sdc-2(y93,RNAi) embryos showed a uniform increase in RNA Pol II levels across most X chromosomal genes [43], suggesting that there is no large distinct set of genes that escape from DCC-mediated repression. In agreement with a continuous effect, our analysis of the published GRO-seq data showed similar upregulation between previously categorized dosage compensated and non-compensated genes (Fig 4A). In mixed stage embryos, dpy-27 RNAi mRNA-seq also showed derepression of X chromosomal genes. Derepression was slightly stronger for previously defined compensated genes, suggesting that posttranscriptional regulation contributes to the final mRNA levels in the DCC mutants.
To further test if the DCC acts on a distinct set of genes on the X, we analyzed the distribution of expression changes caused by DCC disruption. We used a standard expectation-maximization algorithm that mixes multiple normal distributions to model the overall pattern of expression changes. We reasoned that if there is a large group of genes specifically repressed by the DCC, when two distributions are forced on the data, one distribution should reflect expression of DCC regulated genes and another should reflect genes not regulated by the DCC. Indeed, when applied to the whole genome GRO-seq data, two distributions can be modeled. Separation of two distributions with average fold changes of -0.37 and 0.69 closely mirrored the distributions of changes for the autosomes and X chromosome (Fig 4B). This indicated that the analysis is sensitive to finding differentially regulated genes by the DCC. Next, we applied the same method to the X chromosome, using GRO-seq data (Fig 4C), and the mRNA-seq data from dpy-27(y56) L1s (Fig 4D). In both cases, the separated distributions all showed upregulation, suggesting that there are no large group of genes that escape DCC regulation on the X. This does not exclude the possibility that different groups of genes are affected at different levels, but this effect must be continuous, rather than discrete.
In C. elegans, H4K20 is monomethylated by SET-1, and H4K20me1 is converted to di- and tri-methylation by SET-4 [18,19]. Concurrent enrichment of H4K20me1 and depletion of H4K20me3 on the X chromosomes suggested that the DCC increases H4K20me1 levels on the X by reducing SET-4 activity [18] (Fig 5A). X-enrichment of H4K20me1 requires the DCC subunits, including dpy-21 [18]. We analyzed genome-wide expression changes that occur in dpy-21(e428), set-1(tm1821) and set-4(n4600) null mutants [18,19]. Set-1 null mutant is maternal effect sterile, and RNAi knockdown of set-1 leads to germline deficient adults, thus we could not isolate embryos. Therefore, we collected set-1(tm1821) homozygous and heterozygous L3 larvae (see Material and Methods). Western blot analysis in whole-larval extracts showed expected changes in H4K20me1 (Fig 5B) [18,19]. H4K20me1 reduced in dpy-21(e428), increased in set-4(n4600), and was eliminated in set-1(tm1821). Immunofluorescence analysis of intestinal nuclei had also shown that H4K20me1 is higher on X in N2 wild type worms; decreases to the autosomal levels in dpy-21(e428) mutant; and increases and becomes equal between X and autosomes in set-4(n4600) mutant [19].
H4K20me1 reduction due to set-1 or dpy-21 null mutation led to a significant increase in X chromosome expression compared to autosomes in L3 larvae (Fig 5C). We observed a small but significant increase in X expression compared to autosomes in the set-4(n4600) mutant L3 (Fig 5C) and mixed stage embryos (Fig 5D). Expression of previously defined dosage compensated genes [11,23,42] were increased in dpy-21(e428), set-1(tm1821), but not as much in set-4(n4600) mutant (Fig 5E). To test if a subset of X chromosomal genes were responsible for the effect seen in the set-4(n4600) mutant (Fig 5C), we plotted the distribution of expression ratios on the X and autosomes (Fig 5F). This indicated a slight shift between X and autosomes, suggesting that set-4(n4600) has a subtle effect across all genes, rather than a large effect on a subset of X chromosomal genes. Note that mRNA-seq measures relative expression changes, rather than absolute. Therefore X derepression seen in the set-4(n4600) mutant may not be specific derepression of the X, as discussed later in the results. Nevertheless, the effects of set-1 and set-4 mutations indicate that proper regulation of H4K20 methylation is important for X chromosome dosage compensation.
GRO-seq analysis of sdc-2(y93, RNAi) embryos showed that the DCC reduces RNA Pol II at the X chromosome promoters [43]. To test if the expression changes seen in dpy-21(e428) and set-4(n4600) mutants are due to transcription, we performed AMA-1 (RNA Pol II large subunit) ChIP-seq analysis. Average RNA Pol II enrichment across the transcription start and end sites showed a 3’ accumulation that was also noted in the GRO-seq analysis for C. elegans genes (Fig 5G) [43]. RNA Pol II on the X chromosome promoters increased relative to autosomes in dpy-21(e428) mutant (Fig 5G). In the set-4(n4600) mutant (Fig 5G), there was a subtle shift in RNA Pol II levels between the X and autosomes, suggesting that the effect of set-4 is also at the level of transcription. These results do not exclude post-transcriptional effects contributing to increased X expression in the dpy-21(e428) and set-4(n4600) mutants, but do suggest that dpy-21 and set-4 act at the level of transcription.
Unlike other DCC subunits, dpy-21 is not essential, but is required for H4K20me1 enrichment on the X chromosomes [11,18,19]. To understand the role of dpy-21 in the DCC, we compared gene expression changes in dpy-27 RNAi, dpy-21(e428), set-1(tm1821) and set-4(n4600) mutant L3 larvae by clustering differential expression ratios on the X chromosome and autosomes (Fig 6A). We note that while the effect of dpy-27 RNAi and dpy-21(e428) were generally similar on the X, genes in several clusters were affected differently between dpy-27 RNAi and dpy-21(e428). We performed gene ontology (GO) analyses comparing enriched functions in each cluster compared to the whole genome (Fig 6B). Interestingly, the effect of dpy-27 RNAi and dpy-21(e428) on clusters enriched for cuticle genes were opposite, yet both mutations result in a dumpy phenotype.
Since our model for DCC regulation of H4K20me1 posits that the DCC reduces SET-4 activity, X derepression in the set-4 mutant was somewhat unexpected. To study this further, we performed H4K20me1 ChIP-seq analyses in wild type, dpy-21(e428) and set-4(n4600) mutant embryos and L3 larvae. Experimental replicates correlated well with each other (S6A Fig). H4K20me1 is enriched in active gene bodies in multiple species [44]. In C. elegans hermaphrodite L3s, H4K20me1 is enriched on expressed genes, but also across the entire X chromosome including silent genes and intergenic regions (Fig 7A). Such pattern of H4K20me1 enrichment on the X could be achieved by a uniform deposition of H4K20me1 by SET-1 during M phase, followed by DCC-mediated reduction of SET-4 activity across the X chromosome [18]. However, standard ChIP-seq analysis cannot detect a uniform increase, due to the analysis step that normalizes read depth between experiments [45,46]. Therefore, to analyze change in H4K20me1 in the absence of SET-4, we used C. briggsae ChIP extract as a spike-in control, similar to previous approaches (Fig 7B) [45,47]. The replicates correlated with each other (S6B Fig). Here, we assume that the ChIP-seq enrichment calculated from the C. briggsae reads should be the same between set-4(n4600) and wild type, since the spiked in C. briggsae extract is the same. For each ChIP sample, the corresponding input was also sequenced, thus the normalization is internally controlled against any potential difference in the spiked in proportion. We normalized H4K20me1 ChIP enrichment from C. elegans reads to that of C. briggsae reads, and calculated the ratio of H4K20me1 levels between set-4(n4600) and wild type worms (Fig 7B).
In set-4(n4600) L3s, there was ~10-fold increase in H4K20me1 on autosomes and ~6 fold increase on the X chromosome (Fig 7C). This is consistent with the idea that in the wild type worms, the DCC already reduces SET-4 activity on the X by half, thus in the absence of SET-4, H4K20me1 increase on autosomes is approximately two-fold more compared to the increase on the X. Immunofluorescence studies in set-4 null mutants also showed an overall H4K20me1 increase resulting in the equalization of X and autosomal H4K20me1 levels [18,19]. Plotting H4K20me1 enrichment as measured by standard ChIP-seq analysis across transcription start sites illustrates the equalization of H4K20me1 levels between X and autosomes in dpy-21(e428) and set-4(n4600) mutants (Figs 7D and S7A). In these plots, ChIP-seq experiments were performed without the spike-in control, and the genome-wide increase of H4K20me1 in the set-4(n4600) mutant was masked. Adjusting enrichment values by the spike-in ratios obtained from Fig 7C illustrates the inferred H4K20me1 levels in the set-4(n4600) mutant L3s (lighter colors in Fig 7D). The ten-fold increase in H4K20me1 levels across the genome was surprising, based on the observation that western blot analysis of H4K20me1 showed approximately five-fold increase in the set-4 null mutant L3s (Fig 4B). The basis of this difference between the two different assays (ChIP versus western blot) is unclear. A ten-fold increase of H4K20me1 may be possible in the light of previous studies in mammalian cells, which showed that ~80% of H4 is di-methylated at H4K20 [48,49]. If demethylation of H4K20me1 is inefficient in C. elegans, set-4 mutation may result in high levels of monomethylation.
The magnitude of H4K20me1 increase and the difference between X and autosomes was greater in L3s compared to mixed stage embryos in set-4(n4600) mutant (S7B Fig). This may be due to less H4K20me1 X-enrichment in embryos compared to L3s, consistent with H4K20me1 enrichment occurring later in embryonic development [18,29]. Incomplete X-enrichment of H4K20me1 in mixed stage embryos was also reflected by less H4K20me1 ChIP-seq enrichment on the X chromosome in wild type embryos compared to L3 (Fig 7E, left panel).
The delayed H4K20me1 enrichment may also cause a delay in full dosage compensation. X chromosome expression was not fully balanced between sexes during embryogenesis (Fig 3A). While this may be due to differences in sex-biased expression between developmental stages (see discussion), it may also be due to slightly less dosage compensation due to incomplete enrichment of H4K20me1 on the X chromosomes until the comma stage of embryogenesis. Indeed, X chromosome derepression upon dpy-27 depletion was slightly but significantly higher in L3 larvae compared to mixed stage embryos (Fig 3E, Wilcoxon rank sum test p value = 5x10-7). This was in spite of having more efficient DPY-27 knockdown in embryos compared to L3 larvae (S4B Fig). X-enrichment of H4K20me1 may slightly strengthen DCC mediated X chromosome repression in later development.
The small but significant X derepression observed in the set-4(n4600) mutant suggested that X/A balance of H4K20me1 contributes slightly to X chromosome dosage compensation (Fig 4C). To address the mechanism of this contribution, we further analyzed gene expression in the set-4(n4600) mutant. Since mRNA-seq measures relative expression changes, we could not distinguish if the relative X derepression seen in the set-4(n4600) mutant is due to repression from autosomes or an increase from the X. Therefore, we adapted the spike-in normalization scheme for mRNA-seq analysis (Fig 8A). To do this, we mixed 1,000 C. briggsae worms to 10,000 N2 (wild type) or set-4 mutant worms before RNA extraction and mRNA-seq analysis. We used fed L1 worms where we could reliably assume that the number of cells was the same between wild type and the mutant. The replicates correlated well with each other (S8A Fig). Since spiked in C. briggsae worms were the same across all samples, we normalized C. elegans expression levels using the linear regression coefficients that equalized C. briggsae expression between wild type and mutant samples (S8B Fig).
Spike-in analysis of mRNA-seq levels in the set-4(n4600) mutant showed decreased expression from autosomes (Fig 8B), resulting in the relative effect of X derepression seen in Fig 4C. To validate the spike-in mRNA-seq method, we compared it to RT-qPCR analysis for a number of genes. RT-qPCR was performed on the same total RNA preparations that were used for making the mRNA-seq libraries. Since RT-qPCR also relies on the assumption that total mRNA levels are the same between the wild type and the mutant, two C. briggsae genes were used as standard. Results from standard and spike in normalized RT-qPCR results were similar to mRNA-seq analysis (Fig 8C).
We reason that if H4K20me1 acts as a downstream effector of the DCC (i.e. the effect of H4K20me1 on transcriptional repression was the same on the X and autosomes), equalization of H4K20me1 between the X and autosomes should result in the same level of X chromosome derepression. Two observations suggested that while H4K20me1 is important, X/A balance of H4K20me1 is not responsible for X chromosome repression. First, upon equalization of H4K20me1 levels between X and autosomes in the set-4(n4600) mutant, X chromosome is only slightly derepressed (Fig 5C). Second, H4K20me1 depletion in the set-1(tm1821) mutant resulted in greater X derepression compared to equalization of H4K20me1 between X and autosomes in the set-4(n4600) mutant (Fig 5C). Thus H4K20me1 does not act as the sole downstream effector of the DCC. Based on the spike in controlled ChIP-seq and mRNA-seq analyses in the set-4(n4600) mutant, we propose a model that H4K20me1 enrichment only slightly represses X chromosome transcription, but H4K20me1 may positively affect DCC activity, which is primarily responsible for X repression (Fig 8E).
Our results indicate that in early embryogenesis, X chromosome is not fully dosage compensated in C. elegans. Then, how is the X chromosome dosage difference tolerated? First, maternal loading of mRNAs into oocytes help equalize overall X-chromosomal transcript levels between the two sexes, thus genes that are important for the development of both sexes may be provided in the oocyte. Second, X chromosome harbors fewer genes with essential functions for embryogenesis [50]. Third, dosage sensitive genes such as haploinsufficient genes are depleted from the X chromosome [51,52]. We also found that C. elegans X chromosome has fewer genes that encode for multi-subunit protein complexes, which were found to be more dosage sensitive [53–55]. There are only 16 X-chromosomal genes with a yeast homolog of a multi-protein complex [52], representing a significant depletion of such genes from the X (p< 10−5, Fisher’s exact test). For the remaining dosage sensitive genes on the X, in addition to DCC mediated repression, potential post-transcriptional mechanisms may act, as such regulation is extensive in C. elegans [56]. Thus, a combination of different mechanisms buffers the X chromosome dosage difference between hermaphrodite and males during early embryogenesis.
Comparison of gene expression levels between hermaphrodite and male embryos reflects two genetic differences: X chromosome number and sex-biased gene expression. Others and we have shown that the X chromosome contains higher numbers of hermaphrodite-biased genes [51,57]. This may contribute to the higher X expression in hermaphrodites compared to the mixed sex embryos (Fig 3). However, most sexual dimorphism in C. elegans develops in the larval stages (reviewed in [58]), and the adult gonads contribute an especially high amount of sex-biased gene expression [51]. Indeed, mRNA-seq replicates in the hermaphrodite and mixed-sex worms clustered by developmental stage rather than sex until the young adult stage (S1A Fig). Thus, in embryos and larvae, developmental gene expression has a stronger effect on the transcriptome than sex biased gene expression. Nevertheless, there is sex-biased gene expression in embryos, including genes in the sex determination pathway (Fig 2D). Interestingly, the XSEs were dosage compensated during late embryogenesis, suggesting that the DCC acts on these genes regardless of their early hermaphrodite-biased expression [36] (S3 Fig). It is possible that the DCC mediated gene repression is chromosome-wide and superimposed onto other mechanisms of regulation that act at the level of individual genes, such as sex biased gene expression programs. Future studies that uncouple sex from X chromosome number are needed to resolve the relation between dosage compensation and sex-biased gene expression.
Although X chromosome dosage compensation is essential for proper development, it is still not clear how DCC mutations cause lethality (reviewed in [1]). In many animals, proper chromosome dosage is important, as trisomy or monosomy of chromosomes is not tolerated (reviewed in [59]). X chromosome dosage compensation problems in mammals and worms would cause upregulation of X-chromosomal genes. In the case of trisomies, increased expression from a set of dosage sensitive genes may cause problems, as in Down syndrome (reviewed in [60]). Additionally, increased expression from a whole chromosome pressures proteasome pathways needed to degrade the extra proteins, as was shown in yeast [61]. DCC mutations cause derepression of many X chromosomal genes that are involved in wide range of processes (Fig 6). Genes that are upregulated in the dpy-27 mutant include several transcription factors and kinases, whose overexpression may cause off-target effects (discussed in [62]). While it is possible that the combination of a number of dosage sensitive genes on the X are responsible for the DCC phenotypes, DCC-mediated repression is applied across the X chromosome (Fig 4), suggesting that the DCC evolved to repress average X chromosomal gene dosage.
DCC starts binding to the X chromosomes at around the 40-cell stage, which is followed by H4K20me1 enrichment on the X, which occurs between the 100-cell and comma stages [18,29]. Since the DCC reduces SET-4 activity, and SET-4 needs previous monomethylation of H4K20 to catalyze di- and trimethylation, H4K20me1 enrichment on the X can only happen after the initial deposition of H4K20me1 during cell division. The delay in H4K20me1 enrichment during embryogenesis may be due to a requirement for cell divisions to initially deposit H4K20me1. In mammalian tissue culture cells, cell-cycle regulation of SET-1 ensures a dramatic increase in H4K20me1 on mitotic chromosomes [63,64]. H4K20me1 also increases on mitotic chromosomes in C. elegans [18,29]. Therefore, it is possible that multiple divisions are necessary to successively increase H4K20me1 on the C. elegans X chromosomes. A microscopy-based study of H4K20me1 enrichment across multiple divisions in one cell lineage can address this possibility.
Previously, we hypothesized that the DCC increases H4K20me1 by reducing SET-4 activity on the X chromosomes [18]. Analysis of gene expression in the set-4 null mutant suggested an effect on X chromosome dosage compensation (Fig 5C). This effect was indirect, as an average of ~20% reduction across the autosomes resulted in a relative X chromosome derepression compared to autosomes (Fig 8B). Since H4K20me1 increase in the set-4 null mutant resulted in repression, H4K20me1 enrichment on the X chromosome may slightly contribute to repression in hermaphrodites. H4K20 methylation has a role in gene expression in other organisms as well, but the mechanisms remain unclear [65].
It is possible that H4K20me1 positively regulates chromatin compaction, thus reduces accessibility of RNA Polymerase II to the X chromosomes. H4K20me1 is enriched on mitotic chromosomes and the inactive X chromosome in mammals [65]. In C. elegans, H4K20me1 is enriched on mitotic chromosomes and across the entire X (Fig 7A), which is more compact in hermaphrodites [66]. Both set-1 and set-4 resulted in a relative increase in X chromosome volume compared to the autosomes in hermaphrodites [66]. While set-4 is important for X compaction, the null mutant showed only a slight effect on X chromosome dosage compensation and does not have obvious dosage compensation phenotypes in standard laboratory conditions (Fig 5C). Therefore, X compaction by itself does not fully explain DCC mediated X chromosome repression.
Regulation of H4K20 methylation by set-4 may be involved in additional processes. Set-4 knockdown increased lifespan in an RNAi screen [67] and suppressed the developmental delay associated with mutant rict-1, a component of the target of rapamycin complex 2 [68]. Gene expression changes that occur in the set-4 mutant may be the basis for these phenotypes. Alternatively, set-4 may act through other non-histone substrates, as recent experiments in D. melanogaster showed that H4K20A mutation did not completely recapitulate the phenotype of the methyltransferase mutant [69].
Our working model for H4K20me1 in X chromosome dosage compensation is summarized in Fig 8E. Briefly, X-enrichment of H4K20me1 contributes slightly to X chromosome repression, but H4K20me1 by itself is not a major downstream effector of the DCC. Instead, H4K20me1 level is important for DCC-mediated repression of the X chromosomes. One possibility is that H4K20me1 may increase association of condensin DC with chromatin. In agreement with this possibility, there was slightly less DPY-27 binding in the dpy-21(e428) mutant embryos where H4K20me1 is no longer enriched on the X chromosome (S7C Fig). In mammalian cells, HEAT domains found in N-CAPD3 and N-CAPG2 subunits of condensin II bind to H4K20me1 containing histone tail peptides, and this interaction was proposed to increase condensin II binding on mitotic chromosomes, which are enriched for H4K20me1 [70]. The DCC contains two subunits with HEAT domains, DPY-28 and CAPG-1 [20]. Positive regulation of DCC by H4K20me1 may help maintain dosage compensation complex on the X in later developmental stages in C. elegans. Such maintenance mechanisms are involved in epigenetic gene regulation, and have been demonstrated to be important for mammalian X inactivation [71,72].
Previous work and our results suggest that the DCC represses X chromosome transcription. The molecular mechanism by which the DCC represses transcription remains unclear. In various organisms, condensins demonstrate activities such as positive supercoiling [73–75], single stranded DNA reannealing [76,77] and making topological links in DNA [78,79]. It is not known if condensin DC has these activities.
The DCC is enriched at active promoters [80] and the DCC reduces RNA Pol II binding to X chromosome promoters [43]. The DCC ChIP binding scores at promoters positively correlate with the amount of transcription [81]. Positive correlation between condensin binding and transcription was also seen in yeast and mammalian cells [77,82–84]. We had hypothesized that coupling the level of DCC binding to the level of transcription at each promoter could be the basis of the two-fold reduction in gene expression regardless of the expression level [80]. Indeed, GRO-seq analysis indicated that while the DCC binding positively correlates with transcription, the level of DCC mediated reduction in RNA Pol II is similar across genes transcribed at a wide range of levels [43]. Based on the observation that the DCC binds to promoters, and that yeast condensin interacts with TBP [83], it is possible that the DCC is recruited to promoters by a general transcription factor, and then halves the rate of RNA Pol II binding at each cycle of transcription.
Alternative but not exclusive DCC mechanisms have been proposed, such as compacting chromatin to reduce activators’ access to the X chromosome or changing interactions between enhancer and promoters to reduce RNA Pol II recruitment [43]. These models are based on the observations that condensins are required for chromosome condensation, and in other organisms such as D. melanogaster and mammalian cells, condensins regulate long-range chromosomal interactions [85,86]. Recently, the DCC was shown to compact the X chromosome [27], regulate sub-nuclear localization of the X [87], and form topologically associated domains on the X [88]. However, it remains unclear if these large-scale chromatin changes are the cause or consequence of increased transcription from the X chromosomes in the DCC mutants.
In mouse, C. elegans and D. melanogaster, dosage compensation complexes are targeted to and regulate the chromatin structure and transcription from the X chromosomes in one of the two sexes (reviewed in [1]). The composition of the dosage compensation complexes suggests that different complexes coopted different sets of gene regulatory mechanisms to the X chromosomes. In the case of C. elegans, duplication and divergence of a condensin subunit (dpy-27) created the dosage compensation specific condensin complex (condensin DC) that reduces transcription from both X chromosomes in hermaphrodites. Condensins are essential for chromosome condensation during cell division, and regulate chromatin compaction and transcription during interphase. H4K20me1 is a mitosis enriched chromatin mark that is also enriched on the dosage compensated X chromosomes in C. elegans. Our work on the DCC and H4K20me1 contributes to understanding the link between the mitotic and interphase function of condensins in chromosome condensation and gene regulation.
N2, wild type; BS553, fog-2(oz40); CB428, dpy-21(e428)V; MK4 dpy-27(y56)/qC1[dpy-19(e1259) glp-1(q339)] [qIs26] III; SS1075 set-1(tm1821)/hT2g[bli-4(e937) let-?(q782) qIs48] (I;III); MT14911 set-4 (n4600) II; VC199, AF16 C. briggsae wild isolate.
Unless otherwise noted, strains were maintained at 20°C on NGM agar plates using standard C. elegans growth methods. For staged embryo collections, worms were synchronized by bleaching gravid worms, allowing embryos to hatch overnight and growing the synchronized L1s on NGM plates for ~60 hours until the first embryos were seen in hermaphrodite gonads. Early embryos were isolated from bleached young adults, and immediately collected. Comma embryos were obtained by aging early embryos 4 hours in M9 buffer. Larvae were collected after growing hatched L1s for 6 hours (L1 larvae) or 24 hours (L3 larvae) on NGM plates. Young adults were synchronized by growing hatched L1s for 44 hours at 25°C, and collected prior to the accumulation of fertilized embryos in the gonad. Mixed stage embryos were collected by bleaching gravid adults. Developmental stages were assessed by counting DAPI stained nuclei by DAPI staining, assessing germline, male tail and vulva morphology.
Due to lethality, set-1(tm1821) mutants were kept heterozygous with a balancer chromosome that includes a GFP marker (SS1075 strain [18]). set-1/balancer heterozygous larvae were selected by picking GFP(+) animals and set-1 homozygous mutant larvae were identified by picking GFP(-) animals. Since dpy-27(y56) is lethal [12,28], dpy-27(y56) worms were kept as heterozygotes with a balancer chromosome that encodes rol-6(su1006). dpy-27(y56) homozygote embryos were collected by bleaching young adults without the roller phenotype. This resulted in embryos that lacked both maternal and zygotic dpy-27.
For DPY-27 RNAi, the bacterial strain from the Ahringer RNAi library containing dpy-27 RNAi plasmid and control plasmid were grown and induced in 100 ml LB media for 3 hours with 0.1 mM ITPG, concentrated 10-fold and seeded onto a 10 cm NGM plate supplemented with ampicillin, tetracycline and 1mM IPTG. Synchronized L1s were placed on seeded plates and collected at 24 hours (L3 larvae) or grown to gravid adults and embryos were collected by bleaching (mixed stage embryos). At collection time, all samples were washed 3 times in M9 buffer and stored in ten volumes of Trizol (Invitrogen) at -80°C. 2–3 collections were pooled for one biological replicate.
A summary of all the data sets and replicates are provided in S1 File. Raw data files and wiggle tracks of ChIP enrichment per base pair, and RNA-seq FPKM values are provided at Gene Expression Omnibus database (http://www.ncbi.nlm.nih.gov/geo/) under accession number [GSE67650].
For RNA preparation, samples in Trizol were frozen and thawed 3–5 times to freeze-crack worms, and phase containing the total RNA was cleaned up using the Qiagen RNeasy kit. mRNA was purified using Sera-Mag oligo(dT) beads (ThermoScientific) from 0.5–10 ug of total RNA. Stranded mRNA-seq libraries were prepared based on incorporation of dUTPs during cDNA synthesis using a previously described protocol [89]. Single-end 50-bp sequencing was performed using the Illumina HiSeq-2000. Reads were mapped to the C. elegans genome version WS220 with Tophat v1.4.1 [90] using default parameters. Gene expression was quantified using Cufflinks version 2.0.2 [91] for strand-specific reads using default parameters and supplying gene annotations. Gene expression values used were the mean of at least three biological replicates, and are provided in S2 File. Differential expression analysis was performed using DESeq2 version 1.0.17 [92] in R version 3.0.2, and the values are given in S3 File. All analyses were performed with genes that had average expression level above 1 FPKM (fragments per kilobase per million, as calculated by Cufflinks). RNA-seq data clustering was performed by K-means clustering using the Hartigan-Wong method in R version 3.02. GO term analysis was performed using GOrilla [93]. The clusters and GO analyses results are provided in S4 File.
ChIP-seq was performed as previously described [26], using H4K20me1 antibody (Abcam ab9051), a custom DPY-27 antibody [80], and AMA-1 antibody against the large subunit of RNA-Pol II (modENCODE SDQ2357). Single end reads were aligned to the C. elegans genome version WS220 using Bowtie 1.0.1 [94], allowing up to two mismatches in the seed, returning only the best alignment, and restricting a read to map to at most four locations in the genome. Peak calling and genome-wide coverage estimation was obtained with MACS version 1.4.2 [95] using mapped reads from ChIP and input. To estimate the final genome coverage, coverage per base was normalized to the genome-wide median coverage, excluding the mitochondrial chromosome, and the input was subtracted from the ChIP. Replicates were merged by averaging coverage at each base position across replicates. For peak calling, ChIP and input reads were merged from all replicates using samtools merge version 0.1.19 [96], and MACS was used for peak calling. The peak calling was also done per replicate, and peaks from the combined data that also occur in the majority of replicates were retained. For wild type and mutant comparisons, data sets were standardized by z-score transformation of the ChIP enrichment values based on the presumed background, excluding regions covering MACS peaks (e-5 cutoff).
The statistical framework for identifying differentially expressed genes takes into account the average and standard deviation between replicates. Genes that are not found as differentially expressed cannot be labeled as “statistically” similarly expressed, because variability between replicates may cause these genes to fail the statistical test. Therefore, to determine genes that are similarly expressed in hermaphrodite and mixed sex embryos, we constructed a 95% confidence interval around the log2 expression ratio for each gene. For a gene to be called similarly expressed, that confidence interval could not exceed a 30% change in expression. Distribution of similarly, equivalently and differentially expressed results are provided in S5 File.
To determine if there is two classes of DCC regulated genes on the X chromosome, we used an expectation-maximization algorithm to determine the distribution of expression changes upon dpy-27 knockdown. A mixture distribution with two or three components was assumed and the better fit was used. Parameters of the mixture distribution were estimated using the mixtools package [97] in R version 3.0.2.
For RNA-seq, hatched L1s were fed on OP50 seeded plates for 6 hours, collected, counted three times and diluted to the desired number of worms based on the average count. 10,000 C. elegans L1s were mixed in with 1,000 C. briggsae L1s. RNA-seq was performed as described above. mRNA-seq reads were mapped to the C. elegans genome version WS220 and C. briggsae genome version WS224 using TopHat (Trapnell et al 2010) version 2.0.11 and reads mapping to both genomes were discarded from further analyses. Raw counts for each gene were estimated with htseq-count version 0.6.1 [98]. All read counts were normalized based on total reads and the median across three replicates was determined. A linear model was fitted to the highly expressed genes from wild type and mutant C. briggsae spike-in data, whose coefficients were used to correct the C. elegans expression data.
For ChIP-seq, 0.1–0.2 mg of ChIP-extract prepared from C. briggsae embryos were added to 1–2 mg C. elegans ChIP extract. ChIP and corresponding input reads were mapped to the C. elegans genome version WS220 and C. briggsae genome version WS224 using Bowtie version 1.0.1 [94] and reads mapping to both genomes were discarded. To determine a spike-in normalization factor, mapped reads were first normalized to total reads and then the ratio between ChIP and input was calculated for uniquely mapped reads to C. elegans or C. briggsae. The ratio of C. elegans and C. briggsae ChIP/Input between the mutant and wild type was used to correct the genome coverage in the mutant.
We used the total RNA that was isolated for the spike in libraries. 20 nanograms of total RNA was used per 20 microliter one-step RT-qPCR reaction (KAPA Biosystems). To eliminate cross amplification between C. elegans and the C. briggsae spike in RNA, primers were designed to amplify C. elegans that do not amplify in the C. briggsae genome. C. briggsae genes CBG27711 and CBG25839 were selected as the spike in normalization gene based on L1 expression, lack of C. elegans homolog and lack of amplification from the C. elegans template cDNA. No-RT and opposite species template RNA reactions were used as controls to confirm specificity of primers. Relative fold change between set-4(n4600) and wild type was calculated using the ΔCt method and spike in normalized changes were calculated using the ΔΔCt method using the averageΔCt of C. briggsae genes CBG27711 and CBG25839 between wild type and set-4 as the second normalization parameter [99]. Primer sequences are as follows: kin-3 forward TTTCAAGGGACCCGAGCTTC, kin-3 reverse GTGTCGCCCGAGAATATCGT, pmp-2 forward CCTGGAGTGGTTGGAATGCT, pmp-2 reverse TCTTGCCAGTTTGCCCAGAA, hlh-11 forward TTCGTTCGGATAGCGCAGAA, hlh-11 reverse GGTGCCATTCGTGCATTTGT, tag-115 forward GTTCAGGACCCCGGTCAAAA, tag-115 reverse TTGGAAGTGATCGCTGACCG, rig-3 forward GCACCAGGAAACAAGCAAGG, rig-3 reverse TTGCATCACGGACGTGGTTA, ifp-1 forward TCCAGAGCGACAGTAGTGGA, ifp-1 reverse TAGTAGTTCGTGCCGGCTTC, rgs-6 forward AATGTGAGCGAACCCAGCAA, rgs-6 reverse CGCCTCGTCGCCTATTTTTG, dhs-25 forward CTGGGAGGACGAATTGCTGT,dhs-25 reverse CCCCTTCACACTGTCTGCAT, CBG27711 forward ATGAGCATAATTCAGAGAGAAAATTTT, CBG27711 reverse CTAGGCATCTATAATTAGGTT, tag-261 forward CGTAGATGCACGTCTTCAAGC, tag-261 reverse TCTCCGCCTCTTTGAGACCTC, C12D5.09 forward TGACTTTTTCTCGACGGATTTG, C12D5.09 reverse CACACCTTCTTTCGTTGGCG, C06C3.12 forward TCGGTGGACCGATTCTAATGAA, C06C3.12 reverse TTAGCCCACTTCGACTGAGG, B0507.10 forward TCTCAGGAATTGCTAATTCGGC, B0507.10 reverse GGCTCTGTGGCGGTTGATAA, tatn-1 forward CGCTTCTTGAGCAAGCCAAA, tatn-1 reverse CTGCGATCTCACTTGGTGGA, nspc-19 forward CATCCTTGCTGCTCTCTGCT, nspc-19 reverse TTCTCCACCGTTGACACCTG, B0294.1 forward ATCCGCAACCAACTCTCCTG, B0294.1 reverse CAAGGTTTTCTCCATTCTTTGGCA, CBG25839 forward CGCGTCTAAACCAGCCAAAC, CBG25839 reverse AGCTGAGCGAGATGAACACC.
|
10.1371/journal.pcbi.1007063 | Modeling visual performance differences ‘around’ the visual field: A computational observer approach | Visual performance depends on polar angle, even when eccentricity is held constant; on many psychophysical tasks observers perform best when stimuli are presented on the horizontal meridian, worst on the upper vertical, and intermediate on the lower vertical meridian. This variation in performance ‘around’ the visual field can be as pronounced as that of doubling the stimulus eccentricity. The causes of these asymmetries in performance are largely unknown. Some factors in the eye, e.g. cone density, are positively correlated with the reported variations in visual performance with polar angle. However, the question remains whether these correlations can quantitatively explain the perceptual differences observed ‘around’ the visual field. To investigate the extent to which the earliest stages of vision–optical quality and cone density–contribute to performance differences with polar angle, we created a computational observer model. The model uses the open-source software package ISETBIO to simulate an orientation discrimination task for which visual performance differs with polar angle. The model starts from the photons emitted by a display, which pass through simulated human optics with fixational eye movements, followed by cone isomerizations in the retina. Finally, we classify stimulus orientation using a support vector machine to learn a linear classifier on the photon absorptions. To account for the 30% increase in contrast thresholds for upper vertical compared to horizontal meridian, as observed psychophysically on the same task, our computational observer model would require either an increase of ~7 diopters of defocus or a reduction of 500% in cone density. These values far exceed the actual variations as a function of polar angle observed in human eyes. Therefore, we conclude that these factors in the eye only account for a small fraction of differences in visual performance with polar angle. Substantial additional asymmetries must arise in later retinal and/or cortical processing.
| A fundamental goal in computational neuroscience is to link known facts from biology with behavior. Here, we considered visual behavior, specifically the fact that people are better at visual tasks performed to the left or right of the center of gaze, compared to above or below at the same distance from gaze. We sought to understand what aspects of biology govern this fundamental pattern in visual behavior. To do so, we implemented a computational observer model that incorporates known facts about the front end of the human visual system, including optics, eye movements, and the photoreceptor array in the retina. We found that even though some of these properties are correlated with performance, they fall far short of quantitatively explaining it. We conclude that later stages of processing in the nervous system greatly amplify small differences in the way the eye samples the visual world, resulting in strikingly different performance around the visual field.
| Psychophysical performance is not uniform across the visual field. The largest source of this non-uniformity is eccentricity: acuity is much higher in the central visual field (fovea), limiting many recognition tasks such as reading and face recognition to only a relatively small portion of the retina. As a result, central visual field loss, such as macular degeneration, can be debilitating. Even a modest difference in eccentricity can have substantial effects on performance. For example, contrast thresholds on an orientation discrimination task approximately triple at 8° compared to 4° eccentricity [1, 2]. Similar effects are found for a wide range of tasks (for a review on peripheral vision, see [3]).
Interestingly, visual performance differs not only as a function of distance from the fovea (eccentricity), but also around the visual field (polar angle). The polar angle effects can be quite systematic. At a fixed eccentricity, contrast sensitivity and spatial resolution are better along the horizontal than the vertical meridian, and better along the lower than the upper vertical meridian. The two effects have been described by Carrasco and colleagues and called the “horizontal-vertical anisotropy” and the “vertical meridian asymmetry” [2, 4–8]. Such effects, often called performance fields, are found in numerous tasks, including contrast sensitivity and spatial resolution [5, 6, 9–23] (Fig 1), visual search [24–31], crowding [32–34], motion perception [35], visual short-term memory [36], contrast appearance [7] and spatial frequency appearance [37]. These effects can be large. For example, contrast thresholds at 8° eccentricity can be 5 times lower on the horizontal meridian compared to the vertical [5], a larger effect than doubling the eccentricity, from 4° to 8° [2].
The causes of performance differences with polar angle are not known. Some eye factors may contribute to them as they do to differences across eccentricity. For instance, the drop-off in density of cones and retinal ganglion cells with eccentricity contributes to decreased acuity [38, 39]. In this paper, we take a modeling approach to quantify the extent to which optics and photoreceptor sampling could plausibly contribute to the reported performance differences with polar angle.
In the human eye, cone density varies with eccentricity and polar angle. Foveal cones have relatively small diameters and are tightly packed, becoming sparser in the periphery due to both increased size and larger gaps between them [40, 41].
Cone density also differs as a function of polar angle at a fixed eccentricity. From ~2° to 7° eccentricity, density is about 30% greater on the horizontal than the vertical meridian [40, 41] (Fig 2). This 30% difference is about the same as the cone density decrease from 3° to 4° eccentricity along a single meridian. As a result, iso-density contours are elongated by about 30% in the vertical axis compared to horizontal. Because cone density is higher on the meridian where performance is better (horizontal compared to vertical), one might be tempted to conclude that cone density explains the performance difference. We return to this question in subsequent sections.
Before light hits the retina, it has already been transformed by refraction from passing through different media (cornea, vitreous and aqueous humors), by diffraction from the pupil, as well as by optical aberrations (chromatic and achromatic) of the lens and intraocular light scattering (for an overview see [42]). These transformations reduce the optical quality of the image projected onto the retina.
Optical quality is not uniform across the retina [43, 44]. A clear, systematic effect is that both defocus and higher-order aberrations become worse with eccentricity. We assume that myopes and hyperopes wear corrective lenses to achieve good focus at the fovea. Defocus is the largest contribution to image quality [42]. The effects of defocus are largest in the far periphery, but still evident when comparing fovea to parafovea (e.g., 0 vs. 5°, Fig 3).
Most measurements of optical quality in human are either at the fovea or along the horizontal meridian. These measurements show that in addition to the decline in optical quality with eccentricity, there are also hemifield effects: For example, in the periphery, the temporal retina tends to have poorer optics than the nasal retina [44]. There are some [43], but many fewer measurements along the vertical meridian compared to the horizontal meridian. To our knowledge, it is not yet firmly established whether there are systematic differences in optical quality between the vertical and horizontal meridians. However, the fact that optical quality varies with eccentricity as well as between nasal and temporal retina suggests that one should at least consider optics as a possible explanatory factor for performance differences around the visual field.
The meridian differences in cone density are correlated with meridian differences in psychophysical performance, with higher cone density and better performance on the horizontal axis (adjusted r2 = 0.88, Fig 4) [40, 41]. However, a correlation does not necessarily imply an explanation. Without an explicit linking hypothesis or model that can predict how a difference in cone density should affect visual performance on a given task, we cannot know whether this correlation is meaningful for explaining this behavior. Should a decrease in cone density increase contrast thresholds? And if so, should a decrease of about 25% in cone density (difference between horizontal and vertical at 4.5° eccentricity) lead to an increase in contrast threshold of about 25%, as observed psychophysically?
Answering these questions requires a computational model. A computational model of the eye can quantify the extent of each component’s contribution on visual performance and potentially reveal which components limit performance on a given task.
In this study, we quantify the contribution of cone density and optical quality on visual performance according to a computational observer model. We then compare the modeled contributions to the observed quantities, and ask whether the observed differences in cone density and optical quality as a function of polar angle can explain the observed differences in performance.
Several studies have proposed ideal observer models of visual tasks limited only by optics and photoreceptor sampling. When such ideal observer models are compared to human performance as a function of eccentricity, results show that these factors alone (optics and photoreceptor properties) cannot completely account for the differences in psychophysical performance (e.g. [45–48]). Generally, these studies find that human performance falls off more rapidly with eccentricity than would be predicted based only on optical and photoreceptor limits, indicating that there are additional downstream processes whose efficiency varies with eccentricity. These studies have not assessed the degree to which differences in performance as a function of polar angle are explained by front-end factors.
To implement the computational observer model of the human eye, we used the Image Systems Engineering Toolbox for Biology (ISETBIO [49–51]), a publicly available toolbox, to simulate encoding stages in the front-end of the human visual system (available at http://isetbio.org/). We used this model to simulate a 2-AFC orientation discrimination task using Gabor stimuli matched in parameters as reported by Cameron, Tai, and Carrasco [5]. Our computational observer model consists of multiple stages representing the front-end of the visual system: the spectral radiance of the experimental visual stimuli, optical quality of the lens and cornea, fixational eye movements, the cone mosaic with different types of photoreceptors and their isomerization rate for a given stimulus presentation. Finally, the observer model has an inference engine to classify the stimulus as belonging to one of two possible classes (clockwise or counter-clockwise). Our inference engine employs a linear support vector machine (SVM) classifier to learn the two stimulus classes from the cone outputs on training data, and then applies the SVM to left out test data. Hence the classifier, unlike an ideal observer model, does not have explicit information about the stimulus, noise distributions, or the system (optics, cone types).
With a computational observer model, one can show where along the visual pathway information loss happens and how this loss of information is inherited or potentially compensated for in later encoding stages of the visual pathway. Here, we investigate to what extent visual performance (contrast threshold) depends on variations in cone density and optical quality. By systematically varying cone density and optical quality (defocus) independently, we can compare the computational observer model performance to reported differences in the literature on the human eye to quantify the individual contribution of each of these two factors to differences in visual performance across the visual field.
By studying polar angle effects, this study complements prior observer model studies of eccentricity effects; thus, this study provides novel information to our understanding of visual perception across the visual field. Moreover, our simulations include stimulus uncertainty (phase randomization) and fixational eye movements. These factors generate trial-to-trial noise that is far more complex than simple independent Poisson noise at each receptor, making the closed-form solution of an ideal observer model more difficult to implement. Hence, we use a computational (but not ideal) observer model. This brings model performance closer to human performance than typical ideal observer models, and more generally, is an important tool for situations in which ideal observer models cannot be implemented.
To investigate to what extent performance differences in the visual field depend on variations in cone density and optical quality, we developed a computational observer model of the first stages of the human visual pathway. The computational observer was presented with oriented Gabor stimuli, tilted either clockwise or counter-clockwise from vertical, with a phase of 90° or 270° (randomized across trials) to simulate a 2-AFC orientation discrimination task. (The phase was not relevant to the judgment). To compare the performance of the computational observer to human observers, we matched the stimulus parameters to a psychophysics study [5].
Two aspects of our simulated experiments give rise to stimulus uncertainty, small fixational eye movements and phase randomization of the stimulus (90° or 270°). The effects of these are similar. We consider eye movements first.
We model eye movements using an algorithm developed by Cottaris et al. [54], which is included in ISETBIO. This algorithm combines a statistical model of fixational drift by Mergenthaler and Engbert [55] with a model of microsaccades based on statistics reported by Martinez-Conde et al. [56, 57]. The displacement of the stimulus due to eye movements in our simulations is relatively small: within one trial (a single colored line), the retinal displacement tends to be about 2–4 cones or less (Fig 6A). This is small compared to the spatial scale of our stimulus, for which a full cycle corresponds to ~6 cones at 4.5° eccentricity. Given that the trials last only 54 ms, the probability of a microsaccade is low. Hence when both microsaccades and drift are present, eye movements are dominated by drift.
The fixational eye movements have a large effect on the computational observer model. Compared to a model with no phase uncertainty and no eye movements, adding fixational eye movements causes the contrast threshold to about double (Fig 6B, black solid line versus black dashed line). It might be surprising that eye movements have any effect on the model performance: An image translation is equivalent to a phase shift in the Fourier domain and the model discards phase information. However, because the retinal mosaic contains multiple cone types with different sensitivities, a shift in the stimulus causes a change in both the amplitude and phase spectra of the absorption images, affecting the information available to the classifier. Put simply, the possible set of cone responses is more variable in the presence of eye movements (whether or not the cone outputs are transformed into the Fourier domain), making it more difficult for the classifier to cleanly separate the two stimulus classes. Note that in addition to increasing the threshold, the presence of eye movements also flattens the slope of the psychometric function. This is because adding eye movements is not equivalent to lowering the contrast: even at the highest contrasts, the classifier may have some uncertainty because the eye movements can result in a set of cone responses that is very different from the average set of training responses.
The second source of stimulus uncertainty is phase randomization. On each trial, one of two phases 180° apart was randomly selected. Again, for a retina with only one cone type, the change in stimulus phase would translate to a change in the phases of the cone outputs, and would not impair a classifier operating on the Fourier amplitudes. But with a mixed cone array, the two phases result in a pattern of cone responses that differ in ways beyond a simple phases change. Like eye movements, the phase uncertainty makes performance worse (a 5x increase in threshold compared to the condition with no eye movements and no phase uncertainty, solid black vs. solid red line, Fig 6B). Combining the two sources of uncertainty makes performance a little worse than either one alone (red dashed line).
Human observers drastically differ in their ratio of L-, M-cones, even at a fixed retinal location [58, 59]. Our computational observer model assumed a ratio of 0.6:0.3:0.1 L:M:S cones. In separate simulations we found that varying the ratio between cone types within a mosaic affects model performance.
To quantify the effect of cone type on our computational model performance, we varied the ratio of cone types in two ways. First, we tested the effect of cone types using uniform cone mosaics: L-cone only, M-cone only, or S-cone only, compared to a mixed cone array with a typical L:M:S ratio of 0.6:0.3:0.1 (Fig 7A).
There is a very big effect of cone type. A uniform mosaic of only S-cones has a threshold about 4 times higher than one of only L- or M- cones. The M-cone only and L-cone only mosaics result in similar performance (a 10% higher threshold for the M-cones). The large decrease in threshold for S-cone only mosaics is partly explained by chromatic aberration: our model assumes that the eye is in focus at 550 nm (in between L- and M-cone peak spectral sensitivity), causing large amounts of blur for shorter wavelengths that the S-cones are sensitive to. Blurring the image is similar to lowering stimulus contrast and therefore increases the contrast threshold. Additionally, the S-cones have lower efficiency because of the yellowing of the lens and the filtering of the macular pigment. The slopes of the psychometric functions for all three uniform mosaics are similar. This is because the changes in efficiency and in focus are similar to a change in contrast, meaning a remapping along the x-axis, or a horizontal shift in the curve.
Interestingly, the trichromatic mosaic results in a shallower psychometric function than any of the uniform cone arrays, and a threshold comparable to that found for the S-cone only mosaic. That may be surprising given that our trichromatic retina contained only 10% S-cones. Why is its performance so poor? This is because of stimulus uncertainty and eye movements: when the array has only one type of receptor, a phase difference in the stimulus or a position change caused by an eye movement translates to a simple phase difference in the absorption images, which does not impair the classifier. On the other hand, for a trichromatic retina, a phase difference or position difference in the stimulus has a more complex effect on the absorptions, resulting in performance that is worse than expected from simply computing a weighted average of the performance by the three separate cone types.
Second, to better understand the effect of arrays with cone mixtures, we simulated experiments with retinas with only L- and M-cones in various ratios. Best performance was found for uniform retinas (100% L-cones, and then almost as good, 100% M-cones) (Fig 7B). When introducing a mixture of cones in the mosaic, even a small fraction, thresholds increase. For example, a mosaic with 10% M-cones and 90% L-cones increases the threshold from 0.71% (L-cone only) to 1.0% stimulus contrast. The worst performance is found for an L:M ratio of 50:50. The computational observer model shows an approximately quadratic relation between contrast threshold and probability of L-cones in the LM mixture cone mosaic (r2 = 0.83, Fig 7C). These results indicate that our model is sensitive to the variability caused by differences in mean absorption rates across cone types, even with small differences in peak spectral sensitivity and efficiency between L- and M-cones. This pattern, whereby the best performance occurs for uniform retinas, depends on there being stimulus uncertainty (phase differences in the stimulus and/or small fixational eye movements). With no uncertainty, there would be little effect of mixing L- and M-cones.
For individual differences in L:M:S cone ratios to explain the meridional effect in human performance between the horizontal and upper visual field, one would need to have a typical ratio of L:M:S on the inferior retina (upper visual field), but a retina with exclusively L-cones or M-cones along the horizontal meridian. Given that the arrangement of L-, M- and S-cones in human retina is approximately random [60], such a scenario is biologically implausible.
Large levels of defocus worsen visual acuity [61], where defocus levels larger than 0.75 diopters (corresponding to 20/40 vision on the Snellen acuity chart for near sightedness) are usually compensated for with visual aids. Here, we tested the effect of defocus on the 2-AFC orientation discrimination task reported by Cameron, Tai, and Carrasco [5] to investigate whether variations in defocus could explain the decrease in performance with polar angle. If the task is very sensitive to the level of defocus, then small differences in optical quality as a function of polar angle might explain the observed differences in performance.
Defocus affects the modulation transfer function of a typical human wavefront by attenuating the high frequencies (Fig 8A). The Gabor patches in our experiment had a peak spatial frequency of 4 cycles/° (dashed line). For this spatial frequency, the simulated levels of defocus in the observer model cause a modest reduction in contrast.
As expected, large increases in defocus cause the computational observer model to perform worse, evidenced by a rightward shift of the psychometric curve (Fig 8B). When comparing contrast thresholds as a function of defocus level, the computational observer model shows a monotonic relation with defocus, which, for simplicity, we approximate with a linear fit (r2 = 0.86, Fig 8C).
The effect of defocus on model performance is small. To explain an increase of 1.5% in contrast threshold, similar to what is observed psychophysically as a function of polar angle, the computational observer model would require an additional 7 diopters of defocus. This is far higher than any plausible difference in defocus as a function of polar angle at 4.5°. Typically, defocus at 4.5° along the horizontal or vertical meridian is within ~0.2 diopters of defocus at the fovea [43, 44]. The difference between the vertical and horizontal locations at 4.5° would be even less. Assuming a difference in defocus of 0.2 diopters, the optical quality would explain only about 3% of the effect of visual performance as a function of polar angle for this task.
The cone mosaic varies substantially with retinal location. As eccentricity increases, cone diameter increases, as does spacing between the cones, resulting in lower density. We used our computational observer model to quantify the extent to which variations in the cone mosaic could explain the changes in performance with polar angle. We simulated a large range of cone densities, from about 3 times lower to 15 times greater than the typical density at 4.5° eccentricity (i.e., the eccentricity of the psychophysical experiment in [5]). As we varied the cone density, we also varied the cone size and spacing between cones according to the reported relation between density and coverage [40]. The denser mosaics sample the stimulus more finely, with fewer absorptions per cone, because as the cone area decreases it captures fewer photons. (Fig 9A).
Our computational observer model shows a decrease in contrast threshold as a function of cone density (Fig 9B and 9C). However, the effect is relatively small. For every 6-fold increase in cone density, the computational model contrast threshold reduces by 1 percentage point (e.g., from 4% to 3%). The meridional effect on human performance is ~4.4% (upper vertical meridian) vs. 3.4% (horizontal). For cone density to account for this observed difference in human contrast thresholds, there would need to be more than a 500% meridional difference in cone density. This is far greater than the 20–30% reported difference in cone density at 4.5° eccentricity [40, 41, 64]. This indicates that, according to our computational observer model, cone density accounts for less than 10% of the differences in visual performance with polar angle on the orientation discrimination task reported by Cameron et al. [5].
Our goal was to assess the degree to which front-end properties of the visual system explain well-established psychophysical performance differences around the visual field. In particular, we quantified the contribution of two factors in the eye–cone density and optical quality (defocus)–to contrast thresholds measured at different polar angles in an orientation discrimination task as reported by [5]. These front-end factors have been reported to vary with polar angle, and in principle, the observed performance differences could be a consequence of the way the first stages of vision process images. For instance, cone density is higher on the horizontal meridian compared to the vertical meridian (up to 20° eccentricity [40, 41, 64]). Nonetheless, without a model to link these factors to performance, how much explanatory power they have cannot be assessed. We therefore developed a computational observer model to test these potential links. The underling software we used, ISETBIO, has recently been used to model a number of basic psychophysical tasks, including contrast sensitivity [50], Vernier acuity [65], illumination discrimination [66], color perception [67], chromatic aberration [68], visual perception with retinal prosthesis [69], and spatial summation in Ricco’s area [70].
Although cone density along the cardinal meridians correlates with behavior, our model showed that this correlation has little explanatory power: Differences in cone density can only account for a small fraction of the variation in visual performance as a function of meridian. Similarly, variation in optical quality within a plausible biological range has only a very small effect on contrast thresholds in the model of our task. Our observer model puts a ceiling on these two factors at less than 10% of the observed psychophysical effects. To fully explain these visual performance differences with polar angle, our computational model would require a difference of more than 7 diopters in defocus and a difference of more than 500% in cone density for the horizontal compared to the upper vertical meridian. Such large differences are far outside the range of plausible biological variation; defocus at 4.5° eccentricity is typically within 0.1–0.2 diopters of the fovea [44] and cone density at the horizontal meridian is ~20–30% more than the vertical at this eccentricity [40].
The fact that neither optics nor the cone sampling array can explain more than a small fraction of the effect of polar angle on contrast thresholds indicates that downstream mechanisms must explain the majority of this effect. This conclusion is in line with similar work using an ideal observer model to study eccentricity-dependent effects (along the horizontal meridian) in which optics and photoreceptor properties cannot completely account for the eccentricity-dependent performance seen in human contrast sensitivity or spatial resolution [46, 48]. In particular, human performance falls off with eccentricity more sharply than would be predicted by the optics or photoreceptor properties alone. One reason for this is that downstream processing further accentuates the loss of neural resources with eccentricity. For example, the ratio of retinal ganglion cells per cone declines with eccentricity [71]. Accounting for this effect brings predicted performance closer to human in terms of the rate of decline with eccentricity, although ideal observer models that account for RGC density still perform much better than human [45, 46]. More accurate models of human detection performance are achieved by incorporating cortical computations in addition to optical and retinal factors (e.g. [47]).
Could performance differences with polar angle be explained by variation in retinal ganglion cell density? Midget retinal ganglion cells are the most prevalent class across the retina (~80% of the retinal ganglion cell population), have small cell bodies and small dendritic trees, and are hypothesized to set a limit to achromatic spatial acuity [72]. Like photoreceptors, midget retinal ganglion cells sample the visual field asymmetrically. For example, at 4.5° eccentricity, the density of midget retinal ganglion cells on the horizontal meridian is reported to be 1.4 times greater than on the vertical meridian (~1,330 vs. ~950 cells/deg2 on the horizontal vs. inferior retina) [71, 73]. This 40% meridional effect is larger than the 20–30% effect at the level of the cones, indicating that polar angle asymmetries in cone density are accentuated in further retinal processing. We have not included retinal ganglion cells in our model, but given our observer model with the cone array, we speculate that this further meridional difference in ganglion cell density alone will not be sufficient to explain the reported meridional psychophysical effects.
Properties of retinal ganglion cells other than density might also vary with polar angle, such as receptive field size. If so, this too, could contribute to performance differences. For example, retinal ganglion cell receptive field size increases with eccentricity [38]. This increase, combined with random inputs of cone type to ganglion cell, has been proposed to explain the precipitous fall-off in chromatic acuity with eccentricity [74]. Little is known about retinal ganglion cell receptive field size as a function of polar angle in human. As estimated from post-mortem dendritic size, midget retinal ganglion cell receptive field sizes are smaller in the nasal quadrant than other quadrants [75, 76]. In macaque, Croner and Kaplan [77] report differences in midget cell density (e.g., nasal vs. temporal) but do not find receptive field size differences with polar angle. Without clear reports of meridional differences in midget ganglion cell receptive field and dendritic field sizes, it is unlikely that properties of retinal ganglion cells would fully account for the reported meridional effects in visual performance.
A second potential factor is visual cortex. Some aspects of performance fields manifest as amplitude differences in the BOLD fMRI signal in V1. Liu, Heeger and Carrasco [78] reported a 40% larger BOLD amplitude in V1 for stimuli on the lower than the upper vertical meridian. This asymmetry was found for high but not low spatial frequency stimuli, matching psychophysical results. They did not report differences between stimuli on the vertical versus horizontal meridians. Performance fields may also be reflected in the geometry of visual cortex. For example, a template of the V1 map fit to a population of 25 observers showed more cortical area devoted to the horizontal than the vertical meridian, although the authors acknowledged that this could be a fundamental fact about V1 or an artifact of the flattening process used in their analyses [79]. This areal difference has been confirmed in an independent data set [80]. These data also showed that population receptive fields (pRFs) in V1 and V2 are ~10% smaller when comparing horizontal to vertical quadrants. The geometry and the pRF size effects are complementary: greater area and smaller pRFs along the horizontal meridian are both consistent with this part of visual cortex analyzing the visual field in greater detail. However, there are also psychophysical differences between the upper and the lower vertical meridian [2, 6, 7], for which no pRF differences have been reported [80]. Our ongoing work suggests that there are in fact differences in cortical magnification between the upper and lower meridian. It is not yet known whether any of these meridional effects in V1 (such as greater area for the horizontal than vertical meridian) are inherited properties from the retina, or whether V1 further amplifies the polar angle differences.
Just as visual cortex does not sample all locations in the visual field in the same manner, it also does not sample spatial frequencies and orientations perfectly uniformly (e.g. [81]). Interestingly, psychophysical asymmetries with polar angle are reported to differ as a function of spatial frequency (larger asymmetries at higher spatial frequencies), but not as a function of stimulus orientation [4–8]. However, it is unknown whether and how cortical sampling differences would account for the observed behavioral patterns. A model is needed to explicitly link non-uniform coverage across V1 to behavioral performance with regard to stimulus spatial frequency.
In addition to factors in early visual cortex, cognitive factors will also be important to consider in developing a full understanding of visual performance across polar angles. Exogenous covert visual attention does not compensate for discriminability differences across polar angles [4–6, 19], but endogenous covert attention may do so. We are currently investigating this possibility.
Our goal in building a computational observer model was to explicitly link known facts about the biology of the visual system with psychophysical performance. The value of the model is evidenced by the difference in the inference one might have drawn from a purely correlational approach (performance is best where cone density is highest) and the inference drawn from the model (little relation between cone density and performance). Nonetheless, all models are simplifications, and ours is no exception.
First, our model contained only one eye, whereas most of the psychophysical evidence in support of performance fields comes from binocular experiments. But the few studies with monocular stimulus presentation confirm differences in performance across polar angle and show a similar magnitude of the effect as for binocular stimulus presentation [4, 8]. Hence this limitation is unlikely to affect our conclusions. The effects of binocular viewing on our model performance are likely to be complicated. On the one hand, by doubling the number of photoreceptors, the signal to noise ratio would increase, consistent with the fact that contrast thresholds are lower with binocular viewing [82]. On the other hand, differences between the eyes, such as in the optics, the cone mosaic, or eye movements, could result in impaired performance. To quantify performance differences between binocular and monocular experiments, one would need to combine the information from the two eyes at some stage. In the human visual system, signals from the two eyes converge in V1 [83–85], although where exactly in V1 is still debated (e.g. [86, 87]). Because our model does not explicitly model visual processing beyond the photoreceptors, we leave the implementation of a biologically accurate binocular viewing condition for future studies.
Second, we modeled the cone mosaic as a rectangular patch with uniform density for each simulation, whereas the photoreceptors in human retina are organized in a hexagonal grid with a gradual change in density as a function of eccentricity. The uniformly spaced rectangular grid was implemented to save computational resources. The difference between an eccentricity-dependent mosaic and a uniform mosaic can be important for modeling performance near the fovea [67], as density declines rapidly over a short distance [40]. However, further in the periphery, the density changes are modest across a small patch. And given that our model showed that very large differences in the cone array were needed to explain variation in psychophysical performance, it is unlikely that using a hexagonal, eccentricity-dependent array would have altered our conclusions.
Third, we did not model differences in photopigment density or macular pigment density as a function of retinal position. Pigment density has an effect on wavelength sensitivity and overall efficiency [52]. Although our model did not vary pigment density, it did include position-dependent efficiency, implemented by varying the cone coverage, which ranged from close to 1 (no gaps between cones) near the fovea to ~0.25 in the far periphery. Hence, additional variation in efficiency arising from pigment density would be unlikely to have a substantial impact on model performance. Moreover, macular pigment density does not vary systematically with polar angle at iso-eccentric locations [88].
Finally, our computational model only deals with visual processes up to photon absorptions by the cones. Processes up to this point, optics, photon noise, and cone sampling, are well characterized and can be accurately modeled. In future work, we will build on our computational observer model to investigate the contribution of downstream factors, such as post-receptor retinal circuitry and pooling of signals by retinal ganglion cells and visual cortex.
The performance of a classifier depends, in part, on how much knowledge of the task the classifier has access to. Our observer model had far less information than an ideal observer model. By definition, ideal observer models have complete knowledge about the relationship between inputs and outputs (except for the trial-to-trial stochastic noise) and use this knowledge to make optimal decisions, thereby setting an upper limit on performance [28, 29, 46, 53, 89–91]. When ideal observer models are applied to very early signals in the visual system such as cone responses, they typically outperform human observers by a large margin, e.g., by a factor of 10 to 100 [46, 92]. In contrast, our computational observer model performs similar to human observers (~2–4% contrast thresholds in our task). The similarity to human performance should be interpreted with caution, however, since small variations in our model, such as the number of training trials, affect the performance of the model.
Our inference engine has two types of knowledge about the task, one more general about visual processing and one more specific to the particular experiment we simulated. The general (and implicit) knowledge arises from transforming the 2D time-varying cone absorption images to amplitude spectra. Transforming the data in this way preserves most of its representation and it effectively gives the observer model knowledge that spatial frequency and orientation (but not phase) might be relevant for the task. This transform does not indicate which spatial frequencies or orientations are relevant. Although the visual system does not literally compute a Fourier transform of the cone responses, cells in visual cortex are tuned to orientation and spatial frequency in local patches of the image [93, 94], and complex cells in V1 are relatively insensitive to phase [84, 95]. Hence the implicit knowledge we provide to the classifier via transform to the amplitude spectra is an approximation that is conceptually inspired by general processing strategies in the visual system, but is not a specific implementation of a V1 stage with complex cells or knowledge about our particular task. Pilot simulations in which the classifier operated directly on the absorption images resulted in near-chance performance. This is expected, because the phase randomization of the stimuli causes the number of absorptions for any particular cone to be uninformative as to the stimulus orientation.
More specific knowledge in the computational observer model comes from the training trials, which are used to learn the best linear separation (hyperplane) between the two stimulus classes. The plane is defined by a weighted sum of the classifier inputs (amplitude spectra in our case), which can be thought of as an approximation to receptive field analysis by downstream neurons. The high weights learned by the classifier for this task correspond to oriented, band-pass filters, which match properties of the stimuli (Fig 5, panel D). Because the model has incomplete knowledge, values far from the stimulus (very high or very low spatial frequency, and orientations far from the stimulus orientations) have non-zero weights, which are learned during training on a finite number of noisy trials.
Our model differs from ideal observer models. To understand how the differences affect performance, we implemented an ideal observer model as described by Geisler [53]. An ideal observer model for simulations with multiple cone types and fixational eye movements is extremely complex, and hence for this comparison, we examined the case of no eye movements and only a single stimulus phase for each of the two stimulus classes. The ideal observer performs far better than our computational observer, with a threshold about 10x lower (Fig 10, black vs. green line). This difference is caused by the fact that the linear SVM classifier needs to learn the stimulus classes and the noise distributions, rather than starting with such knowledge. If we increase the number of trials fourfold, the SVM computational observer performance improves a little (red vs. green line in Fig 10), because it has more samples to estimate learn the classes. As shown by Cottaris et al. [50] for a related task, the SVM may require thousands to millions of trials to reach ideal performance levels. With the additional uncertainty from multiple stimulus phases and fixational eye movements, the computational observer model performs even worse, as shown in Fig 6. It is this performance—the SVM classifier trained with 200 trials operating on simulations that include fixational eye movements and two stimulus phases—that is similar to human performance. The type of computational observer model implemented here is useful when the ideal observer model is unwieldy or intractable and has the benefit of being potentially more similar to how the human learns the task.
Overall, our model includes a relatively detailed, biologically plausible front-end, which incorporates realistic details about the optics, photon noise, small fixational eye movements, and wavelength- and position-sampling by photoreceptors. This front-end processing was combined with a linear classifier that performs at levels comparable to the human without providing explicit knowledge about the tasks. Future work will incorporate more biologically explicit models of downstream processing, including retinal and cortical circuitry. Such models are likely to reveal that later processing in the nervous system inherits, and possibly amplifies, asymmetries in processing around the visual field that begin in the earliest stages of vision, and thus, to explain a larger portion of the psychophysical asymmetries found in many visual tasks.
The computational observer model relies on the publicly available, MATLAB-based Image Systems Engineering Toolbox for Biology (ISETBIO [49–51]), available at http://isetbio.org/. The ISETBIO toolbox incorporates the image formation process, wavelength-dependent filtering, optical quality, and the spatial arrangement and biophysical properties of cones. We used the ISETBIO toolbox for the core model architecture and supplemented it with experiment-specific custom MATLAB code. The experiment-specific code implements stimulus parameters matched to a prior psychophysical study [5], manipulation of biological parameters to assess their impact on performance, and a 2-AFC linear support vector machine classifier. In the interest of reproducible computational methods, the experiment-specific code, for both simulation and analysis, is publicly available via GitHub (http://github.com/isetbio/JWLOrientedGabor). In addition, the data structures created by the simulation and analyses are permanently archived on the Open Science Framework URL: https://osf.io/mygvu/.
Our simulations were created to match a previous psychophysical study [5]. In that study, stimuli were achromatic oriented Gabor patches. The Gabors were comprised of harmonics of 4 cycles/°, windowed by a Gaussian with a standard deviation of 0.5°, presented at 4.5° eccentricity, at one of 8 locations equally spaced around the visual field (see also Fig 1A). Gabor patches were tilted either 15° clockwise or counter-clockwise from vertical, and presented for 54 ms on each trial. The contrast of the Gabor patches varied from trial to trial. The contrast levels were selected for each observer based on pre-experiment testing, and usually ranged from about 1% to 10% Michelson contrast using a method of constant stimuli. The observer’s task was to indicate the orientation of the Gabor stimulus relative to vertical (clockwise or counter-clockwise) with a button press. Data were analyzed by fitting a Weibull function to the mean performance (% correct) at each contrast level, independently at different locations around the visual field.
The observer model starts with a description of the stimulus, called a ‘scene’ in ISETBIO. The scene is defined by the spectral radiance at each location in space and time (the ‘light field’). The spectral radiance contained wavelengths ranging from 400–700 nm, discretized to 10 nm steps, with equal photons at each wavelength (3.8x1015 quanta/s/sr/nm/m2). The stimulus was discretized into 2-ms time steps and 1.8-arcminute spatial steps (32 samples per degree). The scene comprised Gabor stimuli with parameters described above (Methods section ‘Psychophysical experiment’), oriented either clockwise or counter-clockwise, represented within a field of view of 2° diameter, and presented for 54 ms per trial. The dimensions of the scene were therefore 64 x 64 x 31 x 28 (height x width x wavelength x time). Gabor patches varied in Michelson contrast between 0.05% and 10% (or 0.01% and 10% for simulations with one stimulus phase and no fixational eye movements). We also incorporated a stimulus with 0% contrast stimulus as a sanity check whether our model would perform at chance level. For all stimuli, the mean luminance was 100 cd/m2. Because photon noise and eye movement noise are added later (see Methods sections ‘Optics’ and ‘Cone mosaic’), and because we do not model the scene before or after the stimulus onset/offset, the scene is in fact identical at all 28 time points.
Machine learning algorithms can exploit sources of information that a human observer would be unlikely to use. For example, if the value of a single image pixel happened to correlate with the stimulus class, a classifier could succeed based on only the value of this pixel. We wanted to prevent our classifier from succeeding in this way. In our simulations (unlike the Cameron et al. paper [5]), the phase of the Gabor patches was selected from two values 180° apart (φ = 90° and φ = 270°), randomized across trials. A 180° phase difference means that the two possible stimuli within a class were identical except for a sign reversal. As a result, the expected value of each pixel in each stimulus class was 0 (relative to the background). Similarly, the expected value of the cone absorption rates at each location on the retina within a stimulus class was 0 (relative to the background). Therefore, the linear classifier could not succeed using the absorption level from any single cone. We believe human observers do not perform the task this way either, hence randomizing the phase is likely to make the observer performance more similar to the human performance.
While most of the simulations contained two possible stimulus phases and fixational eye movements, we also explicitly manipulated these factors in two simulations. In the Results section describing the effect of stimulus uncertainty and small fixational eye movements (Fig 6B), we compare the effect of one vs. two stimulus phases and the presence vs. absence of small fixational eye movements on model performance. In the Discussion section ‘The inference engine’ (Fig 10), we compare ideal observer model performance to our computational observer model performance for a simulated experiment with only one stimulus phase and no eye movements.
The optics transform the scene into a retinal image. We first describe the optics used for the simulations in Results sections on the effect of stimulus uncertainty and eye movements, cone type and cone density on model performance (Figs 6, 7 and 9). For these simulations, the optics were matched to a typical human eye with a 3-mm pupil (diameter) in focus at 550 nm using a statistical model of wavefront aberrations [62]. This statistical model is based on measurements from healthy eyes of 100 observers [63], and described by a basis set of Zernike polynomials [96]. The statistical model by Thibos contained the first 15 Zernike coefficients (Z0-Z14, using OSA standard indexing). The simulated human wavefront was used to construct a point spread function (PSF). This PSF was convolved with the scene at every time point to generate the retinal image. After this spatial blurring, the optical image was further transformed by spectral filtering (light absorption by inert pigments in the lens and macula), which primarily reduce the intensity of short-wavelength light. Finally, the optical images were padded by 0.25° on each side with the mean intensity at each wavelength. The padding is needed to handle eye movements, so that cones near the edge of the simulated retinal patch have a defined input even when these cones are moved outside the scene boundaries. The dimensions of the optical image are the same as the dimensions of the scene, except for the spatial padding: 80 x 80 x 31 x 28 (height x width x wavelength x time), which was discretized the same way as the scene.
To investigate the effect of optical quality on visual performance of our task, we systematically added further defocus to the model of human optics (Fig 8). We did this by increasing the Z4 Zernike coefficient (defocus) from 0–2 μm in steps of 0.25 μm (corresponding to 0–6.16 diopters for a 3-mm pupil), while keeping all other Zernike coefficients from Thibos’ statistical model unchanged. Note that using a defocus coefficient of 0 does not result in perfect diffraction limited optics, given that the other aberrations are still non-zero. We manipulated defocus rather than all the higher-order aberrations because at the stimulus eccentricity we simulated (4.5°), defocus is the largest contributor to optical quality [44].
We constructed the cone mosaic as a uniformly spaced rectangular patch with a field of view matched to the stimulus (2x2°). Each cone mosaic contained a random distribution of L-, M- and S- cones with a ratio of 0.6:0.3:0.1. We used the Stockman-Sharpe [97] cone fundamentals to estimate cone photopigment spectral sensitivity, assuming 50% optical density for L- and M-cones, and 40% for S-cones. Peak efficiency was assumed to be equal for each cone class, 66.67% multiplied by the retinal coverage (the fraction of local retina occupied by cones).
For the simulations on the effect of stimulus uncertainty and eye movements, cone type and optical quality on model performance (Figs 6–8), the cone density was 1,560 cells/deg2, approximately matched to the density at 4.5° on the horizontal retina as reported by Curcio et al. [40]. This results in an array of 79 x 79 cones for our 2° patch. The positions of the L-, M-, and S-cones were randomized within the array (but held to fixed ratio). For these simulations, we assumed a coverage proportion of 0.49, meaning that the cone inner segments sampled from about half of the optical image, and missed about half due to the spaces between cones. A coverage of less than 1 acts like a reduction in efficiency, since photons are lost to the gaps between cones. In general, cone coverage decreases with eccentricity as the density of rods increases, filling the spaces between cones.
For the Results section on the effect of cone types (Fig 7), we quantified model performance when the L:M:S cone ratios varied. First, we simulated experiments with mosaics containing a single cone type, i.e. only L-cones, only M-cones, or only S-cones. Second, we systematically varied the L:M ratio in a cone mosaic without S-cones. We quantified model performance for 11 different mosaics, varying from 100% L-cones to 100% M-cones discretized in steps of 10%.
For the set of experiments investigating cone density (Fig 9), we systematically varied cone density spanning 22,500 to 466 cones/deg2 (corresponding to cone arrays ranging from 297 x 297 to 43 x 43). For each cone density, we determined an equivalent eccentricity based on the relation between eccentricity and density on the nasal meridian from Curcio et al. [40]. We then adjusted the cone coverage according to this eccentricity, assuming that coverage declines exponentially as a function of eccentricity, from 1 (fovea, no gaps between cones) to 0.25 at 40°. This approximation is similar to that used by Banks et al. [46], which was based on data from Curcio et al. [40].
The number of absorptions was computed for each cone in two steps. First, the noiseless number of absorptions was computed by multiplying the appropriate cone sensitivity function (L-, M-, or S-cone) by the corresponding location in the optical image (hyperspectral), and scaling this value by the peak efficiency (66.67%). The cone coverage was accounted for by only sampling the optical image at the locations within the cone inner segments. Second, the noiseless values were converted to noisy samples by assuming a Poisson distribution.
The dimensions of the cone array absorptions were 79 x 79 x 28 (rows x columns x time) for the simulations in the Results section on the effect of stimulus uncertainty and eye movements, cone type and optical quality (Figs 6–8). When the cone density varied (Fig 9), the first two dimensions of the cone array size also changed.
We added small fixational eye movements (drift and microsaccades), before computing the isomerization rate for each cone at each time sample. The ISETBIO toolbox provides an algorithm developed by Cottaris et al. [54] that generates eye movement samples based on Mergenthaler and Engbert’s drift model [55] and microsaccade statistics reported by Martinez-Conde et al. [56, 57].
The drift model computes eye movement paths for a single trial with modified Brownian motion process. The eye movement paths were generated in units of arc minutes and then converted to discrete cone shifts in the horizontal and vertical direction. If the amplitude of an eye movement was smaller than the distance between two cones, the displacement was accumulated over multiple time samples, until the threshold was reached, before a new shift was added to the eye movement path.
The drift model was implemented by adding a displacement vector to the current position at each time point. The displacement vector was determined by combining 3 inputs: 2D Gaussian noise, an autoregressive term for persistent dynamics at short time scales, and a delayed negative feedback for antipersistent dynamics at longer time scales. The parameters we used for this model were the ISETBIO defaults, which contained a horizontal and vertical delay defined as X = 0.07 s and Y = 0.04 s, feedback steepness of 1.1, and feedback gain of 0.15. The control function had a mean of 0 and standard deviation of 0.075 and the gamma parameter was set to 0.25. The mean noise position and standard deviation were set to 0 and 0.35, respectively. Before computing the velocity of a drift period, the drift model applied a temporal smoothing filter to the eye movement paths using a 3rd order Savitzky-Golay filter over a velocity interval of 41 ms.
For periods where the drift was stabilized, the eye movement code checked for microsaccade jumps to add to the eye movement path. Whether or not a microsaccade was added depended on when the last microsaccade was. In our experiment, we used the ISETBIO default where the interval between microsaccades followed a gamma function with a mean of 450 ms, with a minimum duration of 2 ms. A microsaccade was defined as a vector where the mean amplitude of a microsaccade was 8 arc minutes. Each vector contained an additional endpoint jitter of 0.3 arc min in length and 15° in direction. The microsaccade jumps were either ‘corrective’ (towards the center of the mosaic) or ‘random’ (any direction). The microsaccade mean speed was defined as 39°/s, with a standard deviation of 2°/s. Given that the defined interval between microsaccades was long compared to the stimulus duration (54 ms), most trials did not contain microsaccades.
A 216 ms warmup period was implemented before the trials began. Eye movements during this period affected the eye position at the start of the trial but were not otherwise included in the analysis.
A simulated experiment comprised 6,000 trials, with 400 trials at each contrast. Most simulated experiments contained 15 contrast levels within the range of 0–10%. More contrast levels were used in a few simulations in which the thresholds were very different between conditions (29 contrast levels). In all experiments there were 400 trials per contrast level. These 400 trials included 200 clockwise and 200 counter-clockwise stimuli, each of which was further subdivided into 100 trials at each of 2 phases. The data from a single contrast level within a single experiment were represented as a 4D array (m rows x m columns x 28 time-points x 400 trials), in which m is the number of cones along one side of the retinal patch (79 in the experiments for Figs 6–8, variable for the experiments in Fig 9).
Within the 6,000 trials of an experiment, all parameters other than the stimulus orientation (clockwise or counter-clockwise) and phase (90° or 270°) were held constant, including the spatial distribution of L-, M-, and S-cones, the optics, the cone density, the cone coverage, and the presence or absence of fixational eye movements. Each simulated experiment was repeated 5 times, so that a single psychometric function summarized 2,000 trials per contrast level. The arrangement of L-, M-, and S-cones was regenerated randomly for each of the 5 repeated experiments. Error bars in Figs 6–9 indicate standard errors of the mean across the 5 experiments. Some stimulus contrasts’ error bars were very small (usually for high stimulus contrasts), which resulted in error bars being masked by the data point.
Classification (clockwise vs. counter-clockwise) via cross-validation was performed separately for each stimulus contrast level (set of 400 trials) in each experiment as follows. First, each m x m image of cone absorptions was transformed into an m x m amplitude spectrum using the 2D fast Fourier transform and discarding the phase information. This left the dimensionality unchanged (m rows x m columns x 28 time-points x 400 trials within a single contrast level). Second, the amplitudes were concatenated across space and time into a 2D matrix (400 trials x 28m2 values per trial). A 400-element vector labeled the trials by stimulus class (1 for clockwise and -1 for counter-clockwise). Third, the 2D matrix and 400-element vector with labels was used for training and testing a linear support vector machine (SVM) classifier on the amplitude images using MATLAB’s fitcsvm with 10-fold cross-validation, kernel function set to ‘linear’, and the built-in standardization option (to z-score each row of the data matrix). The learned classifier weights represented the best linear separation (hyperplane) between the two stimulus classes. With these trained weights, the classifier predicted the stimulus class label for the left-out trials in a given data fold. We used MATLAB’s kfoldLoss function to average the accuracy across the 10-folds, which yielded one accuracy measure (% correct) per contrast level per experiment.
The implementation of an ideal observer model was based on Geisler [53]. First, we computed the ideal observer outputs based on Geisler’s closed-form solution in equation 3 (p. 777). This solution computed the ideal observer discriminability between two stimuli at the level of the cones given only the expected number of absorptions per cone per stimulus. To compute these values, we used a noiseless version of our computational observer in ISETBIO (no photon noise, typical human optics without defocus, no fixational eye movements, single stimulus phase, and a cone mosaic at 4.5° eccentricity). The d-prime for each contrast level was converted to percent correct assuming an unbiased criterion. Second, we also computed the ideal observer percent correct using the sampled (noisy) absorptions, rather than the closed form solution (equation 5, p. 781). As expected, with a reasonably large number of trials, the two implementations converge.
To quantify the contribution of a given factor in the eye, we averaged the classifier accuracy for each stimulus contrast level across the 5 experiments to fit with a Weibull function (Eq 1). This resulted in full psychometric functions for each cone density, cone type, or optical quality level. We calculated error bars for each contrast level as the standard error of the mean across the 5 iterations. For each psychometric function, we defined the contrast threshold as the power of 1 over the slope of the Weibull function β, in our case β = 3, of the performance level expected at chance (0.5 for a 2-AFC task). This results in a contrast threshold taken at ~80% (0.51/3 = 0.7937, defined as α in Eq 2).
y=1-(1-g)∙e-(kxt)β
(1)
Where g is the performance expected at chance (0.5), t is the threshold, β is the slope of the Weibull function, and k is defined as:
k=(-log(1-α1-g))1β
(2)
The contrast thresholds, t, were summarized as a quadratic function as a function of L-cone probability (Fig 7), as linear function of defocus (Fig 8), or an exponential function of cone density (Fig 9, represented as a straight-line on a semi-log axis). The square of Pearson correlation coefficient was used to report proportion variance explained (r2) by the fit. These quadratic, linear, or log-linear fits enabled us to compute the change in cone density or the change in defocus needed to achieve a 1% increase in contrast threshold–similar to the meridional effect observed in human performance (~4.4% at the upper vertical meridian vs. ~3.4% at the horizontal meridian as seen in [5]).
|
10.1371/journal.pcbi.1002577 | Quantifying the Contribution of the Liver to Glucose Homeostasis: A Detailed Kinetic Model of Human Hepatic Glucose Metabolism | Despite the crucial role of the liver in glucose homeostasis, a detailed mathematical model of human hepatic glucose metabolism is lacking so far. Here we present a detailed kinetic model of glycolysis, gluconeogenesis and glycogen metabolism in human hepatocytes integrated with the hormonal control of these pathways by insulin, glucagon and epinephrine. Model simulations are in good agreement with experimental data on (i) the quantitative contributions of glycolysis, gluconeogenesis, and glycogen metabolism to hepatic glucose production and hepatic glucose utilization under varying physiological states. (ii) the time courses of postprandial glycogen storage as well as glycogen depletion in overnight fasting and short term fasting (iii) the switch from net hepatic glucose production under hypoglycemia to net hepatic glucose utilization under hyperglycemia essential for glucose homeostasis (iv) hormone perturbations of hepatic glucose metabolism. Response analysis reveals an extra high capacity of the liver to counteract changes of plasma glucose level below 5 mM (hypoglycemia) and above 7.5 mM (hyperglycemia). Our model may serve as an important module of a whole-body model of human glucose metabolism and as a valuable tool for understanding the role of the liver in glucose homeostasis under normal conditions and in diseases like diabetes or glycogen storage diseases.
| Glucose is an indispensable fuel for all cells and organs, but at the same time leads to problems at high concentrations. As a consequence, blood glucose is controlled in a narrow range to guarantee constant supply and on the other hand avoid damages associated with elevated glucose levels. The liver is the main organ controlling blood glucose by (i) releasing newly synthesized or stored glucose in the blood stream when blood glucose is low (ii) using and storing glucose when blood glucose is elevated. These processes are regulated by hormones, in particular insulin, glucagon and epinephrine. We developed the first detailed kinetic model of this crucial metabolic system integrated with its hormonal control and validated the model based on a multitude of experimental data. Our model enables for the first time to simulate hepatic glucose metabolism in depth. Our results show how due to the hormonal control of key enzymes the liver metabolism can be switched between glucose production and utilization. We provide an essential model to analyze glucose regulation in the normal state and diseases associated with defects in glucose homeostasis like diabetes.
| The human plasma glucose is kept in a narrow range between minimum values of ∼3 mM after prolonged fasting or extensive muscle activity and maximum values of ∼9 mM reached postprandially [1], [2]. Homoeostasis of plasma glucose is crucial for the organism: Hyperglycemia results in non-enzymatic glycosylation (glycation) and thus loss-of-function of proteins [3], glucose induced oxidative damage [4], [5] and other adverse effects [6], [7]. Hypoglycemia leads to an under-supply of tissues with glucose and is thereby of particular danger for neuronal cells, erythrocytes and fibroblasts, using glucose as dominant or even exclusive energy-delivering fuel under normal physiological conditions.
The liver is a central player in buffering plasma glucose contributing either by net hepatic glucose utilization (HGU) or net hepatic glucose production (HGP) depending on the plasma glucose level exceeding or falling below a critical threshold value (in the following referred to as ‘set point’) of ∼6 mM. Switching between HGP and HGU is therefore a switch between positive (i.e. export of glucose) and negative (i.e. import of glucose) net hepatic glucose balance. This crucial metabolic function of the liver is performed by hepatocytes which exhibit high capacity of glycogenesis, glycogenolysis, glycolysis and gluconeogenesis enabling them to transiently store substantial amounts of glucose as glycogen, to synthesize glucose from lactate, glycerol and glucoplastic amino acids and to convert excess glucose into triglycerides [2], [8], [9].
Glucose homeostasis is controlled by several hormones, with insulin and glucagon being the main counteracting players [1], [10]. Insulin is the only known hormone lowering blood glucose, whereas multiple glucose increasing hormones exist. Glucagon plays the primary role in counter-regulation to hypoglycemia. Epinephrine has a secondary role, becoming critical under impaired glucagon responses, but with reduced effectiveness compared to glucagon [11]–[13]. Other counter-regulatory hormones like cortisol or thyroxin play only a minor role for the liver [10], [12], [13]. The plasma concentrations of insulin, glucagon and epinephrine change as direct response to varying blood glucose [1], [10]. In the liver insulin increases the activity of glucose utilizing pathways (HGU, glycolysis, glycogenesis) and decreases glucose producing pathways (HGP, gluconeogenesis, glycogenolysis), whereas glucagon and epinephrine have contrary effects. Main targets of the gluco-regulatory hormones are key interconvertible enzymes of glucose metabolism like pyruvate kinase or glycogen synthase. The kinetics of the interconvertible enzymes, and consequently also the hepatic glucose metabolism, depends on their phosphorylation state [2], [14] which is altered by the hormones.
Despite the crucial role of the liver for glucose homeostasis, a detailed mathematical model of human glucose metabolism of the liver, indispensable for understanding the hepatic role under normal and impaired conditions like occurring in diabetes, has not been developed yet. Available kinetic models of hepatic glucose metabolism are either minimal models [15]–[18], concentrate on parts of the glucose metabolism [19], or lump reactions [20], [21]. Furthermore, all available models concentrate on the glucose-insulin system, and ignore the crucial insulin antagonists glucagon and epinephrine [15]–[19], or do not include hormonal regulation at all [20], [21]. All mentioned models with exception of [19] ignore the crucial dependency of enzyme kinetics on the respective phosphorylation state of the enzyme completely, an essential mechanism for short term regulation in hepatic glucose metabolism.
Here, we present the first model of hepatic glucose metabolism in molecular detail, which includes the crucial control of hepatic glucose metabolism by insulin, glucagon and epinephrine via changes in phosphorylation state of key enzymes based on a new concept of linking hormonal regulation with metabolism.
The model of human hepatic glucose metabolism (Figure 1) consists of 49 localized metabolites (Text S1) and 36 reactions (Text S2) compartmentalized into cytosol, mitochondrion and blood. Abbreviations of metabolites and reactions are defined in the legend of Figure 1. All reactions and transporters are modeled with individual kinetics (Text S3) with parameters from literature research or by fitting to experimental metabolite and flux data (Equation 1, Dataset S1). An annotated SBML of the model is provided in the supplement (Dataset S2). The model includes regulation via allosteric mechanisms (Text S4) and regulation by hormones via phosphorylation and dephosphorylation of interconvertible enzymes (Text S5). Enzymes regulated by allosteric mechanisms are GK, GS, GP, PFK1, PK, PC, PDH, FBP1 and FBP2. Enzymes regulated via changes in phosphorylation state are GS, GP, PFK2, FBP2, PK and PDH. All model fluxes are given in (bw = bodyweight).
The detailed kinetic equations for the reactions and transporters (Text S3) are specific for human liver and based on extensive literature research. All kinetic parameters except the maximal velocities are based on literature data with references given in the respective rate equation. All values in the model were determined by fitting simulated model fluxes to experimentally determined fluxes and simulated model metabolite concentrations to experimentally measured concentrations .
The fit procedure minimizes the sum of quadratic relative differences in fluxes and concentrations, divided by the total number of experimental fluxes or respectively the total number of experimental concentrations (Equation 1). Relative differences were used to avoid a domination of the optimization by large absolute values and to have dimensionless quantities. The contributions of fluxes and concentrations to least-square fit function were weighed equally () assigning same relevance to fluxes and concentrations.(1)The experimental flux data was human specific (Dataset S1), the metabolite concentrations were taken from human and rat liver and are given in (Text S1, Dataset S1). Equation 1 was minimized using the MATLAB® Optimization Toolbox (constraint nonlinear optimization) and resulted in a final value of , indicating that the overall remaining relative deviations of theoretical values from experimental ones were lower than a factor of 2. The fitted values are given in Text S2.
For the time course and steady state simulations the differential equation system (Text S3) was integrated with a variable-order solver for stiff problems based on numerical differentiation formulas with absolute integration tolerance and relative integration tolerance (ode15s MATLAB® R2011a, MathWorks). Initial concentrations for all simulations are given in Text S1. Variation in blood glucose and glycogen are given in the respective figure legends. The external concentrations of blood glucose and lactate were kept at fixed values in all simulations as their time evolution in the blood depends on the metabolic activity of various organs not considered in this model. The cellular redox state (given by the concentrations of NAD and NADH) and cellular energy state (given by the concentrations of the adenine nucleotides) were kept constant during all simulations.
Steady state solutions are defined as states with absolute changes in every concentration smaller than for a time interval . Steady state solutions were tested to be stable against small changes in initial concentrations (1%). All conservation entities were tested to be constant within the tolerances over the integration.
As the release of hormones and their elimination from the plasma are not part of the model, the relationships between plasma glucose level and hormone levels (Figure 2) are described by phenomenological functions called glucose-hormone responses (GHR). The GHRs are sigmoid functions ranging between basal hormone concentration hbase and maximal hormone concentration hmax, monotonically increasing with increasing blood glucose for insulin (Equation 2), monotonically decreasing for glucagon and epinephrine (Equation 3). The GHRs were determined by least-square fit to oral glucose tolerance tests and hypoglycemic, hyperinsulinemic clamp studies (Table 1, Dataset S1). The parameters correspond to , , the inflection point and the Hill coefficient which determines the steepness of the sigmoidal hormone response.(2)(3)Experimental data and standard deviations (Dataset S1) were extracted from figures, tables and supplemental information. Data points correspond to mean data for multiple subjects from the studies. No data points were omitted.
The short-term effects of insulin, glucagon and epinephrine result from changes in the phosphorylation state of key interconvertible enzymes of glucose metabolism, namely GS, GP, PFK2, FBP2, PK and PDH. The interconvertible enzymes exhibit different kinetic properties in the phosphorylated state (P) and dephosphorylated state (DP), thus carrying different fluxes in the phosphorylated state () and dephosphorylated state (). The resulting effective kinetics (Equation 4) is the linear combination of and dependent on the phosphorylation state , with being the phosphorylated and the dephosphorylated fraction of the enzyme.(4)The phosphorylation state is a phenomenological function of insulin, glucagon and epinephrine (Equation 5). Insulin decreases , whereas glucagon and epinephrine increase . Epinephrine acts as a backup system for glucagon with reduced effectiveness . Only the currently dominating hormone (glucagon or epinephrine) was taken into account (max). The hormonal dependencies on the phosphorylation state were modeled by a Michaelis-Menten like hyperbolic function with half-saturation at whereby the saturation parameter was set to for all hormones. The maximal basal hormone concentrations of the three hormones () result from the respective GHR curves and are given in Table 1.(5)We assumed that for given hormone concentrations the fraction of all interconvertible enzymes is equal.
Experimental data (Table 2, Dataset S1) was extracted from figures and tables of 25 independent studies with different tracer methods. See [2] for review and Table 2 for detailed references, used methods and experimental data. Every data point is the mean for multiple subjects from one of the studies. No data points were omitted.
Experimental data (Dataset S1) was extracted from figures and tables from [22] (individual data) and [23] (mean person data with STD) for the glycogenolysis and from [8], [24], [25] (mean person data) for the glycogen synthesis. The only difference between model simulations in states of either net glycogen synthesis or net glycogenolysis simulations are differences in the initial glycogen and blood glucose concentrations. Experimental data for the mixed meal and hyperglycemic, hyperinsulinemic clamp simulations were extracted from [26] using the experimental time courses of glucose as model input. Insulin and glucagon time courses were predicted with the respective hormone dose response curves (Equation 2 and Equation 3). Simulations were started the from initial glycogen concentrations reported in the experiments.
Experimental data (Dataset S1) were extracted from figures and tables from [27]. Basal hormone concentrations of dogs and humans differ due to difference in human and dog physiology. Therefore, in model simulations of the human liver typical insulin and glucagon concentrations for the basal glucose production rate in humans were used (insulin and glucagon ) and concentration changes for the hormones expressed relative to these basal values Changes in blood glucose were calculated from the simulated HGP, for a mean bodyweight of , blood volume of and whole-body basal glucose utilization rate of . In Simulation 1 no experimental data was taken into account, but only a drop in the respective hormones during the infusion period (135–200 min) assumed. In Simulation 2 the experimentally observed changes in hormones and GU relative to basal levels (mean of values during the pre-infusion period) were used.
The HGRC defined by Equation 6 measures the change in net hepatic glucose balance defined through the net rate of glucose exchange between hepatocytes and the blood () elicited by a small change in blood glucose concentration. In model simulations, the HGRC was approximated by a difference quotient and numerically calculated from the steady state GLUT2 fluxes. For the HRGC corresponds to the changes in HGU, for to the change in HGP.(6)
The reactions for citrate efflux, oxaloacetate influx and acetyl-CoA flux were included in the model, to test the model under various TCA cycle loads and changes in acetyl-CoA demand and production being a necessary condition in the model development of a functioning model of hepatic glucose metabolism under typical physiological TCA cycle loads. The malate-aspartate shuttle (MALT) including cytosolic and mitochondrial malate dehydrogenase (MDH) was not modeled in detail. Thus, the rate of the cytosolic isoform of the PEPCK was put to directly depend on mitochondrial oxaloacetate concentration. For the actual simulations these boundary fluxes where set to zero.
To characterize the regulatory importance of various reactions in different states of hepatic glucose metabolism we used metabolic flux response coefficients (Equation 7), describing how the flux rate of an arbitrary reaction changes in response to a small change in the maximal activity of an arbitrary reaction .(7)Response coefficients with respect to the maximal activities of all enzymes were calculated at varying values of external glucose and cellular glycogen content for key fluxes of hepatic glucose metabolism (Figure S9): hepatic glucose uptake/release (), gluconeogenesis/glycolysis () and glycogenolysis/gluconeogenesis ().
We present the first detailed kinetic model of human hepatic glucose metabolism integrated with the hormonal regulation via insulin, glucagon and epinephrine (Figure 1 and Methods) via a new concept of changes in phosphorylation state (Figure 2 and Methods). The model was validated with quantitative and qualitative experimental data from a multitude of studies on hepatic glucose metabolism in overnight fasting, short term fasting and postprandial glycogen storage (Table 3, Figure 3, Figure 4, Dataset S1). In a next step the model was applied to analyze the central role of the liver in glucose homeostasis, especially the hepatic capability (i) to switch between an anabolic glucose mode (HGP) to a catabolic glucose mode (HGU) depending on the blood glucose concentration and hormonal signals (iii) to use glycogen as glucose buffer in this process (iii) to respond to counteracting changes in blood glucose with changes in glucose production (HGP) and glucose utilization, especially in the fasting and postprandial physiological state (Figure 5).
The plasma concentrations of the gluco-regulatory hormones change with changes in plasma glucose concentration. With increasing glucose the insulin level increases (Figure 2C), whereas the levels of glucagon (Figure 2A) and epinephrine (Figure 2B) decrease. Consequently, the phosphorylation state of the interconvertible enzymes (Figure 2D, Equation 5) changes from phosphorylated (94% at 2 mM) to dephosphorylated (5% at 14 mM) with increasing blood glucose. Changes in the phosphorylation state are accompanied by changes of the kinetic properties of the respective enzymes (see Equation 4). Due to the higher activity of the interconvertible enzymes of HGP (GP and FBP2) in the phosphorylated form and the increased activity of the enzymes of HGU (GS, PFK2, PDH, PK) in their dephosphorylated form (Text S5), the hepatic glucose metabolism is shifted from a glucose producing phenotype (HGP) under low glucose concentrations to a glucose consuming phenotype (HGU) at hyperglycemia. This change from an anabolic metabolism of glucose production (HGP) to a catabolic mode of glucose utilization (HGU) is a short term adaptation via changes in the kinetic properties of crucial enzymes of glucose metabolism. By adapting the HGP/HGU, gluconeogenesis/glycolysis and glycogen metabolism to the current hormonal signals and blood glucose concentration, the liver is able to fulfill its important role in glucose homeostasis in a variety of physiological states ranging from hypoglycemic states in overnight and short term fasting to hyperglycemic states postprandial.
The liver is the main glucose supplier in overnight fasting and short term fasting. The produced hepatic glucose (HGP, Figure 3A) results either from de novo synthesis via gluconeogenesis (GNG, Figure 3B) or from degradation of hepatic glycogen via glycogenolysis (GLY, Figure 3C). The relative contributions of these two processes to HGP change over the time course of fasting with gluconeogenesis becoming more and more important, whereas the share of glycogenolysis to HGP decreases (GNG/HGP, Figure 3D).
The simulations for short term fasting under constant glucose concentrations between 5 mM (green) and 3 mM (red), the normal range of fasting glucose concentration [22], [23], are in agreement with the experimental data (Figure 3, Table 2). Taking into account the gradual decrease in blood glucose concentration during fasting from 5 mM to 3.6 mM (blue), an agreement between simulation and experimental data is observed.
HGP decreases with ongoing fasting to a constant basal rate of 7–8 at around 40 h (Table 3D, Table 3F). With increasing blood glucose HGP decreases. Gluconeogenesis rate is constant for given blood glucose concentrations (see [2] for review). Taking into account the gradual decrease in blood glucose over fasting the rate of gluconeogenesis increases gradually (blue). In contrast, glycogenolysis decreases sharply during fasting due to the emptying glycogen stores (see also Figure 4A). As a consequence, the fractional contributions of glycogenolysis and glycogen synthesis to HGP shift from initially equal contribution of both processes to glucose finally completely synthesized de novo. Whereas after an overnight fast (∼10 h) half of the HGP comes from glycogenolysis [2], after 40 h fasting only 10% of HGP result from glycogen, 90% from gluconeogenesis.
Liver glycogen has an important role as a short term glucose buffer for glucose homeostasis. At low blood glucose concentrations, like in the fasting state or during extensive muscle activity, glucose is released from glycogen (Figure 4A), whereas during periods of high blood glucose like postprandial glycogen is synthesized from glucose (Figure 4B).
Over 40–50 h short term fasting the glycogen stores are emptied. During an overnight fast glycogen is utilized and the resulting glucose from glycogenolysis is exported resulting in half-filled glycogen stores after ∼16 h (Table 3E). The rate of glycogenolysis is almost constant and decreases only at low glycogen concentrations (Table 3E).
The simulations for glycogen depletion under hypoglycemia (Figure 4A) for glucose concentrations between 3.6 mM (red) and 5 mM (green) are in agreement with experimental data [22], [23]. The rate of glycogenolysis depends on the blood glucose concentration (see also Figure 3C). With decreasing blood glucose concentration, the rate of glycogenolysis increases, glycogen stores are emptied faster. As a consequence of the drop in glucose concentration over fasting, simulations at low glucose concentration overestimate the decrease in glycogen, simulations at high glucose concentrations underestimate the depletion. When taking the gradual drop of blood glucose from 5 to 3.6 mM into account (blue), the concordance of simulations experimental data further improved.
Blood glucose levels are elevated postprandial and glycogen is stored via glycogen synthesis (Figure 4B). The simulations for glycogen synthesis under hyperglycemia for different glucose concentrations between 5.5 mM (red) and 8.0 mM (green) (normal range of postprandial glucose concentrations) are in good agreement with the experimental data [8], [24], [25], especially the simulation at 7 mM (blue) representing a normal blood glucose value postprandial. With increasing blood glucose the rate of glycogen synthesis increases and the hepatic glycogen stores are filled faster. For medium filled glycogen stores between 200 and 300 mM the rate of glycogen synthesis is constant for a given blood glucose concentration.
To further evaluate the predictive capacity of our model, we performed time-course simulations of experimentally determined glycogen levels in dogs monitored under conditions of hyperglycemic, hyperinsulinemic clamps and administration of a mixed meal diet [26], as can be seen from the respective figures in the supplement (Figure S1, S2, S3, S4, S5, S6), our model simulation were in good agreement with the measured time courses of insulin, glucagon and glycogen. Of note, in these simulations none of the data was used for the calibration of the model. Moreover, successful simulation of these experiments under conditions where the two hormones insulin and glucagon could be varied independently from each other while in vivo their levels are coupled by the glucose level of the blood demonstrates the validity of the phenomenological glucose-hormone response functions (Equation 2 and 3) and interconversion –versus-hormone function γ (Equation 5) used in our model.
To analyze the effects of insulin and glucagon on HGP classical hormone perturbation experiments of hepatic glucose metabolism were simulated [27]. In these experiments a deficiency in either insulin or glucagon or both was achieved by somatostatin infusion, an inhibitor of insulin and glucagon secretion, in combination with hormone replacement infusions. Figure 4C depicts exemplarily the effect of a transient drop of insulin which is characterized by a marked increase in HGP and a consequent rise in plasma glucose concentration. HGP and plasma glucose return to normal after the perturbation. Simulations of other cases (saline control, insulin and glucagon depletion, glucagon depletion and somatostatin in combination with insulin and glucagon restoration) are depicted in Figure S1, S2, S3 and S5, respectively. For all hormone perturbations the time courses of HGP as well as glucose are in good agreement with the experimentally observed changes.
Predicted changes in basal glucose production (HGP) are very similar to the experimentally observed changes (Figure 4C). Insulin and glucagon depletion or glucagon depletion alone reduce the HGP to around 70% of basal values, insulin depletion increases HGP to around 130%. Insulin and glucagon have a dramatic effect on the human hepatic glucose metabolism, with basal glucagon being responsible for about 30% of HGP and basal insulin preventing increased HGP as a consequence of an unrestrained glucagon action [27].
The special role of the liver for glucose homeostasis results from the ability to switch between an anabolic glucose producing mode (HGP) to a catabolic glucose utilizing mode (HGU) depending on the blood glucose concentration and hormonal signals. In this process the hepatic metabolism is altered from glucose production via gluconeogenesis and glycogenolysis at hypoglycemia to glucose utilization via glycolysis and glycogen synthesis at hyperglycemia – a short term switch between metabolic pathways occurring in the range of minutes.
To analyze the temporal response of hepatic glucose metabolism to rapid variations in the external glucose concentrations we performed time-dependent simulations with a step-wise constant concentration profile of blood glucose and constant internal glycogen concentrations. (Figure S7 and S8). To evaluate the impact of hormone induced fast changes in the phosphorylation state of key regulatory enzymes on metabolic regulation we performed these simulations also in the absence of this mode of regulation in the glycogen metabolism, i.e. frozen phosphorylation state of glycogen synthase and glycogen phosphorylase and regulation of these enzyme activities only by allosteric effects. We found clear differences in the simulated time-courses (black and red curves in Figure S7 and S8) indicating that hormonal regulation contributes substantially in the rapid adaptation of the network to abrupt changes of blood glucose. In both cases (i.e. presence or absence of hormonal control) the simulation revealed that even after changing the external glucose concentrations abruptly by 2 mM and more a new steady state was reached within several minutes.
Based on the above finding that the a new metabolic steady-state (except for glycogen) is achieved within a few minutes after abrupt changes of the blood glucose level it appeared justified to treat the metabolic network as being in a quasi-steady state as long blood glucose changes are slow and the glycogen pool is quasi constant.
Hence, the quasi-steady state approximation was applied to simulate HGP/HGU at varying concentrations of blood glucose and cellular glycogen (Figure 5). In these calculations, the external glucose concentration was varied between constant values of 2 to 14 mM and the glycogen concentration kept at constant values between 0 and 500 mM. For each couple of glucose/glycogen values the resulting HGP/HGU (Figure 5A), the contribution of gluconeogenesis/glycolysis to HGP/HGU (Figure 5C), the contribution of glycogenolysis/glycogen synthesis to HGP/HGU (Figure 5D) and the ability of the liver to respond to changes in blood glucose expressed through the response coefficient HGRC (Figure 5B) in this quasi steady-state where analyzed. The fluxes depicted Figure 5 for a given constant glucose and glycogen concentrations are the reached steady state fluxes for this glucose and glycogen concentrations.
The hepatic glucose metabolism switches between anabolic HGP and catabolic HGU at blood glucose concentrations of 6.6 mM for half-filled glycogen stores (Figure 5A). The set point above the normal glucose concentration of around 5–5.5 mM [1], [28] is in line with the role of the liver as glucose producer under normoglycemic conditions [1], [29]. The liver contributes to glucose homeostasis by exporting glucose below the HGP/HGU set point (red) and importing glucose above this threshold (green) (Table 3A). With increasing blood glucose levels, HGP decreases and HGU increases, in accordance with reported suppression of glucose production and increase in HGU with increasing postprandial glucose level [8]. Glycogen has almost no influence on HGU whereas for low blood glucose concentrations HGP strongly depends on glycogen due to increased contribution of glycogenolysis to HGP with increasing glycogen content (Figure 5D). For low glucose concentrations the increase in glycogenolysis with increasing glycogen is strongest for low glycogen levels, more moderate for glycogen concentrations above 200 mM. HGP is maximal for filled glycogen store and low blood glucose.
The switch between gluconeogenesis and glycolysis occurs at 8.5 mM glucose for half-filled glycogen stores (Figure 5C). Gluconeogenesis and glycolysis rate are mainly determined by the prevalent blood glucose and only marginally depend on glycogen. Below 8 mM blood glucose gluconeogenesis is remarkably constant [2] (see also Figure 3B).
For plasma glucose levels below 5.1 mM glucose is released from glycogen stores via glycogenolysis, above 5.1 mM glycogen is synthesized for half-filled glycogen stores (Figure 5D, Table 3B) with rates of glycogenesis and cumulative glycogen as reported (Table 3C). In contrast to gluconeogenesis/glycolysis, glycogen metabolism is markedly affected by the glycogen content. Glycogenolysis is almost constant for partially filled glycogen stores and decreases only for low glycogen concentrations (Table 3H). Glycogenesis increases with increasing glucose concentration (Figure 5D), which is in accordance with the view that blood glucose determines the maximal rate of glycogenesis [8].
Most interestingly the switching behavior for glycogenesis/glycogenolysis and glycolysis/gluconeogenesis is very distinct, with different set points and different dependencies on glucose. Gluconeogenesis is almost constant for low blood glucose and glycolysis increases with increasing glucose. In contrast glycogenolysis is almost constant for high blood glucose and increases with decreasing blood glucose at low blood glucose. In combination of the two processes a HGP/HGU output of the liver is generated with a set point at normal blood glucose concentrations which is able to react over the whole range of physiological glucose concentrations (Figure 5A).
As a consequence of the different set points for gluconeogenesis/glycolysis and glycogenesis/glycogenolysis, hepatic glycogen can be synthesized via two alternative pathways: the direct pathway, in which glucose taken from the blood (HGU) is directly stored as glycogen above the gluconeogenesis/glycolysis set point of ∼8.5–8.8 mM; the indirect pathway, in which glucose synthesized via gluconeogenesis, is stored as glycogen for blood glucose concentrations between 5 mM and the HGP/HGU set point of ∼6.6–7 mM; In the intermediate region between the set points of HGP/HGU and gluconeogenesis/glycolysis glycogen is synthesized via the direct and indirect pathway with varying contributions [8] (Table 3G).
The effects of changes of values on HGP/HGU, gluconeogenesis/glycolysis and glycogenolysis/gluconeogenesis were analyzed via response coefficients [30] (Figure S9, Equation 7). The important findings of this analysis are as follows: (i) The effect of changes in the is strongly dependent on the external concentration of glucose and to a much lesser extent on the glycogen level. (ii) Under hypoglycemic conditions main control of hepatic glucose metabolism is exerted by a completely different set of reactions than in hyperglycemia. (iii) The response of the opposing pathway couples gluconeogenesis/glycolysis is very different to the glycogenolysis/gluconeogenesis. (iv) A small group of enzymes catalyzing irreversible reaction steps (GK, G6PASE, GP, PFK1, FBPASE1, PFK2, and FBPASE2) have a significant influence on the hepatic glucose metabolism whereas the majority of enzymes have only marginal effects. This control analysis is of particular interest for gene expression studies of hepatic glucose metabolism like for instance studies of circadian or feeding induced changes in hepatic expression [31].
The hepatic counter-regulatory capacity to changes in blood glucose was evaluated using the HGRC (Equation 6, Figure 5B), a measurement of the ability of the liver to react to changes in the blood glucose with changes in HGP or HGU, respectively. Our simulations showed that the liver is able to respond over the whole range of physiological blood glucose concentrations with being the lower limit, i.e. a change of 0.01 mM in blood glucose results in a change of HGP/HGU of at least . Intriguingly, our analysis suggest the counter-regulatory response of the liver to variations in the external glucose concentration to be particularly pronounced around the HGP/HGU set point at ∼6.6 mM which is flanked by a strong counter-regulatory response to hypoglycemia and a weaker response to elevated blood glucose levels. The strong counter-regulation to a decrease of blood glucose below 5 mM results in an effective increase in HGP whereby the rise of HGP depends on the glycogen content, with most effective counter-regulation for filled glycogen stores. The liver counteracts falling blood glucose levels to avoid hypoglycemia with a strong response. Furthermore, an increased response to elevated blood glucose levels of above 7.5 mM is seen enabling the hepatocyte to react efficiently to elevated blood glucose levels as occurring postprandially (6–10 mM). The hepatic glucose metabolism has ideal regulatory properties to react to the typical physiological challenges to glucose homeostasis: counter regulation to hypoglycemia in fasting and under extensive muscle activity and counter regulation to postprandial increase in blood glucose.
We present the first detailed model of human hepatic glucose metabolism integrating hormonal regulation by insulin, glucagon and epinephrine based on a novel concept to couple the level of these hormones with the phosphorylation state of interconvertible enzymes. This model enables for the first time the analysis of the hepatic carbohydrate metabolism at molecular level, including hormonal regulation. Furthermore, we provide a novel method to integrate hormonal signals with metabolism based on changes in phosphorylation state.
Model simulations are in good agreement with experimental data from a multitude of studies by different laboratories, researchers and methods. We want to emphasize, that the model was not fitted to single study data, but instead, data from a multitude of different studies covering various aspects of glucose metabolism were used. Thereby, the fundamental properties of human hepatic glucose metabolism and not individual study properties could be captured. Remarkably, the agreement of model simulations with numerous experimental and clinical findings was achieved without any re-fitting of model parameters and under neglect of other gluconeogenic substrates than lactate and regulatory phenomena on slow timescales as insulin-dependent changes in the expression level of metabolic enzymes. The model clearly underlines the importance of short term regulation of metabolism by interconvertible enzymes, being able to adapt hepatic metabolism to hormonal signals and glucose levels, and in this process being able to switch between anabolic and catabolic modes even within metabolic pathways.
We simulated the response of the human liver to changes in blood glucose under varying glycogen concentrations and assessed the contributions of glycogen metabolism and glycolysis/gluconeogenesis to HGP/HGU. Thereby we provide essential data for the understanding of the role of the liver in glucose homeostasis, which is not accessible experimentally. The underlying metabolic network reaches quasi-steady state within minutes after perturbations in plasma glucose (see Figure S7 and Figure S8). The dynamics of hepatic glucose metabolism is therefore mainly determined by the depletion/filling of the glycogen stores and the external glucose concentrations under normal conditions. As a consequence, the quasi-steady state system responses, shown for the key fluxes of the glucose metabolism in Figure 5, provide a good approximation of the state of hepatic glucose metabolism at given concentrations of blood glucose and temporary filling state of the glycogen store.
Furthermore, we analyzed the hepatic response to changes in blood glucose over the whole physiological range of blood glucose concentrations (3–11 mM), thereby integrating the available experimental data from the research field of hypoglycemia (accessible with hypoglycemic, hyperinsulinemic clamps <5.5 mM) and of elevated glucose concentrations (accessible with oral glucose tolerance tests OGTT>5 mM).
A model is always an abstraction of reality describing a certain subset of biological phenomena. The underlining assumptions are crucial for to understand the range of application of the model and the limitations. Main model assumptions and simplifications are (i) usage of phenomenological functions to incorporate hormone-induced signal transduction, (ii) a constant cellular redox- and energy status (iii) modeling of an ‘average’ hepatocyte, i.e. neglecting metabolic zonation of hepatocytes along the sinusoid [32] and (iv) no inclusion of changes in the gene expression of metabolic enzymes.
(i) A crucial part of this model is the phenomenological description of signal transduction (), which takes into account a substantial body of qualitative knowledge about the effects of the individual hormones on hepatic glucose metabolism. Due to the lack of quantitative experimental data on signaling processes as, for example, concentrations, activities and phosphorylation states of key kinases and phosphatases at varying concentrations of glucose, insulin, glucagon and combinations of these factors, a more detailed description was not possible. The model reproduces the experimentally observed dependency of glucose metabolism on the hormones. Especially, the reproduction of the classical insulin and glucagon perturbation experiments of [27], performed as independent validation without taking into account in the modeling process, underlines the validity of such a simplified treatment of the signaling network in the case of the hepatic glucose metabolism. Throughout the model the phosphorylation state of all regulatory enzymes is the same at given concentration of glucagon and insulin. This simplification could be the reason why the activation of the glycogen synthase was overestimated in mixed meal and clamp simulations. A second simplification was that the fraction () of phosphorylated inter-convertible enzymes follow instantaneously the hormone concentrations in the plasma although signal transduction occurs in the range of some minutes [33]. Including time-dependent changes of proteins involved in the cAMP-dependent signaling cascade in a later and more advanced version of the model should result in slightly different time-courses of fluxes and metabolite concentrations elicited by abrupt changes of external metabolite- and hormone concentrations. However, the main results of this study refer to the normal physiological situation where significant changes of the blood glucose level occur in a time window of hours (Figure 3, 4 and 5) with the metabolic network always being in quasi-steady state.
(ii) The model is limited to physiological states of the liver where changes in the energy- and redox state can be neglected. Normally, the hepatic energy state is decoupled from the hepatic glucose metabolism, with -oxidation of fatty acids providing ATP and reduced redox equivalents (NADH). Consequently, the model is not able to simulate conditions where this assumption of energy decoupling is not valid, for example, under hypoxic or ischemic conditions.
(iii) The presented model describes the metabolic net behavior of the liver of a hepatocyte averaged over the liver and is therefore able to simulate the net effect of the liver on glucose balance, namely net HGP and net HGU. In reality the liver exhibits a zonated structure with varying capacities of gluconeogenesis and glycolysis of the hepatocytes along the sinusoid [32]. Some of the effects brought about by an ensemble of hepatocytes equipped with differing capacities of glucose metabolism cannot be simulated with a model of a ‘mean’ hepatocyte. For example, the remaining difference between absolute glucose production and net glucose production (Figure S6) are due to the simultaneous presence of hepatocytes with net HGP (peri-portal hepatocytes) and HGU (peri-central hepatocytes).
(iv) Remarkably, the anabolic/catabolic switch was achieved by fast changes in the phosphorylation state and allosteric regulation of key enzymes of glucose metabolism without temporal changes in gene expression, i.e. changes in the protein levels of enzymes. In our opinion, changes in gene expression as observed to occur with different period lengths during the day [31] play no crucial role in short-term glucose homeostasis, a system which has to react to fast changes in the minute range of glucose supply (postprandial) and glucose demand (muscle activity) by an adaptation of metabolism. Changes in gene expression play a role in adapting hepatic metabolism on longer time scales to for example reduce futile cycles in counteracting pathways (like down-regulation of glucokinase and up-regulation of glucose-6 phosphatase under hypoglycemia) or match glucose metabolism to physiological requirements (up-regulating of glucose-6 phosphatase under hypoglycemic conditions to increase gluconeogenesis). The strong effects of such changes on hepatic glucose metabolism could be seen in the analysis of the response coefficients.
Finally, it has to be noted, that our model comprises fairly active futile cycles in the glycogen metabolism, the PFK1/FBPase1 system and between PEPCK and PK. During model building it was tried to minimize these cycles. Reducing and removing individual cycles in the modeling process compromised the ability of the model to switch between glucose consuming and glucose producing pathways, namely HGP/HGU, gluconeogenesis/glycolysis and glycogen synthesis/glycogenolysis. An essential property of the model is the simultaneous activity of glycogen synthase and glycogen phosphorylase and the accompanied futile cycle, necessary to reproduce the experimental time-courses for glycogen synthesis in fasting and postprandially.
Substrate cycling allows the system to react to fast changes in metabolite levels, the concomitant adaptation of key enzymes via phosphorylation or dephosphorylation causes an additional shift of the net flux in the right direction (see Figure S7). Taken together, metabolic regulation via phosphorylation/dephosphorylation of key enzymes in combination with futile cycles plays a key role in our model, in line with the results presented by Xu et al. [34].
Future work will apply the presented model to diseases of glucose homeostasis like diabetes, integrate the model in models of whole-body glucose metabolism and use individual patient data to generate individualized hepatic glucose models and analyze inter-individual differences in glucose metabolism.
|
10.1371/journal.pntd.0006300 | No evidence for association between APOL1 kidney disease risk alleles and Human African Trypanosomiasis in two Ugandan populations | Human African trypanosomiasis (HAT) manifests as an acute form caused by Trypanosoma brucei rhodesiense (Tbr) and a chronic form caused by Trypanosoma brucei gambiense (Tbg). Previous studies have suggested a host genetic role in infection outcomes, particularly for APOL1. We have undertaken candidate gene association studies (CGAS) in a Ugandan Tbr and a Tbg HAT endemic area, to determine whether polymorphisms in IL10, IL8, IL4, HLAG, TNFA, TNX4LB, IL6, IFNG, MIF, APOL1, HLAA, IL1B, IL4R, IL12B, IL12R, HP, HPR, and CFH have a role in HAT.
We included 238 and 202 participants from the Busoga Tbr and Northwest Uganda Tbg endemic areas respectively. Single Nucleotide Polymorphism (SNP) genotype data were analysed in the CGAS. The study was powered to find odds ratios > 2 but association testing of the SNPs with HAT yielded no positive associations i.e. none significant after correction for multiple testing. However there was strong evidence for no association with Tbr HAT and APOL1 G2 of the size previously reported in the Kabermaido district of Uganda.
A recent study in the Soroti and Kaberamaido focus in Central Uganda found that the APOL1 G2 allele was strongly associated with protection against Tbr HAT (odds ratio = 0.2, 95% CI: 0.07 to 0.48, p = 0.0001). However, in our study no effect of G2 on Tbr HAT was found, despite being well powered to find a similar sized effect (OR = 0.9281, 95% CI: 0.482 to 1.788, p = 0.8035). It is possible that the G2 allele is protective from Tbr in the Soroti/Kabermaido focus but not in the Iganga district of Busoga, which differ in ethnicity and infection history. Mechanisms underlying HAT infection outcome and virulence are complex and might differ between populations, and likely involve several host, parasite or even environmental factors.
| Human African Trypanosomiasis (HAT) occurs in two distinct disease forms; the acute form and the chronic form which are caused by microscopically indistinguishable hemo-parasites, Trypanosoma brucei rhodesiense and Trypanosoma brucei gambiense respectively. Uganda is the only country where both forms of the disease are found, though in geographically distinct areas. Recent studies have shown that host genetic factors play a role in HAT resistance and/or susceptibility, particularly by genes involved in the immune response. In this study, we identified single nucleotide polymorphisms in selected genes involved in immune responses and carried out a case-control candidate gene association study in Ugandan participants from the two endemic areas. We were unable to detect any polymorphisms that were robustly associated with either Tbr or Tbg HAT. However, our findings differ from recent studies carried out in the Tbr HAT another endemic area of Uganda that showed the APOL1 (Apolipoprotein 1) G2 allele to be protective against the disease which merits further investigation. Larger studies such as genome wide association studies (GWAS) by the TrypanoGEN network that has >3000 cases and controls covering seven countries (Cameroon, Cote d’Ivoire, DRC, Malawi, Uganda, Zambia) using the H3Africa customized chip reflective of African genetic diversity will present novel association targets (https://commonfund.nih.gov/globalhealth/h3aresources).
| The tsetse transmitted African trypanosomes are flagellated protozoa, a range of which cause disease in animals (known as Nagana) and humans (Human African Trypanosomiasis, HAT, also known as sleeping sickness). These diseases are responsible for significant morbidity and mortality [1–3] and therefore directly impact on public health and animal productivity. Current reports indicate that annual HAT incidence is on the decline, although under reporting is typical, especially in areas where conflicts and civil unrest interrupt control efforts and regular epidemiological surveys [4–6].
HAT is caused by two microscopically indistinguishable sub-species: Trypanosoma brucei rhodesiense that causes an acute form of the diseases that develops within a few weeks or months of infection, and Trypanosoma brucei gambiense that causes a chronic form of the disease that can take years to become patent. The acute form of the disease is prevalent in Eastern and Southern Africa while the chronic form of the disease is prevalent in West and Central Africa [4]. Uganda is the only country with active foci for both forms of the disease, though in geographically distinct regions.
Studies in the Democratic Republic of Congo (DRC), Cameroon, Cote D’Ivoire, Guinea and Uganda have found evidence for polymorphisms in HP, IL6 and APOL1 associated with outcome of infection [7–12]. In the present study, we investigated the possible association of selected gene polymorphisms with HAT by undertaking a candidate gene association study (CGAS) using case-control samples from the Tbr and Tbg HAT endemic areas of Uganda. The IL10, IL8, IL4, HLAG, TNFA, TNX4LB, IL6, IFNG, MIF, APOL1, HLAA, IL1B, IL4R, IL12B, IL12R, HP, HPR, and CFH genes that were selected have protein products that are involved in the HAT immune response. The CGAS approach was used to compare the frequencies of genetic polymorphisms between cases and controls in order to identify risk variants for HAT in the two Ugandan populations.
This study was approved by the Uganda National Council of Science (UNCST; assigned code HS 1344) following review by the IRB of the Ministry of Health. Participants were identified through community engagement and active field surveys; they gave written informed consent administered in their local language by trained local health workers. In instances where participants were below 18 years of age, consent was sought from a parent or primary guardian. Any individuals for whom it was not possible to obtain consent or blood samples were excluded from the study.
The Tbr HAT endemic area samples were from the traditional Tbr HAT foci in the South East of Uganda [13]. Samples were collected mainly from Iganga district and included individuals from the predominantly Basoga ethnic group, with a few Baganda, Banyole, Balamogi, Basiginyi, Itesot, and Japadhola ethnicities.
The Tbg HAT endemic area samples were from the traditional Tbg HAT foci in the Northwest of Uganda [13]. Samples were collected from Adjumani, Arua, Koboko, Maracha, and Moyo districts and comprised of individuals from the Kakwa, Lubgbara and Madi ethnicities. In both areas, only individuals who were born and lived in these traditional foci were selected, as they were most likely exposed to HAT for most of their lives.
HAT cases were defined as individuals in whom trypanosomes have been detected by microscopy in at least one of the body tissues including, blood, lymph node aspirates or cerebral spinal fluids. Controls were defined as individuals from the endemic area with no history or any signs/symptoms suggestive of HAT. Controls from the Tbg HAT endemic area were required to have no serological reaction to the CATT or Trypanolysis tests.
Blood was drawn by venipuncture and collected in EDTA/heparin vacutainer tubes (BD). Buffy coats were prepared from the whole blood in field laboratories using centrifugation, aliquoted, and then stored in liquid nitrogen in preparation for DNA extraction that was carried out at the Molecular Biology Laboratory, COVAB, Makerere University. The DNA was quantified using Qubit (Life Technologies).
This study was one of five studies of populations of HAT endemic areas in Cameroon, Cote d’Ivoire, Guinea, Malawi and Uganda. The studies were designed to have 80% power to detect odds ratios (OR) >2 for loci with disease allele frequencies of 0.15–0.65 and 100 cases and 100 controls with the 96 SNPs genotyped. The study design included an overall total of 462 samples, 239 samples from Tbr HAT endemic regions (120 cases, 119 controls) and 223 samples from Tbg HAT endemic regions (110 cases and 113 controls).
Power calculations were undertaken using the pbsize routine in Genetics Analysis Package gap version 1.1–16 in R [14].
The selection of the genes depended on prior knowledge of the genes and their association with the HAT. The following genes IL10 [9], IL8 [7], IL4 [15], HLAG [16], TNFA [7], TNX4LB [17], IL6 [7], IFNG [18], MIF [19], APOL1 [8], HLAA [20], IL1B [21], IL4R [21], IL12B [21], IL12R [21], HP [22], HPR [22,23], and CFH [24] were selected.
96 SNPs were selected for genotyping using two strategies: 1) SNPs that had been previously reported to be associated with HAT or 2) in the cases of IL4, IL8, IL6, HLAG and IFNG by using sets of SNPs in LD (r2>0.5) with each other, such that all bases in the gene were in LD with at least one SNP. The SNPs in this second group of genes were selected using a merged SNP dataset obtained from 10X coverage whole genome sequence data generated from 230 residents living in regions (DRC, Guinea Conakry, Ivory Coast and Uganda) where trypanosomiasis is endemic (TrypanoGEN consortium, sequences at European Nucleotide Archive Study: EGAS00001002602) and 1000 Genomes Project data from African populations. Linkage (r2) between loci was estimated using the PLINK v1.9 package (https://www.cog-genomics.org/plink/1.9/) [25] and sets of SNPs that covered the gene were identified. Some SNP loci were excluded during assay development or failed to genotype and were not replaced.
Approximately 1μg of gDNA per sample were submitted to INRA (Plateforme Genome Transcriptome de Bordeaux, France) for genotyping. A multiplex analysis (two sets of 80 SNPs each) was designed using Assay Design Suite v2.0 (Agena Biosciences). SNP genotyping was achieved with the iPLEX Gold genotyping kit (Agena Biosciences) for the MassArray iPLEX genotyping assay, following the manufacturer’s instructions. Products were detected on a MassArray mass spectrophotometer and the data acquired in real time with MassArray RT software (Agena Biosciences). SNP clustering and validation was carried out with Typer 4.0 software (Agena Biosciences). SNPs that failed genotyping at INRA and some additional SNPs were genotyped at LGC Genomics, Hoddesden, UK where SNPs were genotyped using the PCR based KASP assay [26]. A summary of the candidate genes and SNPs is shown in S1 Table.
The raw genotypic data were converted to PLINK format and quality control (QC) procedures implemented using the PLINK v1.9 package [25]. PLINK was used to determine the level of individual and genotype missingness, Hardy-Weinberg Equilibrium (HWE), estimate allele frequencies, and linkage disequilibrium (LD). Testing for population stratification and admixture was carried out using Admixture 1.3 [27] and the plot was visualized using StructurePlot2 [28].
Testing for the association of SNPs with HAT was done using a Fisher’s exact test [29] implemented in PLINK and producing a 95% confidence interval for the odds ratios. Controlling for multiple testing was implemented using a Bonferroni correction (α* = α/n, where α is the level of significance, n is the number of independent SNP association tests and α* is the adjusted threshold of significance) [30]. The Bonferroni correction assumes that each of the statistical tests are independent; however, this was not always true since there was some linkage disequilibrium between the SNPs in IL4, IL8, IL6, HLAG and IFNG which were subject to complete linkage scans. Where the assumption of independence is not true, the correction is too strict potentially leading to false negatives. Thus, an alternative correction for multiple testing was also employed. The Benjamini-Hochberg false discovery rate (FDR) estimates the proportion of significant results (p < 0.05) that are false positives [30,31].
Our study population consisted of 239 individuals from Tbr and 223 from the Tbg HAT endemic areas. The former comprised of 120 cases and 119 controls, who had a mean age of 43 ± 5 years, and a male to female ratio of 1:2. The Tbg HAT endemic area participants comprised of 110 cases and 113 controls, who had a mean age of 37 ± 5 years, and a male to female ratio of 1:1.
Uganda is the only country where both acute and chronic HAT are endemic [32]. The two forms of the disease however occur in geographically isolated regions [32]. The two samples represented two distinct forms of the disease and regions inhabited by distinct ethnic groups (Nilo-Saharan language speakers in the Tbg region and Bantu language speakers in the Tbr region). The cohorts were analyzed separately including initial quality control. Ninety-six (96) SNPs in 15 genes were genotyped from each of the Tbr and Tbg HAT endemic area samples as shown in S1 Table (the Plink MAP and PED files are available in S1 and S2 Data). Before association testing, histograms of the distribution of missing data both by individual and by locus (Supplementary Figures S1 Fig–S4 Fig) were inspected in order to identify appropriate cut-offs to apply in each population. Individuals with missing data or loci with missing data above the cut-off threshold were removed as were loci that were not in HWE, or those that were poorly genotyped [33,34].
Individuals with more than 20% or 15% missing data were excluded from the Tbr and the Tbg HAT endemic datasets, respectively, resulting in a final dataset of 238 (119 cases and 119 controls, 1:2 male to female sex ratio) individuals from the Tbr HAT endemic sample and 202 (99 cases and 103 controls, 1:1 male to female sex ratio) individuals from the Tbg HAT endemic sample (Supplementary Figures S1 Fig and S2 Fig). Similarly, loci that were missing more than 30% or 40% data were excluded from the Tbr and the Tbg HAT endemic area samples (Supplementary Figures S3 Fig and S4 Fig). We used a HWE p-value cut-off of 1 x 10−8 and further selection of loci below the HWE cut off was done basing on their genotype scatter plots to see which loci were to be excluded. In order to get a high LD between marker and causal SNPs, loci that were in a five SNP window after a single step with a variance inflation factor (VIF) [VIF = 1/(1-R2)] beyond 0.2 were excluded from both sample datasets. This was done because a high LD between marker SNPs increases redundancy and reduces power. After quality pruning, 79 loci from Tbr and 85 loci from the Tbg HAT endemic samples were included in the association testing.
Admixture was used to test for population structure that might confound the association study. Eight values of K ancestral populations from 1–8 were tested to identify which had the lowest coefficient of variations (CV) error. CV error was at a minimum for K = 4, but the CV error was very similar for all values of K (0.42–0.46) providing no persuasive evidence for any particular number of ancestral populations. The Admixture plot showed no clear evidence for any gross population structure and therefore no correction for population structure was applied in the analysis.
Six SNPs in the Tbr HAT endemic area and four in the Tbg endemic had raw p < 0.05 but none of these remained significant after Bonferroni correction (Table 1). Surprisingly, there was no evidence for association with any SNP in APOL1.
In this case-control CGAS, we found no evidence for variants associated with Tbr or Tbg HAT in two Ugandan populations. We tested for association between candidate genes and the disease caused by Tbg and Tbr separately as they present two distinct forms of the disease. Tbr and Tbg parasite resistance to human serum is mediated by different mechanisms which place distinct selective pressures on the host genes [35]. Furthermore, the two populations were from different broad ethnolinguistic groups, and were geographically isolated from each other [13]. Admixture analysis found no evidence of population structure with these SNP which might have reduced the power of the study (S5 Fig).
We found no SNP associated with HAT after multiple testing corrections. Our power calculations indicated that we had power to detect odds ratios > 2, however 7 of the 10 SNPs with P <0.05 had odds ratios < 2.0, which the study was not powered to detect. Larger populations would be required to confirm these findings and the data presented could be used to estimate the necessary sample size.
The most striking feature of the data was the absence of any association at APOL1. The APOL1 G2 (deleted allele for indel rs71785313) allele has been shown to be lytic to T. b. rhodesiense in vitro [36] and a recent study in the Soroti and Kaberamaido focus in Eastern Uganda found an association with APOL1 G2 and protection from Tbr HAT with an odds ratio of 0.2 [8]. The present study in the Busoga focus was well powered to discover such a strong effect, but the frequencies of APOL1 G2 in cases and controls were almost equal (8.1% and 8.6%, 95% odds ratio confidence interval: 0.37–2.34) which indicates that an odds ratio as large as seen in Kaberamaido (OR = 0.2, 95% odds ratio confidence interval: 0.07–0.48) is very unlikely to be seen in Busoga (Supplementary data S2 Table). The frequency of the G2 allele in the control population in Kabermaido (14.4%)[8] was higher than in Busoga (8.6%). Although this difference in G2 allele frequency is not significant with the sample sizes that were used (Chisq Test p = 0.12), it may be indicative of real differences between these populations in selection pressure on this allele. Despite their geographical proximity (240km) these populations have divergent ethnic backgrounds; with the Kaberamaido population being Luo speakers which is a Nilotic family language originating in Sudan and Ethiopia and the Busoga population being Niger-Congo-B (Bantu) language speakers with origins in West Africa. These linguistic groups are believed to have diverged over 5,000 years ago giving plenty of time for allele frequencies to diverge. Therefore, despite the well-established function of APOL1 in response to trypanosome infection and the evidence for protection associated with G2 in Kaberamaido [8], the role of APOL1 G2 in response to T. b. rhodesiense infection more generally remains to be clarified.
Despite the suggestively significant associations found at nine SNP loci, none of them passed Bonferroni correction for multiple testing [30]. The FDR_BH shows the rate of type 1 errors or false positives, eg for rs9380142 in HLA-G there is an 18% chance that this is a false positive and conversely a 82% chance that it is a true positive. There was a greater than 38% probability for each of these nine SNPs being associated with HAT [30,31]. The finding of suggestive associations in multiple populations would increase the probability that these are genuine associations with disease [37]. For example, our findings suggest that HLA-G variants may be important in both forms of the disease. These observations should be followed up in future studies.
|
10.1371/journal.pntd.0005439 | GOST: A generic ordinal sequential trial design for a treatment trial in an emerging pandemic | Conducting clinical trials to assess experimental treatments for potentially pandemic infectious diseases is challenging. Since many outbreaks of infectious diseases last only six to eight weeks, there is a need for trial designs that can be implemented rapidly in the face of uncertainty. Outbreaks are sudden and unpredictable and so it is essential that as much planning as possible takes place in advance. Statistical aspects of such trial designs should be evaluated and discussed in readiness for implementation.
This paper proposes a generic ordinal sequential trial design (GOST) for a randomised clinical trial comparing an experimental treatment for an emerging infectious disease with standard care. The design is intended as an off-the-shelf, ready-to-use robust and flexible option. The primary endpoint is a categorisation of patient outcome according to an ordinal scale. A sequential approach is adopted, stopping as soon as it is clear that the experimental treatment has an advantage or that sufficient advantage is unlikely to be detected. The properties of the design are evaluated using large-sample theory and verified for moderate sized samples using simulation. The trial is powered to detect a generic clinically relevant difference: namely an odds ratio of 2 for better rather than worse outcomes. Total sample sizes (across both treatments) of between 150 and 300 patients prove to be adequate in many cases, but the precise value depends on both the magnitude of the treatment advantage and the nature of the ordinal scale. An advantage of the approach is that any erroneous assumptions made at the design stage about the proportion of patients falling into each outcome category have little effect on the error probabilities of the study, although they can lead to inaccurate forecasts of sample size.
It is important and feasible to pre-determine many of the statistical aspects of an efficient trial design in advance of a disease outbreak. The design can then be tailored to the specific disease under study once its nature is better understood.
| Since many outbreaks of infectious diseases last only six to eight weeks, there is a need for trial designs that can be implemented rapidly in the face of uncertainty. The Generic Ordinal Sequential Trial (GOST) is a flexible statistical design for a randomised clinical trial comparing an experimental treatment for an emerging infectious disease with standard care. The details of the design are derived to satisfy a generic power requirement using large sample theory. The accuracy of the approach for moderate sample sizes is then checked using million-fold simulations, and found to be very reliable under a wide range of circumstances. Total sample sizes (across both treatments) of between 150 and 300 patients prove to be adequate in many cases, although more patients may be needed if the majority of patients die or if the majority experience complete recovery, as there is then less evidence available to distinguish between treatments. An advantage of the approach is that any erroneous assumptions made at the design stage about the proportion of patients falling into each outcome category have little effect on the error probabilities of the study, although they can lead to inaccurate forecasts of sample size.
| The 2013–15 Ebola virus disease epidemic in West Africa highlighted the need to be able to develop treatment trial protocols in a matter of weeks, rather than the months or even years that are more usually taken. Clinical research on epidemic infectious diseases has to take place when new cases are occurring. Urgency arises because the outbreak might subside before any lessons about treatment can be learnt, or worse, the outbreak might spiral out of control before effective therapies can be developed.
This paper presents statistical aspects of trial designs that can be developed in advance and then quickly be adapted for a particular outbreak. The Generic Ordinal Sequential Trial (GOST) is a flexible, off-the-shelf statistical design for a randomised clinical trial comparing an experimental treatment with standard care for an emerging infectious disease. Key aspects of GOST are fixed in advance, so that clinicians and statisticians can immediately adopt these generic features, and focus on the optional elements that have to be determined as well as the countless other tasks involved in initiating a clinical trial of this nature. The context envisaged is one where there are only weeks available for preparation, perhaps with limited knowledge of the natural history of the disease. This paper may also be a helpful illustration for research teams with longer to prepare for a trial. In that case, trial statisticians might wish to vary the fixed elements of the design and to explore the consequences using methods described in [1], perhaps applying the statistical code provided in [2].
The full name of GOST, Generic Ordinal Sequential Trial, includes the statistical terms ordinal and sequential. An ordinal scale is a categorisation of outcomes for which there is an intrinsic ranking (or order) of the categories in terms of desirability, but there is no specific numerical value attached to each one. A clinical trial is sequential if it is conducted using a sequence of successive analyses, each of which may resolve the primary clinical question and lead to the termination of the trial. The primary trial endpoint in GOST is an ordinal categorisation of patient outcome as recorded a specified number of days following randomisation, and the sequential monitoring will lead to stopping as soon as it is clear that the experimental treatment has an advantage or that sufficient advantage is unlikely to be detected. The trial is powered to detect a generic clinically relevant difference: namely an odds ratio of 2 for better rather than worse outcomes. Total sample sizes (across both treatments) of between 150 and 300 patients prove to be adequate in many cases, the precise value depending on both the magnitude of the treatment advantage and the nature of the ordinal scale.
As many features of GOST as possible are pre-specified so that much of the statistical section of the trial protocol can be developed in advance, before the nature of the disease is known and without details of the experimental treatment. Other elements, such as details of the ordinal outcome scale, the randomisation ratio and the day on which a patient’s primary assessment will be made, will have to be quickly determined by investigators once an outbreak occurs.
Patients will be randomised between an experimental treatment (E) and standard care (S), stratified by treatment centre and perhaps by one or two other key prognostic factors. Usually the allocation ratio will be set to 1:1 for simplicity and because expected sample sizes are minimised if this choice is made [3]. If, however, the availability of E were limited, then the allocation ratio could be modified to randomise more patients to S than to E.
The primary patient response will be the status of the patient, D days after randomization, classified into one of k outcome groups, C1, …, Ck. D is likely to be set at 7, 14 or 28 days. The outcome categories must be unambiguously defined and each patient must fall into exactly one of them. They must also reflect progressively less desirable states as one moves from C1 (the best outcome) to Ck (the worst outcome). Outcome C1 might reflect complete recovery and Ck death before Day D. Intermediate outcomes might include C2: alive and requiring only basic support and C3: alive but requiring intensive support, where these terms would need careful definition for specific diseases. It is not necessary for the number of patients in every outcome category to be large, and the method remains valid and accurate if one or more categories turn out to be completely empty, provided that at least two categories are well represented. The special case k = 2 allows for a binary outcome such as alive or dead.
Use of a response that is available after a short and fixed duration of follow-up reduces the risk of loss to follow-up and is essential if the trial is to yield an early conclusion. GOST is presented for the case of an ordinal response because expected sample sizes will be reduced if more than two outcome categories can be reliably identified [3]. Furthermore, at the outset of a trial concerning a new infection, it may not be clear whether the key issue will be the prevention of death or the reduction of morbidity. Using a categorisation that distinguishes between a number of outcome states will allow the trial to be informative if life or death proves to be the major issue or if fatalities prove to be rare and the need for intensive therapy becomes the key concern. In normal circumstances, a pilot study of conventionally treated patients might be used to determine a binary endpoint for the trial: here we are concerned to start the definitive randomised study as early in the outbreak as possible.
The probability that a patient on E achieves an outcome category that is any one of C1,…, Cj, is denoted by PEj. Achieving an outcome in any one of categories C1,…, Cj is preferable to being in one of the categories Cj+1,…, Ck, an event that occurs with probability 1 –PEj. The odds of the former event is OEj = PEj/(1 –PEj), and PSj and OSj are defined similarly for patients receiving S. The odds ratio Rj is defined by Rj = OEj/OSj. Notice that these definitions make sense for values of j from 1 to k– 1, but they are not used for j = k as PEk and PSk refer to the probability of a patient being in any of the outcome categories, which must be 1, and the corresponding odds values are undefined. The null hypothesis is that E has no effect, in which case PEj = PSj and so Rj = 1 for each value of j from 1 to k– 1. If this null hypothesis is true then the probability of concluding that E is better than S (an event that will be designated “E wins” hereafter) is set to equal 0.025. This is the one-sided risk of type I error (denoted by α), and the value of 0.025 is chosen for GOST to follow convention.
As well as considering the properties of the design when the treatment has no effect (the null hypothesis), we consider its properties when there is a tendency for patients to achieve better outcome categories on E than on S across the whole outcome scale (the alternative hypothesis). Thus, treatment with E might lead to a greater chance of complete recovery, a greater chance of complete or partial recovery, and a smaller chance of death. Specifically, situations are considered in which all of the odds ratios from R1 to Rk−1 are of equal magnitude (denoted by a common value R) and greater than 1. The design is constructed to ensure that, if R = 2, then the probability that E wins is 0.90. This is the power of the trial. The alternative hypothesis is a compromise between the desires to detect small but worthwhile treatment effects and to complete the trial quickly. For a binary outcome, an odds ratio of 2 corresponds to an increase in success rate from ⅓ on S to ½ on E, or from ½ to ⅔, or from ⅔ to ⅘. Typical sample sizes when GOST is employed are in the range 150–300 (totalled over both treatment groups). The value of 0.90 is chosen for the power of GOST as it is a conventional choice: choosing 0.80 would allow too large a risk of missing a treatment effect as large as R = 2. For an increase in success rate from ½ on S to ⅗ on E (R = 1.5), GOST will conclude that E is better than S with probability 0.47. The sample size would triple to 450–900 if a power of 0.90 were specified for the alternative R = 1.5.
The trial will be monitored using a series of up to 20 interim analyses, equally spaced by newly accrued patient responses. It will be seen in the results section that such a choice will lead to around 20 to 30 new responses (totalled over S and E) being needed between consecutive interim analyses. It will also be seen that typically only 8 to 12 of these analyses will be required before the trial stops. The data requirements for each patient at each interim analysis are modest: patient identification number, date of randomisation, treatment, treatment centre and any other baseline stratification factors, and status on Day D. Setting 20 interim analyses for GOST is a subjective choice of the authors achieving a much quicker reaction to the message of the data than setting just 3 or 4 interim analyses while being more practical than updating the sequential plot every time a Day D report is received.
At each interim analysis two test statistics are calculated. The first is a cumulative measure of the observed advantage of E over S and is denoted by Z. The second quantifies the amount of information about the treatment difference contained in Z, and is denoted by V. Expressions for computing Z and V, allowing for stratification factors, are taken from [4] and presented in equations (E1) and (E2) of the Supporting Information (S1 Text. Supporting technical details). The monitoring of GOST can be depicted by a plot of the values of Z computed at each interim analysis against the corresponding values of V, using the diagram shown in Fig 1. A completed plot is presented in the results section. The stopping rule is represented by two straight lines. If Z lies above the upper line, the trial is stopped and E wins. If the plotted value of Z lies below the lower line, the trial is stopped and it is concluded that no evidence that E is better than S has been found. This design is a special case of the triangular test [1, 2], and it was proposed as the phase III part of a trial strategy for Ebola virus disease [5, 6].
Fig 2 shows the probability that E wins, plotted against the natural logarithm, θ, of the true odds ratio R. When R = 1 (θ = 0) the plotted probability is 0.025 and when R = 2 (θ = 0.693) it is 0.90. Fig 3 shows the probability of stopping at or before selected interim analyses, plotted against the true value of the log-odds ratio θ. In both of these figures values of θ corresponding to selected values of the odds-ratio R are also indicated on the horizontal axis. Although a maximum of 20 analyses is allowed, it is very unlikely that more than 16 will be required. If the treatment is either harmful (R < 1, θ < 0) or very efficacious (R > 2.7, θ > 1), then it is unlikely that more than 4 interim analyses (one fifth of the maximum sample size) will be needed. Fig 4 shows the expected value of V at the end of the trial (that is the average value of the final value of V over many iterations of the same trial) plotted against θ. As discussed later, the values of V in Fig 4 can be converted into expected final sample sizes. During the trial, V will be calculated according to equation E2 in S1 Text, but prior to the trial starting, the relationship between sample size and V can be approximated using equation E3 in S1 Text to yield a plot of expected terminal sample size against θ, as will be illustrated in the results section below.
The primary analysis will be based on the sequential design used, and will feature a one-sided p-value for the null hypothesis of no treatment difference, and a median unbiased estimate and 95% confidence interval for R. In the final dataset, the numbers of new patients recruited to each treatment arm may not be as planned in the protocol, the allocation ratio might not be as intended and the information V accrued might not be as anticipated. Provided that departures from the plan are purely chance deviations rather than being prompted by emerging data, actual values of these quantities will be used in the analysis. Thus it is acceptable if an unexpected surge of recruitment leads to there being more information available for an interim analysis than anticipated, but it is not acceptable for investigators to see a value of Z close to the stopping boundary and to bring forward the next interim analysis in the hope of a quick conclusion. The valid analysis is described in [1] and statistical code for its implementation is provided in [2]. Conduct of the final analysis will need expert statistical input. Unlike the finalisation of the design, there should be sufficient time for the trial statistician to study and practice these methods ahead of the trial reaching a conclusion. Although the final analysis will require technical input, the conclusion of the trial—whether E wins or not—will be immediately apparent from a glance at the plot of Z against V.
When the trial is stopped, there may still be patients under treatment whose outcome is unknown, as well as patients whose status became available during the conduct of the interim analysis and its discussion. Data from these patients will be added into a final “overrunning” analysis [7], provided that they followed the protocol without any change of treatment due to the stopping of the trial. The latter might not be the case if the experimental treatment is suspected of being harmful and it is consequently withdrawn from current patients.
Consider comparing an experimental treatment (E) with standard therapy (S) for Middle East Respiratory Syndrome Coronavirus (MERS-CoV) motivated by a sudden increase in the number and geographical spread of incident cases. Randomisation is 1:1. We choose D = 28 days and outcome categories C1: alive and not receiving ventilation; C2: alive and receiving only non-invasive ventilation; C3: alive and receiving invasive mechanical ventilation and C4: dead. Data from an observational study [8] of 70 patients yield estimates of the probabilities of these four outcomes occurring for patients on S of 0.286, 0.043, 0.214 and 0.457 respectively.
In Table 1, these four outcome probabilities form Column 2. In the first of 12 sets of simulations, one million replicate runs of GOST were conducted in which these outcome probabilities governed the responses both for patients receiving S and for those receiving E. The results are shown in the second column of Table 2. The proportion of trials in which E won was 0.025; equal to the intended one-sided type I error rate, confirming the accuracy of the procedure. In the second set of simulations, outcome probabilities for patients receiving S were unchanged, but a common odds ratio of R = 1.5 was imposed and the respective probabilities 0.375, 0.048, 0.217 and 0.359 (shown in Column 3 of Table 1, and reflecting a shift to better outcomes) were used to generate patient outcomes on E. For the third set of simulations, the outcome distribution on S was again unchanged, but R was increased to 2. The results are shown in Column 4 of Table 2, showing that the intended power of 0.90 was achieved.
Nine more simulation runs were conducted. The outcome distributions for patients on S were changed to those shown in bold in Table 1 under Scenario 2, and then as shown for Scenarios 3 and 4. For each scenario, three outcome distributions on E were explored, corresponding to R = 1 (no treatment effect), 1.5 and 2. Scenario 2 uses a rounded version of the estimated distribution on S to demonstrate that precise values are unnecessary at the design stage. Scenario 3 represents a more extreme situation in which all patients either leave intensive care or die by Day 28, while in Scenario 4, most patients leave intensive care by Day 28, with the other three categories being unusual. Values reported in Table 2 for Scenarios 1 and 2 are virtually indistinguishable, but more patients are needed in the case of Scenario 3 or 4.
When interpreting the simulation results shown in Table 2, it is important to distinguish between what the trial designer anticipated as the truth before starting the trial and what was actually true. All simulations represent trials in which the investigators anticipated that Scenario 1 was true, even if they were wrong. As explained in the Supplementary Information (S1 Text), if Scenario 1 is true, then a maximum sample size of 440 will be sufficient to ensure that V eventually reaches the value where the stopping boundaries in Fig 1 meet, so that a conclusion must be reached. Thus, 22 new patient responses will be needed for each of the 20 interim analyses. The expected final values of the information statistic V shown in Fig 4 can be converted into expected final sample sizes under Scenario 1, and the latter are shown as the red curve in Fig 5. The expected sample size lies well below the maximum sample size of 440 whatever the true treatment effect. Investigators’ pre-trial forecasts of the simulated quantities are shown in the last three columns of Table 2.
Having set the design and forecast its properties assuming Scenario 1, the simulations are then conducted under the twelve different models displayed in Table 1. For Scenario 1, with R = 1, 1.5 or 2, the investigators’ predictions are confirmed as being very accurate: average sample sizes are at most 3 over forecast. Switching to Scenario 2 shows how minor imperfections in the anticipated model have negligible effects. Scenarios 3 and 4 are quite different from the design assumptions, and yet the simulated probabilities that E wins and the simulated average final values of V remain close to predictions. The average sample sizes needed to reach a conclusion are, however, considerably larger than anticipated. Being wrong about the underlying model at the design stage will have little effect on the error probabilities of the study, but it might lead to inaccurate forecasts of sample size. The design reacts to the true nature of the data collected to ensure that the appropriate sample size is collected. Note that neither the predictions nor the simulations of average sample sizes include patients who are receiving treatment at the time of analysis, but who have not yet provided a Day 28 response, nor those recruited during the conduct of what turns out to be the final interim analysis.
Table 3 presents data from a single simulated run of GOST and Fig 6 shows the resulting plot. This fictitious trial stopped at the 11th interim analysis with 242 patients, and E won. Using the approach described in [1], the one-sided p-value is found to be 0.016. The median unbiased estimate of the log-odds ratio θ is 0.568 with 95% confidence interval (0.059, 1.062). For the odds-ratio R, the median unbiased estimate is 1.76 with 95% confidence interval (1.06, 2.89). The simulation did not generate patient data that would be received by the investigators after this analysis, but in practice results would come in from study patients who were still being followed to 28 days at the time the data for the 11th interim analysis were extracted, and those who were recruited while that analysis was being undertaken. Provided that no change was made to the treatment of these patients, they could be included in a subsequent overrunning analysis [7], and this would become the definitive interpretation of the trial results.
We conclude this section with a brief account of the changes that would follow if the investigators chose to dichotomise patient responses into alive at 28 days (C1, C2 or C3), or dead (C4). Taking the rounded outcome probabilities of Scenario 2, and then combining those relating to the first three categories, leads to Scenario 3. GOST can be applied to such binary data, and equations E4 in S1 Text provide simplified versions of the test statistics. However, binary data are less informative than the ordinal version of the data, and it will now take 520 patient responses to ensure that V eventually reaches the value where the stopping boundaries in Fig 1 meet. Thus 26 new responses will be required at each interim analysis. The blue curve of Fig 5 indicates expected final sample sizes for the binary approach, and it can be compared with the red curve that corresponds to both Scenario 1 and Scenario 2, as the two are indistinguishable. The inflation in sample size due to dichotomising the ordinal scale is a factor of 1.18: an 18% increase in sample size. Additional simulations conducted using 26 new binary responses per interim analysis confirmed that the intended type I error rate 0.025 and the power of 0.90 were achieved, but the increase in average final sample sizes relative to those for the ordinal approach reported for Scenarios 1 and 2 in Table 2 ranged from 17% to 26%.
GOST has been devised for trialists in a hurry due to the speed with which a pandemic is emerging. It is intended that they use the GOST design as described in this paper. The investigators have to identify the outcome categories and the day D of their observation. They also choose the allocation ratio and any stratification factors. The rest is as presented above.
Usually, evidence from two or more trials is required for drug registration, although in certain circumstances evidence from just one is considered to be sufficient [9, 10]. It would be important to determine in advance whether a single trial would be sufficient in future outbreaks of infectious diseases. GOST provides an approach that could be used once or repeated in a replicate trial if deemed necessary.
A “platform approach” was suggested for trials of a series of experimental treatments in Ebola virus disease [11]. First a comparison of Treatment E1 with S is conducted. If Treatment E1 wins it becomes the new standard. Treatment E2 is then compared with the current standard, and so on. The α level required to declare a treatment superior to control is fixed at that relevant to a single trial, with no allowance for multiplicity of experimental treatments. GOST could be used as the design for each comparison made within the platform approach, with α being set at 0.025 throughout. Implementations of GOST that allow simultaneous randomisation between multiple experimental treatments and S are also possible.
The triangular test is just one of many sequential methods that could be used as the engine to drive GOST. Alternatives based on α = 0.025 and power 0.90 to detect an odds-ratio of 2 would be natural competitors. The triangular test is chosen because amongst tests satisfying the power requirement above, it minimises the maximum expected sample size, which occurs when R is close to 1.5 [12]. The efficiency of the triangular test is achieved from its asymmetry. Strong evidence is required for E to win, but if superiority is not apparent the trial will stop quickly without recommending E. The design does not seek to distinguish between lack of effect and harm: either way there is no further interest in E and resources are better devoted to other experimental treatments. The triangular test was devised over 50 years ago [13], and has been used extensively in a wide range of studies [14].
Adoption of the GOST design should be subject to approval of a Data and Safety Monitoring Board (DSMB), who consider unblinded data during the ongoing trial. They have the duty to recommend stopping the trial if they feel it unsafe to continue, considering the primary categorisation of status after D days and also data on other endpoints and from patients who have not yet been observed for D days. They will also be asked to confirm any stopping recommendation resulting from the triangular boundaries, taking account of information on patient progress not captured by the primary ordinal response, relevant external information, and indications of major discrepancies in treatment effect across patient subgroups.
The trial will also be overseen by a Steering Committee without access to unblinded trial data. This committee could, however, be provided with data on the sample size and the amount of information V available at each interim analysis. This would provide a reassessment of the relationship between these two quantities, as shown in equation E3 of S1 Text, that does not depend on pre-trial assumptions. To protect the accuracy of the trial, the Steering Committee might authorise a change in the numbers of new patient responses to be collected for each interim analysis to ensure that the increments in V are closer to their intended values. As this would be done without access to unblinded data, no bias would be introduced.
The triangular test itself is very flexible, and the approach can be reworked with different choices for α, power and R, and different numbers and patterns of interim analyses (although the name GOST is reserved for the specific case presented here). Normally distributed data, count data, survival data and other types of response can also be accommodated [1].
|
10.1371/journal.pntd.0000918 | Responses of Human Endothelial Cells to Pathogenic and Non-Pathogenic Leptospira Species | Leptospirosis is a widespread zoonotic infection that primarily affects residents of tropical regions, but causes infections in animals and humans in temperate regions as well. The agents of leptospirosis comprise several members of the genus Leptospira, which also includes non-pathogenic, saprophytic species. Leptospirosis can vary in severity from a mild, non-specific illness to severe disease that includes multi-organ failure and widespread endothelial damage and hemorrhage. To begin to investigate how pathogenic leptospires affect endothelial cells, we compared the responses of two endothelial cell lines to infection by pathogenic versus non-pathogenic leptospires. Microarray analyses suggested that pathogenic L. interrogans and non-pathogenic L. biflexa triggered changes in expression of genes whose products are involved in cellular architecture and interactions with the matrix, but that the changes were in opposite directions, with infection by L. biflexa primarily predicted to increase or maintain cell layer integrity, while L. interrogans lead primarily to changes predicted to disrupt cell layer integrity. Neither bacterial strain caused necrosis or apoptosis of the cells even after prolonged incubation. The pathogenic L. interrogans, however, did result in significant disruption of endothelial cell layers as assessed by microscopy and the ability of the bacteria to cross the cell layers. This disruption of endothelial layer integrity was abrogated by addition of the endothelial protective drug lisinopril at physiologically relevant concentrations. These results suggest that, through adhesion of L. interrogans to endothelial cells, the bacteria may disrupt endothelial barrier function, promoting dissemination of the bacteria and contributing to severe disease manifestations. In addition, supplementing antibiotic therapy with lisinopril or derivatives with endothelial protective activities may decrease the severity of leptospirosis.
| Leptospirosis is a widespread zoonotic infection that primarily affects residents of tropical regions, but is seen occasionally in temperate regions as well. Leptospirosis can vary in severity from a mild, non-specific illness to severe disease that includes multi-organ failure and widespread endothelial damage and hemorrhage. To investigate how pathogenic leptospires affect endothelial cells, we compared the responses of two endothelial cell lines to infection by pathogenic versus non-pathogenic leptospires. Our analyses suggested that pathogenic L. interrogans and non-pathogenic L. biflexa caused changes in expression of genes whose products are involved in cellular architecture and interactions with the matrix, but that the changes were in opposite directions, with infection by L. biflexa primarily maintaining cell layer integrity, while L. interrogans disrupted cell layers. In fact, L. interrogans caused significant disruption of endothelial cell layers, but this damage could be abrogated by the endothelial protective drug lisinopril. Our results suggest that L. interrogans binds to endothelial cells and disrupts endothelial barrier function, which may promote dissemination of the bacteria and contribute to severe disease manifestations. This disruption may be slowed by endothelial-protective drugs to decrease damage in leptospirosis.
| Leptospirosis is a geographically widespread zoonosis that has emerged as a significant public health problem in urban slums, particularly in the tropics. The infection is caused by species of spirochetes belonging to the genus Leptospira. There are more than 200 serovars of Leptospira distributed among both pathogenic and non-pathogenic species [1]. The pathogenicity of different strains can vary considerably depending on the host species and age, and on the infecting serovar [2]. The spirochete's mode of entry is through mucous membranes and cuts or abrasions on the skin [1]. Upon entry, the organisms travel through the bloodstream to multiple sites, and may cause liver and kidney damage, meningitis, and a variety of other inflammatory conditions. If the host survives the acute infection, leptospires can persist in the proximal renal tubules for weeks to months, protected from antibodies and causing little to no inflammation. The bacteria are then shed in the urine, and animal urine contamination of water is the primary source of human exposure.
Although little is known about how Leptospira species establish infection in their hosts, adhesion to the host cell surface and extracellular matrix (ECM) by pathogens is often the first critical step in the initiation of infection. Several groups have investigated the adhesion of Leptospira interrogans to endothelial, fibroblast, kidney epithelial, and monocyte-macrophage cell lines cultured in vitro [3]–[9]. It is likely that pathogenic leptospires can attach to several different types of mammalian receptors to establish the infection. In fact, infectious strains of Leptospira have been shown to adhere to ECM components including collagen type IV, fibronectin and laminin, and also to the plasma protein fibrinogen [4], [10]–[12]. Adhesion to several ECM components is mediated at least in part by the LigA and LigB proteins [11] and a group of additional related proteins that were identified through homology to a laminin binding protein [10], [12].
Several studies have shown that the adhesion of pathogens to mammalian cells will provoke multiple changes in the physiology and/or gene expression of the host. The host-pathogen interactions that define a disease are clearly complex. Microarrays are a powerful tool to explore those host-pathogen interactions by analyzing the transcriptional profiles of host cells or pathogens. Although it has been documented that temperature and osmolarity alter leptospiral gene expression [13], [14], no previously published research has focused on the mammalian cell responses to the bacteria. To understand how human endothelial cells alter gene expression in response to incubation with different strains of Leptospira, human gene arrays were probed with cDNA derived from the RNA purified from infected cells and uninfected controls. In this study, we discuss how global analysis of gene expression allows us to gain insights into host specific responses to infection with pathogenic Leptospira.
The human microvascular endothelial cell line of dermal origin (HMEC-1) [15] was obtained from Dr. Ades (Centers for Disease Control and Prevention, Atlanta, Georgia) and cultured in endothelial basal medium (Clonetics, San Diego, CA) supplemented with 15% heat-inactivated fetal bovine serum (Hyclone, Logan, UT), 1 µg/ml hydrocortisone (Sigma-Aldrich, St. Louis, MO) and 10 ng/ml epidermal growth factor (Sigma-Aldrich). The immortalized human macrovascular endothelial cell line EA.hy926 [16] was kindly provided by Dr. C.-J. Edgell (University of North Carolina, Chapel Hill, NC) and grown in Dulbecco's modified Eagle medium with high glucose supplemented with 10% heat-inactivated fetal bovine serum (Gibco, Grand Island, NY) and HAT Media Supplement (Sigma-Aldrich). Both cell lines were cultured in the medium recommended by the supplier in a humidified atmosphere of 5% CO2 and both cell media were supplemented with 1 U/mL penicillin, 1 µg/mL streptomycin, and 2 mM L-glutamine for routine propagation. Cells to be used for experimental infection with Leptospira strains were cultured without the antibiotics.
The roles of proteoglycans in the endothelial cell response to L. interrogans were tested based on previously published protocols [17]. Briefly, chondroitin sulfate B was shown to bind L. interrogans and to competitively inhibit L. interrogans to mammalian cells, so it was tested for the ability to inhibit the endothelial cell responses to the bacteria described below. In addition, inhibition of proteoglycan synthesis by β-xyloside, which also decreases L. interrogans attachment to mammalian cells, was tested for any effect. Controls included chondroitin sulfate A, to which L. interrogans does not bind, and the sugar analog α-galactoside, which does not affect proteoglycan synthesis.
The reference strain Leptospira biflexa serovar Patoc was obtained from the American Type Culture Collection (ATCC 23582, Manassas, VA), and is a non-pathogenic species. L. interrogans serovar Canicola (pathogenic, strain ATCC 23606 and strain 11203-32) were obtained from the ATCC and Dr. Richard Zuerner (USDA, Ames, IA), respectively. L. interrogans serovar Copenhageni (pathogenic, strain designated Fiocruz L1-130) was provided by Dr. David Haake (UCLA, Los Angeles, CA). Bacterial strains were maintained in ambient air at 30°C. Bacteria utilized for this study were at low passage from the suppliers (≤passage 6) and cultured in EMJH medium [1] supplemented with 100 µg/ml of 5-fluorouracil and 1% rabbit serum (Sigma-Aldrich). For some experiments, the bacteria were radiolabeled by addition of 35S cysteine plus methionine to the medium as described previously [17]. The bacteria were enumerated using a Petroff-Hausser counting chamber and dark field microscopy.
Mammalian cells were plated in T-225 tissue culture flasks (BD Falcon, Bedford, MA) and grown up to 90% or higher confluence. When cells reached desired confluence, the monolayer was washed with PBS and the cells were lifted off the plastic culture flask with 5mM EDTA in PBS. This was done to allow access of the bacteria to endothelial cell surface receptors that are normally involved in attachment to the substratum, i.e. receptors that the bacteria may encounter when penetrating the vasculature. In addition, this approach minimizes degradation of mRNA that occurs during harvesting of adherent cells. After lifting, cells were spun for 10 minutes at 1,000 rpm, resuspended in the cell culture medium without antibiotics, and enumerated using a hemocytometer counting chamber. 2×107 cells per sample were incubated in suspension with either L. biflexa serovar Patoc or L. interrogans serovar Canicola, or without any bacteria, for 1 h and 3 h at room temperature in the cell medium without antibiotics. The MOI (multiplicity of infection) used was 10 bacteria per mammalian cell. After incubation, cells were washed with phosphate buffered saline (PBS) and harvested for RNA isolation. The RNA was purified using RNeasy kit (Qiagen, Valencia, CA) with DNase digestion according to manufacturer's manual. The quality of RNA was checked using a Bioanalyzer (Agilent, Santa Clara, CA).
Human HEEBO (Human Exonic Evidence Based Oligonucleotide) Arrays, consisting of 44,544 70mer probes representing 30,718 known genes, were purchased from Microarrays Inc. (Nashville, TN). 5 to 20 µg of total RNA from uninfected control and infected samples was used to generate cDNA labeled with aminoallyl (aa)-dUTP through a reverse transcription reaction using anchored oligo(dT) primers. The purified aa-dUTP-labeled cDNAs were coupled in 10 µl 0.1 M NaHCO3 with either Cy3 or Cy5 NHS-ester dye. Cy-dye labeled cDNA was purified using a Cyscribe GFX column (Amersham Biosciences, Piscataway, NJ). The two differently labeled cDNAs were mixed and hybridized using Pronto Microarray Hybridization Kit in a hybridization chamber (Corning, Corning, NY), with the same array slide for 38 to 42 hr according to manufacturer's instruction. After a series of washes using the buffers provided in the kit, slides were spun dry and scanned under two laser channels in a Scanarray 4000 scanner (Packard Bioscience, Meriden, CT).
Images were overlaid and analyzed using Imagene (BioDiscovery, El Segundo, CA). Raw gene expression was imported from Imagene to GeneSifter (GeneSifter.Net, VizX Labs, Seattle, WA) for analysis. Data from 3 biological replicate experiments were normalized using Lowess normalization and by the median of the raw intensities for all spots in each sample for each array. The ratio of two fluorescence intensities of each spot reflected the ratio of each gene expressed in the infected and uninfected samples. Genes were considered to be induced or repressed when the ratio of infected/uninfected was at least 1.5 fold (increased or decreased), and the P value was <0.05 by the Student's two-tailed t test. For analysis involving more than one time point and/or condition, the one way ANOVA test was performed. Microarray data are deposited in GEO archive under the accession numbers GSE23172 and GSE23173.
EA.hy926 cells were seeded in tissue culture treated glass slides (BD Falcon) and grown at 37°C as described above. After cells reached 100% confluence, the monolayer was washed three times with PBS and medium without antibiotics was added. Four compartments of each slide were inoculated with 1×107 bacteria (MOI = 10) of either L. biflexa serovar Patoc or L. interrogans serovar Canicola. The remaining four wells were left uninfected to serve as negative controls. In some cases, parallel experiments were performed using cells plated on coverslips in 24 well culture dishes, which allowed centrifugation to facilitate bacterial-endothelial cell contact. At the end of the incubation (1 h, 3 h and 24 h) the slides were washed three times with PBS and fixed with 3% (wt/vol) paraformaldehyde in PBS at room temperature for 30 min. Cells were permeabilized with 0.1% Triton X-100 in PBS, washed three more times with PBS, and blocked overnight at 4°C with HEPES buffered saline (HBS) and 1% bovine serum albumin (BSA). On the next day the slides were washed again with PBS and incubated with fresh blocking solution for 1h at room temperature. After blocking, the layers were probed with either rabbit anti-L. interrogans (a gift from Dr. Richard Zuerner, USDA, AMES, IA) diluted 1∶5000 or anti-L. biflexa antiserum (Biogenesis, Inc., Brentwood, NH) diluted 1∶1000, followed by anti-rabbit IgG-TRITC conjugate (1∶1000) plus phalloidin-FITC (200 U/mL) to stain filamentous actin. After repeated washing in PBS, chambers were removed from the slides and Prolong Anti-Fade (Invitrogen, Carlsbad, CA) was used to mount coverslips. Two different microscopes at two different institutions were used throughout the course of this work. At institution one, images were captured using a Zeiss Axioplan microscope with a digital charge-coupled device camera (Hamamatsu, Hamamatsu City, Japan) and co-localization of the fluorescent labels was done using Volocity software (Improvision Inc., Lexington, MA). At the second institution a Zeiss Axioimager Z1 with an Axiocam HrC camera and a Nuance Multi-Spectral Imaging System (software CRI Inc, Woburn, MA, v.2.6.0) was used.
The endothelial cell lines EA.hy926 and HMEC were plated in 3.0 µm (2×106 pores/cm2) polyester transwell inserts (Corning) and cultured as described above. After reaching 100% confluence, as assessed by lack of penetration of the fluorescent dye FITC-dextran 40,000 (and loss of penetration of the L. biflexa serovar Patoc), the monolayer was washed with PBS and cell medium without antibiotics was added to the inserts and wells. Inserts without cells were used as controls for these experiments. Bacteria were added to an MOI of 50 to allow reliable enumeration of bacteria crossing the cell layers or membranes without cells at early time points, and 10 µL from the insert and from the well were taken after 1 h, 3 h, 6 h, 24 h, 27 h, 48 h and 72 h. In addition to the non-pathogenic strain Patoc and the pathogenic Canicola, Leptospira interrogans serovar Copenhageni was also used to analyze the migration of leptospires through the cell monolayer. Motile leptospires were counted by dark-field microscopy using a Petroff-Hausser chamber. Data are shown for the time points through which the bacteria were motile; after 72 hr there was a progressive decrease in L. biflexa motility.
To determine whether the bacteria were affecting the viability of the endothelial cells, four methods were used. First, adherent and EDTA-lifted endothelial cells infected at an MOI of 10 were washed, then incubated with the vital dye CellTracker Green (CT-CMFDA, 10 µM) plus DAPI (0.02 µg/ml) (Molecular Probes, now part of Invitrogen, Eugene, OR) for 1 hour at 37°C under 5% CO2. The samples were mounted and viewed using the Zeiss Axioplan microscope described above, and live cells (bright green cytoplasm) and dead cells (bright blue nuclei) were enumerated in at least three fields per sample in at least three independent experiments. Second, the cells were stained using the Vybrant Apoptosis Assay Kit 2 (Molecular Probes), which stains for annexin V and membrane permeability. Third, the APO-BrdU TUNEL kit, also from Molecular Probes, was used. A second TUNEL-based kit, Alert DNA Fragmentation kit (Clontech Laboratories, Inc., Mountain View, CA) was also used. For methods two and three, the cells were also assessed using fluorescence microscopy. Finally, cells were harvested, and DNA was purified and analyzed for fragmentation (an assessment of apoptosis) using conventional agarose gel electrophoresis.
We identified statistically significant and reproducible changes in endothelial cell gene expression after incubation with each bacterial strain as compared to the uninfected controls and to each other. The data were analyzed using Webgestalt [18] to identify mammalian cell genes whose products comprise functional pathways in which multiple components showed alterations in gene expression (Table 1). Four pathways that show internally consistent changes in gene expression are the KEGG focal adhesion, regulation of actin cytoskeleton, leukocyte transendothelial migration, and ECM-receptor interaction pathways. They are considered together because a number of genes encode proteins whose functions participate in aspects of cell biology common to these pathways.
Actin microfilaments are one of the three major components of the cellular cytoskeleton. The cytoskeleton participates in maintaining adhesion to and communicating with the extracellular matrix, cell migration, division, and signaling. β-Actin (ACTB) mRNA was decreased in response to L. interrogans but increased in response to L. biflexa, both as compared to the uninfected control cells (Table 2). Guanine nucleotide-binding protein alpha-13 subunit (GNA13) mediates the activation of the small GTPase RhoA [19] which when activated controls the assembly of focal adhesions and actin in the formation of stress fibers [20]. Although RhoA was not differentially regulated in response to the bacteria, Rho GTPase activating protein 5 (RhoGAP5) was differentially expressed following the same pattern as GNA13, in which both genes were downregulated in response to the pathogenic leptospires in comparison to the uninfected controls, and upregulated in response to the non-pathogen. The effect of decreased GNA13 may be to decrease stimulation of Rho, while decreasing the GAP would decrease inactivation of Rho with concomitant decreased cell spreading on the extracellular matrix.
The changes in expression of several additional genes are consistent with changes in cellular architecture as a result of leptospiral infection of these endothelial cells. For example, decreases in the mRNAs for radixin (RDX, a protein that links the actin cytoskeleton to the plasma), caveolins 1 and 2 (CAV1 and CAV2, which couple integrins to the Ras-ERK pathway, titin, the ECM component laminin β1, and integrin subunits αv and β3 (Table 2), were seen in cells infected with L. interrogans Canicola as compared to the uninfected controls. In contrast, the L. biflexa Patoc caused increases in mRNA levels for the same genes in infected cells vs. uninfected controls (Table 2). Together, all of these gene expression patterns are consistent with the hypothesis that one effect of L. interrogans serovar Canicola is to promote actin remodeling and detachment of the cells from the ECM. A fundamental stage in the pathogenesis of Leptospira infections is the ability of the bacteria to cross mucous membranes and underlying epithelial barriers, as well as endothelial cell barriers, and disseminate to different organs. Although Leptospira species are extracellular bacteria apparently devoid of actin modifying exotoxins [21]–[23], and devoid of the specialized secretion systems utilized by many bacterial pathogens to deliver toxins that disrupt the host cell cytoskeleton (as reviewed in [24]–[28]), pathogenic leptospires might be indirectly targeting the cytoskeleton via cell surface attachment mechanisms that co-opt the host cell signaling to achieve the same result.
Decreased cellular adhesion to the ECM and rearrangement of the cytoskeleton may facilitate the migration of Leptospira through endothelial barriers as it disseminates from the site of inoculation. To further explore the possibility that actin rearrangements are triggered by Leptospira infection at the functional level, endothelial cells plated in chamber slides were infected at an MOI of 10 for 1 hour and 3 hours. As shown in Figure 1, the bacteria were clearly more adherent to the cells than to the extracellular space, and the pathogenic bacteria caused dramatically more significant alterations in cellular morphology and integrity of the cell layer than did the non-pathogenic bacteria. The earliest change noted was a reduction in cortical actin (so the cell edges are less defined) and appearance of gaps in confluent cell layers, followed by loss of stress fibers and rounding of the cells. The images shown in Figure 1 are from cell layers that were just below confluence prior to infection, to allow better visualization of changes in individual cells. For example, while the cortical actin has largely disappeared in cells infected with L. interrogans Canicola by 1 hour post-infection, and stress fibers have disappeared and cell rounding is evident by 3 hours, the cells are largely unaffected at the same time points after infection with L. biflexa Patoc (Figure 1). L. biflexa does adhere to mammalian cells in culture less efficiently than does L. interrogans (as shown and reviewed in [17]), but even when bacterial contact with the cells was facilitated by centrifugation, the L. biflexa caused little disruption to cellular morphology and cell layer integrity (data not shown).
Although these and subsequent experiments were performed using adherent cells, the morphologic changes are consistent with changes in mRNA levels seen using lifted cells in the microarray experiments. Despite the alterations in cellular architecture and monolayer integrity, no decrease in endothelial cell viability was found by any of several criteria (see Materials and Methods), even after infection times extended as long as 48 hours (Figure 2). The disruptions in the layers did, however, result in the ability of the pathogenic strain to cross the monolayers more efficiently than did the non-pathogenic bacteria (Figure 3). After a brief period in which the endothelial layer did prevent significant transmigration of the bacteria, the layer rapidly became essentially irrelevant as a barrier to the penetration of the pathogenic bacteria, as the bacterial counts in the lower chamber were unaffected by whether or not cells had been plated on the membrane.
Because Leptospira interrogans has been shown to bind to proteoglycans on the mammalian cell surface [17], we tested a proteoglycan synthesis inhibitor, β-xyloside, for the ability to decrease damage to endothelial cell layers caused by L. interrogans Canicola. β-xyloside inhibits transfer of glycosaminoglycan chains to protein cores; a control sugar analog, α-galactoside, was tested in parallel. As shown in Figure 4, inhibition of proteoglycan synthesis did not fully prevent the damage to the endothelial cell layers caused by L. interrogans. The inhibition of glycosaminoglycan chain attachment does not significantly affect the formation of holes in the cell layer caused by L. interrogans Canicola as assessed visually and by measurement of L. interrogans penetration of the cell layers (data not shown). β-xyloside does cause a reduction of L. interrogans Canicola and Copenhageni attachment to these cells ([17] and data not shown), but does not abolish bacterial attachment, consistent with the hypothesis that additional non-proteoglycan molecules serve as substrates for L. interrogans attachment to cells. Direct bacterial attachment to the cells does appear to be required for the damage to the endothelial cell layers, as supernatants harvested from infected cell layers (infection times of 1–24 hr) and sterilized by centrifugation and filtration through 0.1 µm filters did not affect endothelial cell layer integrity (data not shown). Therefore, non-proteoglycan cell surface receptors are likely to be those primarily involved in the responses of the endothelial cells to L. interrogans attachment, and efforts to identify both the host cell and the bacterial cell molecules involved in these interactions are underway. As noted in the publication reporting the sequence of two L. biflexa Patoc strains [29], there are a number of proteins predicted in the published L. interrogans genomes that are not present in the L. biflexa Patoc genome, including some that are postulated to have potential adhesin activities. These include proteins containing leucine-rich repeats, which are involved in many protein-protein interactions [29]. As stated in the publication of the L. biflexa genome, it is intriguing that a Treponema denticola leucine-rich repeat protein, LrrA, has been identified as an adhesion/tissue penetration factor [29], [30]. It is also possible that additional components of the surfaces of L. interrogans and L. biflexa might have different effects on host cells [31]–[33]. At this point, however, the determinants critical to the effects of L. interrogans-host cell interaction reported here remain to be identified, and neither bacterial adhesins nor host substrates can necessarily be predicted solely on the basis of the primary amino acid sequences.
Several drugs currently in use in humans have been reported to have endothelial barrier protective function; all are in use as anti-hypertensive therapeutics, and some for other therapeutic purposes as well. We therefore tested four different drugs with different mechanisms of action for the ability to prevent the damage to endothelial layers in culture caused by L. interrogans. Lisinopril binds to and competitively inhibits angiotensin 1 binding to angiotensin converting enzyme (ACE), which is expressed by endothelial cells, while telmisartan competitively inhibits angiotensin 2 binding to its receptor AT1. Dopamine is an antagonist of VEGF/VEGFR2-mediated cell layer permeability in treatment of human umbilical vein endothelial cells (HUVECs) in vitro at 10µM, as well as VEGF-mediated angiogenesis in vivo and proliferation of HUVECs at 1 µM in vitro [34], [35]. Furosemide is an anion transport blocker and is used as a diuretic but has anti-hypertensive activity as a consequence, and was used as a control not expected to preserve endothelial layer integrity. While telmisartan, furosemide, and dopamine did not protect the endothelial layers from the damage due to L. interrogans Copenhageni infection, lisinopril did at 100 nM, 1 µM and 10 µM (Figure 5, representing 3 independent experiments, and data not shown). There are several possible explanations for this, including: 1) lisinopril inhibits L. interrogans attachment to the cells, and 2) that attachment is unaffected but the interaction of the bacteria triggers activation of a signaling cascade or release of a mediator whose action or activation is inhibited by lisinopril. We therefore investigated the possibility that lisinopril might prevent endothelial damage by blocking L. interrogans Copenhageni attachment to the cells, but no inhibition of adhesion of 35S-labeled bacteria [17] was seen even at a concentration of lisinopril 10 fold over the concentration used for these experiments (Figure 5).
Although it was tempting to speculate that cell-surface-localized ACE could serve as a receptor for L. interrogans, as the enzyme is expressed by endothelial cells and proximal tubule epithelial cells [36], and is therefore open to possible competition by the lisinopril, this is not consistent with our results to date. However, ACE2 is not inhibitable by lisinopril, but is a receptor for the SARS virus [37], so there is precedent for ACE proteins serving as receptors for pathogens. It is also possible that the effect of lisinopril in our system is not related to ACE inhibition, but is instead due to additional effects of lisinopril, such as inhibition of isoprenoid synthesis, which is required for the post-translational modification of Rho GTPases, which in turn regulate the actin cytoskeleton [38]. In turn, this may lead to increased NO synthesis, which is protective of endothelial function in the face of a variety of insults. Given that doxycycline also has endothelial protective effects [39], and that doxycycline is effective in treating leptospirosis [40], our results may also provide a starting point for investigation into possible combinatorial therapeutic approaches to reduction of endothelial damage and consequent organ damage in human populations during leptospirosis outbreaks. Should this combinatorial approach prove useful in animal models, consideration as a focused approach to the treatment of human leptospirosis is warranted. The 1 µM dose shown in Figure 5 is at the high end of the physiologically relevant dosing range for humans, but administration of an antihypertensive to a patient with clinical manifestations of leptospirosis would be contraindicated, as further depression of blood pressure levels would be potentially lethal. However, in outbreak situations, this agent could potentially help to reduce endothelial damage if administered to affected populations as soon as an outbreak situation is recognized, prior to exposure of the majority of the population to pathogenic Leptospira species. In addition, protective effects of lisinopril were maintained even at a dose of 100 nM, which is well within the range routinely used in humans (Figure 6). It will also be interesting to investigate the possibility that, on a population basis, patients on lisinopril fare better than patients not on this therapy during leptospirosis outbreaks.
Reorganization of the actin cytoskeleton, as indicated by our microarray studies and by phalloidin staining of F actin, is essential to the pathogenesis of diverse bacterial infections, and pathogens use many different strategies to provoke changes in the cellular cytoskeleton in order to facilitate invasion of tissues, invasion of host cells, or evasion of phagocytosis (as reviewed in [24], [41], [42]). A different spirochete, Treponema denticola, produces the protein Msp, which disrupts the actin cytoskeleton in neutrophils and fibroblasts, preventing phagocytosis of the bacterium and inhibiting the cellular migration required to respond to and repair the damage caused by the pathogen and the host response at the site of infection [43], [44]. These activities are likely to facilitate invasion and colonization of periodontal tissues by T. denticola. Previous work by another laboratory demonstrated that L. interrogans Copenhageni crosses MDCK canine kidney epithelial cell layers in culture more rapidly than does L. biflexa Patoc [45], but without significant disruption to the cell layers or the actin cytoskeleton. Consistent with these results, in experiments not shown here we also observed no significant damage to NRK (normal rat kidney) 293 (human kidney) or HEp-2 (human laryngeal) epithelial cell layers infected with L. interrogans Canicola or L. interrogans Copenhageni. The calculations of the proportions of bacteria crossing the cell layers differed between the two studies, but our protocol accounted for the replication of the L. interrogans Canicola and Copenhageni in the co-cultures, while the L. biflexa Patoc did not replicate (data not shown). Thus the endothelial cells tested here respond very differently to the bacteria than did the MDCK epithelial cells, and our results are the first to suggest a mechanism: disruption of actin dynamics by bacterial attachment to the cell surface. Thus, while L. interrogans has not been shown to secrete a toxin that modifies actin, the bacteria are able to manipulate the actin cytoskeleton indirectly. Even the pore forming toxin activity reported for Leptospira [46], [47] does not appear to have as large an effect, as the endothelial cells here were viable throughout the experiments. The leptospires may be able to establish disseminated infection in part due to the binding of the bacteria to one or more mammalian cell surface receptors that in turn, regulate the dynamics of the actin cytoskeleton in the mammalian cell. Deciphering the role of, and mechanisms behind, actin rearrangement in response to pathogenic Leptospira will provide insights into the mechanisms that leptospires uses to disseminate to different organs of the host to cause infection and disease, and provides a possible avenue for therapeutic intervention in conjunction with antimicrobial therapy.
|
10.1371/journal.pcbi.1006960 | Modeling the temporal dynamics of the gut microbial community in adults and infants | Given the highly dynamic and complex nature of the human gut microbial community, the ability to identify and predict time-dependent compositional patterns of microbes is crucial to our understanding of the structure and functions of this ecosystem. One factor that could affect such time-dependent patterns is microbial interactions, wherein community composition at a given time point affects the microbial composition at a later time point. However, the field has not yet settled on the degree of this effect. Specifically, it has been recently suggested that only a minority of taxa depend on the microbial composition in earlier times. To address the issue of identifying and predicting temporal microbial patterns we developed a new model, MTV-LMM (Microbial Temporal Variability Linear Mixed Model), a linear mixed model for the prediction of microbial community temporal dynamics. MTV-LMM can identify time-dependent microbes (i.e., microbes whose abundance can be predicted based on the previous microbial composition) in longitudinal studies, which can then be used to analyze the trajectory of the microbiome over time. We evaluated the performance of MTV-LMM on real and synthetic time series datasets, and found that MTV-LMM outperforms commonly used methods for microbiome time series modeling. Particularly, we demonstrate that the effect of the microbial composition in previous time points on the abundance of taxa at later time points is underestimated by a factor of at least 10 when applying previous approaches. Using MTV-LMM, we demonstrate that a considerable portion of the human gut microbiome, both in infants and adults, has a significant time-dependent component that can be predicted based on microbiome composition in earlier time points. This suggests that microbiome composition at a given time point is a major factor in defining future microbiome composition and that this phenomenon is considerably more common than previously reported for the human gut microbiome.
| The ability to characterize and predict temporal trajectories of the microbial community in the human gut is crucial to our understanding of the structure and functions of this ecosystem. In this study we develop MTV-LMM, a method for modeling time-series microbial community data. Using MTV-LMM we find that in contrast to previous reports, a considerable portion of microbial taxa in both infants and adults display temporal structure that is predictable using the previous composition of the microbial community. In reaching this conclusion we have adopted a number of concepts common in statistical genetics for use with longitudinal microbiome studies. We introduce concepts such as time-explainability and the temporal kinship matrix, which we believe will be of use to other researchers studying microbial dynamics, through the framework of linear mixed models. In particular we find that the association matrix estimated by MTV-LMM reveals known phylogenetic relationships and that the temporal kinship matrix uncovers known temporal structure in infant microbiome and inter-individual differences in adult microbiome. Finally, we demonstrate that MTV-LMM significantly outperforms commonly used methods for temporal modeling of the microbiome, both in terms of its prediction accuracy as well as in its ability to identify time-dependent taxa.
| There is increasing recognition that the human gut microbiome is a contributor to many aspects of human physiology and health including obesity, non-alcoholic fatty liver disease, inflammatory diseases, cancer, metabolic diseases, aging, and neurodegenerative disorders [1–14]. This suggests that the human gut microbiome may play important roles in the diagnosis, treatment, and ultimately prevention of human disease. These applications require an understanding of the temporal variability of the microbiota over the lifespan of an individual particularly since we now recognize that our microbiota is highly dynamic, and that the mechanisms underlying these changes are linked to ecological resilience and host health [15–17].
Due to the lack of data and insufficient methodology, we currently have major gaps in our understanding of fundamental mechanisms related to the temporal behavior of the microbiome. Critically, we currently do not have a clear characterization of how and why our gut microbiome varies in time, and whether these dynamics are consistent across humans. It is also unclear whether we can define ‘stable’ or ‘healthy’ dynamics as opposed to ‘abnormal’ or ‘unhealthy’ dynamics, which could potentially reflect an underlying health condition or an environmental factor affecting the individual, such as antibiotics exposure or diet. Moreover, there is no consensus as to whether the gut microbial community structure varies continuously or jumps between discrete community states, and whether or not these states are shared across individuals [18, 19]. Notably, recent work [20] suggests that the human gut microbiome composition is dominated by environmental factors rather than by host genetics, emphasizing the dynamic nature of this ecosystem.
The need for understanding the temporal dynamics of the microbiome and its interaction with host attributes have led to a rise in longitudinal studies that record the temporal variation of microbial communities in a wide range of environments, including the human gut microbiome. These time series studies are enabling increasingly comprehensive analyses of how the microbiome changes over time, which are in turn beginning to provide insights into fundamental questions about microbiome dynamics [16, 17, 21].
One of the most fundamental questions that still remains unanswered is to what degree the microbial community in the gut is deterministically dependent on its initial composition (e.g., microbial composition at birth). More generally, it is unknown to what degree the microbial composition of the gut at a given time determines the microbial composition at a later time. Additionally, there is only preliminary evidence of the long-term effects of early life events on the gut microbial community composition, and it is currently unclear whether these long-term effects traverse through a predefined set of potential trajectories [21, 22].
To address these questions, it is important to quantify the dependency of the microbial community at a given time on past community composition [23, 24]. This task has been previously studied in theoretical settings. Specifically, the generalized Lotka-Volterra family of models infer changes in community composition through defined species-species or species-resource interaction terms, and are popular for describing internal ecological dynamics. Recently, a few methods that rely on deterministic regularized model fitting using generalized Lotka-Volterra equations have been proposed (e.g., [25–27]). Nonetheless, the importance of pure autoregressive factors (a stochastic process in which future values are a function of the weighted sum of past values) in driving gut microbial dynamics is, as yet, unclear.
Other approaches that utilize the full potential of longitudinal data, can often reveal insights about the autoregressive nature of the microbiome. These include, for example, the sparse vector autoregression (sVAR) model, (Gibbons et al. [24]), which assumes linear dynamics and is built around an autoregressive type of model, ARIMA Poisson (Ridenhour et al. [28]), which assumes log-linear dynamics and suggests modeling the read counts along time using Poisson regression, and TGP-CODA (Aijo et al. 2018 [29]), which uses a Bayesian probabilistic model that combines a multinomial distribution with Gaussian processes.
Particularly, Gibbons et al. [24], uses the sparse vector autoregression (sVAR) model to show evidence that the human gut microbial community has two dynamic regimes: autoregressive and non-autoregressive. The autoregressive regime includes taxa that are affected by the community composition at previous time points, while the non-autoregressive regime includes taxa that their appearance in a specific time is random and or does not depend on the previous time points. In this paper, we show that previous studies substantially underestimate the autoregressive component of the gut microbiome.
In order to quantify the dependency of taxa on past composition of the microbial community, we introduce Microbial community Temporal Variability Linear Mixed Model (MTV-LMM), a ready-to-use scalable framework that can simultaneously identify and predict the dynamics of hundreds of time-dependent taxa across multiple hosts. MTV-LMM is based on a linear mixed model, a heavily used tool in statistical genetics and other areas of genomics [30, 31]. Using MTV-LMM we introduce a novel concept we term ‘time-explainability’, which corresponds to the fraction of temporal variance explained by the microbiome composition at previous time points. Using time-explainability researchers can select the microorganisms whose abundance can be explained by the community composition at previous time points in a rigorous manner.
MTV-LMM has a few notable advantages. First, unlike the sVAR model and the Bayesian approach proposed by Aijo et al. [29], MTV-LMM models all the individual hosts simultaneously, thus leveraging the information across an entire population while adjusting for the host’s effect (e.g,. host’s genetics or environment). This provides MTV-LMM an increased power to detect temporal dependencies, as well as the ability to quantify the consistency of dynamics across individuals. The Poisson regression method suggested by Ridenhour et al. [28] also utilizes the information from all individuals, but does not account for the individual effects, which may result in an inflated autoregressive component. Second, MTV-LMM is computationally efficient, allowing it to model the dynamics of a complex ecosystem like the human gut microbiome by simultaneously evaluating the time-series of hundreds of taxa, across multiple hosts, in a timely manner. Other methods, (e.g., TGP-CODA [29], MDSINE [26] etc.) can model only a small number of taxa. Third, MTV-LMM can serve as a feature selection method, selecting only the taxa affected by the past composition of the microbiome. The ability to identify these time-dependent taxa is crucial when fitting a time series model to study the microbial community temporal dynamics. Finally, we demonstrate that MTV-LMM can serve as a standalone prediction model that outperforms commonly used models by an order of magnitude in predicting the taxa abundance.
We applied MTV-LMM to synthetic data, as suggested by Ajio et al. 2018 [29] as well as to three real longitudinal studies of the gut microbiome (David et al. [17], Caporaso et al. [16], and DIABIMMUNE [21]). These datasets contain longitudinal abundance data using 16S rRNA gene sequencing. Nonetheless, MTV-LMM is agnostic to the sequencing data type (i.e., 16s rRNA or shotgun sequencing).
Using MTV-LMM we find that in contrast to previous reports, a considerable portion of microbial taxa, in both infants and adults, display temporal structure that is predictable using the previous composition of the microbial community. Moreover, we show that, on average, the time-explainability is an order of magnitude larger than previously estimated for these datasets.
We begin with an informal description of the main idea and utility of MTV-LMM. A more comprehensive description can be found in the Methods. MTV-LMM is motivated by our assumption that the temporal changes in the abundance of taxa are a time-homogeneous high-order Markov process. MTV-LMM models the transitions of this Markov process by fitting a sequential linear mixed model (LMM) to predict the relative abundance of taxa at a given time point, given the microbial community composition at previous time points. Intuitively, the linear mixed model correlates the similarity between the microbial community composition across different time points with the similarity of the taxa abundance at the next time points. MTV-LMM is making use of two types of input data: (1) continuous relative abundance of focal taxa j at previous time points and (2) quantile-binned relative abundance of the rest of the microbial community at previous time points. The output of MTV-LMM is prediction of continuous relative abundance, for each taxon, at future time points.
In order to apply linear mixed models, MTV-LMM generates a temporal kinship matrix, which represents the similarity between every pair of samples across time, where a sample is a normalization of taxa abundances at a given time point for a given individual (see Methods). When predicting the abundance of taxa j at time t, the model uses both the global state of the entire microbial community in the last q time points, as well as the abundance of taxa j in the previous p time points. The parameters p and q are determined by the user, or can be determined using a cross-validation approach; a more formal description of their role is provided in the Methods. MTV-LMM has the advantage of increased power due to a low number of parameters coupled with an inherent regularization mechanism, similar in essence to the widely used ridge regularization, which provides a natural interpretation of the model.
We evaluated MTV-LMM by testing its accuracy in predicting the abundance of taxa at a future time point using real time series data. Such evaluation will mitigate overfitting, since the future data points are held out from the algorithm. To measure accuracy on real data, we used the squared Pearson correlation coefficient between estimated and observed relative abundance along time, per taxon. In addition we validated MTV-LMM using synthetic data, illustrating realistic dynamics and abundance distribution, as suggested by Aijo et al. 2018 [29]. Following [29], we evaluate the performance of the model using the ‘estimation-error’, defined to be the Euclidean distance between estimated and observed relative abundance, per time point (see Supplementary Information S1 Note).
We used real time series data from three different datasets, each composed of longitudinal abundance data. These three datasets are David et al. [17](2 adult donors—DA, DB—average 250 time points per individual), Caporaso et al. [16] (2 adult donors—M3, F4—average 231 time points per individual), and the DIABIMMUNE dataset [21] (39 infant donors—average 28 time points per individual). In these datasets, the temporal parameters p and q were estimated using a validation set, and ranged from 0 to 3. See Methods for further details.
We compared the results of MTV-LMM to common approaches that are widely used for temporal microbiome modeling, namely the AR(1) model (see Methods), the sparse vector autoregression model sVAR [24], the ARIMA Poisson regression [28] and TGP-CODA [29]. Overall, MTV-LMM’s prediction accuracy is higher than AR’s (Supplementary Information S1 Table) and significantly outperforms both the sVAR method and the Poisson regression across all datasets, using real time-series data (Fig 1). In addition, since TGP-CODA can not be fully applied to these real datasets (due to scalability limitations), we used synthetic data, considering a scenario of 200 taxa and 70 time points with realistic dynamics and abundance distribution, as suggested by the authors of this method. Similarly to the real data, MTV-LMM significantly outperforms all the compared methods (Supplementary Information S1 Fig).
We applied MTV-LMM to the DIABIMMUNE infant dataset and estimated the species-species association matrix across all individuals, using 1440 taxa that passed a preliminary screening according to temporal presence-absence patterns (see Methods). We found that most of these effects are close to zero, implying a sparse association pattern. Next, we applied a principal component analysis (PCA) to the estimated species-species associations and found a strong phylogenetic structure (PerMANOVA P-value = 0.001) suggesting that closely related species have similar association patterns within the microbial community (Fig 2). These findings are supported by Thompson et al. [32], who suggested that ecological interactions are phylogenetically conserved, where closely related species interact with similar partners. Gomez et al. [33] tested these assumptions on a wide variety of hosts and found that generalized interactions can be evolutionary conserved.
We note that the association matrix estimated by MTV-LMM should be interpreted with caution since the number of possible associations is quadratic in the number of species, and it is, therefore, unfeasible to infer with high accuracy all the associations. However, we can still aggregate information across species or higher taxonomic levels to uncover global patterns of the microbial composition dynamics (e.g., principal component analysis).
In order to address the fundamental question regarding the gut microbiota temporal variation, we quantify its autoregressive component. Namely, we quantify to what degree the abundance of different taxa can be inferred based on the microbial community composition at previous time points. In statistical genetics, the fraction of phenotypic variance explained by genetic factors is called heritability and is typically evaluated under an LMM framework [30]. Intuitively, linear mixed models estimate heritability by measuring the correlation between the genetic similarity and the phenotypic similarity of pairs of individuals. We used MTV-LMM to define an analogous concept that we term time-explainability, which corresponds to the fraction of temporal variance explained by the microbiome composition at previous time points.
In order to highlight the effect of the microbial community, we next estimated the time-explainability of taxa in each dataset, using the parameters q = 1, p = 0. The resulting model corresponds to the formula: taxat = microbiome community(t−1) + individual effect(t−1) + unknown effects. Of the taxa we examined, we identified a large portion of them to have a statistically significant time-explainability component across datasets. Specifically, we found that over 85% of the taxa included in the temporal kinship matrix are significantly explained by the time-explainability component, with estimated time-explainability average levels of 23% in the DIABIMMUNE infant dataset (sd = 15%), 21% in the Caporaso et al. (2011) dataset (sd = 15%) and 14% in the David el al. dataset (sd = 10%) (Fig 3, Supplementary Information S2 Fig). Notably, we found that higher time explanability is associated with higher prediction accuracy (Supplementary Information S3 Fig).
As a secondary analysis, we aggregated the time-explainability by taxonomic order, and found that in some orders (non-autoregressive orders) all taxa are non-autoregressive, while in others (mixed orders) we observed the presence of both autoregressive and non-autoregressive taxa (Fig 4, Supplementary Information S4 Fig), where an autoregressive taxa have a statistically significant time-explainability component.
Particularly, in the DIABIMMUNE infant data set, there are 7244 taxa, divided into 55 different orders. However, the taxa recognized by MTV-LMM as autoregressive (1387 out of 7244) are represented in only 19 orders out of the 55. The remaining 36 orders do not include any autoregressive taxa. Unlike the autoregressive organisms, these non-autoregressive organisms carry a strong phylogenetic structure (t-test p-value < 10−16), that may indicate a niche/habitat filtering. This observation is consistent with the findings of Gibbons et al. [24], who found a strong phylogenetic structure in the non-autoregressive organisms in the adult microbiome.
Notably, across all datasets, there is no significant correlation between the order dominance (number of taxa in the order) and the magnitude of its time-explainability component (median Pearson r = 0.12). For example, in the DIABIMMUNE data set, the proportion of autoregressive taxa within the 19 mixed orders varies between 2% and 75%, where the average is approximately 20%. In the most dominant order, Clostridiales (representing 68% of the taxa), approximately 20% of the taxa are autoregressive and the average time-explainability is 23%. In the second most dominant order, Bacteroidales, approximately 35% of the taxa are autoregressive and the average time-explainability is 31%. In the Bifidobacteriales order, approximately 75% of the taxa are autoregressive, and the average time-explainability is 19% (Fig 4). We hypothesize that the large fraction of autoregressive taxa in the Bifidobacteriales order, specifically in the infants dataset, can be partially attributed to the finding made by [34], according to which some sub-species in this order appear to be specialized in the fermentation of human milk oligosaccharides and thus can be detected in infants but not in adults. This emphasizes the ability of MTV-LMM to identify taxa that have prominent temporal dynamics that are both habitat and host-specific.
As an example of MTV-LMM’s ability to differentiate autoregressive from non-autoregressive taxa within the same order, we examined Burkholderiales, a relatively rare order (less than 2% of the taxa in the data) with 76 taxa overall, where only 19 of which were recognized as autoregressive by MTV-LMM. Indeed, by examining the temporal behavior of each non-autoregressive taxa in this order, we witnessed abrupt changes in abundance over time, where the maximal number of consecutive time points with abundance greater than 0 is very small. On the other hand, in the autoregressive taxa, we witnessed a consistent temporal behavior, where the maximal number of consecutive time points with abundance greater than 0 is well over 10 (Supplementary Information S5 Fig).
The colonization of the human gut begins at birth and is characterized by a succession of microbial consortia [35–38], where the diversity and richness of the microbiome reach adult levels in early childhood. A longitudinal study has recently been used to show that infant gut microbiome begins transitioning towards an adult-like community after weaning [39]. This observation is validated using our infant longitudinal data set (DIABIMMUNE) by applying PCA to the temporal kinship matrix (Fig 5). Our analysis reveals that the first principal component (accounting for 26% of the overall variability) is associated with time. Specifically, there is a clear clustering of the time samples from the first nine months of an infant’s life and the rest of the time samples (months 10 − 36) which may be correlated to weaning. As expected, we find a strong autoregressive component in an infant microbiome, which is highly associated with temporal variation across individuals. By applying PCA to the temporal kinship matrix, we demonstrate that there is high similarity in the microbial community composition of infants at least in the first 9 months. This similarity increases the power of our algorithm and thus helps MTV-LMM to detect autoregressive taxa.
In contrast to the infant microbiome, the adult microbiome is considered relatively stable [16, 40], but with considerable variation in the constituents of the microbial community between individuals. Specifically, it was previously suggested that each individual adult has a unique gut microbial signature [41–43], which is affected, among others factors, by environmental factors [20] and host lifestyle (i.e., antibiotics consumption, high-fat diets [17] etc.). In addition, [17] showed that over the course of one year, differences between individuals were much larger than variation within individuals. This observation was validated in our adult datasets (David et al. and Caporaso et al.) by applying PCA to the temporal kinship matrices. In both David et al. and Caporaso et al., the first principal component, which accounts for 61% and 43% of the overall variation respectively, is associated with the individual’s identity (Fig 6).
Using MTV-LMM we observed that despite the large similarity along time within adult individuals, there is also a non-negligible autoregressive component in the adult microbiome. The fraction of variance explained by time across individuals can range from 6% up to 79% for different taxa. These results shed more light on the temporal behavior of taxa in the adult microbiome, as opposed to that of infants, which are known to be highly affected by time [39].
MTV-LMM uses a linear mixed model (see [44] for a detailed review), a natural extension of standard linear regression, for the prediction of time series data. We describe the technical details of the linear mixed model below.
We assume that the relative abundance levels of focal taxa j at time point t depend on a linear combination of the relative abundance levels of the microbial community at previous time points. We further assume that temporal changes in relative abundance levels, in taxa j, are a time-homogeneous high-order Markov process. We model the transitions of this Markov process using a linear mixed model, where we fit the p previous time points of taxa j as fixed effects and the q previous time points of the rest of the microbial community as random effects. p and q are the temporal parameters of the model.
For simplicity of exposition, we present the generative linear mixed model that motivates the approach taken in MTV-LMM in two steps. In the first step we model the microbial dynamics in one individual host. In the second step we extend our model to N individuals, while accounting for the hosts’ effect.
We first describe the model assuming there is only one individual. Consider a microbial community of m taxa measured at T equally spaced time points. We get as input an m × T matrix M, where Mjt represents the relative-abundance levels of taxa j at time point t. Let yj = (Mj,p+1, …, MjT)t be a (T − p) × 1 vector of taxa j relative abundance, across T − p time points starting at time point p + 1 and ending at time point T. Let Xj be a (T − p) × (p + 1) matrix of p + 1 covariates, comprised of an intercept vector as well as the first p time lags of taxa j (i.e., the relative abundance of taxa j in the p time points prior to the one predicted). Formally, for k = 1 we have X t k j = 1, and for 1 < k ≤ p + 1 we have X t k j = M j , t - k + 1 for t ≥ k. For simplicity of exposition and to minimize the notation complexity, we assume for now that p = 1. Let W be an (T − q) × q ⋅ m normalized relative abundance matrix, representing the first q time lags of the microbial community. For simplicity of exposition we describe the model in the case q = 1, and then Wtj = Mjt (in the more general case, we have Wtj = M⌈j/q⌉,t−(j mod q), where p, q ≤ T − 1).
With these notations, we assume the following linear model:
y j = X j β j + W u j + ϵ j , (1)
where uj and ϵj are independent random variables distributed as uj∼ N ( 0 m , σ u j 2 I m ) and ϵ j ∼ N ( 0 T - 1 , σ ϵ j 2 I T - 1 ). The parameters of the model are βj (fixed effects), σ u j 2, and σ ϵ j 2.
We note that environmental factors known to be correlated with taxa abundance levels (e.g., diet, antibiotic usage [17, 20]) can be added to the model as fixed linear effects (i.e., added to the matrix Xj).
Given the high variability in the relative abundance levels, along with our desire to efficiently capture the effects of multiple taxa in the microbial community on each focal taxa j, we represent the microbial community input data (matrix M) using its quantiles. Intuitively, we would like to capture the information as to whether a taxa is present or absent, or potentially introduce a few levels (i.e., high, medium, and low abundance). To this end, we use the quantiles of each taxa to transform the matrix M into a matrix M ˜, where M ˜ j t ∈ { 0 , 1 , 2 } depending on whether the abundance level is low (below 25% quantile), medium, or high (above 75% quantile). We also tried other normalization strategies, including quantile normalization, which is typically used in gene expression eQTL analysis [45, 46], and the results were qualitatively similar (see Supplementary Information S6 Fig). We subsequently replace the matrix W by a matrix W ˜, which is constructed analogously to W, but using M ˜ instead of M.
Notably, both the fixed effect (the relative abundance of yj at previous time points) and the output of MTV-LMM are the continuous relative abundance. The random effects are quantile-binned relative abundance of the rest of the microbial community at previous time points (matrix W ˜). Thus, our model can now be described as
y j = X j β j + W ˜ u j + ϵ j (2)
So far, we described the model assuming we have time series data from one individual. We next extend the model to the case where time series data is available from multiple individuals. In this case, we assume that the relative abundance levels of m taxa, denoted as the microbial community, have been measured at T time points across N individuals. We assume the input consists of N matrices, M1, …, MN, where matrix Mi corresponds to individual i, and it is of size m × T. Therefore, the outcome vector yj is now an n × 1 vector, composed of N blocks, where n = (T − 1)N, and block i corresponds to the time points of individual i. Formally, y k j = M j , ( k m o d ( T - 1 ) ) ⌈ k / ( T - 1 ) ⌉. Similarly, we define Xj and W ˜ as block matrices, with N different blocks, where corresponds to individual i.
When applied to multiple individuals, Model (2) may overfit to the individual effects (e.g., due to the host genetics and or environment). In other words, since our goal is to model the changes in time, we need to condition these changes in time on the individual effects, that are unwanted confounders for our purposes. We therefore construct a matrix H by randomly permuting the rows of each block matrix i in W ˜, where the permutation is conducted only within the same individual. Formally, we apply permutation πi ∈ ST−1 on the rows of each block matrix i, Mi, corresponding to individual i, where ST−1 is the set of all permutations of (T − 1) elements. In each πi, we are simultaneously permuting the entire microbial community. Hence, matrix H corresponds to the data of each one of the individuals, but with no information about the time (since the data was shuffled across the different time points). With this addition, our final model is given by
y j = X j β j + W ˜ u j + H r + ϵ j , (3)
where u j ∼ N ( 0 m , σ u j 2 I m ) and ϵ j ∼ N ( 0 n , σ ϵ j 2 I n ), and r ∼ N ( 0 m , σ r 2 I m ). It is easy to verify that an equivalent mathematical representation of model 3 can be given by
y j ∼ N ( X j β j , σ A R j 2 K 1 + σ i n d 2 K 2 + σ ϵ j 2 I ) , (4)
where σ A R j 2 = m σ u j 2, K 1 = 1 m W ˜ W ˜ T, σ i n d 2 = m σ r 2, K 2 = 1 m H H T. We will refer to K1 as the temporal kinship matrix, which represents the similarity between every pair of samples across time (i.e., represents the cross-correlation structure of the data).
We note that for the simplicity of exposition, we assumed so far that each sample has the same number of time points T, however in practice the number of samples may vary between the different individuals. It is easy to extend the above model to the case where individual i has Ti time points, however the notations become cumbersome; the implementation of MTV-LMM, however takes into account a variable number of time points across the different individuals.
Once the distribution of yj is specified, one can proceed to estimate the fixed effects βj and the variance of the random effects using maximum likelihood approaches. One common approach for estimating variance components is known as restricted maximum likelihood (REML). We followed the procedure described in the GCTA software package [47], under ‘GREML analysis’, originally developed for genotype data, and re-purposed it for longitudinal microbiome data. GCTA implements the restricted maximum likelihood method via the average information (AI) algorithm.
Specifically, we performed a restricted maximum likelihood analysis using the function “–reml” followed by the option “–mgrm” (reflects multiple variance components) to estimate the variance explained by the microbial community at previous time points. To predict the random effects by the BLUP (best linear unbiased prediction) method we use “–reml-pred-rand”. This option is actually to predict the total temporal effect (called “breeding value” in animal genetics) of each time point attributed by the aggregated effect of the taxa used to estimate the temporal kinship matrix. In both functions, to represent yj (the abundance of taxa j at the next time point), we use the option “–pheno”. For a detailed description see Supplementary Information S3 Note.
We define the term time-explainability, denoted as χ, to be the temporal variance explained by the microbial community in the previous time points. Formally, for taxa j we define
χ j = σ A R j 2 σ A R j 2 + σ i n d 2 + σ ϵ j 2
The time-explainability was estimated with GCTA, using the temporal kinship matrix. In order to measure the accuracy of time-explainability estimation, the average confidence interval width was estimated by computing the confidence interval widths for all autoregressive taxa and averaging the results. Additionally, we adjust the time-explainability P-values for multiple comparisons using the Benjamini-Hochberg method [48].
We now turn to the task of predicting y t j using the taxa abundance in time t − 1 (or more generally in the last few time points). Using our model notation, we are given xj and w ˜, the covariates associated with a newly observed time point t in taxa j, and we would like to predict y t j with the greatest possible accuracy. For a simple linear regression model, the answer is simply taking the covariate vector x and multiplying it by the estimated coefficients β ^ : y ^ t j = x T β ^. This practice yields unbiased estimates. However, when attempting prediction in the linear mixed model case, things are not so simple. One could adopt the same approach, but since the effects of the random components are not directly estimated, the vector of covariates w ˜ will not contribute directly to the predicted value of y t j, and will only affect the variance of the prediction, resulting in an unbiased but inefficient estimate. Instead, one can use the correlation between the realized values of W ˜u, to attempt a better guess at the realization of w ˜ u for the new sample. This is achieved by computing the distribution of the outcome of the new sample conditional on the full dataset, by using the following property of the multivariate normal distribution. Assume we sampled t − 1 time points from taxa j, but the relative abundance level for the next time point t, y t j, is held out from the algorithm. The conditional distribution of y t j given the relative abundance levels at all previous time points, yj, is given by:
y t j | y j ∼ N ( x T β j + Σ t , - t Σ - t , - t - 1 ( y j - X j β j ) , Σ t , - t Σ - t , - t - 1 Σ - t , t ) , (5)
where Σ = W ˜ W ˜ T σ u j 2 + H H T σ r 2 + I σ ϵ j 2 and positive/negative indices indicate the extraction/removal of rows or columns, respectively. Intuitively, we use information from the previous time points that have a high correlation with the new time point, to improve its prediction accuracy. The practice of using the conditional distribution is known as BLUP (Best Linear Unbiased Predictor). Therefore, MTV-LMM could be used to learn taxa effects in a train set (taxa abundance at time points 1, …, t), and subsequently use these learned taxa effects to predict the temporal-community contribution in the next time point in a test set (taxa j at t + 1). We will define the association matrix U (m × m) using BLUP, where uij is the effect of taxa i on taxa j.
The predictive ability of a model is commonly assessed using the prediction error variance, P E V = V a r ( y j - y ^ j ), where y ^ j is the Best Linear Unbiased Predictor of yj. The proportional reduction in relative abundance variance accounted for by the predictions (referred to as R2 in this paper) can be quantified using
R 2 = V a r ( y j ) - V a r ( y ^ j ) V a r ( y j ) = C o v ( y j , y ^ j ) 2 V a r ( y j ) V a r ( y ^ j )
Notably, this definition is equivalent to the squared Pearson correlation.
For every t ∈ {p + 1, ⋯, T}, we calculate y ^ t j, where p ≥ q and the microbial community composition at time t was held out from the algorithm. We next compute R2 between y { p + 1 , ⋯ , T } j and y ^ { p + 1 , ⋯ , T } j.
Given that the model presented in Eq (3) can be extended to any arbitrary p and q, we tested four different variations of this model: 1. p = 0 and q = 1 (no fixed effect, random effects based on 1-time lag), 2. p = 1 and q = 1 (one fixed effect based on 1-time lag, random effects based on 1-time lag), 3. p = 0 and q = 3 (no fixed effect, random effects based on 3-time lags) and 4. p = 1 and q = 3 (one fixed effect based on 1-time lag, random effects based on 3-time lags). We divide each dataset into three parts—training, validation, and test, where each part is approximately 1/3 of the time series (sequentially). We train all four models presented above and use the validation set to select a model for each taxa j based on the highest correlation with the observed relative abundance. We then compute sequential out-of-sample predictions on the test set with the selected model. Based on this metric, we found p = 1 and q = 1 to be the best model for most taxa. We use these parameters when comparing with the other methods such as sVAR and ARIMA-Poisson.
There are three main justifications for the use of multiple time points in the model. First, Gibbons et al. [24] empirically preformed a time-lag analysis and found that for most taxa the autocorrelation disappeared after 3 or 4 days, whereas for some taxa the autocorrelation disappeared after 1 or 2 days. Second, previous studies [26, 27, 49, 50] found that the human microbiome reaches equilibrium within 10 days following small perturbations to the community. It is imperative to model the different taxa in a manner that will fit their temporal patterns. Third, allowing for the use of multiple previous time points increases flexibility so that the model can select the correct time window required for each taxa.
We performed the following phylogenetic analysis. First, in order to test the hypothesis that both autoregressive and non-autoregressive dynamics carry a taxonomic signal, we fitted a linear mixed model, where the kinship matrix is now the phylogenetic distance between pairs of taxa and the outcomes are the time-explainability measurement for each taxa. Second, in order to test the hypothesis that only non-autoregressive dynamics carry a non-random taxonomic signal, we conducted a permutation test by shuffling the taxonomic order assigned to each taxa—generating new random “orders” using 100, 000 iterations. We counted the number of non-autoregressive orders in each iteration, thereby generating a null distribution, which we then used to calculate an exact P-value for the dataset in each iteration.
To measure the alpha diversity, we used Shannon-Wiener index, which is defined as H = −∑pj ln(pj), where pj is the relative abundance of species j. Shannon-Wiener index accounts for both abundance and evenness of the species present. Additionally, we computed the ‘effective number of species’ (also known as true diversity), the number of equally-common species required to give a particular value of an index. The ‘effective number of species’ associated with a specific Shannon-Wiener index a is equal to exp(a).
To calculate the temporal kinship matrix we included taxa using the following criteria. A taxa is present in at least 10% of the time points (removes dominant zero abundance taxa). In the David et al. dataset we included 1051 (out of 2804), in the Caporaso et al. dataset we included 922 (out of 3436) and in the DIABIMMUNE dataset we included 1440 (out of 7244) taxa.
We compared MTV-LMM to two existing methods: sVAR suggested by [24] and Poisson regression suggested by [28]. In the sVAR method, we followed the procedure described in [24], while running the model and computing the prediction for each individual separately, since it can only handle one individual at a time. We then computed an aggregated prediction accuracy score for each taxa, by averaging the prediction accuracy of each individual. In the Poisson regression method, we followed the procedure described in [28], while running the model for all the individuals simultaneously and calculating prediction accuracy for each taxa. We used the taxa that passed the screening suggested in [28] (eliminating any taxa in the data for which there were a small number (< 6) of average reads per sample). In both models, the training set was 0.67 of the data and the test set was the remaining 0.33 of the data. In both cases we used the code supplied by the authors.
We evaluated the performance of MTV-LMM using three real longitudional datasets with 16S rRNA gene sequencing. All data sets are publicly available. The first data set was collected and studied by David et al. (2014) [17] (2 adult donors). The next data set was collected and studied by Caporaso et al. (2011) [16] (2 adult donors). The third data set was collected by the ‘DIABIMMUNE’ project and studied by Yassour et al. (2016) [21] (39 infant donors). In order to compare across studies and reduce technical variance between studies, closed reference OTUs were clustered at 99% identity against the Greengenes database 13_8 [51]. Open reference OTU picking was also run [52], in order to look for non-database OTUs that might contribute substantially to community dynamics. OTU tables were normalized by random sub-sampling to contain 10, 000 reads per sample.
David et al. (2014) dataset [17]. Stool samples from 2 healthy American adults were collected (donor A = DA and donor B = DB). DA collected gut microbiota samples between days 0 and 364 of the study (total 311 samples). DB primarily collected gut microbiota samples between study days 0 and 252 (total 180 samples). The V4 region of the 16S ribosomal RNA gene subunit was used to identify bacteria in a culture-independent manner. DNA was amplified using custom barcoded primers and sequenced with paired-end 100 bp reads on an Illumina GAIIx according to a previously published protocol [53]. ‘OTU picking’ and ‘quality control’ were performed essentially as described [17]. In this work, we used the OTUs shared across donors (2, 804 OTUs).
Caporaso et al. (2011) dataset [16]. Two healthy American adults, one male (M3) and one female (F4), were sampled daily at three body sites (gut (feces), mouth, and skin (left and right palms)). M3 was sampled for 15 months (total 332 samples) and F4 for 6 months (total 131 samples). Variable region 4 (V4) of 16S rRNA genes present in each community sample were amplified by PCR and subjected to multiplex sequencing on an Illumina Genome Analyzer IIx according to a previously published protocol [53]. ‘OTU picking’ and ‘quality control’ were performed essentially as described [16]. In this work, we used the OTUs shared across donors (3, 436 OTUs).
DIABIMMUNE dataset [21]. Monthly stool samples collected from 39 Finnish infants aged 2 to 36 months. To analyze the composition of the microbial communities in this cohort, DNA from stool samples was isolated and amplified and V4 region of the 16S rRNA gene was sequenced. Sequences were sorted into OTUs. 16S rRNA gene sequencing was performed essentially as previously described in [21]. In this work, we used all the OTUs in the sample (7, 244 OTUs).
Code is available in https://github.com/cozygene/MTV-LMM.
We have presented MTV-LMM, a flexible and computationally efficient tool, which can be easily adapted by researchers to select the core time-dependent taxa, quantify their temporal effects and predict their future abundance. Using MTV-LMM we find that in contrast to previous reports, a considerable portion of microbial taxa in both infants and adults display temporal structure that is predictable using the previous composition of the microbial community. In reaching this conclusion we have adopted a number of concepts common in statistical genetics for use with longitudinal microbiome studies. We introduce concepts such as time-explainability and the temporal kinship matrix, which we believe will be of use to other researchers studying longitudinal microbiota dynamics, through the framework of linear mixed models.
Time-explainability can be informative for selecting autoregressive taxa that are essential to understanding the temporal behavior of the microbiome in longitudinal studies. In particular, such taxa can be used to characterize the temporal trajectories of the microbial community. The temporal kinship matrix can be used to uncover low-rank temporal structure. Specifically, as shown in the Results section (Fig 5), applying PCA to the temporal kinship matrix in the DIABIMMUNE infant dataset revealed a clear clustering of the time samples that separate the first nine months of an infant’s life from the rest of the time samples (10-36 months). Further, we have shown that the association matrix estimated by MTV-LMM can be used to uncover global patterns in microbial composition. Using the DIABIMMUNE dataset, we found a strong phylogenetic structure suggesting that closely related species have similar association patterns. Finally, we have demonstrated that MTV-LMM significantly outperforms commonly used methods for temporal modeling of the microbiome, both in terms of its prediction accuracy as well as in its ability to identify time-dependent taxa.
Using MTV-LMM, we have demonstrated that taxa autoregressiveness is a spectrum where certain taxa are almost entirely determined by the community composition at previous time points, some are somewhat dependent on the previous time points, and others are completely independent of previous time points. We further show that MTV-LMM can identify autoregressive taxa in both ‘evolving’ (i.e., infant’s gut) and ‘stable’ (i.e., adult gut) ecosystems. In the former case, i.e., infant gut, the organisms are shifting in abundance over time, which will induce autoregressive dynamics. In this case, where succession is one of the main driving forces, a strong phylogenetic signal is expected. In the latter case, i.e., adult gut, the dynamic is more stationary, with occasional blooms of low-abundance taxa that introduce short-term non-stationary behavior. Notably, the ability of MTV-LMM to identify time-dependent taxa in both scenarios (i.e., ‘evolving’ and ‘stable’) can be utilized to find keystone species that may be responsible for the temporal changes observed in different ecosystems.
It is important to note that MTV-LMM assumes linear dynamics and is built around an AR(p) type of model. However, we recognize that there are also non-linear dynamics in this ecosystem. Nonetheless, it seems that the linear approximation of these dynamics, using the framework of linear mixed models, is capturing a non-negligible signal, which is consistent with other applications of linear mixed models, such as genetics [47] and methylation data [54]. This is demonstrated using both real and simulated longitudinal data where MTV-LMM outperforms methods that directly model these non-linear dynamics. Despite the multiple methodological advancements provided by MTV-LMM, future refinements are possible. These include modeling count uncertainty as well as applying different transformations to the data (e.g., arcsine). This will allow MTV-LMM to model nonlinear correlations and multiplicative errors while accounting for the compositional nature of the data. The instrumental novelty of our method to predict the temporal behavior of taxa is the statistical power that is gained by leveraging the overall community composition as well as all the individuals in the dataset. This suggests that mutual effects of taxa within the microbial community are of major importance in modulating the microbiome’s behavior over time.
|
10.1371/journal.pcbi.1000271 | Efficient Network Reconstruction from Dynamical Cascades Identifies
Small-World Topology of Neuronal Avalanches | Cascading activity is commonly found in complex systems with directed
interactions such as metabolic networks, neuronal networks, or disease spreading
in social networks. Substantial insight into a system's organization
can be obtained by reconstructing the underlying functional network architecture
from the observed activity cascades. Here we focus on Bayesian approaches and
reduce their computational demands by introducing the Iterative Bayesian (IB)
and Posterior Weighted Averaging (PWA) methods. We introduce a special case of
PWA, cast in nonparametric form, which we call the normalized count (NC)
algorithm. NC efficiently reconstructs random and small-world functional network
topologies and architectures from subcritical, critical, and supercritical
cascading dynamics and yields significant improvements over commonly used
correlation methods. With experimental data, NC identified a functional and
structural small-world topology and its corresponding traffic in cortical
networks with neuronal avalanche dynamics.
| In many complex systems found across disciplines, such as biological cells and
organisms, social networks, economic systems, and the Internet, individual
elements interact with each other, thereby forming large networks whose
structure is often not known. In these complex networks, local events can easily
propagate, resulting in diverse spatio-temporal activity cascades, or
avalanches. Examples of such cascading activity are the propagation of diseases
in social networks, cascades of chemical reactions inside a cell, the
propagation of neuronal activity in the brain, and e-mail forwarding on the
Internet. Although the observation of a single cascade provides limited insight
into the organization of a complex network, the observation of many cascades
allows for the reconstruction of very robust features of network organization,
providing valuable insight into network function as well as network failure. The
current work develops new algorithms for an efficient reconstruction of
relatively large networks in the context of cascading activity. When applied to
the brain, these algorithms uncover the structural and functional features of
gray matter networks that display activity cascades in the form of neuronal
avalanches.
| Cascade-like dynamics is characterized by the succession of events, or processes,
that are causally related, and is frequently encountered in many complex systems
(networks) across disciplines. For example, single cells in living organisms
maintain metabolic, protein and gene-interaction networks with mostly unidirectional
signaling cascades in which nodes represent metabolites, proteins and genes
respectively [1]–[3]. At the next higher
level of cell to cell interactions such as the brain, pyramidal neurons in the
cortex connect with thousands of other neurons [4] thereby supporting
cascades of neuronal activity in the form of waves [5], neuronal avalanches
[6] and
synfire chains [7],[8]. Cascade-like dynamics also occurs in many social
networks such as the spread of epidemics [9] and gossip [10] in human
networks as well as human travel itself [11]. This cascading
dynamics carries the signature of the underlying statistical interdependencies
between the interacting nodes, which are summarized by the functional network
topology, represented by adjacency matrix indicating whether
two nodes interact or not, and architecture
[12],
represented by a weighted graph which additionally indicates the magnitude of each
interaction. The relationship between the cascading dynamics and the functional
network is often poorly understood, even though reconstructing the network from the
observed dynamics can provide crucial insights into the causal interactions between
the nodes as well as the overall functioning of a complex system [13]. Of
similar challenge remains the problem of how the functional architecture relates
back to the structural organization of a network, that is to its physical nodes and
physical connections between nodes [14]. While very similar
dynamics can arise from fundamentally different network structures, e.g. for small
neuronal networks with diverse elements [15], for large networks such
as the human cortex the global brain dynamics has been shown to reflect fairly
accurately the underlying structural connectivity, i.e. cortex anatomy [16],[17]. It is
therefore critical to identify new approaches that provide insight into the
functional and structural organization of a network based on the observed dynamics.
Correlations in the dynamics between nodes have been successfully used to identify
functional links in relatively large networks such as obtained from MEG or fMRI
recordings of brain activity (e.g. [18]–[20]). A pure correlation
approach, however, is prone to induce false connectivities. For example, it will
introduce a link between two un-connected nodes, if their activities are driven by
common inputs [21],[22]. More elaborate
approaches such as Granger Causality [23], partial Granger
Causality [24], partial directed coherence (for a review see [25]),
and transfer entropy [26] partially cope with the problem of common input,
however, these methods require extensive data manipulations and data transformations
and have been mainly employed for small networks [27],[28].
Here, we propose a new method that efficiently reconstructs the functional
architecture of a network from the dynamics. In the theoretical part of the
manuscript, we first introduce two different Bayesian approaches to reconstruct the
network topology from the observed cascades: (1) the Iterative Bayesian (IB), and
(2) the Posterior Weighted Averaging (PWA) with equal link priors. We then use PWA
to derive the Normalized Count (NC) approach, a simple and efficient nonparametric
algorithm that requires very little knowledge about the dynamical rules underlying
activity cascades. We show that the NC, which is a hybrid between a Bayesian
approach and a correlation method, performs almost as well as the IB when the exact
probabilistic rules of the dynamics known. Using simulations, we demonstrate the
utility of these algorithms for reconstructing random, small-world and scale-free
network architectures from activity cascades modeled by subcritical, critical, and
supercritical branching processes.
We apply our approach to neuronal avalanches, which are the activity cascades in the
brain. It has been shown [6],[29],[30] that they spontaneously emerge in superficial
layers of cortex, both in vitro (acute slices and slice cultures)
[29]–[32] and in vivo
[33].
They have also been demonstrated recently in the spike activity of dissociated
cortex cultures [34],[35]. The network
architecture that gives rise to neuronal avalanches is currently not known, although
neuronal avalanches have been simulated in networks with scale-free [36],[37], fully connected
[38],
random [39], and nearest-neighbors [37],[40]
topologies. Here we demonstrate a small-world functional topology of neuronal group
formation in neuronal avalanches.
For each pair of nodes, we can determine some scalar measure of connectivity. For
example, these can be node to node correlations, or the FC approach (Equation
19), or the NC approach using Ultimately, we are trying to use these estimates as a measure
of directed influence or causal traffic for each link in the underlying network.
However, these measures will also include a contribution from non-causal
correlations arising when pairs of nodes are active close in time but had a
common ancestor at some prior time during the cascade, or share common inputs
directly. We thus have to determine the statistical significance for each of the
scalar connectivity estimates. The null-model is obtained by randomizing the
recorded activity cascades using constrained pairwise
shuffling. In this randomization procedure, the times of two randomly
selected, active nodes will be switched, such that the node active at time , will be assigned time and vice versa.
This shuffling method is straightforward to implement for continuous time events,
in which case the time interval distribution will be preserved. For binned data,
one will encounter situations where the time bin already has node active, and vice versa, in which case the shuffle is aborted
and a new pair of nodes is sought. Shuffling in this way preserves the average
activity at each node as well as the occupation of time bins with active nodes
and thus the dynamical regime of the underlying branching process (see Results). To obtain the resampled dataset,
the pairwise switching is repeated times, being comparable to the total number of active nodes in the
dataset.
By repeating this procedure, resampled datasets are obtained, each with its corresponding estimate. We use the distribution of the to determine the threshold value for the given significance level . The number of shuffled replicates used to obtain the
connectivity estimate at a significance level , where is the “over-shuffling” factor, usually 5
or 10. We obtain the topology, i.e., the adjacency matrix of the estimated
network at the significance level as(20)and the architecture as(21)Hence the reconstructed network is a weighted, directed graph, , which depends on the prescribed level of confidence, and is
supposed to be a measure of causal traffic in the network. Note that by using
shuffling, we can determine a separate threshold for each link, thus reducing
the bias towards more active nodes and reducing the contribution from
correlations in the absence of interactions. When comparing reconstruction
results using shuffling and individually derived thresholds with results based
on a single common threshold in order to determine the significance of links, we
always used the best possible (oracular) single threshold, since in our
simulations the original network was known. We also investigated in our
simulations if the threshold in the IB approach is indeed optimal and it turns out that
choosing anywhere in the range between 0.1 and 0.9 yields very similar
estimates.
We simulated the branching process dynamics on 4 different network topologies
ranging from a random connectivity with low clustering to a small-world
connectivity with high clustering [44]. For the
Erdös-Rényi (ER) network, nodes were connected randomly with fixed probability resulting in an average node degree and randomly assigned link directionality. In the Watts-Newman
(WN) network [45], each node had outgoing links to its nearest neighbors, after which new links were added randomly
with probability to introduce long-range connections. This algorithm produces a
small-world topology with a high clustering coefficient and an average degree similar to the topology described by Watts and Strogatz [44]. In
our simulations we used . Neither the ER nor the WN topology take into account that
many networks self-organize and expand through growth, e.g. cortical neuronal
networks. We therefore also tested two growth models that achieve a small-world
topology with high clustering coefficients. The Barabasi-Alberts (BA) [46]
model uses a preferential attachment rule in which the probability of attachment
from a new node is proportional to the node degree of the existing nodes. Each
new node establishes new outgoing links starting initially with disconnected or fully connected nodes. The resulting topology
is scale-free in which the degree distribution decays according to a power law
with a slope of −3. Here we use and an all-to-all connectivity for the initial network seed.
The BA model requires a new node to attain some knowledge about the degree
distribution in the network, which might pose a problem for large networks. In
contrast, spatial growth networks [47] do not require
global information about the existing network during development. We used the
Ozik-Hunt-Ott (OHO) network [48], which is initialized with nodes on a circle and all-to-all connectivity. In this
network, a new node, whose location is chosen randomly on the circle, attaches
preferentially to its nearest neighbors with outgoing links, hence its growth rule
is named geographical preferential attachment. The OHO network is not
scale-free, but has a clear small-world property with a high clustering
coefficient that is independent of the number of nodes. Its average node
degree is simply given by for large networks. In our simulations, we used . The initial seed for the OHO network is the network with an all-to-all connectivity. We note that for both
growth models the number of outgoing links was for each node and that both models incorporate a subnetwork
(the initial seed) with maximal clustering that is particularly difficult to
reconstruct in the supercritical dynamical regime.
For each topology, we created specific network architectures by using constant
individual link activation probabilities , or alternatively, by drawing from a uniform distribution, or
truncated Normal distributions (e.g. truncated within the range [0,1] and then
scaled to ).
Different dynamical regimes for each topology were explored on networks with nodes and an average node degree of . The quality of network reconstruction as a function of
reconstruction algorithm, network topology, and network architecture was studied
using nodes and , which approximates the number of electrodes from planar
integrated micro-electrode array recordings for neuronal avalanches and the
corresponding node degree. For the BA and OHO network, the average degree is
discretized since it directly depends on the integer parameter , ( for undirected case). Here we used .
The branching process dynamics was simulated as follows. A source node was selected randomly according to some initiation probability
distribution (see below) and activated. In the next time step, all outgoing
links emanating from will have a chance to activate its neighbors (targets) with the corresponding link activation probability . Each activated target now becomes a source for the next
generation of active nodes, and this is repeated for successive time steps until
no active nodes are found. Heterogeneity in node initiation was simulated by
assigning the node initiation probability from a truncated Gaussian profile, , where is the normalized set of ordered node indices so that all
nodes span the profile from is the heterogeneity parameter. Thus, the probability of
choosing the center node (the most active one) was a factor of times larger than the probability of choosing the two edge
nodes (the least active ones). We used , hence the ratios were ≈1.65, 7.4,
2.7×105 respectively.
We evaluated three different dynamical regimes of the branching process. In the
critical regime, one active node at time on average will lead to exactly one active node in the next
time step and the distribution of avalanche sizes obeys a power law with
a slope of −1.5 [49]. In the ER network, the critical regime is
reached if the average link probability , for and for WN networks, . Conversely, sub- and supercritical regimes of the branching
process were simulated at , respectively. For the BA and OHO networks, a power law
spanning a large range of avalanche sizes was difficult to identify, although
their sub- and supercritical regimes were similar to those in ER and WN
networks. We therefore used for those simulations a value for that yielded the closest fit to a power law size distribution
between the sub- and supercritical regimes (see also Figure 2). A refractory period ensured that
an avalanche ended once, or before, all nodes in the network were activated, a
constraint that assured termination of the process particularly when simulating
supercritical dynamics.
Random node activation independent from the ongoing dynamics, i.e. due to noise
or external inputs, was implemented such that any node on the network could be
activated with probability per time step, expressed as . We used a level of 20% for all simulations with
noise, which translated on average into the random activation of one node every
five time steps, independent from the ongoing dynamics. Note that randomly
activated nodes did not initiate new cascades, otherwise they would increase
reconstruction efficiency since the patterns of activity in the
‘noise-induced’ cascades would also be influenced in the
same manner by the underlying network that we are trying to reconstruct. While
noise was used universally, in some instances we also tested the robustness of
the algorithms to time jitter, implemented such that every active node at time was displaced into time bin with 20% chance.
We applied the NC, FC, IB, and SS algorithms to different instances of the
simulated cascade dynamics on all four network topologies and different
architectures. Because the algorithms were described in detail in the Theory
section, here, we focus on additional, practical issues.
When reconstructing a network using IB, we used a cut-off value for the number of
active nodes considered, , above which the IB iteration is skipped. Those iterations
would take a significant portion of the evaluation time and yield only a slight
gain in the posterior probability. While this diminished somewhat the
performance of the IB particularly in the supercritical regimes, larger values
of would have resulted in impractically long reconstruction
times.
In order to establish significance for various network parameters, we used two
randomization techniques, the Erdös-Rényi randomization (ER)
and the degree sequence preserving randomization (DSPR) [50],[51].
In ER randomization, links were completely randomized in order to obtain an ER
network with an equivalent number of nodes, links, and weight distribution as in
the original network. This randomization destroys any correlations and changes
the node degree distribution. In the DSPR, two directed links were chosen
randomly between four different nodes, and then the target nodes of the two
links were switched preserving the degree distribution. This is repeated many
times, and in our implementation the number of such switches is equal twice the
number of the links in the network (number of links that have not been switched
even once is less than 2%).
Finally, for each of the network reconstructions, the total error, , was expressed as the number of links that differed between
the reconstructed network and the original network relative to the total number of links in ,(22)This error counts both false positives, i.e. an estimated link
does not exist, as well as false negatives, i.e. an existing link was not
identified, and because is usually sparse, the error can far exceed 100% of
the true number of links. The error was averaged over 10 different realizations
for each topology and expressed as mean±standard deviation, if not
stated otherwise. When comparing two networks, neither of which represents the
“gold standard”, we use the following two measures for
comparison. One is, , the percent difference in topology, similar to , but now expressed as the total number of the differences
relative to the number of the links that exist in either of the two networks.
This is a less stringent measure than the , and the maximal error is limited to 100%. The
second is the Pearson correlation coefficient between the link weights among the
common links in the two networks, , or alternatively, among the links that are in either of the
two, .
In order to reduce a potential bias in reconstruction efficiency from arbitrarily
selecting a particular significance level, we chose the best reconstruction
obtained from the significance levels . Using an over-shuffling factor of 10, best reconstructions
for NC and FC were generally obtained at . In our simulations, we can also measure the traffic of causal
activations through any given link by summing all the activations that actually
occurred between its source and target nodes. The resulting traffic for each
link was compared with the reconstructed link weights (see Equation 21) to study
traffic estimates using FC and NC.
Coronal slices from rat dorsolateral cortex (postnatal day 0–2; thick) were attached to a poly-D-lysine coated 8×8
multi-electrode-array (MEA; Multichannelsystems, Germany) and grown at in normal atmosphere in standard culture medium without
antibiotics for 4–6 weeks before recording (for details see [29]–[32]). In short,
spontaneous avalanche activity was recorded outside the incubator in normal
artificial cerebrospinal fluid (aCSF) under stationary conditions (laminar flow
of 1–2 ml/min) for up to 10 hrs. For long-term, pharmacological
experiments a second set of cultures was recorded inside the incubator (for
details on long-term recording conditions see [29]). In short, MEAs
with cultures were placed onto storage trays inside the incubator, which were
gently rocked (≈200 s cycle time). For recording, single cultures grown
on the MEAs for 5–6 weeks were placed into a head stage
(MultiChannelSystems, Inc.), which was affixed to a second tray within the
incubator and which had the exact same motion as the primary storage tray. This
allowed recording from cultures inside the incubator in culture medium under
conditions identical to growth conditions. Bath application of the AMPA
glutamate-receptor antagonist 6,7-dinitro-quinoxaline-2,3(1H,4H)-dione (DNQX, Sigma) was used to reduce synaptic excitability in the
cortical network. DNQX was directly added to the culture chamber. For wash, the medium was replaced with normal pre-conditioned culture
medium. Analysis was based on the following time periods of spontaneous
activity: 2–5 hr before, 15–20 hr during DNQX and
2–5 hr after 19 hr of washing of the drug.
Spontaneous local field potentials (LFP) were low-pass filtered at 50 Hz and
sampled continuously at 1 kHz at each electrode. Negative deflections in the LFP
(nLFP) were detected by crossing a noise threshold of −3 SD followed
by negative peak detection within 20 ms and nLFP peak times and nLFP amplitudes
were extracted. Neuronal avalanches were defined as spatiotemporal clusters of
nLFPs on the MEA. In short, a neuronal avalanche consisted of a consecutive
series of time bins with width that contained at least one nLFP on any of the electrodes.
Each avalanche was preceded and ended by at least one time bin with no activity.
Without loss of generality, the present analysis was done with bin width , estimated individually [30]. ranged between for different sets of cultures. Avalanche size was defined as
(1) the number of active electrodes that constitute an avalanche, i.e. the
number of nLFPs, and (2) as the sum of absolute nLFP amplitudes on active
electrodes. In the former case, size ranged from 1 to 60 (corner electrodes were
missing on the array), whereas in the latter case size ranged from (lowest detection level of an nLFP) up to several thousands of .
During activity cascades, an active node on average can activate less than 1,
exactly 1, or more than 1 node in the next time step in correspondence to the
subcritical, critical, and supercritical dynamical regime of a branching
process. We therefore identified these three dynamical regimes for each of the 4
topologies by calculating the corresponding cascade size distributions on
networks with N = 5000 nodes, and a constant activation probability for all links. For both the WN and ER networks, the critical
probability, , was characterized by a cascade size distribution that
followed a power law with a slope of −1.5 as predicted by theory [49]
(Figure 2; for ER; for WN). Conversely, an exponential distribution characterized
the subcritical regime in which most cascades engaged only few nodes, whereas in
the supercritical regime, a bimodal size distribution revealed that cascades
stayed either relatively small or engaged most of the network. For the BA
network, the distribution of cascades sizes in the subcritical regime followed a
power law with a slope of ≈−3 for sizes <10, suggesting
that cascades in that regime were dominated by the degree distribution (slope
−3). In contrast, the supercritical regime was identified by a bimodal
size distribution. At the transition to the supercritical regime, the BA network
revealed a power law slope close to −1.5 for a small range of
avalanche sizes (10 to 100 at ), which we used to identify the critical dynamics. For the OHO
network, a critical regime was indicated at (mean field prediction was 0.085) at which the cascade size
distribution revealed a corresponding power law with slope of −1.5
(Figure 2), from which
it deviates for large cascade sizes. Thus, given the constraints of a constant , the critical regime in the current simulations represented an
approximation of a true critical dynamics for both the BA and OHO network (Figure 2).
The characteristic size distributions for each dynamical regime suggest a varying
efficiency in reconstructing networks based on the observed activity cascades.
For the subcritical regime, we expect fewer ambiguous situations with multiple
source nodes (Figure 1C) and
thus better accuracy in network reconstruction. These smaller cascades, however,
contain fewer links that can be estimated per unit time, which should slow the
reconstruction progress. The opposite holds for the supercritical regime where
large cascades allow for a larger percentage of links to be estimated per unit
time, while the reconstruction accuracy might decrease due to an increase in
ambiguous situations. Consequently, we expect the critical dynamical regime to
achieve a balance between these opposing tendencies in network reconstruction.
Additionally, in subcritical regime much greater number of initial events will
not propagate at all, in which case a reconstruction step cannot be performed.
Thus, it takes much longer time to collect the same number of STES in the
subcritical regime than it does in critical or supercritical regimes.
We quantified the relationship between the dynamical regime and the
reconstruction efficacy by plotting the total reconstruction error as a function of number of propagation steps, , which is the total number of successive time bins that both
contain at least one active node. This was done for all three regimes and all
four algorithms (Figure 3;
ER topology, , uniform link activation probability for avalanche initiation; see also Figure 4B). For both FC and NC, the
significance of a link was based on 1000 shuffles. For the IB algorithm, the
correct value of was used in the dynamic term (Equation 4).
In our initial evaluation without noise, the IB algorithm was superior in
reconstructing the network in all three dynamical regimes. As predicted from the
cascade size distributions, its reconstruction efficiency was higher in the
critical regime compared to the subcritical regime (Figure 3A, left, open arrows). Importantly,
the IB algorithm further improved in the supercritical regime demonstrating its
robust handling of situations with common inputs, where it achieved a high
efficiency that is possible links were estimated in approximately the same number
of propagation steps in order to reach a reconstruction accuracy of
1%. Similarly, the correlation algorithm FC, while being less
efficient than the IB algorithm, faired better in the critical regime when
compared to the subcritical regime. However, it failed in the supercritical
regime to achieve 1% accuracy even for up to 106
propagation steps demonstrating its sensitivity to correlations due to common
inputs (Figure 3A, left, red
arrow). Importantly, our newly developed NC algorithm clearly overcame
the weakness of the FC algorithm and demonstrated its efficiency in all three
regimes (Figure 3A, left, black
filled arrow). We note that the error reported is calculated with
respect to the number of existing links in the network, i.e. ≈600 links
for out of 3,600 possible links. Hence a reported error of
1% is equivalent to about
1/6 = 0.167% overall error in
deciding whether a link existed or not.
The simple SS algorithm, by avoiding ambiguous situations, performed surprisingly
well for all regimes and was comparable to the performances of the IB and NC
algorithm. However, the SS algorithm was highly sensitive to noise and relied on
the assumption that the observed activations completely arose from the intrinsic
dynamics. In fact, when we repeated our simulations in the presence of
20% noise (Figure 3A,
right), SS failed entirely in all regimes resulting in errors
significantly larger than 100%. Equally important, the IB algorithm
now required 4–5 times more propagation steps to reach an accuracy of
1% in the supercritical regime; a sensitivity to noise that
originated from the iterative development of the priors over time (Figure 3A, right, open arrow).
In the presence of noise, only the NC algorithm robustly reconstructed networks
with similar efficiency in the critical and supercritical regime thereby
performing even better than the IB in the supercritical regime (Figure 3A, right). In
comparison to the standard correlation approach, the NC algorithm provided about
50% improvement in the critical regime and more than a 10-fold
improvement to achieve 3% accuracy in the supercritical regime.
These results demonstrate that NC performed best given (1) its simplicity,
requiring no assumptions about the network connectivity or network dynamics, (2)
its high accuracy for all three regimes, and (3) good reconstruction efficiency
of about 2.7 propagation steps per potential link (total links) for the critical and supercritical regime at
1% reconstruction error.
Correlation methods in network reconstruction commonly utilize a single, global
threshold to identify links, e.g. links are assumed to exist for all pairwise
node correlations that are above a minimal correlation value (e.g. [18],
[20], [52]–[54]). However,
heterogeneous node activation frequencies, as well as other conditions, might
require different significance thresholds for each link. For the networks in
Figure 3, we compared
the efficiency in network reconstruction when establishing link significance
using either shuffling or, alternatively, a fixed, best possible threshold for
both the FC and NC algorithm in the presence of 20% noise. While
shuffling performed slightly worse in the subcritical regime, it significantly
improved reconstruction accuracy in the critical and supercritical regime (Figure 3B). For the FC
algorithm, shuffling was necessary for an accurate estimation in the critical
regime, but it was insufficient in the supercritical regime where the error remained high above 1%, even for large numbers of
propagation steps (Figure 3B, red
arrow). For the NC algorithm, shuffling was required to accurately
reconstruct a network with supercritical dynamics (Figure 3B, black arrow). The results, here
plotted for , were similar for (data not shown). This analysis clearly demonstrates that
correlation based methods benefit from using shuffling estimates for thresholds
in the critical regime. On the other hand, the NC algorithm in combination with
shuffling is required for network reconstructions in the supercritical regime.
The reconstruction results were obtained on a relatively small network with , and a question arises on how well it performs for larger
networks. Since the network model we are trying to reconstruct has binary parameters, it is natural to expect that the number of
needed samples, i.e. propagation steps, for the same reconstruction error should
at least increase proportionally to . Using NC to reconstruct an ER topology from the cascades in
the critical dynamical regime, we demonstrate (Figure 4A) that the number of propagation
steps required for 1% reconstruction accuracy scales approximately
linearly with the total number of potential links in the network, i.e. it scaled
as , making it a potentially useful algorithm for reconstructing
larger networks.
Of particular concern for network reconstruction are situations in which nodes
rarely participate in cascade initiations. For example, initiation sites of
neuronal avalanches differ up to an order of magnitude in avalanche initiation
rate [29],[32]. Such heterogeneity
should make it more difficult to reconstruct the topological neighborhood of
less active nodes. Nevertheless, as shown in Figure 3B, the NC algorithm accurately
reconstructed networks with heterogeneities in node initiation frequency up to a
factor of 268,000∶1 for all three dynamical regimes and with only a
slight increase in computation for critical and supercritical regimes.
Finally, we tested the robustness of the IB, FC and NC algorithms in
reconstructing networks with heterogeneous activation probabilities even though the reconstruction algorithms assume a fixed In addition, we introduced a temporal jitter of 20%
when binning activity cascades as to account for temporal imprecision in cascade
measurements. As before, the noise level was 20% and the node
initiation heterogeneity was set to . Under these conditions, the IB failed (Figure 4C) to reconstruct the networks to
1% accuracy for all dynamical regimes. Similarly, FC was robust in
subcritical and critical regimes, but it failed to reach below a 10%
error in the supercritical regime. In contrast, NC always reached below
1% reconstruction accuracy, and performed the best in all regimes.
The performance of NC can be further improved in supercritical regimes when the
knowledge of the branching parameter, is taken into account, as in (Figure
4C).
The NC algorithm also allowed for a robust and accurate reconstruction of network
topologies that differed from random connectivity. We tested its performance for
4 different topologies and all three dynamical regimes in comparison to the FC
algorithm (Figure 5; and reconstructed with and 1000 shuffles). While the FC algorithm failed for the OHO
topology in the critical regime, the NC algorithm reconstructed all topologies
in the subcritical as well as critical regime (Figure 5B). Significantly, the FC algorithm
failed to reconstruct any of the small-world topologies in the supercritical
regime, while the NC algorithm reconstructed the WN as well as the BA network,
demonstrated here up to an accuracy of 0.1%. Only the OHO network
provided a limit above 1% in the efficacy in network reconstruction
(Figure 5B). This limit
most likely arises because a supercritical dynamics will engage all nodes most
of the time in a highly clustered manner at which pairwise shuffling becomes too
constrained (i.e. shuffling two active nodes between two different time points).
The errors due to reconstruction will most likely be false positives and random
in nature. Hence the overall network parameters (average clustering coefficient,
mean path length, average degree) might or might not be affected significantly
by the errors of this order of magnitude. Accordingly, we plotted the
reconstructed network parameters as a function of propagation steps for the OHO
network in the supercritical regime. As can be seen from Figure 5B, even seemingly high error rates of
10% did not significantly affect the clustering coefficient, while
the average degrees are biased to larger values, indicating that most of the
errors are false positives.
The traffic on a network, i.e. the network flow, is one of the most important
aspects that characterizes network functionality [55]. It was reliably
estimated by NC for all three dynamical regimes and most topologies. We studied
the correlation between the known link activation probabilities and the estimated link weights on an ER network for which link activation probabilities were
drawn either from a uniform distribution or a truncated normal distribution
between [0,1] with (, and 20% noise). In Figure 6A it is shown that for both uniform
and normal distributed activation probabilities, NC did significantly better
than FC in relating the reconstructed weights to the original weights prescribed as , particularly in the supercritical dynamics. Furthermore, when
correlating the estimated with the actual traffic in the network, calculated during the
simulation, we found that NC provided a very good measure of the traffic between
two nodes (slope close to 1; Figure
6B and 6C). In contrast, FC significantly underestimated the traffic
for increasingly higher traffic values (slope ≪1). These results,
obtained on an ER network topology, were also confirmed for small-world
topologies, where NC reliably estimated the traffic on the WN and BA network for
all three dynamical regimes. Only for the supercritical regime on the OHO
network did the NC algorithm estimate the traffic poorly (Figure 6D, black dots). However using further improved the reconstruction in traffic similar to that
of an equivalent ER network (R = 0.72; data not
shown).
Given that the avalanche dynamics can be realized on different topologies (see
Figure 2), we used the
robust performance of the NC algorithm for different dynamical regimes and
widely varying network topologies in order to reconstruct the functional
topology and architecture of real neuronal networks that display neuronal
avalanches recorded with integrated planar micro-electrode arrays (MEA) from
neuronal cortex cultures. Spontaneous activity in these cultures is
characterized by negative deflections in the local field potential (nLFP)
indicative of a local synchronization within a subgroup of neurons near the
electrode (Figure
7A–D; [30]). The organization of nLFPs in the neuronal
network takes on the form of complex spatiotemporal patterns that evolve over
successive time bins (Figure 7E and
7F). These patterns, when interpreted as successive node activations
(see Figure 1B), were used
to reconstruct the functional network topology and network architecture. Under
normal conditions, the dynamics that emerges in this system [29] is
characterized by neuronal avalanches whose sizes obey a power law with a slope
of −1.5 for avalanche sizes measured in terms of integrated nLFP
amplitude or number of nLFPs indicative of a critical state (Figure 7G, [6],[56],[57]).
Importantly, the power law in avalanche sizes correlates with a sequential
activation of local neuronal groups that is analog to a critical branching
process [29]–[32]. In the absence of
any knowledge of the real underlying network organization, we reasoned that the
reconstructed network architecture might be reliable if its features converged
with increasing number of propagation steps in the reconstruction process, e.g.
as shown for the simulated OHO network in Figure 5B. Indeed, the network parameters
such as the clustering coefficient, , and average node degree, , remained largely constant beyond 30,000 propagation steps.
This was in agreement with our simulation results, where NC achieved a smaller
than 1% error estimate for all topologies in the critical regime
within a similar range of propagation steps (Figure 5). Importantly, despite the
relatively small network size of and an average degree of , the clustering coefficient of was significantly higher than what would be expected for
corresponding randomized versions of the network . Similarly, we also plot the excess clustering , a network parameter (not a reconstruction error in ) that measures the clustering coefficient in the network that
is beyond the one of an equivalent randomized version of the network. Results
for indicate that the high clustering coefficient was not simply
due to saturation by adding more and more links into a small network (Figure 8A and 8B). These
networks have nearly a linear relationship between the node degree and its
strength, i.e. the summed weights of all links at a node, , with (Figure
8D) while Figure 8E
shows the node in- and out-degree distributions . The weight distribution of the links revealed an
exponentially decaying tail demonstrating the presence of a few links with large
traffic (Figure 8F).
Given that the relatively high clustering was achieved with a small network
diameter of (Figure 8A and
8B), which was similar to those of the equivalent randomized networks , our findings demonstrate that the neuronal cultures with
neuronal avalanche dynamics establish a small-world topology as previously
reported in abstract form [58],[59]. The functional
network topology of the cortex in vitro cultures (and acute
slices [31]) derived from neuronal avalanches is compared to
the results reported for various neural systems in Table 1. The networks range from full brain
and cortical networks among different anatomical and functional areas of the
brain [16], [44], [60]–[63] to
cortical slices and cultures, as well as the neural network of the nematode
C-elegans [44]. The table also shows the results for 21 cortical
networks binned at (14 were acquired in the course of the previous studies, and
combined with the current set of 7, also re-binned to the same ). The networks and the sources of this data are listed in the
caption. One should note that these networks, with exception of the C-elegans
are not very sparse, in which case the clustering coefficient will depend on the
size of the network, as the table roughly indicates. A better comparison between
these different systems can be achieved by using the excess clustering , found in the range between 0.13 and 0.32, and which shows no
obvious dependence on network size or sparsity.
Functional connectivities are dynamically modulated even on a millisecond time
scale [21],[22]. For example, the
functional connection of a single synapse, i.e. its efficacy to elicit a spike
in a post-synaptic neuron, depends on the depolarization of the post-synaptic
neuron, which itself is linked to the neuron's inputs from within the
network, i.e. level of network activity. This suggests that the functional
small-world topology reconstructed from the dynamical cascades, which captures
the spatiotemporal organization of spiking activity [33], might change with
a change in network activity. On the other hand, local synaptic plasticity
mechanisms such as spike-timing dependent plasticity [64] are expected to
translate successive neuronal activations as reflected in the spontaneous
dynamical cascades into a corresponding increase in synaptic strength thereby
establishing a structural correlate of the observed dynamics. In that case, the
network organization might be expected to be relatively robust to a decrease in
overall activity levels.
By taking advantage of the NC algorithm to reconstruct network architectures in
subcritical and critical regimes, we tested the robustness of the functional
small-world topology to acute changes in network activity. We acutely reduced
the efficacy of excitatory glutamatergic fast synaptic transmission in the
cultured networks by bath application of the AMPA receptor antagonist DNQX
(n = 3 networks). As expected, of DNQX significantly reduced the rate of spontaneous cascades
by . Thus, in order to compensate for the reduced number of
propagation steps per time, networks were reconstructed from ≈20 hr of
activity in the presence of DNQX compared to 2–5 hrs of the control
and wash condition. DNQX also reduced the formation of large avalanches leading
to size distributions more similar to that of a subcritical state, which clearly
deviated from the power law with a slope of −1.5 for the pre and wash
condition (Figure 9A). DNQX
significantly reduced the traffic on the network, which under normal conditions
revealed an exponential distribution (Figures 8 and 9B). Despite these significant reductions in
cascade rate and size as well as link traffic, the small-world topology of the
critical network obtained before and after DNQX, nevertheless, was reliably
reconstructed during DNQX as indicated by the similarity in the clustering
coefficient with increasing number of propagation steps (Figure 9C). On average, , as well as was not different between controls and DNQX . A detailed link-by-link comparison using , between the
“pre”↔“wash” showed an error of and correlations, . Similarly, a comparison between
“pre”↔“DNQX”, and
“DNQX”↔“wash” yielded , respectively. When the comparison were made between the
randomized versions of each network (ER randomization), the results were
virtually the same for all three cases, . These results show that while these networks are far from
identical, their overlap is significantly larger than expected by chance.
In the present study, we developed a method that derives a weighted directed graph
based on the observed cascade dynamics, which successfully overcomes ambiguous
source and target node correlations in all dynamical regimes of a branching point
process. Several methods have been previously employed to cope with the issue of
common inputs when using a correlative approach. For example, using delayed
correlations, Cecci et al. [20] demonstrated power law scaling in human fMRI data
even when links with zero delays indicative of common input were removed. A
three-node motif approach using mutual information allowed to remove potential links
arising from common input resulting in undirected small-world graphs reconstructed
from spontaneous spiking activity in dissociated cultures [52]. Assuming an
Ising-model underlying pairwise node correlations, non-directed functional
connections have been estimated for networks of up to 10 nodes from spontaneous
neuronal activity in vitro
[65],[66] and genetic interactions [67]. Although, the last
approach is able to identify common input situations, it results in non-directed
graphs, in contrast to our approach which also reconstructs directed network
traffic.
The Bayesian approaches described here differ from the so-called Bayesian
networks, or belief networks [68]–[70], which specialize
in the reconstruction of directed, acyclic graphs with a smaller number of
configurations to be explored. In order to reconstruct cyclic graphs,
“loopy” Bayesian network approaches [71] can be used,
however, they are, even in their approximate form, NP-hard [72]. Bayesian networks
are particularly useful in small networks when precise Bayesian inference is
required for each link. In contrast, the IB or PWA approaches in the present
study are meant for the reconstruction of large networks from large datasets.
For that purpose we derived and tested new methods for reconstructing the
functional network topology and traffic from dynamical network cascades. We made
the Bayesian methodology feasible by dividing the observations and the network
into individual target activations with the corresponding active subnetworks
(STES). The essential computational reduction was achieved by using the
assumptions of (a) only the events in the near past (the source nodes) are a
potential cause for an activation event in the cascade and (b) the activation
events of two different target nodes that have common source nodes are
independent. Both assumptions make sense in neuronal networks such as the
cortex, in which events in the near past predominantly influence the present
state of a neuron and where the synaptic transmission of a neuron at different
postsynaptic sites is independent. All these methods rely on the assumption that
the underlying dynamics is stochastic. A fully deterministic dynamics would not
allow to discriminate direct from indirect influences.
To combine individual STES and to obtain the reconstructed network, , we used the IB and PWA approach. They enable one to improve
the reconstruction reliability whenever additional knowledge about the dynamics
(or priors in the case of PWA) becomes available. They are computationally
feasible, since their computational complexity is simply the number of STES, , times the complexity of the individual STES. We will assume
that the needed in an observation for a given reconstruction accuracy
is (as was found for NC, see Figure 4A). Hence, the complexity of the IB
is , where is the average number of over all STES. It will be likely that is a function of in the critical and supercritical regimes, but less so in the
subcritical regime. When , the exponential complexity of IB can be managed to some degree by introducing a cut-off
value, , thus reducing the complexity to , but keeping a large pre-factor . The computational complexity of individual STES in PWA will
in most cases be equal or less than . For NC, the individual STES have complexity , hence, the NC has the same low complexity as FC and other
correlation methods, , but it produces much better estimates of causal traffic and
connectivity, making it a candidate algorithm for the reconstruction of large
networks. Note, that most of the computational demand in NC comes from
shuffling, whose complexity also is . Technical considerations of this algorithm are discussed in
the next paragraph (see also Text S1 for the implementation summary).
The PWA approach can also be extended to include situations when the cascade
propagation speed is highly heterogeneous, i.e. the continuous time approach is
necessary, and/or when the amplitudes of the events need to be considered. This
will require some knowledge, or experimental estimate, on how temporal
differences and event amplitudes will affect the activation probabilities (see
Equation 3). In these cases, the equivalent of the expression in Equation 16 becomes(23)where is the link activation probability for the link connecting the active source node and the target node . This expression is obtained in the limit of . A simple inclusion of the weights can also be obtained by
treating , in which case is not the number of active nodes but the total strength of
the sources . This more general framework, requiring the simulation of
continuous time dynamics and varying amplitudes was beyond the scope of this
manuscript.
Although PWA was derived from Bayesian considerations, strictly speaking it is
not a Bayesian method, particularly not the NC algorithm. When PWA uses uniform
priors, one can argue that it is essentially a maximum likelihood method. The
difference, however, with the maximum likelihood approach is that we use uniform
priors on the links, but not the configurations themselves, which are the
elements of our sample space. Thus, different configurations will get assigned
different prior probabilities. When the prior probabilities for the existence of
any link , are small, or are assigned based on the sparsity of a network, the existence of a link can be established using
a nonparametric measure similar to correlation. Historically, arguments have
been made that, in situations where prior knowledge is not available, a precise
choice of the prior probability is not crucial [73] as long as the
choice is smooth in the region of high likelihood. Thus, a uniform and
sufficiently small probability will lead to essentially the same final estimate
[74].
The general methodology of PWA and IB was derived in our Theory section. We then
tested a particular nonparametric instance of PWA, the NC algorithm, with the
goal of reconstructing large networks from large records of a point process
dynamics. The NC is essentially a weighted correlation measure, with the weight
inversely proportional to the number of potential source nodes. This weighting
is not arbitrary, and if one uses a different weighting factor, e.g. , it does not perform as well as NC (data not shown). If one
assumes small prior probabilities for each link, this result becomes intuitive,
since the posterior probability for the existence of simultaneous links is
negligible, hence each link's probability is inversely proportional to
the number of possibilities, i.e. active source nodes . Importantly, we did not assume that is small, but only that it is equal to the sparsity of the
network and that the dynamics is near the critical point. This indicates that
the validity of the NC algorithm does not rely on the precise choice of . The more elaborate IB approach with fully known dynamics
established a benchmark that was closely met by the NC algorithm. The NC
algorithm returns the link weights that are an approximate measure of the causal
traffic across each link. In this paper we tested, using the simulations of a
branching point process on a network, the case when the activation probabilities
do not depend on the magnitude of the events and the event times are discrete.
More general cases can be addressed using an appropriate activation function in
equation 3, and using a different weighting factor for PWA (see Equation 23).
Shuffling of the original time series is commonly used to establish a priori
statistical distributions for the null-hypothesis. Our results clearly
demonstrate that pairwise shuffling significantly improves the reconstruction
accuracy in the critical and supercritical regime. On the other hand, this
method imposes strong limitations resulting in a conservative model that not
only maintains the average activity rate of each node, which prevents the
introduction of correlations due to rate modulation [22], but also the
exact lifetime and size distribution of cascades, thus ensuring that the
shuffled raster remains in the same dynamical regime. This shuffling method
reaches its limits in the supercritical regime with highly synchronized
cascades, e.g. when almost all nodes become active within 1 time step for most
cascades, in which the constraints of the pairwise shuffling limit its
statistical power. Similarly, pairwise shuffling becomes constrained in the
subcritical regime because of the limited number of nodes participating in
cascades. Alternative methods combined with pairwise shuffling, such as temporal
jittering, using a smaller portion of the raster to determine thresholds, or
limiting total number of shuffles, might improve reconstruction efforts further
in these cases.
The ad hoc use of a global threshold in order to extract a
functional connectivity from correlation matrices is often justified by
providing a range of thresholds for which the obtained results are robust [18],
[20], [52]–[54]. In the present
study, we obtained thresholds for each potential link, which significantly
outperformed the global threshold approach in the critical and supercritical
regimes. The calculation of a probability value using a conservative model, i.e.
maintained firing rate and cascade sizes and durations also naturally allows
these thresholds to be interpreted in terms of significance for individual link
existence. As shown in Figure
8C, topological features were shown to be robust for different
significance thresholds.
Our simulation of the branching process incorporated a refractory period during
which a node remained inactive before being able to participate in a cascade
again. Thus, the simulated dynamics represents a branching process only in the
limit of large number of nodes . Notably, refractory periods for nodes are common in many real
systems, where they arise from energy limitations such as transport capacities
and where they serve several major purposes, such as limiting the rate with
which each node engages in the network dynamics and terminating cascades in the
supercritical regime. In the temporal domain, refractory periods support the
formation of non-recurrent dynamics in an otherwise recurrent network. For
example, in neuronal networks, each neuron after its action potential is not
responsive to the near future neuronal feedback [76], or in epidemics
[9] typically studied in Susceptible-Infected-Removed
models [77], in which infected individuals acquire immunity
against re-infection supporting the view of epidemic spread as an essential
forward cascade with little recurrence. While we have addressed the existence of
different dynamical regimes on different topologies, we have not studied
comprehensively all possible issues that might affect the dynamics of the
network, e.g. network modularity [78]. Despite the
dynamic feed-forward aspects of most cascades, the resulting functional
architecture is not limited to acyclic graphs because potentially recurrent
links between nodes that do not engage in one cascade can be active during other
times.
In the present study, we derived the directed, weighted functional architecture
of superficial cortical layers [29],[31] grown on planar
integrated micro-electrode arrays. We demonstrated that a small-world functional
topology of neuronal avalanches is robust to an acute reduction in network
traffic, suggesting that it potentially arises from a corresponding structural
small-world topology of cortical micro-circuits.
The neuronal avalanche dynamics that arises in these layers in
vitro parallels layer formation in the intact animal [33].
The reconstruction of the architecture was based on neuronal avalanches,
dynamical cascades that form in analogy to a critical branching process [29],[30] for which our
simulations show robust and accurate network reconstruction using the NC
algorithm. The estimated clustering coefficient stabilized as predicted from our
network simulations. Importantly, a similar topology was recovered from acute,
subcritical network dynamics in the presence of DNQX. This suggests that the
subgraph described by a cascade does not depend on the overall state of the
network, but might underlie structural components of the network as formed by
the number and strengths of neuronal connections. A small-world topology
combines short distances between network sites with high clustering that allows
for diverse functionality of subgraphs, as shown recently for sensory activities
in the visual cortex of the cat [79].
Previous studies in dissociated neuronal cultures have quantified dynamical
cascades during spontaneous neuronal activity using a variety of measures such
as conditional probability [80], pairwise delayed-correlation indices [81], and
sequential ordering [82]. Additionally, functional topologies were
derived using correlation methods with global correlation thresholds [83]–[85]. As shown in the
present study, the correlation approach might not adequately address functional
connectivity, particular for dissociated cultures which have been shown to
display supercritical dynamical cascades [82]. Despite these
potential limitations, correlation and mutual information based methods derived
non-directed functional small-world topologies from spontaneous activity in
dissociated cortical cultures [52],[86],
in line with our topological findings for the neuronal avalanche dynamics in
layered cultures. Our study further quantified the network traffic, which was
characterized by an exponential tail distribution similar to what has been found
for the weight distribution in dissociated neuronal cultures [52] and airport traffic networks [55].
These characteristics of the small-world architecture formed by neuronal
avalanches provide important constraints for future simulations of this type of
cortical dynamics.
|
10.1371/journal.pbio.0060231 | Rice XB15, a Protein Phosphatase 2C, Negatively Regulates Cell Death and XA21-Mediated Innate Immunity | Perception of extracellular signals by cell surface receptors is of central importance to eukaryotic development and immunity. Kinases that are associated with the receptors or are part of the receptors themselves modulate signaling through phosphorylation events. The rice (Oryza sativa L.) XA21 receptor kinase is a key recognition and signaling determinant in the innate immune response. A yeast two-hybrid screen using the intracellular portion of XA21, including the juxtamembrane (JM) and kinase domain as bait, identified a protein phosphatase 2C (PP2C), called XA21 binding protein 15 (XB15). The interaction of XA21 and XB15 was confirmed in vitro and in vivo by glutathione-S-transferase (GST) pull-down and co-immunoprecipitation assays, respectively. XB15 fusion proteins purified from Escherichia coli and from transgenic rice carry PP2C activity. Autophosphorylated XA21 can be dephosphorylated by XB15 in a temporal- and dosage-dependent manner. A serine residue in the XA21 JM domain is required for XB15 binding. Xb15 mutants display a severe cell death phenotype, induction of pathogenesis-related genes, and enhanced XA21-mediated resistance. Overexpression of Xb15 in an XA21 rice line compromises resistance to the bacterial pathogen Xanthomonas oryzae pv. oryzae. These results demonstrate that Xb15 encodes a PP2C that negatively regulates the XA21-mediated innate immune response.
| Resistance to pathogens is critical to plant and animal survival. Plants, unlike animals, lack an adaptive immune system and instead rely on the innate immune response to protect against infection. To elucidate the molecular mechanism of plant innate immunity, we are studying the signaling cascade mediated by the rice pathogen recognition receptor kinase XA21, which confers resistance to the bacterial pathogen Xanthomonas oryzae pv. oryzae. We demonstrate that XA21 binding protein 15 (a protein phosphatase 2C) negatively regulates XA21-mediated signaling resistance. This finding provides significant insight into regulation of receptor kinase-mediated immunity.
| Protein kinases regulate most cellular signal transduction pathways including cell growth and proliferation, cellular differentiation, morphogenesis, gene transcription, and immunity [1–3]. Adaptive immunity, restricted to vertebrates, is characterized by the creation of antigen-specific receptors through somatic recombination in maturing lymphocytes [4]. In contrast, innate immunity, common to both animals and plants, is mediated by a set of defined receptors referred to as pathogen recognition receptors (PRRs). Recognition of pathogen-associated molecular patterns (also called microbe-associated molecular patterns) or pathogen-derived avirulence (Avr) molecules by PRRs triggers signal transduction pathways mediated by activation of mitogen-associated protein kinase (MAPK) cascades and transcription factors [4,5]. These pathways lead to a core set of defense responses including accumulation of defense related molecules, increases in reactive oxygen species, calcium fluxes, and programmed cell death (PCD) [4–6].
In animals, recognition of pathogen-associated molecular patterns in extracellular compartments or at the cell surface is largely carried out by the Toll-like receptor (TLR) family containing leucine rich repeats (LRRs) in the extracellular domain [6]. TLRs associate with the interleukin-1 receptor-associated kinase (IRAK) family [7] and with receptor interacting-protein (RIP) kinases [8] via adaptor proteins. In plants, cell surface recognition of pathogen-associated molecular patterns or pathogen-derived Avr molecules is largely carried out by the non-RD class of receptor kinases (RKs) [6,9].
These “non-RD” kinases typically carry a cysteine (C) or glycine (G) before the conserved aspartate (D) residue. In contrast, the larger group of “RD” kinases have an arginine (R) immediately preceding the conserved catalytic aspartate (D) [10,11]. The RD class of kinases includes nearly all receptor tyrosine kinases and most characterized plant receptor serine/threonine kinases [10]. The non-RD class includes members of human IRAKs and RIPs, Drosophila Pelle, and members of plant RKs belonging to the IRAK family [10,12,13].
Plant genome analyses have revealed the presence of a large family of these non-RD IRAK kinases, with more than 45 encoded in the Arabidopsis genome and more than 370 found in the rice genome [10,14]. Members include the Arabidopsis flagellin RK (FLS2), the Arabidopsis elongation factor Tu RK (EFR) [15,16], the rice XA26 and Pi-d2 RKs [17–19], and the rice XA21 RK that mediates recognition of the Gram negative bacteria Xanthomonas oryzae pv. oryzae (Xoo) [18,20]. Because the presence of the non-RD motif in IRAK kinases is correlated with a role of the protein in pathogen recognition, there is great interest in understanding how they are regulated [10,21]. Unlike the majority of RD kinases, where phosphorylation of the activation loop is critical for activation [22], the mechanism of non-RD kinase regulation, in which many non-RD kinases do not autophosphorylate the activation loop, remains to be elucidated [10]. The juxtamembrane (JM) domain, a region of the RK that is N terminal to the kinase domain, has been suggested to be important for regulation of non-RD RKs and to serve as a high affinity binding site for downstream signaling proteins [10,23].
Although non-RD IRAK kinases are clearly essential for innate immunity in both plants and animals, sustained or highly induced immune response can be harmful [24]. It is therefore necessary that PRR signaling through non-RD kinases be under tight negative regulation. For example, misregulation of the non-RD IRAK1 in mice induces activation of nuclear factor κB (NF-κB) and increases inflammatory responses to bacterial infection [25,26].
Dephosphorylation of kinases by protein phosphatases (PPs) is a common mechanism for downregulating kinase-mediated signaling [27]. PPs are classified into two major classes: tyrosine phosphatases and serine/threonine phosphatases, depending on their substrates [27,28]. In humans, a group of MAPK phosphatases (MKPs) including MKP1, MKP5, and dual specificity PPs, negatively regulate the innate immune response [24,27,29].
In contrast to animal systems, negative regulation of kinases involved in plant innate immunity is not well understood. One important class of negative regulators are PP 2Cs (PP2Cs), a group of serine/threonine phosphatases that function as monomers and require Mn2+ and/or Mg2+ for activity [28,30]. One of the best characterized plant PP2Cs is the Arabidopsis kinase-associated PP (KAPP), which interacts with many RKs including FLS2, CLAVATA1 (CLV1), somatic embryogenesis RK 1, brassinosteroid-insensitive 1 (BRI1), and BRI1-associated RK 1 [31–36]. Overexpression of KAPP in Arabidopsis results in loss of sensitivity to flagellin treatment, suggesting that KAPP negatively regulates the FLS2-mediated defense response [33]. Arabidopsis CLV1 controls stem cell identity in shoot and flower meristems. In vitro, a CLV1 fusion protein can phosphorylate KAPP. Conversely, KAPP dephosphorylates the kinase domain of CLV1 in vitro [3,31,37,38]. So far, KAPP is the only PP known to be involved in regulation of RK-mediated signaling and to be associated with RKs in plants [31,33].
The rice XA21 RK is one of a few plant PRRs that has been studied in depth [18]. Despite the clear biological role for XA21 in the rice innate immune response, very little is known about the mechanism by which XA21-mediated resistance is regulated. Although the rice KAPP protein emerged as a good candidate for being a negative regulator of the XA21-mediated innate immune response, it does not interact with XA21 [39]. This suggests the presence of another protein that negatively regulates XA21-mediated signaling pathway.
In this study, we report the identification and characterization of rice XA21 binding protein 15 (XB15), which encodes a novel PP2C. Transgenic rice lines overexpressing Xb15 display compromised Xa21-mediated resistance to Xoo strains carrying AvrXa21 activity. Conversely, Xb15 Tos17 insertion mutants display cell death, constitutive induction of PR genes, and enhanced XA21-mediated resistance upon Xoo infection. Subsequent biochemical experiments show that XB15 can dephosphorylate autophosphorylated XA21 in a temporal- and dosage-dependent manner and that a serine residue in the XA21 JM domain is required for XB15 binding. Our findings are consistent with a model in which XB15 associates with the XA21 JM domain to negatively regulate the XA21-mediated innate immune response and cell death.
The rice XA21 protein is representative of the large class of non-RD RKs predicted to be involved in plant innate immunity [10,18,20]. To elucidate the mechanism of XA21-mediated resistance and to identify its interaction partners, we performed a GAL4-based yeast two-hybrid screen using a rice cDNA library constructed from Oryza sativa spp. Indica line, IRBB21. IRBB21 is an isogenic line of the Indica variety IR24 carrying an introgression at the Xa21 locus [40]. It has been previously reported that both the JM region (the portion of the cytoplasmic domain between the transmembrane sequence and the kinase domain) and the C-terminal region of RKs, can serve as high affinity binding sites for downstream signaling proteins [23]. Therefore, the entire XA21 predicted intracellular region, including the JM, kinase, and C-terminal domains, called XA21K668 (668–1,025 amino acids), was used as bait for this screen. Of the 7 × 107 transformants screened, a total of eight unique clones both grew on selective media lacking histidine and tested positive for β-galactosidase reporter gene activity. These clones were named XA21 binding proteins (XBs). We have previously reported characterization of another Xb, called Xb10, which encodes a WRKY transcription factor [41], and Xb3, which encodes an ubiquitin ligase [42]. No RKs were isolated in this screen.
Here, we report the functional characterization of one of the isolated interacting proteins, XB15, the only PP among the isolated XBs. To assess whether the XA21 JM domain was important for the physical interaction of XB15 with XA21, we generated a new construct, XA21K(TDG) (residues 705–1,025), which lacks the JM domain. XA21K(TDG) was then used as bait in a yeast two-hybrid assay with XB15 (Figure 1A). Activation of the lacZ reporter gene was dependent on the simultaneous presence of pAD-XB15 and pLexA-XA21K668. XA21K(TDG) lacking the JM domain did not interact with XB15. This result suggests that the JM domain contains specific amino acids needed for XB15 binding in yeast and that these residues may serve as a docking site for XB15. The JM region (XA21JM) lacking the kinase domain was not sufficient for the interaction with XB15 (Figure 1A). XB15 also failed to interact with the catalytically inactive mutant XA21K668K736E, in which lysine 736 is substituted with glutamic acid [43], suggesting that XA21 kinase activity is also important for the interaction.
On the basis of these results, we hypothesized that a residue(s) in the JM domain phosphorylated by XA21 may serve as a docking site for XB15. It has previously been reported that XA21 autophosphorylates residues Ser686, Thr688, and Ser689 in vitro [44]. To test if these residues are important for XB15 binding, we assessed whether mutations in these sites affected binding activity. We found that all three XA21 single mutants, XA21S688A, XA21T688A, XA21S689A, as well as the triple mutant XA21S688A,T688A,S689A maintained interaction with XB15. We next assessed the role of three previously uncharacterized Ser and Thr residues (Ser697, Ser699, and Thr705) in the JM region of XA21. When Ser697 is mutated to Alanine, interaction with XB15 is abolished (Figure 1A). This result indicates that Ser697 in the XA21 JM domain is critical for interaction with XB15.
To confirm that the absence of interaction was not due to lack of expression of the fusion constructs in yeast cells, we performed protein gel blot analysis with anti-XB15 or anti-LexA antibodies. All yeast cells transformed with AD-Xb15 carried an 80-kDa band cross-reacting with the anti-XB15 antibody (unpublished data). In Western blot analysis using an Anti-LexA antibody, yeast cells expressing LexA-XA21K668 or -XA21 variants, displayed bands with molecular weights corresponding to the correct size of each fusion protein (Figure 1B). Yeast cells expressing the vector control produced a protein of 24 kDa that reacted with the anti-LexA antibody.
Xb15 carries a 3,219-bp open reading frame that consists of three introns with lengths of 94 bp, 1,070 bp, and 138 bp and four exons. The ORF is predicted to encode a 639 amino acid protein (Figure 2A) with a molecular mass of 69.2 kDa and an isoelectric point of 5.4. XB15 is similar in overall structure to known plant PP2Cs, with a conserved C terminus (240–630 amino acids), which is predicted to have phosphatase catalytic activity (Figure 2A, underlines). It has a unique N terminus (1–239 amino acids) with no similarity to proteins with known function. The catalytic domain is interrupted by approximately 100 amino acids with no similarity to any sequences in the database (Figure 2A, with italics). XB15 contains six amino acids residues (four aspartic acids, one glutamic acid, and one glycine) known to interact with two Mg2+ or Mn2+ ions to form a binuclear metal center in Arabidopsis POL and human PP2Cα [45–47].
We next compared the PP2C domain of XB15 with other evolutionary related proteins from the reference plant species rice and Arabidopsis, such as Arabidopsis POL and POL-like proteins (PLLs) (Figure 2B). XB15 shows significant similarity to PP2Cs from other organisms including Arabidopsis POL (47% identity) and shares several features with the Arabidopsis POL and PLLs such as unique intron/exon boundaries and an insertion in the same region of the PP2C catalytic domain. Figure 2B shows the phylogenetic relationship among POL and PLL genes from rice and Arabidopsis based on the amino acid sequence of the last three exons. Arabidopsis POL and PLL1 cluster in one branch of the tree, consistent with the similar primary phenotype of pol and pll1 mutants—phenotypic suppression of clv mutants [48,49]. These two proteins have also been shown to regulate the balance between stem-cell maintenance and differentiation and are closely related to Wuschel, which encodes a homeodomain transcription factor expressed in shoot meristems [50]. XB15 groups with Arabidopsis PLL2, PLL3, PLL4, and PLL5, showing the greatest similarity to PLL4 and PLL5 with 57.5% and 56.3% identity, respectively. pll4 and pll5 mutants develop abnormal leaves that are altered in shape [48]. To date, no function has been reported for Arabidopsis PLL2 or PLL3, or any of the rice PLLs including XB15, leaving the functional role of these proteins unknown.
To explore the role of Xb15 in resistance, we examined whether constitutive overexpression of Xb15 would affect resistance to Xoo. We first generated an Xb15 tagged N-terminal tandem affinity purification (NTAP) construct under the control of the ubiquitin promoter (Ubi) [51,52] and then introduced this construct into the rice cultivar Kitaake using our established transformation procedures [53,54]. The resulting transgenic plants (NTAP-XB15, pollen donor) were crossed with another Kitaake line possessing an N-terminal Myc-epitope-tagged XA21, under control of its native promoter (Myc-XA21, pollen recipient). The presence of the Myc-Xa21 and/or Ntap-Xb15 gene in the resulting F1 was confirmed by the PCR analysis (unpublished data). Next, we tested the expression of NTAP-XB15 and Myc-XA21 in the cross (XA21/XB15 17A-21 and 19A-72) using an anti-Myc antibody and peroxidase-anti-peroxidase (PAP, for detecting the TAP tag) (Figure 3A). Bands corresponding to the predicted molecular mass of Myc-XA21 and NTAP-XB15 were detected at approximately 95 and 140 kDa, respectively confirming that the crossing was successful and that the progeny express both NTAP-XB15 and Myc-XA21. Although equivalent amounts of protein were analyzed, no signals were observed in nontransgenic Kitaake plants.
Six-week-old progeny from self-pollinated XA21/XB15 17A-18, 17A-21, 19A-64, and 19A-72 were then inoculated with Xoo PR6 strain PXO99 expressing AvrXA21 and examined for cosegregation of genotype with phenotype by PCR analysis and measurement of the length of Xoo-induced lesions. All Myc-Xa21 plants overexpressing Ntap-Xb15 (Myc-Xa21/Ntap-Xb15, +/+) by PCR analysis showed accumulation of NTAP-XB15 by protein gel blot analysis and displayed enhanced susceptibility to Xoo PR6 compared to transgenic plants carrying Myc-Xa21 alone (+/−) (Figure 3B). Western blot analysis revealed a higher accumulation of NTAP-XB15 protein in XA21/XB15 19A line compared to XA21/XB15 17A, which correlated with longer leaf lesions developed by Xoo PR6, indicating that XB15 negatively regulates the XA21-mediated defense pathway in a dosage-dependent manner.
We could barely distinguish a difference in XA21 accumulation with homozygous or heterozygous Myc-XA21 transgenic plants (Figure S1). Therefore, to rule out the possibility that this phenotypic difference is caused by a Xa21 dosage effect, we selected segregants (F2) carrying Myc-Xa21 and analyzed the next generation (F3) to determine which segregants were heterozygous or homozygous for Myc-Xa21. After PCR genotyping 20 progeny from each line, we identified 17A-18–1, 17A-18–2, and 19A-72–5 as homozygous for Myc-Xa21 (unpublished data). The remaining segregants are heterozygous for Myc-Xa21. Inoculation analysis revealed that there is no correlation between the number of copies of Myc-Xa21 and the enhanced susceptibility phenotype in the NTAP-XB15 overexpression lines (Figure 3B). These results indicate that the enhanced susceptibility observed in the crossing population (F2) is caused by NTAP-XB15 overexpression and is not due to the presence of fewer copies of Myc-Xa21.
Figure 3C shows a picture of two typical leaves from each of the following inoculated rice plants: F2 segregants of XA21/XB15 17A and 19A, and transgenic plants overexpressing NTAP only (NTAP) 16 d after Xoo PR6 inoculation. While the Myc-XA21 plants lacking Ntap-Xb15 (+/−, XA21/XB15 17A-18–2 and 17A-21–10) were highly resistant, showing short lesions (approximately 2 cm), inoculated leaves of crossed lines (+/+), XA21/XB15 17A-21–7 and 19A-72–5, developed typical water-soaked, long lesions (approximately 8–10 cm). In inoculated NTAP-XB15 lines lacking Xa21 (−/+, XA21/XB15 19A-72–3 and 17A-21–8), there was no significant difference from segregants not carrying Myc-Xa21 and Ntap-Xb15 (−/−) in lesion lengths. These results demonstrate that overexpression of XB15 reduces XA21-mediated resistance to Xoo.
We quantified the effect of XB15 overexpression on bacterial growth by monitoring bacterial populations on XA21/XB15 17A and 19A plants (F2) inoculated with Xoo PR6 (Figure 4A). For both growth curve and lesion length analysis up until 4 d after inoculation (DAI), there was no significant difference in bacterial growth in any of the lines. However, at 12 DAI, Xoo PR6 populations in the Kitaake and NTAP lines reached approximately 1.5 × 109 colony-forming units per leaf (cfu/leaf): whereas, population in Myc-XA21 plants leveled off at fewer than 8.7 × 107 cfu/leaf. In XA21/XB15 19A carrying Myc-Xa21 and Ntap-Xb15 (+/+, 19A-72–5, −6, −8, and −11), Xoo PR6 populations grew to 8.5 × 108 cfu/leaf, a 10-fold increase compared to segregants carrying Myc-Xa21 alone (+/−, 19A-72–1, −10 and 17A-21–6, −10). A significant difference in bacterial growth between the lines was observed continuously up to 16 DAI. We also measured the length of Xoo-induced lesions on all rice lines (Figure 4B). The progeny of XA21/XB15 19A and 17A (+/+, 19A-72–14 and 17A-21–15) displayed enhanced susceptibility to Xoo PR6 with lesions ranging in length from 4–10 cm compared to the segregants carrying Myc-Xa21 alone (+/−, 17A-21–10), which showed 2–3 cm lesion lengths. Segregants overexpressing Ntap-Xb15 but lacking Myc-Xa21 (−/+) did not show significant difference in bacterial populations and lesion lengths, as compared to Kit and NTAP control.
To further investigate the in vivo function of Xb15, a knockout mutant line was identified from a collection of rice mutants generated by random insertion of the endogenous retrotransposon Tos17 (http://tos.nias.affrc.go.jp/∼miyao/pub/tos17/) [55]. One mutant line (NF9014) carried a Tos17 insertion in the third exon of the Xb15 gene (Figure 5A). This insertion resulted in a lack of expression of the full-length Xb15 mRNA compared to the wild type Nipponbare plants when tested with appropriate primers by reverse transcription-PCR (RT-PCR) (Figure 5B). 18S ribosomal RNA (18S rRNA) was used as an internal control. The homozygous insertion mutant line (−/−) showed a severe cell death phenotype marked by the appearance of small necrotic lesions that were apparent at the vegetative stage (Figure 5C). Of the 24 segregants of 6-16-30, six out of eight of the heterozygotes (+/−) displayed the cell death phenotype. Two of the heterozygotes displayed a severe cell death phenotype similar to that observed in the homozygous individuals (−/−). In contrast, all nine homozygous (−/−) plants displayed a cell death phenotype and seven of these were very severe (Figure S2).
We next tested if the cell death phenotype correlated with alterations in PR gene expression. Total RNA was extracted from the mutant line and the Nipponbare control. RT-PCR was performed with primers targeting PR genes as molecular markers. Figure 5D shows that the expression of full-length Xb15 was not detected in the Tos17 mutant line. In contrast, all the defense-related genes tested, PR1a, PR1b, PR10, Betula verrucosa 1 (Betv1, major birch allergen), and probenazole-inducible protein 1 (PBZ1) were highly expressed in the insertion mutant lines (−/−) but barely expressed in Nipponbare wild type (WT) or the no-insertion null plants (+/+). Both the internal controls (18S rRNA and elongation factor 1 α [EF1α]) showed constitutive expressions in all tested plants.
To confirm that the cell death phenotype of the Tos17 insertion mutant is due to the loss of function of XB15, we constructed a plasmid for the gene-specific knockdown of Xb15 by RNA interference (RNAi) with an inverted repeat of a specific fragment of Xb15 cDNA under control of the Ubi promoter. The protein level of XB15 in Xb15 RNAi transgenic rice plants was severely reduced compared with the Kitaake control (Figure S3A). We observed a similar cell death to that detected in the Tos17 insertion mutant line in several independent RNAi lines (Figure S3B). Taken together, these results show that loss-of-function of XB15 induces expression of defense-related genes and elicits a cell death phenotype, suggesting a negative regulatory role for Xb15 in the defense signaling pathway.
We have shown that the Xb15 Tos17 insertion mutant line (NF9014) accumulates PR gene transcripts (Figure 5). On the basis of this result, we hypothesized that these lines would be more resistant to Xoo. However, because Nipponbare plants are already moderately resistant to Xoo PR6 and because we are pushing the limits of sensitivity of our bacterial assay, we were unable to detect enhanced resistance in these lines (unpublished data).
To further explore the role of Xb15 in resistance, we carried out additional experiments in the Kitaake genetic background, which is highly susceptible to Xoo PR6 (Figures 3 and 4). In this experiment, we crossed the Xb15 Tos17 insertion mutant line (NF9014–1) (pollen recipient) with a Kitaake line containing Myc-XA21 under control of its native promoter (pollen donor). We then tested if the loss of XB15 function alters the XA21-mediated defense response. We were able to simultaneously test for both reduced and enhanced resistance by taking advantage of the fact that the resistance conferred by XA21 is developmentally regulated. At the juvenile two-leaf stage (∼2 wk old), the plants are fully susceptible. Full resistance develops only at the adult stage [56].
We therefore inoculated progeny (F4) from the cross (NF9014–1/Myc-XA21) at 3 wk old when XA21 resistance is not yet fully active. At this stage of development, XA21 rice plants are only partially resistant (approximately 40% of resistance to that of 6-wk-old plants) [56]. Because lesion development continues to 16 d (Figure 4), we measured the lesion lengths at 18 d to capture as large a difference as possible (Figure 6A).
We found that progeny carrying Xa21 and the Xb15 Tos17 mutation (xb15/xb15, Xa21/−), displayed enhanced resistance to Xoo PR6 with lesions ranging length from 1–2 cm compared to the segregants carrying Xa21 alone (Xb15/Xb15, Xa21/−), which showed approximately 4-cm lesion lengths. The difference was statistically significant. In contrast, although enhanced resistance was observed in the Xb15 Tos17 mutant lines lacking Xa21 (xb15/xb15, −/−) as compared to the wild-type Xb15 line (Xb15/Xb15, −/−), this difference was not statistically significant (Figure 6A). This result indicates that, using our currently available methods, the enhanced resistance caused by the Xb15 mutation can be detected only in the presence of XA21.
To further validate the enhanced resistance phenotype conferred by the Xb15 Tos17 mutation in the XA21 genetic background, we measured bacterial populations in these lines. In Figure 6B, we show that Xoo populations multiplied to 9.2 × 107 cfu/leaf in the XA21 plants carrying the Xb15 Tos17 mutation (xb15/xb15, Xa21/−). In contrast, in the absence of the Xb15 Tos17 mutation (Xb15/Xb15, Xa21/−), bacterial populations reached to 3.5 × 108 cfu/leaf a 3.8-fold increase compared to the Xb15 Tos17 mutant lines. These results clearly show that loss of XB15 function enhances XA21-mediated resistance.
To elucidate the mechanism of XB15 negative regulation of XA21-mediated resistance, we initiated in vitro biochemical studies. We first examined whether Xb15 encodes a functional PP2C and if alterations in XA21-mediated resistance are caused by the putative PP2C activity of XB15. Full length XB15 (GST-XB15FL) tagged with an N-terminal tagged glutathione-S-transferase (GST) recombinant fusion protein and GST protein alone were expressed in E. coli. A smaller clone expressing the PP2C catalytic region alone, without the N-terminal extension (GST-XB15ΔN) was also expressed to check its possible regulatory function. We purified and assayed the recombinant proteins for phosphatase activity by measuring the release of phosphate from a phosphorylated synthetic peptide [57]. Figure 7A shows that in the presence of the substrate, both GST-XB15FL and GST-XB15ΔN, but not GST alone, catalyze the release of 240–350 pmoles of phosphate. This reaction was inhibited by the serine/threonine phosphatase inhibitor, sodium fluoride (NaF), but not by vanadate, a tyrosine phosphatase inhibitor. These results indicate that Xb15 encodes a protein serine/threonine phosphatase not a tyrosine phosphatase, as predicted on the basis of a sequence analysis that places XB15 in the PP2C clade.
To experimentally test which class of serine/threonine phosphatases XB15 falls into, we incubated the recombinant GST-XB15FL, GST-XB15ΔN, and GST proteins with the substrate in various reaction buffers optimized for activity for different protein PP classes, PP2A, PP2B, and PP2C (Figure 7B) [57]. As predicted, XB15 enzyme activity was detected in PP2C buffer but not in PP2A and PP2B buffer. The presence of Mg2+ in the PP2C buffer was required for full activity. Significant inhibition was observed in PP2B buffer with saturating amounts of calmodulin and additional Ca2+. While GST-XB15FL showed a slightly lower PP activity than GST-XB15ΔN, unlike POL, inhibition of phosphatase activity by the N-terminal domain appears to be minimal (Figure 7A and 7B).
To test whether plant-expressed XB15 has PP2C activity, we used the NTAP-XB15 transgenic plants described above. The NTAP tag includes two IgG binding domains from Staphylococcus aurous protein A and a calmodulin binding peptide linked by a TEV protease cleavage site [51,52]. This tag has been successfully used for purification of native protein complexes from yeast, plants, and animals [52,58,59]. The NTAP-tagged XB15 (NTAP-XB15) and NTAP alone were purified with IgG-agarose and assayed for their phosphatase activity in PP2C buffer. Figure 7C shows that NTAP-XB15 purified from transgenic rice plants carries PP2C activity, but not purified NTAP alone or boiled NTAP-XB15. These experiments demonstrate that Xb15 encodes a serine/threonine protein PP2C.
XB15 was originally isolated as one of XA21 binding proteins using the yeast two-hybrid screen. To rule out the possibility of false positive caused by nonspecific binding to XA21, the specificity of the interaction between XA21 and XB15 was confirmed using a GST pull-down assay with immobilized GST-XA21K668. GST-XA21K668 and HIS-XB15ΔN were expressed in E. coli and purified using glutathione and Nichel nitrilotriacetic (Ni-NTA) agarose beads, respectively. The purified recombinant HIS-XB15ΔN protein was incubated with glutathione beads bound to either GST-XA21K668 or GST alone in a binding buffer containing 2 mM MgCl2. Figure 8A shows that the recombinant protein HIS-XB15ΔN specifically interacts with GST-XA21K668 (lane 2) but not GST (lane 3) or glutathione beads (lane 1).
To further investigate the association between XA21 and XB15 in vivo, we used the Myc-XA21 transgenic plants described above. Anti-Myc antibody detected a 140-kDa polypeptide in transgenic plants carrying Myc-Xa21 but not in the control line Kitaake (Kit) (Figure 8B). When the Myc-XA21 protein is enriched after immunoprecipitation (IP) with agarose conjugated anti-Myc antibody, an additional 100-kDa product was detected. We have previously reported that the 140-kDa polypeptide is Myc-XA21 and the 100-kDa polypeptide is a proteolytic cleavage product of Myc-XA21 (Myc-XA21cp) [42]. We then developed a polyclonal antibody against a synthetic peptide of XB15 (anti-XB15) to detect the XB15 protein in planta and confirmed its specificity against XB15 from rice extracts. Anti-XB15 detected a major band of 70 kDa, which is close to the predicted size of XB15 (69.2 kDa) (Figure 8C). In the NTAP-XB15 overexpression line 17A and 19A, there was a significant increase in NTAP-XB15 (95 kDa) compared to endogenous XB15 (70 kDa).
When rice protein extracts were incubated with agarose conjugated anti-Myc antibody, an immune complex containing a 140-kDa polypeptide was precipitated and detected after western blot analysis with anti-Myc antibody in Myc-XA21 transgenic plants inoculated with Xoo PR6 but not in the Kitaake control (Figure 8D). Although the same amount of total protein extract was used for each immunoprecipitation, the amount of Myc-XA21 protein precipitated by anti-Myc-antibody accumulated to greater amounts at 12 and 24 h after Xoo PR6 inoculation as compared to the 0-h time point. Next, we examined whether XB15 was co-immunoprecipitated with XA21 in the immune complex. A 70-kDa polypeptide was detected with the anti-XB15 antibody in Xoo PR6-inoculated protein extracts. This band was not detected in Kitaake lacking Myc-Xa21 and was barely detectable in mock-treated Myc-XA21 plants. The association between XB15 and XA21 was detectable by 12 h after inoculation with Xoo PR6 and had significantly increased after 24 h. As a control, similar experiments were performed with agarose-conjugated anti-Myc antibody presaturated with Myc peptide to check for a nonspecific interaction of XB15 with Myc peptide; we detected no interaction between XB15 and the Myc peptide (unpublished data). These results demonstrate a direct interaction between XB15 and XA21 in leaves, consistent with our in vitro observations.
We have previously shown that XA21 encodes a protein kinase and that autophosphorylation of the XA21 kinase domain occurs via an intramolecular mechanism [43]. Because XB15 is associated with XA21 both in vitro and in vivo, because the overexpression of XB15 compromises the XA21-mediated resistance, and because XB15 possesses PP2C activity, we hypothesized that XB15 could directly dephosphorylate XA21. We tested this hypothesis by checking if XB15 could dephosphorylate the in vitro autophosphorylated XA21. After expression in E. coli and purification, GST-XA21K668 bound to glutathione beads was autophosphorylated with [γ-32P] in vitro. Autophosphorylation was monitored by radioactive incorporation of γ-32P-ATP into the bead-binding fraction (unpublished data) and by SDS-PAGE. The exposure to X-ray film shows that the fusion protein, GST-XA21K668 is capable of autophosphorylation (Figure 9A, 0 min).
The 32P-autophosphorylated GST-XA21K668 was incubated with or without 1 μg of purified recombinant HIS-XB15ΔN in the presence of PP2C buffer for the indicated time (Figure 9A). While XA21K668 remained labeled for over 60 min in the absence of XB15 (unpublished data), dephosphorylation of XA21K668 was detectable after 1 min of incubation with XB15. Additionally, dephosphorylation of XA21K668 by HIS-XB15ΔN was not detected in the presence of serine/threonine PP inhibitor, NaF (20 mM) or when HIS-XB15ΔN (1 μg) was boiled (unpublished data) suggesting that the dephosphorylation of XA21 is caused by the PP2C activity of XB15. In further support of this conclusion, increasing the amount of HIS-XB15ΔN (1–10 μg) resulted in a dosage-dependent dephosphorylation of GST-XA21K668 (Figure 9B). Although XA21-dependent phosphorylation of XB15 was examined with recombinant proteins purified from E. coli at various time points and different conditions, we did not detect transphosphorylation of XB15 by XA21 (unpublished data). From these results, we conclude that XA21 is a substrate of XB15 and that the XB15 phosphatase can effectively dephosphorylate XA21.
To further elucidate XB15 function, we assessed its subcellular localization. According to MultiLoc (http://www-bs.informatik.uni-tuebingen.de/Services/MultiLoc) [60] and PSORT (http://www.psort.org/) [61], XB15 has a putative nuclear-localization sequence (Figure 2) and is predicted to be a nuclear protein. To investigate the in vivo cellular distribution of XB15, a targeting experiment was performed using smGFP2, as a fluorescent marker [62]. We fused the entire Xb15 coding region without the termination codon to the smGFP2 gene (Figure S4A), and the resulting construct was introduced into rice Kitaake protoplast cells by PEG-mediated transformation [63,64]. Localization of the fusion protein was determined by visualization with an Olympus FV1000 confocal microscope. We introduced the smGFP2 gene alone into protoplast cells as a control. As shown in Figure S4B, the fusion protein was mainly localized to the plasma membrane of the rice protoplast cell, whereas the control smGFP2 was uniformly distributed throughout the cell except the large central vacuole. As a positive control for plasma membrane targeting, we cotransfected protoplasts with vector expressing a H+-ATPase-red fluorescent protein (dsRed) fusion protein, which also localized to the plasma membrane [65]. A close overlap was observed between the green and red fluorescent signals of XB15-smGFP2 and H+-ATPase-dsRed, respectively. This experiment was repeated with protoplasts extracted from transgenic rice carrying Xa21 gene and same localization of XB15 to the plasma membrane was observed (unpublished data).
We next tested if the large green fluorescent protein (GFP) tag would inhibit XB15 phosphatase activity. XB15-smGFP2 and smGFP2 alone were purified from transiently transformed protoplasts with agarose conjugated anti-GFP antibody and assayed for their phosphatase activity in PP2C buffer (Figure S5). We found that this purified XB15-smGFP2 carries PP2C activity similar to that observed for NTAP-XB15 (Figure 7C), but not purified smGFP2 alone or boiled XB15-smGFP2. Taken together, these results suggest that the tagged protein is functional and that, in vivo, XB15 is primarily targeted to the plasma membrane. Because its membrane targeting appears to be constitutive external signals are likely not required for its membrane translocation. This observation is also consistent with the predicted plasma membrane-localization of the intact XA21 protein [44].
Recognition of pathogen-associated molecules by PRRs is a critical component of innate immunity in animals and plants [66]. The rice XA21 protein is representative of a very large class of plant PRRs that carry the non-RD motif in the kinase domain [10]. The regulation of this important subclass of kinases has not yet been elucidated. Here, we have shown that rice employs a novel PP2C, XB15, to attenuate the XA21-mediated innate immune response.
XB15 possesses an active PP2C domain in its C-terminal region. On the basis of sequence comparison and phylogenetic analysis among PP2Cs from the rice and Arabidopsis databases, XB15 shows relatively high similarity with Arabidopsis POL, PLL4, and PLL5 (Figure 2B). All cluster into a small subgroup of the rice/Arabidopsis phylogenetic tree.
Abnormal leaf development was observed in Arabidopsis knockout mutants for pll4 and pll5 and transgenic plants overexpressing Pll5 [48]. Despite the sequence similarity with POL and PLLs, we observed no disorders of leaf development in the Xb15 knockout mutant and RNAi plants. Additionally, transgenic plants overexpressing Xb15 displayed normal leaf development. These results indicate that XB15 has distinct functions to that of the Arabidopsis POL and PLLs.
The apparent differences in the function of XB15 and Arabidopsis POLs may be caused by their different localizations. Arabidopsis POL contains a putative nuclear-localization motif [47] and is predicted to be a nuclear protein according to the prediction methods for subcellular location, MultiLoc, and PSORT. In contrast, the fusion protein, XB15-smGFP2 is mainly localized to the plasma membrane of rice protoplast cells, suggesting that the in vivo function of XB15 is primarily related to the plasma membrane itself or to other membrane-localized proteins (Figure S4). Based on sequence analysis, XA21 is predicted to localize to the plasma membrane like other RKs, but its subcellular localization has not yet been demonstrated in vivo. Our in vivo data of XB15 combined with the putative localization of XA21 suggest that the interaction between XB15 and XA21 occurs at the plasma membrane, and may be similar to KAPP, which interacts with its target RKs at the plasma membrane to attenuate their signaling [34,36,67].
In animals, some receptor tyrosine kinases are controlled through autophosphorylation of their JM domains. For example, the mouse ephrin and KIT receptors, and human Fms (Feline McDonough Sarcoma)-like tyrosine kinase 3, interact with protein tyrosine phosphatases via phosphorylated tyrosine residues in their JM domains. These interactions negatively regulate receptor tyrosine kinase-mediated signaling [68–71]. In plants, only a few RKs have been shown to autophosphorylate residues in their JM regions. These include barley HvLysMR1, legume symbiosis RK, Arabidopsis BRI1, and rice XA21 [43,72–76]. In spite of the presumed importance of the JM domain in plant RK-mediated signaling, no downstream target proteins that bind the JM regions and regulate RK signaling in a phosphorylation-dependent manner have yet been identified.
XB15 failed to interact with the XA21K(TDG) mutant lacking the XA21 JM region and the catalytically inactive mutant XA21K668K736E (Figure 1), suggesting that XA21 autophosphorylation in the JM region is critical for the XB15/XA21 interaction [43]. Although autophosphorylation of XA21 has not been demonstrated in yeast, there have been many reports showing that plant proteins maintain their function when they are expressed in yeast as heterologous proteins [77–79]. For example, an Arabidopsis kinase, GSK3/shaggy-like protein rescues the phenotype of its yeast homolog mck1, suggesting that plant kinases in yeast cells undergo the necessary post-translational modification for their in vivo functions [78]. Autophosphorylation of the JM residues Ser686, Thr688, and Ser689 was previously shown to be important stabilizers of XA21 protein levels [44]. XA21 proteins with mutations in these residues maintained interaction with Xb15, suggesting that these phosphorylation sites mainly function in the stability of XA21 protein but are not directly involved with mediating the downstream signal transduction cascade. In contrast, when Ser697 is mutated to Alanine, interaction with XB15 is abolished, indicating that Ser697 is essential for XA21 binding with XB15 in yeast. The Ser697 mutation does not affect kinase activity of XA21S697A, suggesting that Ser697 serves as a high affinity binding site rather than as a regulator of kinase activity (unpublished data, X. Chen, P.E. Canlas, M. Chern, D. Ruan, R. Bart, et al.). Taken together, these results suggest that XA21 requires its own kinase activity and residues in the JM region for interaction with XB15.
In support of this hypothesis, our preliminary results suggest that transgenic rice carrying the XA21 variant, XA21S697A, display enhanced resistance to Xoo PR6 at the 3-wk-old stage (unpublished data, X. Chen, P.E. Canlas, M. Chern, D. Ruan, R. Bart, et al.). These results indicate that Ser697 is essential for XA21 binding with XB15 and that in the absence of XB15 binding to XA21, negative regulation is compromised leading to enhanced resistance.
In both animals and plants, PCD is a highly regulated process involved in immunity and other functions [66]. Many mutants exhibiting spontaneous PCD, initially isolated in maize [80], have now been identified in other plants, including Arabidopsis, barley, and rice [80–86]. Analysis of these mutants has led to the identification of genes that regulate cell death [87].
The Xb15 Tos17 insertional mutant line, NF9014, and Xb15 RNAi transgenic rice lines, RNAiXB15, display spontaneous cell death on the leaves during development under green house condition (26–28 °C) in the absence of obvious stress or disease. All of the tested defense-related PR genes commonly associated with the PCD and defense response [87–89] are constitutively expressed in the NF9014 mutant. In addition, progeny (F4) carrying both Xa21 and the Xb15 Tos17 mutation derived from the cross (NF9014–1/Myc-XA21) shows enhanced resistance to Xoo PR6, indicating that XA21-mediated resistance is enhanced in the absence of XB15 (Figure 6). In the absence of XA21, we do see enhanced resistance, although this difference is not statistically significant. We believe this unexpected result is due to the limitations in sensitivity of our assay.
The PCD-like phenotype observed in plants lacking Xa21 suggests the presence of an additional signaling cascade that is negatively regulated by a functional XB15 (Figure 5). There are many examples of PP2Cs that function as negative regulators of multiple kinase cascades. In addition to Arabidopsis KAPP, which associates with at least five RKs [31–36], mouse PP2Cε dephosphorylates and negatively regulates MKKKs, apoptosis signal-regulating kinase 1 and transforming-growth-factor-β-activated kinase 1 (TAK1) [90]. In this report, we provide evidence that XB15 regulates at least one, but likely more, RK-mediated pathways, controlling PCD-like lesions.
Figure 10 compares our working model for XA21-mediated innate immunity with the human TLR4 and Arabidopsis FLS2 signaling cascades. TLR4 contains an extracellular LRR that is critical for transmitting the lipopolysaccharide signal (LPS) across the cell membrane [91]. An adaptor molecule, myeloid differentiation factor 88 (MyD88), associated with the Toll/interleukin-1 receptor (TIR) intracellular domain of TLR4, recruits IRAK4 [5]. The activated IRAK4 rapidly phosphorylates the non-RD kinase IRAK1, and leaves the receptor complex to interact with the tumor necrosis factor receptor associated factor 6 (TRAF6) [92]. TRAF6 is autoubiquitinated and forms a complex with TAK1, which functions as a MAPK kinase kinase [93,94]. The activated kinase TAK1 mediates downstream events, such as the activation of inhibitor of κ kinase, p38, and Jun-N-terminal kinases, which lead to the activation of transcription factors including NF-κB and activating protein-1 [5,95].
Similar to TLR4, Arabidopsis FLS2 also carries an extracellular LRR domain that recognizes a pathogen-associated molecule; in this case a small peptide called flg22 [15]. FLS2 contains an intracellular non-RD kinase, which, when activated, transduces the signal to a MAPK cascades [96–98]. WRKY transcription factors in the nucleus are activated by the MAPK cascades and turn on downstream target genes [97]. KAPP has been shown to negatively regulate FLS2-mediated signaling through overexpression studies, although the mechanism has not yet been elucidated [33].
In rice, the XA21 LRR domain is responsible for race-specific recognition of Xoo strains carrying AvrXA21 [20,99]. XA21/AvrXA21 binding is hypothesized to activate the non-RD kinase domain leading to XA21 autophosphorylation and/or transphosphorylation of downstream target proteins [20,100]. In support of this hypothesis we have observed a strong interaction between XB15 and XA21 in Xoo PR6-inoculated rice plants but not in untreated rice (Figure 8D), suggesting that Xoo inoculation induces XA21/XB15 complex formation. Our data indicate that Ser697 in the XA21 JM region is required for XB15 binding.
XA21 transphosphorylates the RING finger ubiquitin ligase XB3, which is required for effective XA21-mediated resistance [41,92]. We have also shown a direct, regulatory role of XB10 (OsWRKY62) in the XA21-mediated response [41], suggesting another layer of conservation between the Arabidopsis FLS2 and rice XA21 signaling pathways. In the nucleus, XB10 and other WRKY transcription factors (unpublished data, Y. Peng, L.E. Bartley, and P.C. Ronald) [41] either activate or repress defense-related genes such as PR1 and PR10.
The XA21-mediated defense response is also regulated by rice negative regulator of resistance (NRR) that interacts with NH1 [53,54], the rice ortholog of NPR1, a key regulator of the SA-mediated defense pathway (unpublished data). This result demonstrates cross-talk between the XA21- and NH1-mediated pathways.
In contrast to animal PPs that inactivate phosphorylated MAPKs to negatively regulate the innate immune response [24], XB15 directly interact with the non-RD kinase domain of XA21. This study suggests that XA21 is a substrate of XB15 and that the phosphatase activity of XB15 attenuates the XA21-mediated innate immune response. Future studies will be directed at identifying the XA21 phosphorylated residue(s) targeted by XB15 for dephosphorylation.
Rice (Oryza sativa L.) plants were maintained in the green house. The growth chamber was set on a 16-h light and 8-h dark photoperiod, a 28/26 °C temperature cycle, and 90% humidity. Healthy and well-expanded leaves from 6-wk-old rice plants were used for Xoo PR6 inoculation and nucleic acid or protein extraction. XA21K668 and XA21K(TDG) were amplified using the primer pairs 5′-CACCATGTC ATCACTCTACTTGCTTA-3′/5′-TCAGAATTCAAGGCTCCCA-3′ and 5′-CACCATGACAGATGGTTTCGCGCCGACC-3′/-TCAGAATTCAAGGCTCCCA-3′, respectively, and cloned into pENTR/D-TOPO/D vector (Invitrogen). XA21K668K736E, XA21K668S686A, XA21K668T688A, XA21K668S689A, XA21K668S686A,T688A,S689A, and XA21K668S697A were constructed by site-directed mutagenesis using the Quick Change kit (Stratagene), according to the manufacturer's protocols. The specific primers for the mutagenesis were 5′- GTTGCAGTGGAAGTACTAAAGCTTGAAAATCC-3′/5′- GGATTTTCAAGCTTTAGTACTTCCACTGCAAC-3′ (for XA21K668K736E), 5′- AAGGGAGCCCCTGCAAGAACTTCCATG-3′/5′- CATGGAAGTTCTTGCAGGGGCTCCCTT-3′ (for XA21K668S686A), 5′- GCCCCTTCAAGAGCTTCCATGAAAGGC-3′/5′- GCCTTTCATGGAAGCTCTTGAAGGGGC-3′ (for XA21K668T688A), 5′- CCTTCAAGAACTGCCATGAAAGGCCAC-3′/5′- GTGGCCTTTCATGGCAGTTCTTGAAGG-3′ (for XA21K668S689A), 5′- AAGGGAGCCCCTGCAAGAGCTGCCATGAAAGGCCAC-3′/5′- GTGGCCTTTCATGGCAGCTCTTGCAGGGGCTCCCTT-3′ (for XA21K668S686A,T688A,S689A), and 5′- CACCCATTGGTCGCTTATTCGCAGTTG-3′/5′-CAACTGCGAATAAGCGACCAATGGGTG-3′ (for XA21K668S697A). The positive clones were verified by DNA sequencing and then using Gateway LR Clonase (Invitrogen), moved into the yeast two-hybrid vector pNlexA carrying the BD domain (Clontech). Positive clones were confirmed again by DNA sequencing. AD-Xb15 was from the clone pAD-GAL4–2.1 identified from a cDNA library. The purified plasmid DNAs from AD vectors and BD vectors were cotransformed into the yeast cell pEGY48/p8op-LacZ (Clontech) using the Yeast transformation kit, Frozen-EZ yeast transformation II (Zymo Research). We followed the detailed procedure from the manual of Matchmaker LexA Two-Hybrid System (Clontech).
Full-length cDNA corresponding to Xb15 was amplified by PCR with primers 5′-CACCATGGGCAACTCCCTCGCCTG-3′/5′-TTACACGCAGGATCTCCAAATC-3′. PCR fragments were purified and subcloned into the pDEST15 or 17 (Invitrogen), which expresses the recombinant protein with an N-terminal GST or HIS tag, respectively. The resulting expression vectors were transformed into the bacterial host strain BL21 (DE3) pLysS (Invitrogen), and expression of protein was induced at midlog phase (1 mM isopropyl β-D-thiogalactosiadse, 3 h, 28 °C). Recombinant proteins were purified by affinity chromatography using Glutathione Sepharose 4B (Amersham) or Ni-NTA bead (Qiagen).
For the purification of the recombinant protein from rice, leaves of 5- to 6-wk-old NTAP-XB15 transgenic plants were harvested essentially as previously described [51,52]. Five grams fresh weight of rice shoots were ground to a fine powder in liquid nitrogen. Crude protein extracts were prepared in four volumes of Extraction Buffer I (20 mM Tris-HCl, [pH 8.0], 150 mM ethylene-diamine-tetra-acetic acid (EDTA), 2 mM benzamidine, 0.1% IGEPAL, 10 mM β-mercaptoethanol, 20 mM NaF, 1 mM phenylmethanesulfonylfluoride [PMSF], 1% Protease cocktail (Sigma), 2 μg/ml leupeptin, 2 μg/ml antipain, and 2 μg/ml aprotinin). The extract was passed through a fine sieve, filtered through two layers of miracloth, and centrifuged twice at 13,000 g for 30 min at 4 °C. The cleared supernatant was mixed with 50 μl of IgG Sepharose beads (Amersham) and incubated at 4 °C for 2 h on Labquake Rotisserie (Barnstead Thernolyne). After centrifugation at 3,000 g for 30 sec, IgG supernatant was discarded and the collected IgG beads were washed four times in 5 ml Extraction Buffer I lacking protease inhibitors and twice in 1 ml of 5 mM ammonium acetate (pH 5.0). The protein was eluted with 1 ml of 0.5 M acetic acid (HOAc) (pH 3.4), neutralized with 100 μl of 1 M Tris-HCl (pH 8.0), and concentrated by acetone precipitation. The purified NTAP-XB15 protein was used for PP activity assay and protein gel blot analysis.
Purified recombinant proteins, GST-XA21K668, HIS-XB15, and GST were used for pull-down assays as described [101], except that buffers were supplemented with 2 mM MgCl2. Bound proteins were eluted by boiling in SDS sample buffer, separated by SDS-PAGE, and detected by immunoblotting with a anti-histidine antibody (Quiagen). Recombinant GST or GST-XA21K668 were incubated with glutathione-sepharose beads (Amersham) overnight at 4 °C and washed three times for 10 min at 4 °C with phosphate buffer saline (PBS). The beads were equilibrated with Binding Buffer (20 mM Tris [pH 7.5], 150 mM NaCl, 2 mM MgCl2, 0.5 mM dithiothreitol) to which a protease inhibitor mixture (Roche) was added. Recombinant HIS-XB15 was incubated with GST or GST-XA21K668 (20 μg) bound to glutathione beads. After 2 h of incubation at 4 °C, the beads were washed four times for 20 min in binding buffer. Proteins bound to the bead were eluted with SDS sample buffer, separated by SDS-PAGE, and processed for immuno-blotting.
To co-immunoprecipitate Myc-XA21 and XB15, total proteins were extracted from 5 g of leaf tissue in 25 ml of ice-cold Extraction Buffer II (0.15 M NaCl, 0.01 M Na-phosphate [pH 7.2], 2 mM EDTA, 1% Triton X-100, 10 mM β-mercaptoethanol, 20 mM NaF, 1 mM PMSF, 1% Protease cocktail [Sigma], 2 μg/ml leupeptin, 2 μg/ml antipain, and 2 μg/ml aprotinin). After filtering through Miracloth (Calbiochem) followed by centrifugation twice at 13,000 g for 20 min at 4 °C, the supernatant was mixed with 50 μl of agarose conjugated anti-Myc antibody (Santa Cruz) and incubated at 4 °C for 2 h. The beads were then washed four times in 1 ml of Extraction Buffer II without proteinase inhibitors. The proteins were eluted with 4× Laemmli loading buffer. Protein blot analyses were performed. To co-immunoprecipitate XB15-smGFP2, total proteins were extracted from protoplasts transformed with Xb15-smGFP2 in 5 ml of ice-cold Extraction Buffer II and 20 μl of agarose conjugated anti-GFP antibody (Santa Cruz). For the anti-XB15 antibody, synthetic peptides and monospecific antibodies were made as a service by Pacific Immunology. Detailed information about their methods can be obtained at Pacific Immunology (http://www.pacificimmunology.com/). Epitope selection was directed to hydrophobic, flexible regions of the proteins. The epitope, TRALLARTEKFQDSADL was used. Peptides were synthesized and conjugated to keyhole limpet hemocyanin. Rabbits were immunized with peptide in complete Freund's adjuvant, followed by three boosts in incomplete Freund's adjuvant. Monospecific antibodies were affinity purified to the synthetic peptides bound to C3-SEP-PAK cartridges. Antibodies were conjugated to horseradish peroxidase for use in Western blots.
For Xoo inoculation, rice plants were grown in the greenhouse normally until they were 6 wk old, unless stated otherwise, and transferred to the growth chamber. The Xoo strain Philippine race 6 (PR6) was used to inoculate rice by the scissors-dip method [18,53]. Only the top two to three expanded leaves of each tiller were inoculated. For Xoo colony counts from inoculated leaves, 20 cm of leaf tissue from the top, including lesions and tissue showing no lesions, was ground up and resuspended in 10 ml water to harvest bacteria. The extract was diluted accordingly and plated out on peptone sucrose agar (PSA) plates containing 15 mg/l cephalexin.
The transgenic lines Myc-XA21 T330-16-1 was used as the pollen recipient in a cross with pollen donor NTAP-XB15 17A and 19A. Over 50 seeds were recovered from the each T330-16-1/17A and T330-16-1/19A cross. The nature of the F1 hybrid was confirmed by PCR amplification of 820-bp fragment spanning part of the Myc tag and Xa21 in the Myc-Xa21 construction using primers 5′-GAGCAAAAGCTGATTTCTGAGGAGGAT-3′/5′-ACCACCTAGCTTGTTTTCTCTGAC-3′ and 650-bp fragment spanning part of the NTAP tag and Xa21 in the Ntap-Xb15 construction using primers 5′-ATGCCCAAGCCCCAAAGGACTACG-3′/5′-GAAGCTTGGACGGCGCCACCCATACGAC-3′.
Rice transformation was constructed as described previously [53]. Agrobacterium EHA105 was used to infect rice callus for transformation. Transformants of rice cultivars Kitaake carrying Ntap-Xb15 or Myc-Xa21 were selected using hygromycin.
For NTAP-XB15 detection, rabbit PAP soluble complex (Sigma) was used at a final dilution of 1:5,000 for 2 h. Bands were visualized using the SuperSignal West Pico Chemiluminescent Substrate (Pierce) according to standard protocol.
PCR was performed using the gene-specific oligonucleotide primers, 5′-GGGATCCAATGGGCAACTCCCTCGCCTG-3′/5′-GGGATCCACGCAGGATCTCCAAATC-3′. Subsequently, the termination codon of the Xb15 cDNA was removed. The PCR-amplified product was fused in-frame to the coding region of soluble-modified green fluorescent protein (smGFP2) (kindly provided by K.H. Paek) [63,64]. Transient expression of green fluorescent protein (GFP) fusion constructs and H+-ATPase-dsRed (kindly provided by I. Hwang) were performed by introducing the plasmid into the rice protoplasts using the PEG-mediated transformation method [63,64].
Images were collected with an Olympus FV1000 confocal microscope. GFP was imaged under the following conditions: excitation: 488 nm; DM 405/488/543; emission: 500–530 nm. DsRed was imaged under the following conditions: excitation: 543 nm; DM 405/488/543; emission: 560–620 nm. Images were collected through a 40× (NA: 1.00) oil immersion lens. All images are the result of 2 kalman line averages, and, where appropriate, sequential scans were used to prevent cross talk. Images were analyzed using the Olympus Fluorview software (Ver 1.4a) and coded green (for GFP) or red (for dsRed).
For RT-PCR analysis, total RNAs were extracted from leaves after each treatment and then the RT reaction was performed following the manual for QuantumRNA 18S Internal Standards (Ambion). PCR analyses were performed with primers pairs, 5′-TTATCCTGCTGCTTGCTGGT-3′/5′-GGTCGTACCACTGCTTCTCC-3′ (for PR1a), 5′-AGGTATCCAAGCTGGCCATT-3′/5′-GGCGTAGTCGTAGTCGCTCT-3′ (for PR1b), 5′-CGCAGCTCACATTATCAAGTCAGA-3′/5′-GAAGCAGCAATACGGAGATGGATG-3′ (for PR10), 5′-GCAGGGAGCGTATACAAGACCAA-3′/5′-CACGCCACAGTAACATGACCACAA-3′ (for Betv1), 5′-CAACAGTCGAAGGGCAATAATAAGTC-3′/5′-ACTGCCACACCTCCCACATTG-3′, 5′-ATGGCTCCGGCCTGCGTCTCCGA-3′/5′-GGCATATTCGGCAGGGTGAGCGA-3′ (for PBZ1), 5′-TCATTCGATGGATCAGTCGGG-3′/5′-ATGCTCTGGTCACCTTCAGCG-3′ (for Xb15), and 5′-CAACAGTCGAAGGGCAATAATAAGTC-3′/5′-ACTGCCACACCTCCCACATTG-3′ (for EF1α). The amplified products were then resolved by gel electrophoresis.
Phosphatase activity was measured according to the instructions provided by the manufacturer by using a nonradioactive serine/threonine phosphatase assay system (Promega). The color was allowed to be developed for 15 min, and the absorbance was measured at 600 nm with plate reader (Bio-Rad). The composition of the buffers used in the assay was: PP2A 5× buffer (250 mM imidazole [pH 7.2], 1 mM EGTA, 0.1% 2-mercaptoethanol, 0.5 mg/ml BSA), PP2B 5× buffer (250 mM imidazole [pH 7.2], 1 mM EGTA, e50 mM MgCl2, 2 mM CaCl2, 250 μg/ml calmodulin, 0.1% 2-mercaptoethanol), PP2C 5× buffer (250 mM imidazole [pH 7.2], 1 mM EGTA, 25 mM MgCl2, 0.1% 2-mercaptoethanol, 0.5 mg/ml BSA).
Purified agarose-bound fusion proteins, GST-XA21K668, were washed with kinase buffer (50 mM HEPES [pH 7.4], 10 mM MgCl2, 10 mM MnCl2, 1 mM dithiothreitol). Autophosphorylation experiments were carried out in 30-μl volumes containing 20 μl of agarose-bound protein (5 μg) and 20 μCi of [γ-32P]ATP (6,000 Ci/mmol) (PerkinElmer Life Science). The reaction was stopped after 30 min by adding 10 μl of Laemmli loading buffer and boiling for 5 min. The proteins were separated by SDS-PAGE (7.5% or 10%). After staining with Coomassie Brilliant Blue G-250, the gel was dried and exposed to x-ray film.
To dephosphorylate the phosphorylated fusion proteins by HIS-XB15ΔN, the 32P-labeled XA21K668 proteins were washed with PP2C buffer and incubated with HIS-XB15ΔN. The resulting proteins were resolved by SDS-PAGE as described above.
A 1,920-nt cDNA fragment encoding full-length XB15 protein was amplified from a rice cDNA using primers, 5′-CACCATGGGCAACTCCCTCGCCTG-3′/5′-TTACACGCAGGATCTCCAAATC-3′. The PCR product was cloned into pENTR/D-TOPO (Invitrogen) according to the instructions provided by the manufacturer and the insert confirmed by sequencing. For Overexpression in rice, the Xb15 cDNA in pENTR/D-TOPO was recombined into the final Ubi-NTAP-1300 vector using Gateway LR Clonase (Invitrogen). Ubi-NTAP-1300 is a pCAMBIA-1300 (AF234296) derivative with an additional expression cassette containing the maize ubiquitin promoter and a nopaline synthase 3′-polysdenylation region, to which the NTAP/Gateway cassette was added [52].
The rice cDNA locus identification numbers (TIGR; http://rice.tigr.org/) are as follows: Xb15, Os03g60650; PR1a, Os07g03710; PR1b, Os01g28450; PR10, Os12g36830; Betv1, Os12g36850; PBZ1, Os12g36880; EF1s, Os03g08010. The GenBank (http://www.ncbi.nlm.nih.gov/Genbank) accession number for Xb15 cDNA sequence is NP_001051726.
|
10.1371/journal.pntd.0005951 | Pre-clinical antigenicity studies of an innovative multivalent vaccine for human visceral leishmaniasis | The notion that previous infection by Leishmania spp. in endemic areas leads to robust anti-Leishmania immunity, supports vaccination as a potentially effective approach to prevent disease development. Nevertheless, to date there is no vaccine available for human leishmaniasis. We optimized and assessed in vivo the safety and immunogenicity of an innovative vaccine candidate against human visceral leishmaniasis (VL), consisting of Virus-Like Particles (VLP) loaded with three different recombinant proteins (LJL143 from Lutzomyia longipalpis saliva as the vector-derived (VD) component, and KMP11 and LeishF3+, as parasite-derived (PD) antigens) and adjuvanted with GLA-SE, a TLR4 agonist. No apparent adverse reactions were observed during the experimental time-frame, which together with the normal hematological parameters detected seems to point to the safety of the formulation. Furthermore, measurements of antigen-specific cellular and humoral responses, generally higher in immunized versus control groups, confirmed the immunogenicity of the vaccine formulation. Interestingly, the immune responses against the VD protein were reproducibly more robust than those elicited against leishmanial antigens, and were apparently not caused by immunodominance of the VD antigen. Remarkably, priming with the VD protein alone and boosting with the complete vaccine candidate contributed towards an increase of the immune responses to the PD antigens, assessed in the form of increased ex vivo CD4+ and CD8+ T cell proliferation against both the PD antigens and total Leishmania antigen (TLA). Overall, our immunogenicity data indicate that this innovative vaccine formulation represents a promising anti-Leishmania vaccine whose efficacy deserves to be tested in the context of the “natural infection”.
| Although vaccination is accepted as a potentially effective approach to prevent leishmaniasis, to date there is no vaccine available for human disease. The research on the topic is therefore extremely important, and the design and testing of new vaccine approaches, as well as non-traditional immunization schemes continues to be as relevant as before. This study proposes an innovative vaccine approach for human visceral leishmaniasis, not only due to its multi-antigen nature which contemplates both parasite and vector derived proteins, but also because it explores the possibility of the use of Influenza virosomes as antigen-delivery vehicles. A strong TLR-4 agonist completes the vaccine formulation. Here we show the rationale-behind this vaccine approach, the safety of all the vaccine components in our in vivo context, and immunogenicity studies of the optimized vaccine candidate in mice that explored the contribution of the virosome to the antigen-elicited immune responses. Additionally, we tested an unusual immunization scheme that potentiated the final vaccine-elicited immune responses. This prime-boost immunization approach gives relevance to the use of both parasite and vector derived antigens together as an anti-Leishmania vaccine, and proposes a new strategy for vaccination in endemic areas, where people are constantly exposed to sand fly bites.
| Leishmaniasis is a spectrum of pathological outcomes caused by different Leishmania spp., intracellular parasites with a complex life cycle requiring a susceptible host and a permissive vector [1]. Visceral leishmaniasis, the most severe form of the disease, fatal if untreated, is caused by L. donovani and L. infantum, parasite species that migrate to the liver, spleen and bone marrow [2–4]. It has a worldwide distribution, being endemic in 74 countries, representing more than 37% of the total Earth terrestrial area [5]. Every year an estimated 0.2 to 0.4 million new VL cases occur and more than 20 000 people die, mostly in developing nations where access to healthcare is limited [6]. Furthermore, scarce and sometimes ineffective treatment options challenge leishmaniasis control [7].
Vaccination is considered one of the most cost/effective ways to control Leishmania infection. However, no human leishmaniasis vaccine is currently available. Several candidates have been proposed during the past few decades [8]. Some were shown to be immunogenic and have conferred protection against Leishmania in rodent models. Nevertheless, most of them were discarded after proving to be ineffective in large animals [8, 9]. Furthermore, most of these studies shared a limitation which may have been responsible for the overestimation of the vaccine candidates effectiveness: they had a binomial focus (host-parasite) and disregarded the contribution of the vector, essential in vaccine efficacy determination as highlighted by Peters et al who showed the loss of protection of a potentially-good vaccine candidate when tested in the context of vector-transmitted leishmaniasis [10].
Leishmania parasites are transmitted by sand flies from the genera Lutzomyia and Phlebotomus in a specific vector-Leishmania spp. pairing [11]. During the sand fly blood meal, parasites together with vector derived factors, including saliva, are introduced into host skin [11–13]. Previous exposure to sand fly salivary components has been shown to confer protection against vector-transmitted Leishmania [14, 15]. Furthermore, in recent studies, protection against natural transmission of Leishmania has been attained by vaccination with defined salivary molecules in animal models for both cutaneous leishmaniasis and VL [16, 17]. Interestingly, these proteins were shown to improve the protection induced by live anti-Leishmania vaccines [18, 19]. Fundamentally, the Th1 immune response elicited against a salivary molecule can adversely impact parasite establishment in the host.
This study proposes a novel vaccine candidate based on defined antigens of both parasite (KMP11 and LeishF3+, the latter a fusion protein consisting of Nucleoside hydrolase, Sterol 24-c-methyltransferase and Cysteine protease B), and sand fly vector (salivary protein LJL143) origins, formulated into Influenza virosomes and adjuvanted with GLA-SE, a TLR-4 agonist. The sand fly antigen LJL143 was shown to produce a long lasting Th1 immune response in dogs, which impacted parasite growth in vitro [20]. One of the parasite-derived antigens, KMP-11, was already demonstrated to be individually effective against VL in the pre-clinical context [21], as were each of the individual components of the second parasite-derived antigen, the fusion protein LeishF3+ [22–24]. Additionally, LeishF3+ predecessor antigen (a fusion protein consisting of Nucleoside hydrolase and Sterol 24-c-methyltransferase, but not Cysteine protease B), was considered safe and immunogenic in the clinical context (Phase I trial) [25], as were also both the adjuvant and the virosomes [26, 27]. Influenza virosomes represent a unique vaccine delivery system, flexible but robust, that allows loading of a wide variety of antigens [28, 29]. The VLP-based antigen formulation has the potential to generate both CD4+ and CD8+ specific memory T cells, the latter due to the potentiation of cross-presentation events [30]. The immune response elicited by the immunization should induce a Th1 phenotype due to the adjuvant chosen and the presence of the sand fly salivary antigen [26, 31]. In theory, an immunized individual bitten by an infected sand fly and exposed to parasites and vector saliva, will quickly mount both a strong Th1 anti-Leishmania, and a strong Th1-DTH anti-sand fly saliva immune responses, resulting in prevention of infection establishment.
Here, we explore the safety and antigenicity of the vaccine candidate, using ex-vivo and in-vivo approaches.
Animal experiments were performed in accordance with the IBMC.INEB Animal Ethics Committee and the Portuguese National Authorities for Animal Health guidelines (directive 2010/63/EU). BPC and ACdS are accredited for animal research (Portuguese Veterinary Direction—DGAV, Ministerial Directive 113/2013). DGAV approved the animal experimentation presented in this manuscript under the license number 0421/000/000/2013.
The study with human Peripheral Blood Mononuclear Cells (PBMCs) was approved by the Hospital de Fuenlabrada (Madrid, Spain) Ethics and Research Committee (protocols APR12-65 and APR14-64), and all participants gave written informed consent to be involved.
Four different virosomal preparations have been specifically designed for this study, three of them containing each of the individual antigens, and one containing all the three antigens together. Briefly, a solution containing 1 mg of inactivated Influenza virus A/H1N1/California was pelleted at 286 000g for 1 hour, dissolved in presence of 0.5 ml of PBS containing 0.1 M of Octaethyleneglycol mono (n-dodecyl) ether (OEG; Sigma Aldrich, MO, USA), and then mixed with 32 mg of phosphatidylcholine (Lipoid Ag, Steinhausen, Switzerland) dissolved in 1.5 ml of PBS-OEG 0.1 M. The mixture was centrifuged at 100 000g for 30 min and the supernatant containing Haemagglutinin and Neuraminidase was recovered. For the individual virosomal formulations, the obtained supernatant was then mixed with 2 mg of Leish-F3 (or Leish-F3+), or KMP11 or LJL143 in presence of detergent. Virosomes were then formed by detergent removal and sterile-filtered. The virosome particles containing the mixture of the three antigens were produced similarly, from 1 mg of starting influenza protein mixed with 1 mg of each antigen (Leish F3+, KMP11 and LJL143). Size determination and distribution of the particle population was performed using a Zetasizer Nano instrument (Malvern Instruments, Malvern, UK). Parasite derived and/or VD proteins content in virosome particle was determined by SDS-PAGE Coomassie Stained.
Subjects included in this study were residents of a L. infantum post-outbreak area. Up to 14 healthy endemic individuals (theoretically never exposed to Leishmania), 11 asymptomatic subjects (positivity to the in vitro PBMC proliferation assay to soluble Leishmania antigen) and 21 cured VL patients (clinically diagnosed with VL; presence of Leishmania confirmed in blood by PCR; three months after successful treatment with liposomal amphotericin B) were included in the antigenicity assays. Blood samples were collected at the hospital blood bank and the internal medicine department (Hospital of Fuenlabrada, Madrid). PBMCs were prepared by density gradient centrifugation of heparinized blood samples (Lymphocyte Isolation Solution, RAFER, Spain). PBMCs were adjusted up to 2×106 cells/ml in complete medium (RPMI 1640 supplemented with 100 U/ml penicillin, 100 μg/ml streptomycin, 2mM L-glutamine, 25mM HEPES and 10% heat inactivated fetal calf serum), and cultured in 96-well plates at a density of 2×105 cells per well for 5 days with either KMP11 (10 μg/ml), LeishF3 (10 μg/ml), LeishF3+ (10 μg/ml), LJL143 (10 μg/ml), soluble leishmanial antigen—SLA (10 μg/ml) or PHA-M (5 μg/ml) in a final volume of 200 μl per well. The supernatants of the in vitro cell cultures were collected and stored at -20°C for cytokine quantification. Interferon-γ, granzyme B, TNF-α, and IL-10, were quantified in culture supernatants, using the BD Cytometric Bead Array Human Flex Set (BD Biosciences, NJ, USA) following the manufacturer’s instructions. Data were acquired using a FACSCalibur flow cytometer and analyzed using the Flow Cytometric Analysis Program Array (BD Biosciences, NJ, USA).
Six to eight weeks old male BALB/c mice (Charles River Laboratories, France) were maintained under specific-pathogen free conditions at the IBMC facilities, with water and food ad libitum.
Animals were immunized intramuscularly with a maximum of 50 μl of the respective formulation in the thigh. The volumes administered were based on the concentration of the antigen/adjuvant preparations, and adjusted to equivalent final volumes with PBS. Unless otherwise stated, BALB/c mice were immunized three times at four weeks intervals, in the two thighs alternately. Four weeks after the last immunization, animals were euthanized by cervical dislocation, under volatile anesthesia (Isoflurane, Piramal healthcare, Northumberland, UK).
Two major in vivo experimental set ups originated this work, the first one for optimization purposes, and the second one as the actual pre-clinical trial.
Blood from mice was collected through intracardiac puncture under isoflurane anesthesia. One hundred μl of blood were immediately dispensed to a pre-heparinized tube to be used for general hematological determinations. The remaining volume was left to clot, and serum was then collected and stored at -80°C for posterior immunoglobulin titration.
Mice were disinfected using 70% ethanol. Thereafter, abdominal skin was cut with sterile scissors and removed to expose the abdomen. Peritonea were then opened using a new pair of sterile scissors and tweezers, and the spleens harvested to pre-weighed 15 mL falcon tubes containing 5 ml of complete RPMI (Lonza, Switzerland) [10% heat-inactivated FBS (Lonza, Switzerland), 2 mM L-glutamine, 100 U/ml penicillin and 100 μg/ml streptomycin (BioWhittaker, Walkersville, MD, USA)] supplemented with 50 μM 2-mercaptoethanol (Sigma-Aldrich, MO, USA). Falcon tubes were re-weighed to obtain spleen masses. Splenic single cell suspensions were then obtained using FalconTM Cell Strainers (Fisher Scientific, MA, USA), and their concentrations determined using an EVETM automatic cell counter (NanoEntek, Seoul, Korea).
Ten million cells per animal were pelleted, washed twice with PBS and stained for 10 minutes at 37°C with CFSE (1 μM; in 1 ml of PBS; Thermo Fisher Scientific, MA, USA). Complete RPMI was then added to cell suspensions to stop the reaction, that were then pelleted, re-suspended in complete RPMI and incubated at 4°C for 5 minutes. Afterwards, cells were once more centrifuged (5 min, 350 g), resuspended in complete RPMI supplemented with 50 μM 2-mercaptoethanol (Sigma-Aldrich, MO, USA), and plated into u-bottom 96-well plates at the final amount of 2.5 x 105 cells per well. After plating, different stimuli were added, depending on the experiment: individual non-formulated antigens (KMP11, LJL143 or LeishF3(+); 10 μg/ml), a pool of the three recombinant antigens (KMP11+LJL143+LeishF3+; 10 μg/ml each), and total Leishmania antigen (TLA; equivalent to 10 parasites per cell). Concanavalin A (3 μg/ml) and complete RPMI medium were added as positive and negative controls, respectively. From each animal, a single CFSE staining was performed, being only then the cells divided and stimulated. This warrants that any proliferating cells, regardless the condition, comes from the same initial suspension. Cells were incubated (37°C and 5% CO2) for three (positive controls) or four days (remaining stimuli), and then the originated cell culture supernatants were collected and stored at -80°C for cytokine quantification. Each determination was performed in duplicate. Proliferating CD4+ and CD8+ cell populations were determined by Flow Cytometry, based on the premise that CFSE intensity gradually decreases after each cell division.
The anti-mouse monoclonal antibodies used to perform this study were all purchased from BioLegend (CA, USA) unless otherwise stated: FITC labeled anti-MHC-II(I-Ad) (AMS-32.1, BD Biosciences, NJ, USA), anti-IFN-γ (XMG1.2) and anti-IL-17A (TC11-18H10.1); PE labeled anti-CD8 (53–6.7, BD), anti-Siglec-F (E50-2440, BD), anti-IL-4 (11B11) and anti-IL-6 (MP5-20F3); PerCP-Cy5.5 labeled anti-Ly6C (HK1.4) and anti-TNFα (MP6-XT22); PE-Cy7 labeled anti-CD3 (HA2) and anti-CD11b (M1/70); APC-Cy7 labeled anti-CD11c (N418); APC labeled anti-CD19 (6D5), anti-IL-5 (TRFK5) and anti-IL-10 (JES5-16E3); BV510 labeled anti-CD4 (RM4-5) and Pacific BlueTM labeled anti-Ly6G (1A8).
To analyze lymphoid and myeloid cell populations, two panels of antibodies were designed. The lymphoid panel was composed of anti-CD8, -CD3, -CD4, and -CD19. The Myeloid panel comprised anti-CD11b, -CD11c, -Siglec-F, -Ly6C, -Ly6G and -MHC-II. Surface staining of splenic cells was performed in PBS + 0.5% BSA (20 min, 4°C) followed by 15 min fixation with 2% PFA. For intracellular staining (non-specific cytokine production), splenocytes were cultured for 2h with PMA/Ionomycin (50/500 ng/ml) and then Brefeldin A (10μg/mL) was added for 2 additional hours. Cells were surface stained, fixed and permeabilized with 1% saponin (Sigma-Aldrich, MO, USA) and then intracellularly stained [33]. Samples were acquired in a FACSCanto (BD) and analyzed with FlowJo software v10 (TreeStar, OR, USA).
An initial gate plotting FSC-A versus SSC-A was performed. Afterwards, singlets were selected by plotting FSC-A versus FSC-H and the remaining cell populations were resolved. T lymphoid cell populations were defined as CD3+/CD4+ and CD3+/CD8+ while B cells were defined as CD19+. Non-specific cytokine production by T cells was assessed within CD3+/CD4+ and CD3+/CD8+ cells. Myeloid cell populations were gated as eosinophils (Siglec-F+/SSC-Hint/high), neutrophils (CD11bhigh/Ly6Ghigh/Siglec-F-), DCs (CD11c+/MHC-IIint/high) and monocytes/macrophages (CD11b+/CD11c-/Ly6G-/Siglec-F-). Proliferating T cells (CD4+ or CD8+) were defined as CFSEint/low/neg (FITC channel), always comparing each condition with the respective negative control.
Cytokines were quantified, according to the manufacturer, using the commercial kits: Mouse IL-10 DuoSet ELISA, (R&D Systems, MN, USA), IL-12p70, IL-4 and IFN-γ ELISA MAX Deluxe (BioLegend, CA, USA).
Uncoagulated murine blood samples were used to obtain a complete blood evaluation, including haemoglobin and hematocrit levels, and total red blood cell, white blood cell, and platelet counts, using an automated blood cell counter (Sysmex K1000, Hamburg, Germany).
For specific immunoglobulin titration assays, high protein binding 96-well plates were coated overnight at 4°C, individually with each one of the three antigens comprising the vaccine formulation (1 μg/ml), with a pool of the three antigens (1 μg/ml each), or with soluble Leishmania antigen (SLA; 1 μg/ml); all solutions were prepared in NaHCO3 0.1 M. Additionally, total IgG levels were also determined, using as a coating agent α-mouse IgG (1 μg/ml; Southern Biotech, AL, USA). Plates were then washed with PBS Tween 0.1%, blocked with 1% gelatin in PBS (blocking buffer) for 1 hour at 37°C and re-washed. Each serum was then serially diluted (twofold, 7 dilutions) in blocking buffer. Wells filled with just blocking buffer were used as blanks. Plates were incubated for 1 hour at 37°C and re-washed. Afterwards IgG and isotypes, IgM and IgE were detected using horseradish peroxidase (HRP) coupled α-mouse antibodies [diluted 1:5000 (IgM, IgE, IgG1, IgG3, IgG2b and IgG2a; Southern Biotech, AL, USA) or 1:8000 (IgG; Southern Biotech, AL, USA) in blocking buffer; incubated for 30 minutes, at 37°C]. The plates were washed for a last time, and the substrate (orthophenyldiamine (OPD) in citrate buffer) was added for 10 minutes, time after which the reaction was stopped with HCl 3 N. Absorbance values were determined at 492 nm in a SynergyTM 2 Multi-Mode Reader (BioTek instruments, VT, USA).
The value of the last dilution factor for which the corrected optical density was equal or higher than 0.1 was the defined titer of the antibody (endpoint titer), as has been previously described [34].
Results are generally expressed per individual animals/samples, with a representation of the group mean value ± standard deviation. Statistical differences were analyzed using GraphPad Prism v6.01 (CA, USA). Mice experimental groups were compared using either the one-way ANOVA or the unpaired t-test. Comparisons between human samples were performed using Mann–Whitney test.
Different experimental groups were designed for the pre-clinical tests of the optimized vaccine candidate in mice to assess, besides the safety of the vaccine components, the influence of several variables in the final outcome of the immunization, such as the contribution of the virosome to the induction of immunogenicity or the possibility of immunodominance of the sand fly salivary protein (S2 Fig). In parallel, the effect of a prime with the sand fly-salivary protein in the final vaccine-elicited immune response was evaluated (S2 Fig). Defined above as essential to increase the antigens immunogenicity, the adjuvant (GLA-SE) was administered to all groups. Human immunogenicity studies dictated the replacement of LeishF3 by LeishF3+ as one of the three antigens of the optimized-vaccine.
Because in Leishmania spp. endemic areas the majority of infected persons do not develop clinical symptoms and previous infection leads to robust immunity against the parasite, vaccination is considered as one of the most viable ways to control Leishmania infection. However, to date there is no anti-Leishmania vaccine available for humans [3, 8]. This work proposes an innovative vaccine concept, consisting on Virus-Like Particles (VLP) loaded with 3 different antigens, two from the parasite and one from the sand fly vector, adjuvanted with a TLR4 agonist, as a strong candidate to fill in the existing gap in terms of human anti-Leishmania vaccines.
Although already demonstrated as a useful adjuvant in the context of anti-Leishmania vaccination, we considered it essential to determine the effect of GLA-SE on vaccine-elicited immune responses, mainly because the vaccine candidate we propose is much more complex than the one previously tested (single recombinant fusion protein) [25], with a multi-antigen nature and a virosomal component, which may itself have an adjuvant effect [37]. As expected, the adjuvant generally improved the antigen-elicited immune response, both in terms of specific cellular and humoral responses elicited by non-formulated antigens (S1i and S1ii Fig; PA versus P). Furthermore, similar results were obtained for virosome-formulated proteins (Fig 1i and 1ii; VPA versus PA) indicating, on one hand, the essentiality of the adjuvant in this vaccination context, and on the other that the Influenza VLP are working mainly as vehicles, and not as adjuvants in this context.
For almost two decades in vaccinology, the effect of the antigen dosage in the final outcome of the immunization has been studied and discussed, always in parallel with the concept of antigen affinity [38, 39]. Here, in order to define the optimal vaccine composition, based on the specific responses elicited, we tested two doses of antigens and adjuvant. A lower antigen/adjuvant dose, although is worse regarding the humoral immune response elicited (Fig 1ii, S1ii Fig), promotes a stronger specific cellular immune response against both LJL143 and LeishF3 (Fig 1i, S1i Fig). These results point therefore to the idea that “less is more”, once it is generally accepted that cellular immunity is essential for Leishmania elimination [8], and justify the choice of the lower antigen/adjuvant dosages used in the pre-clinical trials per se.
In agreement with previous observations [40–42], we demonstrated the immunogenicity of KMP11 in humans. In fact, KMP11 was the antigen that generated a better response in the VL patients PBMCs stimulation experiments, with a significant increase in IFN-γ production by cells collected from cured VL patients compared with cells from matching endemic controls (Fig 2i). Such an observation was paramount to the final decision to include this particular antigen as a component of the innovative vaccine candidate. A possible explanation for the weak KMP11 immunogenicity detected in mice, which contrasts with previous studies in animals, is the use of the recombinant protein in opposition with the use of different DNA-based or heterologous recombinant live-vaccine approaches [21, 43–45].
On the other hand, the non-expressive response obtained in human ex vivo immunogenicity studies against LeishF3 (results generally similar between VL patients and controls; Fig 2i) was unexpected. A previous study showed the individual immunogenicity of NH and SMT, the two components of the fusion protein LeishF3, and successfully defined it as immunogenic and safe in a Phase I human clinical trial [25]. However, while the ex vivo immunogenicity assessment done by Coler and colleagues [25] was performed in a cohort from a L. donovani endemic area in Bangladesh, ours was performed using a cohort from a L. infantum endemic area in Spain, which may explain the lower-than-expected reactivity detected. These results led to the characterization of the immunogenicity of a LeishF3 “upgraded version” named LeishF3+ using the same human cohort, and the final substitution of LeishF3 by LeishF3+ in the vaccine formulation due to the observed improvement of the detected responses (Fig 2ii).
Interestingly, PBMCs from some individuals of the three different studied groups, including the controls (Old World human samples) responded to LJL143 (Fig 2i), a salivary protein from Lutzomyia longipalpis, the vector of VL in the New World. The sand fly salivary Lufaxin-like proteins are found in both the New and Old Worlds sand flies [46]. Within this family, LJL143 from L. longipalpis and PpeSP06, the homologous salivary protein from P. perniciosus, the main vector of VL in the Mediterranean Basin, share an amino acid sequence conservation of 45%. In line with this evidence, the reactivity, equally detected in samples from infected and non-infected individuals, is a potential indicator of immune cross-recognition of LJL143, with which in theory, the studied population has not been in contact before. These results further support the inclusion of this antigen in the vaccine formulation, stressing the sand fly salivary protein LJL143 as a potential “broad-spectrum antigen”.
All the above mentioned justifies the final composition of the optimized innovative vaccine used in the definitive pre-clinical trials in mice for extrapolation of its safety profile and characterization its in-depth immunogenic profile. Furthermore, several variables in the final outcome of the immunization, such as the contribution of the virosome to the induction of immunogenicity, or the possibility of immunodominance of the sand fly salivary protein, were considered.
The values of hematological studies, splenic cell populations, CD4+ T cell non-specific reactivity and IgE specific titers (Table 1, S3i–S3iv Fig), determined in the pre-clinical trials, potentially indicate the safety of each vaccine component (proteins, adjuvant and virosome). The absence of specific IgE titers deserves to be highlighted, due to the correlation of antigen-specific IgE and vaccine-associated anaphylatic reactions development, shown particularly, but not exclusively for anti-Influenza vaccines [47]. To further explore the safety of the vaccine candidate, a repeated dose toxicity study in rabbits, complying with the WHO Expert Committee on Biological Standardization [48] is ongoing.
The different optimized formulations tested are indeed immunogenic, eliciting overall significant specific humoral and cellular immune responses (Figs 3i–3iv and 4i and 4ii). Regarding the humoral responses detected, they were generally mixed in nature (IgG1/IgG2a), indicating a mixed Th1/Th2 phenotype. Furthermore, the improvement of the specific humoral response against LeishF3+ induced by the VLP-based antigen formulations (Fig 3iii) deserves to be highlighted, as a possible advantage of the use of formulated antigens without forgetting, however, the debatable relevance of the humoral immune responses in the context of VL [49]. In respect to cellular immune responses detected, they shown distinct magnitudes, depending on each of the individual antigens. Reproducibly, the sand fly-derived antigen induced a more robust response than the parasite-derived ones (LJL143 ≥ LeishF3+ > KMP11; Fig 4ii). Of note, the responses obtained against LeishF3+ were higher than those previously obtained against LeishF3 (Fig 4ii versus S1i Fig; PA (1+1+1) versus PA). This difference in the magnitude of the responses detected against the three different antigens seems not to be an immunodominance problem, since both cellular and humoral immune responses detected against LeishF3+ (the parasite derived antigen showing significant responses) were similar for the groups that received VPA (1+1+1) and VPA (1+5+5) (doses of LJL143, KMP11 and LeishF3+, respectively; Figs 3iii and 4ii).
Although it is a dogma that the protection against Leishmania spp. requires antigen-specific CD4+ and CD8+ T cell responses, the correlates of immunity to human VL are yet to be completely understood [50]. Therefore, the vaccine correlates of protection are still a debatable issue that takes bigger proportions when we add the translatability of animal pre-clinical trials to the equation. This said, the balance between specific IFN-γ and IL-10 production, has been used as predictive of vaccine efficacy in mice [51, 52]. In our study, through cell proliferation assays we detected in non-primed groups, a higher production of IL-10 than IFN-γ in response to the pool of antigens or to LeishF3+ alone, an either comparable or prevalent IFN-γ over IL-10 response against LJL143 (depending on the experimental group; Fig 5i and 5ii) and a limited but prevalent IFN-γ over IL-10 response against total parasite antigens (TLA; Fig 6ii). The Th1 directed response induced by stimulation with TLA, the experimentally closest experimental set up to the infectious process (deposition of whole parasites in the skin) is a promising indication of vaccine effectiveness. Nevertheless, we cannot ignore the apparent main Th2 response induced by the pool of antigens and LeishF3+, and either mixed or Th1 responses induced by LJL143. One curious observation is that, while the IL-10 levels quantified in the cell proliferation against the pool of antigens are 1.5 fold higher than the sum of levels determined in the cell proliferation against the individual antigens (excluding KMP11), the same comparison gives similar IFN-γ (Fig 5i and 5ii). This particular observation, together with evidence showing that stimulation with high antigen doses leads to enhanced IL-10 production by Th1 CD4+ cells [53], makes us speculate on the occurrence of a possible regulatory mechanism in vitro as a way to control a vigorous immune response and prevent inflammation-mediated damage.
In parallel, in the pre-clinical trial, we evaluated the effect of priming with the sand fly salivary protein in the final vaccine-elicited immune responses. Interestingly, the previous administration of the sand fly saliva derived antigen may be beneficial for the generation of a better response against the parasite-derived antigens, particularly in terms of cellular immunity (higher CD4+ T cell proliferation against LeishF3+, KMP11 and TLA; Figs 4ii and 6i). Nevertheless, the IL-10 response detected, particularly against the pool of antigens and LJL143, was higher in the primed animals (Fig 5i and 5ii). On the other hand, the specific IFN-γ response generated by TLA increased tendentiously comparing primed with non-primed animals, and was 5 fold higher than the TLA induced IL-10 response (Fig 6ii), making this vaccination approach interesting to be tested in terms of anti-Leishmania effectiveness, in the context of natural infection (parasites delivered by the sand fly in the presence of salivary proteins).
Overall our results indicate that the innovative vaccine candidate tested here represents a promising anti-Leishmania vaccine. Some questions remain that need to be further explored, such as the potential benefits or implications of the predicted constant vector exposure in endemic countries, as well as a probable exposure to Influenza virus, to the final vaccine-induced responses.
|
10.1371/journal.pcbi.1000264 | Developmental Robustness by Obligate Interaction of Class B Floral Homeotic Genes and Proteins | DEF-like and GLO-like class B floral homeotic genes encode closely related MADS-domain transcription factors that act as developmental switches involved in specifying the identity of petals and stamens during flower development. Class B gene function requires transcriptional upregulation by an autoregulatory loop that depends on obligate heterodimerization of DEF-like and GLO-like proteins. Because switch-like behavior of gene expression can be displayed by single genes already, the functional relevance of this complex circuitry has remained enigmatic. On the basis of a stochastic in silico model of class B gene and protein interactions, we suggest that obligate heterodimerization of class B floral homeotic proteins is not simply the result of neutral drift but enhanced the robustness of cell-fate organ identity decisions in the presence of stochastic noise. This finding strongly corroborates the view that the appearance of this regulatory mechanism during angiosperm phylogeny led to a canalization of flower development and evolution.
| The development of organs, their position, and boundaries in multicellular organisms are defined by genes that can sustain their own activation over long periods of time, termed genetic switches. A good case in point is provided by the genetic machinery controlling the development of flowers in higher plants. In Arabidopsis thaliana and other plants, a particular class of these genes—DEF-like and GLO-like floral homeotic genes—regulates the development of petals and stamens. These genes are self-activating via a heterodimer of their protein products, making the activity of each one of them fully bound to the activity of the other one. The reason for their total functional interdependence has long remained unclear, as the expression of both genes is jointly controlled by shared transcription factors in addition to the heterodimer. In principle, one gene alone could provide their switching functionality. In this study, we use computer modeling to show that the obligate heterodimerization mechanism found in DEF- and GLO-like genes reduces the susceptibility of the genetic switch to failure caused by stochastic noise. This would have provided the system an evolutionary advantage over a single gene with the same functionality.
| Depending on the nature of the interactions of their constituents, gene regulatory circuits can display a variety of dynamical behaviors ranging from simple steady states, to switching and multistability, to oscillations. Temporal or spatial patterning during development requires activation of genes at a particular time or position, respectively, and the inhibition in the remaining time or part. Regulatory genes involved in such processes often show a switch-like temporal or spatial dynamics, which requires a direct or indirect positive non-linear feedback of the genes on their own expression, e.g. via dimers of their own product [1]. Switch-like behavior can be displayed by a single gene [2],[3], but many gene regulatory switches have a more complex structure. Due to the small number of molecules involved, these switches are inherently stochastic and their behavior under noisy conditions can strongly depend on their genetic architecture [4]–[6]. In some cases the complex regulatory interactions have been quite well documented, but the functional implications of the corresponding regulatory circuitry have remained enigmatic. A good case in point is provided by some floral homeotic (or organ identity) genes from model plants such as Arabidopsis thaliana (thale cress; henceforth termed Arabidopsis) and Antirrhinum majus (snapdragon; henceforth called Antirrhinum).
Floral homeotic genes act as developmental switches involved in specifying organ identity during flower development. According to the ‘ABC model’, three classes of floral organ identity (or homeotic) genes act in a combinatorial way to specify the identity of four types of floral organs, with class A genes specifying sepals in the first floral whorl, A+B petals in the second whorl, B+C stamens (male reproductive organs) in the third whorl, and C alone carpels (female organs) in the fourth floral whorl [7]. The combinatorial genetic interaction of floral homeotic genes may involve the formation of multimeric transcription factor complexes that also include class E (or SEPALLATA) proteins, as outlined by the ‘floral quartet’ model [8].
In Antirrhinum, there are two different class B genes termed DEFICIENS (DEF) and GLOBOSA (GLO). In Arabidopsis these genes are represented by APETALA3 (AP3), the putative orthologue of DEF, and PISTILLATA (PI), the putative GLO orthologue. For simplicity, we will refer to DEF-like and GLO-like genes from here on. DEF-like and GLO-like genes represent paralogous gene clades that originated by the duplication of a class B gene precursor 200–300 million years ago [9],[10]. All class B genes identified so far, like most other floral homeotic genes, belong to the family of MADS-box genes, encoding MADS-domain transcription factors [11],[12].
Mutant phenotypes reveal that DEF-like and GLO-like genes are essential for the development of petals and stamens, since def and glo loss-of-function mutants all produce flowers with petals converted into sepals and stamens transformed into carpels [13]–[17]. When co-expressed in the context of a flower, DEF and GLO are not only required, but even sufficient for specifying petal and stamen identity, as revealed by transgenic studies (e.g., [18]).
Induction and stable maintenance of switch-gene expression are typically two independent processes, depending on a transient external signal and autoregulation, respectively [19]. Whenever a transient activating signal is above a threshold, the gene activity switches from the OFF- to the ON-state. The signal is required only for initiation, but not for maintenance of gene activity. Due to the autoregulation, the gene's response becomes in a wide range independent of the exact strength of the input signal. During later stages of flower development (in Arabidopsis from stage 5 on), mRNA of DEF- and GLO-like genes is detected only in whorls 2 and 3 [15],[16]. This is so because upregulation and maintenance of class B gene expression in Arabidopsis and Antirrhinum during later stages of flower development depends on both DEF and GLO, due to an autoregulatory loop involving these proteins (Figure 1C). The proteins encoded by class B genes of Arabidopsis and Antirrhinum are stable and functional in the cell only as heterodimers, i.e., DEF-GLO complexes, because both nuclear localization and sequence-specific DNA-binding depend on obligate heterodimerization [19],[20]. Class B protein heterodimers bind to specific cis-regulatory DNA sequence elements termed ‘CArG-boxes’ (consensus 5′-CC(A/T)6GG-3′). Except PI, the promoter regions of all class B genes of Arabidopsis and Antirrhinum contain CArG-boxes that are involved in positively regulating class B gene expression [21]–[23]. These data, together with the total functional interdependence of the two class B gene paralogues, strongly corroborate the hypothesis that positive autoregulatory control of class B genes involves heterodimers of class B proteins that bind to CArG-boxes in the promoters of class B genes (Figure 1C) [14]. Since PI lacks CArG-boxes in a minimal promoter region, the autoregulatory feedback may work indirectly in this case [23],[24].
Obligate heterodimerization of their encoded products involved in positive autoregulation explains why DEF-like and GLO-like genes are functionally non-redundant and totally interdependent. This raises the question as to how and why such a regulatory system originated in evolution. Studies on the interaction of class B protein orthologues from diverse gymnosperms and angiosperms suggested that, following a gene duplication within the class B gene clade, obligate heterodimerization evolved in two steps from homodimerization via facultative heterodimerization [25]. Meanwhile obligate heterodimerization of DEF-like with GLO-like proteins has also been observed outside of the eudicots Arabidopsis and Antirrhinum in diverse groups of monocots, suggesting that it originated quite early or several times independently during angiosperm evolution [26].
So why then did obligate heterodimerization evolve? In principle, it could represent a neutral change in protein-protein interactions that occurred by random genetic drift [25]. This cannot be excluded at the moment, but for several reasons, it appears not very likely. Even though obligate heterodimerization originated early or several times independently within class B proteins, it did not occur in any other class of floral homeotic proteins, suggesting some kind of functional specificity. Moreover, it occurs within evolutionary especially ‘successful’ (e.g., species-rich) groups of angiosperms, suggesting that it might provide some selective advantage.
Winter et al. [25] suggested that obligate heterodimerization in combination with autoregulation may have provided a selective advantage because of the fixation of class B gene expression patterns and thus the spatial domain of the floral homeotic B-function within the flower during evolution. Mutational changes in the promoter region of only one class B gene that expand the gene's expression domain may leave the late and functionally especially relevant expression domain of the class B genes unchanged, because expression of the other partner would be missing in the ectopic expression domain. Only parallel changes in both types of class B genes, which are much less likely than changes in single genes, could lead to ectopic expression of the B-function under the assumption of obligate heterodimerization and strong autoregulation. Thus obligate heterodimerization may have evolved in parallel, or even as a prerequisite, of the canalization of floral development and thus standardization of floral structure in some groups of flowering plants [25].
Amending this ‘evolutionary’ explanation of obligate heterodimerization, we put forward and test a set of stochastic in silico models of class B gene and protein interactions as shown in Figure 1, thus testing the hypothesis that obligate heterodimerization also provides advantages during development by providing robustness against wrong cell-fate decisions caused by stochastic noise. The models enabled us to study the influence of noise in isolation from other factors, and allowed the comparison of three major stages in the envisioned path of evolutionary transitions (Figure 1): (A) One ancestral gene positively regulates its transcription via a homodimer of its own gene product; (B) Two genes positively regulate their transcription via homo- and heterodimers of both types of products; this very likely represents the situation directly after duplication of the ancestral gene; (C) Obligate heterodimerization of the two products for regulation, i.e., the situation in extant Arabidopsis and Antirrhinum. Since only a small number of individual transcription factors is actually in the nucleus at any time [24],[25], stochastic fluctuations play a large role in the behavior of gene regulatory circuits, and may have an influence on their evolutionary dynamics [5],[27].
Each model consists of a set of reactions for transcription factor binding, transcription, dimerization, and decay (Table S1), where translation is modeled in one step together with dimerization for efficiency (details in Methods section). In turn, each reaction is associated with a propensity function (Tables S2 and S3), which yields the probability of an occurrence of that reaction in a time step. Using the Gillespie algorithm [28], the exact order and timing of reactions is then stochastically determined, based on the propensities. To model transient activation of the circuits, we simulate an inflow of activating molecules (summarizing all different activating transcription factors other than DEF/GLO that act on the respective genes) over 50 minutes of simulated time. After this time, the inflow is switched off and the system equilibrates, i.e., reaches a state in which no change occurs except for stochastic fluctuations (always reached after 72 hours of simulated time). If at this point gene product dimers are still present, the circuit is considered as active (full expression), otherwise it is inactive (no expression of class B genes). Linear stability analysis of the corresponding differential equation system reveals that both the active and the inactive state constitute stable fixed points in all three systems, with an unstable fixed point in between (data not shown).
The activation of the DEF and GLO genes depends on a temporally limited concerted action of many more genes and proteins besides the class B genes themselves, which have been described from an evo-devo perspective [12] and by mathematical modeling [29]. To keep the focus on the self-regulation of the genetic switch, we summarize these in one common or two distinct activators for both genes, respectively. In the first experiment we used a common regulator to temporally activate both genes, and investigated the switching behavior of the three circuits with regard to the number of available activatory input molecules. Looking at the probability of reaching full expression (Figure 2A), the most probable state in the one-gene circuit switches from no steady-state expression (resulting in a non-class B cell identity) to full expression (class B, i.e., petal or stamen cell) at approximately 10 input molecules. Gene duplication without further mutational changes leads to a 3 times lower switching threshold (Figure 2A), which may entail a drastically increased zone of class B gene expression in the flower. Mutations leading to obligate heterodimerization again increase the activation threshold to the previous level, thus restoring the class B gene expression region (Figure 2A). Therefore, in contrast to the facultative heterodimerization circuit, obligate heterodimerization results in the same switching threshold and thus the same domain of expression as just one autoregulatory gene. This result is in contradiction to an intuitive expectation that two genes can produce twice as many dimers as a single gene. With obligate heterodimerization, however, the heterodimers assemble from translated products of one DEF and one GLO mRNA intermediate, while the homodimer in the one-gene system is produced from two translated proteins of the same type. Because mRNA is not used up in translation, this leads to equal production rates for the heterodimer in the obligate heterodimerization system and the homodimer in the one-gene system.
To look at the robustness of the switching decision against stochastic noise, we calculated the decision uncertainty (binary entropy), thus more uncertainty implies less robustness. Focusing on the two circuits with identical expression domains, this uncertainty is nearly equal in the first and third circuit for small numbers of activatory input molecules, until the peak of uncertainty is reached. In contrast, the probability for a decision against class B gene mediated cell identity despite large numbers of activatory input molecules is significantly higher in the one-gene circuit than in the circuit with obligate heterodimerization. With 60 activatory molecules, the probability for such a ‘false negative’ in the former circuit is still 10%, while the latter one achieves nearly 100% correct decisions under our conditions (Figure 2B).
Hence, comparing one autoregulatory class B gene with the circuit after duplication and reduction to obligate heterodimerization, our model suggests that an important difference lies in the response to larger numbers of activatory molecules, where the latter system exhibits a clearly reduced tendency to switch off by mistake. This is explained by the fact that although the circuit needs both DEF-like and GLO-like proteins to sustain activation, its two pools of gene products provide a buffer to temporary stochastic failure of one of the two genes. This is especially important during the initial phase of activation, where circuits that are supposed to lock themselves into permanent expression are susceptible to a run of ‘bad luck’, i.e., the supposedly-active genes are inactive over a longer period of time. Obligate heterodimerization of gene products therefore provides a way to gain robustness against wrong cell identity decisions while retaining the original expression domain of one autoregulatory gene.
Even though the mechanisms of the initial activation of DEF-like and GLO-like genes appear to be quite similar, they are very likely not identical [23], since the initial expression patterns of DEF- and GLO-like genes are slightly different. In Arabidopsis flowers at an early developmental stage 3, AP3 (DEF-like) is expressed in the organ primordia of whorls 2 and 3, but also in parts of whorl 1, while PI (GLO-like) is expressed in whorls 2–4 at the same stage [15],[16]. In contrast, the AP3 orthologue DEF is expressed weakly in the organ primordia of whorl 4 (carpels) and very weakly in those of whorl 1 (sepals), while the PI orthologue GLO is expressed in sepal but not carpel primordia of early stages during Antirrhinum flower development [14],[19]. To investigate the consequences of independent input into both genes, we explored a model setting in which the DEF-like and the GLO-like gene are activated independently by two input signals. Our experiments showed that immediately after gene duplication, the mode of integration represents a logical ‘OR’, meaning that both inputs can independently switch on the circuit (Figure 3A). In this case, each input has the role of the one input present before duplication. After the transition to obligate heterodimerization, a logic ‘AND’ function is achieved (Figure 3B), thus both inputs are needed for activation.
In conclusion, we are providing here, to the best of our knowledge, the first rationale, developmental genetic explanation for the intricate design of a genetic switch controlling class B floral homeotic gene expression in core eudicots, involving obligate heterodimerization and positive autoregulatory feedback of two duplicate genes or their protein products, respectively. The increased robustness against unwanted deactivation by chance found in case of obligate heterodimerization strongly suggests that this mechanism has a distinct advantage when the number of available regulatory molecules is small, leading to less cells of wrong identity in a floral organ and therefore to sharper organ identity transitions. It should be noted that since the mathematical model applies to any system with obligate heterodimerization and positive feedback, the conclusions drawn here also transfer to any such system. However, to the best of our knowledge, the phenomenon of obligate heterodimerization together with positive feedback seems quite rare in genetic regulation outside of flower development, potentially due to the high cost of maintaining this system together with a strong dependence of the predicted fitness gain on external factors that might be specific for the situation depicted here.
In the standard ABC model, class A and C genes are mutually antagonistic [7],[30], while class B genes have no floral homeotic ‘repressor’, possibly explaining the class-specific need for sharpened expression domains and thus obligate heterodimerization, which is not found in the other two gene classes. However, Zhao et al recently reported that the antagonistic expression of class A and class C genes is involved in defining the expression domain of class B genes in Arabidopsis [31], suggesting that our observation may not be sufficient to explain the obligate heterodimerization of class B proteins. Taking a different perspective, the evolution of a regulatory ‘AND’ function out of an ‘OR’ function may have provided the plant with a more stringent control of the class B floral homeotic genes depending on different induction signals. The fact that there must be different inputs into DEF- and GLO-like genes is obvious from gene expression studies (see above), but its functional importance may have escaped the attention of previous investigations because of the coordinate upregulation and functional importance of DEF- and GLO-like genes in the second and third floral whorl. Our results suggest that identifying these different induction pathways, and clarifying their molecular mechanisms (e.g., trans-acting factors and cis-regulatory DNA motifs in DEF-like and GLO-like genes being involved) would enable an important step forward in understanding class B floral homeotic gene function in flowering plants.
The functional implication of these different input signals, and hence also of our hypothesis, could be tested by transgenic experiments. For example, Arabidopsis class B gene mutants in which both the AP3 and the PI gene have been brought under the control of the AP3 or the PI promoter rather than every gene under its own promoter (as in the wild-type) should affect the spatial or temporal development of petals or stamens, or both. Transgenic plants mutated at the pi locus (pi-1) in which wild-type PI is expressed under control of the AP3 promoter (5D3) have already been reported [32]. These plants were used only as control for other experiments and have therefore not been described in much detail concerning the traits of interest here. However, it is clear that the 5D3::PI pi-1/pi-1 plants do not just show petals in the second floral whorl and stamens in the third floral whorl, as wild-type plants do; rather, they frequently develop sepal/petal mosaics in the second whorl, and mosaic organs or even carpels in the third whorl. These observations support our hypothesis concerning the functional importance of different induction pathways controlling the expression of DEF- and GLO-like genes for a proper development of organ identity in whorls two and three. More detailed analyses should be done to better understand how exactly the transgenic plants deviate from wild-type plants, and why. In addition, complementary transgenic studies in which AP3 is expressed under control of the PI gene promoter (pPI) should be performed in order to determine whether the pPI::AP3 ap3/ap3 plants have also developmental defects. The construction of a transgenic plant with switched promoters (i.e., pAP3::PI pPI::AP3 ap3/ap3 pi/pi) would also be of great interest. Due to the apparently symmetric roles of AP3 and PI, one might speculate that this phenotype shows less deviation from the wild type than the transgenic plants with both genes under the control of a single promoter.
If the origin of obligate heterodimerization of class B proteins during evolution provided some plants with selective advantages, one may expect that this had an impact on the molecular evolution of these proteins, which indeed seems to be the case. Class B floral homeotic proteins are MIKC-type MADS-domain proteins characterized by a defined domain structure, including a MADS (M), Intervening (I), Keratin-like (K) and a C-terminal (C) domain [11],[12]. The K-domain mediates heterodimerization of GLO- and DEF-like proteins and has been postulated to fold into three amphipatic α-helices termed K1, K2 and K3 [33]. In accordance with the expectations mentioned above, phylogenetic data indicate that after the duplication leading to DEF-like and GLO-like gene lineages, positive selection acted on the sections of these genes encoding the K-domain [34]. Intriguingly, one site under positive selection [34] is in a subdomain of K1 (“position 97-102” according to ref. [33]) proposed to be critical for heterodimerization specificity of DEF- and GLO-like proteins, as revealed by yeast two-hybrid analyses [33].
Given that the duplicates resulting from one homodimerizing protein would be capable of homo- as well as heterodimerization, our results suggest that positive selection should have enforced the loss of the homodimerization ability, since our model with duplicated class B genes and obligatory heterodimerization implies a sharper switching characteristic and a more constrained domain of class B gene expression than the one with facultative heterodimerization. It has been proposed that within the subdomain of K1 mentioned above, the interaction of Glu-97 in PI and Arg-102 in AP3 facilitates specific heterodimerization between AP3 and PI and prevents formation of homodimers [33]. For these sites, however, positive selection has not been detected [34]. Clearly, the relationships between the molecular evolution and biophysical interactions of DEF- and GLO-like proteins deserve more detailed studies in the future.
All in all, our findings strongly support the view that the unexpected complexity of the floral homeotic gene switch considered here was not simply produced by random genetic drift but evolved because it provided the plant with a clear selective advantage. This might have led to the establishment of this regulatory motif in a whole range of plant species. In line with this notion, it is intriguing that at least some basal angiosperms do not have sharp, but ‘fading borders’ of expression of orthologues of DEF-like and GLO-like genes as well as gradual transitions in organ identity [35]. This underlines the hypothesis [25] that the mechanism described here improves developmental robustness and thus helped to canalize the development and hence also the evolution of flowers within angiosperm evolution.
The model investigated in this work is simulated using the Gillespie algorithm [28], implemented as a C++ function linked to MATLAB (The MathWorks, Inc. 2008). This method, which simulates an exact instance of the stochastic master equation, explicitly accounts for each reaction event and thus represents stochastic effects in full detail. A list of all modelled reactions is given in Table S1, and the full model is shown in Figure S2. Transcription factor binding and unbinding are simple reaction processes, where we assume that exactly one functional copy of both DEF and GLO genes are available. For simplicity, we assume that only activated DNA is transcribed; however, experiments with basal transcription rates have led to qualitatively similar results. The decision to model translation and dimerization in one step was taken to simplify the model while keeping the focus on transcriptional rather than translational regulation. This entails that we only model DEF and GLO mRNA and the dimerized proteins, but not the single DEF and GLO proteins. The slight loss of accuracy here has been unavoidable, as we needed to keep the model computationally tractable for the large numbers of replicated experiments. All constituents of the model decay with a linear rate. For details on all kinetic rate constants, see the Text S1 and Tables S1–S3. We conducted 10,000 experiments for each parameter combination.
The different types of regulation are achieved by enabling or disabling the binding and activation of one type of gene by either a transcription factor homodimer produced by itself, a heterodimer of the products of both genes, or a homodimer of the proteins encoded by the other gene. Concerning initial activation, the class B genes are regulated by a number of (possibly interacting) transcription factors, some of which are still unknown. Since the aim of this contribution is to investigate the effect of autoregulation on gene activity, we summarize the effects of all upstream transcription factors in two specific input factors, IDEF and IGLO, and a common input factor, IC.
As developmental switches, the B-genes are transiently activated by their inputs, which are switched off after activation. Depending on the level of gene activity reached by that time, this activity either stays high or decays to a low value again, corresponding to on- and off-states of the genes. To model the transient activation, an inflow of (on average) N activatory molecules (of type IDEF, IGLO or IC, respectively) over a period of T minutes was simulated. After time T, the inflow is switched off and the system is left alone, reaching steady state. Figure S1 shows example time courses for all three modes of regulation considered here.
All three systems investigated in this work represent autoactivatory circuits, which are used by the plant to establish the expression (ON-state) or non-expression (OFF-state) of homeotic genes in certain floral whorls. Therefore, a decision has to be made, depending on the number of activatory input molecules initially coming into the system. For low numbers of input molecules, the decision should be ‘OFF’, for higher numbers it should be ‘ON’.
To measure the uncertainty of this decision, we use the binary entropy function. Let X be a random variable that takes value 1 with probability p, value 0 with probability 1−p, i.e., a Bernoulli trial. The entropy of X is defined asIn our case, X taking value 1 means that the system reaches ON-state, value 0 means OFF-state. Repeating the simulation 10,000 times, we compute the probability p for each specific number N of activatory input molecules IC (Figure 2A). Using the formula above, this translates to the binary entropy, or decision uncertainty (Figure 2B).
Alternative approaches which could potentially lead to additional insights into the functionality of the DEF-GLO system include the application of control theory [36] or an analytical calculation of the first and second stochastic moments, which should confirm the experimental results in this paper.
|
10.1371/journal.pntd.0003084 | Suppressing Dengue-2 Infection by Chemical Inhibition of Aedes aegypti Host Factors | Dengue virus host factors (DENV HFs) that are essential for the completion of the infection cycle in the mosquito vector and vertebrate host represent potent targets for transmission blocking. Here we investigated whether known mammalian DENV HF inhibitors could influence virus infection in the arthropod vector A. aegypti. We evaluated the potency of bafilomycin (BAF; inhibitor of vacuolar H+-ATPase (vATPase)), mycophenolic acid (MPA; inhibitor of inosine-5′-monophosphate dehydrogenase (IMPDH)), castanospermine (CAS; inhibitor of glucosidase), and deoxynojirimycin (DNJ; inhibitor of glucosidase) in blocking DENV infection of the mosquito midgut, using various treatment methods that included direct injection, ingestion by sugar feeding or blood feeding, and silencing of target genes by RNA interference (RNAi). Injection of BAF (5 µM) and MPA (25 µM) prior to feeding on virus-infected blood inhibited DENV titers in the midgut at 7 days post-infection by 56% and 60%, and in the salivary gland at 14 days post-infection by 90% and 83%, respectively, while treatment of mosquitoes with CAS or DNJ did not affect susceptibility to the virus. Ingestion of BAF and MPA through a sugar meal or together with an infectious blood meal also resulted in various degrees of virus inhibition. RNAi-mediated silencing of several vATPase subunit genes and the IMPDH gene resulted in a reduced DENV infection, thereby indicating that BAF- and MPA-mediated virus inhibition in adult mosquitoes most likely occurred through the inhibition of these DENV HFs. The route and timing of BAF and MPA administration was essential, and treatment after exposure to the virus diminished the antiviral effect of these compounds. Here we provide proof-of-principle that chemical inhibition or RNAi-mediated depletion of the DENV HFs vATPase and IMPDH can be used to suppress DENV infection of adult A. aegypti mosquitoes, which may translate to a reduction in DENV transmission.
| Arboviruses utilize homologous host factors of the mammalian and insect cellular machinery to complete the infection cycle. Studies in both mammalian and insect cell lines have shown that virus infection can be suppressed through inhibition of host factors by chemical compounds that therefore could be developed into transmission blocking agents. However, similar studies have not been conducted in adult mosquitoes. Here we investigated the effect of four chemical compounds (bafilomycin, mycophenolic acid, castanospermine, and deoxynojirimycin), known to inhibit the host factors vacuolar H+-ATPase (vATPase), inosine-5′-monophosphate dehydrogenase (IMPDH) and glucosidases, on dengue virus replication in adult mosquitoes. We found that bafilomycin and mycophenolic acid suppressed dengue virus replication in adult mosquito guts when they were injected prior to dengue virus infection; however, castanospermine and deoxynojirimycin did not. Ingestion of bafilomycin and mycophenolic acid also inhibited virus replication. We showed that the predicted target genes of bafilomycin and mycophenolic acid function as virus host factors in adult mosquitoes through RNAi-mediated gene silencing. Inhibition of vATPase also decreases mosquito longevity and fecundity, thereby further compromising vector capacity. Our study demonstrated that chemical compounds or double stranded RNAs (dsRNA) can be used to suppress virus infection through inhibition of host factors in adult mosquitoes, thereby rendering such approaches interesting for the development of novel transmission-blocking strategies.
| From a global health perspective, dengue virus (DENV) is currently the most important arbovirus transmitted by mosquitoes. Approximately 3.6 billion people are at risk of DENV infection, and 100 million people are infected annually [1]. Given the lack of registered antivirals or vaccines against DENV, a major effort to reduce DENV transmission has been concentrated on mosquito vector control. Although suppression of mosquito populations represents the most widely used dengue control strategy, this approach is hampered by insecticide resistance and the rapid adaptation and expansion of mosquitoes to urban areas [2]. Thus, the development of novel methods to reduce DENV transmission is urgently needed. Here, we investigated a novel transmission-blocking method that targets mosquito proteins (HFs) used by DENV for viral replication and transmission instead of directly targeting the mosquito or DENV for destruction.
DENV incubates in a mosquito for about 14 days before the mosquito is able to transmit the virus to a human host. The virus is ingested by the mosquito through infected blood, from which it infects the insect's midgut epithelial cells. There the virus replicates and then disseminates throughout the mosquito, including the salivary glands, where it further replicates and is then transmitted to a new human host [3]. During this extrinsic incubation period, the mosquito mounts an immune response against the virus that results in suppression of infection to various degrees. Previous studies have shown that the Toll, Janus kinase/signal transducer and activator of transcription (JAK/STAT), and RNA interference pathways control DENV restriction mechanisms in mosquitoes [4]–[6].
Several studies have identified mosquito genes that are essential for arbovirus replication and transmission and can therefore be considered DENV host factors (HFs) [7]–[10]. For example, mosquito prohibitin is a DENV HF that acts as a receptor protein to mediate DENV cell entry [7]. The mosquito and mammalian vacuolar H+-ATPase (vATPase) functions as a DENV HF by acidifying endosomes, a process that is important for viral fusion and the release of the viral genome into the cytoplasm [11], [12]. DENV also utilizes HFs that are involved in de novo pathways for RNA synthesis during viral replication [13]. Host glucosidase has also been shown to act as a DENV HF and is responsible for the proper folding and glycosylation of virus proteins [14], [15]. Other mosquito proteins have also been shown to act as virus agonists, but the mechanisms by which they influence virus infection remains unknown. For example, an A. aegypti cathepsin, an MD2-like protein, and NPC1-like factors have been shown to act as DENV agonists [16]. Thus, DENV HFs and agonists represent potential chemical- and vaccine-based transmission-blocking targets that could be developed into novel mosquito-based dengue control strategies.
In several studies, chemical compounds targeting DENV HFs in mammalian or insect cells have been shown to suppress DENV infection to various degrees [14], [15], [17]–[22]. However, the possible anti-dengue activities of such chemicals have not been studied in adult mosquitoes in order to assess their usefulness for transmission blocking. A chemical approach to inhibiting DENV HFs in the mosquito could circumvent the ecological impact of insecticide use and the probability of the virus's developing resistance to the blocking mechanism.
We investigated the ability of four putative DENV HF-inhibitor compounds, bafilomycin (BAF), mycophenolic acid (MPA), castanospermine (CAS), and deoxynojirimycin (DNJ), to block DENV infection of the mosquito midgut. We administered these compounds by various treatment methods, including injection and ingestion by sugar feeding or blood feeding. BAF is a well-characterized inhibitor of HFs for various viruses in mammalian and insect cell lines [18], [23]–[26]. It acts by interfering with vATPase function, thereby inhibiting the acidification of the endosome, a necessary step for viral entry [10], [27], [28]. MPA also exerts antiviral activity in mammalian and insect cells [13], [17], [29] by inhibiting the enzyme inosine 5′-monophosphate dehydrogenase (IMPDH), which is involved in the de novo pathway of guanosine nucleotide synthesis that is essential for viral RNA synthesis as well as host DNA and RNA synthesis [13]. CAS and DNJ are inhibitors of glucosidases that act as DENV HFs by ensuring the proper folding and glycosylation of viral proteins [14], [15].
This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. Mice were only used for mosquito rearing as a blood source according to approved protocol. The protocol was approved by the Animal Care and Use Committee of the Johns Hopkins University (Permit Number: M006H300). Commercial anonymous human blood was used for dengue virus infection assays in mosquitoes, and informed consent was therefore not applicable. The Johns Hopkins School of Public Health Ethics Committee has approved this protocol. Mosquito collections were performed in residences after owners/residents permission.
Several A. aegypti strains (Rockefeller/UGAL [Rock], Singapore [SIN], and Puerto Triunfo [PTri]) [30] were used to test the function(s) of vATPase subunits by RNA interference (RNAi)-mediated gene silencing. The anti-DENV compounds assays and IMPDH gene silencing assay were performed with our standard lab strain, the Rock strain. The mosquitoes were maintained on a 10% sucrose solution at 27°C and 95% humidity with a 12-hr light/dark cycle [5]. Infected mosquitoes were dissected to collect the midguts at 7 days post-infection or the salivary glands at 14 days post-infection. Each mosquito body was dipped in 70% ethanol and rinsed twice in 1×PBS. The dissection was performed in 1 drop of 1× PBS, and the dissected midguts or salivary glands were transferred to a microcentrifuge tube containing 150 µl of MEM and stored at −80°C until used for virus titration. The C6/36 (A. albopictus) cell line that was used for DENV propagation was grown in minimal essential medium (MEM, Gibco) with 10% heat inactivated FBS, 1% L-glutamine, 1% penicillin-streptomycin, and 1% non-essential amino acids at 32°C with 5% CO2. The BHK-21 (baby hamster kidney) cell line that was used for plaque assays was maintained on Dulbecco's modified Eagle's medium (DMEM, Gibco) supplemented with 10% FBS, 1% L-glutamine, 1% penicillin-streptomycin, and 5 µg/ml plasmocin (Invitrogen) at 37°C and 5% CO2.
DENV serotype 2 (New Guinea C strain, DENV-2) was propagated in the C6/36 cell line. One milliliter of virus stock was used to infect a 75-cm2 flask of C6/36 cells at 80% confluence. Infection was allowed to proceed for 6 days, at which time the cells were harvested, centrifuged at 800 g for 10 min, and mixed 1∶1 with commercial human blood supplemented with 10% human serum and 1% 100 mM ATP (Thermo scientific). The infectious blood meal was maintained at 37°C for 30 min prior to feeding 5- to 7-day old mosquitoes.
Bafilomycin (BAF) was obtained from Cayman Chemical, and mycophenolic acid (MPA), castanospermine (CAS), and deoxynojirimycin (DNJ) were obtained from Sigma Aldrich. Compounds were resuspended in 100% dimethyl sulfoxide (DMSO) and further diluted with PBS to yield a 10% DMSO in 1XPBS solution. Control mosquitoes were injected with the 10% DMSO in 1XPBS solution. Cold-anesthetized 6-day-old female mosquitoes were injected in the thorax and then transferred to paper cups and stored at 27°C overnight for next-day DENV infection.
For ingestion of compounds through the sugar meal, various concentrations of BAF and MPA were mixed with a 10% sucrose solution to yield a 2.5% DMSO solution. Four-day-old females were fed on the solution for 2 days prior to, and every second day until the day of dissection after, ingestion of a DENV-infected blood meal. For ingestion of compounds through the DENV-infected blood meal, various concentrations of BAF and MPA were directly mixed with the blood to yield a 2.5% DMSO solution. Control groups were provided with either sucrose or blood meals containing an equal amount of DMSO.
The function of vATPase subunits and IMPDH as DENV HFs was assayed using RNAi-mediated gene silencing as described previously [31]. In brief, double-stranded RNAs targeting the following genes encoding various vATPase subunits were synthesized from PCR-amplified gene fragments using the HiScribe T7 in vitro transcription kit (NEBioLabs): vATPase subunit ac39 (vATP-ac39, AAEL011025), vATPase proteolipid subunit (vATP-V0B, AAEL012113), vATPase subunit f (vATP-f, AAEL002464), vATPase 16-kDa proteolipid subunit (vATP-16, AAEL000291) and IMPDH (AAEL009273), plus the GFP gene as a control. The primer sequences are listed in Table S1. Approximately 69 nl of dsRNAs (3 µg/µl; 200 ng/mosquito) in water was injected into the thorax of cold-anesthetized 4-day-old female mosquitoes using a nano-injector (Nanoject; Drummond Scientific) with a glass capillary. Three days after injection, mosquitoes were fed on a DENV-2-supplemented blood meal or were sacrificed for gene-silencing efficiency assays. After non-fed mosquitoes were removed, the blood-fed mosquitoes were maintained in the insectary under the conditions mentioned above for 7 days before midgut dissection or for 14 days before salivary gland dissection. Gene silencing was verified 3 days after dsRNA injection by real-time quantitative RT-PCR. Three independent biological replicate assays were performed for each gene, with the A. aegypti ribosomal S7 gene as the internal control for normalization [32]. Mosquito samples were collected in RLT buffer (Qiagen) and then stored at −80°C until extraction. Total RNA was extracted from tissue samples using the RNeasy Mini Kit (Qiagen) according to the manufacturer's protocol. For cDNA synthesis, extracted RNA samples were treated with Turbo DNase (Ambion) at 37°C for 1 h before reverse transcription with a MMLV Reverse Transcriptase kit (Promega) according to the manufacturer's instructions. The cDNA was then used to determine gene expression by qPCR using the SYBR Green PCR Master Mix with gene-specific primers. The primers for each gene are presented in Table S2.
DENV-NGC titers in midguts were determined by plaque assay on BHK-21 cells. Frozen midguts were thawed, homogenized in DMEM with a Bullet Blender (NextAdvance), serially diluted, and then inoculated onto cells seeded to 80% confluence in 24-well plates (100 µl per well). Plates were rocked for 15 min at room temperature, and then incubated for 45 min at 37°C and 5% CO2. Subsequently, 1 ml of DMEM containing 2% FBS and 0.8% methylcellulose were added to each well, and plates were incubated for 5 days at 37°C and 5% CO2. Plates were fixed with a methanol/acetone mixture (1∶1 volume) for at least 1 h at 4°C, and plaque-forming units were visualized by staining with 1% crystal violet solution for 10 min at room temperature.
Longevity, fecundity, and egg hatchability assays were performed after vATPase and IMPDH were silenced with dsRNAs as described previously [33]. For the longevity assay, 20 four-day old Rock strain adult female mosquitoes that had been injected with dsRNAs targeting either GFP, vATPase (V0B), or IMPDH were kept in a wax-lined cardboard cup at 27°C with 70% humidity and maintained on a sterile 10% sucrose solution. Three biological replicates were performed, and all groups were monitored daily for survival until all mosquitoes had perished. The survival percentage represents the mean survival percentage for all three biological replicates as described previously [33]. For the fecundity assay, 30 four-day old GFP dsRNA-, vATPase (V0B) dsRNA-, or IMPDH dsRNA -injected mosquitoes were allowed to feed on human blood through an artificial membrane feeder for 30 min. The fed mosquitoes were transferred to individual 50 ml centrifuge tubes (one mosquito per tube) outfitted with moistened filter paper at the base of the tubes, and incubated under normal rearing conditions. Eggs oviposited on filter paper were counted after 2 days using light microscopy. Female mosquitoes that did not produce eggs on day 2 were maintained and re-examined on day 3. After each count, eggs were submerged in a standard larval pan for rearing according to standard methods. First- to second-instar larvae were counted to determine the larval hatch rate. The fecundity and larval hatch-rate assays were performed in three biological replicates, and the number of eggs laid by each female and the respective hatch rates were used to calculate mean values.
The midgut or salivary gland DENV titers of the control groups and experimental groups were compared using the plaque assay results from at least three biological replicates. Mann-Whitney U-tests and Kruskal-Wallis tests with Dunn's multiple comparison test were used when appropriate. A 2-way ANOVA was performed to analyze the longevity assay data in addition to Kaplan–Meier survival analysis and Wilcoxon tests (Table S3). A Mann-Whitney U-test and a Student's t-test were used to calculate p-values and determine the significance of fecundity and fertility, respectively. Statistical analyses were conducted using the GraphPad Prism statistical software package (Prism 5.05; GraphPad Software, Inc.). Statistical significance is indicated with asterisks: *, p<0.05; **, p<0.01; ***, p<0.001. Descriptive statistics for DENV infection assays are presented in supplementary table S2.
vATPase subunit ac39 (vATP-ac39, AAEL011025), vATPase proteolipid subunit (vATP-V0B, AAEL012113), vATPase subunit f (vATP-f, AAEL002464), vATPase 16-kDa proteolipid subunit (vATP-16, AAEL000291) and IMPDH (AAEL009273).
To assess the anti-dengue function of BAF, we microinjected 5 µM or 25 µM of BAF into the thorax of 4-day-old female mosquitoes 1 day before allowing them to feed on DENV-infected blood. Microinjection of compounds was used in the first screening as the most effective approach to standardize the timing and quantity of injected compounds. As compared to the untreated controls (n = 42), DENV-fed mosquitoes injected with 5 µM or 25 µM BAF showed a significantly suppressed midgut infection, by 56% (n = 49, p = 0.0002) and 54% (n = 24, p = 0.037), respectively, at 7 days after the infectious blood meal (Fig. 1A). While injection of either 5 µM, 25 µM, 125 µM, or 625 µM BAF did not affect mosquito longevity for up to 16 days (Fig. S1A), injection of 25 µM BAF caused a decreased blood-feeding propensity that resulted in a smaller sample number; therefore we did not use BAF concentrations >25 µM for further DENV inhibition assays. Altogether, these results suggest that BAF has anti-DENV activity in adult mosquitoes.
Thoracic microinjection of mosquitoes with 50 µM or 250 µM of MPA 1 day prior to feeding on virus-infected blood reduced DENV titers of the midgut tissue by 60% in both cases (n = 39, p = 0.0003 and n = 32, p = 0.0003) when compared to the untreated controls (n = 54) at 7 days after feeding on the infectious blood meal (Fig. 1B), indicating that MPA functions as a DENV inhibitor in adult mosquitoes as it does in mammalian and insect cells [13], [17]. Injection of the higher 1.25 mM and 6.25 mM concentrations of MPA resulted in an inconsistent blood feeding propensity, and therefore they could not be used for DENV inhibition assays. Injection of 50 µM, 250 µM, 1.25 mM, or 6.25 mM MPA did not affect the mosquito lifespan for up to 16 days after treatment (Fig. S1B).
We also assessed the ability of two known alpha-glucosidase inhibitors, CAS and DNJ, to influence DENV infection of adult females [14]. Injection of adult female mosquitoes with 250 µM CAS (n = 35, p = 0.0792) or DNJ (n = 40, p = 0.8882), or with 1 mM CAS (n = 38, p = 0.6035) or DNJ (n = 36, p = 0.0039) 1 day prior to feeding on virus-laden blood did not suppress the DENV infection of the midgut tissue when compared to control DMSO-treated mosquitoes (n = 36 or 28) at 7 days after the infectious blood meal (Fig. 1D, E). The data suggest that whereas these two compounds suppress mammalian alpha-glucosidase, they may not function effectively against the mosquito alpha-glucosidase ortholog, or else DENV may utilize alternative enzymes in the mosquito. In an earlier study, we found that RNAi-mediated silencing of an A. aegypti alpha-glucosidase ortholog of a D. melanogaster gene, previously shown to function as a DENV HF in a fly cell line, did not influence the adult mosquito's susceptibility to DENV infection in the Rock, Sin, or PTri strain of A. aegypti [30].
We also investigated the effect of BAF and MPA exposure on salivary gland DENV titers at 14 days post-infection. Injection of 5 µM of BAF or 250 µM of MPA one day prior to feeding on DENV infected blood resulted in reduced viral titers in the salivary glands by 90% (n = 55, p = 0.0005) and 83% (n = 63, p = 0.013), respectively, compared to controls (n = 56). This reduction was greater than that observed in midguts after the same compound treatment (BAF: 56%; MPA: 60%) (Fig. 1F). It is likely that further reduction in salivary gland titers would occur if the compound treatments had taken place at 6–10 days after feeding on DENV infected blood since it would likely result in a more efficient HF inactivation in the infected salivary glands.
We wanted to investigate whether injection of a cocktail of BAF and MPA would result in a synergistic anti-DENV effect that would be greater than that of each compound when used independently. Single treatment with 5 µM BAF (n = 35) or 250 µM MPA (n = 39) reduced DENV titers by 60% and 44%, respectively, when compared to controls (n = 42) (Fig. 1C). Surprisingly, the cocktail containing 5 µM BAF and 250 µM MPA reduced DENV titers by only 44% (n = 41) (Fig. 1C), a level comparable to single treatment with 250 µM MPA, suggesting that there may be some drug interaction between the two compounds or that injection of both compounds resulted in a greater mosquito detoxification activity that diminished the active concentration of one or both compounds in DENV-infected cells.
To determine whether the anti-dengue potency is dependent on whether the compound is injected prior to or after exposure to DENV, we injected mosquitoes with BAF or MPA 1 day after ingestion of a DENV-infected blood meal. Mosquitoes were fed with DENV-infected blood, and the following day, they were injected with 5 µM BAF or 50 µM MPA, or with DMSO as a control. There was no significant difference in DENV titers between the control mosquitoes and those injected with either BAF (n = 39, p = 0.3047) or MPA (n = 33, p = 0.245) (Fig. 2), indicating that the effective antiviral action of these compounds requires application prior or simultaneous with DENV infection.
Next, we investigated whether BAF and MPA could mediate anti-DENV activity when ingested with either a sugar meal prior to ingestion of a DENV-infected blood meal or together with a DENV-infected blood-meal. For delivery of compounds via a sugar meal, female mosquitoes were maintained on a 10% sucrose solution supplemented with either 50 µM of BAF or 250 µM of MPA or with DMSO (control group) for 2 days prior to, as well as after, feeding on DENV-infected blood. Ingestion of BAF or MPA via the sugar meal decreased DENV titers by 22% (n = 27, p = 0.0553) and 37% (n = 54, p = 0.0351), respectively, when compared to the controls (Fig. 3A, B).
Similarly, ingestion of 50 µM of BAF or 250 µM of MPA, or of DMSO (control group) via the DENV-infected blood also reduced DENV titers by 22% (n = 28, p = 0.0396) and 50% (n = 30, p = 0.0313), respectively, when compared to the controls (Fig. 3C, D).
BAF has been shown to inhibit viral entry in vitro by binding to the multi-subunit vATPase enzyme complex [10], [18], [24], [26], . To date, one vATPase subunit (vATP-g, AAEL012819) has been experimentally confirmed to influence DENV infection in A. aegypti mosquitoes [30]. To provide more robust evidence for the activity of vATPase as a DENV HF and to determine whether depletion of other vATPase subunits also negatively affects DENV infection of adult mosquitoes, we performed dsRNA-mediated gene silencing on four additional vATPase subunits. dsRNA-mediated silencing of vATP-ac39, vATP-V0B, vATP-f, and vATP-16 reduced DENV titers by 85%, 78%, 76%, and 98%, respectively, in the Rock (laboratory mosquito) strain (p values<0.01) (Fig. 4A,B). Moreover, silencing of vATP-ac39 and vATP-V0B in two DENV-susceptible field-derived strains, SIN and PTri, also significantly reduced their DENV titers by 98% and 82% in SIN, and by 87% and 61% in PTri, respectively (Fig. 4C, D) [30]. MPA has been suggested to block DENV infection in human cells by inhibiting IMPDH1 and IMPDH2 [34], [35]. There is only one ortholog of IMPDH (AAEL009273) in A. aegypti mosquitoes, and its RNAi-mediated silencing resulted in an 80% suppression of DENV infection (p = 0.0001, n = 43) when compared to control GFP dsRNA-injected mosquitoes (n = 53) (Fig. 4E). Our data therefore demonstrate that the targets of BAF and MPA in the A. aegypti mosquito are likely to be vATPase and IMPDH, respectively.
Next, we investigated whether co-silencing of vATPase and IMPDH would result in a greater DENV inhibition than when these genes were silenced independently. Simultaneous knockdown of vATP-V0B and IMPDH reduced DENV titers by 96.1%, while silencing of the individual genes decreased DENV titers by 89% (vATP-V0B) and 60% (IMPDH) (Fig. 5). Our data suggest that the synergistic silencing effect of these two HFs on DENV infection is marginal, likely because of a lesser depletion efficiency of each gene transcript and/or protein when co-silenced, or a possible alternative infection route of a small proportion of DENV.
We also measured salivary gland DENV titers after silencing the target genes through injection of dsRNA at 3 days prior to feeding on DENV containing blood. Silencing of vATP-V0B and IMPDH reduced viral titers by 93% (n = 51, p<0.0001) and 87% (n = 63, p<0.0001), respectively. This reduction was greater, although marginal, than that observed in midguts (vATP-V0B: 78%; IMPDH: 80%) (Fig. 4F). It is likely that further reduction in salivary gland titers would occur if the dsRNA treatments had taken place at 6–10 days after feeding on DENV infected blood since it would likely result in HF gene silencing in the infected salivary glands.
In order to assess possible fitness effects of HF inactivation we investigated the effects of vATPase and IMPDH silencing on mosquito fitness as a measure of longevity, fecundity, and fertility. Silencing of vATP-V0B significantly affected longevity (2 way ANOVA, p<0.0001), fecundity (98% reduction, p<0.0001) and fertility (19% reduction, p = 0.0475) while silencing of IMPDH showed no or only marginal effects on longevity (p = 0.8124), fecundity (14% reduction, p = 0.0141) and fertility (p = 0.3366) compared to control mosquitoes (Fig. 6).
A variety of arthropod-transmitted virus HFs have been discovered, and some of them have potential as virus transmission-blocking targets [14], [15], [36]–[38]. Here we have investigated the potency of the drugs BAF, MPA, CAS, and DNJ, all of which are known to block DENV infection in mammalian cells, in inhibiting infection of the mosquito vector. vATPase plays a fundamental role in virus membrane fusion as a vacuolar proton pump that acidifies the vacuole [11]. When viruses enter the cell by endocytosis, membrane fusion is necessary to release the virions from the endosome to the cytoplasm, and a key factor in membrane fusion is vacuole acidification [12]. Inhibition of vATPases with BAF, derived from Streptomyces griseus [27], has been shown to suppress DENV infection in various mammalian and insect cell lines. BAF suppresses DENV by 80% in the A. albopictus cell line C6/36 HT [18] and also inhibits Sindbis virus infection in mammalian BHK cells, but not in A. albopictus C7-10 cells [24]. MPA was developed as a transplant rejection preventive drug and is an inhibitor of inosine-5′-monophosphate dehydrogenase (IMPDH) that catalyzes the synthesis of xanthosine monophosphate (XMP) from inosine monophosphate (IMP) [37], [38]. This is a rate-limiting step for the de novo synthesis of guanine nucleotides and is required for DNA and RNA synthesis [34]. MPA treatment of various mammalian cells has been shown to suppress DENV infection [13], [17]. Treatment with MPA also inhibits Sindbis virus infection of the A. albopictus cell lines, LTC 7 and Ama 18, suggesting MPA can act as an antiviral compound across taxa [39]. In our present study, Injection of either 50 or 250 µM MPA suppressed DENV infection in adult mosquitoes, and there was no difference of the reduction produced by the two concentrations, suggesting that IMPDH inhibition by MPA is already saturated at 50 µM.
CAS and DNJ are inhibitors of alpha-glucosidases that function as DENV HFs in mammals [14], [15]. An alpha-glucosidase has also been identified as a possible DENV HF in the D. melanogaster cell line S2 [10], and its ortholog is induced in A. aegypti midgut during DENV-2 infection [40], suggesting that it may represent a DENV HF in the mosquito. However, our results showed that none of the alpha-glucosidase inhibitors suppressed DENV infection of adult mosquitoes (Fig. 1D, E), nor did RNAi-mediated silencing of the alpha-glucosidase gene (AAEL010599 and AAEL015337), an ortholog of a D. melanogaster DENV HF, result in altered susceptibility to the virus [30]. These results suggest that DENV can utilize alternative enzymes in A. aegypti that are resistant to DNJ and CAS treatment. However, we cannot exclude the possibility that another A. aegypti alpha-glucosidase that is resistant to CAS and DNJ treatment serves as a DENV HF.
Effective antiviral action of BAF and MPA required administration prior, or simultaneously, to virus exposure, whereas compound injection into mosquitoes at 24 h after feeding on DENV infected blood did not result in a significant DENV suppression in the midgut tissue. BAF inhibits virus entry and MPA inhibits replication, and at 24 h after ingestion of the virus-laden blood, a sufficient number of viruses may already have succeeded in entering the cells and replicating, thereby exceeding the threshold above which the drug's antiviral effect is ineffective. However, it is likely that BAF and MPA treatment after feeding on DENV infected blood could result in virus suppression in the salivary glands. We did not inject mosquitoes with the compounds at the time of infected blood ingestion to avoid interfering with the feeding. However, ingestion of the compounds together with DENV via a blood meal resulted in suppression of DENV. Thus, inhibition of HFs with BAF and MPA must occur prior to infection, or during the early stage of infection, in order to block viral infection of the mosquito.
Despite the fact that BAF (an inhibitor of virus cell entry) and MPA (an inhibitor of viral RNA synthesis) inhibit the virus at two independent stages of infection, no synergistic antiviral effect was observed when the compounds were administered simultaneously. However, in contrast to the co-injection of the two compounds, RNAi-mediated co-silencing of the putative drug target genes vATPase and IMPDH did result in a marginally greater antiviral effect when compared to silencing each gene individually (the p-value of the difference in virus titer between the co-silenced and single-silenced mosquito cohorts was just above the significant level: p = 0.0535). This discrepancy is a likely result of a drug interaction between BAF and MPA that negatively affects the activity of one of the two, or of both drugs when administered together. Alternatively, the large amount of xenobiotics in co-injected mosquitoes may have augmented the insect's detoxification system, thereby neutralizing the action of any of the administered compounds.
Although injection may be the most efficient way to deliver these compounds, it is not a natural route that could be applied to mosquitoes in nature; however, in nature, one of the following three routes would be possible: ingestion through the nectar, through blood feeding, or through surface exposure as in the case of insecticide treatment. Ingestion of toxic mosquitocidal substances through artificial nectar feeding has been deployed for mosquito control [41], [42], but ingestion of mosquitocidal or transmission-blocking substances through blood feeding has only been addressed by one previous study that investigated the effect on the mosquito but not virus infection [43]. Thus, we investigated the possibility of compound delivery through blood feeding. This route of mosquito exposure to compounds would simulate a situation when mosquitoes would feed on a dengue infected individual being treated with therapeutic and transmission-blocking medication. Our data show that this would have the potential to suppress DENV replication and thus limit the likelihood of DENV transmission.
The greater antiviral efficacy of the injected versus ingested compounds is most likely due to the fact that the injection brings the compound directly into the hemolymph, where it can be absorbed by the basal side of the midgut epithelial cells. In contrast, in order to exert antiviral activity through HF inhibition, compounds ingested via the blood would have to traverse the chitinous peritrophic matrix and be absorbed by the lumenar side of the midgut epithelial cells, in addition to persisting in the blood meal [44]–[46]. Furthermore, a large proportion of the ingested compounds may be excreted through diuresis rather than being absorbed by the midgut epithelium [47]. Administration through the sugar meal will most likely result in a lower overall uptake of a compound, since the amount of ingested sugar is smaller than that of blood and it occurs over a longer period that might not allow the active concentration of the compound to reach potent levels.
By silencing the respective genes in adult mosquitoes prior to DENV infection, we confirmed that the predicted targets of BAF and MPA are associated with DENV replication. BAF inhibits the vATPase by binding to a proteolipid subunit of the vATPase V0 domain [27], [48]. vATPases are multisubunit enzymes with two domains. In yeast, the peripheral catalytic V1 domain has eight subunits and the integral membrane V0 domain has six subunits [Reviewed in 49]. We have previously confirmed that vATPase subunit G (AAEL012819) is a DENV HF [30]. In the current study, we have tested four additional subunits of the vATPase, including two of the proteolipid subunits (AAEL012113, AAEL000291) of V0 that are known to be inhibited by BAF. vATP-f is an ortholog of the vATPase V1 domain subunit in yeast, and the other tested genes are also orthologs of vATPase V0 domain subunits. The A. aegypti orthologs of the V0 domains, AAEL011025, AAEL012113, and AAEL000291, have been shown to be up-regulated in the DENV-susceptible Moyo-S strain in response to DENV infection, further indicating that vATPases play important roles in viral infection [50]. Our RNAi-mediated gene silencing of all four tested vATPase subunits resulted a significant inhibition of DENV, suggesting that the function of vATPase enzyme as a whole complex is required for efficient DENV infection in A. aegypti. This gene silencing-mediated infection phenotype was consistent across laboratory-adapted and field-derived mosquito strains, emphasizing a crucial role for this enzyme complex in DENV infection of adult mosquitoes. Thus, vATPase is a highly potent candidate target for DENV HF inhibitors. Effects on IMPDH is the target enzyme of MPA and a known virus HF involved in viral replication [51], [52], and its depletion through RNAi also reduced DENV titers, suggesting that it represents a druggable DENV HF in adult A. aegypti.
Silencing of the vATPase subunit negatively affected mosquito fitness; a decreased mosquito lifespan was observed from 17 days post injection and onwards. This observation is consistent with findings from earlier studies on yeast [53], [54], Neurospora [55], mice [56] and insects including Drosophila melanogaster, Tribolium castaneum, Acyrthosiphon pisum, Manduca sexta, Peregrinus maidis, Diabrotica virgifera, Bemisia tabaci and Bactrocera dorsalis [57]–[61]. Defective vATPase in these organisms resulted in greater mortality during various life stages including embryonic, larval and adult stages, indicating that vATPase function is critical for homeostasis. Impaired function of vATPase in adult female mosquitoes also impeded oviposition. Following vATPase subunit silencing, fecundity, as measured by the median number of eggs produced, was reduced by 98%. Reduced fecundity as a result of vATPase silencing has been reported in horn flies and corn planthoppers [59], [62]. In corn planthoppers, silencing of vATPase caused immature oocyte formation [59]. Our data show that chemical or dsRNA-mediated inactivation of vATPase has the potential to result in mosquito population suppression. Earlier studies have also proposed that dsRNAs targeting of vATPases could be used to target insect pests [58]–[60]. The synergistic effects of vATPase inactivation on mosquito fitness and DENV infection potentiates the targeting of this HF for transmission blocking. However, in the case of chemical vATPase inactivation the non-specific effects due to the high degree of vATPases conservation across species must be considered.
In summary, we show here for the first time that several vATPase subunits and IMPDH represent potent HFs in adult female A. aegypti and that they can be targeted by drugs to inhibit DENV infection and potentially transmission. The development of dengue control strategies based on the chemical or RNAi-mediated inhibition [63]–[65] of virus HFs has several advantages, since it does not have the ecological impact of insecticides that indiscriminately kill insects, and it is not subject to the virus's ability to develop resistance. HF-targeted compounds or dsRNAs can be delivered through artificial nectar [66]. Alternatively, they can be engineered, or administered through an appropriate medium, for effective diffusion through the insect's cuticle and then be applied as sprays or impregnated into bed nets, as in the case of insecticides. Many DENV HFs, for example vATPase and IMPDH, are shared between the mosquito and human hosts and therefore open the way for the discovery of drugs that have both therapeutic (i.e., blocking the virus in the human host) and transmission-blocking (i.e., blocking the virus in the mosquito vector) activity, thereby maximizing the disease-control potential. Although BAF and MPA exert anti-viral activity in the mosquito, they are likely not optimal, from a drug efficacy or economic standpoint, for dengue control. However, a knowledge of vATPase and IMPDH as ubiquitous HFs in humans and mosquitoes, along with the structural attributes of BAF and MPA and other known anti-dengue compound point to the possibility of developing effective chemicals that could exert both therapeutic and transmission blocking activity for dengue control.
|
10.1371/journal.pntd.0005962 | Quantitative multiplexed proteomics of Taenia solium cysts obtained from the skeletal muscle and central nervous system of pigs | In human and porcine cysticercosis caused by the tapeworm Taenia solium, the larval stage (cysts) can infest several tissues including the central nervous system (CNS) and the skeletal muscles (SM). The cyst’s proteomics changes associated with the tissue localization in the host tissues have been poorly studied. Quantitative multiplexed proteomics has the power to evaluate global proteome changes in response to different conditions. Here, using a TMT-multiplexed strategy we identified and quantified over 4,200 proteins in cysts obtained from the SM and CNS of pigs, of which 891 were host proteins. To our knowledge, this is the most extensive intermixing of host and parasite proteins reported for tapeworm infections.Several antigens in cysticercosis, i.e., GP50, paramyosin and a calcium-binding protein were enriched in skeletal muscle cysts. Our results suggested the occurrence of tissue-enriched antigen that could be useful in the improvement of the immunodiagnosis for cysticercosis. Using several algorithms for epitope detection, we selected 42 highly antigenic proteins enriched for each tissue localization of the cysts. Taking into account the fold changes and the antigen/epitope contents, we selected 10 proteins and produced synthetic peptides from the best epitopes. Nine peptides were recognized by serum antibodies of cysticercotic pigs, suggesting that those peptides are antigens. Mixtures of peptides derived from SM and CNS cysts yielded better results than mixtures of peptides derived from a single tissue location, however the identification of the ‘optimal’ tissue-enriched antigens remains to be discovered. Through machine learning technologies, we determined that a reliable immunodiagnostic test for porcine cysticercosis required at least five different antigenic determinants.
| Human and porcine cysticercosis caused by Taenia solium is a parasite disease still endemic in developing countries. The cysts can be located in different host tissues, including different organs of the central nervous system and the skeletal muscles. The molecular mechanisms associated with the tissue localization of the cysts are not well understood. Here, we described the proteome changes of the cysts obtained from different host tissues from infected pigs using quantitative multiplex proteomics. We explored the diversity of host proteins identified in the cyst’s protein extracts and we also explored the immune-localization of several host-related proteins within the cysts, and propose their possible function. We identified several proteins and antigens enriched for a given tissue localization. Several synthetic peptides designed from these tissue-enriched antigens were tested trough ELISA. Using a combination of peptide mixtures and machine learning technologies we were able to distinguish non cysticercotic and cysticercotic pig’s sera. The tissue-enriched proteins/antigens could be useful for the development of improved immuno-diagnostic tests capable of discriminate the tissue-localization of the cysts.
| Human and porcine cysticercosis caused by the larval stage of Taenia solium, is acquired by the ingestion of this parasite’s eggs. After activation by several gastrointestinal agents, the oncospheres penetrating the intestinal wall later establish in different tissues and organs including the skeletal muscles (SM) and the brain. In humans, establishment of cysts in the central nervous system (CNS) causes neurocysticercosis (NC), a serious and pleomorphic disease that can become highly debilitating [1]. Heterogeneity of human NC has been associated, at least in part, with the number and localization of the cysts in the CNS [2], as well as to many other factors including a complex immune response directed to a number of cyst’s antigens [3, 4, 5, 6].
The molecular factors associated with the tissue localization of the T. solium cysts remain poorly understood [7]. Other pathogenic microorganisms (S. pneumoniae, Campylobacter jejuni, Escherichia coli, Trypanosoma brucei, etc.), show tissue preference linked to a number of specific pathogen’s proteins [8, 9, 10, 11, 12].
Information available on proteomics changes of flatworm parasite infections is limited. However, we know that parasites respond to hormones, cytokines and other host’s molecules [13]. The availability of several tapeworm genomes [14] has allowed to detail this complex host-parasite cross-communication including insulin, EGF/FGF, TGF-b/BMP, among others (for an updated review see [15]). Insulin responsiveness has been described for Schistosoma mansoni, Taenia crassiceps and Echinococcus multilocularis [16, 17, 18]. The differential effects of steroid hormones during parasite infections is also well documented [19]. Some parasites also have the ability to respond to host cytokines; for example, S. mansoni has receptors to TNF-α and TGF-β and proteomic and genomic changes have been reported in response to those cytokines [20, 21, 22].
The advent of high throughput proteomic techniques greatly widens our power to approach these old questions in molecular helminthology. In this context, body fluids of the host may affect proteome expression of infectious agents, for example, E. coli growing in a media supplemented with urine show a differential proteome signature [23]. Furthermore, several proteomic changes of Streptococcus pyogenes have been reported in response to serum supplementation [24]. The molecular factors associated with the tissue localization of helminth parasites within the host tissues has been less explored; in the case of Trichinella spiralis several changes have also been reported between parasites isolated from different host tissues [25]. However, important advances in helminth proteomics, including metacestode cystic/vesicular larval forms, have been reported [26–29].
It is conceivable that the host tissue’s molecular environment modulates the protein expression of pathogens, including parasites. Accordingly, specific proteomic profiles of parasites could be associated with a certain tissue localization.
Understanding the proteome changes of parasites in different host tissues, can provide insights not only on the molecular networking occurring in complex host parasite relationships, but it could also be useful for the design of more effective vaccines, drugs, as well as for the improvement of available diagnostic procedures.
Here we benefited from isobaric quantitative proteomics to elucidate the proteomic changes of T. solium cysts obtained of SM and CNS of pigs. A protein profile was found associated with each tissue localization, allowing the identification of 42 tissue-enriched antigens and the design of 14 synthetic antigenic peptides that were evaluated for antibody recognition using infected and uninfected pig’s sera. Our results indicated that an optimal immunological diagnosis for porcine cysticercosis requires at least five different epitopes from several tissue-enriched antigens. A remarkable finding was the conspicuous and abundant presence of host proteins in the protein extracts of the cysts; 891 host proteins were identified and quantified. We present initial findings suggesting that several intact host’s proteins might play a significant role in tapeworm’s physiology.
Methanol−chloroform precipitation of the reduced and alkylated protein extracts was performed prior to protease digestion. Samples of 400 μg of each protein extract were resuspended separately in 100 μL of 8 M urea in 50 mM HEPES, pH 8.2. After solubilization, the protein extracts were diluted to 4 M urea with 50 mM HEPES, pH 8.2, and digested at RT for 3 h with endoproteinase Lys-C (Wako, Japan) at 5 ng/μL. The mixtures were then diluted to 1 M urea with 50 mM HEPES, pH 8.2, and trypsin was added at a 50:1 protein-to-protease ratio. The reaction was incubated overnight at 37°C and stopped by the addition of 100% TFA to a final pH < 2. Peptides were desalted using 50 mg tC18 SepPak solid-phase extraction cartridges (Waters, Milford, MA) and lyophilized. Desalted peptides were resuspended in 100 μL of 200 mM HEPES, pH 8.2. Peptide concentrations were determined using the microBCA assay (Thermo Fisher Scientific, Waltham, MA). One-hundred micrograms of peptides from each sample was labeled with TMT reagent. TMT-10 reagents (0.8 mg, from Thermo Fisher Scientific) were dissolved in anhydrous acetonitrile (40 μL), of which 10 μL were added to the peptides along with 30 μL of acetonitrile (final acetonitrile concentration of approximately 30% (v/v)). The labeling reaction proceeded for 1 h at room temperature and then was quenched with hydroxylamine (Sigma, St. Louis, MO) to a final concentration of 0.3% (v/v). The TMT-labeled samples were mixed equally, vacuum centrifuged to near dryness, desalted using 200 mg solid-phase C18 extraction cartridge (Sep-Pak, Waters), and lyophilized.
The TMT-labeled peptides were fractionated using BPRP HPLC. An Agilent 1100 pump equipped with a degasser and a photodiode array (PDA) detector (set at 220 and 280 nm wavelength) from Thermo Fisher Scientific (Waltham, MA) were used. Peptides were subjected to a 50 min linear gradient from 5% to 35% acetonitrile in 10 mM ammonium bicarbonate pH 8 at a flow rate of 0.8 mL/min over an Agilent 300 Extend C18 column (5 μm particles, 4.6 mm ID, and 220 mm in length). Beginning at 10 min of peptide elution, fractions were collected every 0.38 min into a total of 96 fractions, which were consolidated into 24, of which 12 nonadjacent samples were analyzed. Samples were dried via vacuum centrifugation. Each eluted fraction was acidified with 1% formic acid and desalted using StageTips [31], dried via vacuum centrifugation, and reconstituted in 4% acetonitrile, 5% formic acid for LC−MS/MS analysis.
All mass spectrometry data were collected on an Orbitrap Fusion mass spectrometer (Thermo Fisher Scientific, San Jose, CA) coupled to a Proxeon EASY-nLC II liquid chromatography (LC) pump (Thermo Fisher Scientific). Peptides were eluted over a 100 μm inner diameter micro-capillary column packed with ∼0.5 cm of Magic C4 resin (5 μm, 100 Å, Michrom Bioresources) followed by ∼35 cm of Accucore resin (2.6 μm, 150 Å, Thermo Fisher Scientific). For each analysis, we loaded ∼1 μg of the peptide mixture onto the column. Peptides were separated using a 90 min gradient of 6−26% acetonitrile in 0.125% formic acid at a flow rate of ∼350 nL/min. The dynamic exclusion duration was set at 90 s, with a mass tolerance of ±7 ppm. Each analysis used the multinotch MS3-based TMT method [32] on an Orbitrap Fusion mass spectrometer, which has been shown to reduce ion interference compared to MS2 quantification. The scan sequence began with an MS1 spectrum (Orbitrap analysis; resolution 120000; mass range 400−1400 m/z; automatic gain control (AGC) target 2 × 105; maximum injection time 100 ms). The 10 most-abundant MS1 ions of charge states 2−6 were fragmented, and multiple MS2 ions were selected using a Top10 method. MS2 analysis was composed of collision induced dissociation (quadrupole ion trap analysis, AGC 4 × 103; normalized collision energy (NCE) 35; maximum injection time 150 ms). Following acquisition of each MS2 spectrum, we collected an MS3 spectrum as described previously [32], in which multiple MS2 fragment ions were captured in the MS3 precursor population using isolation waveforms with multiple frequency notches. MS3 precursors were fragmented by high energy collision-induced dissociation (HCD) and analyzed using the Orbitrap (NCE 55; AGC 5 × 104; maximum injection time 150 ms, resolution was 60,000 at 400 Th).
Instrument data files were processed using a SEQUEST-based in-house software pipeline [33]. Spectra were converted from.raw to mzXML using a modified version of ReAdW.exe. A database containing all predicted ORFs for entries from the parasite (T. solium genome database; http://www.genedb.org/Homepage/Tsolium; downloaded March 31, 2015) and the host (Sus scrofa database; http://www.uniprot.org/proteomes/?query=taxonomy:9823; downloaded March 31, 2015) was used. This database was concatenated with another database composed of all protein sequences in the reverse order. Searches were performed using a 50 ppm precursor ion tolerance for total protein level analysis. The product ion tolerance was set to 0.9 Da. These wide mass tolerance windows were chosen to maximize sensitivity besides SEQUEST searches and linear discriminant analysis [34, 35]. TMT tags on lysine residues and peptide N termini (+229.163 Da) and carbamidomethylation of cysteine residues (+57.021 Da) were set as static modifications, while oxidation of methionine residues (+15.995 Da) was established as a variable modification.
Peptide-spectrum matches (PSMs) were adjusted to a 2% false discovery rate (FDR) [35]. PSM filtering was performed using a linear discriminant analysis, as described previously [33], while considering the following parameters: XCorr, ΔCn, missed cleavages, peptide length, charge state, and precursor mass accuracy. For TMT-based reporter ion quantitation, we extracted the signal-to-noise (S/N) ratio for each TMT channel and found the closest matching centroid to the expected mass collapsed to a 1% peptide FDR and then collapsed further to a final protein-level FDR of 1%. Moreover, for protein assembly, principles of parsimony were used to produce the smallest protein set, necessary to account for all observed peptides. Proteins were quantified by summing reporter ion counts across all matching PSMs using in-house software, as described previously [36]. Briefly, a 0.003 Th window around the theoretical m/z of each reporter ion (126, 126.1278 Th; 127N, 127.1249 Th; 127C, 127.1310 Th; 128N, 128.1283 Th; 128C, 128.1343 Th; 129N, 129.1316 Th; 129C, 129.1377 Th; 130N, 130.1349 Th; 130C, 130.1410 Th; 131, 131.1382 Th) was scanned for ions, and the maximum intensity nearest the theoretical m/z was used. PSMs with poor quality, MS3 spectra with TMT reporter summed signal-to-noise ratio less than 387, or no MS3 spectra were excluded from quantitation [32]. The RAW files will be made available upon request. Protein quantitation values were exported for further analysis in Excel, Perseus 1.5.2.4 and GraphPad prism v6. Proteins with more than three missing channels were discarded, in the case of identifications based in a single peptide, that peptide was present in at least 7 samples. The selection of the tissue-enriched proteins (Fig 1) was based on the comparison of fold changes between CNS and SM cysts using a multiple T-test and Benjamini-Hochberg correction with a 5% of FDR (there were 5 CNS samples, unfortunately, one sample of CNS cysts was discarded at the end, due to poor data quality).
Proteins with a P-value <0.01 (n = 261) and proteins without changes (lowest coefficient of variation, n = 50) were chosen to predict their antigenic regions. A detailed explanation is found in the S2 Fig. Initially, the antigenicity algorithm [37] and the B cell epitope algorithm [38] were used to quantitatively estimate the proteins with the higher antigenicity. Only the proteins predicted by both algorithms with a high percentage of antigen/epitope were selected (n = 42). The peptide selection was based on the following criteria: length of at least 15 amino acids (average size of predicted antigenic regions = 14.1); coincidence in the prediction of at least 5 amino acids by both algorithms, and at least, 10 amino acids should be predicted by one of the algorithm. The resulting peptides were submitted to an algorithm that was trained with a set of synthetic peptides of proven utility in diagnostic procedures as well as with a set of peptides that were not useful [39]; peptides with the highest probability of recognition by antibodies were selected. A total of ten peptides were selected (one from each protein); 4 peptides derived from proteins that were abundant in SM cysts, 4 from CNS cysts and 2 from proteins that did not show change in both tissues. All peptides were purchased from GenScript (USA).
Proteins from the host (Sus scrofa) were annotated using the PantherGo algorithm [40, 41]; in the case of T. solium proteins, only proteins with a P-value<0.01 were submitted to Argot2 algorithm [42–44] using a threshold of 200. Disulfide bonds, N-linked glycosylation sites, transmembrane regions, signal peptides, and GPI-anchoring sites were predicted for selected proteins using several algorithms [45–56].
The insoluble fraction of T. solium cysts, the VF and the synthetic peptides were tested by ELISA. Briefly, 1.5 μg of the insoluble fraction and of the VF, as well as 500 ng of each synthetic peptide were used to coat separate wells of microtiter plates. After overnight incubation at 4°C with mild agitation, the plates were washed, blocked for 2 h with 1% albumin in PBS-0.05% Tween 20 (PBST) and incubated with different pig sera, diluted 1:200 in PBST and incubated overnight at 4°C. A HRP coupled anti-pig IgG hyperimmune serum was used (diluted 1:4,000) as secondary antibody. The reaction was developed using OPD (0.4 mg/mL) for about 3–5 min and stopped with 3N HCl. Absorbance at 492 nm was determined in a Multiskan FC (Thermo-Fisher Scientific).
The total saline extract was passed through a column of Protein G coupled to Sepharose 4B. The bound IgG was eluted using 0.1 M glycine pH 2.3 and immediately neutralized with Tris 1M, pH 7.3. Fractions containing the bound IgG were concentrated using an Amicon system (10 kDa cutoff) and washed several times using PBS, pH 7.3. The purified IgG was quantified by Non-Interfering protein assay (GBiosciences).
The IgG purified from the cysts protein extracts was tested for antibody activity through conventional ELISA and western-blotting procedures. For ELISA, microtiter plates were separately coated using 1.5 μg of VF or the insoluble protein fraction of cyst tissue (see above) in carbonate buffer, pH 9.6. After overnight incubation at 4°C with mild agitation, the plates were washed three times using PBST and blocked using 1% albumin in PBST for 1 h at room temperature. After another washing cycle, the plates were incubated overnight at 4° C with the IgG fraction purified from the T. solium cysts. A pool of sera from 15 cysticercotic pigs was also used in similar assays for comparison. Dilutions for both the IgG purified from the cysts and the pool of sera from the infected pigs are shown below. After washing, the plates were incubated with a HRP-coupled rabbit anti-pig IgG hyperimmune serum (diluted 1:1000) for 1h at room temperature. The reaction was developed using OPD (0.4 mg/mL) for about 3–5 min and stopped with 3N HCl. Absorbance at 492 nm was determined in a Multiskan FC (Thermo-Fisher Scientific).
In the case of western-blotting, 20 μg of the VF and the insoluble fraction were resolved through 12% SDS-PAGE and transferred onto a nitrocellulose membrane. The membranes were blocked overnight using 10% of skim milk in PBS and incubated with 10μg/mL of the IgG purified from T. solium cysts in 10% skim milk at room temperature. After three washings using PBS-Tween 0.1%, the membrane strips were incubated with the rabbit anti-pig IgG secondary antibody (diluted 1:85,000) for 2h at room temperature. The antigen-antibody reaction was developed using a West femto chemiluminiscence kit (Thermo) following the manufacturer’s instructions.
For each subset of k antigenic peptide measures and n subjects, predictive accuracy was measured by Leave One Out Cross Validation. Here, n independent training/testing procedures using a Support Vector Machine (SVM) were performed. Each training set consisted in all except one individual value, and testing set being the individual left out. Accuracy is computed as the fraction of times each individual test was correctly classified for that particular selection of k peptides. Source code for SVM implementation is found in S1 Data and is contained in scikit-learn package [SKLEARNREF] with default parameters [57].
Cysts (CNS and SM) were fixed in Zamboni solution. Afterwards, all samples were dehydrated and embedded in paraffin. Heat-mediated antigen retrieval was performed on 5-μm sections, using a 0.1 M sodium citrate solution (pH 6.0) in a high-pressure sterilizer (120°C for 5 min) and endogenous peroxidase was consumed by incubation with 0.3% (v/v) H2O2 in PBS for 10 min at RT. Afterwards, the tissue section on slides were washed three times with PBS and maintained in a blocking solution (0.1% BSA in PBS, Sigma-Aldrich) for 10 min. After washing with PBS, the slides were incubated overnight with the primary antibody (the list of antibodies used could be found in S1 Table), diluted 1:50, in PBS-0.1% BSA, at 4°C. After washing several times with PBS, the tissue sections were incubated with the corresponding second antibody (HRP-conjugated) 1:1000 for 30 min at 37°C. Peroxidase activity was visualized by incubating the samples for 2 min with 3-diaminobenzidine tetrahydrochloride (DAB, MP Biomedicals). Reaction was stopped with water, and sections were counterstained with hematoxylin, dehydrated, cleared, and mounted with resine (Gold Bell). The single labeled sections were examined and photographed under light microscopy (Nikon Eclipse 80i) using a digital color video camera (Nikon Digital Sigth). The second antibody controls could be found in S3 Fig.
All relevant information are within the manuscript or supplementary material. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE [58] partner repository with the dataset identifier PXD00527. The source code for the support vector machine is available in the S1 Data; protein and peptide identifications can be found in S2 and S3 Data; the subset of highly antigenic proteins can be found in S4 Data; the functional annotation of the tissue-enriched proteins in S5 Data and several cestode proteome comparison in S6 Data.
Briefly, 50 μg of VF and insoluble fraction of cysts tissue were resolved by 12% SDS-PAGE, the gels were transferred onto nitrocellulose membranes. The membranes were blocked overnight using 10% of skim milk in PBS and incubated with the primary antibody diluted 1: 3000 in 10% skim milk at room temperature (S1 Table). After three washings using PBS-Tween 0.1%, the membrane strips were incubated with the corresponding secondary antibody diluted 1:50,000 in PBST. The antigen-antibody reaction was developed using a West femto chemiluminiscence kit (Thermo) following the manufacturer’s instructions.
We compared our proteomic dataset with several comprehensive proteomic studies performed in Echinococcus granulosus [26], E. multilocularis [27], Mesocestoides corti [28] and the theoretical secretome of T. solium [59]. In the case of Echinococcus and Mesocestoides proteomes, the sequences of the identified proteins in those studies were obtained from WormBase (www.parasite.wormbase.org/). After, those sequences were blasted against the T. solium database (http://www.genedb.org/blast/submitblast/GeneDB_Tsolium). Then, the top-ranked protein was considered the T. solium homologue of a certain Echinococcus or Mesocestoides protein (the complete list can be found in S6 Data).
The Animal protocol was revised and approved by the Ethical Committee for the Care and Use of Laboratory Animals at the Institute for Biomedical Research, Universidad Nacional Autónoma de México, under the license number ID 199 which follows the guidelines stated by the National Institutes of Health Guide for the Care and Use of Laboratory Animals.
The proteome of the larval phase of T. solium was a complex mixture of parasite and host proteins. Using a TMT-multiplexed strategy, we were able to identify and quantify over 4,200 proteins across the nine cyst samples. Among these proteins, 3,368 were identified as parasite proteins, whereas 891 proteins were of host origin (Fig 1A). To our knowledge, this is the largest number of identified and quantified proteins in a multiplex assay for a cestode parasite to date. All the proteins were found across all the samples. More than 99.4% of proteins (including host or parasite) were identified in the nine samples. However, in the case of proteins TsM_000997700 and TsM_000195700, single peptides were only present in 8 samples and absent in only one. The protein changes observed between parasites obtained from different hosts and tissues were relatively discrete; the large majority of the identified and quantified cyst proteins remained at similar levels of expression, i.e., more than 3,200 proteins were found within a fold change of -1:1 (Fig 1B and 1C) for the SM and CNS cysts of the five pigs analyzed. Quantified host proteins were more variable between cysts from different tissues than between cysts from different pigs, thereby supporting the reproducibility of protein changes across the study (Fig 1B and 1C).
However, several parasite and host proteins were enriched for certain tissue localization of the cysts. For example, protocadherin alpha 6 (P = 0.0003), actin type 5 (p = 0.0002), a component of the γ-tubulin complex (P = 0.0001), a subunit of the splicing factor 3A (P<0.0001) and a protein associated with microtubules (P = 0.0001), were more abundant for cysts dissected from SM. On the other hand, proteins significantly associated to cysts dissected from the CNS of the pigs included a DNA topoisomerase 1 (P = 0.0001), alanine aminotransferase (P = 0.0002), an aldo-keto reductase (P = 0.0002), dnaJ protein (P = 0.0001), a protein containing a carbohydrate kinase domain (P<0.0001) and two hypothetical proteins (P<0.0001 and P = 0.0003) (Fig 1D). Two groups of host proteins were also found enriched for each tissue localization of the cysts (Fig 1E).
Several studies consistently detected intact host proteins in protein extracts of Taeniid parasites [30, 60–64]. In a recent report, we identified 17 host proteins in the vesicular fluid of T. solium cysticerci [63]. The results reported here, to our knowledge, are by far the largest set of host proteins reported within the tissues and fluids for any cestode parasite, suggesting a highly complex and close contact between the porcine and the cysts proteins.
It has been proposed that the host proteins are up-taken through a non-specific mechanism such as fluid phase endocytosis [60]. In our dataset, the host proteins were more variable than cysts proteins, this could be associated with a differential composition of the cysts microenviroment, CNS vs SM (S2 and S3 Data). Gene ontology analysis allowed determining that a diversity of host proteins were present; categories included metabolic enzymes involved in pathways like glycolysis or fructose/galactose metabolism, as well as signaling proteins including those participating in the integrin and ubiquitin-proteasome system. Functions of the uptaken host proteins included RNA binding, chaperones, oxidoreductases, ribosomal and isomerases (Fig 2A), those pathways were also enriched for the skeletal muscle proteome of the pigs [65], suggesting that the up-taken host proteins reflect the composition of the cysts micro-environment (S5 Data).
It has been proposed that the cysts use the up-taken host proteins, including immunoglobulins, as a source of amino acids [66, 67]. The amino acid composition of the host proteins resulting from cysts obtained from CNS and SM was very similar; the biggest differences were found for aspartic acid and serine, reaching only an increase of 1.2% and a decrease of 1.5%, respectively (compared with S. scrofa proteome). The similar amino acid composition of up-taken host proteins in CNS and SM cysts is consistent with the concept of an unspecific mechanism for the host protein uptake by the cysts (Fig 2B).
We also explored the question of the host proteins tissue localization. Tissue localization studies were carried out to determine the distribution of several host proteins within the cyst’s tissues. We selected host proteins related to iron metabolism in the host: haptoglobin (Hp), hemoglobin (Hb), hemopexin (Hpx), hepcidin and ferritin; other host proteins such as LDL, albumin and IgG, were used for comparison. Two experimental approaches were employed: cyst protein fractionation followed by western blotting and immuno-localization in tissue-sections. Through western blotting we found the majority of host proteins in the vesicular fluid, as well as in the soluble fraction of cyst’s tissue (Fig 2C). Abundant host proteins such as IgG and albumin were detected in the three protein fractions tested (soluble and insoluble fraction of cyst´s tissue and in the vesicular fluid). Hemopexin (Hpx) was found in the tissue’s extracts (soluble and insoluble fractions) and was not detected in the vesicular fluid; in the case of the insoluble fraction of cysts tissue, the immuno-reactive band was detected at the same molecular weight as the positive control (serum) and in the soluble fraction a band with a slightly increased molecular weight was detected. LDL was found present in the cyst’s tissue soluble protein extract and scarce in the vesicular fluid (Fig 2C).
In the case of the host proteins related to iron metabolism, the most abundant ones found in the cyst’s tissues were Hp and Hb; for Hp several bands were recognized (most of the bands were shared with the positive control), but others only appeared in the cysts extracts. In the case of Hb, immunoreactive bands were detected in the cysts tissue. After immune-histochemical analysis we found a conspicuous distribution of Hp and Hb in the cyst’s tissues, particularly in tissue surrounding the spiral canal of the invaginated scolex. Hepcidin and ferritin were also found in the cyst’s subtegumentary tissue, but in the case of ferritin in cysts extracts, the immuno-reactive bands had lower molecular weight than the band in serum (Fig 2C and 2D). All tested host proteins (with the exception of ferritin) were detected in their expected molecular weight in gels, suggesting that at least a fraction of the protein is intact.
This idea was explored further using as a model the uptaken host IgG. IgG was located on the outer surface of cysts in both CNS and SM cysts, with a more intense signal in SM cysts (Fig 3A). This localization was consistent with a previous report [68]. To explore the antibody activity of the uptaken host IgG; we purified host IgG from total saline extracts of SM cysts through affinity chromatography using protein G (Fig 3B). The purity of the isolated pig IgG was evaluated by SDS-PAGE and western blot (Fig 3C). Heavy and light IgG chains were detected at the expected molecular weights (50 and 25 kDa), suggesting that pig IgG was uptaken intactly. Two assays were performed to test the recognition of T. solium antigens by the purified pig IgG: ELISA and western blot using vesicular fluid (VF) or the insoluble fraction of the cysts as parasite antigens and the purified IgG as primary antibody (Fig 3D). The purified pig IgG reacted both, with the VF and the insoluble fraction of cysts antigens in a saturable and specific way. For western blotting, both antigenic fractions, VF and the insoluble fraction reacted with the purified IgG from the cysts (as shown in Fig 3E). As expected, several bands were recognized and were shared with the band obtained using sera from cysticercotic pigs (Fig 3D), indicating that uptaken host immunoglobulins retained their antigen-binding activity.
To explore the molecular functions and biological processes of the T. solium proteins identified for each tissue localization, we performed a k-means clustering analysis using a subset of cysts proteins with a significant fold change (P-value<0.01, n = 261). Two major clusters were identified, one for each tissue localization (Fig 4). For CNS cysts (Fig 4A), 116 proteins were abundant for parasites of this tissue localization (compared with SM parasites). These proteins were associated with metabolic processes, transport and phosphorylation. In the case of SM cysts (Fig 4B), methylation, signal transduction and microtubule-based processes were the most frequently observed categories. For comparison, 48 proteins with the lowest coefficient of variation that remained in similar levels between cysts from different tissues were selected, including proteins associated with ubiquitination, phosphorylation and mitochondrial processes (Fig 4C, S5 Data). The relevance of those pathways in explaining both preferential localizations of the cysts, in terms of adaptation and survival within the host tissues, deserves future study.
As mentioned above, several host and parasite proteins were associated with the tissue localization of the cysts. To investigate more thoroughly that possibility, several previously described antigens (reviewed in [69]) were queried for in our database. As shown in Fig 3, several relevant antigens in cysticercosis (i.e., paramyosin, GP50 and a calcium binding protein), were more abundant in SM cysts than in CNS cysts (Fig 5A). Moreover, we mined our database to search for other antigens enriched for each tissue localization. The case of the tetraspanin family appeared to be especially interesting as tetraspanins are relevant antigens in schistosomiasis [70] and hydatidosis [71]. Our results showed the expression of five members of the tetraspanin family; of these, two proteins were enriched in the CNS localization and one in the SM localization of the cysts, while the other two have no detectable changes. These proteins: TsM_000744700 and TsM_001075800 were subsequently analyzed (Fig 5B). Initially, the large extracellular loop of these proteins was deduced using several algorithms (see Materials and Methods). In those tetraspanins, the protective domain (the protein region associated with protection in vaccination trials) is located within the large extracellular loop [70, 71]. Therefore, this portion of the protein was analyzed for antigenicity using algorithms for B cell epitope-prediction. Two peptides were chosen from each protein and synthesized through a commercial service (TsM_000744700: VQGPSDYDGK, NAVQKFECCGVQ; TsM_001075800: YNPNTPEGKGPA, FCCRKDQDCPITE) (Fig 5C). To explore if these four peptides (p744700-1 and 2; p1075800-1 and 2) were recognized by antibodies in the sera of cysticercotic pigs, two groups of 15 sera (cysticercotic and non cysticercotic) from pigs bred in rural endemic areas were used. As shown in Fig 5D, the four peptides were recognized by several infected animals, with the peptide FCCRKDQDCPITE (p1075800-2) having the strongest antibody recognition.
The differential abundance of several antigenic proteins (GP50s, paramyosin, E/S protein M13, calcium-binding protein, tetraspanins, etc.) between SM and CNS cysts, provides evidence about the presence of antigens that were enriched for SM or CNS cysts. Our next goal was the identification of highly antigenic tissue-enriched proteins; here we defined a tissue-enriched protein as one with differential abundance between CNS and SM cysts (p value<0.01). The high antigenicity was defined using B cell epitope and antigenicity predictors, see Materials and Methods. First, we selected 261 proteins with a significant fold change (P value<0.01) and 48 proteins with the lowest coefficient of variation, for comparison. Those proteins were analyzed through several antigenicity algorithms (see Materials and Methods and S2 Fig). Several proteins were predicted as strongly antigenic by one algorithm (antigen/epitope content ≥ 70%). Another group of proteins was also predicted as highly antigenic by both algorithms, although with a lower antigen/epitope content (50%) (Fig 6A).
Using this approach, 40 highly antigenic proteins were identified (the complete list of proteins can be found in S4 Data). To experimentally test the reliability of our antigenic prediction, 10 proteins were selected (Fig 6A and 6B). Of these proteins, four were enriched for CNS cysts, four were enriched for SM cysts and two proteins with a low coefficient of variation between CNS and SM cysts, none of these proteins had been studied before in T. solium. Then, a single epitope was selected for each protein (selected epitopes had to be predicted by both algorithms). After the synthesis of the antigenic peptides (Fig 6B), they were evaluated for antibody recognition by ELISA, using the same two groups of sera mentioned above. For a peptide to be considered as a valid antigen, the difference between the recognition of the cysticercotic and the non cysticercotic pig sera had to be statistically significant.
Nine of 10 peptides (with the exception of one based on pinin) were significantly recognized by the sera from cysticercotic, in comparison with the sera from non cysticercotic pigs (Fig 7). These data, suggest that the proteins from which the peptides were originated are immunologically recognized in porcine cysticercosis; then we have identified several tissue-enriched antigens; interestingly, a peptide that was enriched for SM (p165800) and other for CNS (p223100) cysts, produced the highest difference between the cysticercotic and non cysticercotic pig sera (Fig 7A and 7B). The recognition by the IgG present in the sera of cysticercotic pigs, suggested that there is a subset of tissue-enriched antigens for cysts located in different host tissues. However, whether the rest of the predicted antigenic proteins are valid antigens in cysticercosis requires further screening.
The tissue-enriched antigens could be exploited for the improvement of current diagnostic tools for cysticercosis. A diagnostic test for cysticercosis would ideally include antigenic determinants for each possible tissue localization of the cysts, as well as antigens that are not affected by the tissue-localization of the cysts. To explore the diagnostic potential of those tissue-enriched antigens, several combinations of peptides were used in mixtures. The initial mixture was made using the peptides that were previously found to produce the highest optical densities, when tested separately with the same cysticercotic pig sera: p223100, p165800, p1075800-2 and p239000 (1: 1: 1:1); two concentrations were employed (Fig 8). As shown in Fig 8A, the lowest concentration produced better results. However, not all sera from the cysticercotic animals produced a significantly positive reaction when compared with the non cysticercotic pig’s sera. Other combinations of synthetic peptides were also tested, including a mixture of the 14 peptides that produced the worst performance (Fig 8B). We also tested combinations of peptides from SM-abundant proteins (Fig 8C), or/and SM constitutive proteins (Fig 8D). Interestingly, when the mixture included the best peptide for each tissue localization, p223100 for CNS and p165800 for SM, 14 out of 15 cysticercotic and non cysticercotic pig sera were clearly differentiated (Fig 8E). However, we were not able to improve the performance obtained using the two crude protein extracts from the cysts, indicating that the 'ideal' antigenic subset will require further investigation.
In this study, each pig showed a differential response for each peptide and peptide mixture (Fig 8F); the idiosyncrasy of individual humoral host immune responses against T. solium cyst antigens is well known [3], as it is in other infectious diseases [72–74]. In addition, the considerable genetic/antigenic variation between cysts obtained from different endemic areas it is frequently reported [75–77].
The immuno-diagnosis of an infectious disease is often performed using a single antigen, i.e., using a single protein/peptide or antibody to discriminate between healthy and infected hosts. In the case of infections caused by E. granulosus, the use of AgB and 8 kDa proteins have been tested as diagnostic agents. However, new methodological approaches are needed for parasite infections such as schistosomiasis, echinococcosis and cysticercosis, to discriminate between infected hosts with low-parasite loads [1, 2; 78, 79].
A novel approach involves the machine-learning models that have proven useful in the diagnosis and prediction of several diseases. Several algorithms have been developed for the diagnosis of breast [80], colorectal [81] and non-small cell lung cancer [82]. Distinct antigenic response patterns (ARP) may constitute better representations of the pathogen’s fingerprints than single-peptide responses. Thus, a multi-antigenic peptide testing (MAPT) can identify such ARP for each infection.
To explore the viability of a MAPT using our synthetic peptides, we constructed an antigen response space (ARS), where each individual is represented by a single point. The position of each individual point (one pig’s serum) depends of its antigenic response to several peptides. For k antigenic peptides considered, k OD measures determine the individual position in ARS. In this sense, an ARP can be defined as a particular region in ARS where all infected hosts are present. Machine-learning algorithms can be directly used to identify boundaries of such regions. It should be noticed that a given ARS could represent a corresponding antigenic peptide subset.
To explore the potential of the synthetic peptides to define an ARS for pig’s cysticercosis, all possible combinations of peptides were evaluated. Analysis was performed taking 14 single peptide measures, 5 peptide mixtures measures, and a combination of both. Thus, 16,383, 31 and 524,287 possible ARS representations were evaluated in each case. Evaluations were performed with leave-one-out cross validation and a support vector machines as a classifier (see Materials and Methods). Number of errors achieved (expressed as percentage) by the best and worst combination for all possible number of peptides were used simultaneously for ARS constructions. As depicted in Fig 9A–9C, the use of peptide combinations usually produced better results than individual peptides. For comparison, we produced an image of ARS using peptide combinations and cysts protein extracts (Fig 9D and 9E). Using this approach we were able to discriminate infected versus non-infected pigs. This discrimination resulted in a similar performance to the one obtained using complex crude cysts extracts (Fig 9D and 9E).
Based on our current preliminary results, developing an accurate immunodiagnostic test for cysticercosis, requires a number of specific antigens from cysts of different tissue localizations within the host, as well of antigens from cysts obtained from different geographical areas. In addition, the use of novel-analytical tools such as machine-learning models, can efficiently discriminate between healthy and infected hosts.
In contrast, we tested this approach using sera from non-cysticercotic (n = 8) and neurocysticercotic patients (n = 12). As expected, our ARS approach produces better results than using a single peptide or peptide mixture. However, sensitivity was about 75% (S4 Fig) indicating that selection of adequate peptides/proteins for diagnosis in humans will require separate studies.
Several studies have recently focused on deciphering the proteome in Taeniid parasites. The proteomes of the whole larva, the protoscolex, the pre-adult stage and some immunogenic proteins have being characterized for Echinococcus spp. [83–87]. The proteome of T. solium has been less explored, although the composition of the excretion/secretion products [88], the proteins of activated oncospheres [89] and a small group of immunogenic proteins have been reported [62]. Moreover, an algorithm for the identification of unique mass spectra for Taeniid parasites has been developed [90]. In addition to the fact that these are still initial efforts, a relevant aspect that remains uncharacterized refers to the proteomic changes associated with the tissue localization of the cysts in the host. Knowing these changes might be essential to understand the tissue preference of the cysts. Although T. solium cysts can establish in a variety of tissues they appear to show preference towards the skeletal muscles and several locations in the central nervous system, including the brain.
In this report, we describe the proteome of T. solium cysts obtained from CNS and SM of infected pigs. We used state-of-the-art quantitative isobaric proteomics to identify and quantify more than 4,200 proteins in a single assay. This is the largest number of identified and quantified proteins so far described for a cestode parasite. A challenging finding is the high amount of host proteins in all crude extracts of T. solium cysts [60–63]. The presence of intact host proteins in the extracts from cestode parasites has been known for six decades [60–64, 91]. Some recent reports on Echinococcus spp. proteomics have also identified a variety of host proteins in the protein extracts of this cestode, for example, 43 proteins were identified in the cysts fluid [27] and up to 293 proteins were identified as excretion/secretion (E/S) products [59]. In our study, we were able to quantify 891 proteins of host origin, the highest number of host proteins identified for a cestode parasite, which brings back the interest on the role of host proteins in the cyst’s physiology. However, if Taenia spp. contains more host proteins than Echinococcus spp. remains to be elucidated; both studies were performed using different proteomic (as well as sampling) strategies. Here, we benefited from high throughput and state-of-the-art multiplexed proteomics that enabled us to identify a significant number of cyst and host proteins.
Herein, one out of each five proteins identified resulted of host origin. These identified host proteins are involved in a number of metabolic, physiologic, signaling and regulatory processes for the pig. It is worth remembering that the T. solium genome revealed a greatly simplified organism, lacking a number of metabolic processes (biosynthesis of amino acids, fatty acids, etc.) as a result of its evolutionary adaptation to parasitism [14]. It is conceivable that the host proteins present in the cysts (associated with metabolic and signaling functions) could play a role for the parasite, beyond being a mere source of amino acids. In the case of E. granulosus, the identified host proteins were also associated with metabolic processes, response to stimuli and regulation of biological processes [26]. It is conceivable that some host proteins retain their function and could play a role on metacestode physiology.
In order to explore this idea, we carried out functional assays and tissue localization studies for a group of host proteins. For example, highly abundant host proteins like albumin and IgG were found in all protein fractions obtained from the cysts, indicating that their presence is ubiquitous in parasite’s tissues and fluid. Another example were LDL and hemopexin, which were found in the cyst’s tissue but were scarce in the vesicular fluid. Cestode parasites have a reduced capacity for lipid biosynthesis [14]. In the case of Schistosoma mansoni (trematode), several proteins have been identified as LDL-binding proteins [92, 93] and LDL was found associated with parasite’s tegument [94]. In our study, we found the LDL protein associated with cysts tissue. However, if Taenia spp. parasites have a subset of specialized proteins to bind and uptake LDL from the host remains to be seen.
Hpx was abundant in the cysts tissue extracts, while scarce in the VF; interestingly, in the soluble fraction of cysts tissue, the Hpx band showed a slight increase in the apparent molecular weight, while in the insoluble fraction the band was detected at the same molecular weight as in the control serum. In the case of Hp and Hb, several bands were detected (especially for Hp); this finding can be explained by the presence of Hp in different forms: the free form, as well as in complexes with Hb. Furthermore, some bands could be the result of Hp/HpHb complex degradation by cysts proteases. We have recently described that intact and functional Hp are present in the cyst’s tissue and could be associated with the iron uptake by the cysts [64]. Here we widen the scope of our investigation on the possible involvement of host proteins in the cyst’s iron metabolism. We carried out tissue localization studies for several host proteins associated with the iron metabolism (hepcidin, ferritin, hemoglobin and haptoglobin). The four proteins were detected in the cyst’s tissue, being Hb and Hp the most abundant and widely distributed within the cysts, suggesting that a relevant portion of iron uptake by the larvae might be supported by these host proteins.
Hepcidin is a master regulator of iron metabolism produced by the liver and its active form is a peptide of 25 amino acids. Hepcidin binds to ferroportin and induces its lysosomal degradation, thus decreasing the iron export by the target cell [95]. However, hepcidin signaling appears to be restricted to mammals [96]; at least, no reports about a cestode homologue ferroportin are available; future studies are need to explore the role (if any) of hepcidin in cestodes biology. On the other hand, ferritin is considered the major iron storage protein [97]. Using specific antibodies, we found ferritin present in the subtegumentary tissue of cyst bladder wall, suggesting that it could also play a role for the parasite, the bands that were detected in protein extracts of cysts had a decrease in the molecular weight (compared with the control). However, after searching the T. solium genome database for ferritin, we found that the cyst has two homologues with a very similar predicted molecular weight, therefore, host and parasite ferritins appears to be undistinguishable in molecular size (≈20 kDa). Therefore, it is possible that the immuno-localized ferritin is a combination of host and cysts ferritin stocks. Whatever the source of ferritin is, its localization in close contact with the host tissue suggests that cysts accumulates specialized molecules for iron storage.
Albumin appears to be involved in the maintenance of parasite’s osmotic pressure in the vesicular fluid, fulfilling a similar function to the role it plays for the host [61]. In addition, immunoglobulins have been proposed as a source of amino acids for the cysts [66, 67].
We determined the antigen-binding activity of host IgG purified from total protein cyst’s extracts through ELISA and western blotting, testing their ability to react with cysts antigens using two protein crude extracts. Purified IgG from the cysts showed a clear antibody activity through the recognition of several antigenic bands (those bands were also shared when sera from cysticercotic pigs were used), suggesting that at least a part of the uptaken host IgG were specific antibodies directed against cysts proteins.
Several other host proteins could also retain their function. Many other uptaken host proteins could play a physiological role, for the parasite. It could also be that potentially deleterious host proteins are simply removed from the host-parasite interface. Ascertain if the host proteins play functions in the physiology of the cysts is an open area of research that could disclose a number of unexpected results.
With respect to the T. solium cysts proteins identified and quantified in our proteomic assays, the changes found between CNS and SM cysts were discrete; more than the 90% of the identified and quantified proteins (>3,100) were grouped within a fold change of -1 and 1. A tissue-enriched protein pattern (including cysts and host proteins) was associated to each cysts tissue localization. These protein patterns could represent a homeostatic adaptation to the biochemical conditions in different tissue environments (SM vs CNS). Our dataset was compared with others previous proteomic reports for helminths [26–28; 59]. From our dataset, 167 proteins were considered excretion/secretion proteins (compared with the T. solium theoretical secretome [59]) (S6 Fig, and S6 Data). In addition, almost 30% of the T. solium gene products were considered ‘hypothetical’, meaning that those genes could not be functionally annotated. Among those hypothetical proteins, the expression of 357 proteins was validated here (S6 Fig and S6 Data). After comparison with the proteomes of E. granulosus, E. multilocularis and M. corti, 14 proteins were common to the four cestodes (S6 Data); those proteins included: a 14-3-3 family member, a fatty acid binding protein, a protein with EGF domain, a lactate dehydrogenase, etc. Indicating that the proteome of cestode parasites are usually highly complex mixtures of parasite and host proteins. The relative consistence of high amounts of host proteins and the relevance of those common proteins identified between the four cestodes need future investigation.
After this initial characterization, several antigens were found enriched in the SM localization of the cysts (paramyosin, a calcium-binding protein, E/S antigens, etc.).
Similarly, other proteins belonging to different families (tetraspanins and GP50s antigens) were also differentially found between CNS and SM cysts. Tetraspanins are integral membrane proteins directly exposed to the host [70, 71] and have been considered vaccine candidates in several helminth infections [98–100]. However, since tetraspanins are highly polymorphic, vaccination trials have produced controversial results [101]. For T. solium, the tetraspanin family has been poorly characterized, only one member (T24) with a good performance as a diagnostic antigen [102], has been described. In this report, five members of the tetraspanin family were quantified; two were associated with the CNS localization of the cysts and one with the SM localization. To investigate if those proteins are antigens during cysticercosis, two peptides were chosen from the amino acid sequence of either TsM_000744700 or TsM_001075800. The four peptides were recognized by antibodies occurring in the sera of naturally infected pigs (four pigs showed a significant recognition of both proteins). The strongest recognition was associated with p1075800-2, derived from the SM-enriched tetraspanin. Regarding T. solium T24 tetraspanin, glycosylation strongly influenced its antibody recognition [102]. It remains to be seen whether the antibody-recognition of the tetraspanins studied here can be increased using recombinant and glycosylated forms.
The diagnostic antigen GP50, is a GPI-anchored glycoprotein with affinity to Lens culinaris lectin, this protein is a promising candidate agent for the immunodiagnosis of NC. Nevertheless, it showed a poor performance when sera from patients having a single viable cyst in the CNS were tested [103, 104]. Two GP50 proteins were found among our proteomic data (S5 Fig). The “canonical” GP50 was associated with the SM localization of the cysts, while a “truncated” form was slightly increased in the CNS cysts. The truncated CNS-abundant GP50 lacked a predicted GPI-anchoring site, in addition to less disulfide bonds and predicted N-linked glycosylation sites. A diagnostic test based on the combination of these two GP50s deserves further study.
We hypothesized that several tissue-enriched antigenic proteins can be used as markers for the tissue localization of cysts. This idea was explored through a combination of theoretical and experimental approaches. Initially, 261 proteins with significant fold changes were selected for epitope prediction using two algorithms; 40 proteins with high antigenic or high epitope content were found enriched either in SM or CNS cysts. Ten proteins were then selected for experimental testing through the synthesis of a single antigenic peptide chosen for each protein. All peptides were recognized by antibodies in the sera of infected animals, but with great variability. When tested with sera from cysticercotic and non-cysticercotic pigs, 9 out of 19 peptides were significantly recognized (p value < 0.01) by the sera from infected animals. The only exception was the peptide derived from a pinin: p455200. These results support the existence of tissue-enriched antigens. The immunodiagnostic potential of those peptides was also tested: peptides were used alone or in mixtures for recognition by the same group of cysticercotic and non-cysticercotic pig’s sera. The best results were obtained with several peptide combinations; in fact, the best combination included peptides derived from SM and CNS-enriched cysts proteins. This can only be considered as an initial study about the potential utility of tissue-enriched antigens; future studies are being conducted using the full recombinant proteins to increase the sensitivity of the immunodiagnostic tests.
A number of antigens have been tested as diagnostic agents. We hypothesized that testing multiple antigens, could produce a better strategy to distinguish between healthy and cysticercotic pigs. A machine-learning strategy was employed and 540,701 combinations were analyzed. Peptide mixtures produced better results than individual peptides. Using five mixtures of peptides allowed to discriminate between cysticercotic and non cysticercotic sera. Our results suggested that, at least five different antigenic determinants are required in order to develop an efficient diagnostic test, able to differentiate between the two groups of sera. Unfortunately, as those synthetic peptides were tested with sera from naturally infected pigs, we do not have relevant information about those infections, such as primary vs secondary infection, co-infections, and nutritional status.
The tissue-dwelling, larval phase of cestodes is usually characterized by host immunomodulatory activities [105]. The observation that some SM and CNS enriched proteins were recognized by antibodies present in the sera of infected animals, could have implications on our understanding about the modulation of the host immune response by cestode parasites.
High-throughput proteomics using a TMT-multiplexed strategy allowed the identification and quantification of over 4,200 proteins across the nine samples of T. solium cysts. The T. solium cyst’s proteome constitutes a mixture of host and parasite proteins (one of each 5 proteins were of host origin). T. solium cysts obtained from either naturally or experimentally infected pigs have several proteins and antigens enriched for SM or CNS localizations. The identified host proteins were highly diverse and were involved in a number of metabolic and signaling processes. Through immuno-localization studies carried out for several host proteins, we found that they are localized in a variety of cysts tissues (tegumentary or subtegumentary tissues in the bladder wall or in the scolex). Other host proteins were detected in the vesicular fluid. Here, we also showed that several host proteins are uptaken intactly, as an example, IgG retained their antige-binding activity. Exploring the functional activity of host proteins in the cyst’s tissue certainly deserves further studies.
The parasite’s antigens that were found enriched for a certain tissue, could be used for the design of highly effective immunodiagnostic methods by combining the peptides/proteins derived from SM and CNS enriched antigens. Using several peptide mixtures and machine-learning models we were able to distinguish between cysticercotic and non cysticercotic pigs with an efficiency that is comparable to the current diagnostic methods using complex cyst’s crude extracts, however, the appropriate subset of tissue-enriched antigens remains to be identified. Development of an optimal immunodiagnostic test for human and porcine cysticercosis requires the use of SM and CNS enriched antigens; variation of those antigens in the cysts isolated from different endemic areas remains to be analyzed, though. This assessment could be crucial for the improvement of the current diagnostic tests. Characterization of the antigenic proteins enriched with each tissue localization of the cysts, is worth studying not only for the design of more efficient immunological tests for human and porcine cysticercosis, but also because they could be involved in complex tissue specific immunomodulatory processes.
|
10.1371/journal.pntd.0000350 | Mefloquine—An Aminoalcohol with Promising Antischistosomal Properties in Mice | The treatment and control of schistosomiasis, an often neglected tropical disease that exacerbates poverty, depends on a single drug, praziquantel. The large-scale use of praziquantel might select for drug-resistant parasites, hence there is a need to develop new antischistosomal compounds. Here, we report that the antimalarial drug mefloquine possesses promising antischistosomal properties in mice.
A single dose of mefloquine (200 or 400 mg/kg) administered orally to mice infected with adult Schistosoma mansoni or adult S. japonicum resulted in high or complete total and female worm burden reductions (72.3%–100%). Importantly, high worm burden reductions were also observed for young developing stages of S. mansoni and S. japonicum harbored in the mouse. Both mefloquine erythro-enantiomers resulted in high and comparable total and female worm burden reductions when given to mice with either a sub-patent or patent S. mansoni infection.
Our findings hold promise for the development of a novel antischistosomal drug based on an aminoalcohol functionality. Further in vitro and in vivo studies have been launched to elucidate the possible mechanism of action and to study the effect of mefloquine on S. haematobium and other trematodes. It will be interesting to investigate whether mefloquine, which is widely and effectively used for the treatment of malaria, has an impact on schistosomiasis in areas where both malaria and schistosomiasis co-exist.
| Schistosomiasis is a chronic and debilitating disease that occurs in tropical and subtropical areas. The disease is caused by an infection with a parasitic worm and affects over 200 million people. The treatment and control of schistosomiasis relies on a single drug, praziquantel. This drug is increasingly used, and hence there is mounting concern about the development of resistance to praziquantel. Here we report that mefloquine, a marketed drug for prophylaxis and treatment of malaria, shows promising antischistosomal properties in laboratory studies with mice. When mefloquine was orally administered at a single dose of 200 or 400 mg/kg to mice infected with young or adult stages of the parasitic worm Schistosoma mansoni or S. japonicum, we found very high worm burden reductions. We also found high worm burden reductions when mefloquine enantiomers were given to mice infected with adult S. mansoni. Further studies are needed because our results might be of public health relevance. Indeed, mefloquine is widely used for prophylaxis and treatment of malaria, often in areas where both malaria and schistosomiasis co-exist. In such areas, it might be possible that the use of mefloquine against malaria reduces the burden of schistosomiasis.
| Schistosomiasis is a chronic and debilitating disease that exacerbates poverty [1],[2]. Although close to 800 million individuals are at risk of contracting the disease and over 200 million people are thought to be infected, schistosomiasis is often neglected [3],[4]. The causative agent is a digenetic trematode of the genus Schistosoma. The three main species parasitizing humans are S. haematobium, S. japonicum, and S. mansoni. Morbidity due to schistosomiasis includes hepatic and intestinal fibrosis (S. mansoni and S. japonicum), and ureteric and bladder fibrosis and calcification of the genitourinary tract (S. haematobium) [5]. The global burden of schistosomiasis has been estimated at 1.7 to 4.5 million disability-adjusted life years (DALYs), but even the higher estimate might be an underestimation of the true burden [2],[6],[7].
The treatment and control of schistosomiasis virtually relies on a single drug, praziquantel. The pressing need to develop new antischistosomal compounds has been emphasized [8]–[11], particularly in view of blanket application of praziquantel within the frame of ‘preventive chemotherapy’ [12], a strategy that might select for drug-resistant parasites. Additionally, there is an important deficiency in the therapeutic profile of praziquantel. The drug targets the adult worm, but has only minor activity against the young developing stages (i.e., schistosomula); hence, retreatment is necessary to kill those parasites that have since matured. There is no dedicated drug discovery and development program pursued for schistosomiasis, either by the pharmaceutical industry or through public-private partnerships. However, despite the paucity of a concerted effort to develop novel antischistosomal drugs, a number of compounds with promising antischistosomal properties have been identified by academia, such as the synthetic trioxolanes [13], the cysteine protease inhibitor K11777 [14], alkylaminoalkanethiosulfuric acids [15], praziquantel analogs [16] and, most recently, the oxadiazoles [10]. Nonetheless, to develop a new antischistosomal drug from lead drug candidates will take at least another decade. Underlying reasons are the scarce resources available for schistosomiasis and other neglected tropical diseases and the high failure rates of compounds during preclinical and clinical testing [17].
Interestingly, the artemisinins (e.g., artemether and artesunate), which are essential components of malaria treatment and control [18], also possess antischistosomal properties [19],[20]. Detailed in vivo studies revealed that schistosomula are particularly susceptible to the artemisinins, whereas moderate worm burden reductions are still apparent for adult worms [21]. A number of clinical trials carried out in different African settings confirmed that both artemether and artesunate have an effect against patent infections with S. haematobium and S. mansoni [20],[22]. Artemisinin-based combination therapies (ACTs) have been adopted as first-line drugs for uncomplicated Plasmodium falciparum malaria in most malaria-endemic countries as a strategy to avoid the selection of parasite drug resistance [18]. Since large parts of Africa are co-endemic for malaria and schistosomiasis [22] and both Plasmodium and Schistosoma parasites degrade hemoglobin, a putative target for several antimalarial drugs, we were motivated to test other antimalarials that are commonly employed in combination with an artemisinin derivative for their potential antischistosomal activities. We then followed up on the promising in vivo activity of mefloquine, first to elucidate the dose-response relationships of single-dose mefloquine against juvenile and adult S. mansoni and S. japonicum and, second, to assess the stage-specific susceptibility of both parasites to mefloquine. Finally, we evaluated the activity of mefloquine (+) and (−) erythro enantiomers against different developmental stages of S. mansoni.
Halofantrine, mefloquine HCl, and mefloquine enantiomers were obtained from Hoffmann La Roche (Basel, Switzerland); pyronaridine was provided by the National Institute of Parasitic Diseases, Chinese Center for Disease Control and Prevention (Shanghai, China); and pyrimethamine, sulfadoxine, and sulfamethoxypyrazine were obtained from Dafra Pharma (Turnhout, Belgium). For the in vivo studies with S. mansoni, lumefantrine was provided by the Novartis Institute for Biomedical Research (Basel, Switzerland); amodiaquine, chloroquine, and quinine were purchased from Sigma (Buchs, Switzerland); and atovaquone was purchased in a local Swiss pharmacy (Wellvone®). For the in vivo studies with S. japonicum, lumefantrine was obtained from Kunming Pharmaceutical Corporation (Kunming, China); and amodiaquine, atovaquone, chloroquine, and halofantrine were provided by the National Institute of Parasitic Diseases, Chinese Center for Disease Control and Prevention (Shanghai, China).
All drugs were prepared as suspensions in 7% (v/v) Tween 80 and 3% (v/v) ethanol before oral administration to mice (10 ml/kg).
Experiments with S. mansoni (Liberian strain) were carried out at the Swiss Tropical Institute (Basel, Switzerland), in accordance with Swiss national and cantonal regulations on animal welfare (permission no. 1731). Female mice (NMRI strain, n = 290, weight ∼20–22 g) were purchased from RCC (Itingen, Switzerland). Mice were kept under environmentally-controlled conditions (temperature ∼25°C; humidity ∼70%; 12-hour light and 12-hour dark cycle) and acclimatized for one week before infection. The animals had free access to water and rodent diet.
The experiments with S. japonicum (Anhui strain) were undertaken at the National Institute of Parasitic Diseases, Chinese Center for Disease Control and Prevention. Male mice (Kunming strain, n = 125, weight ∼20–22 g) were purchased from Shanghai Experimental Animal Center of the Chinese Academy of Sciences (Shanghai, China).
Cercariae of S. mansoni and S. japonicum were obtained from infected intermediate host snails in our laboratories as described previously [13].
Each mouse was infected subcutaneously with ∼80 S. mansoni cercariae. Twenty-one days (pre-patent infection) and 49 days (patent infection) after the experimental infection, groups of 3–5 mice were treated orally with the drugs to be tested at single oral doses (25–400 mg/kg). To study the stage-specific susceptibility of S. mansoni, mice were treated with a single 400 mg/kg oral dose of mefloquine either 2 days or 1 day before infection, 3 hours after infection or at days 7, 14, 21, 28, 35, 42, and 49 post-infection. For the hepatic shift experiments groups of mice were treated with 400 mg/kg mefloquine 49 days post-infection, and the worm distribution was analyzed on days 1, 3, 7 and 14 post-treatment. Results of the hepatic shift experiments are used as an additional criterion for evaluating antischistosomal drugs since this test shows how quickly the forced dislodgement of worms occurs [23]. For each experiment, groups of 5–10 untreated mice served as controls.
At 21 days post-treatment, animals were killed by the CO2 method and dissected, and worms were sexed and counted as described elsewhere [13]. In vivo experiments with S. mansoni were carried out in duplicates. The results from the second set of experiment are summarized in Supporting Tables S1, S2, and S3.
Mice were infected percutaneously with ∼40 S. japonicum cercariae each. To investigate the dose-response relationship of mefloquine against juvenile and adult S. japonicum, single 25–400 mg/kg oral doses were given to mice 14 days (pre-patent infection) and 35 days (patent infection) post-infection. To assess the efficacy of mefloquine against different stages of S. japonicum, mice were treated with a single oral dose of 400 mg/kg mefloquine 2 days or 1 day before infection, 3 hours after infection, and at days 3, 7, 14, 21, 28 and 35 post-infection. In each experiment, infected but untreated mice served as controls.
Twenty-one days post-treatment, mice were killed and the worms recovered from the hepatic and portomesenteric veins by the perfusion technique [24]. To study the hepatic shift in adult S. japonicum, groups of mice were treated with 400 mg/kg mefloquine 35 days post-infection, and the worm distribution was analyzed on days 1, 3, 7, and 14 post-treatment.
For statistical analysis we used version 2.4.5 of the Statsdirect statistical software package (Cheshire, United Kingdom). The Kruskal-Wallis (KW) test, which compares the medians of the responses between the treatment and control groups, was used. A difference in median was considered to be significant at a significance level of 5%.
The in vivo antischistosomal efficacy of 11 antimalarial drugs is summarized in Tables 1 and 2. Drugs were administered orally at a single dose of 400 mg/kg to mice harboring adult S. mansoni, and worm burden reductions, including changes in worm distributions, were assessed. Amodiaquine, atovaquone, lumefantrine, pyrimethamine, pyronaridine, sulfadoxine, and sulfamethoxypyrazine showed no antischistosomal activity. Quinine and halofantrine resulted in total and female worm burden reductions ranging between 51.7% and 74.9% and changes in the worm distribution. The highest activity (total and female worm burden reduction of 77.3% and 100%, respectively) was observed with a single dose of mefloquine (400 mg/kg), which was statistically significant (p<0.05). The chemical structures of the four aminoalcohols quinine, halofantrine, lumefantrine, and mefloquine are shown in Figure 1.
In view of the promising antischistosomal activity of mefloquine, its properties were further characterized, with an emphasis on dose-response relationships in juvenile (21-day-old) and adult (49-day-old) S. mansoni harbored in mice (Table 3, Table S1). In the juvenile infection model, total and female worm burden reductions of 94.2–100% were achieved with a single-dose oral regimen (100 mg/kg and above). At a dose of 50 mg/kg, the total and female worm burden reductions were 30.8% and 38.3%, respectively. At the lowest dose investigated (25 mg/kg) mefloquine showed no effect on juvenile S. mansoni in the mouse. The difference in total and female worm burdens between mice infected with 21-day-old juvenile S. mansoni that were treated (25–400 mg/kg) and those mice left untreated was highly significant (KW = 9.51, p = 0.002 and KW = 8.16, p = 0.004, respectively).
Oral administration of mefloquine at a single dose (200 mg/kg and 400 mg/kg) to mice infected with adult S. mansoni resulted in total and female worm burden reductions of 72.3–100%. No or only moderate total and female worm burden reductions (4.9–56.3%) were achieved with a single dose of 25, 50 or 100 mg/kg mefloquine. There was a highly significant difference between the total and female worm burden of mefloquine-treated mice (25–400 mg/kg) and control mice in the adult infection model (KW = 12.49, p<0.001 and KW = 9.46, p = 0.002, respectively).
Tables 4 and Table S2 summarizes the activity of mefloquine when given 2 days or 1 day before infection, shortly after infection (3 hours post-infection) and until 49 days post-infection. These experiments were carried out with a single oral dose of 400 mg/kg mefloquine as this dose achieved the highest reductions in worm burden against juvenile and adult S. mansoni. A single oral dose of mefloquine was highly active against mice harboring either a 7-, 14-, 21-, 28-, 35-, 42-, or 49-day-old S. mansoni infection (total and female worm burden reductions ranged between 83.9% and 100%). Mefloquine administration to mice 2 days or 1 day before infection or 3 hours after infection showed moderate total and female worm burden reductions (35.9–46.5%). Regardless of the timing of mefloquine administration, i.e., shortly before infection or administration when mice were infected with juvenile or adult S. mansoni, total and female worm burden reductions were highly significant (p<0.001).
Both (−)(11S, 2′R) and (+)(11R,2′S) erythro-enantiomers of mefloquine resulted in high and comparable total and female worm burden reductions (p<0.001) when given to mice infected with either juvenile or adult stages of S. mansoni. At a single dose of 100 mg/kg and above, mefloquine (−)(11S, 2′R) achieved worm burden reductions of 57.1% to 100% in mice harboring 21-day-old juvenile and 49-day-old adult S. mansoni. At a single dose of 50 mg/kg, total worm burden reductions of 8.7% to 19.9% were observed in the juvenile and adult infection model, respectively (Table 5).
Mefloquine (+)(11R,2′S) yielded total and female worm burden reductions of 64.0% to 100% in mice harboring 21-day-old juvenile and 49-day-old adult S. mansoni at a single dose of 100 mg/kg and above. At the lowest dose investigated, 50 mg/kg of mefloquine (+)(11R,2′S), we still observed moderate total and female worm burden reductions (36.0–49.1%).
Table 6 summarizes the dose-response relationship of mefloquine against 14-day-old juvenile and 35-day-old adult S. japonicum in the mouse model. At the lowest dose investigated (25 mg/kg), total and female worm burden reductions of 16.9% and 10.9%, respectively, were observed against juvenile S. japonicum. Slightly higher total and female worm burden reductions (48.9% and 43.6%, respectively) were found when the dose of mefloquine was doubled (50 mg/kg). Administration of single doses of 100 mg/kg to 400 mg/kg achieved high total and female worm burden reductions of 86.7% and 95.1%. There was a highly significant difference (p<0.001) in the medians of the total and female worm burden between treated (50–400 mg/kg) and untreated control mice that were infected with S. japonicum 14 days previously.
Administration of mefloquine at a single-dose of either 25 mg/kg or 50 mg/kg to mice infected with adult S. japonicum resulted in total and female worm burden reductions ranging between 16.7% and 39.7%. A higher dose of mefloquine (100 mg/kg) given to mice harboring adult S. japonicum resulted in total and female worm burden reductions of 70.9% and 73.7%, respectively. Total and female worm burden reductions of 87.5% to 100% were obtained with mefloquine at 200 mg/kg and 400 mg/kg in the adult infection model. Total worm counts recovered from treated (25–400 mg/kg) mice were significantly different from non-treated control mice infected with adult S. japonicum (KW = 10.86, p = 0.001 and KW = 9.40, p = 0.002, respectively).
A single oral dose of 400 mg/kg mefloquine was used to investigate the stage-specific susceptibility of S. japonicum because this dose showed the highest activities against juvenile and adult stages of S. japonicum in our previous experiments. Mefloquine achieved high total and female worm burden reductions ranging between 77.3% and 100% when administered to mice infected with S. japonicum for 3 to 35 days (Table 7). Lower total and female worm burden reductions (26.3–49.2%) were observed when mefloquine was administered 1 or 3 days before infection (days 1 and 3 prior to infection) or shortly after infection (3 hours post-infection). The difference in total and female worm burden reductions between treated and untreated control mice was highly significant regardless of the point in time mefloquine was administered (both p<0.001).
In Table 8 and Table S3 we summarize the distribution of adult schistosomes in the liver and mesenteric veins on days 1, 3, 7, and 14 following mefloquine administration in S. mansoni- and S. japonicum-infected mice. The hepatic shift commenced one day post-treatment, and nearly all worms had shifted to the liver on day 3 post-treatment in both infection models. The worm size was smaller, and the majority of worm pairs were separated. On day 14, the majority of the worms had been eliminated.
We report promising antischistosomal properties of mefloquine, a marketed drug for prophylaxis and treatment of malaria. Oral administration of a single dose of mefloquine (400 mg/kg) to mice infected with either juvenile or adult stages of S. mansoni and S. japonicum, two of the three most important schistosome species [5], resulted in very high or complete total and female worm burden reductions. Interestingly, a recent study, which used a lower dose of mefloquine (150 mg/kg), reported no effect on the worm burden, but a reduction in egg fecundity in the first three developmental stages of S. mansoni in the murine model [25]. The discrepancy of the activity of mefloquine, when administered at doses of 100 mg/kg or 200 mg/kg to mice infected with adult S. mansoni observed in our study (worm burden reduction of 45% and 72%, respectively) and the previous investigation (no worm burden reduction) [25] remains to be elucidated. Previous research has shown that mefloquine also exhibits a broad spectrum of antimicrobial activity [26], as well as activity against larval and adult stages of Brugia patei and B. malayi in vitro [27].
If our results can be confirmed in proof-of-concept studies in humans, initially with current antimalarial dosages, it is conceivable that mefloquine can play a role in public health because the drug is widely and effectively used in malaria-endemic settings [28],[29], and because of the fact that malaria and schistosomiasis co-exist over large parts of sub-Saharan Africa and elsewhere [22]. The highest activities in S. mansoni- and S. japonicum-infected mice were observed when mefloquine was given at a single oral dose of 200–400 mg/kg, which correspond to 16–31 mg/kg in humans (dose calculator: http://www.fda.gov/cder/cancer/animalframe.htm). At present, the recommended dosage of mefloquine is 25 mg/kg when used in human treatment of malaria.
In contrast to other recently portrayed schistosomicides such as the oxadiazoles [10] or the cysteine protease inhibitor K11777 [14], which thus far have only been tested intraperitoneally and in multiple doses, mefloquine at a single oral dose resulted in high worm burden reductions. Moreover, the consistently high worm burden reductions observed against all development stages of the schistosome worms in the rodent model seems to be an advantage of mefloquine over praziquantel; the latter only displaying high activity against very young stages (skin penetration) and adult schistosomes [30],[31]. Actually, the minor activity of praziquantel against juvenile (2- to 3-week-old) schistosomes is believed to be a key factor explaining observed treatment ‘failures’ in areas highly endemic for schistosomiasis and that require frequent retreatments [11],[32],[33]. For comparison, the stage-specific susceptibility of praziquantel and mefloquine are juxtaposed in Figure 2. It is evident that mefloquine exceeds benchmark criteria set forth by the World Health Organization (WHO) for highly active lead compounds (defined as worm burden reduction of >80% in the adult S. mansoni-mouse model following five consecutive doses given intraperitoneally or subcutaneously) [34]. We approached or exceeded this benchmark level with a single dose of 400 mg/kg given orally to mice infected with either adult or juvenile stages of S. mansoni and S. japonicum.
Single oral doses of the antimalarials amodiaquine, atovaquone, sulfadoxine, sulfamethoxypyrazine, pyronaridine and pyrimethamine showed no activity in the S. mansoni-mouse model. A single oral dose of chloroquine also had no activity in the S. mansoni-mouse model, while previous work documented antischistosomal properties when the drug was given as multiple intraperitoneal doses [35]. On the other hand, quinine and halofantrine, which are structurally related to mefloquine, exhibited promising antischistosomal properties in the mouse (total worm burden reductions >50% following a single-dose oral regimen). The lack of antischistosomal activity of oral lumefantrine, which also belongs to the class of aminoalcohols, might be explained by the low oral bioavailability of this drug [36]. Further studies with intraperitoneal lumefantrine in S. mansoni-infected mice are ongoing in our laboratories. Interestingly, oxamniquine, a drug that has been widely and effectively used for the treatment and control of schistosomiasis mansoni, particularly in Brazil where more than 12 million people have been administered this drug [37], also contains an aminoalcohol functionality [38].
Importantly, previous investigations on the morphology of schistosomes recovered from host animals after administration of praziquantel [39] and observations made with schistosomes collected from mefloquine-treated mice point to different mechanisms of actions. Results obtained from preliminary morphological investigations (no data shown) indicate that mefloquine exerts a rapid action on schistosomes, resulting in marked alterations of the digestive tract and the reproductive system of the worms. The detailed mechanism of action of mefloquine and related aminoalcohols on schistosomes remains to be investigated. Still today, the exact mechanism of action of mefloquine on Plasmodium is not known, though interference with hemoglobin digestion seems to play a role [40]. It was demonstrated that the antimalarial chloroquine inhibits the formation of hemozoin, a heme detoxification aggregate in S. mansoni female homogenates [35], hence future studies should elucidate whether mefloquine also targets hemozoin formation. It is interesting to note that adult female S. mansoni were more affected by mefloquine at doses of 100–400 mg/kg than male adult S. mansoni. This phenomenon was not observed when treating juvenile stages of either S. mansoni or S. japonicum. Differences in drug susceptibility between male and female S. mansoni have been reported previously, e.g., following hycanthone treatment [41] and point to a sex-specific interference of the drug with the target, or different drug targets.
Interestingly, moderate worm burden reductions were found when a single dose of mefloquine was given one or two days before or three hours after infection of mice with either S. mansoni or S. japonicum. Although mefloquine has a long half-life of 6.5 to 22.7 days in healthy volunteers [42], the half-life of this drug in mice is much shorter, namely 17 hours [43]. Further studies should be launched to clarify whether active metabolites, the significant enterohepatic recirculation of mefloquine found in rodents [44] or a treatment induced alteration of the immune response, which has, for example, been described for praziquantel, [45] may be a contributing factor to the efficacy and be partly responsible for the moderate worm burden reductions recorded when mefloquine is given pre-infection. Pro-inflammatory effects of mefloquine have been described in both S. mansoni-infected and non-infected mice [25].
Mefloquine is generally well tolerated by adults and children. However, there is evidence that mefloquine may result in harmful events in the gastrointestinal and central nervous systems. Observed adverse events include insomnia, nausea, vomiting, diarrhea, headache, dizziness, rash, pruritus, and abdominal pain. Severe neuropsychiatric symptoms as seizures and hallucinations are rare (occurring in 1 per 10,000 patients) [46]. Adverse events seem to be dose-dependent, and it was suggested that doses greater than 15 mg/kg should be divided [42]. To date, the neurotoxicity of mefloquine cannot be explained, and it remains to be elucidated whether there is steroselectivity in the neurotoxic potencies of the enantiomers. However, the brain penetration of the (+) enantiomer was found to be much higher than that of the (−) enantiomer in two post-mortem human cerebral biopsies [47]. We have demonstrated that both enantiomers and the racemic hydrochloride possess a similar activity against juvenile and adult S. mansoni. Interestingly, this finding contrasts with the activity of mefloquine in Plasmodium yoelii-infected mice, where the racemic hydrochloride of mefloquine showed a two-fold to three-fold higher activity when compared to the enantiomers, which had similar activity to each other [48].
In conclusion, we have documented promising in vivo antischistosomal efficacy of the antimalarial drug mefloquine. New studies have been launched with an aim to elucidate the effect of mefloquine on the third major schistosome species, i.e., S. haematobium, and a number of the biologically-related trematodes such as Clonorchis sinensis and Opisthorchis viverrini. Additionally, in vitro studies and an evaluation of the potential of praziquantel-mefloquine combination therapy have been initiated. Finally, a proof-of-concept study will be launched in an African setting, where malaria and schistosomiasis co-exist, similar to our preceding work with the artemisinins [49],[50]. A word of caution should be mentioned here, as the malaria community has argued that antimalarial drugs should not be used against schistosomiasis because of concern that this strategy might select for Plasmodium-resistant parasites. However, millions of people have been, or will be, treated with mefloquine and mefloquine-artesunate combinations in areas where both malaria and schistosomiasis co-exist [22]. Hence, the potential ancillary benefit of the antimalarial drug mefloquine should be investigated against schistosomiasis.
|
10.1371/journal.ppat.1003370 | Crosstalk between the Circadian Clock and Innate Immunity in Arabidopsis | The circadian clock integrates temporal information with environmental cues in regulating plant development and physiology. Recently, the circadian clock has been shown to affect plant responses to biotic cues. To further examine this role of the circadian clock, we tested disease resistance in mutants disrupted in CCA1 and LHY, which act synergistically to regulate clock activity. We found that cca1 and lhy mutants also synergistically affect basal and resistance gene-mediated defense against Pseudomonas syringae and Hyaloperonospora arabidopsidis. Disrupting the circadian clock caused by overexpression of CCA1 or LHY also resulted in severe susceptibility to P. syringae. We identified a downstream target of CCA1 and LHY, GRP7, a key constituent of a slave oscillator regulated by the circadian clock and previously shown to influence plant defense and stomatal activity. We show that the defense role of CCA1 and LHY against P. syringae is at least partially through circadian control of stomatal aperture but is independent of defense mediated by salicylic acid. Furthermore, we found defense activation by P. syringae infection and treatment with the elicitor flg22 can feedback-regulate clock activity. Together this data strongly supports a direct role of the circadian clock in defense control and reveal for the first time crosstalk between the circadian clock and plant innate immunity.
| Plants are frequently challenged by various pathogens. The circadian clock, which is the internal time measuring machinery, has been implicated in regulating plant responses to biotic cues. To better understand the role of the circadian clock in defense control, we tested disease resistance with Arabidopsis mutants disrupted in CCA1 and LHY, two key components of the circadian clock. We found that consistent with their contributions to the circadian clock, cca1 and lhy mutants synergistically affect resistance to both bacterial and oomycete pathogens. Disrupting the circadian clock caused by overexpression of CCA1 or LHY also results in severe disease susceptibility. Thus, our data further demonstrate a direct role of the circadian clock mediated by CCA1 and LHY in defense regulation. We also found that CCA1 and LHY act independently of salicylic acid mediated defense but at least through the downstream target gene GRP7 to regulate both stomata-dependent and -independent pathways. We further show that defense activation by bacterial infection and the treatment with the elicitor flg22 can also feed back to regulate clock activity. Together our study reveals for the first time reciprocal regulation of the circadian clock and plant innate immunity, significantly expanding our view of complex gene networks regulating plant defense responses and development.
| Plants are challenged by various pathogens on a daily basis. Accumulating evidence implicates a role of the circadian clock in regulating plant innate immunity. The circadian clock is the internal time measuring machinery important for plant growth and development. However, our understanding of the molecular basis of how the circadian clock controls plant innate immunity is still in its infancy.
Plants have evolved various mechanisms, some pre-formed and others induced, to ward off pathogen invasion. An example of pre-formed surface structures is the stomate, the natural opening important for photosynthetic gas exchange. This opening can provide a portal for pathogens to enter leaves; however, plants can also control the aperture of stomata to physically limit pathogens [1], [2]. One type of induced defense is activated when plants recognize pathogen-associated molecular patterns (PAMPs), which are conserved molecules or structures present in groups of related microbes. This defense, also termed PAMP-triggered immunity (PTI), can be highly effective against non-adapted pathogens and provides a basal level of defense even against adapted pathogens [3], [4]. Another type of induced defense is activated by plant resistance (R) proteins, which specifically recognize secreted pathogen effectors and subsequently activate effector-triggered immunity (ETI). ETI, also termed R gene-mediated resistance, is a stronger and faster elaboration of PTI, and frequently results in hypersensitive cell death at the infection site [5], [6], [7]. The small molecule salicylic acid (SA) has been linked to signal transduction in PTI and ETI [8], [9], [10].
The circadian clock has profound influence on the fitness of organisms [11], [12], [13], [14], [15], [16]. The core of the circadian clock is the central oscillator, which in Arabidopsis, is composed of multiple interconnected negative feedback loops that orchestrate biological adjustments independently of external stimuli [17], [18]. Of these clock components, CIRCADIAN CLOCK ASSOCIATED1 (CCA1) and its close homolog LATE ELONGATED HYPOCOTYL (LHY) are transcription factors that are involved in multiple feedback loops and function synergistically to regulate clock activity [19], [20], [21].
The role of the circadian clock in controlling plant innate immunity has long been proposed based on circadian-regulation of defense gene expression [22], [23], [24], [25], [26]. Direct evidence from several research groups has recently emerged to support such a role of the circadian clock. Under free running conditions, wild type Arabidopsis exhibits temporal oscillations in susceptibility to Pseudomonas syringae infection, which are disrupted by overexpression of CCA1 [27]. Misexpression of several clock genes, including CCA1, compromises resistance to the bacterial pathogen Pseudomonas syringae and/or to the oomycete pathogen Hyaloperonospora arabidopsidis (Hpa) [27], [28], [29]. Interestingly, although lhy mutants exhibit similarly shortened circadian period as cca1 mutants, LHY was not shown to play a defense role against Hpa [28]. This raises the question of whether CCA1 is a dual function protein, affecting both the circadian clock and other non-clock related processes, as shown in the case of another central oscillator component GIGANTEA [30]. cca1-conferred disease susceptibility might be attributed to a role of CCA1 in regulating non-clock related processes rather than to its direct involvement in the circadian clock [31].
To better understand the role of CCA1 and LHY-mediated circadian clock in defense control, we tested plants misexpressing CCA1 and/or LHY for disease resistance to P. syringae and Hpa. We show that CCA1 and LHY loss-of-function mutants synergistically affect basal resistance and R gene-mediated defense against both pathogens. Disrupting the circadian clock caused by overexpression of CCA1 or LHY also results in severe disease susceptibility to P. syringae. The defense role of CCA1 and LHY against P. syringae is at least partially through circadian control of stomatal aperture but is SA-independent. Furthermore, we found that clock activity is modulated by P. syringae infection or treatment with the elicitor flg22. These data further establish the role of the circadian clock in defense control and for the first time reveal crosstalk between the circadian clock and plant innate immunity.
To evaluate defense roles of CCA1 and LHY, we constructed the cca1-1lhy-20 mutant via a genetic cross in a Col-0 background that also contains the LUCIFERASE reporter gene driven by the CCA1 promoter (ProCCA1:LUC). The single loss of function mutants, cca1-1 and lhy-20, have shortened circadian periods of ProCCA1:LUC expression in constant light (LL) [11]. In LL, we confirmed that cca1-1lhy-20 had a much-shortened period (19.9±0.11 hr), compared with wild type (wt) Col-0 (24.4±0.09 hr) (Figure S1A and [19]). Although experiments in LL are important for establishing the involvement of the circadian clock in specific phenotypes, such experimental conditions can also be limiting. In entraining conditions (e.g., a 12 hr L/12 hr D cycle; LD), the altered period of clock mutants like cca1-1 and lhy-20 is not seen due to the entraining cycle, which imposes a 24 hr period (Figure 1). The clock remains important in such LD conditions, though, because the clock determines the phase of specific events with respect to as dawn and dusk. Mutants with altered period in LL typically exhibit altered phase in LD, with short period mutants exhibiting a leading (early) phase and long period mutants exhibiting a lagging (late) phase [32]. Moreover, interactions between the endogenous circadian clock and external LD cycles can results in phase differences, sometimes dramatic, when measured in LD versus LL. For example, the phase of maximal hypocotyl elongation during early seedling growth was shifted 8–12 hours between LD and LL conditions [33], [34]. In their natural environment, plants do not usually encounter LL. Therefore in evaluating the role of the circadian clock on plant defense against pathogens, it is critically important to study plant-pathogen interactions in LD and to consider the potential influence of the circadian clock on the phases of rhythmic events that might influence the plant response to pathogen challenge.
We show here that in LD the phases of cca1-1 and lhy-20 single mutants were leading with respect to that of wild type Col-0, and that the cca1-1 lhy-20 double mutant exhibited a much earlier phase than either single mutant, consistent with the synergistic contribution of CCA1 and LHY in regulating clock activity (Figure 1 and Figure S1B). Early phase was also reported with other cca1lhy mutants [20], [21]. In addition, we found that plants overexpressing CCA1 (CCA1ox), which display arrhythmic clock activity in LL [35], also showed arrhythmic expression of ProCCA1:LUC in LD with an acute peak in response to lights on (Figure 1 and S1B). Low ProCCA1:LUC activity in CCA1ox is consistent with CCA1 being a negative regulator of its own expression [35]. These results emphasize that altered function of the circadian clock can manifest in both LL and LD conditions.
To test disease resistance of cca1-1 and lhy-20 plants, we performed infection experiments at Zeitgeber Time 1 (Zeitgeber Time is the time relative to dawn; ZT1 is 1 hr after lights on) or ZT13 (1 hr after lights off), two times of day associated with drastic changes of light regime. Plant leaves were pressure-infiltrated with virulent P. syringae pv. maculicola ES4326 strain DG3 (PmaDG3) [36]. The infected plants were placed in either LD or LL. Bacterial growth assays at 3 days post infection (3 dpi) revealed no significant difference among Col-0, cca1-1, lhy-20, and cca1-1lhy-20 in either LD or LL (Figure 2 and Figure S2).
Under natural conditions, P. syringae enters the apoplast of leaves through openings such as stomata and wounds. It is known that stomatal aperture is regulated by the circadian clock [37], [38]. Therefore, infiltration of bacteria directly into plant tissue might bypass the influence of the circadian clock on stomatal defense. To test this possibility, we spray-infected with PmaDG3 Col-0, cca1-1, lhy-20, and cca1-1lhy-20 at ZT1 and ZT13 in LD. We found that Col-0 supported over 10-fold more bacterial growth with ZT1 infection than with ZT13 infection (Figure 3A and 3B), suggesting that Col-0 is more resistant at night than at dawn when spray-infected. Although we did not observe significant difference in bacterial growth between Col-0 and cca1-1 and lhy-20 single mutants, the double mutant cca1-1lhy-20 showed enhanced susceptibility to PmaDG3 when sprayed at ZT13 (Figure 3A to 3C). Consistent with this result, we found that PmaDG6 (an avirulent strain recognized by the resistance protein RPS2 in Col-0) [36]) grew significantly more in cca1-1lhy-20 than in Col-0 and the single mutants with ZT13 infection (Figure 3D and 3E). Together these data suggest that CCA1 and LHY share redundant functions to regulate both basal and RPS2-mediated defense against P. syringae.
To further substantiate the role of CCA1 and LHY in defense regulation, we tested disease resistance of plants overexpressing CCA1 (CCA1ox) or LHY (LHYox), which were shown to have arrhythmic clock activity in LL [35], [39]. CCA1ox plants also exhibited clock arrhythmicity in LD (Figure 1 and S1B). Disease resistance assays indicate that CCA1ox plants were more susceptible to PmaDG3 than Col-0 with infiltration infection in LD or LL (Figure 2 and S2). CCA1ox plants were also more susceptible than Col-0 to PmaDG3 and to PmaDG6 when spray-infected at ZT1 or ZT13 in LD (Figure 3).
LHYox plants are in the Landsberg erecta (Ler) background, with which we used P. syringae pv. tomato DC3000 (DC3000) to test disease resistance because this strain induces stronger disease symptoms in our hands than does PmaDG3. Similar to CCA1ox plants, LHYox plants had more bacterial growth than Ler when infiltrated with DC3000 at ZT1 or ZT13 in LD (Figure 4A). In addition, spray-infection at ZT1 or ZT13 in LD also gave similar results (Figure 4B). Together, disruption of the circadian clock by misexpressing CCA1 and/or LHY compromises disease resistance to P. syringae, supporting a direct role of the circadian clock in defense regulation.
Our data show that cca1-1lhy-20 was more susceptible with spray-infection and CCA1ox and LHYox plants displayed enhanced susceptibility with both spray and infiltration infections. These suggest that both stomata-dependent and -independent defense can be affected by misexpression of either of these two core oscillator genes. Consistent with this notion, a previous study showed that CCA1ox plants had increased CO2 assimilation and stomatal conductance [13]. To further test whether the defense role of CCA1 and LHY is linked to the control of stomatal pore size, we measured plant stomatal aperture at ZT1 and ZT13 in LD. Consistent with Col-0 being more resistant with spray-infection at ZT13 than at ZT1, we found that stomatal aperture of Col-0 was much smaller at ZT13 than at ZT1 (Figure 5A). Compared with Col-0, the cca1-1 and lhy-20 mutants and CCA1ox plants showed similar stomatal aperture at ZT1 but had greater stomatal aperture at ZT13 (Figure 5A). These data suggest that disrupting clock activity mediated by CCA1 and LHY could make plants less responsive to dark-induced stomatal closure at night, thereby enhancing access of P. syringae to the leaf interior.
To further determine how these mutants respond to P. syringae infection, we measured stomatal aperture in the presence of PmaDG3. PmaDG3 treatment was performed at ZT4 after plants had been exposed to light for four hours to ensure the opening of the stomata (Figure S3). At 1 hr post infection (1 hpi), we observed a 48.1% suppression of stomatal aperture in Col-0, compared with mock treatment (Figure 5C top and Table S1). However, this suppression was much reduced in cca1-1 and lhy-20 and largely blocked in cca1-1lhy-20 and CCA1ox. P. syringae-induced stomatal closure was transient since both mock and PmaDG3-treated leaves showed similar stomatal aperture at 3 hpi (Figure 5C bottom). Although exhibiting similar stomatal aperture at ZT1 and ZT13 (Figure 5B), the LHYox plants also showed reduced suppression of DC3000-induced stomatal closure at 1 hpi (16.9%), compared with Ler control (51.6%) (Figure 5D and Table S1). Hence, these results indicate that disrupting the circadian clock by CCA1 and LHY misexpression impairs plants' capacity of inducing stomatal closure in response to P. syringae.
CCA1 but not LHY was previously shown to regulate resistance to the oomycete pathogen Hpa [28]. To test whether a contribution of LHY to Hpa resistance could be discerned in the double mutant cca1-1lhy-20, we sprayed seven-day-old seedlings at ZT7 in LD with the virulent strain Hpa Emco5 or the avirulent strain Hpa Emoy2 (recognized by the R protein RPP4 in Col-0). We observed significantly more susceptibility to both Hpa strains in the cca1-1lhy-20 double mutant, compared to Col-0 and the single mutants (Figure 6A and 6B) while the CCA1ox plants were substantially more resistant to Hpa Emco5 (Figure 6A). Our data are broadly in agreement with those previously reported [28]. The reason that we did not observe a significant difference between Col-0 and cca1-1 could be due to the difference in the infection time and/or Hpa strains used - Wang et al inoculated plants with the avirulent strain Hpa Emwa1 at dawn [28] while we used Hpa Emco5 (virulent) and Emoy2 (avirulent) in the afternoon in our experiments. Nevertheless, these data, together with the P. syringae data described earlier, demonstrate that CCA1 and LHY contribute synergistically to basal resistance and R-gene mediated defense against both bacterial and oomycete pathogens. What surprises us is the difference in response to P. syringae (decreased resistance) and Hpa (enhanced resistance) strains observed in CCA1ox plants. We speculate that there are distinct mechanisms that these plants use to defend against the two pathogens.
Identification of defense-related genes controlled by CCA1 and LHY is critical to gain better understanding of the mechanism of action of CCA1 and LHY in defense regulation. To this end, we analyzed promoters of 571 genes for CCA1-binding site (CBS) and evening element (EE), two cis elements known for CCA1 and LHY binding [40], [41], [42]. These 571 genes had been previously selected to construct mini-microarrays, consisting of three groups, selected (337 defense-related genes based on microarray experiments), empirical (127 empirical marker genes for various pathogen responses), and normalization (107 non-defense related genes whose expression levels were relatively stable among experiments with pathogen infection) [43]. The online tool POBO [44] was used to analyze up to 3000 bp from the promoter regions of these genes, which do not include the coding sequences of neighboring genes, for an enrichment of CBS or EE motifs. The background for this analysis was generated using pseudo-clusters of 100 promoters of up to 3000 bp in length of randomly sampled Arabidopsis genes (1000 bootstrap replications were used in the sampling). Compared with the background, the CBS motif was found as often as expected by chance in the selected and empirical gene promoters (Figure 7A and 7B) but the motifs were found less frequently in the normalization gene promoters (Figure 7C and Table S2). When compared to the normalization genes, there was a greater than 40% increase of the cluster mean for the CBS motif in both selected and empirical genes. These observations suggest that although defense-related genes (selected and empirical genes) are not particularly enriched with the CBS motif, the non-defense related genes (the normalization genes) are slightly depleted of the motif. The enrichment of the EE motif was more pronounced in both selected and empirical genes, with about 200% increase of the cluster means when compared to the normalization genes (Figure 7D–7F and Table S2). Thus, these results suggest that defense-related genes are preferentially regulated by CCA1 and LHY. However, since the sample size in each group is small, caution should be taken when extrapolating this interpretation to the whole genome level.
The frequency of CBS or EE motif per promoter region was quantified from the above three sets of genes (Figure S4). Among the genes analyzed, we found that GRP7 (At2g21660; also known as COLD AND CIRCADIAN REGULATED 2 [CCR2]) [45], [46] had the most overrepresentation of the EE motif, with four EE within a 300 bp promoter region. One CBS motif was also found at 1294 bp of the GRP7 promoter. GRP7 is a key constituent of a slave oscillator regulated by the circadian clock [45], [47] and also has been demonstrated to have roles in regulating floral transition and plant defense [48], [49]. Expression of GRP7 was previously shown to be circadian regulated with a shortened circadian period in a cca1lhy double mutant and a disrupted pattern in CCA1ox plants [12], [20], [50]. However, GRP7 had never been explicitly established as a target gene of CCA1 and LHY. Our northern analysis confirmed circadian expression of GRP7 and showed that such expression was slightly affected by the cca1-1 mutation and became arrhythmic in CCA1ox in LL (Figure S5). We also observed disrupted expression of GRP7 in CCA1ox plants in LD (Figure 8A). Thus, these data further confirm that GRP7 is regulated by CCA1.
GRP7 was previously demonstrated to regulate stomatal activity [51]. We found that similar to cca1-1lhy-20 and CCA1ox plants, stomatal aperture of grp7-1 was greater than that of Col-0 at ZT13 (Figure 8B). In response to PmaDG3 infection, grp7-1 displayed 14.2% suppression of stomatal aperture whereas Col-0 showed 48.1% suppression at 1 hpi (Figure 8C, S3, and Table S1), suggesting that grp7-1 has reduced responsiveness to PmaDG3 in stomatal closure. We further found that grp7-1 was significantly more susceptible to PmaDG3 than Col-0 when spray-infected at ZT13 in LD (Figure 8D). Together our bioinformatic analysis and experimental evidence indicate that GRP7 is a target of CCA1 and/or LHY that regulates stomatal activity and modulates plant defense.
SA is a key signaling molecule involved in both basal resistance and R gene-mediated defense. The accelerated cell death 6-1 (acd6-1) mutant shows constitutive defense, high levels of SA, and extremely small size that is sensitized to the change of SA defense [52], [53]. Thus, acd6-1 has been used as a convenient readout to gauge the effect of some known defense genes in regulating SA-mediated defense [54], [55], [56]. To determine whether CCA1 and LHY act through SA, we crossed cca1-1lhy-20 to acd6-1 and obtained homozygous double (acd6-1cca1-1 and acd6-1lhy-20) and triple (acd6-1cca1-1lhy-20) mutants. We found that both double and triple mutants resembled acd6-1, displaying dwarfism and accumulating similar SA levels (Figure 9A and B). However, when spray-infected with PmaDG3 at ZT13, the double mutants were slightly more susceptible while the triple mutant was much more susceptible than acd6-1 (Figure 9C). These results corroborate a synergistic interaction between CCA1 and LHY in clock and defense regulation. They also suggest that the defense role of CCA1 and LHY is largely SA-independent. Consistent with this notion, we found that in the absence of acd6-1, the SA levels are comparable among Col-0, cca1-1, lhy-20, cca1-1lhy-20, and CCA1ox in LD (Figure S6A). In addition, although more susceptible to P. syringae infection, LHYox plants were dwarf, showed spontaneous cell death, and accumulated high levels of SA (Figure 4C, 4D, and S6B). Together, these results indicate that CCA1 and LHY act independently of SA to regulate resistance to P. syringae.
Our data and those from other groups clearly indicate that plant innate immunity is an output event regulated by the circadian clock. However, it is not known whether this regulatory relationship is reciprocal with defense activation feeding back to affect clock activity. To test this, we infected Col-0 expressing the ProCCA1:LUC reporter with both virulent and avirulent P. syringae strains. Bioluminescence analysis indicated that the period of ProCCA1:LUC was significantly shortened in the presence of the virulent strain PmaDG3 or the avirulent strain PmaDG6 at a high dose (OD = 0.1) (Figure 10 and Table S3). Similarly, infection of Col-0 seedlings expressing ProGRP7:LUC also resulted in period shortening of ProGRP7-controlled luciferase activity (Figure S7 and Table S3). These results suggest that clock activity is modulated by both basal and RPS2-mediated defenses.
To further investigate which defense signaling pathway(s) are involved in the feedback-regulation of clock activity, we treated Col-0/ProCCA1:LUC seedlings with flg22 or benzo (1,2,3) thiadiazole-7-carbothioic acid (BTH). Flg22 is a 22-aa synthetic peptide from the conserved region of flagellin proteins of P. syringae and elicits plant basal defense in a wide variety of plant species [4], [57]. BTH is an agonist of SA that efficiently activates SA signaling [58]. We found that flg22 at both doses (1 µM and 10 µM) significantly shortened the period of CCA1 expression. However, BTH treatment (10 µM and 300 µM) did not change CCA1 promoter activity (Figure 11A and Table S3). To further test if SA could affect clock activity, we used a cotyledon movement assay [59] to gauge clock activity in the acd6-1 mutant, which constitutively accumulates high levels of SA [52], [53]. We found that acd6-1 showed similar period, phase, and amplitude of the rhythm for cotyledon movement to Col-0 (Figure 11B and S8). Taken together, these data indicate that activation of flg22-triggered basal defense but not SA signaling can feedback to regulate clock activity.
Increasing evidence has implicated a role of the circadian clock in regulating plant innate immunity. Of the components in the central oscillator of the circadian clock, CCA1 is the first shown to affect plant defense against P. syringae and Hpa [27], [28]. However, its close homolog, LHY, has not been shown such a role, despite the fact that loss-of-function mutants in both genes displayed similarly shortened period. Thus, it was unclear whether plant innate immunity is regulated by the circadian clock mediated by CCA1 or by non-clock related function of CCA1. Here we show that disrupting clock function by misexpression of CCA1 and/or LHY leads to compromised immunity, thus further establishing a direct role of the circadian clock in defense regulation. Our data suggest that one of the mechanisms by which CCA1 and LHY regulate plant innate immunity is through affecting stomatal defense with the downstream target gene GRP7. We further demonstrate that defense activation by P. syringae infection and flg22 treatment shortens circadian period. Thus this study reveals for the first time crosstalk between the circadian clock and plant innate immunity.
Typical studies of the circadian clock have been performed under constant light (LL) conditions to emphasize the endogenous nature of the clock. In LL, perturbations of the circadian clock typically result in altered period length; for instance, loss of CCA1 or LHY function shortens circadian period. However, plants typically grow in LD cycles in which the environmental cycle entrains even a mutant clock to a 24-hour period. Under such LD conditions, perturbations in the circadian clock can manifest as alterations in phase for reporter gene expression (Figure 1 and S1B and [20], [21]) as well as changes in a variety of other traits, including flowering time, metabolism, stomatal activity, gene expression patterns, and defense responses [12], [13], [20], [29], [50], [60], [61]. Thus, the effects of disrupted circadian clock could become apparent under LL and LD conditions.
Several studies indicate that the circadian clock mediated by CCA1 and LHY regulates plant defense in both LL and LD ([27], [28] and this study). For instance, Bhardwaj et. al. showed that CCA1ox plants were more susceptible to P. syringe infection than wt in LL [27]. Here we extend this observation by showing that CCA1ox plants had enhanced susceptibility to P. syringae in both LL and LD (Figure 2, 3, and S2). In LD, enhanced susceptibility was also observed in cca1-1lhy-20 and LHYox plants to P. syringae strains (Figure 3 and 4) and in cca1-1lhy-20 to Hpa strains (Figure 6). Consistent with our data, a single cca1 mutant showed compromised resistance to a different Hpa strain and affected expression of some defense-related genes in LD [28]. Together these studies firmly establish that plant innate immunity is an output regulated by the circadian clock under LL and LD conditions.
While we mainly focus our analyses in this report on defense phenotypes regulated by CCA1 and LHY in LD, we also agree that we should use caution when interpreting our results since the effect of the circadian clock can manifest differently under different light conditions, including both differing daylengths and light intensities. For instance, it is possible that the degree of susceptibility to pathogen infection and the severity of stomatal change in response to dark and P. syringae infection could be different in LD from those in LL in cca1-1lhy-20 (compared with wt). Alternatively, the amplitude, period, and/or phase of defense gene expression could be different in cca1-1lhy-20 (compared with wt) in LD from those in LL. Even different LD conditions could have different effects on clock activity. For example, Michael et al. [62] showed that the set of cycling transcripts increased with the number of different cycling conditions examined. We found that in 12 hr L/12 hr D, expression of GRP7 retained rhythmicity in CCA1ox, compared with that in wt, although the waveform was altered with baseline expression increased (Figure 8A). However, Green et al observed more pronounced alterations in phase of GRP7 expression in CCA1ox (compared with that in wt) in seedlings growing in long or short daylengths (16 hr L/8 hr D or 8 hr L/16 hr D), with maximal transcript accumulation in the dark [12]. Such differences in the patterns of GRP7 transcript abundance could also be due to other reasons besides light conditions. Nonetheless, these observations together with those of Michael et al. [62] emphasize that to better understand the role of the circadian clock in defense control, analyses of defense phenotypes with plants misexpressing CCA1, LHY, and/or other clock genes should be carried out in LL, DD, and different LD conditions for a comprehensive comparison.
Although encountering pathogens at different times in a day, Arabidopsis plants were suggested to be more resistant in the morning than at night. To support this conclusion, wt plants demonstrated higher resistance and/or defense responses when infiltrated during the day than at night [27], [63]. We also observed similar results in plants infiltrated with P. syringae in LL or LD (Figure 2, 4A, and S2), thus supporting this conclusion. However, with spray-infection in LD, we observed the opposite phenotype; wt plants were more resistant at night than in the morning (Figure 3, 4B and 8D). During spray-infection, P. syringae initially lands on the leaf surface. Further invasion depends on the success of the bacteria in gaining entry into the host tissue via natural openings, such as stomata [1], [2]. Consistent with enhanced disease resistance to sprayed P. syringae, plants in the evening have much smaller stomatal pore sizes than in the morning (Figure 5A, 5B, and 8B).
These two seemly contradicting results actually coalesce to suggest different mechanisms that plants use to defend against pathogens at different times of day, depending on the mode of pathogen invasion. As summarized in Figure 12A, at night plants might rely more on closed stomata to physically restrict pathogen invasion but stomata-independent defense is relatively low. If a pathogen can breach stomatal restriction (i.e. being pressured into host tissue via infiltration in the laboratory) at night, it can be more virulent to the host. However, with stomata widely open during the day, plants apparently compensate for enhanced pathogen access to the leaf interior with enhanced stomata-independent defense that is stronger during the day than at night. This cycling in host resistance means that plants can be more resistant to epiphytic pathogens at night than during the day. But in the presence of apoplastic pathogens, plants can activate stronger defense during the day than at night. Taken together, we conclude that plants rely on distinct mechanisms, involving stomata-dependent and stomata-independent defenses, to respond to pathogen attacks at different times of day.
Our data suggest that both stomata-dependent and -independent defense can be affected by CCA1, LHY, and its downstream target GRP7. Consistent with such a role of CCA1 and GRP7, these proteins are expressed in guard cells [51]. It is conceivable that CCA1 and/or LHY proteins directly affect the abundance of GRP7 via binding to its promoter at different times of day, which in turn regulates stomatal aperture and thereby stomatal defense (Figure 11B). Since both CCA1 and GRP7 proteins are also found in other cell types besides the guard cells [51], [64], it is possible that CCA1/LHY/GRP7 also contribute to stomata-independent defense.
GRP7 is unlikely to be the only target of CCA1 and LHY to regulate pathogen defense. First, our bioinformatic analysis suggests that a number of defense genes besides GRP7 might be preferentially regulated by CCA1 and LHY (Figure 7). And second, plants overexpressing GRP7 are not more susceptible to P. syringae (J. Alfano and H. Kang, personal communications) while CCA1ox and LHYox plants are more susceptible to P. syringae (this study). Thus, CCA1 and LHY presumably act through multiple downstream target genes to regulate plant defense. Identification of these additional defense genes controlled by CCA1 and LHY should advance our understanding of the mechanisms by which the circadian clock regulates plant defense.
Rhythmic variation in stomatal aperture is known to be regulated by the circadian clock [13], [37], [65]. Besides CCA1 and LHY, other genes encoding components of the central oscillator may also affect stomatal defense. For instance, a mutation in EARLY FLOWERING 3 (ELF3) was recently shown to suppress stomatal closure and disease resistance [27], [66]. ELF3 might act through the FLOWERING LOCUS T gene, which is highly expressed in stomata of the elf3 mutant and has been shown to affect stomatal activity [66]. In addition, the timing of cab expression1-1 (toc1-1) mutant also shows defects in stomatal aperture [59], [67]. It is tempting to speculate that ELF3-mediated defense is related to its role in stomatal control and TOC1 could also contribute to plant defense. However, further experiments are necessary to validate these speculations. Nevertheless, these observations suggest that the circadian clock can influence stomatal activity and possibly also stomatal defense via different pathways (Figure 12B).
Stomata have been proposed as a critical battleground during plant-bacterium interactions [1], [2]. However, it is not known whether stomatal defense can also restrict the invasion of pathogens with different life styles from those of bacteria. The oomycete pathogen Hpa does not enter host organs through stomata; rather, germinating spores produce hyphae that penetrate between host epidermal cells and extend through the intracellular space in the mesophyll layer. However, near the end of the infection cycle, hyphal tips emerge through the stomata to the exterior of the leaf and then differentiate into spore-bearing structures [68], [69]. Thus, it is possible that host control of stomatal aperture could influence this stage of the life cycle. Although the role of stomata in defense against Hpa has not been well established, the fact that cca1-1lhy-20 showed enhanced susceptibility to Hpa infection relative to the single mutants and wt suggests such a role of the circadian clock. Interestingly, while conferring enhanced disease susceptibility to P. syringae, CCA1ox heightened resistance to Hpa (Figure 6A and [28]), suggesting that CCA1ox plants employ different mechanisms to defend against these two pathogens. However, it is not clear whether the enhanced Hpa resistance conferred by CCA1ox is related to the circadian clock or to another function resulting from CCA1 overexpression.
While regulating multifaceted physiological activities of plants, the circadian clock can also be influenced by external signals, such as changes of light, temperature, hormones, and nutrients [32], [70], [71], [72], [73], [74]. Here we show that infection with both virulent and avirulent P. syringae strains shortens circadian period in Arabidopsis (Figure 10 and S7). We further found that such feedback regulation can be recapitulated with flg22 treatment (Figure 11A). Thus, defense activation can also serve as an input signal to regulate clock activity besides being an output of the circadian clock.
Since flg22-triggered callose deposition and expression of genes involved in flg22 sensing and signal transduction were previously shown to be under circadian clock control [27], we conclude that the clock-defense crosstalk involves flg22 signaling (Figure 12B). Production of SA is circadian regulated [75], however, activation of SA defense does not affect clock activity (Figure 11 and S8 and [74]). Therefore, SA is an output of the circadian clock but does not serve as an input factor. Since our data showed that CCA1 and LHY act largely independently of SA, we speculate that other circadian clock components may act through SA as an output in defense control.
What would be the advantages for plants to have clock-defense crosstalk? A properly tuned circadian clock enhances growth vigor and confers better survival rate and competitive advantage [11], [12], [13], [14], [15], [16]. Regulation of defense by the circadian clock suggests that timing of effective defense against pathogens is crucial for host fitness in the presence of pathogens. However, defense is an energy-costly process intricately connected with plant growth and development. A feedback regulation of the circadian clock by defense activation could be important for the host to balance growth, development, and defense responses, for instance, to redirect the energy from costly disease resistance to primary metabolism. Consistent with this idea, several phytohormones are potential components of the clock-defense circuitry. For instance, auxin regulates clock activity as an input [74] while auxin production and signaling are affected by the circadian clock and thus are clock output events [73], [76], [77]. Other hormones, such as abscisic acid, brassinosteroids, cytokinins, and gibberellic acid, have been shown to serve as clock inputs [74], [78]. Interestingly, cytokinin affects the phase but not the period of the clock [74], [79], [80]. However whether these hormones are also on the output pathways of the circadian clock remains to be investigated. On the other hand, ethylene and jasmonic acid production and/or signaling are on the output pathways of the circadian clock [29], [75], [81], [82], [83] although ethylene does not serve as a clock input in Arabidopsis [82]. The role of jasmonic acid as a clock input is currently unknown. All these phytohormones have been implicated in defense control besides their critical roles in regulating plant growth and development [84], [85], [86]. Therefore further investigating the roles of these phytohormones in clock-defense crosstalk should shed light on the molecular mechanisms by which plants employ to regulate growth, development, and responses to pathogen invasion. Such information could potentially lead to a better control of plant growth and resistance to devastating pathogens, ultimately enhancing productivity of plants.
Unless otherwise indicated, all plants used on this paper are in the Columbia-0 (Col-0) background and were grown in growth chambers with a 12 hr light/12 hr dark cycle, light intensity at 200 µmol m−2 s−1, 60% humidity and 22°C. Single mutants (acd6-1, lhy-20, and grp7-1) and plants overexpressing CCA1 (CCA1ox) or LHY (LHYox) were described previously [11], [35], [39], [48], [52]. cca1-1 was originally a Wassilewskija allele but was introgressed into Col-0 via five sequential backcrosses. The mutants cca1-1lhy-20, acd6-1cca1-1, acd6-1lhy-20, and acd6-1cca1-1lhy-20 were made by genetic crosses and confirmed with PCR markers corresponding to individual mutations (Table S4 and [54]). CCA1ox (line #34) and grp7-1 seeds were from Elaine Tobin and James Alfano, respectively.
P. syringae strains were grown at 28°C with King's B medium (10 g proteose peptone, 1.5 g K2HPO4, 3.2 ml 1 M MgSO4, and 5 g glycerol per liter) containing the appropriate antibiotics for selection. Freshly cultured bacteria were collected, washed once, and resuspended to desired final concentrations in 10 mM MgSO4 for infiltration and spray infections or in sterile water for stomatal aperture measurement and bioluminescence analysis. For infiltration infection, the bacterial solution was pressured into the abaxial side of the fifth to seventh leaves of a plant with a 1 ml needleless syringe. For spray infection, the bacterial solution was mixed with Silwet L-77 (Lehle Seeds) at a final concentration of 0.04% and sprayed onto plants until the leaf surface was evenly wet. Bacterial growth and disease symptoms were analyzed as described previously [53]. Log transformed bacterial growth was used in statistical analysis.
Hyaloperonospora arabidopsidis (Hpa) strains were propagated and prepared as previously described [56], [87]. Seven day-old soil-grown seedlings were sprayed with a spore suspension (5×104 spores/ml in water) containing the virulent strain Hpa Emco5 or the avirulent strain Hpa Emoy2. Seven days post inoculation, sporangiophores on both sides of cotyledons were counted to determine the level of resistance. Hpa infections were conducted as blind experiments where plant genotypes were unknown to the experimenters until the completion of the experiments. All bacterial and Hpa infection experiments were repeated at least three times unless otherwise indicated.
RNA extraction and northern blotting were performed as described [54]. Radioactive probes were made by polymerase chain reaction (PCR) with a specific antisense primer for a gene fragment in the presence of [32P] dCTP. Primers used for making probes were listed in Table S4.
Stomatal aperture was measured with 25-day-old plants as previously described [1]. Briefly, the fifth to seventh leaves were taken at the indicated times and mounted onto a glass slide at the abaxial side using Telesis 5 silicone adhesive (Premiere Products, Inc., CA). The top layer of the leaf was scratched off with a razor blade. Images of at least three random regions of the bottom layer of the leaf were taken immediately with a camera (Canon Digital Rebel xsi, Japan) connected to an inverted microscope (Olympus Model IMT-2). P. syringae treatment was performed at ZT4 when plants had been exposed to light for 4 hr to ensure that most stomata opened. The fifth to seventh leaves of plants were collected and immersed in PmaDG3 or DC3000 resuspensed in sterile water (108 cfu/ml) or in water as mock treatment. At 1 hpi or 3 hpi, treated leaves were harvested and processed for stomata imaging. At least three leaves per genotype and per time point were taken for stomatal images. The stomatal aperture was determined by the ratio between the width and the length of a stoma, which was measured with the assistance of ImageJ (version 1.45).
Seedlings expressing the reporter gene LUCIFERASE (LUC) under the control of promoters from CCA1 or GRP7 (At2g21660; also called CCR2) [45], [46] were grown on MS media with 2% sucrose in a 12 hr light/12 hr dark cycle at 22°C for 7–10 days. Seedlings were soaked in P. syringae resuspended in sterile water in the presence of 0.04% Silwet L-77, flg22 (1 µM or 10 µM), or benzo(1,2,3)thiadiazole-7-carbothioic acid (BTH; a SA agonist) (10 µM or 300 µM), and transferred to 96-well plates containing 200 µl of MS media and 30 µl of a 2.5 mM D-luciferin solution. Mock treatments were conducted with sterile water with or without 0.04% Silwet L-77. The seedlings were subsequently transferred to LL at 22°C. LUC activity was detected at 1 hr intervals for 7 days with a TopCount luminometer (Perkin Elmer Life Sciences) and analyzed with MetaMorph image software [88]. Flg22 was purchased from GenScript USA Inc. and BTH was a kind gift from Robert Dietrich (Syngenta).
For cotyledon movement, surface sterilized Arabidopsis seeds were grown on MS media with 2% sucrose for 6 days in a 12 hr light/12 hr dark cycle at 22°C and were transferred to 24-well cloning plates, one seedling per well. The seedlings were entrained for two more days in the 12 hr light/12 hr cycle at 22°C and were subsequently released into LL at 22°C. Cotyledon movement was recorded with multiple surveillance cameras every 20 min for 7 days and post-run image analysis was performed as described [88].
Up to 3000 bp promoter sequences of 571 genes [43] were downloaded from Athena (http://www.bioinformatics2.wsu.edu/cgi-bin/Athena/cgi/analysis_select.pl) [89]. These genes were grouped into three sets, selected (337 defense-related gene based on microarray experiments), empirical (127 empirical marker genes for various pathogen responses), and normalization (107 non-defense related genes whose expression levels were relatively stable among experiments with pathogen infection) [43]. Promoters of these genes were analyzed for the enrichment of CBS (AA[AC]AATCT) or EE (AAAATATCT) motifs, using the online tool POBO (http://ekhidna.biocenter.helsinki.fi/poxo/pobo/) [44]. Pseudo-clusters of 100 promoters of up to 3000 bp in length of Arabidopsis genes, which do not contain the coding sequences of the neighboring genes and were sampled randomly from the entire Arabidopsis genome with 1000 bootstrap replications, were analyzed to generate the background as a control for each motif. The number of the CBS or EE motifs in gene clusters was quantified, using a Perl program.
|
10.1371/journal.pcbi.1000370 | Representation of Time-Varying Stimuli by a Network Exhibiting Oscillations on a Faster Time Scale | Sensory processing is associated with gamma frequency oscillations (30–80 Hz) in sensory cortices. This raises the question whether gamma oscillations can be directly involved in the representation of time-varying stimuli, including stimuli whose time scale is longer than a gamma cycle. We are interested in the ability of the system to reliably distinguish different stimuli while being robust to stimulus variations such as uniform time-warp. We address this issue with a dynamical model of spiking neurons and study the response to an asymmetric sawtooth input current over a range of shape parameters. These parameters describe how fast the input current rises and falls in time. Our network consists of inhibitory and excitatory populations that are sufficient for generating oscillations in the gamma range. The oscillations period is about one-third of the stimulus duration. Embedded in this network is a subpopulation of excitatory cells that respond to the sawtooth stimulus and a subpopulation of cells that respond to an onset cue. The intrinsic gamma oscillations generate a temporally sparse code for the external stimuli. In this code, an excitatory cell may fire a single spike during a gamma cycle, depending on its tuning properties and on the temporal structure of the specific input; the identity of the stimulus is coded by the list of excitatory cells that fire during each cycle. We quantify the properties of this representation in a series of simulations and show that the sparseness of the code makes it robust to uniform warping of the time scale. We find that resetting of the oscillation phase at stimulus onset is important for a reliable representation of the stimulus and that there is a tradeoff between the resolution of the neural representation of the stimulus and robustness to time-warp.
| Sensory processing of time-varying stimuli, such as speech, is associated with high-frequency oscillatory cortical activity, the functional significance of which is still unknown. One possibility is that the oscillations are part of a stimulus-encoding mechanism. Here, we investigate a computational model of such a mechanism, a spiking neuronal network whose intrinsic oscillations interact with external input (waveforms simulating short speech segments in a single acoustic frequency band) to encode stimuli that extend over a time interval longer than the oscillation's period. The network implements a temporally sparse encoding, whose robustness to time warping and neuronal noise we quantify. To our knowledge, this study is the first to demonstrate that a biophysically plausible model of oscillations occurring in the processing of auditory input may generate a representation of signals that span multiple oscillation cycles.
| In recent years, there has been a growing interest in understanding how temporal information of sensory stimuli is encoded by sensory corticies (see, e.g., [1]–[8]). It has been shown that information about the features of the external stimulus is encoded in the fine temporal structure of the neural response (see, e.g., [8]–[15]). We are especially interested here in stimuli that have a natural hierarchy of temporal scales, such as speech and its components, including phones, diphones, words etc. Sensory processing has also been shown to be associated with the appearance of gamma oscillations in various sensory corticies (see, e.g., [16]–[20]). This raises the question whether the gamma oscillations can be directly involved in the representation of time-varying stimuli, including stimuli whose time scale is larger than that of a gamma cycle.
Such a model was suggested by Hopfield [5], and later was studied in the contex of diphone discrimination [21]. In this model subthreshold oscillatory input acts to coordinate the firing of cells so that a downstream neuron can read out a population code based on synchrony of firing. The implementation of this idea had a memory of about 200 ms, in a way that varied along a given stream of speech; the time scale of the memory depended on a dynamically changing “Lyapunov exponent”; the more negative this quantity, the shorter the memory and the more stable the representation. Thus, the longer memory was also associated with a less stable and less transparent representation. Here we build on the ideas in that paper about the synchronizing effects of gamma oscillations. However, to represent a signal having a natural time scale of more than one gamma period, we use multiple periods explicitly in the representation.
The aim of this paper is to show that this idea can be implemented robustly in the context of biophysically reasonable networks of neurons. The gamma oscillations are a product of the network, rather than an external input, and correspond to spiking events in the network, not subthreshold oscillations. We use a dynamical model of a network of spiking cells [22] that responds to a one-dimensional time-varying input in the shape of a sawtooth. Such a signal models the response of one cochlear frequency-band to a short speech stimulus, such as a diphone, that lasts several gamma cycles. We show that the oscillations produced by the network tend to discretize the neural response to the sawtooth. From this, we get a binary response of the population, based on which cells fire in which cycles. Using a simple measure of discriminability, we examine the reliability of the representation, and show that reliability requires an onset signal, something that is well known for sensory signals (see, e.g., [14],[23],[24],[25]). We also show that the representation is robust to moderate noise and time warp. In the Discussion, we compare the ideas of this paper with other work on coding (or recognition) of temporal patterns. We also discuss how hierarchies of oscillations in the nervous system may relate to the natural hierarchy of timescales in speech (phone, diphone, syllable, word, and sentence) and possible mechanisms for reading out the kind of code we suggest.
Ultimately, we would like to study the representation of a diphone. A diphone is a speech segment, roughly from the middle of a phoneme to the middle of the phoneme following it. In a single cochlear frequency-band, the temporal fluctuations of the sound energy of a diphone can be represented in caricature by a single sawtooth waveform that mimics the dynamics of energy as it enters and leaves the frequency band. In this study we focus on the representation of sawtooth-shaped signals. Different sawtooths will be represented by a single shape parameter, , that specifies the time of the energy peak in the sawtooth from the beginning of the sawtooth, in units of the sawtooth period (see Figure 1). Unless otherwise stated we use a typical duration of 50 ms for the sawtooth stimulus, although we have tested the network response for slightly shorter and longer stimulus durations 40–100 ms. The advantage of using a simplistic abstract model for the input stimulus, instead of, for example, a real intensity profile taken from speech, is that it allows for systematic investigation of the representation which, in turn, facilitates the clear understanding of the properties of the representation.
The functional architecture of the network is depicted in Figure 1. The excitatory-inhibitory interactions are sufficient to generate and sustain oscillations in the gamma frequency range. Specifically, oscillation period was about 18 ms. Hence, the duration of the external stimulus (typically 50 ms) is about three network cycles. The oscillations are generated via a mechanism known as PING (Pyramidal-Interneuronal Network Gamma). Essentially, input from the excitatory cells cause the inhibitory population to fire and generate a volley of inhibition that synchronizes the network activity (see [22] for a fuller description).
Excitatory cells are further divided into three functional subpopulations according to their different inputs. The background subpopulation receives high DC current and is responsible for generating the intrinsic gamma oscillations. The onset subpopulation receives an onset signal and is responsible for resetting the oscillation phase to synchronize it with the stimulus onset. The last subpopulation is the coding population that receives the time dependent sawtooth input current. A more detailed description of the network and its dynamics appears in the Materials and Methods section below.
Figure 2 shows three examples of the population response to the external stimuli, in the absence of internal noise. The x-axis is time and every line shows the spiking events of a different cell in the population during the same trial. The cells are ordered according to their functional subpopulation. At the bottom (cells 1–30) is the excitatory background population that, together with the inhibitory population (top - cells 71–80), generate the intrinsic gamma oscillations. The onset-response population (cells 31–45) are responsible for resetting the phase of the intrinsic oscillations, thus, synchronizing them to the onset of the external stimulus. Cells in the coding population (25 cells, no. 46–70) are plotted in an increasing order of their ‘sensitivity’ from bottom (cell 46 - least sensitive) to top (cell 70 - most sensitive).
The three Figures 2A, 2B, and 2C show the population response to stimuli with three different shape parameter values , and , respectively. For a very fast-rising stimulus (Figure 2A, ), cells in the coding population will tend to fire in the first cycle immediately after the onset. For a slower-rising stimulus (Figure 2B, ), few cells will fire in the first cycle and most cells will fire in the second cycle after the onset. For a stimulus that rises even slower (Figure 2C, ), few cells will fire in the second cycle and most cells will fire in the third cycle after the onset.
Thus, intrinsic oscillations discretize the coding population response in the following sense: the external stimulus overlaps approximately three gamma cycles. Every cell can fire at most a single spike during every cycle. The specific spike pattern of every cell depends on its identity (i.e., different cells in the coding population have different sensitivity due to different DC input levels) as well as on the stimulus shape. Hence, the list of which cell fired during what cycle contains information about the stimulus shape. Below we define a binary representation of the neural response that will be used to quantify the information content of the response.
We represent the neural response by a binary matrix of size: [number of coding cells]×[three gamma cycles] (25×3 in our model). Matrix element () indicates whether cell in the coding population fired (1) or did not fire (0) in the cycles following the stimulus onset. This choice of binary representation ignores information that may exist on a time scale finer than the gamma cycle.
Figure 3 demonstrates the binning procedure (complete description of the procedure appears in Materials and Methods section, below). The mean firing time of the onset population (plus 4.5 ms) defines the start of the first bin. The boundaries of the bins are defined by the mean spike times of the inhibitory cell population plus 4.5 ms (vertical dotted lines in Figure 3A). Figure 3B shows the binary representation of the network response in Figure 3A. The activity of every cell in the coding population during the three gamma cycles in which stimulus is presented is shown by a single row. Every row is divided into three columns that show the firing of the cell during each cycle in black (fired) and white (did not fire).
The information content can be quantified by measuring the discriminability of the binary representation of stimuli with different shapes. We chose a very simple readout mechanism, based on template matching. Every stimulus is associated with an internal binary template (see Materials and Methods). For a given response, the estimated sawtooth shape parameter is defined as the one associated with the closest template. Hamming distance was used as the distance measure between templates and input response. These choices were made due to their simplicity and the fact that they emphasize the binary nature of the neural responses. Neither the template nor the distance measure was chosen to optimize the estimation accuracy. We do not mean to suggest that the central nervous system uses this particular readout mechanism. Nevertheless, this readout is an appropriate metric for assessing the accuracy of population response in representing sawtooth-shape waveforms.
A convenient description of the readout discrimination power is the confusion matrix, (see Materials and Methods). Figure 4 shows the confusion matrix for A three alternative shape parameter values: and B nine alternative shape parameter values: . The probability of a correct classification provides a scalar summary of the of the confusion matrix. In the three alternative tasks, A, the system is always correct, the probability of correct classification is (chance level is 1/3). In the more difficult nine alternative task B performance decreases, (chance level 1/9). However, errors in estimating the shape parameter, , have a magnitude: (where is the estimated shape parameter; see Materials and Methods equation 7). As can be seen from the confusion matrix, although the error rate increases, the errors are small, typically (the first off-diagonal elements in the confusion matrix).
Figure 5A shows the the percent correct classification in an alternative () forced choice task, as a function of . For large , the percent correct decays to zero inversely with the number of alternatives, . This results from a finite resolution in the representation of the shape parameter . The confusion matrix in the case of alternatives is shown in Figure 5B. As in Figure 4B, we observe that the confusion matrix has relatively large elements mainly close to the diagonal. Hence, although there is considerable probability of error, the magnitude of the error is typically small. This finite resolution can be quantified by the root mean square (RMS) of the estimation error, , where denotes average of over different trials and phase relations. Here we obtain . In order to obtain this resolution a reliable representation is required. Below we show the necessity of the phase resetting mechanism by the onset population for obtaining a reliable representation of the shape parameter.
Since network oscillations are intrinsic and the stimulus is external, the oscillation phase at the time of stimulus onset is arbitrary. In the absence of a phase resetting (synchronizing) mechanism, the same stimulus may elicit very different responses, depending on exact phase relation. This added variability of the neural responses to the stimulus increases the dispersion of the responses to the same stimulus around the template and can be thought of as added noise. Hence, the templates become less representative and the readout performance decreases. Figure 6 shows the confusion matrix in the three alternative task, , in the absence of the onset signal (see Figure 4A for comparison). As can be seen from the figure, the probability of correct classification decreased dramatically: , relative to , in the case with the onset signal. Nevertheless, performance is still above chance (chance level is 1/3).
It is important to note that the onset signal does not need to precede the stimulus. The requirement is that the onset signal activates the onset population before the coding population responds to the stimulus. In a diphone, typically, onset is shared among all frequency bands; hence, it provides a clear and robust signal. In a recent work Chase and Young [25] have demonstrated how an onset signal can be accurately reconstructed from the response of a population of inferior colliculus cells of the cat and then used to estimate the external stimulus.
Thus the onset response assists in stabilizing a reliable representation of the stimulus shape by the neural responses. However, it does not erase all traces of the past. Even with the presence of the onset signal, the neural response to the stimulus depends on the phase relation, but to a smaller extent. This variability in the neural responses to the same stimulus is, in part, responsible for the finite resolution of the representation in the absence of intrinsic noise. Yet another factor that limits the resolution with which the network can represent the stimulus shape is our choice of binary representation. For example, one may imagine two close but different stimuli which elicit neural responses that differ by their exact spike times but fire during the same gamma cycle; these will be indistinguishable in our binary representation. Below we show that this insensitivity to exact spike timing is advantageous in representing time-warped stimuli.
Time warp is a very common perturbation in speech signal. A desired property of speech representation is robustness to such perturbations. In order to study the robustness of our representation we modified the stimulus duration and measured our readout performance, keeping the same templates. Figure 7A shows the quality of representation, in terms of percent correct classification in the three alternative task, as a function of the stimulus duration. All network parameters remained unchanged. The templates were obtained from the network response to 50 ms stimulus duration, as in previous sections. As can be seen from the figure, probability of correct discrimination is maximal when the stimulus duration is 50 ms and decreases as the stimulus duration is changed. Nevertheless, there exists a large range of durations 45–75 ms in which probability of correct discrimination is well above chance level.
The type of errors caused by time warping of the stimulus depends on the specific time stretch. To see this, it is convenient to further classify errors into three groups: immediate-up, immediate-down and other. In the alternative forced choice task, errors in which stimulus was estimated to be were classified as immediate-up (down). Figure 7B shows the error type distribution as a function of stimulus duration. As in Figure 7A, all network parameters remained unchanged and the templates were obtained from the network response to 50 ms stimulus duration. From the figure, one can see that immediate-down error rate (blue) increases when the stimulus duration is increased, whereas immediate-up error rate (red) increases when stimulus duration is decreased in the alternative forced choice task. Thus, error type follows the direction of time warping.
Figures 7C and 7D show the percent correct and error type distribution as in Figures 7A and 7B, respectively, in the alternative forced choice task. Results in the case are similar to the . Probability of correct discrimination, , peaks at the duration used to obtain the templates, 50 ms, as the stimulus duration is changed, decreases. The immediate-down error rate is increased when stimulus duration is increased and vice versa for immediate-up error rate. Similarly, there exists a range of stimulus durations (of about 45–65 ms) for which probability of correct classification is well above chance level. However, this range is smaller for the case than it is for the case. This difference is discussed below.
Robustness to time warp comes at the expense of the resolution of the representation. This can be seen by comparing Figures 7A and 7B. When a higher resolution ( alternatives) is required, the range of durations in which the readout is robust to time warp is decreased, relative to the lower resolution case ( alternatives), see above. This notion can be further quantified by studying the RMS estimation error as a function of the amount of time warp of the stimulus. Figure 8 shows the RMS estimation error, , as a function of the amount of time warp of the stimulus duration. As can be seen from the figure, for stimulus durations of 50–70 ms the resolution fluctuates around its maximum ( is minimal). The resolution decreases ( increases) as the amount of time warp increases in its magnitude, both above 70 ms and below 50 ms.
All of the above numerical simulations quantifying the network ability to represent time varying stimuli were done in a deterministic model, in the absence of intrinsic noise to the neural dynamics. For example, every inhibitory cell fired during every gamma cycle and every excitatory cell in the gamma generating population fired every other cycle. In a more realistic model [22],[26],[27] firing will be sparse and noisy, with oscillations that appear only on the network level. Thus, one should think of every cell in our deterministic model as an “effective cell”, representing the firing of a group of sparsely firing neurons. However, intrinsic noise that may cause spike time jitter, addition or deletion of spikes can have drastic detrimental effect on the quality of a temporal code [28],[29]. It is therefore important to test the sensitivity of this representation to intrinsic noise. Figure 9 shows the percent correct classification as a function of the input noise level for three, five and nine alternatives (top to bottom). As expected, the probability of correct discrimination is a monotonically decreasing function of noise level. Nevertheless, good performance levels are retained for moderate noise levels. Note, for three alternatives decreased by less than 5%, for five alternatives decreased by 23% and for nine alternatives decreased by 33%. This corresponds to a natural tradeoff of the representation resolution and robustness to intrinsic noise fluctuations.
Oscillations in the brain have been suggested to play a central role in various cognitive tasks, including attention [17],[18], navigation [30], memory [31] and motor planning [32]. In the context of speech processing, oscillations appear naturally, as almost all models of speech processing use oscillations, a clock signal or a pacemaker either explicitly or implicitly to take advantage of the natural hierarchy of timescales in the speech signal. Empirical findings suggest that oscillations in the auditory system may play an important role in spoken-language comprehension [20],[33]. The gamma frequency range (40–90 Hz), in particular, is widely found in the context of sensory processing [16]–[20].
The aim of this paper is to explore the use of oscillations in creating a representation of a time varying signal whose length is longer than the oscillation period. Using a family of signals, each in the shape of a sawtooth, but with different slopes, we have constructed a code using several gamma oscillations, with a total time interval about that of the signal. The gamma oscillations discretize the firing of a population of neurons, leading to a 3-bit binary representation. The representation of the shape parameter consists of a list of which cells in the coding population fired during what gamma cycle.
Typically, cells will fire at most once during the entire presentation of the stimulus. Hence, stimulus identity can be estimated by measuring the time interval between the firing of the onset cells and the firing of the coding cells. Every cell in the coding population is characterized by its sensitivity to the external stimulus, e.g., in Figure 2 cells 46–70 in the coding population are arranged in increasing order of sensitivity. This sensitivity dictates the firing order of cells in the coding population. Thus, the neural representation of the shape parameter is not arbitrary, but consists of natural firing order.
Our representation is sensitive to spike times with a resolution of a single gamma cycle (Figure 3). This finite temporal resolution limits the sensitivity with which temporal aspects of external stimuli can be coded (Figure 4). On the other hand, it provides robustness to fluctuations that affect the exact spike times. Those fluctuations include: stimulus variability, e.g., time warping (Figure 7), as well as intrinsic noise (Figure 9). There exists a natural tradeoff between the resolution of the representation and the robustness to fluctuations (Figure 8).
In our numerical simulations we made certain choices that are required to define the system but are not essential for our qualitative results. We chose to represent the external stimulus by neural responses that extend over internal gamma cycles. The specific choice of gamma cycles is arbitrary and our approach could be easily generalized to cycles. Larger values imply that the stimulus can be represented to a finer resolution. However, finer resolution comes at the expense of robustness to noise and time-warping perturbations. The neurons in our simulations follow Hodgkin-Huxley dynamics (see Materials and Methods below). This choice is also not essential to our main conclusions. Other choices for the neural dynamics, such as integrate and fire, may generate representations that are different in their fine details but still preserve the central qualitative features reported here. Namely: the oscillations discretize the output, forming a binary representation that is robust to moderate levels of noise and time warping perturbations of the external stimulus and is characterized by a tradeoff of sensitivity and robustness. The essential features of our network are the architecture of a PING mechanism for generating the gamma oscillations and the manner in which the external stimulus interacts with the internal oscillations.
Speech is an important example of a time-varying signal. There is a natural hierarchy of timescales in speech: phone, diphone, syllable, word, and sentence. The time duration of phones and diphones is on the order of a few gamma cycles, while the duration of a word is roughly that of a theta cycles. Oscillations on different timescales in the auditory cortex have been shown to be organized hierarchically: delta modulates theta, theta modulates gamma [34]. These data support a view of a network with nested oscillations on different timescales [35]–[39]. Though a diphone can be correlated with a beta frequency period or multiple gamma frequency periods, we chose to explore the role of gamma frequency oscillations, since gamma oscillations are known to be prominent in early sensory processing (see, e.g., [16]–[20]), and to help produce cell assemblies [40].
The nesting of oscillations has a potential relationship to robustness to time warping. Empirical studies of speech [41],[42] as well as of birdsong [43] have shown positive correlations in time warping fluctuations of short speech and birdsong segments. For example, the degree of time warping of a specific syllable in Zebra finch song can be predicted, to a large extent, by the degree of time warping of previous syllables. Similarly, in speech, time warping fluctuations of nearby short speech segments are correlated. The correlated time stretch can be predicated by estimating a ‘tempo variable’, such as the prosody, that varies on a longer timescale. Such a tempo variable can be used by an oscillatory network to modulate its oscillation frequency to compensate for the time warp of the stimulus. The mechanism that we suggest for the time encoding lends itself naturally to such a tempo variable, since the PING gamma has increasing frequency with increased drive; any mechanism that can increase drive with faster prosody will produce more robustness to time warp variability of the auditory stimulus. The frequency of a slower but correlated rhythm, such as theta [44], could act as such a tempo variable. We note that theta rhythms and gamma rhythms sometimes covary in their frequencies [45]. The beta frequency may be associated with the onset signals.
In mainstream models of spoken-word recognition the speech waveform is processed by a front-end, providing a representation from which a phonetic transcription is generated. The sequence of phones recognized is then integrated into a form that results in a ‘pointer’ to a specific item in the lexicon. Phonetic transcription is usually accomplished by a search within a vocabulary of acoustic models of the phones. These models are statistical in nature, and the probabilistic model is acquired by training [46],[47]. While such Hidden Markov Models (HMMs) have shown themselves to be highly effective, it is reasonable to question certain properties of their basic structure as a model for biological systems of speech processing. The conditional independence assumption imposed by HMMs is a poor model for the dynamics in the speech signal [48]. It is also extremely difficult to model long-range dependencies with an HMM [49]. Thus, methods which can better model temporal-spectral dynamics inherent in speech are highly desirable.
Our long-term goal is to use the physiological aspects of speech processing to improve our understanding of speech representation. In the work discussed here, a first step in this endeavor, we quantify how our model represents a cartoon signal mimicking the response of one cochlear frequency-band to speech input. Many difficult questions have to be answered before we can implement this model as a front-end to a speech recognition system. For example, what is the discrimination power of the model for more realistic signals at the input of a single cochlear channel, e.g., for a set of signals that are different in shape, in duration, in amplitude? Can our model provide a stable representation with respect to time scale variations that conform with realistic phonemic variation (usually not a uniform time warp in nature)? How to synchronize an onset signal with the signals across several cochlear channels (with relative time alignment dictated by the speech source)? How to integrate across all cochlear channels? A system based on the principles of neuronal processing that answers these questions also has the potential to create a paradigm shift in the way that speech is processed by machines.
A closely related model was suggested by Hopfield [5]. The focus of this model was on readout of the activity of multiple integrate-and-fire neurons, each of which integrates over time the time-varying signal for a single “channel”. From the perspective of representing speech, the Hopfield model is complete; it suggests an architecture, with a subthreshold gamma oscillator at the core, in which all frequency bands are integrated via a well defined readout mechanism. Although we do not have a complete system yet, a comparison can be made between our model and Hopfield's for a single frequency-band signal. Hopfield used subthreshold oscillations to synchronize the firing across channels, forcing the cells to fire in a “window of opportunity”. Though the equations that embody the model have some memory beyond one cycle, the memory corresponds to a small negative Lyapunov exponent, which is also associated with lack of robustness. Thus, it is unclear how well this performs for a longer time-varying signal. In contrast, our model is not focused on readout, but on representation. The oscillations are used to discretize the signal across several periods, rather than to synchronize many channels. Spike times are determined by both the endogenous gamma rhythm and the external input. This mechanism allows the external stimulus to modulate the frequency of the intrinsic oscillation, unlike the fixed period in the Hopfield model.
The idea that a stimulus may be coded by a sequence of firings in discrete epochs has been discussed in the context of olfaction by Bazehnov et al. [50],[51]. There are two central differences between their work and ours: First, the Bazhenov et al. papers deal with a set of signals that all have the same temporal properties: they have a rise time of 100 ms and a decay time of 200 ms, unlike the sawtooth signals of the current work. Second, in [50],[51], the different signals excite different (possibly overlapping) sets of cells in the coding population, unlike the signals in the current paper, which all excite the same set of cells, but have different effects on them. Thus, the information in the signals is different from that of the Bazhenov papers and the coding strategy is different, even though both result in discretization. The differences in strategy are appropriate for the differences in the kinds of signals to be encoded: the energy in a given auditory frequency band has a varying temporal structure across the set of signals, for which a sawtooth of different shapes provides a characterization. There is no such structure in olfactory signals.
In the current work we did not simulate a neural network implementation of our readout mechanism. How can our readout be implemented? The approach taken by Hopfield lends itself to a simple readout mechanism based on simultaneity. Since our code has more than one “bit”, a more complex readout mechanism is necessary. There are many suggestions in the literature that might be modified to work for this example [14],[52].
Stimulus identity, in our model, can be estimated by measuring the time from the firing of the onset cells to the firing of the coding cell. This could be achieved, for example, by an integrator that starts integrating time at the onset response and stops integration at the response of the coding population neurons. Thus, a class of potential readout mechanisms is that of neuronal integrators. Of particular interest is a single cell integrator model of Loewenstein et al. [53] based on slow calcium dynamics in a dendrite of a single cell. In their model [53], calcium level along the dendrite transitions from high to low and the location of the transition point along the dendrite is determined by integration over time of dendritic inputs. Thus, the firing rate of the cell corresponds to the time integral of the cells' dendritic inputs. Readout of a multiple-bit code might make use of input to multiple dendritic branches. Other ways to estimate such times use long-term potentiation and depression [54] and physiological slow conductances [55]. The above are more appropriate to the current model than the Tempotron [56], which can distinguish arbitrary time varying inputs, but is unable to discriminate well temporal features that extend beyond its integration time.
In this work we studied a very simplified stimulus model. The envelope amplitude of a diphone stimulus in a single frequency channel was approximated by a sawtooth. Incorporating a wider range of envelope repertoire as well as ranges of amplitude and several frequency bands will result in a much richer temporal code and will, most likely, require a larger neural population. However, this richness of detail may impair the clarity of our results. Moreover, meaningful theoretical investigation along these lines requires a better empirical understanding of cortical oscillations during speech perception to yield the essential constraints for theory. For example, when studying a model of several frequency channels we must choose whether or not the onset stimulus and the oscillations are shared among the different channels. Different choices may lead to different results, without reason to choose one over another. The question of whether oscillations are shared is an empirical question. To pursue in a meaningful manner the theoretical framework begun in the current work requires empirical effort to characterize the interaction of neural oscillations with time varying stimuli across several frequency channels. The current framework motivates such empirical work by suggesting ways in which an external stimulus can interact with the dynamics that encodes the signal.
|
10.1371/journal.pntd.0005587 | A nonsense mutation in TLR5 is associated with survival and reduced IL-10 and TNF-α levels in human melioidosis | Melioidosis, caused by the flagellated bacterium Burkholderia pseudomallei, is a life-threatening and increasingly recognized emerging disease. Toll-like receptor (TLR) 5 is a germline-encoded pattern recognition receptor to bacterial flagellin. We evaluated the association of a nonsense TLR5 genetic variant that truncates the receptor with clinical outcomes and with immune responses in melioidosis.
We genotyped TLR5 c.1174C>T in 194 acute melioidosis patients in Thailand. Twenty-six (13%) were genotype CT or TT. In univariable analysis, carriage of the c.1174C>T variant was associated with lower 28-day mortality (odds ratio (OR) 0.21, 95% confidence interval (CI) 0.05–0.94, P = 0.04) and with lower 90-day mortality (OR 0.25, 95% CI 0.07–086, P = 0.03). In multivariable analysis adjusting for age, sex, diabetes and renal disease, the adjusted OR for 28-day mortality in carriers of the variant was 0.24 (95% CI 0.05–1.08, P = 0.06); and the adjusted OR for 90-day mortality was 0.27 (95% CI 0.08–0.97, P = 0.04). c.1174C>T was associated with a lower rate of bacteremia (P = 0.04) and reduced plasma levels of IL-10 (P = 0.049) and TNF-α (P < 0.0001). We did not find an association between c.1174C>T and IFN-γ ELISPOT (T-cell) responses (P = 0.49), indirect haemagglutination titers or IgG antibodies to bacterial flagellin during acute melioidosis (P = 0.30 and 0.1, respectively).
This study independently confirms the association of TLR5 c.1174C>T with protection against death in melioidosis, identifies lower bacteremia, IL-10 and TNF-α production in carriers of the variant with melioidosis, but does not demonstrate an association of the variant with acute T-cell IFN-γ response, indirect haemagglutination antibody titer, or anti-flagellin IgG antibodies.
| Melioidosis is a high-mortality infectious disease in Southeast Asia and northern Australia caused by Burkholderia pseudomallei, which is a flagellated, rod-shaped Gram-negative bacterium. Understanding protective host immune responses to melioidosis is fundamental for effective vaccine development. A previous study demonstrated a strong relationship between a TLR5 stop codon polymorphism that encodes a truncated receptor for bacterial flagellin and protection against death from melioidosis. In this study, we confirmed the relationship of this genetic variant with survival from acute melioidosis in adult patients in northeast Thailand, and identified an association with a lower rate of bacteremia. We also demonstrated that this variant was associated with an increase in peripheral lymphocyte count, but we did not find an association with B. pseudomallei-specific lymphocyte responses; i.e., IFN-γ secreted T cell response, indirect haemagglutination titers or anti-flagellin IgG antibodies. In addition, patients with the TLR5 variant have significantly lower levels of IL-10 and TNF-α cytokines in plasma. Our findings further the understanding of the role of TLR5 in protective host immune responses against fatal melioidosis, and inform efforts to develop novel vaccines and therapeutics for melioidosis.
| Melioidosis is caused by the Gram-negative, flagellated bacillus and environmental saprophyte, Burkholderia pseudomallei, which the US Centers for Disease Control and Prevention (CDC) have identified as a Tier 1 bioterrorism agent. Clinical presentations of melioidosis range from acute sepsis to chronic and persistent infections, and the overall mortality rate can exceed 40% in endemic regions including northeast Thailand [1–3]. Pre-existing conditions such as diabetes, renal disease, excessive alcohol use and increasing age are known risk factors [1,2]. Further expansion of endemic boundaries of melioidosis [4–7], increasing prevalence of diabetes [8], and population ageing [9] lead to an urgent demand for a vaccine against melioidosis, especially in at-risk populations.
Understanding host defense mechanisms against B. pseudomallei infection is crucial for vaccine design and development, to allow selection of the best vaccine platform including adjuvant, and may drive development of novel therapeutics. Emerging evidence suggests the importance of membrane-bound Toll-like receptors (TLRs) in defense against B. pseudomallei infection in vitro and in vivo [10–12], and the TLR5 ligand flagellin has potential as a vaccine adjuvant [13]. Single nucleotide variants (SNV) in TLR genes may influence the innate immune response by altering the magnitude and quality of intracellular signaling cascades with implications for susceptibility to infection and disease outcomes [14]. A recent analysis demonstrated a significant association of the TLR5 SNV c.1174C>T with protection against organ failure and death in melioidosis [15]. This variant encodes a stop codon at position 392, truncating the receptor in the extracellular domain [16]. c.1174C>T is associated with lower TLR5-mediated innate immune responses in vitro and in healthy subjects whose blood was stimulated ex vivo [15]. This hypofunction in TLR5 signaling may result in lower immunopathology and in turn a reduction in sepsis-induced organ failure and death. Furthermore, reduced TLR5 signaling could result in lower levels of the regulatory cytokine interleukin-10 (IL-10), leading to less suppression of the host immune defense against the bacteria [15]. However, the relationship between c.1174C>T and innate immune responses has not been studied in patients with melioidosis.
TLRs activate signals crucial for the initiation and modulation of adaptive immune responses such as TLR-dependent dendritic cell control of T-cell activation [17]. Many individuals living in northeast Thailand become seropositive to B. pseudomallei at a young age, indicating that environmental exposure to the bacterium and the development of adaptive immune responses in the absence of clinical infection is common [18]. A previous study in this cohort reported reduced T-cell responses in patients with acute melioidosis that did not survive [19], raising the possibility that c.1174C>T may protect against death by enhancing T-cell mediated immunity against B. pseudomallei. Therefore it was important to characterize the association between c.1174C>T and adaptive immune responses in melioidosis.
The objective of this study was to confirm in an independent, prospectively designed cohort the previously reported association of c.1174C>T with survival in acute melioidosis, and to determine whether c.1174C>T is associated with innate and adaptive immune responses in patients with melioidosis.
The study was approved by the ethics committees of Faculty of Tropical Medicine, Mahidol University (Submission number TMEC 12–014); of Sappasithiprasong Hospital, Ubon Ratchathani (reference 018/2555); and the Oxford Tropical Research Ethics Committee (reference 64–11). The study was conducted according to the principles of the Declaration of Helsinki (2008), and the International Conference on Harmonization (ICH) Good Clinical Practice (GCP) guidelines. Written informed consent was obtained for all patients enrolled in the study.
The prospective recruitment of patients with melioidosis for immunological studies at Sappasithiprasong Hospital, Ubon Ratchathani, Thailand has been described previously [19]. Two hundred in-patients aged 18 years or older with melioidosis were enrolled, at a median of 5 days (IQR 3–6, range 2–13) after admission. Melioidosis was defined as isolation of B. pseudomallei from any clinical sample (blood, sputum, throat, endotracheal, bronchoalveolar lavage, pus, or urine), submitted to the laboratory. HIV status was not tested but previous work in the hospital has shown HIV rates are low and HIV is not a major risk factor for melioidosis [20]. Whole blood samples were collected at the time of enrollment (week 0), as well as again at weeks 12 and 52 after admission to hospital in surviving patients. 194 patients were successfully genotyped and analyzed in this study.
Genomic DNA was extracted from blood samples using QIAamp DNA Blood Midi kit (QIAgen, Hilden, Germany) according to the company’s instruction and stored at -20°C. The TLR5 c.1174C>T (rs5744168) SNV was genotyped using TaqMan® SNP genotyping assay (Applied Biosystems, CA, USA) on a CFX96 Touch Real-Time PCR Detection System (BioRad, Hercules, USA). The SNV context sequence was TGAATGGTTGTAAGAGCATTGTCTC[A/G]GAGATCCAAGGTCTGTAATTTTTCC.
The magnitude of cellular responses to B. pseudomallei was determined by ex vivo IFN-γ ELISPOT assay, as previously described [19]. Briefly, 96-well Multiscreen-I plates (Millipore, UK) were coated with 1D1K anti-human IFN-γ (Mabtech, AB, Sweden) and stored at 4°C overnight. Fresh peripheral blood mononuclear cells (PBMC) at 2 x 105 cells per well were added in duplicate and whole heat-inactivated B. pseudomallei (HIA-Bp) clinical isolates 199a and 207a [21] at concentration of 20 μg/ml were then added. Phytohemagglutinin (PHA) at final concentration of 5 μg/ml and RPMI-1640 were used as positive and negative controls, respectively. A T cell peptide pool (CEF, (Mabtech) at concentration of 1 μg/ml was used as control antigens. After 18 hours, secreted IFN-γ was detected following the manufacturer’s protocol (Mabtech) and read under CTL ELISPOT reader. Results are expressed as IFN-γ spot-forming cells (SFC) per million PBMC.
Titers of antibodies against B. pseudomallei were assessed by IHA following a standard protocol at the Mahidol-Oxford Tropical Medicine Research Unit, as modified from a protocol previously described [18,22]. Briefly, two-fold dilutions of patient serum were added to 96-well U bottom microplate containing 25 μl of HIA-Bp-sensitized sheep red blood cells. Plates were left at room temperature for 2 hours before incubation at 4°C overnight. The results were recorded as the highest dilution when a positive reaction was observed. The cut off was set at a dilution of 1:40.
Plasma levels of IgG antibodies specific to flagella of B. pseudomallei were determined by rapid Enzyme-Linked Immunosorbant Assay (ELISA), as described in a previous study [23] using recombinant flagellin (rFliC) as the coating antigen. The fliC gene (BPSL3319) was PCR amplified from B. pseudomallei K96243 genomic DNA, cloned into pBAD/HisA (Invitrogen, USA) and expressed in E. coli as previously described [24].
To perform ELISA, the purified rFliC antigen was added to wells of a 96-well U-bottom immunoplate (Nunc MaxiSorp U-bottom 96-Well plates; Thermo Scientific, Denmark) at a concentration of 15 μg/ml and incubated overnight at 4°C. Between each step, the ELISA plate was washed with 0.05% Tween-20 in PBS 4 times. After blocking at 37°C for 2 hours with 5% skim milk in PBS, patients’ plasma was diluted 1:300 and added to the pre-coated ELISA plate in duplicate then incubated at room temperature for 2 hours. The secondary antibody, HRP-conjugated rabbit antihuman IgG (DAKO, Copenhagen, Denmark), was diluted 1:2000 then added to the plate and incubated for 30 minutes. ELISAs were developed using TMB substrate. Results were determined as absorbance value (OD450). Pooled plasma from five melioidosis patients and five healthy controls were used as positive and negative controls, respectively.
Heparinized plasma for immunoassays was separated from blood by density centrifugation within three hours of blood draw. Cytokine levels in the plasma were quantified by using ELISA kits according to manufacturers’ instructions; The Human IL-10 and TNF-α Instant ELISA kits (eBioscience, San Diego, CA, USA), human granulocyte colony-stimulating factor (G-CSF) ELISA kit (Abcam, Cambridge, MA, USA), and human transforming growth factor beta 1 (TGF-β1) DuoSet ELISA kit (R&D systems, Minneapolis, MN, USA). Concentrations of cytokines were calculated from standard curves.
Categorical variables were displayed as counts and proportions, and were compared using Pearson’s chi squared test or Fisher’s exact test. Non-normally distributed continuous data were reported as median and interquartile range (IQR). The significance of differences between two groups was analyzed by Mann-Whitney U-test in Graphpad Prism Version 6 (San Diego, USA). In addition, immunological data was divided into tertiles and the distribution of the TLR5 genotype was compared between the highest third and lowest third of responses by Mann-Whitney U-test. To test the association of genotype with outcome, we performed univariable logistic regression and multivariable logistic regression adjusting for age, sex, diabetes and pre-existing renal disease using Stata version 14.0 for Window (StataCorp LP, TX, USA). Survival analysis was assessed with log-rank test of Kaplan-Meier curve by Stata version 11.1. A P value <0.05 was considered significant.
To confirm the previously reported association of the TLR5 variant c.1174C>T (rs5744168) with protection against death in acute melioidosis patients, we genotyped the variant in 194 Thai patients with culture-proven melioidosis admitted at Sappasithiprasong Hospital. Of these, 168 (86.6%) were genotype CC, 25 (12.8%) were CT and one (0.5%) was TT. The characteristics of the melioidosis patient cohort have been described in detail elsewhere [19], with key clinical information including demographics and risk factors shown in Table 1. Despite receiving appropriate antibiotic treatment, 25.3% (49/194) of patients died within 28 days of admission to hospital (28-day mortality). A further 12 patients died between days 29 and 90 after admission resulting in a 90-day mortality rate of 31.4% (61/194).
We confirmed Hardy–Weinberg equilibrium in survivors (P = 1) before testing the association of c.1174C>T with mortality. When 28-day mortality was selected as outcome, 16.6% of survivors were CT or TT genotypes, whereas 4.1% of non-survivors were these genotypes (P = 0.055, Table 2). We also observed the same pattern in analysis of 90-day mortality: 17.3% of survivors were heterozygotes or minor homozygotes, compared with 4.9% of non-survivors (P = 0.03). In a dominant genetic model (combining CT and TT subjects into the same group), the c.1174C>T variant was significantly associated with survival at both 28 days and 90 days [odds ratio (OR) for death 0.21, 95% Confidence Interval (CI) 0.05–0.94, P = 0.04 for 28-day mortality, and OR 0.25, 95% CI 0.07–086, P = 0.03 for 90-day mortality].
We also plotted the Kaplan Meier survival curve by c.1174C>T genotype for melioidosis subjects (Fig 1). The risk of death for subjects carrying the CC genotype was significantly higher than those carrying CT or TT genotypes by the log-rank test (P = 0.03).
We next evaluated the association between the c.1174C>T variant and bacteremia. In this cohort of melioidosis patients, 99 patients (51%) had bacteremia and bacteremia was tightly associated with mortality (P < 0.0001, Table 1) [19]. Thus, bacteremia can be considered an intermediate outcome measure for control of B. pseudomallei. Our results in the present study show that 8.1% (8/99) of patients with bacteremia were CT or TT genotypes, compared with 19.0% (18/95) of patients who had no bacteremia (P = 0.04, Table 2). We found that the c.1174C>T variant was associated with a lower rate of bacteremia in an unadjusted dominant model (OR for bacteremia 0.38, 95% CI 0.15–0.91, P = 0.03). Taken together, these results demonstrate a decrease in fatality and bacteremia in patients carrying the TLR5 c.1174C>T variant compared to those without the variant.
To take into account other drivers of immune responses to bacterial infection, we next tested the association of the c.1174C>T variant with death using an adjusted multivariable model including potential confounding variables. The odds ratio point estimates of the effect of c.1174C>T on mortality did not change appreciably from the univariable model. The c.1174C>T variant showed a borderline evidence of an association with 28-day mortality (P = 0.06) but remained significantly associated with 90-day mortality (P = 0.04) and bacteremia (P = 0.04) when sex, age and the major pre-existing conditions of diabetes and renal disease were incorporated into the model (S1 Table).
We also found a significant association between increasing age and 28-day mortality (OR 1.037 95% CI 1.009–1.066 for each year of age) and 90-day mortality (OR 1.030 95% CI = 1.012–1.067 for each year of age) in the multivariable-adjusted logistic regression model. In our cohort, the odds of bacteremia for melioidosis patients with pre-existing renal disease were 3.7 times higher than for those without renal disease (P = 0.003, S1 Table). We did not observe any significant association gender or diabetes and 28-day mortality, 90-day mortality or bacteremia.
We then examined the relationship between c.1174C>T and absolute neutrophil and lymphocyte counts during acute melioidosis. As shown in Table 3, the median peripheral blood neutrophil count of patients with the CC genotype was comparable to that of patients with CT or TT genotypes (P = 0.87). In contrast, absolute lymphocyte counts in patients with the CC genotype were significantly lower than those of patients with CT or TT genotypes (Table 3; P = 0.02).
T cell responses as quantified by IFN-γ ELISPOT and IHA titers to B. pseudomallei are lower in patients with melioidosis who do not survive [19]. In the previous study, we found undetectable or very low mean IFN-γ ELISPOT response in healthy controls (15 SFC per million PBMC) when compared with melioidosis patients (133 SFC per million PBMC) [19]. We therefore evaluated the association of c.1174C>T with these adaptive immune responses to B. pseudomallei in melioidosis patients. At the time of enrollment with acute melioidosis (median day 5 of hospitalization), the median IFN-γ ELISPOT response of subjects with the CC genotype was not different from those having CT or TT genotypes after stimulation with either the T-cell control peptide pool CEF (P = 0.32) or heat-killed B. pseudomallei (P = 0.49; Table 3, S1A and S1B Fig).
When subjects’ responses were grouped in tertiles according to IFN-γ ELISPOT results, we did not observe any difference in low (≤ 7.5 SFC/106 PBMC) and high responders (≥65 SFC/106 PBMC) by genotype. The median IFN-γ ELISPOT result of low responder with CC and CT or TT genotypes was 1 (IQR 1–2.5) and 3 (IQR 1–5), respectively (P = 0.47); and those result of high responder with CC and CT or TT genotypes was 201.25 (IQR 92.5–380) and 262.5 (IQR 105–625), respectively (P = 0.43).
We assessed the relationship of c.1174C>T with IHA titer. There was no difference based on genotype as the median IHA titer at the time of enrollment of patients carrying CC was 160 (IQR 40 to 1280) and those of individuals carrying CT or TT was 80 (IQR 18 to 640, P = 0.30; Table 3 and S1C Fig). Grouping the IHA titer by low (titer ≤ 1:160) and high responder (titer > 1: 160) status also did not demonstrate any relationship with genotype. Likewise, plasma levels of IgG antibodies to B. pseudomallei flagellin, a known ligand of TLR5, (anti-FliC), obtained at enrollment, were not statistically significantly different between melioidosis patients with CC (median 0.53, IQR 0.23–1.45) and CT or TT genotypes (median 0.92, IQR 0.42–1.45, P = 0.10, Table 3 and S1D Fig). We also found that plasma anti-FliC antibody levels were not significantly different between survivors (median 0.57, IQR 0.25–1.37) and fatal cases (median 0.48, IQR 0.26–1.59, P = 0.64).
We also evaluated the relationship between c.1174C>T and the kinetics of the T-cell IFN-γ response and IHA titer over one year (sample were collected at week 0, 12 and 52). We did not see a significant difference of kinetics in IFN-γ ELISPOT response between these two patients groups (S2A Fig). However, we found that the median IHA titer of patients with CT or TT genotypes (median 10.5, IQR 1–160) was significantly lower than those having CC genotype (median 320, IQR 80–640) at week 12 after admission (P < 0.001) but not at the other time points (S2B Fig).
Although we found a reduced IHA titer in survivors of melioidosis with CT or TT genotypes at week 12 after admission, the results in the present study do not demonstrate an effect of c.1174C>T on the measured T-cell IFN-γ response, IHA titer, or plasma levels of anti-FliC IgG antibodies during the acute phase of B. pseudomallei infection.
Stimulation of whole blood with B. pseudomallei has been previously shown to induce lower levels of monocyte-normalized IL-10 and granulocyte colony-stimulating factor (G-CSF) in healthy individuals with CT or TT genotypes [15]. To determine whether this association of c.1174C>T with experimentally induced inflammatory responses holds during acute melioidosis, we measured the levels of these two cytokines in the plasma of melioidosis patients in our cohort. IL-10 levels were significantly lower in carriers of CT or TT (Table 3 and S3A Fig; median 8.6, IQR 0.34 to 21.74) than in those with CC (median 17.0, IQR 6.0 to 35.2, P = 0.049). However, we did not observe a difference of plasma G-CSF levels by genotype (Table 3 and S3B Fig; P = 0.83).
We next assayed plasma levels of the pro-inflammatory cytokine TNF-α that has been previously associated with death in melioidosis [25]. We found that plasma TNF-α levels in patients with the CC genotype (Table 3 and S3C Fig; median 4.73, IQR 0.605 to 7.66) were significantly higher than those having CT or TT genotypes (mostly undetectable). In addition, the patients with bacteremia also showed a trend toward higher levels of TNF-α in plasma (median 3.65, IQR 0–7.28) compared with those with no bacteremia (median 0.53, IQR 0–4.51, P = 0.06). We also measured plasma levels of transforming growth factor beta 1 (TGF-β1), another immunoregulatory cytokine, but we did not observe a significant difference between patients with CC compared to CT or TT genotypes (Table 3 and S3D Fig; P = 0.21).
Data in this study confirm a previous report [15] demonstrating a significant association between the nonsense TLR5 c.1174C>T variant and survival in melioidosis patients. We also identified a relationship between c.1174C>T and lower rates of bacteremia, which represents improved control of the infection. These data underscore the importance of TLR5-dependent signaling in driving clinical outcomes in human melioidosis. In melioidosis patients, c.1174C>T was also associated with lower plasma levels of both pro-inflammatory cytokine TNF-α and anti-inflammatory cytokine IL-10, implicating differential activation of innate immunity in the mechanism of increased survival attributable to c.1174C>T. In this population with likely broad subclinical exposure to B. pseudomallei and the development of adaptive immunity, suppressed T-cell responses to B. pseudomallei are associated with death from acute melioidosis [19]. However, we did not find evidence that the TLR5 c.1174C>T variant drives the T-cell IFN-γ response, IHA titer, or anti-FliC antibody response during acute melioidosis, suggesting that the mechanism of enhanced survival in carriers of the TLR5 variant may be independent of these adaptive immunological responses.
Our finding of an inhibitory effect of c.1174C>T on IL-10 production in acute melioidosis extends findings from a previous study [15] in which blood from healthy individuals carrying the c.1174C>T variant released less IL-10 upon stimulation with B. pseudomallei. Together these data suggest a possible role for TLR5-driven IL-10 release in modulating risk of death in melioidosis [25,26]. Low concentrations of IL-10 in plasma may diminish suppressive activity of immune responses, resulting in augmentation of pro-inflammatory activity, and control of bacterial infection. However, our study does not establish causation, and more in-depth investigation is required to clarify the mechanism of TLR5-dependent IL-10 function in melioidosis.
In contrast to whole blood stimulation studies in healthy subjects [15], we found significantly lower levels of plasma TNF-α in melioidosis patients carrying c.1174C>T. This result agrees with studies in rheumatoid arthritis and Salmonella infections, in which flagellin-induced TLR5 ligation leads to upregulation of TNF-α in monocytes or macrophages [27–29]. TNF-α plays a key role in neutrophil recruitment in the inflammatory response to infections; nevertheless, it can also enhance bacterial growth [30,31]. Interestingly, we also found a strong trend towards higher plasma TNF-α levels and the presence of bacteremia. As bacteremia is tightly linked with death, this is consistent with a previous study reporting increased TNF-α levels in non-survivors of melioidosis [25]. In our study, carriers of c.1174C>T had no effect on G-CSF or TGF-β1 production. It is postulated that the release of these cytokines might pass through or compensate by other pathways.
TLR5 plays a critical role in connecting innate and adaptive immunity in other bacterial infections. Accumulating evidence demonstrates that flagellin ligation of TLR5 can simultaneously initiate MyD88-dependent and Spleen tyrosine kinase (Syk)-dependent pathways leading to pro-inflammatory cytokine secretion and antigen presentation to flagellin-specific CD4 T cells, respectively [32–34]. TLR5 activation can also lead to suppression of adaptive immune responses by pathways involving IL-10 as discussed, myeloid-derived suppressor cells (MDSC) [35] and regulatory T-cells (Treg) [36]. However, we did not observe an association between the TLR5 c.1174C>T genotype and IFN-γ secreting B. pseudomallei-specific T-cell responses nor serum anti-B. pseudomallei antibody titers during the acute stage (week 0) of bacterial infection in this study.
The stimuli used in the assays of adaptive immunity in this study may have induced too broad a response to identify the distinct downstream responses of c.1174C>T variant. The T-cell IFN-γ response and IHA titer in our study were assessed using heat-killed whole-cell B. pseudomallei, which contains a large number of immunogenic antigens. Many bacterial antigens including lipopolysaccharide (LPS) and acyl hydroperoxide reductase (AhpC) can elicit both B- and T-cell responses via other immunogenic pathways besides TLR5 [37,38].
We observed comparable plasma levels of anti-FliC IgG between patients with CC and CT or TT genotypes during acute melioidosis. Sanders et al [39] demonstrated that Salmonella flagellin elicits a strong IgG response in TLR5-/- mice, indicating that TLR5 is not required for antibody responses to flagellin. Therefore, it is postulated that B. pseudomallei flagellin may also promote humoral immunity via a TLR5-independent pathway similar to that reported during Salmonella infection.
Although we did not see a relationship between TLR5 genotype and IHA titer during acute melioidosis, we found an association between c.1174C>T variant and reduced IHA titer in survivors during convalescence from disease (12 weeks after admission). This could be due to the impact of TLR5 engagement on antibody production and secretion of terminally differentiated plasma cells compared with B-cells at an earlier maturation stage [40]. However, further study on the detailed mechanism of TLR5 triggering on memory B cells is required.
The TLR5 c.1174C>T variant was associated with a higher absolute lymphocyte count, but not with T-cell responses. The increased number of lymphocytes in patients carrying the variant may result from the reduced suppressive effect of IL-10 induced during acute melioidosis. Further studies should aim to characterize this increased lymphocyte population with a particular focus on B-cells, NK cells or subsets of T cells that do not produce IFN-γ. In our study, the TLR5 c.1174C>T variant did not influence the quantity of neutrophils in melioidosis patients.
TLR5 c.1174C>T might play a critical role only in inflammatory cytokine responses against melioidosis, contributing to control of bacterial infection before adaptive immunity takes place. Otherwise, the relationship between TLR5 c.1174C>T and adaptive immunity may be present but our study had insufficient power or did not measure the relevant T-cell or antibody response. Additional studies focusing on the relationship between the TLR5 c.1174C>T and adaptive immune responses against B. pseudomallei flagellin may uncover an association. Further studies will address the direct and crucial link between innate and adaptive immunity of TLR5 in B. pseudomallei.
In summary, the results of our study provide critical confirmation of the association of TLR5 c.1174C>T genotype with protection against death in acute melioidosis patients. Our results also suggest that the genotype c.1174C>T in melioidosis patients is associated with reduced production of both pro-inflammatory cytokine TNF-α and anti-inflammatory cytokine IL-10 at the early stage of infection. Although TLR5 genotype is associated with protection against melioidosis, other factors underlying host defense mechanisms merit exploration in further studies.
|
10.1371/journal.ppat.1002496 | Viral Cyclins Mediate Separate Phases of Infection by Integrating Functions of Distinct Mammalian Cyclins | Gammaherpesvirus cyclins have expanded biochemical features relative to mammalian cyclins, and promote infection and pathogenesis including acute lung infection, viral persistence, and reactivation from latency. To define the essential features of the viral cyclin, we generated a panel of knock-in viruses expressing various viral or mammalian cyclins from the murine gammaherpesvirus 68 cyclin locus. Viral cyclins of both gammaherpesvirus 68 and Kaposi's sarcoma-associated herpesvirus supported all cyclin-dependent stages of infection, indicating functional conservation. Although mammalian cyclins could not restore lung replication, they did promote viral persistence and reactivation. Strikingly, distinct and non-overlapping mammalian cyclins complemented persistence (cyclin A, E) or reactivation from latency (cyclin D3). Based on these data, unique biochemical features of viral cyclins (e.g. enhanced kinase activation) are not essential to mediate specific processes during infection. What is essential for, and unique to, the viral cyclins is the integration of the activities of several different mammalian cyclins, which allows viral cyclins to mediate multiple, discrete stages of infection. These studies also demonstrated that closely related stages of infection, that are cyclin-dependent, are in fact genetically distinct, and thus predict that cyclin requirements may be used to tailor potential therapies for virus-associated diseases.
| Many viruses encode homologs of human oncogenes, including the gammaherpesvirus viral cyclin genes. These viruses cause lifelong infection associated with chronic diseases, including malignancies, which are exacerbated in immune deficiency. The conserved viral cyclins were first recognized nearly two decades ago, and despite extensive interest and study, their essential features for virus infection and disease have been elusive. We used a mouse model of these viruses to make recombinant viruses with viral or human cyclins knocked into the endogenous locus. We then determined the requirements for cyclins by genetic complementation in three distinct viral cyclin dependent aspects of infection. We report that the viral cyclins of different gammaherpesviruses are able to support all three stages of infection. However, none of the human cyclins can, and instead comprise distinct complementation groups that are functional in non-overlapping aspects of infection. We showed that gammaherpesvirus encoded cyclins are functionally conserved, and that their essential unique property is the assimilation of the functions of distinct mammalian cyclins within a single multifunctional gene. Finally, in dissecting the requirements for viral cyclins during gammaherpesvirus infection, we demonstrated that related stages of infection are genetically separable and therefore may be susceptible to specific therapeutic manipulation.
| Gammaherpesviruses are oncogenic viruses that establish lifelong infection of the host. Primary gammaherpesvirus infection of healthy adult hosts results in an acute stage of lytic virus replication which is then cleared, with lifelong latent infection established primarily in B lymphocytes. A transient mononucleosis-like stage is associated with establishment of latent infection with Epstein Barr virus (EBV) and the murine gammaherpesvirus 68 (gHV68). The latent stage of infection is controlled by an active immune response, and immune deficient hosts suffer increased virus reactivation from latency and persistent infection (evidenced by ongoing production of infectious virus), both of which are associated with disease. Viral cyclin genes are conserved among gamma-2-herpesviruses, including the human Kaposi's sarcoma-associated herpesvirus (KSHV), and Epstein Barr virus (EBV), a closely related human gammaherpesvirus, uses positional homologs to up regulate expression of host D-type cyclins. Cyclins are the regulatory partners of the catalytic cyclin dependent kinases (cdks), which together regulate cellular DNA replication and cell division. Viral cyclins share the greatest sequence similarity to one another and to mammalian D-type cyclins, yet are functionally most similar to mammalian cyclins A and E [1]–[4]. Relative to mammalian cyclins, the viral cyclins confer increased kinase activity and demonstrate broader cdk binding and substrate specificity, as well as increased resistance to cyclin-dependent kinase inhibitors [5]–[10]. The viral cyclin (v-cyclin) protein of the mouse model gHV68 is abundantly expressed in lytic virus replication and in reactivation from latency [11], and v-cyclin transcript is also detected in latently infected B cells [12]. The first gammaherpesvirus viral cyclin gene was described in 1992 [13], since which time numerous activities of the viral cyclins have been discovered and proposed as important in gammaherpesvirus pathogenesis. However, to date, no study has addressed whether the unique biochemical features of the v-cyclin are essential to promote infection or if mammalian cyclins, with more restricted activities, are capable of promoting infection. This issue is particularly important given the increasing evidence that mammalian cyclins have an unexpected degree of plasticity and redundancy in promoting cell cycle progression [14] [15], yet specific cyclins are required for cell- or tissue-specific functions [16], [17]. The emerging picture of the mammalian cyclins in cell cycle, development and cancer present a compelling case for understanding the specific activities of the unique viral cyclins.
While extensive biochemical characterization of viral cyclins revealed multiple unique characteristics of viral cyclins relative to mammalian cyclins when expressed in isolation, the precise function of the viral cyclins in the context of virus infection has only more recently been elucidated. In transgenic studies in which the viral cyclins are constitutively expressed in mice, both the gHV68 and KSHV viral cyclins are tumorigenic [11], [18]. This observation, coupled with the known cell cycle promoting effects of the viral cyclins and viral cyclin expression in some gammaherpesvirus associated tumors, initially lead to a focus on the oncogenic effects of the viral cyclin during infection. However, given that gammaherpesvirus infection in healthy individuals rarely induces malignancy, the viral cyclin is very likely to have roles in promoting viral infection and pathogenesis. To rigorously assess the genetic contribution of the viral cyclin in the context of virus infection and gammaherpesvirus pathogenesis, we have made extensive use of the gHV68 mouse model and have now shown that the v-cyclin of gHV68 plays a critical role in several distinct aspects of virus infection. We demonstrated that virus production in acute pulmonary infection is dependent on the v-cyclin [19]. Additionally, we noted a dramatic decrease in the survival of persistently infected endothelial cells upon infection with v-cyclin-deficient virus [20]. Finally, we and others observed a profound defect in viral reactivation from latency in the absence of the v-cyclin [21], [22]. The requirement for the v-cyclin is manifested in many disease states, that is, the v-cyclin-deficient virus is attenuated in lethal pneumonia [19], arteritis [23] and chronic pulmonary disease [24], chronic mortality in immune deficient mice [25], and in atypical lymphoid hyperplasia [26] found in immune deficient mice and pathologically similar to EBV-induced post-transplant lymphoproliferative disease. In contrast, the v-cyclin is dispensable for viral replication, the establishment of latency [22] and the development of pulmonary lymphoma in immunodeficient mice [27].
To rigorously dissect the essential cyclin feature(s) required for the v-cyclin during virus infection, we generated a panel of recombinant viruses in which the v-cyclin of gHV68 was precisely replaced with the viral cyclin of KSHV (k-cyclin) or with multiple different mammalian cyclins. By testing the capacity of different viral and mammalian cyclins to substitute for the function(s) of the endogenous v-cyclin of gHV68 in known v-cyclin dependent parameters, we determined that the viral cyclins of gHV68 and KSHV were able to interchangeably fulfill all v-cyclin dependent parameters of infection. On the other hand, analysis of viral recombinants expressing mammalian cyclins revealed varying capacity to support v-cyclin dependent stages of infection. Unexpectedly, distinct and non-overlapping cyclins were capable of functioning in different stages of infection, an observation which allowed us to genetically separate reactivation from latency and viral persistence. In total, these studies demonstrate that the viral cyclins are uniquely multifunctional and mediate their complete function by possessing properties of multiple mammalian cyclins.
We generated a complete panel of recombinant viruses to genetically test cyclin requirements in promoting gammaherpesvirus infection (Figure 1A and S1, Table S1). Using bacterial artificial chromosome mediate mutagenesis, we generated six viral recombinants in which different viral or mammalian cyclins precisely replaced the endogenous cyclin gene. This method placed different cyclins under control of the endogenous v-cyclin promoter and viral polyA signal to faithfully recapitulate the transcriptional regulation of this gene. To facilitate uniform and sensitive detection of cyclin expression among these recombinant viruses, a 3x-FLAG epitope tag was fused to the amino terminus of each cyclin [28]–[30]. The cyclins included in this recombinant panel were based on similarity in either sequence or function to the v-cyclin: 1) the gHV68 v-cyclin, 2) the viral cyclin of KSHV (k-cyclin), 3) the mammalian cyclins D2 and D3, based on sequence similarity [6], [10], [31] and their predominant expression in lymphocytes [32], the major reservoir for gHV68 latency, and 4) the mammalian E and A cyclins based on structural and functional similarity [33]. Quantitative analysis of virally expressed cyclin mRNAs, via the shared 3x-FLAG sequence, demonstrated similar RNA expression of all 3x-FLAG tagged cyclins during virus infection at 12 and 48 hours post-infection. Further, all 3x-FLAG-cyclin RNAs were expressed at low levels 12 hours post-infection, and were abundant by 48 hours post-infection (Figure 1B), consistent with the early-late gene kinetics previously established for the gHV68 v-cyclin [11]. These data demonstrated that the FLAG-tagged cyclins are equivalently transcribed during virus infection. Viral cyclin protein expression during infection was detectable by immunoflourescence at 12 and 24 hours post-infection (Figure S2C), and demonstrated a similar and primarily nuclear/perinuclear pattern. Specificity of cyclin protein expression from each recombinant virus was verified by western analysis of independent duplicate infections using both FLAG- and cyclin-specific antibodies (Figure 1C), and abundant expression of each cyclin protein was demonstrated during infection of fibroblasts (Figure 1D) and endothelial cells (Figure S2A). Because viral and mammalian cyclins are presumed to function via binding of cellular cdks, we performed kinase interaction analyses for each of these viruses at 24 hours post-infection of two different cell types. Infection of both fibroblasts and endothelial cells demonstrated the expected interaction partners, with the viral cyclins binding cdks more efficiently than their cellular counterparts (Figure S2B). Notably, the cdks associated with the viral cyclins were distinct from each other, and neither of the viral cyclins shared a common interaction profile with any of the mammalian cyclins during virus infection. Variation in relative protein abundance and in kinase binding of the 3x-FLAG cyclins is consistent with known differences in protein stability and partner preferences, and notably did not correspond to function during virus infection in subsequent studies. Finally, we previously showed that the v-cyclin is not required for virus replication in vitro [22]. To ensure that insertion of other cyclin genes did not alter viral replication, we compared virus replication of WT virus with the panel of cyclin recombinants in multiple cycles of replication and found indistinguishable replication among these viruses (Figure 1E). Thus, replacement of the v-cyclin of γHV68 with other cyclin genes does not alter virus replication in vitro.
As we recently reported, infection of immunodeficient mice with gHV-cycKO virus resulted in a significant defect in acute virus production in the lung and failed to cause the lethal pneumonia that results from WT infection and WT levels of acute virus production [19]. Therefore, to investigate the cyclin requirements for acute virus production and lung pathology, we infected IFN-g-/- mice with the panel of recombinant cyclin viruses. We tested gHV-cycK, the cyclin with the greatest overall similarity to the v-cyclin, gHV-cycD3 and D2 for sequence similarity and cell type relevance, and gHV-cycA and E for functional similarity. We infected IFN-g-/- mice with the recombinant cyclin viruses for 8 days, previously identified as the time at which virus titer and lung pathology differed most between WT and gHV-cycKO infection (Figure 2A) [19]. The severity of pathology, or acute pneumonia, was most profound following infection with the gHV-cycV, with similar morphology in the gHV-cycK-infected lungs (interstitial and airway edema, hypercellularity, tissue condensation and severe inflammatory infiltrates marked by neutrophils; Figure 2B), and mice in these groups demonstrated hunched posture and ruffled fur at time of sacrifice. Less severe pathology was observed in lungs infected with viruses expressing mammalian cyclins (inflammatory cell infiltrates and edema primarily surrounding vessels; Figure 2B) and these mice did not show physical symptoms. Similarly, virus production in acute lung infection was fully restored by the viral cyclins, whereas all viruses expressing mammalian cyclins were impaired relative to those expressing the viral cyclins (all statistically significant relative to gHV-cycV, p≤0.05), but did partially restore acute lung titers relative to gHV-cycKO (Figure 2C). These data suggest that mammalian cyclins have only a modest ability to function in viral infection and that the viral cyclins are unique in their ability to facilitate viral pathogenesis in lungs at early times post-infection.
We previously reported that endothelial cells are able to support persistent gHV68 infection, a process that is dependent on the v-cyclin [20]. gHV68 infection results in a characteristic alteration in endothelial cell morphology, marked by modified gene expression and adherence-independent growth. These persistently infected endothelial cells remain viable for an extended time and are not lysed by virus infection, yet are productively infected and release abundant infectious virus. We next sought to determine the capacity of the recombinant cyclin viruses to promote survival and persistent infection in endothelial cells. Growth and survival of non-adherent surviving endothelial cells was measured at nine days post-infection (Figure 3A). Cell survival (percent annexin V- and PI-negative cells) following infection with the recombinant cyclin viruses is shown in Figure 3B. As in acute pulmonary infection, the KSHV k-cyclin and the gHV68 v-cyclin were both fully functional in endothelial cell persistent infection; however, no mammalian cyclins showed modest or intermediate capacities. Instead, mammalian cyclins A and E, which bear functional similarity to the viral cyclins, were fully functional (≥50% viability) in promoting persistent endothelial cell infection. In contrast, cyclins D3 and D2, which share the most sequence similarity to the viral cyclins, conferred no advantage over a cyclin deficient virus (Figure 3C). These data demonstrated that the viruses expressing D-type cyclins were completely defective in promoting endothelial cell survival (Figure 3C), equivalent to the defect observed with a virus completely deficient for the v-cyclin. In contrast, not only both viral cyclins, but also mammalian cyclins E and A, were able to fully restore endothelial cell persistence (Table 1) to the level conferred by the gHV68 v-cyclin.
Latent infection with gammaherpesviruses is a complex process that is normally established in vivo, and is best measured after primary lytic infection has resolved. We previously used ex vivo analysis of cells infected with wild-type or cyclin deficient virus to show that the v-cyclin is critical to reactivation from latency [22] in both healthy and immune deficient mice [22], [25]. This requirement for the v-cyclin is surprising, given the presumed expression of the homologous host cyclins during infection. And while other viral genes are also required for reactivation from latent infection, to date, no other single gene has been found to play an equivalent role. To determine the required cyclin function in reactivation during infection of mice, we first established that the recombinant cyclin viruses behaved as expected in vivo; that is, no mortality was observed during six weeks of infection, the relative cellularity of splenic and peritoneal cells was consistent with WT infection at both 16 and 42 days post-infection (data not shown), and while infected cells are scarce, the FLAG-tagged v-cyclin can be detected in vivo (Figure S3A) during the peak of infection. Furthermore, we verified that reactivation of the FLAG-tagged recombinant virus was equivalent to that of the original WT virus (Figure S3B), and based on our previous demonstrations that the viral cyclin is not required for the establishment of latency [22], [25], we used a subset of these viruses (representing both those that do and do not complement reactivation from latency) to show that, as expected, latency was established normally (Figure S3C; no significant differences found in the frequency of latently infected cells). We previously demonstrated that the v-cyclin is required for both reactivation from latency and for persistent infection in immune deficient mice. Therefore we hypothesized that cyclin requirements for reactivation might be synonymous with those for persistence. This would predict complementation in reactivation by mammalian cyclins E and A, that is, that gHV-cycE and gHV-cycA would be significantly increased over gHV-cycKO. Infected peritoneal exudates cells (PECs) were plated on highly permissive MEF indicator cells for measurement of viral cytopathic effect (CPE) (Figure 4A). Reactivation analyses of the full panel of recombinant cyclin viruses are shown in Figure 4B, with cyclin recombinant viruses that restored v-cyclin function in reactivation shown in Figure 4C and those that fail to complement shown in Figure 4D. To our surprise, gHV-cycE and gHV-cycA failed to support reactivation, as did gHV-cycD2, with reactivation less than or equal to the cyclin deficient virus for each of these viruses (Figure 4D). Reactivation frequencies of the gHV-cycKO and the non-complementing viruses were from extrapolated values, as the cells reactivating fell short of 63% even at the highest concentration. We found that the viral cyclins of both gHV68 and of KSHV complemented v-cyclin function in reactivation, and that gHV-cycV and gHV-cycK infections resulted in reactivation frequencies that did not statistically differ from each other. In addition, gHV-cycD3 was the only mammalian cyclin virus that differed significantly from the gHV-cycKO virus in supporting reactivation from latency (Figure 4C). As expected from previous studies, the frequency of latently infected cells, or latency establishment, was similar between PECs infected with complementing versus non-complementing recombinant viruses (Figure S2D). These data demonstrate that reactivation from latency is supported by a cyclin activity common to the viral cyclins and mammalian cyclin D3 (Table 1), but that is not shared with mammalian cyclins E, A or D2.
Gammaherpesviruses establish lifelong infections in their host and are considered to be etiological agents for a variety of disease states, ranging from inflammatory conditions to malignancies, particularly in immunosuppressed individuals [34], [35]. While the precise mechanisms by which these viruses establish a chronic infection remains an ongoing area of investigation, one gene that clearly influences chronic infection is the viral cyclin, encoded by the human virus Kaposi's sarcoma associated herpesvirus and murine gammaherpesvirus 68. In KSHV, the k-cyclin is expressed during latency and reactivation from latency, and the k-cyclin regulates latency in KS cell lines [36], [37]. By using murine gHV68 infection of mice to assess the role of the v-cyclin in multiple stages of infection, we and others have found that the v-cyclin is necessary for multiple facets of chronic infection and pathogenesis, including acute virus production in the lung during immune deficiency [19], endothelial cell survival and viral persistence [20], and reactivation from latency [22]. Notably, the gHV68 v-cyclin is also required for chronic pathogenesis in immunosuppressed individuals, including the induction of chronic inflammatory conditions (e.g. in IFNgRKO; [25], [38]) and tumorigenesis (e.g. in BALB/b2M KO mice; [26]).
Given the varied roles that this gene has in promoting optimal gammaherpesvirus chronic infection and pathogenesis, there is a pressing need to understand the molecular mechanisms by which the v-cyclin mediates these diverse outcomes. While numerous reports have identified biochemical differences in the viral cyclins relative to host cyclins, to date there have been no studies to define which of these enhanced features of the viral cyclin are critical for gammaherpesvirus infection and pathogenesis. In fact, as reported here, analyses of kinase binding by viral and mammalian cyclins expressed under identical conditions during virus infection indicated multiple distinct patterns that do not correspond to subsequent functional studies. Additionally, kinase binding and activation may well differ in particular cell types and infection states in vivo, many of which are not readily amenable to biochemical analysis. Recently, mouse knock-outs and knock-ins have led to major advances in our understanding of cyclins and cdks, such that cyclins are now implicated not only in cell cycle progression, but in development, tissue specificity, tumorigenesis, and DNA damage and transcription, in the presence or absence of cdk partners. Using this successful approach, in this report, we tested the capacity of both viral and mammalian cyclins to function in multiple v-cyclin dependent stages of infection. Based on the enhanced biochemical features of the viral cyclins relative to mammalian cyclins, and the fact that host cyclins are present within the virus infected cell, we hypothesized that the viral cyclins of gHV68 and KSHV might be uniquely capable of functioning during virus infection. Indeed, when we first analyzed the ability of the various recombinant viruses to undergo replication in the lungs of immunosuppressed mice, we found that only the viral cyclins of either gHV68 or KSHV, and not mammalian cyclins, were capable of conferring wild-type levels of virus production and consequent increased pneumonia in infected lungs. These observations are consistent with the idea that the viral cyclins mediate their functions during infection through unique biochemical features, such as kinase binding, not present in mammalian cyclins. Notably, the viral cyclins of both gHV68 and KSHV were interchangeable in these tests of genetic complementation, despite this and previous reports identifying potential differences in their cdk binding partners and substrates [37], [39]–[41]. These data identify a genetically conserved mechanism of the gammaherpesvirus cyclins for in vivo infection and pathogenesis.
While only the viral cyclins provided optimal virus production in the immunosuppressed lung, further investigation of v-cyclin dependent stages of infection revealed a surprising ability of mammalian cyclins to mediate different stages of infection (model represented in Figure 5). On the one hand, host cyclins E and A was capable of fully promoting persistent infection of endothelial cells, while host cyclin D3 was capable of promoting reactivation from latency. Strikingly, the ability of host cyclins to mediate these distinct processes was non-overlapping, such that host cyclins capable of functioning in viral persistence were not capable of functioning in reactivation and vice versa. These data clearly demonstrate that, when expressed in the correct spatiotemporal manner (by insertion in the endogenous v-cyclin locus), host cyclins are able to mediate distinct subsets of v-cyclin dependent functions in vivo.
The existence of distinct genetic complementation groups of mammalian cyclins for optimal infection strongly suggests that these processes are mediated by distinct molecular mechanisms. Based on the ability of host cyclins E and A, but not D-type cyclins, to promote viral persistence it is worth asking what this complementation pattern might tell us about how the v-cyclin promotes persistence. What unique features do cyclins E and A have that differ from the D-type cyclins? First, cyclins E and A (and the v-cyclins) differ in kinase partners, but all confer stronger kinase activation and longer half-lives than the D-type cyclins. While cyclins E and A may function by promoting cell cycle progression, herpesvirus infection is also associated with cell cycle arrest [42]. A second possible explanation for cyclins E and A in promoting endothelial cell persistence might be the fact that a cellular DNA damage response is important in promoting early herpesvirus DNA replication [43], and these cyclins are important in the DNA damage response [44], induction of which correlates to strength of kinase activation [45]. It is also worth noting that viral persistence is dependent on host autophagy machinery and an ability to survive substrate detachment as well [46]. These studies clearly demonstrate that the requirement for cyclin function in endothelial cell persistent infection corresponds to capacity for kinase activation. Based on this, we hypothesize that viral persistence may be particularly sensitive to therapeutic kinase inhibitors.
Reactivation of virus replication from latently infected cells is a critical function of the v-cyclin in vivo, and correlates well with pathologies of chronic infection [25], [26], [38]. Because latent infection over time is concentrated in quiescent memory B cells, one straightforward possibility is that the v-cyclin is required in reactivation simply to stimulate quiescent cells into cycle. However, past reports have indicated that induction of the cell cycle by immunoglobulin cross linking or Toll-like receptor stimulation is insufficient to overcome the defect in reactivation that is observed with the v-cyclin-deficient virus [21], [47]. If cell cycle progression via cdk activation is the sole requirement for reactivation, then redundancy in cell cycle function [14] predicts that proper expression of any cyclin should substitute for v-cyclin in reactivation. Instead we found that only the viral k-cyclin and mammalian cyclin D3 were able to genetically function in promoting reactivation from latency. These data compellingly indicate that reactivation from latency is dependent on a highly restricted cyclin activity possessed by cyclin D3 that does not correspond to kinase requirement in cell cycle progression, in which cyclins E and A can substitute for D-type cyclins and cyclin D/cdk4 or cdk6 complexes are not required for cell cycle [16]. This observation is consistent with the demonstration of wild-type reactivation following infection with mutant v-cyclin viruses that are impaired in cdk binding in vitro [48], and with reports of cdk-independent functions of D-type cyclins [49]–[51]. The inability of cyclin D2 to compensate in reactivation is not likely a feature of poor protein stability [52], because this is a shared feature of the D-type cyclins. Instead, our data suggest that the ability of mammalian cyclin D3 to function in reactivation is based on a unique role for cyclin D3, such as transcription regulation [53]–[55], activation of unconventional kinase partners [56], [57], or cell type-specific function. It is worth noting that a unique role for cyclin D3 in promoting virus infection has also been observed in promoting herpes simplex virus reactivation [58]. Given that B lymphocytes are the major latency reservoir of the gammaherpesviruses, it is notable that cyclin D3 is specifically required in lymphocyte development [59], and particularly in germinal center B cells, a prominent early reservoir for viral latency [60], [61].
Beyond specific insights into the mechanisms by which the v-cyclin promotes chronic infection, this study also revealed a fundamental new insight in gammaherpesvirus infection, by demonstrating that viral reactivation from latency and viral persistence are genetically separate processes. To date, these processes are frequently intertwined spatially and temporally, making it difficult to discern their interrelationship. These distinct cyclin functions suggest a new explanation for the partial complementation of the mammalian cyclins during the acute phase of replication in the lung. While the general assumption that primary lytic virus replication is cleared and then followed by latent infection, Flano, et al. provided evidence that lytic and latent infection occur simultaneously and early in the lungs, and that latently infected cells are apparent as early as three days post-infection [62]. Our study further supports this finding, and provides genetic evidence that reactivation from latency, generally considered only in later stages of infection, contributes to virus production during the early stages of primary infection. Previously, persistent infection (as defined by detection of infectious virus late in infection) and reactivation were both increased in immune-deficient mice, consistent with increased reactivation from latency resulting in increased persistent infection, or vice versa [38]. Two viral homologs of host genes, the viral cyclin and the viral bcl-2, are both required in persistence and in reactivation. The first indication that these are separable processes was identified by recent analysis of the viral bcl-2 homolog, in which different v-bcl-2 mutants were capable of supporting either persistence or reactivation [63]. Further, we demonstrated that both the v-cyclin and the v-bcl-2 are required for optimal virus production and lethal pneumonia in immune-deficient hosts [19]. First, it is remarkable that the genetic requirements for both the v-cyclin and the v-bcl-2 are separable in these distinct aspects of virus infection. Second, these data illustrate that in immune-deficient mice, acute virus production cannot be solely attributed to primary virus replication, but may be the sum of replication, persistence, and reactivation. Genetic separation of these functions raises the potential that chronic disease previously associated with both persistence and reactivation may be dependent on one or the other. The cyclin recombinant viruses now provide a mechanism to determine the relative contribution of reactivation and persistence in various disease processes, and may provide insight for therapeutic interventions specifically tailored to the cyclin susceptibility of each.
In total, our findings demonstrate that the multifunctional nature of the viral cyclins described in in vitro biochemical studies corresponds to genetically distinct and required functions during virus infection, and that both the gHV68 and KSHV viral cyclins share this multifunctional capacity in infection. Additionally, this study revealed distinct genetic complementation groups of the mammalian cyclins, demonstrating that mammalian cyclins can fulfill the biochemical features of the v-cyclin in infection. These studies reveal that the unusual biochemical features of viral cyclins, such as broad substrate specificity and increased kinase activity, are not absolutely required to mediate specific processes within viral infection. And yet, cyclin D3 restored v-cyclin dependent reactivation less effectively than did the viral cyclins, suggesting that unique biochemical feature(s) of viral cyclins may be required to facilitate robust activity in reactivation. These data also indicate that v-cyclin features, such as resistance to cell cycle inhibitors or enhanced kinase activity, are necessary for optimal gammaherpesvirus pathogenesis.
The unique capacity of the viral cyclins to encompass functions of multiple mammalian cyclins probably explains the evolutionary advantage of encoding viral cyclins within the viral genome. Whereas only the viral cyclins can perform all v-cyclin dependent parameters of infection, our data also suggest that expression of endogenous host cyclins could complement v-cyclin-dependent functions in vivo. This idea is consistent with our observations that neither persistent infection nor reactivation from latency is completely abrogated in absence of the v-cyclin. Since mammalian cyclins can genetically replace the v-cyclin in distinct stages of infection, we hypothesize that methods of interfering with mammalian cyclin-mediated processes may also be effective at inhibiting specific functions of the v-cyclin. The ultimate test of this idea will be specific chemical inhibition of specific v-cyclin functions, and whether such inhibition indeed decreases persistent infection and reactivation levels below that of v-cyclin deficient viruses. Finally, the distinct cyclin requirements in different v-cyclin stages of infection provide potential for specific treatment of different gammaherpesvirus pathologies using existing therapeutic inhibitors specific to certain host cyclins and cdks [64]. Because cyclins and cdks are well-conserved and are host proteins, this strategy circumvents potential virus escape and may also prove useful for treatment of herpesviruses that do not encode cyclins within their genomes.
This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All animal studies were conducted in accordance with the University of Colorado Denver Institutional Animal Use and Care Committee under the Animal Welfare Assurance of Compliance # A3269-01. All surgery was performed under isoflurane anesthesia, and all efforts were made to minimize suffering.
gHV68 clone WUMS (WT; ATCC VR1465), gHV68 containing a stop codon within ORF 72 (cycKO), and parental and epitope-tagged recombinant viruses (Figure S1) were passaged and grown, and the titer was determined as previously described [22], [31]. NIH 3T12 (ATCC CCL-164), Vero-Cre cells (Dr. David Leib, Dartmouth Medical School of Medicine) and mouse endothelial cell lines MB114 [65] were grown in Dulbecco's modified Eagle medium (DMEM) supplemented with 5% fetal bovine serum (FBS), 100 U/ml penicillin, 10 ug/ml streptomycin sulfate, and 2 mM L-glutamine. Mouse embryonic fibroblasts (MEFs) were isolated from C57/BL6 mice as previously described [66] and cultured in DMEM supplemented with 10% FBS, 2 mM L-glutamine, 10 U/mL penicillin, 10 µg/mL streptomycin sulfate, and 250 ng/mL amphotericin B. Infection of MB114 endothelial cells was carried out at a multiplicity of infection (MOI) of 5 plaque forming units (PFU) per cell, as previously described [20]. The inoculum was removed after one hour of infection at 37°C, and the cell monolayers were cultured in complete media after rinsing with PBS. Intact and non-adherent cells were collected at six days post-infection, at which time cells and media were collected [20].
C57BL/6 and IFN-g−/− mice on a BALB/c background (strain C.129S7(B6)-Ifngtm1Ts/J) were purchased from The Jackson Laboratory (Bar Harbor, ME). Eight to ten week old mice were infected by intraperitoneally (i.p.) with 1×106 PFU of virus in 0.5 ml of DMEM/5% FBS for reactivation studies and intranasally (i.n.) with 4×105 PFU of virus in 40 µl of DMEM/5% FMS for acute infection studies. Upon sacrifice, lungs for which virus titers were to be determined were placed in 1 ml of DMEM/5% FBS on ice and frozen at 80°C [19]. Peritoneal cells (PECs) were harvested by peritoneal lavage with 10 ml of DMEM/1% FBS [22].
Viral DNA was generated by infection of 3T12 cells at an MOI of 0.05 for each recombinant virus. Infected-cell cultures were harvested at 50% CPE, and DNA was prepared as previously described [31]. 5–10 ug of viral DNA was restriction enzyme-digested for four hours. Digests were electrophoresed on 0.8% agarose gels with biotinylated DNA ladders (New England Biolabs, Ipswich, MA). The DNA was alkaline transferred to Zeta-Probe membrane (Bio-Rad Laboratories, Hercules, CA) using the Turboblot apparatus (Schliecher & Schuell, Keene, NH), according to the manufacturer's recommendations. Probes were from gHV68 genome coordinates 101656 to 105385 (cyclin region probe) or 11100 to 16328 region (left end probe). Each probe was biotinylated and quantitated, according to manufacturer's instructions for the KPL Detector HRP chemiluminescent blotting kit (KPL, Inc., Gaithersburg, MD).
gHV68 clone WUMS (ATCC VR-1465) and recombinant viruses were passaged and grown, and titer determined as previously described [31]. Plaque assays were performed on 3T12 cells as previously described [19], [25]. Lung homogenates were serially diluted, and plated onto NIH 3T12 cells in 12 well plates in triplicate. The limit of detection of the assay is 50 PFU. Viral replication in vitro was determined by infection of 3T12 cells at a MOI of 0.05 PFU per cell to measure multiple cycle replication. Cells and supernatants were collected at various times post-infection and frozen at 80°C. Samples were subjected to four cycles of freezing and thawing prior to quantitation by plaque assay.
Plasmids and BAC DNA were introduced into cells using the calcium phosphate method. 293T, Vero-Cre or 3T12 cells were plated in 6-well plates and transfected at 50–80% confluency. The DNA mixture for each well, which consisted of the 1–10 ug DNA, 2 M CaCl2, and sterile H20, was combined with 2× HEPES balanced saline buffer (0.3 M NaCl, 0.05 M HEPES, 0.003 M Na2HP04, H20; pH 7.05–7.15) and added to each well dropwise while gently swirling plates. At 16 hours post transfection, cells were washed with 1× phosphate buffered saline, the media were replaced with DMEM/5% FBS, and cells were examined by fluorescence microscopy at various times post-transfection to monitor transfection efficiency or BAC-GFP deletion.
Total RNA was isolated from infected 3T12 cells using TRIzol Reagent (Invitrogen) and then purified using the RNeasy Micro Kit (Qiagen). An ABI Prism 7900 sequence detector (Applied Biosystems, Foster City, CA, USA) was used for measurement of the fluorescence spectra in a thermal cycler during PCR amplification (University of Colorado Cancer Center Quantitative PCR Core Facility). Forward and reverse primers and probe (Applied Biosystems) specific to the 3x-FLAG-CMV 7.1 epitope were designed according to the recommendations of the TaqMan PCR chemistry design and optimized using the Primer Express software (Applied Biosystems). Primer and probe sequences used were 3x-FLAG7.1FWD-CTACAAAGACCATGACGGTGATTATAA; 3x-FLAG7.1REV.NEW-TCGCGGCCGCAAGC; 3x-FLAG7.1PROBE-6-carboxyfluorescein-CATGACATCGATTACAAGGATGACGATGAC-6-carboxy-tetramethylrhodamine. Amplification reactions and thermal cycling conditions were performed as per the manufacturer's recommendations. A standard curve was created using the fluorescence data from 10-fold serial dilutions of a 24 hour 3x-FLAG-v-cyclin infection. The 24 hour 3x-FLAG data were normalized to the 12 hour 3x-FLAG data, and is represented as the ratio of 3x-FLAG RNA to the total amount of 18S rRNA per sample.
The following antibodies were used: mouse anti-Flag (M2, Sigma-Aldrich), rabbit anti-Flag (Cell Signaling Technology, Inc, Danvers, MA), rabbit anti-v-cyclin[11]), rabbit anti-k-cyclin [gift from Sibylle Mittnacht [67], rabbit anti-cdk1/cdc2, goat-anti-cdk2, goat anti-cdk4, rabbit anti-cdk6, rabbit-anti-cyclin A, rabbit anti-cyclin D3, rabbit anti-cyclin E (Santa Cruz Biotechnology, Inc., Santa Cruz, CA), mouse anti-beta-actin (Sigma), and donkey anti-rabbit-HRP, donkey anti-mouse-HRP, donkey anti-goat-HRP (Jackson ImmunoResearch Laboratories, Inc, West Grove, PA). Protein expression was detected by lysing cells in ELB buffer (50 mM HEPES pH 7.2, 250 mM NaCl, 2 mM EDTA, 0.1% NP-40) for 20 minutes on ice, and boiling for 10 minutes. Equal cell equivalents or equal amounts of protein, based on RC-DC protein assay (Bio-Rad) were loaded per lane. Samples were separated by electrophoresis on 7.5%–15% denaturing polyacrylamide gels and transferred to Immobilon-P membranes (Millipore Corp., Bedford, MA) by semi dry protein transfer (Panther Semi Dry Electroblotter, Thermo Fisher Scientific, Inc., Portsmouth, NH), and analyzed ECL Plus western blotting detection reagents (GE Healthcare, Piscataway, NJ). 10% of each cell lysate was set aside for lysate loading controls in immunoprecipitations. Remaining lysates in ELB containing protease inhibitors (1 mM DTT, 10 mM NaF, 50 ug/mL PMSF, 1 ug/mL aprotinin, 1 ug/mL leupeptin) were precleared with protein A sepharose CL-4B beads (GE Healthcare) for one hour at 4°C with agitation. Lysates were then clarifed, and incubated for one hour at 4°C with anti-Flag Ab (Sigma) prior to the addition of sepharose beads and overnight incubation. Beads were washed four times in cold ELB with inhibitors, boiled for 10 minutes in Laemmli buffer (0.25 M Tris-HCl pH 6.8, 2% SDS, 10% glycerol, 5% b-mercaptoethanol, 0.002% bromophenol blue) and subjected to SDS-PAGE. Cells infected for immunohistochemical detection of 3x-FLAG cyclins were fixed using 3∶1 methanol: glacial acetic acid. 20 ug/ml of mouse anti-FLAG (Sigma) was added to the cover slips and visualized with goat anti-mouse Alexa Fluor 568 (1∶1000; Invitrogen). For ex vivo FLAG detection, four-six µm sections were deparaffinized and before antigen retrieval using 10 mM citrate buffer. Tissues were blocked using 10% 2.4G2 and 5% goat serum in PBS prior to staining with rabbit anti-FLAG at 1∶500 (Cell Signaling) followed by biotin goat anti-rabbit at 1∶50 (BD Pharmingen) and streptavidin-RPE (Invitrogen) at 1∶100. All slides were mounted with ProLong Gold antifade reagent with 4′-6-diamidino-2-phenylindole (DAPI, Invitrogen) and images were obtained using an Olympus IX81 inverted motorized scope with spinning disk (Olympus, Center Valley, PA), a Hamamatsu ORCA IIER monochromatic CCD camera (Hamamatsu, Bridgewater, NJ) and Intelligent Imaging Slidebook v.4.067 (Intelligent Imaging Innovations, Denver, CO).
The frequency of cells containing viral DNA was determined by a limiting-dilution nested-PCR assay that amplifies gHV68 gene 50 sequences with approximately single-copy sensitivity, as described previously [22], [68]. Briefly, peritoneal cells (PECs) were harvested from latently infected mice and plated as a limiting-dilution series of cells. The cells were lysed prior to PCR amplification, and the first-round PCR product served as a template for the second round of PCR amplification. Control reactions of uninfected cells (negative control) or plasmid DNA (pBamHIN) of known copy number (positive control) were included in each experiment [25].
Quantitation of gHV68 reactivation from latency was performed as previously described [22], [68], [69]. Briefly, PECs were harvested from infected mice at day 42–50 post-infection, and single-cell suspensions were generated. Two-fold serial dilutions of infected cells were plated onto MEFs and scored for CPE after 21 days of co-culture. To detect preformed infectious virus, parallel samples were mechanically disrupted as previously described [25].
Two parameter viability studies using propidium iodine (PI) and annexin V were performed as previously described [20] and analyzed by FlowJo (Treestar, Ashland,OR).
For histologic examination, lungs were fixed in 10% formalin, paraffin embedded, sectioned (4–6 µm) and stained with H&E for analysis using a Zeiss Axiocam HR camera and KS 300 Imaging System 3.0 software [27]. Pulmonary disease was evaluated by board certified pathologist, Dr. Carlyne Cool.
All data was analyzed by using GraphPad Prism software (GraphPad Software, San Diego, CA). Viral titers were statistically analyzed with a one-way ANOVA test. Differences in endothelial cell survival were statistically analyzed by unpaired t-test. The frequencies of reactivation and genome-positive cells were statistically analyzed by paired t-test. Frequencies of latently infected and reactivating cells were obtained from the cell number at which 63% of the wells scored positive for either reactivating virus or the presence of the viral genome based on the Poisson distribution. Data were subjected to nonlinear-regression analysis to obtain the single-cell frequency for each limiting-dilution analysis.
Genbank accession numbers for proteins studied within this manuscript: gHV68 cyclin AAB66456; KSHV cyclin ADQ57958; human cyclin A2 AAM54042; cyclin E1 AAH35498; cyclin D3 AAA52137; cyclin D2 AAH89384.
|
10.1371/journal.pbio.1000278 | Coherence Potentials: Loss-Less, All-or-None Network Events in the Cortex | Transient associations among neurons are thought to underlie memory and behavior. However, little is known about how such associations occur or how they can be identified. Here we recorded ongoing local field potential (LFP) activity at multiple sites within the cortex of awake monkeys and organotypic cultures of cortex. We show that when the composite activity of a local neuronal group exceeds a threshold, its activity pattern, as reflected in the LFP, occurs without distortion at other cortex sites via fast synaptic transmission. These large-amplitude LFPs, which we call coherence potentials, extend up to hundreds of milliseconds and mark periods of loss-less spread of temporal and amplitude information much like action potentials at the single-cell level. However, coherence potentials have an additional degree of freedom in the diversity of their waveforms, which provides a high-dimensional parameter for encoding information and allows identification of particular associations. Such nonlinear behavior is analogous to the spread of ideas and behaviors in social networks.
| Perception and behavior are thought to arise from transient associations among sub-groups of nerve cells in the brain. However, identifying which of the many active neurons are associated at any given time and how poses a challenge. Here we show that when the composite activity of a local group of cortical neurons, measured as a complex waveform in the extracellular field, exceeds a threshold, its activity pattern extending up to hundreds of milliseconds occurs without distortion at other cortical sites via fast synaptic transmission. We call these all-or-none propagated patterns “coherence potentials”, in analogy to action potentials at the single cell level. In contrast to action potentials, which are stereotypical and thus capable only of binary coding, coherence potentials are diverse and complex waveforms that can serve as a high-dimensional parameter for encoding information. The non-linear relationship between local activity and its extent of replicated spread suggests a “tipping point” that bears analogy to the propagation of innovations and economic behavior in social networks, which can spread rapidly once they have garnered a local critical mass.
| Since its introduction by Hebb [1], the transient formation of cell assemblies has been one of the most fundamental and provocative hypotheses to understand cortex function. The existence and identification of such assemblies, however, has been extremely difficult as it requires a criterion that separates the activities of neurons inside the cell assembly from those outside the assembly. Assuming that neurons within an assembly undergo similar changes in their activities, spectral coherence in the local field potential (LFP) reflecting temporal similarities irrespective of amplitudes [2] and synchronization in the spiking activity of individual neurons [3]–[6] have each been used independently to identify functionally associated neuronal groups at different cortical sites during discrete responses to stimuli (for review see [7]–[9]). Many studies focused particularly on similarities in the LFP, as temporal attributes of the LFP waveform have been shown to carry substantial information about a stimulus or behavior [10]–[12], often serving as better predictors of behaviors than spikes [13]–[15]. For instance, in the primate motor and premotor cortex transient phase-locking across cortical sites can be seen in the β- or γ- frequency band of the LFP during voluntary movements and behavioral task performances [16]–[19], which has been proposed to reflect neuronal interactions [20]. In the EEG, transient phase-locking correlates with “gestalt” perception clearly linking similarities in phase across cortical sites to higher cortical function [21]. Importantly, spontaneous or ongoing activity shares many similarities with stimulus-evoked activity [22]–[26], and indeed, transient phase-locking has been observed during ongoing cortical activity in awake monkeys [10],[23]. Because transient phase-locking has been found to include different cortical sites at different times, these dynamics have been widely appreciated in the literature under the concept of “metastability” [8],[27]–[34], emphasizing the idea that cortical dynamics might be sequentially ordered as strings of transient associations.
If similarity is the defining property of associative activity, then the exact replication of activity at different sites might be the ultimate degree in similarity that can be utilized by the cortex. Such precise replication of activities requires the transient phase-locking over a wide range of frequencies in addition to maintaining the amplitude of the signal. Therefore, we analyzed the similarity of the complete LFP waveform across cortical sites during spontaneous cortical activity. Ongoing activity was recorded from chronically implanted arrays in the motor, premotor, and somatosensory regions of two monkeys sitting quietly in a recording chair as well as from organotypic cultures from rat cortex. We demonstrate a sigmoidal relationship between the amplitude of a deflection in the spontaneous LFP, reflecting the aggregate neuronal activity at a single cortical site and the occurrence of the identical LFP waveform and amplitude at other sites with millisecond delays. We show that this phenomenon of all or none, loss-less propagation of activity at high amplitudes arises by virtue of synaptic transmission. These results suggest a “tipping point” in the cortical network dynamics analogous to that found for the spread of ideas, innovation, and economic behavior in social networks, where a small increase in the number of participating agents can suddenly lead to widespread cascades of adoption.
Ongoing LFP activity (∼40 min) was recorded from two macaque monkeys sitting in a monkey chair without having to attend to stimuli or perform motor commands. An array of thirty-two microelectrodes spanning 64 mm2 was implanted in the left motor cortex (M1left) and one array of sixteen electrodes spanning ∼34 mm2 was implanted in the left somatosensory cortex (S1left) in monkey A. Four arrays of sixteen electrodes spanning 34 mm2 were located in the left and right dorsal premotor and motor cortex areas, respectively, of monkey B (see [35] for a sketch of the in vivo array positions and electrode configurations). For the in vitro analysis, spontaneous LFP activity was recorded from organotypic prefrontal and somatosensory cortex cultures of rat grown for many weeks on sixty channel planar integrated microelectrode arrays spanning 2 mm2 [36].
The LFP has been shown to carry substantial information about the underlying spike activity [14],[37],[38]. In our datasets in vivo, amplitudes of negative deflections in the LFP (nLFPs) correlated with both the rate of spike firing as well as the number of distinct units, indicating that the nLFP amplitude reflects the firing rate and degree of synchronization in the local neuronal population [35]. Similarly, the nLFP strongly correlates with the time of neuronal spiking in the in vitro cultures [39]. We thus defined our periods of interest as continuous negative excursions from the baseline (i.e., nLFPs; Figure 1A) whose peaks exceeded thresholds defined in terms of successive multiples of the standard deviation of the signal (SD). These nLFPs varied substantially in duration. Mean duration increased with thresholds up to −1.5 SD and then changed little, if at all, for nLFPs with higher amplitudes reaching about 170 ms in monkeys A and B and 60 ms in vitro (Figure 1D).
For each amplitude threshold, we randomly selected up to 250 suprathreshold nLFPs on each channel (per threshold 4,000–8,000 nLFPs in vivo, and 10,000–15,000 nLFPs in vitro). For each of these nLFPs (we call them triggers), we compared its waveform to the identical period recorded at each other electrode (time-aligned, Figure 1B) by calculating the correlation coefficient R of the time series, which provides a measure of the temporal similarity of the waveform independent of amplitude (examples in Figure 1C and Figure S1). As controls, we calculated similarities between trigger nLFPs and randomly selected duration-matched periods at other electrodes (Figure 1B, red) as well as similarities between random, time-aligned segments, whose durations distributed similarly to the trigger nLFPs (Figure 1B, green; see also Material and Methods “Correlation Analysis and Controls”). At high nLFP detection thresholds, many of the time-aligned nLFPs were highly correlated. The distribution of these correlations showed a characteristic peak close to 0.9 that was not present in the controls (Figure 1E; representative distributions for M1left in monkey A at −4 SD and a single culture at −9 SD). We thus systematically evaluated the average fraction of sites that were correlated R≥0.8 for time-aligned comparisons (Figure 1E, gray area) as we increased the amplitude threshold from low to high SD (Figure 1F). Remarkably, when the nLFP amplitude exceeded approximately −1.5 SD, there was a rapid transition to a large fraction of correlated sites in the time-aligned comparison that was absent in the controls (Figure 1F, open squares; p<10−4 for ≤−2 SD, both controls; Kolmogorov-Smirnov (KS) test). This demonstrates an amplitude dependent, sigmoidal transition from a regime of low spatial coherence to one of high spatial coherence (sigmoidal fit R>0.99, broken line; χ2/DoF<10−4 all cases compared to R values between 0.86 and 0.91 for linear fits). This transition was also found for the somatosensory cortex S1left in monkey A (Figure S2A), suggesting that, in line with the in vitro data, the transition is not limited to motor cortex. Note that we use the term coherence here to refer to waveform similarity in a general sense rather than the more technical spectral coherence, which is limited to a particular frequency band.
In order to take into account potential time delays between trigger nLFPs and similar waveforms at other electrodes, we repeated our analysis, allowing the window of comparison (whose size was determined by the duration of the trigger nLFP) to shift by up to ±10 ms relative to the trigger nLFP (Figure 1B) in order to identify the temporal shift that gave rise to the highest correlation (best-match). The sigmoidal transition to a state of high spatial coherence was robust to this best-match comparison (Figure 1F, filled black squares). Importantly, the significant increase in the fraction of correlated sites for best-match comparisons beyond −1.5 SD suggested that a number of highly correlated nLFPs were shifted in time. We note that the fraction of sites with highly correlated waveforms also increased as the threshold was lowered from −1.0 SD towards 0 SD particularly for best-match comparisons (Figure 1F, open arrow in left panel), which simply reflects the increased likelihood of finding correlations for shorter and shorter segments within a relatively larger time window (cf., Figure 1D).
A systematic analysis of time shifts up to ±200 ms relative to the time-aligned position revealed that the fraction of highly correlated sites increased up to a maximal shift of ±50 ms but changed little thereafter for both monkeys and in vitro (Figure 2A). Thus, most correlated periods at other electrodes occurred within ±50 ms of a trigger nLFP. Note that the maximal average fraction of correlated sites as a function of trigger amplitude is achieved when there are about 30%–40% available sites on the array, and thus the upper part of the sigmoidal function cannot be explained by a ceiling effect due to limited array size. The average distribution of time differences between the trigger nLFP and highly correlated nLFPs at other sites centered around 0, suggesting that any site on the array can be preceded or followed by highly correlated events. The distribution was similar for pairs of electrodes, indicating that the time differences arose from functional rather than anatomical factors such as differences in cortical depth (Figure S3). As expected, the distribution in temporal shifts corresponding to best-match comparisons closely matched the peak-to-peak nLFP intervals (Figure 2B, Figure S2B).
When highly correlated nLFPs identified as above were arranged according to their position in time, the distribution of inter-event intervals revealed that similar nLFPs arose on the array in quick succession with delays rarely exceeding 10 ms (Figure 2C; see also Methods “Time-Shift Analysis” for calculation of delays). In fact, the distribution was skewed heavily towards zero; i.e., many similar nLFPs peaked in the same time bin. We note that such simultaneous occurrence of nLFPs does not necessarily exclude a propagation process as an underlying mechanism (see below). As demonstrated in Figure S4, the spread between highly correlated nLFPs often entailed a one-to-many cascade rather than a strict one-to-one sequence resulting in many zero-delay nLFPs within a cascade. In addition, propagation could have occurred faster than the temporal resolutions of our recordings, which were 2 ms in vivo and 1 ms in vitro. Time delays ranging between 1 and 50 ms translate to propagation speeds of 0.02–1 m/s in vivo and 0.004–0.2 m/s in vitro given inter-electrode distances of 1 mm and 0.2 mm, respectively. This range is generally consistent with stimulus evoked spike response latencies in the organotypic slice [40] as well as stimulus evoked wave initiation in the acute slice [41].
We next performed a similar analysis for amplitude, comparing the peak amplitude of the trigger nLFP to the largest negative peak within the time-aligned window of comparison at other sites (Figure 3A). When we restricted this analysis only to sites with correlated waveforms (R≥0.8), we found that the normalized peak amplitude values distributed narrowly around 1 (Figure 3B; M1left monkey A; shown in log scale; 0 = log2(1)), indicating that the amplitudes of correlated waveforms were highly similar. When the analysis was extended to include all sites, irrespective of the correlation of their waveforms, the distribution retained its peak around 1, but now broadened because of the inclusion of many amplitude ratios <1 originating from non-similar waveforms with on average smaller amplitudes than the trigger nLFP. As a further control, time-aligned comparisons of random segments with durations comparable to those of trigger nLFPs lacked a sharp peak at 1 demonstrating a low a priori probability of waveforms having similar peak amplitudes. Importantly, the fraction of sites with similar peak amplitudes (i.e., within ±50%, Figure 3B shaded area) increased non-linearly as a function of nLFP amplitude threshold in parallel to the increase in waveform similarity (Figure 3C, open arrow; data for monkey B and in vitro shown in Figure S5). This transition was equally pronounced when only highly correlated waveforms were considered (R≥0.8; Figure 3C, filled arrow), and indeed, amplitude similarity was greater for the more correlated waveforms. These results indicate a non-linear transition to a regime where there is not only high similarity in waveform at different sites but amplitude as well.
Finally, we demonstrate that the sigmoidal transition to a spatially extended regime of high coherence was not due to an overall greater similarity of nLFPs with large amplitudes, as one might expect due to an improved signal-to-noise ratio. We estimated the a priori likelihood of finding similar nLFP waveforms at distant times as a function of amplitude. We compared nLFP triggers at each threshold to nLFPs with similar amplitude at each other electrode randomly chosen from time periods at least 200 ms away from the peak of the nLFP trigger. The fraction of such comparisons that were highly correlated (R≥0.8) was 2- to 3-fold smaller even for the highest amplitudes both in vivo and in vitro (Figure 1E, blue; R = 0.12±0.01 to 0.27±0.01 for amplitudes ≤−4 SD in vivo; 0.52±0.01 to 0.6±0.01 for amplitudes ≤−9 SD in vitro). This implies that while large amplitude nLFPs clustered in time were highly similar, temporally distant clusters were highly dissimilar.
We have thus far shown that, as the nLFP amplitude increases, there is an abrupt transition to a regime where a pattern of activity appears at a large number of sites with millisecond delays without distortion of its temporal structure or substantial change in amplitude. We will call these highly similar temporally clustered nLFP waveforms with similar large amplitude “coherence potentials” to reflect their wide spatial coherence. Given that the extent of spatial coherence can vary considerably, we operationally define a coherence potential as an nLFP waveform with suprathreshold amplitude (right side of the sigmoidal function) that occurs at least at one other site in the network. In Figure 4, we provide a detailed visualization of individual coherence potentials as they are identified in the successive occurrence of 100 suprathreshold nLFPs (i.e., ≤−3 SD) during a segment of recording from M1left in monkey A (note that no information on the spatial positions of the nLFPs is used in this representation). First, we plotted in matrix form the time difference between all nLFP peaks (Figure 4A), which reveals temporal clusters of nLFPs (i.e., within a few milliseconds of one another, black–red squares along the diagonal). The corresponding correlation matrix (Figure 4B, see also Material and Methods “Visualization of Coherence Potentials”), which measures the similarity of nLFP waveforms to each other, reveals that temporally clustered suprathreshold waveforms (separated by <10 ms) tend to be highly similar in waveform (red). In contrast, waveforms of successive temporal clusters (generally separated by >50 ms, white) were no more similar to one another than randomly chosen nLFPs. Thus, in most cases, coherence potentials are readily identified in the ongoing activity by simply over-plotting successive suprathreshold nLFPs arising with a maximum delay of 10 ms from each other (note that single waveforms, by definition, cannot classify as coherence potentials and are marked by an asterisk). These over-plots readily demonstrate the similarity of waveforms and their amplitudes within a coherence potential, and the diversity of waveforms across coherence potentials (Figure 4C).
In other instances, the temporal clusters of suprathreshold nLFPs were less precisely delimited (Figure 4D; e.g., clusters a and c) reflecting temporally intermingled coherence potentials that gave rise to a “checker board”–like organization in the correlation matrix (Figure 4E). Here, although over-plots based on a simple temporal criterion (i.e., maximum delay of 10 ms) did not result in identical waveforms, sorting the waveforms based on a correlation threshold criterion (here R≥0.8) readily uncovered the multiple coherence potentials that were temporally intermingled (Figure 4F). Thus, temporal cluster a was composed of five distinct waveform groups of which three were coherence potentials, while temporal cluster c consisted of five distinct waveforms of which one grouped with cluster b and three with cluster d (see Figure S6 for more raw coherence potential traces in vivo and in vitro).
We then studied the waveform characteristics of coherence potentials in more detail and asked in a first step whether the negative excursion of the nLFP described their full extent in time (cf., Figure 4A–C; clusters “a”, “b”, “n”). For each suprathreshold nLFP (≤−4 SD in vivo; ≤−9 SD in vitro), we systematically increased the time of comparison before and after the nLFP peak for up to ±500 ms (Figure 5A) or until R<0.8. The identified durations with R≥0.8 distributed with a heavy tail spanning a wide range up to 500 ms with median values of 200 ms (Monkey A, M1left) and 178±22 ms (monkey B, average of four arrays), respectively, and 92±12 ms in vitro (n = 6 cultures). Thus, coherence potentials can last for up to many hundred milliseconds. During this period, most coherence potentials included a combination of one negative and one positive excursion from the baseline (>90%; Figure 5C). Indeed, hardly 10% of coherence potentials extended over more than three baseline crossings, the minimal expected number for a two-cycle oscillation, suggesting that they did not originate from an underlying oscillation.
Significantly, the period of highest correlation was initiated at a fast rise phase generally around the −1 SD mark but continued for up to several hundreds of milliseconds after the last threshold crossing (Figure S7). Thus, while the nLFP alone does not describe the full extent of the correlation, it is a sufficient approximation for the purpose of this study and provides a useful method of extracting periods of interest.
The power spectrum density (PSD) of coherence potentials rarely revealed dominant oscillations. We calculated the PSD for coherence potentials whose negative component (i.e., nLFP) was at least 256 ms long, using exactly 256 ms for the power spectrum analysis. The shortness of these segments restricts the reliable region of the PSD to frequencies >4 Hz for which the PSD showed a systematic decay without any dominant frequency band in the aggregate (Figure 5D) as well as at the level of individual coherence potentials. A similar decay was observed for periods of equal durations uncorrelated with other sites (i.e., −0.3≥R≤0.3, red) and the complete recording (Figure 5D). In order to get a more accurate view of frequencies below 4 Hz, longer segments of ±4 s around the trigger nLFP were also considered, but again the decay was devoid of dominant frequencies (see Figure S8A, S8B for examples of individual segments). Thus, while coherence potentials do not reflect a specific oscillatory period, this does not exclude the possibility that some coherence potentials could have oscillatory components. Indeed, autocorrelation analysis revealed oscillatory components at various frequencies in a small fraction of individual coherence potentials (Figure S8C). Importantly, when coherence potentials were phase-shuffled, the correlations between sites were destroyed (Figure 5E), indicating that coherence potentials involve specific phase relationships among frequencies.
The non-linear transition to high spatiotemporal coherence crucially depended on fast excitatory and inhibitory synaptic transmission. In culture, where controlled pharmacological manipulation was possible, reduction of fast excitatory synaptic transmission with 2 µM of the AMPA receptor antagonist DNQX (n = 3) reduced the fraction of sites correlated with R≥0.8 by over 50%, particularly at higher amplitudes (Figure 6A; p<10−5 for nLFP triggers ≤−9 SD; note that in order to avoid any distortion due to changes in the LFP amplitude distribution in the presence of the drug, amplitude values used for both pre-drug and drug comparisons are the absolute amplitude values corresponding to pre-drug SDs). Similarly, the fraction of correlated sites with similar amplitude was also dramatically decreased by DNQX application (p<10−4; Figure 6B). Such a dependency on AMPA-receptor mediated activity indicates that the appearance of waveforms of similar time course and amplitude at different sites is primarily a consequence of fast synaptic transmission and cannot be simply explained by volume conduction, overlapping electrode fields, or electrode filtering characteristics.
Partial reduction of fast inhibitory synaptic transmission also had strong effects, distinct from those of DNQX (Figure 6C–E). Application of 5 µM of the GABAA receptor antagonist Picrotoxin (PTX, n = 3, raw trace examples in Figure S9) destroyed the sigmoidal amplitude dependence such that the fraction of sites correlated was no longer dependent on the nLFP trigger amplitude (Figure 6C). The maximal spatial coherence was reduced by ∼30% for high amplitude nLFP triggers (Figure 6C; p<10−2 for nLFP triggers ≤−9 SD) and concomitantly increased at small amplitudes. In addition, PTX significantly reduced the amplitude similarity of highly correlated waveforms (R≥0.8; Figure 6D, p<0.05) introducing much greater amplitude variability. Indeed, temporally clustered sequences of high amplitude nLFPs had far lower waveform similarity relative to pre-drug conditions (Figure 6E, cf., Figure 4B). Thus, fast synaptic inhibition is required for the existence of distinct low-coherence and high-coherence regimes.
The finding that waveforms of similarly low amplitude and comparable durations have higher spatial coherence in the presence of PTX indicates that noise, which would be expected to obscure correlations for low amplitude signals, is insufficient to explain the low spatial coherence seen for small amplitudes under normal conditions. Thus, the sigmoidal transition to higher coherence at higher amplitudes cannot be explained by an increase in signal to noise ratio. This is also supported by the large deviation of the overall LFP amplitude distribution from a Gaussian fit both in vitro and in vivo (Figure S10), which suggests that commonly occurring Gaussian noise would have little influence in the range of −2 to −3 SD, where the non-linear transition occurs.
The previous section demonstrated that the regime of high spatial coherence reflects a loss-less, undistorted propagation of activity that depends crucially on fast AMPA and GABAA mediated synaptic transmission. We next systematically examined the relationship between successively occurring nLFPs (Figure 7) to determine if there was a sequential dependence that was unique to the high threshold regime of coherence potentials. Evidence for sequential dependence would also give credence to an intrinsic propagation process in vivo, where sub-cortical input was present.
Given that correlated waveforms tended to be nLFPs of similarly high amplitude (Figure 3) and that time delays between sites corresponded closely to peak alignment of the nLFP waveforms (Figure 2B), sequences were defined as nLFPs identified at any site on the array whose peaks crossed a certain threshold and occurred in successive time bins of 2 ms (Figure 7B). For simplicity, sequences with only one nLFP peak in the first time bin were used (>50% of all sequences), although successive time bins often contained multiple nLFPs (cf., Figure S4).
We first examined sequences of nLFP peaks that exceeded a threshold identified by the top of the sigmoid transition to high spatial coherence (Figure 7A; −4 SD in vivo; −9 SD in vitro; cf., Figure 1F). We reasoned that if nLFP waveform similarity across sites arose as a sequential, loss-less process intrinsic to the cortical network, then nLFPs occurring n ms apart that formed part of a sequence (Figure 7B, top) would be more stable in their waveform similarity across each step in the sequence relative to amplitude-matched nLFPs separated by similar time intervals that were not part of a sequence (Figure 7B, bottom). Remarkably, nLFPs within a sequence were highly stable in their correlation relative to the initial nLFP; the median correlation of nLFPs at each position in the sequence relative to the first nLFP decreased only slightly from ∼0.94 to 0.86 in nine time steps in vivo (≈18 ms, Figure 7C, top) and from ∼0.78 to ∼0.75 in twelve time steps in vitro (≈12 ms, Figure 7C, bottom). In contrast, correlations between nLFPs separated by correspondingly larger intervals were significantly worse than their within sequence counterparts (p<0.02 up to p<10−5 by KS test), decaying to no better than comparisons of random nLFPs of similar amplitude within 15 to 30 ms (Figure 7C, open squares; median random correlations were 0.47 and 0.49±0.013 in vivo and 0.58±0.03 in six cultures, p<10−4 when compared to within sequence correlation for all cases). In sharp contrast to this behavior of coherence potentials, sequences constructed predominantly of small or subthreshold nLFPs, i.e., before the transition to the regime of high spatial coherence (Figure 7A), decayed progressively to random in a manner that was no different from interval-matched comparisons both in vivo as well as in vitro (Figure 7D). Thus, the sigmoidal increase in spatiotemporal coherence reflects a transition from a regime of progressive distortion of the waveform time course within sequences to spatially extended sequences that stably maintain waveform amplitude and time course.
In vitro, where there was no source of sub-cortical input, this indicates a clear sequential process intrinsic to the cortical network. In vivo, such sequential dependence is not likely to arise from sub-cortical input and similarly points to cortical propagation. However, we also considered one additional possibility: that subcortical input arrives at different sites on successive cycles of a subcortical oscillation (e.g., thalamocortical spindles) creating an indirect sequential dependence. Inconsistent with this scenario, the time delays between successive nLFPs in individual sequences that had at least ten correlated nLFPs (R≥0.8) were widely distributed (Figure S11; cf., also Figure 2B) indicating the absence of any characteristic time scale. Thus, all evidence points to a sequential process that is intrinsic to the cortical network both in vivo and in vitro.
Propagation in the cortex has been widely described as wave-like. We therefore looked to see if the propagation moved progressively in a wave front. Because time delays between successive coherence potentials were broadly distributed (cf., Figure 2B), it is not possible to determine the true order of occurrence of individual coherence potentials within a large sequence. Thus, we restricted our analysis to cases of coherence potentials where there were only two highly correlated nLFPs either in the same or contiguous time bins. Remarkably, in both monkeys, over 70% of pairs of correlated nLFP waveforms occurring in the same or contiguous time bins were not at adjacent electrodes, irrespective of the temporal bin width (Figure 8A, 8B, shown are time bin Δt = 2 and 20 ms). Moreover, the median correlations of large amplitude nLFPs (≤−3 SD) occurring in the same or adjacent time bins were not different irrespective of the distance separating them (Figure 8C). Thus, there was no evidence of distance dependent distortion. This points, surprisingly, to non-contiguous or saltatory propagation quite unlike a wave, and is clearly contrary to expectations of volume conduction and overlapping electrode fields where similar waveforms would be found at adjacent sites.
We have previously shown in the present in vivo data that higher nLFP amplitudes are systematically associated with higher local firing rates as well as a higher number of simultaneously firing neurons or synchrony [35]. This suggests that the amplitude reflects a certain degree of spike synchrony or aggregate spike activity, lending credence to the possibility that particular waveform patterns reflect a particular pattern of spike activity. We thus looked for evidence of a preservation of the underlying pattern of spike activity associated with coherence potential propagation.
While our LFP recordings represent the aggregate activity of several hundred neurons in the local field of the electrode, only a very small number of units close to the electrode can be resolved, representing a very small sampling of the neurons contributing to the aggregate activity. Most electrodes resolved either one or two units, while a small fraction resolved three or four units. Indeed, often there were no units detected in association with a coherence potential indicating a lack of participation of the resolved units. Our analysis was thus necessarily limited to those coherence potentials that had associated unit activity. These were spread across many sequences offering a large diversity of waveforms and therefore waveform correlations within and across different, diverse coherence potentials. We thus made comparisons of the aggregate pattern of unit activity as a function of the similarity of the associated coherence potential waveforms (i.e., nLFPs with peak amplitude ≤−4 SD; cf., Figure 1F). Comparisons were made by calculating the dot product of the summed unit rasters binned at 6 ms controlling carefully for the number of units and duration of comparison, both factors that affect the a priori expectation of similarity (see Material and Methods for details). Indeed we found that the more correlated the coherence potential waveforms at different sites, the more similar the pattern of aggregate unit activity (Figure 9A; linear fit, M1left monkey A: R = 0.96, p<10−4; monkey B: R>0.94, p<0.005, all arrays). A similar pattern was observed for binning of units anywhere between 2 and 10 ms bins (unpublished data), 6 ms, shown here, represented a midpoint. Some examples of cases with preserved temporal pattern of spikes are shown in Figure 9B.
These findings argue strongly in support of a most unexpected conclusion, that when the activity of many neurons in a local field are sufficiently synchronized, the aggregate activity of these neurons is able to propagate to distant sites without distortion of the overall temporal pattern or substantial change in the number of participating neurons.
Here, we have identified a quasi-discrete network object in the LFP, which we term the coherence potential that arises at high amplitudes with a non-linear relationship. A coherence potential is characterized by a complex waveform comprised of negative–positive excursions that exceeds a certain local amplitude threshold and is associated with a synaptic transmission dependent saltatory, cascade-like spread through the cortical network without distortion of its temporal structure or substantial change in amplitude. The transition from a regime where waveforms propagate with progressive distortion (subthreshold potentials) to one without such distortion (coherence potentials) occurs according to a sigmoidal function of LFP amplitude, suggesting a threshold dependent process. This phenomenon could not be explained by enhanced signal to noise at higher amplitudes (Figures 1F, 3, 6, S10), volume conduction, overlapping electrode fields, electrode characteristics (Figures 3, 6, 8), or common input from sub-cortical sites (Figures 6, 7, S11). Rather, its occurrence in both explant cortical tissue and the cortex of awake monkeys suggest that it is an intrinsic property of cortical networks. This has phenomenological parallels with the action potential, which describes a large amplitude potential across a local patch of excitable membrane that represents a non-linear regime in which activity propagates in a loss-less manner. However, unlike action potentials, which are stereotypical waveforms, coherence potentials are highly diverse in their waveforms. Distinct coherence potentials, formed of sequentially occurring nLFPs, often occurred in rapid succession as a stream of dynamical associations (Figure 4) representing an identifiable switching of the cortical network from one dynamical state to another.
The propagation of coherence potentials was saltatory in nature, rather than wave-like, frequently “jumping” across neighboring sites (Figure 8B) such that the fidelity of propagation was not substantially impacted by physical distance (Figure 8C). How might this phenomenon be reconciled with wave-like propagation ubiquitously reported in the cortex using techniques ranging from voltage sensitive dye imaging to LFPs and spikes recordings [10],[42],[43]? While this remains to be clearly understood, we provide a hypothetical framework that could reconcile these two findings. Coherence potentials arise at high thresholds of −4 SD in vivo and −9 SD in vitro, thereby representing only a very small fraction of nLFP activity in the cortex (<1%). This means that most of the activity one measures is sub-threshold in nature. Thus, in the absence of knowledge of a threshold dependent process of loss-less propagation, one would include both sub-threshold and supra-threshold activity, whereby alternative dynamics, such as wave-like propagation, might dominate. We therefore suggest that sub-threshold network activity may spread in a wave-like manner analogous to the spread of a sub-threshold depolarization in a neuron. We note that both waves and coherence potential propagation depend crucially on inhibition [41]. It will be interesting in the future to understand the mechanistic similarities and differences.
The amplitude dependence of the transition to loss-less propagation indicates that coherence potentials are characterized by a certain level of synchronization in the local field [35]. Theory has suggested that synchrony might propagate differently from single spikes. For example, synfire chain models in feed-forward networks report that the size of the initial synchronized group determines propagation fate in a threshold dependent manner [44],[45]. In these basic models, however, there is convergence to a simple waveform as activity propagates, which is in contrast to coherence potentials, which can be highly diverse in their waveforms, and where the original temporal properties of the waveform or underlying synchronized group are preserved in the propagation to other sites. This interpretation equates the time course of a local nLFP with the synchronization of inputs arriving at a single layer of a synfire chain and where the coherence potential itself, which spatially extends over different sites, would reflect the various, spatially distinct layers of the chain. Alternatively, and in line with the general argument that only local connectivity between pyramidal neurons is high enough to sustain a synfire chain [44],[46], the nLFP time course could reflect the synchronization time course of a local synfire chain. In that case, coherence potentials would reflect phase-locking of distant synfire chains as suggested in studies on compositionality by simulating several chains that are weakly coupled through long-range connections [47].
Additionally, once recruited in this manner, sites may become locked into recurrent interactions that “attract” each site to a common pattern of activity for a short period of time. Such an interpretation is consistent with the long durations of the coherence potentials (several tens to hundreds of milliseconds) relative to the rapid time scales of propagation (generally a few milliseconds) and is supported by the high degree of reciprocal connectivity observed in cortical networks [48]–[51]. This interpretation has parallels to the principles underlying the construction of attractor neural networks or Hebbian cell assemblies [1],[52]. Evidence for recurrent interactions has also been described in the propagation of waves across cortical areas [53]. Such inter-areal phenomena also raise interesting questions about how propagation of coherence potentials across cortical regions may occur and how coherence potential originating at different cortical regions may interact.
Finally, we note that in each of the datasets used in this study, nLFPs have been shown to organize as neuronal avalanches [35],[54]. Neuronal avalanche organization is a statistical property characterized by scale invariance in the temporal and spatial clustering of nLFPs suggesting that propagation could in theory span the entire cortex on a wide range of time scales [55]. Furthermore, partial reduction of fast GABAergic transmission abolishes the neuronal avalanche dynamics [35],[54] as well as the sigmoidal dependency of coherence potentials (cf., Figure 6C). It is thus of considerable importance to further understand the sigmoidal, non-linear transition to the coherence potential regime that reflects propagation without distortion and the general scale-invariant organization of nLFPs in the context of neuronal avalanches that may govern the organization of successive coherence potentials.
The greater similarity of aggregate unit firing patterns at sites with similar LFP waveforms (Figure 9) suggests that coherence potential propagation reflects a preservation of the temporal characteristics of the underlying neuronal activity. Given that electrode noise and the spatial arrangement of the participating neurons relative to each electrode contribute to differences in waveforms observed at distinct sites [56], the precision with which coherence potentials propagate in the network might be even higher than we measured. The amplitude dependence and the initiation of high correlation periods on the rise phase only milliseconds before the peak (Figure S7) suggests that it is the synchronization of activity itself that acts as the trigger. Indeed, at the level of the individual neuron, increasing synchronization of input gives rise to a non-linear increase in precision and reliability of the input-output relationship by altering the threshold for spike formation [57]–[60]. Computational models also suggest a crucial role for fast inhibition in regulating the precision and reliability of output [61]–[63], consistent with the dependence of coherence potential propagation on GABAA dependent transmission. However, even with these findings at the single cell level, it remains to be understood how locally, synchronized activity can be directed to another site in an all-or-none manner.
Unlike the stereotypical action potential, which can serve only as a binary code, the diversity of coherence potential waveforms represents a high dimensional degree of freedom, which could be used to encode information. In contrast, coherence potentials demonstrate the maintenance of spectrally complex waveforms and therefore indicate phase-locking or spectral coherence between sites across a broad range of component frequencies. While coherence potentials in our data showed no dominant frequency, this does not preclude them from exhibiting clear oscillatory components in the waveform under certain stimulus or behavioral conditions. Indeed in a small fraction of cases, we did find an oscillatory component (Figure S8). We propose that oscillatory behavior may represent a certain class of coherence potentials, analogous to a burst of action potentials, where the duration of waveform correlation may be extended across a greater number of positive–negative cycles. This proposition, of course, must be tested empirically but points to a new approach to LFP analysis, where, by not restricting analysis to a particular frequency band, it might be possible to make finer distinctions in the information contained in the complex structure of the LFP waveforms.
A fundamental question that arises is how to reconcile the occurrence of similar associations both in stimulus deprived explant networks and the awake animal in a functional context. Here we suggest that just as the action potential arises by virtue of the intrinsic properties of the cell in a dish but takes on meaning in the intact organism by virtue of an input-output feedback mechanism, coherence potentials, which likely represent an aggregation of spiking activity, arise by virtue of intrinsic properties of the cortical network and take on meaning in a similar fashion. That temporal properties of the LFP waveform often serve as better predictors of behaviors than spikes [10],[12]–[15],[64]–[66] is consistent with such an explanation, suggesting that coherence potential propagation in vivo represents the transfer of meaningful information in the cortical network.
Sequences of propagated coherence potentials occurred in rapid succession as a stream of dynamical associations (Figure 4) representing an identifiable switching of the cortical network from one dynamical state to another, which is a hallmark of “metastability” [8],[27]–[34]. However, the question remains how the cortex might make use of the specific phenomenon of coherence potentials, where many sites mirror one another with millisecond delays, reflecting transmission of information without any distortion. One possibility is that such propagation may serve as a form of working memory. However, to spark debate and discussion, we put forth a more radical hypothesis, drawing from the analogy to the spread of ideas, innovation, and economic behavior in social networks, which tend to display a threshold or “tipping point” such that a small increase in the number of participating agents can suddenly lead to widespread cascades of adoption [67],[68]. The initial adoption of the idea or behavior often arises among geographically proximal agents [69],[70] on networks with small-world principles [71] similar to those observed in cortical networks [72]–[75]. “Tipping points” are commonly modeled under conditions where individual agents make binary decisions (e.g., to buy or not to buy) by integrating the incoming information from their local neighborhood of friends [76], in close parallel to the “integrate and fire” property of neurons. We propose that perhaps information competes to pervade the cortex analogous to the way that ideas compete to pervade society, originating from various sensory areas in the way that ideas arise in society from specialized regions or areas focused on particular subjects or activities. Perhaps it is the case that when a sufficient number of sites across multiple cortical regions have adopted a particular module of cortical information, this now has the ability to impact the behavior of the organism. Such a hypothesis is not entirely unwarranted. For instance, coordinated cortical activity has been shown to be associated with motor behavior [10],[17], the spread of voltage-sensitive dye signals from sensory to motor areas of cortex and their amplitudes have been found to be correlated with animal whisking behavior [77], and the diversity found in average LFP waveforms across many electrodes have been shown to encode behaviors [78]. We note that the diversity of coherence potential waveforms in vivo was much greater than in vitro indicating a richer associative environment and we suggest that a comprehensive characterization of coherence potential waveforms and their spatial spread in relation to behavior is likely to be highly instructive with respect to a broad range of cognitive functions.
The ACUC of Duke University approved all procedures. Arrays of monopolar tungsten electrodes (30 µm in diameter, 1 MΩ impedance; 1 mm spacing) were chronically implanted into the cortex of two adult rhesus monkeys (Macaca mulatta) [79]. Briefly, arrays were inserted 1.5 mm deep into the leg representation area of the motor cortex (M1left) and 1 mm deep in the somatosensory cortex (S1left) in the left hemisphere of monkey A. In monkey B, four arrays were similarly implanted in the arm representation of the left and right motor cortex (M1left, M1right) and dorsal premotor cortex (PMdleft, PMdright) (for sketch in electrode arrangements and array positions, see [35]). During the ∼40-min-long recording sessions analyzed here, the monkeys were awake and seated in a monkey chair with the light in the recording room turned off. LFPs were sampled at 500 Hz (National Instruments) and band-pass filtered between 1 and 100 Hz using the idealfilter function in MATLAB. Results were subsequently verified for M1left in monkey A using a phase-neutral filter (MATLAB filter function “filtfilt,” Figure S12). No difference was found between the two filter approaches.
Extracellular spiking activity was sampled at 40 kHz and band-pass filtered with a 2-pole low-cut and a 4-pole high cut filter at 0.4–8 kHz. Off-line unit discrimination was based on principal component analysis and spike-template matching (for details, see [79]). Simultaneous recordings were carried out in M1left and S1left (monkey A) and the four arrays in M1 and PMd (monkey B). LFP and unit activity was sampled simultaneously from every other electrode resulting in 32 electrodes for M1left and S1left for monkey A and 16 electrodes per array in monkey B using a dual amplifier Plexon system (Figures 1A and S5; filled circles). An epidural stainless steel T-bolt, at least 20 mm from all recording areas, served as a common ground (for a detailed sketch of electrode location and array positions, see [35]). The 68 units resolved in M1left (32 electrodes) fired on average at 4.3±3.7 Hz. For the four arrays in monkey B, on average 47±13 units were resolved (16 electrodes/array), which fired at a rate of 5.7±1.8 Hz. Unit firing rates were not significantly different between monkey A and monkey B (Student's t test, p = 0.3).
Some periods in the recording exhibited higher amplitude, slow-wave activity that was characteristic of sleep-spindles (90% of power at 1–10 Hz [80]); periods with >50% channels showing this type of activity accounted for 6% to 12% of the recordings in monkeys A and B, respectively, and contributed between 5.5% and 9.5% of high amplitude nLFPs (≤−3 SD), not substantially different from the rest of the recording.
Organotypic cortex slice cultures were prepared as previously described [36],[54]. Briefly, coronal section of somatosensory and prefrontal cortex from postnatal day 1–2 old rats were grown on planar integrated 60-channel microelectrode arrays (8×8 grid; 30 µm diameter titanium nitride electrodes; Multichannel Systems). One enlarged electrode served as the common ground. At an electrode spacing of 200 µm as used in the present study, there is no detectable overlap between electrode fields [81]. After 4–6 wk of cultivation, spontaneous LFP activity emerged and was recorded for up to 5 h and filtered between 1 Hz and 50 Hz. To study the effects of changes in synaptic transmission, after 2 to 5 h of baseline recordings, either DNQX or picrotoxin (PTX) was added directly into the medium of the cultures (final concentration 2 µM and 5 µM, respectively) and recordings were continued for 2 to 5 h.
Correlation analysis was carried out using in turn each electrode as the trigger electrode and selecting a large number of nLFPs (n≥250) of each amplitude from the recording on this electrode as triggers. Correlations R were then calculated between the nLFP trigger (X) and a period of equal duration (Y) on each other electrode aswhere and var indicates the variance of the signal. To prevent double counting of nLFPs, the distributions of correlations were obtained for each trigger electrode and the fraction of sites correlated R≥0.8 were averaged across electrodes for Figure 1F. The controls (Figure 1B, E, F) were done as follows. To control for random similarity (Figure 1B, E, F; red), a random period of similar duration to the nLFP trigger was selected by beginning from a random position chosen from anywhere in the recording that was at least 100 ms away from the peak of the nLFP trigger. To control for the large variability in the duration of trigger nLFPs (Figure 1D), a factor that could greatly influence correlations, we generated random time-aligned and time-shifted comparisons across electrodes that had the same distribution of durations as the nLFP triggers but were not nLFP triggered. Because these were not nLFP triggered, they controlled only for the durations seen at particular nLFP amplitude thresholds (Figure 1C, E, F; green) and not the amplitudes themselves. To control for the possibility of an overall greater similarity between large amplitude deflections (Figure 1F; blue), for each nLFP trigger we made a comparison to an amplitude-matched nLFP at each other electrode and then identified the fraction of cases that were ≥0.8. To do so, all nLFPs arising on the comparing electrode that were of comparable amplitude (within ±10%) and whose peaks were at least 100 ms away from the nLFP trigger peak were identified and a random selection was made from this group. Because amplitude-matched nLFPs were often of different durations, the period of comparison was defined as the duration of the trigger nLFP before and after its peak. For the controls in Figure 7, all nLFPs of comparable amplitude on all electrodes were pooled together and 500 randomly selected pairs were compared, creating a distribution of random correlations for different interval times. In this case, variability of nLFP duration was dealt with by using segments corresponding to the length of time from the first start to the last end of the nLFP pair.
Time-shift analysis (Figures 1B, 1F, 2A) was carried out by moving the window of comparison on each comparison electrode one time bin at a time in both directions relative to the nLFP trigger window position (bin width Δt = 1 ms in vitro, 2 ms in vivo) and identifying the shift corresponding to the highest correlation. For the random and duration controls, the window was similarly shifted relative to the originally chosen random position. Intervals between successive correlated periods were calculated by assuming a one-to-one sequential process (Figure 2C). For example, if there was one nLFP in the first time bin and three correlated nLFPs in the second time bin, the inter-event intervals would be counted as 1×Δt between the first nLFP and the first nLFP in the second time bin (nLFP order was arbitrarily chosen based on electrode number) and 0 between the second and third nLFP and third and fourth nLFP in that second time bin, giving a distribution of delays of 2 zeros and 1 Δt.
In order to quantify amplitude similarity as a function of nLFP trigger amplitude (Figure 3), peak amplitudes were identified as the minimum within the time-aligned window of comparison. Since the duration of the nLFP, which set the window of comparison, was generally many tens to hundreds of milliseconds, and since best-match time delays were usually <50 ms (Figure 2), this ensured that in most cases, the true peak was identified. The peak amplitude of the comparison period was then normalized by the peak amplitude of the nLFP trigger.
Best-match correlations and peak-to-peak time differences were calculated for each pairwise combination of 100 (Figure 4A, B) and 30 (Figure 4D, E) nLFPs ≤−3 SD whose peaks occurred on the array during a 15 s and 250 ms period, respectively. Because nLFP periods were of variable duration, correlations were calculated for the period corresponding to the maximal nLFP duration preceding and following the nLFP peak of the nLFPs in the pair. Over-plots shown begin 50 ms before the peak crossing to 250 ms after the peak in order to ensure the plots are peak aligned and to encompass the complete duration of the longest nLFP. Note that most of the nLFPs were much shorter than 300 ms, while the matched period in the over-plots often extended beyond the nLFP indicating periods of high correlation beyond the nLFP.
Power spectrums of coherence potentials with negative deflection periods of at least 256 ms in length were calculated using the Fast Fourier transform (FFT, MATLAB) for a period of exactly 256 ms (Figure 5D). The PSD was also calculated using the FFT of the autocorrelation function with similar results (see Figure S8). Phase dependency of coherence potentials was studied (Figure 5B) by calculating the FFT of the original coherence potentials, shuffling phase angles, and obtaining a new LFP time course using the inverse FFT. Correlations were obtained for original coherence potentials and corresponding phase-shuffled waveform periods.
Several controls and considerations went into this analysis (Figure 9). In order to ensure that we were working with the maximal amount of unit data given the very small sampling of units relative to the population contributing to the LFP, we made comparisons of the aggregate or summed unit activity at each electrode rather than individual units. For each electrode, this was done simply by summing the number of units in each time bin Δt resulting in a single unit activity vector for each electrode. At bin widths Δt = 2–10 ms used here (Δt = 6 ms shown), these vectors mainly contained values of 0 and 1, occasionally the values 2 and 3. The dot product between two vectors was used to quantify the similarity in unit activity patterns between electrodes. Because coherence potentials varied in duration, corresponding unit activity vectors varied in their number of time bins. In addition, the number of units detected between electrodes also varied, affecting the a priori expectation of empty time bins. Since duration and number of units impact the a priori expectation of the dot product, i.e. similarity, it was essential to precisely control for these factors. We kept the duration of comparison constant by choosing only those coherence potentials that were at least 200 ms in duration and comparing the period starting 50 ms before the nLFP peak until 150 ms after the peak. We further restricted the comparisons to only those cases where there were exactly two unit occurrences during this period (presumably two spikes, n = 520 in monkey A and n = 200–400 in monkey B for 3 out of 4 arrays. In the remaining array of monkey B there were n = <100 cases, which was inadequate for the analysis across the range of coherence potential correlations). An example with a dot product of one is shown in Figure 9A. Single unit occurrences were excluded from analysis because it did not allow conclusions about temporal structure, while three or more unit occurrences were too rare for statistically relevant aggregation (<75 in all cases).
For statistical significance between dissimilar distributions, the KS test was used. For curve fitting, regression analysis was performed using a standard linear or sigmoidal function (Origin). Significance was established at p<0.05.
|
10.1371/journal.ppat.1002405 | Kaposi's Sarcoma Herpesvirus microRNAs Target Caspase 3 and Regulate Apoptosis | Kaposi's sarcoma herpesvirus (KSHV) encodes a cluster of twelve micro (mi)RNAs, which are abundantly expressed during both latent and lytic infection. Previous studies reported that KSHV is able to inhibit apoptosis during latent infection; we thus tested the involvement of viral miRNAs in this process. We found that both HEK293 epithelial cells and DG75 cells stably expressing KSHV miRNAs were protected from apoptosis. Potential cellular targets that were significantly down-regulated upon KSHV miRNAs expression were identified by microarray profiling. Among them, we validated by luciferase reporter assays, quantitative PCR and western blotting caspase 3 (Casp3), a critical factor for the control of apoptosis. Using site-directed mutagenesis, we found that three KSHV miRNAs, miR-K12-1, 3 and 4-3p, were responsible for the targeting of Casp3. Specific inhibition of these miRNAs in KSHV-infected cells resulted in increased expression levels of endogenous Casp3 and enhanced apoptosis. Altogether, our results suggest that KSHV miRNAs directly participate in the previously reported inhibition of apoptosis by the virus, and are thus likely to play a role in KSHV-induced oncogenesis.
| MiRNAs are small, non-coding RNAs that regulate gene expression post-transcriptionally via binding to complementary sites in target mRNAs. This evolutionary conserved regulatory system is present in most eukaryotes, and it has recently been shown that certain viruses have evolved to express their own miRNAs. Due to their non-immunogenic nature, viral miRNAs represent an efficient tool for the virus to control its environment. Here we show that KSHV miRNAs are involved in the control of apoptosis both when expressed in stable cell lines and in the context of viral infection. Using a microarray based approach we identified putative cellular targets, among which the effector caspase 3 is targeted by three of the viral miRNAs. Finally, we showed that blocking these miRNAs in infected cells resulted both in increased Casp3 levels and a higher apoptosis rate. These findings indicate that miRNAs of viral origin are key players in cell death inhibition by KSHV.
| The development of cancer is linked to six major hallmarks that explain how cells transgress from a normal to a neoplastic state, including (i) sustained proliferative signaling, (ii) evasion of growth suppression, (iii) activated invasion and metastasis, (iv) enabled replicative immortality, (v) induced angiogenesis and (vi) resistance to cell death [1]. There is ample evidence that programmed cell death or apoptosis functions as a barrier to cancer development (reviewed in [2]). Many different factors, including environmental ones, contribute to the origin and progression of cancer. For example, infection by microbial pathogens sometimes leads to tumor development. Several viruses have been recognized as causal agents of specific types of cancer, and up to 20% of all human cancers are associated with single or multiple viral infections. One such oncogenic virus is Kaposi's sarcoma-associated herpesvirus (KSHV), the primary etiological agent of Kaposi's sarcoma, which is a highly angiogenic tumor most probably arising from the endothelium and developing primarily in immunocompromised individuals. KSHV-infection is also associated with aggressive lymphomas such as primary effusion lymphoma and multicentric Castleman's disease [3]. Like many viruses, KSHV has been shown to inhibit apoptosis, and possesses a truly impressive arsenal to do so (reviewed in [4], [5]).
Viruses have acquired an extraordinary capacity to evolve and adapt to their host, which translates into an incessant battle between the infected organism and the virus. One of the latest discoveries reflecting this continuous arms race is that certain mammalian viruses encode for miRNAs. In mammals, miRNAs constitute one of the most important classes of regulatory RNAs [6], [7]. Their biogenesis involves the processing of a large primary transcript into a stem-loop pre-miRNA, ultimately leading to the mature single stranded ∼22 nt miRNA (reviewed in [7]–[11]). This functional miRNA is incorporated into an RNA-induced silencing complex (RISC) that invariably contains a member of the Argonaute protein family. Once loaded, the active RISC can be directed towards its messenger RNA target to regulate, predominantly negatively, its translation (see references [12], [13] for review). The fact that target RNAs are frequently destabilized justifies the use of large-scale approaches to look at global changes in transcriptomic profiles as a way to identify miRNA targets [14]. To date, the vast majority of reported miRNA/mRNA interactions involve binding of the miRNA to the 3′ untranslated region (UTR) of the transcript through an imperfect base-pairing mechanism in which nucleotides 2 to 8 of the miRNA (the seed) appear to play an important role [15]. However, other types of interactions, such as binding in the coding sequence or in the 5′ UTR, or with bulges in the seed region, have also been reported [16]–[18].
The use of small non-coding RNAs such as miRNAs to regulate gene expression makes perfect sense for viruses, allowing them to modulate the cellular environment in a non-immunogenic manner [19]. The first virus-encoded miRNAs were identified in Epstein-Barr virus [20], and subsequent studies concluded that many herpesviruses, including Kaposi's sarcoma herpesvirus (KSHV) encode miRNAs (reviewed in [21]). KSHV has been shown to encode 12 miRNAs [22]–[25], which are clustered in the vicinity of the major KSHV latency transcript, K12. KSHV-miR-K12-1 to miR-K12-9, and miR-K12-11 are located in the intron of the larger kaposin transcript, while miR-K12-10 maps to the coding region, and miR-K12-12 resides within the 3′ UTR of the K12 coding sequence. Some cellular targets of KSHV miRNAs have been identified, mostly for miR-K12-11, which shares an identical seed sequence with the cellular miRNA miR-155 [26], [27].
Here, we show that KSHV miRNAs also contribute to the inhibition of apoptosis in infected cells. We show that cell lines expressing KSHV miRNAs are less sensitive to both caspase-dependent and -independent apoptosis induction by staurosporine or etoposide. Using a microarray approach, we identified caspase 3 (Casp3) as a target of some of these viral miRNAs. Casp3 is a well-known effector caspase (reviewed in [28]) that is critical for apoptosis induction. Using site-directed mutagenesis, we found that KSHV miR-K12-1, K12-3 and K12-4-3p are responsible for Casp3 regulation. Finally, by blocking the function of these miRNAs in infected cells, we showed that both Casp3 levels and apoptosis were increased.
We generated inducible HEK293 cells (FLP-293) expressing the intronic KSHV miRNAs under a doxycycline-inducible CMV promoter. To this end, the sequence spanning the ten intronic miRNAs miR-K12-1 to 9 and miR-K12-11 (K10/12) (Figure 1A) was inserted into the pcDNA5/FRT/TO plasmid, and used to transfect Flp-In T-Rex-293 cells. Stable cell lines were obtained by hygromycin selection, and subsequently named FLP-K10/12. As a negative control, we generated stable cells transfected with a pcDNA5 plasmid with no insert, that we then named FLP-pcDNA. We verified by northern blot analysis that doxycycline treatment readily induced the expression of the miRNAs to level similar to that found in the KSHV-infected BCBL-1 cells [29] (Figure 1B). In all following experiments, we used a final concentration of doxycycline of 1 µg/mL. We also measured by northern blot analysis the level of KSHV miRNAs expression in the induced FLP-K10/12 cells and compared it to KSHV-infected BCBL-1 and BC3 cells [30]. We found that expression of the miRNAs was slightly higher than in BCBL-1, but lower than in BC-3 cells (Figure S1), suggesting expression close to physiological levels. To assess the effect of KSHV miRNA expression on apoptosis, we first grew the FLP-pcDNA and -K10/12 cell lines in the presence of doxycycline to induce expression of the viral miRNAs, and then treated them for 8 h with 2 µM of staurosporine, a well-described inducer of apoptosis [31], or DMSO as a control. To measure the effect of this treatment on apoptosis we used Annexin V binding assay, which allows quantification of the level of phosphatidylserine exposure at the outer membrane side, a well characterized event of early apoptosis [32]. In addition, cells were labeled with propidium iodide (PI), staining both apoptotic and necrotic cells. Statistical analysis of six independent cell-sorting experiments revealed that Annexin V binding levels following staurosporine treatment were not significantly different in the presence or absence of doxycycline for the control FLP cell line (Figure 1C). In contrast, concerning the FLP-K10/12 cell line, a statistically significant decrease in Annexin V levels after staurosporine treatment was observed following doxycycline-induced expression of the microRNAs (Figure 1C). Figure 1D shows one representative experiment of the six biological replicates. In order to get an independent measure of apoptosis, we monitored the activity of effector caspases using a DEVD-aminoluciferin substrate for Casp3 and Casp7 that is measurable by a luciferase assay. As shown in Figure 1E, the luminescent Casp3/7 activity induced by 2 or 5 µM of staurosporine treatment of the stable FLP-K10/12 cells was sharply decreased (2.5 to 3 times) upon doxycycline induction of the KSHV miRNAs expression, while it remained unchanged in the control FLP-pcDNA cells.
To monitor the effect of KSHV miRNAs on apoptosis in a cell line more physiologically relevant for KSHV infection, we used the previously described DG-75-K10/12 cells -a Burkitt lymphoma cell line [33], [34] lentivirally transduced with a construct expressing KSHV intronic miRNAs [35]-, and measured the effect of KSHV miRNAs expression on apoptosis in either the DG-75-K10/12 cells or the DG-75-EGFP control cells. Statistical analysis of four independent experiments confirmed that staurosporine treatment readily induced phosphatidylserine exposure in the control cell line, but that this induction was significantly reduced in the K10/12-expressing cells (Figure 2A). A representative experiment of the biological replicates (Figure 2B) shows that the percentage of Annexin V positive cells dropped almost two-fold in DG-75-K10/12 cells vs. DG-75-EGFP cells after 8 h of staurosporine treatment. As opposed to the FLP-293 cells, we were unable to induce Casp3/7 activity with staurosporine in the DG75 cells (data not shown). In line with this observation, it has been reported previously that in this particular cell line, the apoptotic protease-activating factor 1 (APAF-1) was sequestered at the plasma membrane, which prevents caspase activation [36].
In order to identify putative cellular targets of KSHV miRNAs involved in the miRNA-induced anti-apoptotic phenotype, we used a microarray approach on the two main cell types that are infected in vivo by KSHV: endothelial cells and B lymphocytes. In addition to the already described DG-75-K10/12 and DG-75-EGFP cells, we also generated by lentiviral transduction endothelial cells EA.hy926 [37] expressing the K10/12 construct or EGFP as a control. In order to determine the relative expression of KSHV miRNAs were expressed in the DG-75-K10/12 cells, we cloned the small RNA population of these cells and analyzed it by Solexa deep-sequencing. As can be seen in Table S1, KSHV miRNAs represented more than 18% of the total miRNAs in this cell line, which is slightly less than what has been previously described for BCBL1 cells ([38] and data not shown). All intronic miRNAs accumulated to measurable levels with the exception of miR-K12-11, which seemed to be expressed at a low level. We then measured by qRT-PCR the levels of some KSHV miRNAs expression in EA.hy926 and DG-75-K10/12 cells compared to BCBL-1 cells (Table S2). The levels of viral miRNAs expression in both cell lines correlated very well (r = 0,93) (Figure S2).
DG-75 and EA.hy926 EGFP control- and miRNA- expressing cell lines were analyzed in triplicate on Affymetrix Human Genome U133 Plus 2.0 microarrays. The clustering of the gene expression profiles primarily correlated with the cell line (DG-75 vs. EA.hy926), but also within each cell line with the expression of KSHV miRNAs (Figure S3). In addition, the changes in gene expression levels following the KSHV miRNAs transduction correlated weakly (r = 0.19) but significantly (p<10−15 at Pearson's test) between the two cell lines (Figure S4).
Target recognition by miRNAs involves a number of determinants, the most important of which appears to be perfect base-pairing of nucleotides 2–7 of the miRNA (the seed), together with either an adenosine opposite miRNA nucleotide 1, or an additional base pair involving the 8th nucleotide of the miRNA [15]. In single miRNA transfection experiments one typically observes that the mRNAs that carry matches to the transfected miRNA are significantly down-regulated in response to transfection compared to mRNAs that do not carry such matches. To determine whether the KSHV miRNAs significantly influenced gene expression levels in a complex experiment such as ours, in which multiple miRNAs are simultaneously induced, we designed the following test. We first computed a KSHV miRNA sensitivity score for each mRNA, defined as the sum over all KSHV miRNAs, the number of matches of the 3′ UTR to the seed of the KSHV miRNA multiplied by the relative abundance of the KSHV miRNA. The relative abundances of the KSHV miRNAs were determined using the DG-75-K10/12 small RNAs deep-sequencing data. The KSHV miRNA sensitivity scores are reported in Dataset S1. We then compared the change in expression level of the 1000 mRNAs with highest KSHV miRNA sensitivity score and of mRNAs with no seed matches to the KSHV miRNAs in the 3′ UTR and found that the KSHV miRNA sensitive mRNAs were significantly down-regulated in both KSHV miRNA expressing DG-75 and EA.hy926 cells (p<10−3 and p<10−15, respectively in Wilcoxon's rank sum test). We observed however, that the 3′ UTRs of the 1000 mRNAs with highest KSHV sensitivity were on average ten times longer than the 3′ UTRs with no seed matches (Figure S5). To test whether differences in 3′ UTR length alone could account for the down-regulation of the KSHV sensitive mRNAs, we computed the average fold change of 1000 mRNAs sampled in such way that their 3′ UTR length distribution was the same as that of the KSHV sensitive mRNAs (Figure 3A, blue bars). We repeated this procedure 1000 times and found that the set of 1000 KSHV sensitive mRNAs still exhibited a stronger down-regulation compared to mRNAs of similar 3′ UTR length (Figure 3A, red bars) (p = 0.036 and 0.002, respectively for the expression changes computed from the DG-75 and EA.hy926 samples). Therefore, the 3′ UTR length alone cannot explain the magnitude of down-regulation of the most KSHV sensitive mRNAs in response to KSHV miRNA expression. These results indicated that KSHV miRNAs exert a detectable effect on mRNA expression in these cell lines and motivated us to proceed with further characterization of candidate direct targets.
As KSHV putative direct targets we extracted transcripts that were significantly down-regulated significantly in the replicate experiments, and which contained at least one seed-match to one of the KSHV miRNAs. We identified 704 putative direct targets in DG-75 cells (Figure 3B), and 980 putative direct targets in EA.hy926 cells (Figure 3C). A complete list of putative direct targets can be found in Dataset S2 for DG-75 cells and in Dataset S3 for EA.hy926 cells. The overlap between the two datasets contained 153 putative direct targets (Dataset S4).
In order to validate direct cellular targets of KSHV miRNAs, we turned to classical reporter assays in HEK293 cells (293A cells). We chose, among genes involved in pathways such as cell cycle regulation, DNA damage repair, and apoptosis, a subset of the 3′ UTR sequences identified as putative direct targets by our previous analysis. These candidates were then cloned 3′ to the firefly luciferase gene in the dual-reporter vector psiCHECK-2, also encoding a Renilla luciferase as a standard. We cloned and tested the full length 3′ UTR of sixteen candidate targets, which were tested in multiple independent assays. We first assessed that the K10/12 construct could repress the activity of luciferase sensors containing bulged complementary sequence (with a bulge at positions 9 to 12) to some of the KSHV miRNAs. For all of the KSHV miRNAs tested, except miR-K12-9, we could show a strong repression in the presence of pcDNA-K10/12 (Figure 4A). The lack of miR-K12-9 activity could relate to its lower expression in the context of the K10/12 construct (Tables S1 and S2). As opposed to what would have been expected based on the DG-75-K10/12 small RNA sequencing data, miR-K12-11 appeared to be functional in the FLP-K10/12, and we confirmed that it accumulated in higher amounts in these cells compared to the DG-75-K10/12 cells (data not shown). As a positive control for the luciferase assays with the selected putative targets, we used SPP1, a previously validated target of KSHV miRNAs [39]. The validation assays showed that only a subset of the 3′ UTRs tested resulted in a measurable repression of luciferase activity (Figure 4B). Among all the tested candidates, we observed the most important and reproducible down-regulation for two genes, Rad51AP1, involved in DNA damage repair, and Casp3, one of the main effectors involved in apoptosis induction. The RAD51AP1 reporter showed a down-regulation of 30 to 40% across luciferase experiments, while the Casp3 reporter showed a down-regulation of 40 to 50% (Figure 4B). We thus hypothesized that the anti-apoptotic phenotype of KSHV miRNA-expressing cells could be in part caused by the regulation of Casp3, and decided to continue this study by focusing on this protein.
The initial analysis of Casp3 3′ UTR revealed 8mer or 7mer seed-matches [15] for miR-K12-4-3p (one M8A1 site), miR-K12-1 (two M8 sites), and miR-K12-3 (one A1 site). In addition, 6mer seed-matches to miR-K12-1, miR-K12-2 and miR-K12-10a could be found (Figure 5A). In order to further identify regions of the Casp3 3′ UTR that were susceptible to regulation by KSHV miRNAs, we subdivided the 3′ UTR in three parts and cloned them in the reporter vector. None of the tested fragments showed such a strong repression as the full-length sequence, suggesting that all putative miRNA binding sites are required for efficient repression, or that the binding sites function optimally only in their natural context (Figure S6). We then transfected pcDNA constructs expressing individual miRNAs (miR-K12-1 to -6, K12-9 and K12-10) to identify whether a single, or multiple miRNAs, mediated Casp3 regulation. We found that as suggested by the seed-matches quality, miR-K12-1, K12-3 and K12-4-3p (in decreasing order of repression observed) were able to significantly regulate the expression of the reporter fused to the 3′ UTR of Casp3 (Figure 5B). Expression of miR-K12-2 and K12-10, or of miRNAs with no predicted seed-matches (miR-K12-5, K12-6 and K12-9) had no effect on the Casp3 sensor.
Subsequently, we aimed at determining which of the five putative binding sites for miR-K12-1, K12-3 and K12-4-3p were most important for Casp3 downregulation. To this end, we mutagenized each individual seed-match by introducing three point mutations to disrupt miRNA binding in the luciferase sensor containing Casp3 3′ UTR (Figure 6A). The resulting luciferase reporters were tested with miRNA expression constructs for either the 10 intronic miRNAs, or the individual miR-K12-1, K12-3 and K12-4-3p. As shown in Figure 6B, only the 3′ proximal binding site for miR-K12-1 appears to be functional, as the Casp3 Mut K12-1 3′ luciferase sensor could not be regulated by the pcDNA-K10/12 or the pcDNA-K12-1 constructs. The binding site for miR-K12-3 was also validated, as the mutant luciferase sensor for this miRNA is not regulated by the pcDNA-K10/12 or the pcDNA-K12-3 construct (Figure 6C). Finally, the binding site for miR-K12-4-3p was validated, although it seems to be less potent than the two others in terms of luciferase regulation (Figure 6D). In conclusion, we showed that Casp3 3′ UTR is regulated via three binding sites for (from 5′ to 3′) miR-K12-4-3p, K12-3 and K12-1. The positions of these sites explain why the luciferase assay done with the Casp3 3′ UTR fragments (Figure S6) did not reveal obvious differences as each individual fragment contained one of the three validated sites.
In order to measure the effect of KSHV miRNAs on endogenous Casp3, we first performed real-time quantitative PCR analysis of 293A cells following primary infection and antibiotic selection of rKSHV infected cells [40]. We found that the level of Casp3 transcript decreased two fold following infection (Figure 7A, left panel). We also measured the level of Casp3 mRNA in the doxycycline-inducible FLP cells, and observed a similar down-regulation upon induction in FLP-K10/12 cells, but not in control FLP-pcDNA cells (Figure 7A, right panel). We then measured Casp3 protein levels in FLP-K10/12 and DG-75-K10/12 cells, and observed a significant down-regulation in three independent experiments (average of 0.63-fold, p = 0.0005 and 0.69-fold, p = 0.0046 respectively) (Figure 7B and C). We then turned to HUVEC endothelial cells, one of the two main cellular types infected in vivo by KSHV, and performed western blot analysis of primary or E6/E7 HUVEC cells stably transduced with either the EGFP, or the K10/12 lentiviral construct. In four independent experiments, the level of Casp3 protein was significantly down-regulated in K10/12 cells compared to the control EGFP cells (average of 0.61-fold, p = 0.0007) (Figure 7D).
In order to assess whether the down-regulation of Casp3 in naturally KSHV infected cells was caused by the specific presence of the three previously identified miRNAs, we used an antisense approach to inhibit specifically miR-K12-1, K12-3 and K12-4-3p. We thus employed either classical full-length 2′-O-methylated (2′OMe) antisense oligoribonucleotides [41], or short Locked Nucleic Acid oligonucleotides directed only against the seed of each individual miRNAs (tiny LNAs) [42]. In three independent experiments, transfection of a cocktail of 2′OMe oligonucleotides against miR-K12-1, K12-3 and K12-4-3p (2′OMe-miR-K12-1/3/4) in BC-3 cells resulted in a modest but measurable increase of Casp3 protein level compared to a control 2′OMe oligonucleotide (2′OMe-miR-67) (1.4-fold on average, p = 0.0486) (Figure 7E). The advantage of using tiny LNAs to inhibit miRNA function over the 2′OMe oligonucleotides is based on the fact that they do not require transfection to enter the cells. We therefore tested the inhibition efficiency of tiny LNAs on luciferase sensors in HEK293 cells and found that they could readily revert the targeted miRNA regulation (Figure S7). BC-3 cells grown in a medium containing a cocktail of tiny LNAs each directed against one of the three KSHV miRNAs listed above (LNA-miR-K12-1/3/4) also showed an 1.8-fold increase in Casp3 mRNA (Figure 7F) accompanied with a somewhat milder increase in the protein levels compared to control tiny LNA (LNA-miR-67) (1.3-fold on average, p = 0.0018) (Figure 7G; left panel for 48 h, and right panel for 6 days). Taking these results altogether, we can definitely conclude that Casp3 is regulated at both mRNA and protein levels by the KSHV-encoded miR-K12-1, K12-3, and K12-4-3p.
In order to test the biological relevance of the repression of Casp3 by these KSHV-encoded miRNAs, we decided to look at Casp3 cleavage or its direct and indirect endogenous cleavage substrates, such as respectively Poly[ADP-ribose] polymerase-1 (PARP-1) or genomic DNA. We thus treated BC-3 cells with a cocktail of tiny LNAs directed against the three Casp3-targeting viral miRNAs, and measured PARP-1 cleavage following staurosporine treatment for 8 h. In the absence of staurosporine, inhibition of KSHV miRNAs had no or little effect on PARP-1 levels (Figure S8, left panel). Upon treatment, we found that cells pre-treated with anti-KSHV specific tiny LNAs (LNA-miR-K12-1/3/4), but not with the control tiny LNA (LNA-miR-67), accumulated slightly more of the PARP-1 cleavage product (Figure S8, right panel). We also tested the effect of this inhibition using KSHV-infected immortalized lymphatic endothelial cells (iLECs) by measuring the appearance of cleaved Casp3 and the extent of apoptosis-induced genomic DNA nicks following a 24 h etoposide treatment. iLECs represent one of the most relevant cell types implicated in KSHV pathogenesis [43]. We observed an increase in the number of cleaved Casp3 positive cells (Figure 8A) and TdT-mediated dUTP nick end labeling (TUNEL) positive cells (Figure 8B) over mock-treated (DMSO) controls when miR-K12-1, K12-3 and K12-4-3p were inhibited with the tiny LNA cocktail (LNA-miR-K12-1/3/4), over the control. In three independent experiments, the mean fold induction of etoposide-induced TUNEL positive cells (over the DMSO treated control) following inhibition of miRNAs (LNA-miR-K12-1/3/4) was significantly greater (2.30-fold, p = 0.041) than in cells treated with the control tiny LNA (Figure 8C). These data suggests that the KSHV-encoded miR-K12-1, K12-3 and K12-4-3p contribute to protection of etoposide-induced apoptosis in KSHV infected iLECs.
Viral miRNAs have only recently attracted attention in studies into viral genetics, and their importance during the course of infection remains to be fully established. Almost all of these miRNAs were found in viruses belonging to the herpesvirus family; viruses that are associated with latency and that suggest long-term disease progression. Like other members of the gammaherpesvirus subfamily, KSHV is associated with a number of neoplastic disorders including Kaposi's sarcoma and B-cell lymphomas [3]. Some cellular targets of KSHV miRNAs have been previously reported. For example, miR-K12-11 has been shown to target a subset of genes that are also targeted by its homologous human miRNA, miR-155, that shares an identical seed region with this miRNA [26], [27]. Among the validated targets of miR-K12-11 are two transcription factors, BACH1 and Fos. Although Fos itself provides a potential link between KSHV infection and oncogenesis, the authors did not show that KSHV miRNAs directly participate in cancer progression. The study of Samols and colleagues identified another potential candidate as a KSHV miRNA target that could contribute to cell transformation [39]. Indeed, with a microarray-based approach similar to the one that was used in this study, they found that thrombospondin (THBS1), a gene involved in angiogenesis, is regulated by KSHV miRNA expression. However, the analysis was performed in HEK293 cells, which are not representing cells naturally infected by KSHV. More recently, the Ganem laboratory also reported on the identification of cellular targets of KSHV miRNAs using a transcriptomic-based approach, with the Bcl2-associated factor BCLAF1 as one of the identified targets of several KSHV miRNAs [44]. Other targets of KSHV miRNAs that have been identified very recently are p21, IκBα, TWEAKR and Gemin 8 [35], [45]–[47].
The aim of this study was to define the role played by KSHV miRNAs in apoptosis inhibition. The apoptotic processes can be executed intracellularly by the release of various factors (e.g. cytochrome c or SMAC/DIABLO) from mitochondria, or extracellularly through transmembrane death receptors, which are activated by their ligands. In both the intrinsic and extrinsic pathways, caspases are recruited and activated, and in turn they cleave substrates leading to the execution of apoptosis. In the intrinsic pathway, cytochrome c leaks from mitochondria [48], and binds to the adaptor apoptotic protease activating factor-1 (APAF1) to form the multi-protein structure, coined the apoptosome. The latter recruits Casp9, which in turn activates downstream effector caspases 3, 6 and 7 [49]. In the extrinsic pathway, ligands such as TRAIL and FasL activate specific pro-apoptotic death receptors at the cell surface [50]–[52], which results in the binding of the intracellular domains of the receptors to the adaptor protein Fas-associated death domain [53]. This leads to the assembly of the death-inducing signaling complex DISC, and to the recruitment of initiator caspases 8 and 10 [54]. Upon stimulation of these two caspases, effector caspases 3, 6 and 7 are activated. Thus, the intrinsic and extrinsic pathways converge at the level of the effector caspases, which highlights Casp3 as a critical factor in the control of apoptosis. In this study, we observed that KSHV miRNAs have a negative effect on apoptosis, as HEK293 cells and DG-75 B lymphocytes expressing these viral miRNAs are partially protected from apoptosis induction by staurosporine. We also measured Casp3 activity in the HEK293 cells, and showed that the presence of KSHV miRNAs resulted in a sharp decrease of Casp3/7 activity upon staurosporine induction. While our data does not rule out that the observed effect in HEK293 cells is due to a decreased activity of Casp7, the evidence available to date indicates that Casp3 activity is predominant over Casp7 activity, and that Casp3 is likely the major executor of apoptosis [55]. However, we were unable to monitor the effect of KSHV miRNA on Casp3/7 activity in DG-75 cells. Indeed, these cells are resistant to caspase activation by the intrinsic pathway [36], and accordingly, we could not induce Casp3/7 cleavage with staurosporine. This result confirms that Annexin V levels do not only measure caspase-dependent apoptosis, and therefore indicates that KSHV miRNAs are regulating both caspase-dependent and -independent apoptosis.
To discover cellular targets of KSHV miRNAs, we used a microarray-based approach to identify transcripts regulated by KSHV miRNAs in both the B lymphocyte DG-75 cell line and the endothelial EA.hy926 cell line. Based on their expression profiles, the samples primarily clustered according to the cell line (DG-75 or EA.hy926), and, within these two clusters, according to the presence of KSHV miRNAs. Using small RNA deep-sequencing data, we determined the relative abundance of each miRNA within the expressed cluster, which enabled us to show that transcripts containing seed-matches to KSHV miRNAs within their 3′UTR were significantly more down-regulated that transcripts without such binding sites. This enabled us to generate a list of putative targets to follow in further functional assays. Our validation rate was relatively low, reflecting presumably the fact that many miRNAs (virus-encoded and endogenous) changed in these experiments, leading to complex secondary effects. We looked for seed-match sites within the coding sequences of down-regulated transcripts and could identify a few (listed in Dataset S1), but the validation of these sites can prove challenging. Nevertheless, we validated two candidate targets that are biologically relevant for KSHV infection, Rad51AP1 and Casp3. Rad51AP1 is a DNA binding protein that participates in RAD51-mediated homologous recombination, and is important for the preservation of genome integrity [56]. Because KSHV has been shown to induce DNA damage response through the expression of v-cyclin [57], the down-regulation of Rad51AP1, which will require further validation, might be important in the context of viral infection. In light of our initial aim to define the role of KSHV miRNAs in apoptosis inhibition, we focused our efforts on the characterization of Casp3 as a target of KSHV miRNAs. We confirmed that a Casp3 3′ UTR luciferase reporter construct is regulated by three KSHV miRNAs, and we identified three miRNAs, miR-K12-1, miR-K12-3 and miR-K12-4-3p, as being responsible for this regulation, as well as their binding sites within Casp3 3′ UTR.
We then showed that endogenous Casp3 was also regulated by KSHV miRNAs, both at the mRNA and protein levels, and in different cell types. We also showed that inhibition of miR-K12-1, K12-3 and K12-4-3p in KSHV-infected cells resulted in an upregulation of Casp3 expression, which in turn translated into an increase in apoptosis, as assessed by cleaved Casp3 quantification and TUNEL assay analysis. These findings are consistent with a report that described the role of KSHV in conferring a survival advantage to endothelial cells [58]. In this report, Wang et al. showed that the level of Casp3 activity was decreased in KSHV-infected HUVEC cells subjected to staurosporine treatment (or other apoptotic insults). The regulation of Casp3 is not the only explanation for KSHV miRNAs-mediated inhibition of apoptosis, especially since we showed that caspase-independent apoptosis was also affected. It is of course highly probable that other factors in the apoptosis pathway are also targeted by KSHV miRNAs. For example, Abend et al. recently reported that KSHV miR-K12-10 targeted the TNF-like weak inducer of apoptosis (TWEAK) receptor [47], which indicates another level of regulation of one certain type of apoptosis.
In summary, our findings demonstrate that KSHV miR-K12-1, K12-3 and K12-4-3p target the effector caspase 3. The down-regulation of Casp3 by KSHV miRNAs results in a decrease in apoptosis activity in different cell types including endothelial cells that are biologically relevant for KSHV infection in vivo. The specific inhibition of these miRNAs in infected cells increased Casp3 levels and cell death. Apoptosis is frequently inhibited in tumor cells, and our results are in agreement with a recent report that indicates that the active form of Casp3 is detected less frequently in Kaposi sarcoma lesions in patients from Brazil [59]. Our data therefore suggests that apoptosis regulation by the viral miRNAs could contribute to the malignant phenotype triggered by KSHV infection. In the long term, delivery of specific inhibitors of these viral miRNAs in KSHV-infected patients to restore apoptotic clearance of the virus by the immune system could be an interesting novel therapeutic approach.
DG-75 and BCBL-1 cells (obtained through the NIH AIDS Research and Reference Reagent Program (Cat# 3233 from McGrath and Ganem)) were grown in RPMI 1640 medium containing 10% fetal calf serum (FCS), 100 UI/mL penicillin, 100 µg/mL streptomycin and 2 mM L-Glutamine. BC-3 cells (ATCC) were grown in the same media with 50 µM ß-Mercaptoethanol. EA.hy926, QBI-HEK 293A (QBiogene), Flp-In T-REx-293 (Invitrogen), and HEK293 cell lines were grown in DMEM supplemented with 10% FCS and penicillin/streptomycin. Primary and E6/E7 HUVEC cells (from Promocell) were cultured in a humidified 5% CO2 atmosphere at 37°C in endothelial basal medium (Promocell) supplemented with 10% FCS, gentamicin, amphotericin and supplement kit provided with the media. To obtain immortal lymphatic endothelial cells (iLECs) primary human LEC cells (Promocell) were immortalized by the HPV oncogenes E6/E7 as previously described (Moses et al., 1999). iLEC cells were maintained in endothelial basal medium (Promocell) supplemented with 5% human AB serum (HS; Sigma, St. Louis, Mo.).
Wildtype KSHV was produced from BCBL-1 cells induced with 20 ng/mL PMA. The virus-containing supernatant was collected after three days by ultracentrifugation (21,000 rpm at 4°C for 2 h), and resuspended in TNE buffer (150 mM NaCl, 10 mM Tris pH 8, 2 mM EDTA, pH 8). For the KSHV infection iLEC cells were plated in 6-well plates one day before the infection using multiplicity of infection (MOI) 1 in the presence of 8 µg/mL polybrene (Sigma). The infection was performed as spin-infection by centrifugation at 2500 rpm (Heraeus Multifuge 3 S-R; Thermo Scientific) for 30 min at room temperature. Cells were then returned to 37°C, 5% CO2, and after 4 h of incubation fresh complete media was added. The virus-containing medium was removed the next day, and replaced with fresh complete media. The extent of KSHV infection was monitored by expression of the latent nuclear antigen-1 LANA-1 in the nuclei of KSHV-infected cells (K-iLECs) and detected by immunofluorescence using anti-LANA antibody (13-210-100, Advanced Biotechnologies Inc).
rKSHV.219 infected HEK293 cells were reactivated by incubating them in DMEM medium containing 1 mM sodium butyrate and 20 ng/mL TPA (tetradecanoyl phorbol acetate) for 24 h, and four more days with media containing sodium butyrate only. The supernatant was collected, filtrated through 0.45 µM filter, and 8 µg/mL polybrene was added before adding the supernatant to QBI-HEK 293A cells seeded one day before. After 4 h, the medium was replaced and the cells grown at 37°C for 2 days. As soon as green fluorescent started to appear, 1 µg/mL puromycin was added to the medium. Cells were harvested for RNA analysis after at least 21 days under puromycin selection.
Cell lines stably expressing the ten intronic KSHV miRNAs were generated using the “Virapower” lentiviral transduction system with the vector pLENTI6/V5 (Invitrogen) and Gateway cloning. The miRNA encoding intronic region was amplified by a two-step PCR using cDNA prepared from KSHV infected BCBL-1 cells (PCR primers: KSHV miRK_for and KSHV miRK_rev for the first PCR and attB1_external for and attB2_external rev for the second PCR), cloned into pDONR207 and transferred to pLENTI6/V5-DEST (Invitrogen). PCR primers are provided in Table S3. The control lentiviral vector pLENTI6/V5-EGFP was a kind gift from Oliver Rossmann. In order to generate lentiviruses for transduction of cells with KSHV miRNAs, the ViraPower Lentiviral Gateway Expression System (Invitrogen) was employed according to the manufacturer's instructions. The packaging mix contained plasmids pLP1, pLP2 and pLP/VSVG. Virus-containing medium was cleared with a 0.45 µm filter and added with polybrene (8 µg/mL) to DG-75 (1×106 cells/mL) or EA.hy926 (3×105 cells/mL) target cells for transduction. In the case of EA.hy926 cells, the plates were centrifugated 30 min at 2500 rpm to increase transduction efficiency. Two days after transduction, when the EGFP signal in the control cells became visible, Blasticidin (1 µg/mL) was added to the medium to select for the transgene and gradually raised to a final concentration of 7.5 µg/mL for DG-75 cells and 3 µg/mL for EA.hy926 cells after six days. Efficiency of selection was determined by analyzing the proportion of EGFP expressing control cells by fluorescence activated cell sorting (FACS). Cell lines were used for experiments when 100% of control cells expressed EGFP.
Primary HUVEC cells were transduced with lentiviruses (pLenti6-vector; Invitrogen) encoding EGFP or 10/12 KSHV miRNA cluster (K10/12) and maintained under blasticidin selection (5 µg/mL) in endothelial basal medium supplemented as above. The cells were replenished with fresh medium every second day and passaged when necessary.
The Flip-In stable cell lines were generated using the Flp-In T-REx-293 cell line (Invitrogen) and according to the manufacturer's instructions (Invitrogen). Briefly, cells were seeded one day before at 106 cells/well in 6-well plates. Cells were transfected with 3.6 µg and 0.4 µg respectively of pOG44 (Invitrogen) and each pcDNA for each cell line with lipofectamine 2000 (Invitrogen). The media was replaced 24 h after transfection, and cells were passaged into 10 cm dishes 24 h later to achieve a desired confluency of maximum 25% prior selection. Hygromycin (Invivogen) was added at a concentration of 200 µg/mL and then raised 2–3 days later at a concentration of 250 µg/mL. The media was replaced each 3–4 days until 2–3 mm wide foci appeared. Cells were then passaged into 75 cm2 flasks for amplification. Efficiency of the selection was then assayed by β-galactosidase staining, for the loss of β-galactosidase activity, and/or by northern blot for the detection of the miRNA.
The microarray data were submitted to the gene expression omnibus database (http://www.ncbi.nlm.nih.gov/geo) under the accession number GSE18946. We imported the CEL files into the R software (R Development Core Team (2008). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0, http://www.R-project.org) using the BioConductor affy package [57]. The probe intensities were corrected for optical noise, adjusted for non-specific binding and quantile normalized with the gcRMA algorithm [58].
Per gene log2 fold change was obtained through the following procedure. We first fitted a lowess model of the probe log2 fold change using the probe AU content. We used this model to correct for the technical bias of AU content on probe-level log2 fold change reported by [59]. Subsequently, probe set-level log2 fold changes were defined as the median probe-level log2 fold change. Probe sets with more than half of the probes (6) mapping ambiguously (more than 1 locus) to the genome were discarded, as were probe-sets that mapped to multiple genes. We then collected all remaining probe sets matching a given gene, and averaged their log2 fold changes to obtain an expression change per gene. For sequence analyses, we selected for each gene the RefSeq transcript with median 3′ UTR length corresponding to that gene.
Controls and transductions were performed in triplicates in both cell lines (DG-75, EA.hy926), and we used limma [60] to compute differential regulation p-values. Finally, for each cell line, we only analyzed genes which had at least one probe set that was called present in either all replicates of miRNA transduction or all replicates of the control (or both).
To generate the pcDNA-K10/12, the KSHV intronic miRNA cluster was PCR-amplified from BAC36 DNA [60] and ligated into the Bam HI and Xho I sites of the pcDNA5/FRT/TO (Invitrogen). The primer sequences were (sense and antisense primers are indicated in respective order): 5′-ATATGGATCCGAATGCGTGCTTCTGTTTGA, 5′-ATATCTCGAGTTTACCGAAACCACCCAGAG. The empty pcDNA vector was obtained by digesting the pcDNA-K10/12 with Pme I, followed by ligation of the plasmid. For KSHV miRNA individual expression vectors, a region of approximately 300 nt surrounding each pre-miRNA (or the miRNA cluster) was PCR-amplified from BAC36 DNA. attB1/2 sequences were added by nested PCR and the resulting PCR product were cloned into pDONR207 (Invitrogen) and then recombined in pLenti6/V5-DEST using Gateway technology (Invitrogen). The attB1/2 primer sequences are (sense and antisense primers are indicated in respective order): 5′-ACAAGTTTGTACAAAAAAGCAGGCT, 5′-ACCACTTTGTACAAGAAAGCTGGGT. The specific primers are indicated in Table S3. The individual miRNA expression cassettes were then subcloned via PCR amplification from pLenti6/V5-DEST expressing vectors and ligated into the Xho I and Apa I sites of pcDNA5/FRT/TO, and with the primers indicated in Table S3. To generate luciferase reporter plasmids, psiCHECK-2 (Promega) was modified by inserting the Gateway cassette C.1 (Invitrogen) at the 3′-end of the firefly luciferase gene into the Xba I site of psiCHECK-2. attB-PCR products were cloned into pDONR/Zeo (Invitrogen) and recombined in the modified psiCHECK-2 vector by Gateway cloning. The 3′ UTR sequence of the different candidates were obtained from the Ensembl database (www.ensembl.org) and were nested PCR-amplified from QBI-HEK 293A cells' genomic DNA with the primers indicated in Table S3 and attB1/2 primers. The imperfect match sensors for KSHV miRNA were obtained by annealing the oligonucleotides indicated in Table S3 and PCR-based addition of the attB sequences using attB1/2 primers. The resulting PCR product was then cloned by Gateway recombination sequentially in pDONR/Zeo and psiCHECK-2 plasmids.
Mutagenesis was performed using QuikChange Lightning Site-Directed Mutagenesis Kit (Agilent Technologies) according to the manufacturer's instructions and using the oligonucleotides indicated in Table S3. Briefly, we mutagenised in the Casp3 luciferase reporter construct the nucleotides predicted to pair to position 3 to 5 of the miRNA sequence to prevent pairing of the miRNA seed sequence on Casp3′s predicted target sites.
QBI-HEK 293A cells were seeded in 48-well plates at 105 cells/well and then incubated a few hours. When cells were adherent, co-transfection of 25 ng of the reporter constructs and 250 ng of the pcDNA-K10/12 (or pcDNA as control vector) were performed using Lipofectamine 2000 (Invitrogen). After 48 h, cells were then washed in PBS and lysed with 65 µL of passive lysis buffer (Promega), and 10 µL were assayed for firefly and Renilla luciferase activity, using the dual-luciferase reporter assay system (Promega) and a luminescence module (Glomax, Promega). The relative reporter activity was obtained by first normalizing to the transfection efficiency with the Renilla activity, and then, to the firefly activity obtained for the empty control reporter, in presence of the pcDNA-K10/12 or pcDNA, to normalize for the effect of transfection of these expression vectors.
For Western blot analysis of HUVEC cells, cells were extracted in ELB lysis buffer (150 mM NaCl; 50 mM HEPES, pH 7.4; 5 mM EDTA and 0.1% NP40) and 30 µg of proteins was separated on 12% SDS-PAGE and transferred on to nitrocellulose membranes according to standard protocols. Primary antibodies used in Western blotting were anti-caspase-3 (MAB4603; Millipore) and anti-γ-tubulin (GTU-88; Sigma-Aldrich). HRP-conjugated anti-mouse (AP308P; Chemicon) immunoglobulin was used as a secondary antibody. Filters were visualized on SuperRX film (Fuji) using the ECL chemiluminescence system (Pierce, Rockford, IL). The intensity of the chemiluminescence signals was quantified with FluoChem 880 imager and software (Alpha Innotech Corporation).
For Western blot analysis of DG-75, BC-3 or FLP-293 cells, cells were extracted in passive lysis buffer (50 mM Tris, 150 mM, NaCl, 5 mM EDTA and 0.5% NP40, 10% Glycerol and 10 µM MG132) and 15 µg or 45 µg of proteins, respectively from BC-3 or FLP-293 cells, was separated on 10% SDS-PAGE for PARP analysis, or on 15% SDS-PAGE for Casp3 analysis, and transferred on to nitrocellulose membranes according to standard protocols. Primary antibodies used in Western blotting were anti-caspase-3 (06-735; UpState), anti-PARP-1 [61] and anti-γ-tubulin (GTU-88; Sigma-Aldrich). IRDye 800CW-conjugated anti-rabbit and anti-mouse (926-32213 and 926-32212; Li-Cor Biosciences) immunoglobulins were used as secondary antibodies. The intensity of the fluorescence signals was quantified with Odyssey Infrared Imaging system and Odyssey v3.0 software (Li-Cor Biosciences).
RNA was extracted using Trizol reagent (Invitrogen) and Northern blotting was performed on 5 to 10 µg of total RNA as described before [23], [62]. Probes were 5′ 32P-radiolabelled oligodeoxynucleotides antisense to the miRNA sequence or to part of the U6 snRNA sequence. Blots were analyzed and quantified by phosphorimaging using a FLA5100 scanner (Fuji).
Small RNA cloning was conducted from 50 µg of DG-75-K10/12 total RNA as previously described [63]. Small RNA sequencing was performed at the Institut de Génétique et de Biologie Moléculaire et Cellulaire (IGBMC, Illkirch, France) using an Illumina Genome Analyzer II with a read length of 36 base pairs (bp).
An in-house Perl analysis pipeline was used to analyze the data produced by small RNA sequencing. After 3′ adaptor removal and size selection (exclusion of trimmed reads shorter than 15 nt), non-redundant sequences were mapped to the genomes from which they may derive and to other RNAs already annotated, using Nexalign (http://genome.gsc.riken.jp/osc/english/software/src/nexalign-1.3.5.tgz) permitting up to 2 mismatches. The Homo sapiens and KSHV genome sequences were respectively downloaded from the UCSC repository (assembly version hg19) and the GenBank database. The following sources of annotated transcripts were used: miRBase v.16 for miRNAs, GenBank v.180 for Homo sapiens rRNA, tRNA, sn-snoRNA, scRNA and piRNA, and Repbase v.16.01 for Homo sapiens and common ancestral repeats. By doing so, small RNAs that mapped unambiguously to sequences from one single functional category were easily classified, while the other ones were identified by applying this annotation rule based on the abundance of various types of sequences in the cell: rRNA > tRNA > sn-snoRNA > miRNA > piRNA > repeat > pathogen genome > host genome > unknown.
For inhibition of miRNAs, BC-3 cells were cultured in 6-well dishes and transfected with the 2′O-methylated oligonucleotides (provided by G. Meister) against individual KSHV miRNAs using Oligofectamine (Invitrogen). Oligonucleotides were used at a final concentration of 60 nM and transfections were performed according to manufacturer's instructions. Total proteins were extracted for analysis 48 h after transfection.
For inhibition of miRNAs with tiny LNAs, 2×106 BC-3 cells were seeded in 6-well plates and incubated with the inhibitors (Table S3) against individual KSHV miRNAs or the control C. elegans miR-67. Oligonucleotides were used at a final concentration of 1.5 µM and incubated in the medium for 48 h, or 6 days by replacing twice the medium (day 2 and 5), prior to harvesting the cells.
The effects of the KSHV miRNAs on apoptosis were analysed by both measurement of caspase 3/7 activity and Annexin V/propidium idiode (PI) staining. Cell death was induced by adding 2 to 5 µM staurosporine (Sigma) for 8 h; DMSO was used as a control. For Annexin V binding analysis, 105 HEK293 cells were seeded in 12-well plates, incubated overnight prior to addition of staurosporine or DMSO. Cells were harvested by trypsinization, washed in PBS, and resuspended in binding buffer (10 mM Hepes/NaOH (pH 7.4), 140 mM NaCl, 2.5 mM CaCl2) containing Annexin V conjugated with Allophycocyanin diluted at 1/100 (BD Biosciences, Le Pont-de-Claix France) and 2 µg/mL PI (Sigma-Aldrich, Lyon, France). The cells were incubated for 15 min in the dark and analyzed with a FacsCalibur flow cytometer (Becton Dickinson, Le Pont-de-Claix, France).
Statistical analysis for Annexin V geo means collected in individual experiments were performed using a Wilcoxon signed-paired rank test, as distribution of measurements in each condition did not fit normality tests. Differences were considered significant when p<0.05.
For caspase 3/7 activity assay, 2.5×104 cells were seeded in 96-well plates, and staurosporine or DMSO immediately added. Caspase 3/7 activity was then measured using Caspase-Glo 3/7 Assay Kit (Promega) and normalized to the protein concentration determined by DC Protein Assay (Bio-Rad).
KSHV infected immortalized (by stable expression of HPV16 E6/E7) human Lymphatic Endothelial Cells (K-iLEC) were seeded one day before at 5×104 cells/well on 24-well plates. For inhibition of miRNAs, K-iLEC cells were treated with two doses of Tiny LNA oligonucleotides (48 h+48 h) at a final concentration of 1,5 µM. Apoptosis was induced with 500 µM Etoposide (Sigma Aldrich) and DMSO was used as a vehicle control (mock). Cells were fixed with 4% Paraformaldehyde (EMS, Hatfield, PA) 24 h after the treatment with Etoposide or mock. Coverslips were blocked 30 minutes with 5% goat serum and incubated first with 1∶800 diluted Cleaved Caspase-3 (Asp175) rabbit monoclonal antibody (Cell signaling) for 1 h at room temperature, then with a 1∶1000 dilution of a goat anti-rabbit secondary antibody coupled to Alexa Fluor 594 (Invitrogen). Alternatively, apoptosis was detected with TdT-mediated dUTP nick end labeling (TUNEL) assay according to manufacturer's instructions of the kit (In situ Cell Death Detection Kit, TMR red, Roche, Mannheim, Germany). The fluorochromes were visualized with a Zeiss Axioplan 2 fluorescent microscope (Carl Zeiss, Oberkochen, Germany). Images were acquired with a Zeiss Axiocam HRc, using Zeiss AxioVision (version 4.5 SP1) and Adobe Photoshop software (version 7.0; Adobe, San Jose, CA).
|
10.1371/journal.ppat.1001174 | Tyrosine Sulfation of the Amino Terminus of PSGL-1 Is Critical for Enterovirus 71 Infection | Enterovirus 71 (EV71) is one of the major causative agents of hand, foot, and mouth disease, a common febrile disease in children; however, EV71 has been also associated with various neurological diseases including fatal cases in large EV71 outbreaks particularly in the Asia Pacific region. Recently we identified human P-selectin glycoprotein ligand-1 (PSGL-1) as a cellular receptor for entry and replication of EV71 in leukocytes. PSGL-1 is a sialomucin expressed on the surface of leukocytes, serves as a high affinity counterreceptor for selectins, and mediates leukocyte rolling on the endothelium. The PSGL-1–P-selectin interaction requires sulfation of at least one of three clustered tyrosines and an adjacent O-glycan expressing sialyl Lewis x in an N-terminal region of PSGL-1. To elucidate the molecular basis of the PSGL-1–EV71 interaction, we generated a series of PSGL-1 mutants and identified the post-translational modifications that are critical for binding of PSGL-1 to EV71. We expressed the PSGL-1 mutants in 293T cells and the transfected cells were assayed for their abilities to bind to EV71 by flow cytometry. We found that O-glycosylation on T57, which is critical for PSGL-1–selectin interaction, is not necessary for PSGL-1 binding to EV71. On the other hand, site-directed mutagenesis at one or more potential tyrosine sulfation sites in the N-terminal region of PSGL-1 significantly impaired PSGL-1 binding to EV71. Furthermore, an inhibitor of sulfation, sodium chlorate, blocked the PSGL-1–EV71 interaction and inhibited PSGL-1-mediated viral replication of EV71 in Jurkat T cells in a dose-dependent manner. Thus, the results presented in this study reveal that tyrosine sulfation, but not O-glycosylation, in the N-terminal region of PSGL-1 may facilitate virus entry and replication of EV71 in leukocytes.
| Enterovirus 71 (EV71) is a major causative agent of hand, foot, and mouth disease and a diverse array of neurological diseases, including fatal encephalitis, in children. EV71 has increasingly caused large outbreaks of hand, foot, and mouth disease particularly in the Asia-Pacific region. Recently, we identified human P-selectin glycoprotein ligand-1 (PSGL-1) as a functional receptor for EV71. PSGL-1 on immune cells is a key molecule involved in early inflammatory events and the PSGL-1–selectin interaction is regulated by post-translational modifications of PSGL-1. Here, we found that a post-translational modification, tyrosine sulfation, at the N-terminal region of PSGL-1 is critical for its binding to EV71 and subsequent viral replication in lymphocytes. Important roles for tyrosine sulfation in protein-protein interactions have been widely accepted; however, involvement of tyrosine sulfation of the receptor in the virus-receptor interaction has been reported only for HIV-1. Therefore, this is the second and unique example of the involvement of tyrosine sulfation in specific virus-receptor interactions. Our results shed new light on biological roles for tyrosine-sulfated proteins in cell tropism and the pathogenesis of EV71.
| Enterovirus 71 (EV71) is a small, nonenveloped, positive-stranded RNA virus that belongs to human enterovirus species A of the genus Enterovirus in the family Picornaviridae. EV71 is a major causative agent of hand, foot, and mouth disease (HFMD), a common febrile disease affecting mainly young children. HFMD is characterized by a skin rash on the palms and soles, and ulcers on the oral mucosa. HFMD due to EV71 and other enteroviruses is usually mild and self-limited; however, EV71 infection may also cause severe neurological diseases including polio-like paralysis and fatal brainstem encephalitis in young children and infants (reviewed in [1], [2]). Over the last decade, many EV71 outbreaks involving a number of fatal encephalitis cases have been reported throughout the world, especially in the Asia-Pacific region, including in Malaysia, Taiwan, Vietnam, and mainland China [2], [3], [4].
Using an expression cloning method by panning with a cDNA library from human Jurkat T cells, we recently identified human P-selectin glycoprotein ligand-1 (PSGL-1) as a functional cellular receptor for EV71 [5]. In addition, Yamayoshi et al. [6] identified scavenger receptor class B, member 2 (SCARB2) as another cellular receptor for EV71 by screening EV71-susceptible transformants after transfecting mouse L929 cells with genomic DNA from human RD rhabdomyosarcoma cells. SCARB2 is ubiquitously expressed on a variety of tissues and cells [7], whereas the tissue distribution of PSGL-1 is mainly limited to immune cells such as leukocytes and dendritic cells [8]. We have also demonstrated that some EV71 strains (PSGL-1–binding strain; EV71-PB) use PSGL-1 as the primary and functional receptor for infection of Jurkat T cells, but other EV71 strains (PSGL-1–non-binding strain; EV71-non-PB) do not, suggesting phenotypic differences in PSGL-1 usage among EV71 strains. Thus, the identification of two distinct cellular receptors for EV71, PSGL-1 and SCARB2, has provided important clues in the elucidation of the molecular basis of early virus-host interactions and pathogenesis of EV71. However, little is known about the biological significance of the two EV71 receptors.
PSGL-1 is a sialomucin membrane protein that is expressed as a homodimer comprised of two disulfide-linked subunits. Interaction of PSGL-1 with selectins and chemokines is a key event during early inflammation of immune cells [8], [9], [10], [11]. The N-terminal region of PSGL-1 is critical for PSGL-1 binding to P-, E- and L-selectins, and post-translational modifications such as O-glycosylation and tyrosine sulfation in the N-terminal region of PSGL-1 contribute the efficient binding to selectins [12], [13], [14], [15]. We have previously shown that the N-terminal region of human PSGL-1 (amino acids 42–61) containing a potential O-glycosylation residue (T57) and three potential tyrosine sulfation sites (Y46, Y48, and Y51) is directly responsible for PSGL-1 binding to EV71-PB [5]. Therefore, in the present study, we investigated the involvement of post-translational modifications of PSGL-1 in the binding to EV71-PB using a series of PSGL-1 mutants and an inhibitor of sulfation.
Tyrosine sulfation is an important late post-translational modification of secreted and membrane-bound proteins expressed in various mammalian cells and tissues and occurs in the trans-Golgi network [16], [17]. Tyrosine sulfated proteins have been described in many mammalian species, and important roles for tyrosine sulfation in protein-protein interactions have been widely accepted, particularly for various chemokine receptors and their ligands that mediate leukocyte migration during inflammation. Furthermore, it has been well established that tyrosine sulfation of the N-terminal region of the chemokine receptor, C-C chemokine receptor 5 (CCR5), plays critical roles in the function of CCR5 as a coreceptor for virus entry and replication of CCR5-tropic human immunodeficiency virus type 1 (HIV-1) variants [18].
Here we demonstrate that tyrosine sulfation of the N-terminal region of PSGL-1 facilitates PSGL-1–EV71 interaction and viral replication of EV71-PB in Jurkat T cells. To our knowledge, this is the second direct example of the involvement of tyrosine sulfation in specific virus-receptor interactions, a modification that mediates viral entry and replication in target cells.
For binding to selectins, PSGL-1 requires post-translational modifications with sialyl Lewis x-containing O-glycans at T57. α1,3-fucosyltransferase (FUT7) is involved in the biosynthesis of sialyl Lewis x determinants (Fig. 1A) [19], [20]. Prevention of O-glycosylation by alanine substitution at T57 (T57A) eliminates binding of PSGL-1 to P-selectin without affecting tyrosine sulfation [12]. First, we generated and expressed a PSGL-1-T57A mutant (Fig. 1A) in 293T cells (293T/T57A) to examine the role of O-glycosylation on T57 for PSGL-1 binding to EV71-1095, a representative strain of EV71-PB [5]. As a positive binding control, we used a soluble form of recombinant P-selectin (P-selectin-Fc). P-selectin-Fc did not bind to any PSGL-1 transfectants in the presence of 2 mM EDTA (Fig. 1B). P-selectin-Fc bound weakly to 293T cells transiently expressing PSGL-1 (293T/PSGL-1) in the presence of Ca2+ but not to 293T/T57A cells (Fig. 1B). Double expression of PSGL-1 and FUT7 in 293T cells resulted in the efficient binding of P-selectin-Fc to PSGL-1 in a calcium-dependent manner (Fig. 1B). Even in the presence of Ca2+ and FUT7, P-selectin-Fc did not bind to 293T/T57A cells (Fig. 1B). These observations are consistent with previous findings that interaction of PSGL-1 with P-selectin is calcium-dependent and requires appropriate O-glycosylation of PSGL-1 at T57 [10], [12]. In contrast, EV71-1095 showed marked binding to 293T/PSGL-1 cells in a calcium-independent manner, even in the absence of FUT7 (Fig. 1B). EV71-1095 also bound to 293T/T57A cells (Fig. 1B). These results indicate that, unlike the interaction between PSGL-1 and P-selectin, the interaction between PSGL-1 and EV71-1095 does not require Ca2+ and the O-glycans at T57 of PSGL-1.
To examine the role of sialic acids on the cell surface, including sialyl Lewis x moieties in the potential O-glycans at T44 and T57 of PSGL-1, on EV71 binding to 293T/PSGL-1 cells, we tested EV71 binding to the cells pretreated with sialidase. Sialidase treatment removed cell-surface sialyl Lewis x (Fig. 2A) and reduced P-selectin-Fc binding to 293T/PSGL-1 cells (Fig. 2B). On the other hand, EV71-1095 binding to the sialidase-treated cells was not reduced regardless of the removal of sialyl Lewis x (Fig. 2C). Although treatment with sialidase decreased EV71 infection to DLD-1 cells [21], sialic acids on the cell surface of 293T/PSGL-1 cells are not necessary for the binding of PSGL-1 to EV71-1095.
In addition to O-glycosylation of PSGL-1, sulfation of the three tyrosines (Y46, Y48, and Y51) in the N-terminal region of PSGL-1 is required for high affinity binding to P- and L-selectins [13], [14], [15], [22], [23]. To assess the role of tyrosine sulfation of PSGL-1 in the PSGL-1–EV71 interaction, we treated 293T/PSGL-1 cells with sodium chlorate, an inhibitor of sulfation that blocks PSGL-1 binding to P-selectin [13]. As described previously, sodium chlorate had no apparent effect on PSGL-1 expression on the cell surface (Fig. 3A). On the other hand, sodium chlorate reduced sulfated tyrosines on the cell surface (Fig. 3B) and inhibited EV71-1095 binding to 293T/PSGL-1 cells in a dose-dependent manner (Fig. 3C). These observations indicated that sulfation of PSGL-1, in addition to its expression on the cell surface, is important for EV71 binding.
We then determined the requirement for the putative sulfated tyrosines (Y46, Y48, or Y51) in the N-terminal region of PSGL-1 for its binding to EV71-1095. We generated PSGL-1 mutants with phenylalanine substitutions at one or more tyrosines and a mutant with a deletion of this region (Fig. 4A). We transfected 293T cells with expression plasmids containing the PSGL-1 mutants and used them for the EV71 binding assay using flow cytometry. 293T cells transfected with an empty vector expressed little or no detectable tyrosine sulfated proteins on the cell surface (Fig. 4B). Similar to the binding of PSGL-1 to P-selectin [13], [14], substitution of the tyrosines with phenylalanine prevented tyrosine sulfation and PSGL-1 binding to EV71-1095 (Fig. 4B). Substitution of one or two tyrosines slightly reduced (Y46F) or impaired (Y48F, Y51F, Y4648F, or Y4651F) the binding of PSGL-1 to EV71-1095 regardless of the apparent expression of tyrosine sulfated proteins on the cell surface (Fig. 4B). Substitution of two or three tyrosines (Y4851F or FFF) or deletion of the region (d46–51) reduced tyrosine sulfated proteins on the cell surface and completely disrupted the PSGL-1–EV71 interaction (Fig. 4B). We also examined the role of tyrosine sulfation in PSGL-1 binding to other EV71-PB strains. Binding of SK-EV006, C7/Osaka, KED005, and 75-Yamagata strains to 293T/PSGL-1 cells was also inhibited by sodium chlorate (Fig. S1). These strains bound to 293T/T57A cells but not to 293T cells expressing the PSGL-1-FFF mutant. Taken together, these findings demonstrate that, in contrast to O-glycosylation at T57, tyrosine sulfation of PSGL-1 is essential for the efficient binding to EV71-PB strains.
We next examined whether sulfation of PSGL-1 is required for PSGL-1-dependent replication of EV71-PB in Jurkat T cells. Jurkat T cells were infected with EV71 and cultured in the presence of sodium chlorate to inhibit the sulfation of PSGL-1. Sodium chlorate treatment did not affect PSGL-1 expression on Jurkat T cells (Fig. 5A). On the other hand, sodium chlorate significantly inhibited the replication of EV71-1095 in a dose-dependent manner (Fig. 5B). The replication of other EV71-PB strains was also inhibited in the presence of sodium chlorate (Fig. 6). In contrast, replication of EV71-non-PB strains (EV71-02362 and EV71-Nagoya), which can replicate in Jurkat T cells in a PSGL-1-independent manner [5], was not affected by sodium chlorate (Figs. 5C and 6). This observation supports that sodium chlorate inhibited replication by blocking EV71-PB entry into the cells. To confirm that sodium chlorate is acting at the receptor level, we transfected Jurkat T cells with genomic RNA of EV71-1095 and examined viral titers at 24 h posttransfection in the presence or absence of 30 mM sodium chlorate. Although infectious viruses were recovered in the presence of sodium chlorate, the mean viral titer in the presence of sodium chlorate was over 10 times lower than that of the control experiments (data not shown). Although sodium chlorate inhibited EV71-PB-binding to PSGL-1 expressing cells (Figs. 3C and 5B (0 h postinfection)), we could not rule out the possible involvement of the sodium chlorate treatment during the later stages of viral replication. Further studies are needed to elucidate the inhibitory mechanism of action of sodium chlorate in a receptor dependent or independent manner during different stages of viral replication of EV71.
Replication of the G-10 strain of coxsackievirus A16, which may use another unidentified receptor(s) to infect Jurkat T cells [5], [24], was significantly inhibited by sodium chlorate (Fig. 6). This result suggests that some sulfated molecules other than PSGL-1 might be involved in the replication of coxsackievirus A16 in Jurkat T cells in a PSGL-1-independent manner.
We have shown that tyrosine sulfation, but not O-glycosylation, of the N-terminal region of PSGL-1 is critical for EV71-PB binding to PSGL-1 and for virus entry and subsequent replication of EV71-PB in Jurkat T cells. First, unlike P-selectin-Fc, EV71-PB bound to a PSGL-1 mutant with an alanine substitution at the potential O-glycosylation site (T57) in a calcium-independent manner (Figs. 1 and S1). Second, removal of sialyl Lewis x by sialidase did not reduce PSGL-1 binding to EV71 (Fig. 2). Third, a sulfation inhibitor, sodium chlorate, significantly impaired EV71-PB binding to PSGL-1 in a dose-dependent manner (Figs. 3 and S1). Fourth, EV71-PB binding to PSGL-1 was inhibited when phenylalanine substitutions were made at one or more potential tyrosine sulfation sites in the N-terminal region of PSGL-1 (Figs. 4 and S1). Finally, PSGL-1-dependent viral replication of EV71-PB strains in Jurkat T cells, but not EV71-non-PB strains, was inhibited by sodium chlorate (Figs. 5 and 6).
Human PSGL-1 is one of the most characterized tyrosine sulfated proteins at the molecular level [11]. The involvement of O-glycans and sulfated tyrosines in the structural and functional basis of PSGL-1 binding to its natural ligands has been extensively studied, and distinct requirements for tyrosine sulfation for PSGL-1 binding to selectins have been elucidated. Among the three potential sulfated tyrosines of human PSGL-1, Y46 and Y51, but not Y48, are important for PSGL-1 binding to L-selectin along with a core-2 based O-glycan with sialyl Lewis x at T57 [22]. On the other hand, the crystal structure of the lectin and EGF domains of P-selectin co-complexed with the N-terminal domain of PSGL-1 revealed a critical involvement of sulfated tyrosines at Y48 and Y51 for direct molecular contact with P-selectin [11]. The corresponding interactions via sulfated tyrosines are not formed in E-selectin binding in the crystal structure of the PSGL-1–E-selectin complex [11]. Thus, tyrosine sulfation is critical for PSGL-1 binding to L- and P-selectins, but not to E-selectin [14]. In our study, we have shown that sulfated tyrosines at Y48 and Y51 play a critical role in PSGL-1 binding to EV71-PB. However, O-glycosylation at T57 and sialyl Lewis x moieties on the potential O-glycans of PSGL-1 were not required for the PSGL-1–EV71 interaction, suggesting distinct structural requirements between EV71 and P-selectin for PSGL-1 binding. To elucidate the structural basis of the PSGL-1–EV71 interaction, further studies will be needed to identify genetic determinants in EV71 capsid proteins required for PSGL-1 binding using both EV71-PB and non-PB strains.
Yang et al. [21] have recently reported that EV71 may use sialylated glycans as receptors for infection in intestinal DLD-1 cells. In our current study, we showed that potential O-glycans at T57 and sialic acids are not critical for binding to EV71-PB (Figs. 1 and 2). However, our study does not exclude possible contributions of sialic acids and other proteins with or without O-glycans on the cell surface of various cells during the course of EV71 replication in a PSGL-1-dependent or -independent manner [21], [24], [25].
In contrast to the structural requirements of O-glycans for PSGL-1 binding to selectins, all three sulfated tyrosines, but not O-glycans at T57, are required for PSGL-1 binding with the skin-associated chemokine, CCL27 [9]. PSGL-1 facilitates P-selectin-mediated T cell migration in the inflamed skin [26], [27] and interacts with the chemokine CCL27 to regulate skin-homing T cells [9]. HFMD pathogenesis due to EV71 can be characterized as acute skin inflammation. Therefore, it is possible that binding of EV71-PB with PSGL-1-positive skin-homing T cells and/or Langerhans cells, and subsequent viral replication in those cells, may participate in HFMD pathogenesis and progression. The status of tyrosine sulfation of PSGL-1 on those cells may modulate cell migration and PSGL-1-dependent replication of EV71-PB in the inflamed skin.
An important role for tyrosine sulfation of a specific cellular receptor in viral entry and replication has been demonstrated for the first time in a co-receptor for HIV-1, CCR5 [18]. CCR5 is a functional receptor for macrophage inflammatory protein (MIP)-1α and MIP-1β, and is expressed on memory/effector T cells, macrophages, and immature dendritic cells [28]. The N-terminal region of CCR5 is highly modified by tyrosine sulfation and O-glycosylation, and sulfated tyrosines play critical roles in CCR5 interactions with chemokines [18]. Site-directed mutagenesis and treatment with sodium chlorate revealed that sulfation of tyrosine residues in the N-terminal region of CCR5 is required for efficient CCR5 binding to MIP-1α and MIP-1β, and to HIV-1 gp120-CD4 complexes, without affecting the expression of CCR5 [18]. Likewise, the efficacy of HIV-1 entry was significantly reduced in cells expressing CCR5 mutants with one or more phenylalanine substitutions at four potential tyrosine sulfated residues compared to that in cells expressing native CCR5 [18]. Tyrosine sulfation may be a common phenomenon in chemokine receptors expressed on immune cells such as leukocytes, platelets, and dendritic cells [16]. Therefore, tyrosine sulfation seems to regulate not only the migration of immune cells but also the infectivity of viruses.
Although the occurrence of severe EV71 infection with a number of fatal cases mainly in children continues to be a major public health threat in the Asia Pacific region, no vaccines or antiviral agents are currently available for EV71 [29]. Our data suggest that the virus-receptor interaction may be a promising target for potential antiviral agents. Thus, soluble PSGL-1 as one such agent may have an inhibitory effect on EV71-PB replication [5]. In our current study, we have demonstrated the possible involvement of tyrosine sulfation of PSGL-1 on EV71 entry into target cells, and accordingly, we showed the inhibitory effect of a tyrosine sulfation inhibitor on viral replication of EV71-PB strains in Jurkat T cells. Thus, the elucidation of the structural and functional basis of virus-receptor interactions will provide novel and unique antiviral approaches for the treatment of severe EV71-associated diseases.
293T cells were maintained in Dulbecco's modified Eagle's medium (DMEM, Wako) supplemented with 10% fetal calf serum (FCS). Jurkat T cells were maintained in RPMI medium (Sigma) supplemented with 10% FCS.
All EV71 strains (Table 1) and the coxsackievirus A16 prototype strain (G-10) were propagated in RD or Vero cells. Because some of the strains produced diffuse plaques on RD cells, the viral titers were determined by a microtitration assay using 96-well plates and RD cells, as previously described [30]. Briefly, 10 wells were used for each viral dilution, and the viral titers were expressed as 50% cell culture infectious dose (CCID50). For flow cytometry, we used concentrated viruses unless otherwise stated. To prepare virus concentrations, viruses were ultracentrifuged, and the amount of EV71 virions was measured.
The anti-EV71 monoclonal antibody (mAb) MA105 (mouse IgG2b) was generated from mice immunized with EV71-1095 (Y. Tano et al., unpublished data) Immunization to mice, fusion, selection of hybridomas, and propagation of hybridomas in the ascite fluid of the mice, were outsourced to Nippon Biotest Laboratories Inc., Tokyo, Japan. The anti-human PSGL-1 mAb KPL1 and anti-sialyl Lewis x mAb CSLEX1 were purchased from BD Biosciences. Anti-human PSGL-1 mAb PL2 was purchased from Beckman–Coulter. Anti-sulfotyrosine mAb Sulfo-1C-A2 [31] was purchased from Millipore. For the negative control, mouse IgG1 (MOPC-21) and IgG2a (G155–178) were purchased from BioLegend and BD Biosciences, respectively. Recombinant P-selectin-Fc was purchased from R&D Systems.
For directional cloning using a CpoI recognition site [32], we introduced a CpoI recognition-compatible (SanDI) site into the pcDNA3.1(+) plasmid (Invitrogen). The BamHI-EcoRI fragment of pcDNA3.1(+) was replaced with 5′-ggatccgggtcccggtaagaattc-3′ (BamHI+SanDI+gg+Stop+EcoRI) to produce pcDNA3.1SS. Human FUT7 cDNA was amplified from Jurkat T cell cDNA with the primers FUT7-F1 (5′-atacggtccggccatgaataatgctgggcacggc-3′) and FUT7-R1 (5′-tgacggaccgtcaggcctgaaaccaaccct-3′). The FUT7 ORF was sub-cloned into a SanDI site in pcDNA3.1SS to produce pcDNA-FUT7. The sequence of the cloned FUT7 ORF was identical to that of FUT7 (NM_004479).
The primers used for mutagenesis/deletion are provided in Table S1. Briefly, cDNA of human SELPLG was cloned into pEF6-Flag-3S [5] to produce pEF-PSGL-1 [5]. Mutations and deletions were introduced into the N-terminal region of human PSGL-1 with PCR, and the mutated SELPLG cDNA was cloned into pEF6-Flag-3S.
293T cells were transfected with expression plasmids using Lipofectamine 2000 (Invitrogen), and DMEM medium was replaced with fresh medium 4 h after transfection. The cells were collected 24 h after transfection by pipetting, and were used for flow cytometry. For inhibition of tyrosine sulfation of PSGL-1, 293T cells were treated with 10–50 mM sodium chlorate in DMEM for 1 day. Four hours after transfection with pEF-PSGL-1, the medium was replaced with medium containing sodium chlorate, and the cells were further incubated for 20 h.
The cells were washed once with flow cytometry buffer (FC buffer; PBS(−) supplemented with 2 mM EDTA, 2% FCS, and 0.1% NaN3) and incubated with the indicated mAb on ice for 30 min. After washing with FC buffer, the cells were incubated with secondary antibodies conjugated with Alexa Fluor 488 (Invitrogen). To detect sialyl Lewis x, the cells were incubated with secondary antibodies conjugated with R-phycoerythrin (SouthernBiotech). To detect PSGL-1 by two-color flow cytometry, PL2 was labeled with a Zenon mouse IgG1 R-phycoerythrin labeling kit (Invitrogen). To detect P-selectin-Fc binding, PBS(−) supplemented with 2 mM CaCl2, 2% FCS, and 0.1% NaN3 was used instead of FC buffer. The cells were washed and analyzed with FACSCalibur (Becton Dickenson).
293T cells (5×105) transfected with the indicated expression plasmid were washed once with FC buffer and incubated with the EV71-1095 preparation (1×107 CCID50) supplemented with 0.1% NaN3, or concentrated viruses (containing 0.5 µg of VP1 protein) per 50 µl FC buffer. The cells were washed and stained for 30 min on ice with Alexa Fluor 488-conjugated MA105.
Cells were processed as in the EV71-binding assay and flow cytometry described above. Prior to the addition of EV71, P-selectin-Fc, or mAb, cells (2.5×106) were incubated with 50 mU/ml of Vibrio cholerae sialidase (Roche) in 500 µl of DMEM supplemented with 2% FCS for 1 h at 37°C and then washed once.
Jurkat T cells (4×104 cells) were inoculated with viruses at 1 CCID50/cell for 1 h on ice, washed, and incubated in medium (200 µl in a 48-well plate) at 34°C. For inhibition of tyrosine sulfation of PSGL-1, the cells were pretreated with 10–30 mM sodium chlorate in medium for more than 3 days, inoculated with viruses, washed, and maintained in medium supplemented with sodium chlorate. For mAb inhibition, the cells were pretreated with 10 µg/ml mAb for 1 h, washed, and maintained in medium with 10 µg/ml mAb. At the indicated time, the infected cells and supernatants were freeze-thawed, and viral titers were determined by CCID50 titration in RD cells. All infection assays were carried out in triplicate unless otherwise stated, and the mean viral titers were compared using Student's t-test (two-tailed). P values<0.01 were considered statistically significant.
|
10.1371/journal.pcbi.1006115 | Age density patterns in patients medical conditions: A clustering approach | This paper presents a data analysis framework to uncover relationships between health conditions, age and sex for a large population of patients. We study a massive heterogeneous sample of 1.7 million patients in Brazil, containing 47 million of health records with detailed medical conditions for visits to medical facilities for a period of 17 months. The findings suggest that medical conditions can be grouped into clusters that share very distinctive densities in the ages of the patients. For each cluster, we further present the ICD-10 chapters within it. Finally, we relate the findings to comorbidity networks, uncovering the relation of the discovered clusters of age densities to comorbidity networks literature.
| Age and sex of a patient can be directly related to susceptibilities to certain medical conditions. We present a method to generate clusters of human phenotype, based on the age of the population. This method helps extract knowledge on age and sex from the data. The age and sex correlations with disease conditions can help in a task of predicting the susceptibility of incoming patients to conditions.
| Studies of groups of diseases occurring together, or disease comorbidities, have traditionally focused on studies of small groups of diseases using techniques of hypothesis-testing [1–6]. The repeated existence of particular comorbidities is important to diagnoses and better index diseases [7, 8]. Databases of electronic medical records contain phenotypic information for humans—namely, patient clinical histories. A novel method to analyze health records is to built the human phenotypic disease network, where nodes represent the diseases and edges indicate comorbidity relations [9]. More recent studies analyze databases on electronic health records to uncover systematic associations in the complete set of known diseases [6, 10–12]. In this context, several methods of information sciences can be used to uncover patterns in electronic patient records. The main interest of these studies is to discover correlations between diseases that can help in prevention and can also inform systems biology frameworks [13]. More recently, computational methods are being used to reduce the costs of healthcare by helping to identify outliers in medical records [14].
Up to date, most of the samples of electronic patient records studied in the literature have used a narrow set of the general population of patients. For example, Hidalgo et al. covered 3 years of medical care claims of patients who were 65 years or older, this biased the information towards population of the elderly. Later, Roque et al. generated fine grained patient stratification and disease co-occurrence statistics of patients from the Sankt Hans Hospital, which is the largest Danish psychiatric facility [15]. Their results focus into phenotypes associated with mental and behavioral disorders or the chapter V of the ICD-10 standard classification catalog. Datasets with more complete sample of the population have become more recently available. Electronic records with time spans of decades allowed, for the first time, to uncover patterns centered on the number of key diagnoses that can detect diseases earlier in a patient’s life [16]. While Chmiel et al. [17] analyzed two years of medical claims of the entire population in Austria. They measured how the comorbidity network change its structure with the age of the patients. This information was used to build a diffusion model that explains a large percentage of the variance of all the disease incidents in a population. In that case, the comorbidity networks were built while pre-defining the age intervals of the patients analyzed. In this work, we present a clustering method by identifying the similarities in the age densities of the actual phenotypic records. We find groups of medical conditions that occur in the unsupervised age groups emerging from the data. These groups are in turn associated with a small set of chapters of the ICD-10 standard classification catalog.
The wisdom of doctors when it comes to assessing susceptibility to diseases have been influenced by the years of practice and observation of many cases on daily basis. Doctors’ knowledge of the susceptibility to diseases at different ages/sexes serves as an essential prior to perform diagnostics of incoming patients. Similar symptoms for patients might lead to different diagnosis depending on the age and sex of the patient, a patient who is 70 years old is much more likely to suffer a heart attack than a 10 year-old even if both patients are suffering the symptom of chest pain. We show here that this common knowledge can be inferred from the data. Besides the symptoms a patient is having, the age and sex can aid the diagnostic process. We present a framework that automatically uncovers the relationship between health conditions and the age/sex of a patient. To that end, we group the health conditions based on their similarities in population age densities. Then, we construct the comorbidity graph in the same way found in the literature [9, 17] to investigate the relationship of comorbidity coefficient values to the discovered clusters of conditions.
For each of the 1.7 million patients there is a log for each visit to the doctor within the 17 months of study from March, 2013 to July, 2014. The data corresponds to medical claims from one of the largest healthcare insurance companies in Brazil. Each health record in the database has several attributes pertaining to the data of the visit, it is synthesized via ICD-10 codes that detail the condition and the purpose of the visit. ICD-10 codes have a range of 23 thousands of different identifiers each representing a health condition of a patient. In addition, the data has the age and sex of each patient. The total number of visits is 6.6 millions, resulting in 47 million conditions.
In Fig 1A we show the age distribution of the entire population of patients in the data. With the age distribution peaking at 34 which is the median age, in agreement with the median age of the entire population of Brazil.
Fig 1B shows the cumulative distribution of the frequencies of ICD-10 codes in the data. About 50% of the ICD-10 codes had a frequency of less than 100 times among all patients’ visits, while the rest of the 50% of the codes makes up for 90% of the records in the data set.
The ICD-10 coding scheme is structured in a tree and the top level contains 22 chapters. The chapters of ICD-10 have common characteristics pertaining to the same organ/system or relating to the nature of the visit. The size of the chapter nodes in Fig 2 corresponds to the frequency of observing the chapter in the data. The thickness of the edges between the nodes corresponds to the frequency of co-occurrence of the chapters in the patients records. Chapter XXI has the highest frequency in the data, it is described as factors influencing health status and contacts with health services, such as performing routine checkups. Chapter X contains the group of conditions relating the respiratory system; VII are diseases of the eye and adnexa; XIII are diseases of the musculoskeletal system and connective tissue and XVIII are abnormal clinical and laboratory findings not elsewhere classified. The description of all the chapters of disease codes is included in (S1 Table).
By inspection, each ICD-10 code has a distinctive signature of density on the age dimension that spans the various age groups from birth onward. Fig 3 shows example age density signatures of Chickenpox and Glaucoma. As expected, Glaucoma is more prevalent for the older group [18] and Chickenpox in kids [19, 20]. The shapes of these distributions hint that there is a pattern of higher likelihood of patients of a certain age for different diseases.
We further analyze the age densities of ICD-10 codes in the data to segment ICD-10 codes into groups of conditions with similar age densities. As a robustness measure, we consider the analysis by excluding all codes of chapters XVIII-XXII. The excluded chapters include symptoms (e.g. R codes), procedural details (such as complications or adverse drug effects) and also personal factors (general examinations and such). We represent the age distribution as a vector of 100 elements, each element has the probability of a patient of the corresponding age within the population of patients having the code. This is defined as probability p(age|patient ∈ c) where age is the age of the patients, c is a disease code and patient ∈ c is the set of patients that had a visit labeled as c.
We cluster the densities for each ICD-10 code based on the vector representation of the age density p(age|patient ∈ c). We use Hierarchical Agglomerative Clustering (HAC) to group the codes into clusters. The method is further discussed in the material and methods section. The age distribution of the codes clusters into six main groups as shown in Fig 4. Clusters A and B show two clusters of codes having higher density towards the lower spectrum of ages. Cluster C shows a group of codes that have age densities concentrated in the ages 20 to 40. Cluster D has diseases that are almost uniformly distributed across the ages. Cluster E has codes with densities concentrating in the range of ages over 60 and cluster F has codes with age densities concentrating over 70. The kernel density estimation of the probability density of the clusters is included in (S1 Fig).
Fig 5 illustrates a few examples of the high prevalent ICD-10 codes from the clusters discovered in the data. For each cluster, Fig 5 shows the clustering dendrogram with a depth of six, branches in the dendrogram with a depth higher than six are represented by the disease that is most common in their respective branch. The branches are labeled by their clusters from A to F. Within cluster A, J21 acute bronchiolitis and H65 otitis media nonsuppurative were observed in 0.4 and 1.2 percents of the population respectively, both have a concentration towards the lower ages as shown previously. Cluster B has J06 acute infections of the upper airways with 8 percents of the population of patients. Furthermore, it has A09 diarrhea and J03 acute tonsillitis each with around 5.9 percents respectively. The noticeably increase of the percentage of patients is due to the population age distribution shown in Fig 1. Cluster C with O82 Cesarean delivery has around 0.8 percents of the population of patients, the cluster is consistent with the defined age range between 20 and 40. Cluster D has H52 disorders of refraction and accommodation with 10.6 percents and J01 acute sinusitis with 6.7 percents of the population of patients. As expected, as the clusters have more density around the peak of the age distribution of the population, the number of patients per code in the clusters becomes higher. Cluster E with age density towards the elderly has M54 back pain with 10.8 percents and M25 other joint disorders with 4.7 percents as the most common. Cluster F with age density in the oldest group has I10 essential hypertension (primary) with 10.4 percents and N39 other disorders of the urinary tract with around 3.5 percents. Pneumonia is third in around 1.8 percents.
Fig 6 shows the decomposition of the clusters in terms of sex and age distribution of each cluster, which has the expected results. Further, we show the probability of association between clusters and the ICD-10 chapters agreed by the World Health Organization [21], we use the Fisher exact test to measure the association between a chapter of codes to our identified clusters. Clusters have increasing mean age except for cluster C where the age range concentrated around 34. Cluster C is dominated by female patients. This is explained by the high probability of association with ICD-10 codes in chapter XV pertaining to pregnancy and childbirth and postpartum. Interestingly, from A to D each cluster has their own signature of few associated chapters, while E and F are associated with more chapters proper of aging.
This section sheds light on the age related characteristics of the edges in comorbidity networks [9, 17]. We first construct the comorbidity network through the measure of relative risk between conditions. Further details about the measure of relative risk are included in the materials and methods. Fig 7 shows a sample of the comorbidity network. In the figure, we only show the edges with highest two thousand relative risk values in the quantified comorbidities. The figure is splitted into two parts A and B. Part A shows the intra-cluster edges and part B shows the inter-cluster edges. The sample selection of edges and nodes display are done for visualization purposes.
To relate the clusters of diseases reported earlier to the study of comorbidity networks, we study the distribution of relative risk for inter-cluster versus intra-cluster comorbidities. Fig 8 shows the distributions of the relative risk of inter-cluster versus intra-cluster comorbidities. For each cluster in the data, we quantify the distribution of the relative risk of intra-cluster comorbidities (in red) and plot it against the distribution of the relative risk of inter-cluster comorbidities (in gray).
We find a clear variation in the divergence between the density of relative risk for inter-cluster comorbidities to the intra-cluster ones. The closer the distribution of age for patients in a cluster to a uniform distribution, the less the divergence in relative risk between inter and intra cluster edges. The divergence is highest in clusters A, C and F. They belong to the clusters that identify infants, women in reproductive age and the elderly. It is followed by clusters E and B. With E grouping age density towards the elderly with M54 back pain patients and M25 other joint disorders, while B groups conditions with concentration in teenage years and early adulthood. Cluster D is the closest in age distribution to a uniform, and has the minimal divergence to the distribution of inter-cluster comorbidity density, which has patients in H52 disorders of refraction and accommodation and J01 acute sinusitis.
This paper presents an approach towards investigating groups of diseases based on their relation to age and sex using the records of medical visits from a diverse population. We show that besides the symptoms, age and sex can rank the susceptibility to conditions in a diagnostic process. Using Hierarchical Agglomerative Clustering, we uncover 6 significant groups of medical conditions that present strong similarities on the age density of the patients. Each group of these medical conditions has meaningful associations with few of the 22 standard chapters used to categorize diseases. To find these associations we use the Fisher exact test. We relate the found groups of conditions to the study of comorbidity networks. Pairs of conditions tend to have higher relative risk with varying magnitudes when conditions are in the same group (intra-cluster conditions) compared to conditions that are not in the same group (inter-cluster conditions). This in a sense means that the correlations of conditions in terms of sex and gender partially explain the higher relative risk values discovered in comorbidity networks [9, 17]. Our findings build prior knowledge related to age and sex for automated diagnostics in a Bayesian setting to predict the condition of a patient given their symptoms. The code and data of the study are available for access at http://www.github.com/fha/brazil_health_study.
This paper studies a population of 1.7 million patients in Brazil, containing 47 million of health records with detailed medical conditions for visits to medical facilities for a period of 17 months. The data were analyzed anonymously for the privacy of patients’ data.
To uncover common patterns of the age distribution of ICD-10 codes, we used a Hierarchical Agglomerative Clustering (HAC) approach to group the codes based on the similarities of age distributions. Each code is represented by a vector v of length 100 where each cell represents p(age = i|patients ∈ c) where patients ∈ c is the set of patients with the condition on their records.
HAC cluster vectors, where each vector is a representation of the probability mass function of a code in the data. The vector representation of the probability mass function of the ages of a ICD-10 code is as follows:
p ( a g e | p a t i e n t ∈ c o d e ) = [ p 1 , p 1 , . . . . , p 100 ] (1)
Where pi = p(age = i|patient ∈ code) for a given code. At initialization, HAC assigns each vector object to a cluster, and sequentially merging them into clusters until all codes form one cluster. For measuring the distance d between two vector representations of age density, we use euclidean distance. The Ward distance criterion of clusters is dependent on the within cluster distances and the across clusters distances. Ward algorithm is conservative when merging clusters, thus it tends to find very compact clusters [22].
HAC provides a hierarchy structure of the clustered codes as illustrated in Fig 4. To determine the number of clusters k that best divide the data, we calculate the total within-cluster distances for k from 1 to 20. The total of distances drops as k increases until it does not decrease significantly. We select k that corresponds to the point where the total distances stops decreasing significantly. This method is known as the elbow curve method.
To quantify the comorbidity between conditions, we employ a similar measure to what is used in the literature [9, 17]. We used the relative risk measure to quantify the comorbidity between conditions in the dataset. The formula for quantifying the relative risk between two conditions is given by:
R R i j = C i j N P i P j (2)
Where Cij is the number of patients having both i and j diseases, N is the total number of patients in the data. Pi is the prevalence of condition i and Pj is the prevalence of condition j.
|
10.1371/journal.ppat.1005763 | Open Source Drug Discovery with the Malaria Box Compound Collection for Neglected Diseases and Beyond | A major cause of the paucity of new starting points for drug discovery is the lack of interaction between academia and industry. Much of the global resource in biology is present in universities, whereas the focus of medicinal chemistry is still largely within industry. Open source drug discovery, with sharing of information, is clearly a first step towards overcoming this gap. But the interface could especially be bridged through a scale-up of open sharing of physical compounds, which would accelerate the finding of new starting points for drug discovery. The Medicines for Malaria Venture Malaria Box is a collection of over 400 compounds representing families of structures identified in phenotypic screens of pharmaceutical and academic libraries against the Plasmodium falciparum malaria parasite. The set has now been distributed to almost 200 research groups globally in the last two years, with the only stipulation that information from the screens is deposited in the public domain. This paper reports for the first time on 236 screens that have been carried out against the Malaria Box and compares these results with 55 assays that were previously published, in a format that allows a meta-analysis of the combined dataset. The combined biochemical and cellular assays presented here suggest mechanisms of action for 135 (34%) of the compounds active in killing multiple life-cycle stages of the malaria parasite, including asexual blood, liver, gametocyte, gametes and insect ookinete stages. In addition, many compounds demonstrated activity against other pathogens, showing hits in assays with 16 protozoa, 7 helminths, 9 bacterial and mycobacterial species, the dengue fever mosquito vector, and the NCI60 human cancer cell line panel of 60 human tumor cell lines. Toxicological, pharmacokinetic and metabolic properties were collected on all the compounds, assisting in the selection of the most promising candidates for murine proof-of-concept experiments and medicinal chemistry programs. The data for all of these assays are presented and analyzed to show how outstanding leads for many indications can be selected. These results reveal the immense potential for translating the dispersed expertise in biological assays involving human pathogens into drug discovery starting points, by providing open access to new families of molecules, and emphasize how a small additional investment made to help acquire and distribute compounds, and sharing the data, can catalyze drug discovery for dozens of different indications. Another lesson is that when multiple screens from different groups are run on the same library, results can be integrated quickly to select the most valuable starting points for subsequent medicinal chemistry efforts.
| Malaria leads to the loss of over 440,000 lives annually; accelerating research to discover new candidate drugs is a priority. Medicines for Malaria Venture (MMV) has distilled over 25,000 compounds that kill malaria parasites in vitro into a group of 400 representative compounds, called the "Malaria Box". These Malaria Box sets were distributed free-of-charge to research laboratories in 30 different countries that work on a wide variety of pathogens. Fifty-five groups compiled >290 assay results for this paper describing the many activities of the Malaria Box compounds. The collective results suggest a potential mechanism of action for over 130 compounds against malaria and illuminate the most promising compounds for further malaria drug development research. Excitingly some of these compounds also showed outstanding activity against other disease agents including fungi, bacteria, other single-cellular parasites, worms, and even human cancer cells. The results have ignited over 30 drug development programs for a variety of diseases. This open access effort was so successful that MMV has begun to distribute another set of compounds with initial activity against a wider range of infectious agents that are of public health concern, called the Pathogen Box, available now to scientific labs all over the world (www.PathogenBox.org).
| Preclinical development for drugs in neglected diseases remains a slow process due to a lack of access to compounds, and legal complications over intellectual property ownership. One way to accelerate drug discovery is to provide open access to bioactive molecules with public disclosure of the resulting biological data. The data from open access of bioactive molecules can help prioritize which compounds to investigate further through medicinal chemistry for the original indication and can also uncover other indications for compound development. It was in this spirit of providing open access of malaria-bioactive compounds, and disseminating the results in the public domain, that the Malaria Box project was initiated by the Medicines for Malaria Venture.
Since 2007, over 6 million compounds were screened against asexual-stage Plasmodium falciparum, at two pharmaceutical companies (GlaxoSmithKline [1] and Novartis [2]), and two academic centers (St. Jude, Memphis [3], and Eskitis, Australia [4]), resulting in over 20,000 compounds active in the low- to sub-micromolar range. The structures of the 20,000 anti-malaria hits were made available in ChEMBL (www.ebi.ac.uk/chembl), but discussions with biology groups had underlined the importance of access to the compounds themselves for testing. Cluster analysis and commercial availability reduced this to a set of 400 representative compounds, the ‘Malaria Box’, which was distributed freely to researchers who provided a rationale for screening [5]. This paper presents a summary and analysis of the collected results of the Malaria Box screening from 55 groups who performed a wide variety of assays, the large majority of which are presented in this paper. The collective results are greater than the sum of the individual assays, because each compound can be queried for activity, pharmacokinetic, and safety data to gauge its suitability as a starting point for subsequent medicinal chemistry optimization efforts.
The Heat Map (S1 Table) reports the data from over 290 assays run on the Malaria Box compounds; a snapshot is shown in Fig 1. The results are color coded, where the compounds with the highest activity are coded red and those with relative inactivity green. In the center of the box in S1 Table, the numerical value for the compound is given. It can be seen immediately that some compounds have activities in several biological assays across multiple species and these tend to have activity against mammalian cells as well, whereas other compounds have a rather limited spectrum of activity and are less toxic to mammalian cells.
The data demonstrated in S1 Table are provided by 55 groups who have performed 291 assays to screen the Malaria Box. The vast majority of the data are presented for the first time in this paper. In supplementary data S1 Table, note that columns with data presented for the first time in this paper, representing 236 assays, are colored pink on the top row; published /in press data columns, 55, are grey, with citations provided. Presenting the combined dataset provides insights into the hit rates in these various assays while allowing rapid access to the data by the wider scientific community.
The Heat Map (S1 Table) presents the Malaria Box chemicals grouped by chemical relatedness. Of the 400 compounds, over 100 are closely-related paired molecules so immediate structure-activity-relationships (SAR) can often be seen from hits with these pairs. The Heat Map identified obvious correlations in chemistry and biology between compounds (both Mechanism-of-Action and phenotypic activity). Some biological assays are relatively similar; for example, there were a large number of different P. falciparum gametocyte assays (S1 Table, columns AV-CB), which also cluster, although not perfectly. As such, the aggregate screening data help overcome inter-laboratory bias and identify outstanding activities. For example, compounds that were active in multiple gametocyte assays represent more solid positives than a compound that was active in only one screening assay. However, the gametocyte assays were often performed using different techniques and screening concentrations (see S1 Methods and Results, for details) and one assay may be preferred over another to select compounds with gametocyte activity. Thus having the aggregate data presented together with the individual protocols is more valuable than just having each individual data set to look at sequentially.
Early safety data were obtained by testing all compounds against 73 human cell lines at 10 μM or above, and developing zebrafish embryos were exposed at 5 μM, providing further clues on potential safety issues. A frequent cardiotoxicity safety concern is QTc prolongation, and all compounds were screened for hERG inhibition [6], which is a proxy for this risk (S1 Table column GI). The efficacy and safety of anti-malarial compounds could be altered in endemic regions when administered to patients who are also treated for HIV (Human Immunodeficiency Virus) or TB (tuberculosis), due to drug-drug interactions in the liver. To flag such interactions, we employed two recent breakthrough models: a bioengineered microscale human liver in a high-throughput assay format that accurately captures human drug-drug interactions not detectable in animals or cell lines [7] and a custom-made, robotic high-throughput Luminex bead-based method for profiling the expression of 83 human liver drug-metabolizing enzymes [8]. Combining these tools, we profiled the Malaria Box compounds for induction or inhibition of drug-metabolizing pathways (S1 Table, columns GL-HA) and thereby ranked compounds for potential for drug interactions with existing HIV and TB regimens, to enhance selection of compounds with the lowest safety risks. We also scored the Malaria Box compounds for acute hepatoxicity by monitoring morphology and daily albumin and urea secretion from hepatocytes (S1 Table, columns FQ-FS).
G protein-coupled receptors (GPCRs) represent the largest human drug target class [9]; they affect neurological and cardiovascular physiology and are included in routine safety pharmacology panels [10]. Therefore, in vitro affinity determinations on 23 selected human off-target GPCRs were performed on a subset (10%) of MMV compounds (S1 Table, columns HC-HZ). One of the most severe GPCR-related adverse effects is cardiac valvulopathy linked to 5-HT2B activation [11, 12]. Therefore, some of the MMV compounds with significant binding affinity for the 5-HT2B receptor were also tested on the corresponding functional assay to determine a potential agonistic effect. In addition, predictions of compound glutathione reactivity and epoxidation potential were calculated for each of the Malaria Box compounds (S1 Table, columns IB-IC). These combined safety results alert us to compounds with issues that hopefully can be resolved in subsequent medicinal chemistry programs.
Prior to in vivo pharmacology evaluation it is important to know that an effective plasma concentration can be reached; this exposure was measured in rodents for all compounds, from a single high oral dose (140 μmol/kg). Around one third of the compounds generated high plasma Cmax (>1 μg/ml) and/or high overall exposure (S1 Table, columns GD-GE). This is a higher than expected percentage of compounds with measureable oral bioavailability than if compounds were randomly selected, and probably reflects the large number of drug-like leads selected for the Malaria Box. The combination of in vitro potency and bioavailability provides a rough dosing estimate, informing subsequent decision-making around selection of development leads.
The combined analysis of all of these safety and pharmacokinetic data allows selection of the most promising compounds to advance to medicinal chemistry, and which parameters should be monitored and improved during a medicinal chemistry program.
The activity of Malaria Box compounds against the asexual, erythrocytic stage of P. falciparum was confirmed by five laboratories on seven different P. falciparum strains. There were sometimes 5-10-fold differences in the effective concentration that caused a 50% reduction in growth (EC50) in each assay, and these may have been due to variations in the readouts for the screening assays (LDH release, MitoTracker or Sybr Green dye incorporation, hypoxanthine incorporation, DAPI imaging assay), variations in the protein concentration in the assay medium (affecting the free compound concentration), the time the compound incubated time, or other differences. However, usually the results were consistent and strain-independent. We have documented which sub-stage of the asexual lifecycle the compounds acted upon (S1 Table, columns AA-AE). This information is important in identifying compounds that may overcome existing resistance against artemisinin and other antimalarials. For instance, compounds that target early ring stage intra-erythrocytic parasites and have fast-killing dynamics are sought after because, like artemisinins, they kill parasites rapidly and may reduce patient mortality. Table 1 shows compounds that also target liver stages of the parasite’s life cycle.
P. berghei liver stage (LS) inhibition, using parasite-encoded luciferase activity as a readout of infection in HepG2 cells, was independently determined by two groups at very different screening concentrations (Hanson: 5 μM, Winzeler: 50 μM). Forty-three compounds, roughly 10% of the compound library, inhibited infection by at least 50% at 5 μM and 90% at 50 μM (referred to as LS double actives). HepG2 cell toxicity, (50% or greater reduction in HepG2 abundance based on direct or indirect readouts) was observed with 63% of Malaria Box compounds at 50 μM, while only 10% were toxic at the 5 μM concentration. After excluding those that showed significant toxicity in HepG2 cells at both 5 and 50 μM, Malaria Box compounds were stratified by potential mode-of-action annotation (S1 Table, column M). Five potential modes of action stood out as enriched in LS double actives: (i) cysteine protease inhibitors (cruzain, rhodesain): 1.8% of all Malaria Box compounds (7/400) and 4.7% (2/43) of LS double actives; (ii) possible respiratory-dependent targets (Δ IC50 in low oxygen vs. normal oxygen): 0.8% (3/400) of all Malaria Box compounds and 4.7% (2/43); (iii) targeting yeast respiration: 3.5% of all Malaria Box compounds (14/400) and 9.3% of LS double actives (4/43); (iv) suspected or known PfDHODH (dihydroorotate dehydrogenase) inhibitors: 2.5% of all Malaria Box compounds (10/400) and 9.3% of LS double active (4/43); and (v) suspected or known cytochrome bc1 inhibitors: 4.3% of all Malaria Box compounds (17/200) and 16.3% of LS double actives (7/43). Compounds with activity against PfATP4, now the most common intra-erythrocytic asexual target seen in phenotypic screens, were not found amongst the LS double actives.
There is a great need for antimalarials that kill dormant, liver-stage P. vivax (hypnozoites), but there is a lack of assays that measure this activity. Only nine compounds (Table 1) show simultaneous activity against gametocytes, liver, and asexual stages, whilst lacking evidence of toxicity in zebrafish and broad cytotoxicity to mammalian cells. These would be compounds to prioritize for in vitro and in vivo screening against P. vivax hypnozoites and would benefit from additional MoA studies.
Gametocytocidal drugs would block transmission from the human to the mosquito and break the parasite’s life cycle. The data shown in Table 1 include series with activities on both gametocyte and liver stages, and some of the data intriguingly challenges existing assumptions. For instance, MMV007116 in this category is a mitochondrial (bc1) inhibitor (S1 Table, column M, line 168) and has activity in a number of gametocytocidal assays, but other bc1 inhibitors are not generally gametocytocidal, suggesting another MoA for this compound. We also see 4-aminoquinolines as inhibitors of some gametocyte assays, although the parent 4-aminoquinoline compound chloroquine is known not to be gametocytocidal for P. falciparum. Again, this may imply a different MoA for some 4-aminoquinoline compounds or perhaps multiple modes of action for certain compounds. These findings re-emphasize the strength of looking at assay data in a wider context in Open Source drug discovery.
Data from one-hundred-nineteen MoA assays for compounds from the Malaria Box are included, identifying potential targets for 135 of them (S1 Table and S1 Methods and Results). The MoA assay data are presented in Column M of S1 Table, and further information about the screens and their results are given in S1 Methods and Results. These screens included biochemical screens for enzyme inhibition, protein-protein interactions, behavior by altered yeast or malaria organisms, and a variety of other screens. Some associations are strong and have been followed up with additional experimentation (e.g. MMV008138 and its target Pf-IspD [13–15]), but most target associations are still tentative. Indeed, some listed MoA activities occur only at higher concentrations than activity in cell-based screens and therefore are unlikely to explain that compound’s activity against a pathogen or tumor cell. In addition, many MoAs have been inferred for malaria, but are less likely to apply to the diverse groups of organisms screened with the Malaria Box compounds.
Surface plasmon resonance (SPR) was used to identify nine compounds which inhibit four sets of protein-protein interactions (PPI), without overlap between sets (S1 Methods and Results), suggesting that molecules were identified that specifically target these protein-protein interfaces. Compounds inhibiting P. falciparum (autophagy-related proteins) Atg8-Atg3 PPI were MMV007907, MMV001246 and MMV665909 (S1 Table, column M). They had a pronounced effect on all stages of gametocyte development, which supports the idea of PfAtg8-Atg3 being involved in remodeling and vesicular trafficking in gametocyte development. Six compounds inhibited in vitro translation in P. falciparum lysates by more than 60% at a concentration of 1 μM (S1 Table, column L; [16]). One of these protein translation-inhibiting compounds, MMV007907, is interesting in that it had activity against both liver and gametocyte stages as well as a broad range of other pathogens, and has low toxicity to human cell lines. Twenty-six compounds either inhibited the mitochondrial electron transport chain (bc1, 11 compounds) or DHODH (15 compounds). Since both the bc1 and DHODH pathways converge on pyrimidine biosynthesis, it is interesting that almost all bc1 inhibitors had anti-liver stage and anti-male gametocyte activity, while the anti-male gametocyte property was generally lacking in most DHODH inhibitors [17–19].
PfATP4 is a P. falciparum plasma membrane protein with genetic variants that confer resistance to several new clinical and preclinical antimalarials [20–24]. PfATP4 has been proposed to function as a Na+:H+ pump, effluxing Na+ from (and importing H+ into) the malaria parasite [21]. Parasites exposed to 28 MMV Malaria Box compounds have shown ion-homeostasis changes similar to those observed with likely PfATP4 inhibitors (indicated in column K, S1 Table) [25], and thus are inferred to be PfATP4 inhibitors. Analysis of the 281 assays’ results with these compounds, reported here, allows detailed conclusions about the potential effects of ATP4 inhibition in Plasmodium as well as other organisms. From the Malaria Box data summarized here, it is evident that the 28 PfATP4-associated hits tended to be inactive against the variety of non-Apicomplexan protozoa, helminths, insects, yeast and bacteria that were tested. An exception was Trypanosoma cruzi, that was growth-inhibited by almost 40% of the PfATP4 inhibitors (11/28), compared to an overall hit rate of 20%. It should be noted that the non-Plasmodium Apicomplexan parasites against which the majority of the compounds were tested–Cryptosporidium parvum, Toxoplasma gondii, Theileria equi and three species of Babesia–were not, in general, particularly susceptible to the PfATP4-associated hits. There is not, to our knowledge, any evidence that the other Apicomplexan parasites against which the Malaria Box was tested are exposed to a high-Na+ environment within their host cells, and this may explain the lower sensitivity to inhibition of a Na+ efflux mechanism. In contrast, infection of an erythrocyte by Plasmodium is followed by an increase in the Na+ concentration in the erythrocyte cytosol as a result of the induction of broad-specificity (Na+-permeable) ‘New Permeability Pathways’ in the host erythrocyte membrane [26–28]. This suggests that perturbation of Na+ efflux through inhibition of PfATP4 is uniquely, highly detrimental to intra-erythrocytic malaria parasites.
There is prior evidence that PfATP4-associated compounds are active against gametocyte stages of P. falciparum [5, 22–24, 29–32]. Twenty-five of the 28 PfATP4-associated hits (89%) caused some inhibition of male gamete formation at 1 μM (i.e. had positive % inhibition values; S1 Table). It should be noted, however, that approximately half of the PfATP4-associated hits have IC50 values for the killing of asexual parasites that are similar to or higher than the 1 μM concentration used in the gamete formation assay. Only 65% of the PfATP4 non-hits tested had positive values for inhibition of male gamete formation at 1 μM. An increase in extracellular pH is known to trigger the exflagellation of male P. falciparum gametes, raising the possibility that an increase in intracellular pH in male gametocytes or gametes, resulting from PfATP4 inhibition, triggers premature exflagellation, leading to parasite death. Thus, it is possible that an increase in intracellular pH in male gametocytes or gametes resulting from PfATP4 inhibition triggers premature exflagellation leading to their death.
Malaria box compounds were also screened against asexual stages using metabolomic and chemogenomic profiling (Fig 2). Using metabolomic profiling to examine the metabolic responses to the 80 compounds in plate A, six of seven compounds believed to target PfATP4 [25] showed a distinct metabolic response characterized by an accumulation of dNTPs, and a decrease in hemoglobin-derived peptides (Fig 2A, S2 Table). Twenty-one compounds clustered with atovaquone, an inhibitor of the bc1 complex of the electron transport chain, exhibiting an atovaquone-like signature characterized by the dysregulation of pyrimidine synthesis. Of these 21 atovaquone-like compounds, 17 were also identified by other groups as targeting the electron transport chain or pyrimidine synthesis. For chemogenomic profiling, a collection of 35 P. falciparum single insertion piggyBac [33] mutants were profiled with 53 MMV compounds and three artemisinin (ART) compounds [Artesunate (AS), Artelinic acid (AL) and Artemether (AM)] for changes in IC50 relative to the wild-type parent NF54 (Fig 2B, S3 Table, S4 Table). Five Malaria Box compounds (MMV006087, MMV006427, MMV020492, MMV665876 and MMV396797) were identified as having similar drug-drug chemogenomic profiles to the ART-sensitivity cluster (Fig 2B). These compounds may be rapid killers, like artemisinin, and should be explored further for confirmation, and whether they can overcome artemisinin-resistance for ring-stage killing.
Four groups carried out screens on S. cerevisiae strains engineered to help elucidate the MoA of test compounds. One screen established that 35 Malaria Box compounds were active on a multiple ABC-transporter deficient strain (also known as the ‘monster strain’) S. cerevisiae [34]. Since yeasts are generally resistant to compound inhibition due to transporters, this monster strain can now be analyzed for MoA of inhibition by these 35 compounds. A second study measured selective growth inhibition of S. cerevisiae using different carbon sources. Growth was measured in three different growth media: rich or minimal media using dextrose as a carbon source, or minimal media using ethanol and glycerol as carbon sources. Compounds affecting growth in a media-specific manner may represent inhibitors of key metabolic pathways. A third group used a yeast strain expressing the Pf phosphoethanolamine methyltransferase (PfPMT) to screen for phosphocholine (PC) synthesis inhibitors. This screen relies on the incapability of this yeast strain to synthesize PC in the absence of exogenous choline, and thus depends on the malaria PfPMT for survival. Screening the Malaria Box compounds, and a variety of controls including wild-type PMT and choline supplemented media, led to the identification of MMV007384, MMV007041 MMV396736, MMV396723, MMV000304, MMV000570, MMV000704, MMV666071, MMV000445, MMV667491, and MMV666080 as possible PfPMT inhibitors. Finally, a fourth group screened S. cerevisiae grown either on ethanol-containing media requiring respiration or glucose-fermentative media not requiring respiration, and identified 12 compounds that gave superior inhibition on ethanol media suggesting that these compounds inhibit a respiratory target. Seven of these were not associated with any other targets; the others were potential inhibitors of DHODH (3, 49), bc1, and IspD.
The Malaria Box was screened against 16 additional protozoa, all of which are of medical or veterinary significance. Compounds with activity against three or more protozoa were usually toxic for the zebrafish or non-cancer mammalian cell lines, underlining the need to limit the concentrations used in assays, to avoid meaningless positives. Table 2 lists compounds with activity against protozoa that were nontoxic to zebrafish and most mammalian cells. In the Cryptosporidium parvum assay there were numerous active compounds, but none were completely devoid of toxicity for zebrafish and mammalian cell lines. MMV665917 had a >20-fold Selectivity Index (SI) for C. parvum over mammalian cells. Trypanosoma cruzi actives were non-overlapping between groups, and are listed separately, but T. brucei actives overlapped extensively with other screens and are presented together. There were seven non-toxic hits that were active against extracellular amastigotes of Leishmania infantum, but no non-toxic compounds were active on intracellular macrophage growth of L. infantum. There were five non-toxic Malaria Box compounds active against T. gondii (MMV666095, MMV007363, MMV007791, MMV007881 and MMV006704). Many of the compounds that were active on Neospora caninum raised no toxicity flags on the accompanying host cell fibroblast screen, but many were toxic at 10 μM or below for mammalian cells and zebrafish. The remaining non-toxic N. caninum actives that bear further investigation include: MMV019670, MMV000911 and MMV006309. Most compounds active against Entamoeba histolytica, Naegleria fowleri, or exflagellation of Chromera velia were toxic. An exception was MMV665979, an outstanding hit for Naegleria fowleri, with limited toxicity elsewhere in the dataset. With respect to screening Babesia and Theileria, ten novel anti-Babesia and anti-Theileria hits with nanomolar IC50s were identified: MMV666093, MMV396794, MMV006706, MMV665941, MMV085203, MMV396693, MMV006787, MMV073843, MMV007092 and MMV665875. The most interesting hits were MMV396693, MMV073843, MMV666093, MMV665875 and MMV006706 with mean SIs greater than 230 and IC50s ranging from 43 to 750 nM for both bovine Babesia and equine Babesia and Theileria parasites. Additionally, 64, 45 and 49 Malaria Box compounds exhibited IC50s lower than those of diminazene aceturate (the most widely used antibabesial drug) against the in vitro growth of B. bovis, B. bigemina and T. equi, respectively.
In vitro screening of Open Access Malaria Box compounds against Babesia bovis, B. bigemina, Theileria equi and B. caballi has led to the discovery of 10 novel potent anti-babesial hits exhibiting submicromolar potency against both bovine Babesia and equine Babesia and Theileria. In vitro follow up of the many of the hits identified in this study for B. bovis, B. bigemina, B. caballi, and T. equi parasites, revealed IC50s lower than that obtained with the previously described drug-leads luteolin, pyronaridine, nimbolide, gedunin and enoxacin [35]. The ten potent hits for bovine Babesia and equine Babesia and Theileria identified in this study exhibited IC50s lower than that obtained with the apicoplast-targeting antibacterials (ciprofloxacin, thiostrepton, and rifampin), miltefosine, fusidic acid or allicin [36–39].
Many Malaria Box compounds were active on helminths at 10 μM, but most of these were also toxic for mammalian cells or zebrafish. The remaining non-toxic compounds had activity against Brugia malayi (lymphatic filariasis) and Ancylostoma ceylanicum (hookworm; Table 1). But no non-toxic compounds were found with consistent activity against Schistosoma mansoni, Strongyloides stercoralis, Trichuris muris, Haemonchus contortus, or Onchocerca linenalis. There remains the possibility that some of the toxic hits against these species can be addressed by medicinal chemistry.
With respect to activity against mycobacteria and bacteria, although every screen delivered actives, the majority were again discarded because of a toxicity signal against zebrafish and/or mammalian cells. The exceptions were non-toxic Malaria Box compounds that were active against Wolbachia (Table 1). Wolbachia bacteria are targeted as anti-filarials in order to deprive nematodes causing river blindness and elephantiasis from essential nutrients provided by this bacterium [40].
The US National Cancer Institute has screened 59 human tumor cell lines (‘NCI60’) against the Malaria Box compounds at 10 μM (S1 Table and S1 Methods and Results). Among the 133 compounds further evaluated for dose-responses, and the ten of these then tested in confirmatory assays (S1 Text), MMV007384 was selected for potency and focused activity against colon cancer cells, and has been advanced to an in vivo proof-of-concept experiment.
Academic drug discovery is highly fragmented. Many biology groups, especially those in disease-endemic countries, excel in developing highly disease-relevant pathogen models suitable for low- to medium-high throughput screening, but suffer from lack of access to innovative compounds. If they do have access to compounds, then they may fail to share the results, or lack drug development skills. The Malaria Box Project demonstrates how an open source approach allows effective data sharing: this publication serves as much to share the data among the 180+ co-authors as with the wider scientific community. By publishing in concert this ensures early publication and also sharing of ideas and expertise in drug discovery. New insights and series have been obtained for malaria (nine pan-stage active molecules which had not been previously prioritized). Moreover, screening against pathogens for additional neglected diseases has been catalyzed and hits found. The sharing of data from safety screens flags compounds that probably work through a general toxicity mechanism, and those compounds can be down-prioritized at an early stage. This is key to prioritizing compounds for medicinal chemistry, since the paucity of good starting points against some parasites has encouraged groups to screen at what may be inappropriately high drug concentrations. Another advantage of having a standardized, publically available library and dataset is that this allows benchmarking assay sensitivity, setting compound concentrations for expanded screens and deciding on acceptable hit criteria [41].
We saw some discrepancies in the values obtained for the same compounds in similar assays that were carried out by multiple groups, such as activity against asexual or gametocyte forms of P. falciparum, Trypanosoma spp., and mammalian cells. In this sense, compounds that were positive in more than one assay would clearly be more likely to represent a true positive than compounds that were positive in only one screen. Some of these apparent discrepancies were probably due to variations in the techniques used for the screens. For instance, many methods used to measure gametocytocidal activity measure a specific metabolic activity. Because the metabolism will be affected by many factors that will lead to differences in output, including media composition (albumax versus serum), how old the media used was, purity of the gametocytes (how much asexual contamination and cell debris is present). In addition, the tested compounds varied widely in their propensity to bind to protein in the assay medium, and large differences in the protein content in two assays could lead to differences in unbound compound. Only the free compound would likely be available for activity in biological assays. Some assays had extensive follow up, and if a compound was tested and activity confirmed with a dose-response, it is more likely to be a true positive than a compound flagged as positive from a single screening run. This complex dataset highlights the need to consider integrating more standardized criteria, such as similar (free) compound concentrations, assay media, or compound exposure duration, into future screening initiatives of this nature. This could potentially reduce inter-assay differences, and facilitate more direct data comparison across the different platforms. However it is clear in the case of gametocyte screens, that different assays that interrogate different biological processes do not necessarily achieve the same result for a given compound, even when the assay conditions have been standardized [42]. And trying to standardize assays may be counterproductive with the goal of convincing multiple groups to run their assays on a given set of compounds.
The MoAs associated with compounds (S1 Table, Column M) vary from very strong associations such as chemical-genetic evidence, to relatively weak associations, such as activity in a single biochemical screen at relatively high compound concentration. Thus most of the associations should not be taken as definitive MoA of the compounds for their biological activities. All associations were presented because not only could they be hypothesis-building for the discovery of a compound’s disease-relevant MoA, but also because the Malaria Box compounds now represent a rich source of bioactive compound tools.
With its outcomes continually evolving, the Malaria Box has already made an impact by stimulating medicinal chemistry for many diseases. We are aware of such new medicinal chemistry programs against pathogens such as Plasmodium [43–45], Babesia, Toxoplasma [46], Trypanosoma [47–49], Cryptosporidium [31], Schistosoma [50], filaria, Echinococcus, helminths, bacteria, cancer and other diseases [30]. Ensuring that data becomes freely available is a challenge, and this paper represents the first such summary of over 290 screens against the compound collection, highlighting new activities and new MoAs. For the future, three goals are important. First, to track these compound series to ascertain whether any of these hits do become leads of drug development candidates. Second, data must be rapidly published, even with follow-up incomplete. Finally and most importantly, this model can be taken further. A second collection of 400 compounds, the Pathogen Box, (www.PathogenBox.org) based on compounds known to be active in phenotypic screens against an expanded set of pathogens responsible for neglected and tropical diseases has now become available from the Medicines for Malaria Venture. It is hoped this can be the start of equally fruitful collaborative networks.
See S1 Methods and Results for further details.
The Malaria Box is a set of 400 compounds that were previously shown to be active against asexual stages of P. falciparum in vitro. The process for Malaria Box compound selection was published previously [5], with 200 drug-like compounds as starting points for oral drug discovery and development and 200 diverse probe-like compounds for use as bioactive tools research. The selection was made to represent the broadest cross-section of structural diversity and, in the case of the drug-like compounds, properties commensurate with excellent oral absorption and the minimum presence of known toxicophores. One limiting factor was that compounds had to be commercially available; this limited the chemical space displayed in the original set of 20,000 malaria bioactives.
The Malaria box was shipped to 193 different research groups in 29 different countries as frozen 96-well plates with the compounds dissolved at 10 mM in 20 μl DMSO (dimethylsulphoxide). Two years after shipping the first Malaria Box, the 193 groups were re-contacted and asked if they wanted to participate in a group publication disseminating and comparing the results from the Malaria Box screens. Forty-seven of these groups did not reply to our multiple requests. Fifty-nine groups had not yet initiated screening, but 26 of these had only received the Malaria Box in the preceding three months. Thirty-one groups had publications in preparation and 39 papers have already been published [5, 14, 16, 25, 30–32, 42–46, 48–75]. Fifty-five groups agreed to contribute data and participate in this paper and provided data from 291 assays.
The compounds were then screened in biochemical and biological screens as documented in detail in S1 Methods and Results. More detailed methods are provided for screens presented in this paper than for those whose results have already published. In addition, S1 Methods and Results provides data for both positive and negative controls obtained for each assay. In most assays, a single-concentration screen was run first and bioactives were identified. Some work was stopped after the primary screen, but most groups went on to perform confirmatory assays, and many provided hit concentrations that achieve 50% activity (S1 Table). The assays included a variety of cell-based pathogen screens covering multiple taxonomic groups, including Plasmodium (multiple life-stages), other protozoa, bacteria, mycobacteria, HIV, and also multicellular-organism screens such as helminths and a mosquito (See Fig 1 and S1 Table).
|
10.1371/journal.pntd.0006636 | The current epidemiological status of urogenital schistosomiasis among primary school pupils in Katsina State, Nigeria: An imperative for a scale up of water and sanitation initiative and mass administration of medicines with Praziquantel | Human schistosomiasis, a debilitating and chronic disease, is among a set of 17 neglected tropical infectious diseases of poverty that is currently posing a threat to the wellbeing of 2 billion people in the world. The SHAWN/WASH and MAM programmes in the study area require epidemiological data to enhance their effectiveness. We therefore embarked on this cross-sectional study with the aim of investigating the prevalence, intensity and risk factors of urogenital schistosomiasis.
Interviewed 484 respondents produced terminal urine samples (between 10.00h – 14.00h) which were analyzed with Medi ─Test Combi 10 and centrifuged at 400 r.p.m for 4 minutes using C2 series Centurion Scientific Centrifuge. Eggs of S. haematobium were identified with their terminal spines using Motic Binocular Microscope. Data were analyzed with Epi Info 7. In this study, the overall prevalence and arithmetic mean intensity of the infection were 8.68% (6.39─ 11.64) and 80.09 (30.92─129.28) eggs per 10ml of urine respectively. Urogenital schistosomiasis was significantly associated with knowledge about the snail host (χ2 = 4.23; P = 0.0398); water contact activities (χ2 = 25.788; P = 0.0001), gender (χ2 = 16.722; P = 0.0001); age (χ2 = 9.589; P = 0.0019); economic status of school attended (χ2 = 4.869; P = 0.0273); residence distance from open water sources (χ2 = 10.546; P = 0.0012); mothers’ occupational (χ2 = 6.081; P = 0.0137) and educational status (χ2 = 4.139; P = 0.0419).
The overall prevalence obtained in this survey shows that the study area was at a low-risk degree of endemicity for urogenital schistosomiasis. Beneath this is a subtle, latent and deadly morbidity-inducing heavy mean intensity of infection, calling for urgent implementation of WHO recommendation that MAM with PZQ be carried out twice for School-Age Children (enrolled or not enrolled) during their primary schooling age (once each at the point of admission and graduation). The criteria for classifying endemic areas for schistosomiasis should also be reviewed to capture the magnitude of mean intensity of infection rather than prevalence only as this may underplay its epidemiological severity.
| In 1851, human schistosomiasis was discovered in Egypt by a German surgeon named Theodor Bilharz. Therefore, it is alternatively called ‘bilharziasis’. Being a disease that is closely associated with poverty in the tropics, chiefly the sub-Saharan Africa, Urinary Schistosomiasis is transmitted to humans who source for non-potable water in open and unwholesome water bodies infested with the infective stage of schistosomes which are excreted by Bulinus globosus, a freshwater snail and suitable intermediate host. Before now, many endemic foci of the disease have been discovered with many unidentified yet. Meanwhile, the initiative on Sanitation, Hygiene and Water in Nigeria (SHAWN) in the study area is suffering a major setback due to poor data availability despite being due to round up her activities by early 2018. Consequently, we embarked on this survey to determine: the extent at which Urinary Schistosomiasis is prevalent, its severity, and factors that were responsible for its transmission. We found out that the study area was at a low-moderate risk of endemicity for the disease. However, mean intensity of infection was too high. Based on these, a need for Praziquantel is indisputable. Criteria for categorizing areas endemic for the disease should be revised to embrace the novelty of capturing mean intensity of infection.
| Human schistosomiasis, though less fatal, but debilitating and chronic in nature, is among a set of 17 neglected tropical infectious diseases of poverty that is currently posing a threat to the wellbeing of 2 billion people in the world [1]. The aetiological agent of urogenital schistosomiasis is the infective stage (cercaria) of Schistosoma haematobium, a digenetic trematode plathyhelminth whose intermediate hosts are some species of gastropod snails in the genus Bulinus [2,3].
Maturity of S. haematobium worms takes place at the portal vein of a human host. Subsequently, each male worm carries a female partner in its genaecophoric canal to the veins of the pelvis where the latter lays a very large number of eggs equipped with terminal spines. Consequently, they rupture the endothelial linings of the vesical and urethral walls, leading to granulomatous inflammation, ulceration, pseudopolyposis, and haematuria [4–6]. As a result of its associated morbidities, an estimate of 280,000 people die annually from human schistosomiasis [7].
Currently, sub-Saharan Africa (SSA) is regarded as the poorest part of the world with 73% of its population living on USD 2 per day [8]. This index of abject poverty is of no doubt a key player in the reported astounding 112 million cases of urogenital schistosomiasis (less than a decade ago) out of the 203.15 million (85% of global figure) active cases of human schistosomiasis in the region [9, 10].
Schistosomiasis was reportedly transmitted into the Northern part of Nigeria by invading Fulani herdsmen from upper Nile valley in the ancient times [11]. However, documented evidence was not obtained until 1881when a German immigrant, Gustav Nachtigal, commented on a high incidence of haematuria in Borno, a northeastern State of Nigeria close to Lake Chad [12]. In 1963, the first map showing the distribution of schistosomiasis in Nigeria was produced by Professor Cowper [13]. Currently, Nigeria has 29 million cases of schistosomiasis, the highest in SSA, with 101.3 million people living in close proximity to various endemic foci [14].
The endemicity of schistosomiasis was ascertained in Katsina State as far back as 1963 when prevalences of 95% and above 90% were recorded in Katsina and Kankia Local Government Areas [12].
In 2010, UNICEF signed a contribution agreement with the Government of United Kingdom and Northern Ireland to implement Sanitation, Hygiene and Water in Nigeria (SHAWN) between March 2014 and November 2018. Among others, the initiative aims at providing Water, Sanitation and Hygiene (WASH) facilities for 3,500 schools and 1, 200 health centers. The benefit of this initiative is that pupils would have the capacity to promote personal, domestic, and environmental cleanliness. Invariably, schistosomiasis control programme will be enhanced because the risk of contracting the disease will be drastically reduced to the barest minimum. However, only 11 out of 34 LGAs in Katsina State were supported by the SHAWN initiative as at February 2015 [15].
The SHAWN/ WASH initiative vis a vis Mass Administration of Medicines (MAM) with Praziquantel (PZQ) in Katsina State require epidemiological data to achieve their aims. Nonetheless, there is limited information in the literature about the current epidemiological status of urogenital schistosomiasis among Primary School pupils in Katsina State. Therefore, we embarked on this study with the aim of investigating the prevalence, intensity and risk factors of urogenital schistosomiasis in the study area.
Dutsin-Ma Local Government Area (LGA) [12.45°N and 7.50° E] is located in the eastern part of Katsina State in North western part of Nigeria. It covers a total surface area of 527 km2 and is inhabited by a population of 169,829 people as at 2006 National Census [16]. It is predominantly inhabited by Hausa-Fulani tribes whose major occupations are trading, farming, and animal rearing. With a characteristic tropical continental climate type, Dutsin-Ma LGA (see Fig 1) has a temperature range of 29─31°C and mean annual rainfall of 700mm which commences in May and ends by September [17]. As a result of these, its soil is typically sandy and shallow, supporting only cereals like sorghum, millet and grasses. However, short, drought-resistant and scattered trees are present. This vegetation type makes the study area most suitable for pastoralists.
Dutsin-Ma is a rural area characterized with a poor network of healthcare facilities. Generally, potable water is not readily available to average inhabitant of the area. Residents depend on water from unwholesome sources like dams, lakes, and ponds. On the average, a Mono pump, where available, serves over 30 households.
In the study area, MAM with Praziquantel (PZQ) was last carried out in December, 2014.
In this study, a total number of 491 primary school pupils with age range of 6─15 years were recruited from 5 communities in Dutsin-Ma Local Government Area of Katsina State. The mean age ± standard deviation (S.D) was 9.57 ± 2.14 years and 40.29% were females. However, 7 of these pupils did not submit urine samples following interview with questionnaires (see Fig 2). Of noteworthy are the facts that 88.64% had lived in the study area for more than 4 years; 60.33% lived more than 100m away from potentially infested open water sources; only 5.37% had a previous access to Praziquantel while 13.08% and 50.81% reportedly experienced macrohaematuria and dysuria respectively.
Level of illiteracy among mothers of the respondents doubled that of their fathers while access to tertiary education in their fathers almost doubled mothers’. The most common paternal occupational category was “brown collar jobs while mothers of respondents were majorly housewives (see Table 1).
The results about knowledge, attitudes and health-seeking practices of the respondents about urogenital schistosomiasis are shown in Table 2. It was found that 426 (88.02%) and 405 (83.68%), a majority had the awareness of urogenital schistosomiasis (locally called tsagiyya or sarajini) and the snail intermediate host respectively. Meanwhile, the prevalence of the disease was found to be significantly associated with knowledge about the snail host (χ2 = 4.23; P = 0.0398). Of the total number of respondents, 46 (9.50%) had visited health facilities, 18 (3.72%) opted for traditional healing, while 37 (7.64%) received no medication for urogenital schistosomiasis.
The results displayed in Table 3 showed that urogenital schistosomiasis was significantly associated with water contact activities for domestic, recreational and farming purposes (χ2 = 25.788; P = 0.0001). Respondents who indulged in the habit of swimming for recreation recorded the highest prevalence of 24.59% (17.20─33.20) while being 3 times more likely to be infected with urogenital schistosomiasis compared with those who only had contact with closed, non-infested water sources [COR (95% CI): 3.39(2.01─5.72)]. Those who combined swimming with play in shallow waters had the second highest prevalence (see Fig 3) of 23.36% (15.70─32.50) and as well 3 times more likely to be infected [COR (95% CI): 3.18(1.82─5.49)]. Nonetheless, those who combined open and closed water sources, against expectation, recorded the highest mean intensity of the infection [27.83(15.78─49.09) eggs/ 10ml of urine sample] followed by respondents with experience of water contact by irrigation [26.92(12.92─56.08) eggs/ 10ml of urine sample].
Results of Table 4 convincingly show that gender (χ2 = 16.722; P = 0.0001); age (χ2 = 9.589; P = 0.0019); economic status of school attended (χ2 = 4.869; P = 0.0273); residence distance from open water sources (χ2 = 10.546; P = 0.0012); mothers’ occupational (χ2 = 6.081; P = 0.0137) and educational status (χ2 = 4.139; P = 0.0419) were additional determinant factors of urogenital schistosomiasis in the study area.
In this study, the overall Prevalence and Arithmetic Mean Intensity of the infection were 8.68% (6.39─ 11.64) and 80.09 (30.92─129.28) eggs per 10ml of urine. Of the 42 respondents infected, 27 (64.29%) belonged to the ‘light intensity’ status while 15 (35.71%) fell to the ‘heavy intensity’ category.
Highest values of prevalence were obtained in: males: 90.48% (77.38─ 97.34); age group 11─15: 54.76% (38.70─70.20); public schools: 88.09% (74.40─96.00); respondents whose residence distance are < 100m from open water sources: 64.29% (48.03─78.45); those with brown collar jobs as fathers’ occupation: 61.90% (45.64─76.43); respondents whose mothers were housewives: 59.52% (43.28─74.37); and respondents whose fathers: 50.00% (34.19─ 65.81) and mothers: 54.76% (38.67─70.15) attained secondary level of education.
Similarly, highest mean intensities of urogenital schistosomiasis were recorded in age group 11─15: 36.37(18.25─72.47) eggs per 10ml (see Fig 4); respondents whose mothers were housewives: 31.43 (16.30─60.61) eggs per 10ml; and respondents whose fathers attained secondary level of education: 41.34 (20.99─81.39) eggs per 10ml. However, contrary to expectation, highest mean intensities of the infection were recorded in females:
33.01(2.88─378.16) eggs per 10ml; private schools: 76.19(23.28─249.34) eggs per 10ml; respondents whose residence distance were ˃ 100m from open water sources: 28.97(12.23─68.62) eggs per 10ml (see Fig 4); those with white collar jobs as fathers’ occupation: 42.08 (15.55─113.85) eggs per 10ml and respondents whose mothers attained tertiary level of education: 255.15 (0.00─9.46e+07) eggs per 10ml.
Table 5 presents data on the strength of association between urogenital schistosomiasis and determinant factors that were found to be significantly associated with it. Male respondents were discovered to be 7 times [AOR (95% CI): 7.23 (2.54–20.60)] more likely to be infected with the cercariae of S. haematobium compared to their female counterparts. Age group 11–15 years [AOR (95% CI): 2.81 (1.48–5.34)], respondents from public school sector [AOR (95% CI): 2.82(1.15─8.25)] and those living < 100m from open, infested water sources [AOR (95% CI): 3.02(1.56─5.85)] were found to be about 3 times more likely to be infected compared to others in their respective categories.
In Mothers’ occupation category, respondent with “other brown collar jobs” [AOR (95% CI): 8.66(1.13─66.49)] and “housewives” as mothers’ occupation [AOR (95% CI): 8.85(1.18─66.33)] were about 9 times more likely to be infected compared to those whose with “white collar jobs” as mothers’ occupation. It is puzzling that respondents whose mothers attained secondary level of education were 6 times [AOR (95% CI): 6.09(1.40─26.45)] more likely to be infected compared to those whose mothers attained a tertiary education level.
With respect to the risk of infection with S. haematobium, males who lived in the study area were found to be 6 times more at risk than females [RR (95% CI): 6.41(2.33─17.67)]. In similarity to the odd of infection, age group 11–15 years [RR (95% CI): 2.55(1.43─4.53)], respondents from public school sector [RR (95% CI): 2.63(1.06─6.55)] and those living < 100m from open, infested water sources [RR (95% CI): 2.74(1.49─5.01)] were found to be about 3 times more at risk of being infected compared to others in their respective groups. Respondent with “other brown collar jobs” [RR (95% CI): 7.90(1.07─58.51)] and “housewives” [RR (95% CI): 8.06(1.11─58.54)] as mothers’ occupational categories were about 8 times more at risk of being infected compared to those with “white collar jobs” as mothers’ occupation. The results of our multivariate analysis further showed that subjects whose mothers attained secondary level of education were 5 times [RR (95% CI): 5.48(1.32─22.72)] more at risk of being infected compared to those whose mothers attained a tertiary level of education. Finally, subjects who suffered macrohaematuria were found to be about 2 times [RR (95% CI): 1.72(1.04─2.85)] more at risk of contracting urogenital schoistosomiasis compared to those who experienced dysuria.
In recent years, the World Health Organization recommended five Public Health interventions to speed up the prevention, control, elimination and eradication of NTDs. Our findings will, however, be discussed in the context of three of such interventions that are considered more related: innovative and intensified disease management, provision of safe water, sanitation and hygiene and preventive chemotherapy.
The concept of IDM was devised less than two decades ago to act as a “catalyst” in the control strategies of diseases that are proving difficult to eliminate although effective tools are available. The fundamental principle of this intervention is to ensure that barriers to control strategies are destroyed using the level of awareness about the disease in the population at risk, knowledge of the current epidemiological status of the disease, and the availability and efficacy of available medicines, vaccines and diagnostic tools among others [22].
In the context of this study, knowledge about haematuria and Bulinus spp. was used as a strong indicator of awareness about urogenital schistosomiasis. Our findings showed that 426 (88.02%) of the school children surveyed had a knowledge of the disease through the local name for haematuria, tsagiyya, while 405 (83.68%), had the awareness of the snail intermediate host. The latter was significantly associated with the disease (see Table 2). The logicality of these findings cannot be subjected to a doubt because infection with the cercariae of S. haematobium is practically impossible without a transmission cycle involving the suitable snail intermediate host. The strong association clearly suggests that infected respondents most likely came across the snail intermediate host while swimming, bathing, fishing, and/or fetching water in infested streams, rivers, ponds, lakes, and so on. Hunting snails in aquatic terrains might also have been responsible for this. In summary, the cause of this association could be linked to a significant contact (with the snail host versus its aquatic medium) that was enough to encourage transmission of the disease. This knowledge about the snail intermediate host is a strong boost because SAC could be integrated into future control programmes for the disease with the aim of targeting the snails with molluscicides. Besides, physical control measure by handpicking could be enhanced.
Our results on the knowledge of the current epidemiological status of urogenital schistosomiasis showed that the prevalence and arithmetic mean intensity of the disease were 8.68% (6.39─11.64) and 80.09 (30.92─129.28) eggs per 10ml of urine sample respectively (see Table 4). It is obvious that the study area is hypo-endemic (prevalence < 20%) for urogenital schistosomiasis [23] while, based on the arithmetic mean of 80.09 (30.92─129.28) eggs per 10ml of urine sample, an average infected respondent had a heavy intensity of infection. As expected, male respondents had higher prevalence (90.48% Vs 9.52%) and mean intensity of infection (81.87 Vs 63.25).
We also showed through the adjusted Odds Ratio that for every female respondent infected, seven males were infected. Previous studies have reported higher odds of infection with S. haematobium in male students [24–26].
Meanwhile, previous findings have shown that increase in intensity of infection with urogenital schistosomiasis increased the risk of Female Genital Schistosomiasis (FGS) evidenced by cervical inflammation, intraepithelial neoplasia, postcoital bleeding and genital ulceration [27]. This has been identified as a risk factor in Human Immuno-deficient Virus (HIV) transmission to women [28]. In males, genital schistosomiasis has been reported to induce pathology of the seminal vesicles and the prostate with irreversible long-term consequences which may culminate in bladder cancer, urethral fibrosis and hydronephrosis [28]. Studies have also shown that such high infection intensities make eggs of schistosomes to be trapped in tissues of the liver, spleen and peritonium with severe and complex pathological consequences which may degenerate to late-stage sequelae. Besides, school-age children suffer academic setbacks as well as Iron status deterioration. Organs damaged may not recover until at least six months after cure with Praziquantel [21]. The Risk Ratio obtained from this study revealed that male inhabitants of the study area were six times more at risk of infection with urogenital schistosomiasis (see Table 5). This could be explained by the already known fact that boys are more involved in water contact activities like swimming, fishing, playing in shallow water, irrigation, and making of bricks to build mud houses than girls.
Currently, there are no vaccines against S. haematobium. However, there exist effective diagnostic tools and chemotherapeutic intervention with PZQ, the global drug of choice [21, 29]. Apart from the gold diagnostic standard of microscopy [29–31], a rapid, non-invasive dipstick called Medi-Test, though expensive but readily available in capital cities of Katsina State, was used for the survey as stated earlier. The study area is known for some spurious sentiments about MAM and Public Health survey. Consequently, chemotherapeutic interventions are sometimes seen as attempt to render females infertile. Sometimes, physical attacks with weapons are launched at researchers/ healthcare workers [24]. This probably informed the low PZQ coverage of 5.37% recorded in this survey. As a matter of fact, the record we had on our school-based questionnaires showed that three of the five schools surveyed did not benefit from the previous MAM carried out in 2014. One of the other two who benefitted claimed to have distributed it but response from the pupils during interview negated such claim. It was also likely that it was not done due to the fear of stiff resistance from the mostly uneducated misinformed and indoctrinated parents.
The juxtaposition of this peculiar challenge with the fact that Nigeria ranks number three globally among the countries in most urgent need of chemotherapeutic intervention [22] shows there is still a long way to go to achieving the roadmap for the control of schistosomiasis.
Water, sanitation and hygiene (WASH) have been identified as part of the key factors determining the state of health in any epidemiological settings [22]. The implication of this is that poor or no access to potable water supply tend to contribute to high prevalence and mean intensity of infection with urogenital schistosomiasis. Our findings in this present study proved this to be true. It showed that respondents who had experience of swimming recorded the highest prevalence, followed by those who combined both the recreational activities of swimming and playing in shallow water. Those who were involved in farm-related activities of fishing had the third placed prevalence. Respondents who combined open, potentially infested with closed water sources recorded the highest mean intensity of the infection, being closely trailed by those who came into contact with open, unwholesome water through the activity of irrigation. Generally, respondents whole came into contact with open water sources like rivers, ponds, streams and lakes were two or three times more likely to be infected with the infective larval form of S haematobium (see Table 3) compared to their counterparts who had contact with well, tap water, bore hole and sachet water (closed water sources).
We stated earlier that only 11 out of 34 LGAs in Katsina State were covered by the SHAWN as at February 2015. This report vis a vis the data obtained in this present survey (more importantly; the heavy mean intensities of infection) pointed to the fact that the initiative has a long way to go in view of its deadline of November 2018 given to round up its intervention in the Nigerian State.
Extrapolation of our findings to the whole communities in the study area will amount to painting a wrong epidemiological picture of urogenital schistosomiaisis [23] partly because of its age-sensitivity and focal nature. To corroborate this, previous studies among high school children have identified the study area as being at “moderate-risk” for the disease [24, 25, 32] as opposed to the “low risk” reported in this present study involving primary school pupils.
This study employed urine centrifugation technique for egg concentration due to a lack of polycarbonate filters that would have provided a better picture of mean intensity of infection. Thus, it is likely that the mean intensity of urogenital schistosomiasis in this survey is under-reported here.
The overall prevalence obtained in this survey shows that the study area was at a low-risk degree of endemicity for urogenital schistosomiasis. Beneath this is the subtle, latent and deadly morbidity-inducing heavy mean intensity of infection which encompassed all determinant (risk) factors identified in this study, thus calling for urgent intervention. Based on these premises, we uphold the recommendation of Crompton & World Health Organization (WHO) on the execution of chemotherapeutic intervention with PZQ twice for School-Age Children (enrolled or not enrolled) during their primary schooling age (once each at the point of admission and graduation) in the study area [33]. We deemed it fit that, as in the case of loiasis, an NTD, the criteria for classifying endemic areas for schistosomiasis be reviewed to capture the magnitude of mean intensity of infection rather than prevalence only as this may underplay its epidemiological severity. While concluding that SHAWN/WASH and MAM programmes in the study area are seriously lagging behind as evidenced in our findings, we strongly recommend that the State Government vis a vis Non-Governmental Organizations (NGOs), as a matter of urgency, form a problem-solving alliance to provide water, sanitation and hygiene facilities as well as PZQ with appropriate networking to the grass-root level. Health education should also be carried out at the grass-root level to create awareness about the various factors that predispose people to the infection.
|
10.1371/journal.pgen.1001367 | 14-3-3 Proteins Regulate Exonuclease 1–Dependent Processing of Stalled Replication Forks | Replication fork integrity, which is essential for the maintenance of genome stability, is monitored by checkpoint-mediated phosphorylation events. 14-3-3 proteins are able to bind phosphorylated proteins and were shown to play an undefined role under DNA replication stress. Exonuclease 1 (Exo1) processes stalled replication forks in checkpoint-defective yeast cells. We now identify 14-3-3 proteins as in vivo interaction partners of Exo1, both in yeast and mammalian cells. Yeast 14-3-3–deficient cells fail to induce Mec1–dependent Exo1 hyperphosphorylation and accumulate Exo1–dependent ssDNA gaps at stalled forks, as revealed by electron microscopy. This leads to persistent checkpoint activation and exacerbated recovery defects. Moreover, using DNA bi-dimensional electrophoresis, we show that 14-3-3 proteins promote fork progression under limiting nucleotide concentrations. We propose that 14-3-3 proteins assist in controlling the phosphorylation status of Exo1 and additional unknown targets, promoting fork progression, stability, and restart in response to DNA replication stress.
| Stalling and collapse of DNA replication forks is an important source of genome instability and has been implicated in early steps of carcinogenesis. The maintenance of stable intermediates upon stalled replication requires the coordinated action of a number of proteins and proper inhibitory control of dangerous enzymatic activities. In this study, we uncover an evolutionarily conserved mechanism through which 14-3-3 proteins modulate the checkpoint-mediated phosphorylation of, and in turn limit the activity of, an exonuclease (Exo1) previously implicated in pathological processing of stalled replication forks and other sensitive DNA intermediates. This represents an unprecedented link in the field of DNA repair and genome stability, providing a molecular rationale to the yet undefined role of 14-3-3 proteins in the maintenance of genome integrity after replication stress. In analogy to Exo1, our data suggest that additional factors at replication forks may be subjected to similar regulation, pointing to the 14-3-3 proteins as central components of the checkpoint triggered in response to replication stress.
| DNA lesions can cause stalling and collapse of the replication fork and lead to chromosome breaks, mutations, genome rearrangements and eventually cancer [1]. To prevent this, a replication checkpoint has evolved as surveillance mechanism to control components of the replisome [2] and to allow coordinating replication with cell cycle progression and DNA repair. Maintenance of stable replication intermediates when DNA synthesis is impeded, requires replisome stability and checkpoint-dependent phosphorylation [3]. Although crucial targets for this checkpoint function await identification, nuclease activities are particularly likely to require fine-tuning, to avoid unscheduled DNA processing under DNA replication stress [4].
Exo1 is a Rad2 family DNA repair nuclease able to remove mononucleotides from the 5′ end of the DNA duplex [5] that was originally identified in the Schizosaccharomyces pombe [6] and subsequently in humans [7]. Exo1 is implicated in several DNA repair pathways including mismatch repair, post replication repair, meiotic and mitotic recombination and double strand break repair [8]–[12]. Saccharomyces cerevisiae Exo1 acts redundantly with Rad27 in processing Okazaki fragments during DNA replication [13]. More recently, Exo1 was shown to be recruited to stalled replication forks where it counteracts fork reversal [4]. Human EXO1 activity is controlled by post-translational modifications, with ATR-dependent phosphorylation targeting it to ubiquitin-mediated degradation upon replication fork stalling [14], [15], and ATM-dependent phosphorylation apparently restraining its activity during homologous recombination [16]. Analogously, Mec1-dependent phosphorylation inhibits yeast Exo1 activity at uncapped telomeres [17]. Studies in budding yeast showed that EXO1 deletion suppresses the sensitivity of rad53, but not mec1, mutant cells to agents causing reversible or irreversible stalling of replication forks [18]. Taken together, this evidence indicates that Exo1 activity is tightly controlled under DNA replication stress and DNA damage.
Eukaryotic 14-3-3 are highly conserved proteins that establish phosphorylation-dependent interactions and modulate the functions of proteins involved in processes such as metabolism, protein trafficking, signal transduction, apoptosis and cell-cycle [19]. Seven 14-3-3 isoforms exist in mammalian cells, but only two in yeast. Structural analysis showed that 14-3-3 proteins self-assemble into flexible homo- and hetero-dimers forming a central groove that is able to adapt two extended peptides [20], [21]. This feature confers them the ability to act as adaptors that integrate signals from different pathways [22], [23]. 14-3-3 proteins can also bind cruciform DNA [24] and replication initiation proteins such as Mcm2 and Orc2 [25]. Upon DNA damage and DNA replication stress, 14-3-3 proteins are required for cell cycle restart, suppression of genomic instability and viability [26]. Moreover, 14-3-3 proteins genetically and physically interact with the checkpoint protein Rad53 [27] as well as the acetyltransferases and deacetylases Esa1 and Rpd3 upon replication perturbations [28]. Although these data point to an important role of 14-3-3 during replication stress, the exact mechanism of 14-3-3 action remains unknown.
In this study, we identify 14-3-3 as novel interaction partners of Exo1 and demonstrate that they regulate phosphorylation of the nuclease. We provide evidence for an accumulation of Exo1-dependent ssDNA gaps at stalled forks in yeast 14-3-3 deficient cells and we show that this causes persistent checkpoint activation and recovery defects. We also show that 14-3-3 proteins control progression and stability of replication forks under conditions of limiting nucleotide availability. Taken together, our data demonstrate that 14-3-3 have a crucial role in regulating the function of proteins at stalled forks, among which Exo1 is a key target.
To identify novel interaction partners for human EXO1 we designed a yeast two-hybrid screen with GAL4-bait fusion proteins that contain either full-length EXO1 or ΔN-EXO1 (EXO1366–846), which lacks the entire catalytic domain. Since the former was not expressed (data not shown), we used the latter to screen a blood peripheral cDNA library. Three 14-3-3 proteins were the highest hits (Table S1), with the β- being more represented than the ε- and ζ-isoform. The presence of an established EXO1 binding protein among the hits, MLH1 (Table S1), confirmed the reliability of this screen.
To independently verify these data, we performed co-immunoprecipitation experiments. Given the low abundance of EXO1 in the cell [14], we transiently transfected HEK-293 cells with an Omni-tagged EXO1 construct [14] and immunoprecipitated the expressed protein using a pan-14-3-3 antibody. The data showed that Omni-EXO1 and 14-3-3 proteins could be recovered as a complex (Figure 1A).
To assess the physiological significance of the EXO1/14-3-3 interaction we selected Sacc. cerevisiae, a system where only two 14-3-3 proteins are present, namely Bmh1 and Bmh2. In preliminary experiments we examined whether yeast Exo1 and 14-3-3 proteins interact. A C-terminal Myc- or a HA-tag was added to the endogenous EXO1 or BMH1/BMH2 genes, respectively. Immunoprecipitation experiments showed that Exo1 formed complexes with Bmh1 or Bmh2 in a HU-dependent manner (Figure 1B). We next explored a possible direct Bmh/Exo1 interaction by Far Western blot analysis. Exo1-Myc immunoprecipitated from control or HU-treated cells was resolved by SDS-PAGE and denatured/renatured on PVDF. Probing the membrane with purified GST-Bmh1 revealed that a direct interaction with Exo1 occurred both in the case of untreated and HU-treated cells (Figure S1). These data possibly indicate that an Exo1 domain normally hidden in non-treated cells may become available for interaction with 14-3-3 proteins upon HU-treatment. Such domain may be artificially exposed during Far Western analysis due to incomplete renaturation of Exo1.
Taken together, these data suggest that the EXO1/14-3-3 interaction is conserved from yeast to mammalian cells. While the interaction is HU-independent in mammalian cells, it requires HU in yeast. This may reflect the different modes of EXO1 regulation in the two systems [15], [17].
Genetic and flow cytometric analysis evidenced the sensitivity of 14-3-3-deficient cells to DNA replication stress, with distinct bmh1 (bmh2Δ) alleles showing different defects upon nucleotide depletion (HU) or treatment with DNA damaging agents (UV or methylmethansulfonate, MMS) [26]. However, despite the evidence that 14-3-3 proteins bind origins of replication and cruciform DNA [29], suggesting a regulatory role in DNA replication [25], the issue of possible direct involvement of 14-3-3 in fork stability or processing under genotoxic stress conditions remained to be clarified. Given the comprehensive molecular characterization of yeast Exo1 as component of the replisome and of its role, in checkpoint defective cells, in the processing of forks stalled by nucleotide depletion [4], we focused our investigations on the bmh1-280 bmh2Δ double mutant (bmh1bmh2 hereafter). This mutant shows normal cell cycle progression in unperturbed conditions, but selective sensitivity and cell cycle recovery defects in response to HU [26]. The mutant Bmh1-280 protein carries a single point mutation (E136>G) in helix αE at a residue neighboring amino acids that form salt bridges and hydrogen bonds with the ligand [21]. Interestingly, immunoprecipitation experiments showed that the mutant Bmh1-280 protein did not interact with Exo1 in untreated nor HU-treated cells (Figure 1C). Thus, we asked whether the cell cycle recovery defects of this mutant reflect a direct role of 14-3-3 proteins at replication forks and whether Exo1 is also implicated in these processes. We performed neutral-neutral bidimensional gel electrophoresis (2D gel) on the early origin of replication ARS305, which is known to be activated in HU-treated cells [3]. Although the 2D gel pattern looked normal in HU-treated bmh1bmh2 cells, we observed that replication intermediates (RIs) in 14-3-3 defective cells were still present close to the origin 60 min after HU removal and were only restarted at 90–120 min (Figure 2A and data not shown). This suggests that misregulation of the replisome, without dramatic physical processing of the forks, might be sufficient to impair fork restart. This effect was not detectably suppressed by EXO1 deletion (Figure 2A).
Flow cytometric analysis of HU-released cells confirmed the slow recovery of the bmh1bmh2 strain and showed that lack of Exo1 per se did not alter the pattern of cell cycle progression (Figure 2B). On the other hand, EXO1 deletion in a bmh1bmh2 background led to a partial rescue of the recovery defect, particularly evident at late time points (≥120 min) after release from HU (Figure 2B). This evidence prompted us to ask whether EXO1 deletion in this background may affect Rad53 activity. Western blot analysis with total and phosphospecific Rad53 antibodies [30] showed that, compared to wild type cells, Rad53 was hyperphosphorylated in HU-treated bmh1bmh2 cells and that its dephosphorylation was retarded during the HU-recovery phase (Figure 2C and 2D), thus correlating with the described replication restart defect. Importantly, deletion of EXO1 in 14-3-3-deficient cells re-established to a great extent the pattern of rapid Rad53 dephosphorylation in the recovery phase (Figure 2C and 2D), substantiating the flow cytometry data (Figure 2B).
Overall these data suggest that 14-3-3 proteins are directly implicated in the effective restart of stalled DNA replication forks. Alternatively, they may assist rapid Rad53 dephosphorylation, which is in turn required for fork restart upon HU removal [31]. The latter interpretation is however unlikely as EXO1 deletion, which markedly restores Rad53 dephosphorylation upon HU removal, does not detectably improve the defective fork restart observed in 14-3-3 deficient cells on the ARS305 replicon. Thus, in the 14-3-3 defective background, Exo1 activity does not directly impact the rate of fork restart, but slows down checkpoint inactivation and delays cell cycle resumption.
Exo1 is controlled in a phosphorylation-dependent manner upon replication fork stalling in mammalian cells [14] and upon a variety of genotoxic insults in yeast [17]. We obtained evidence that yeast Exo1 is phosphorylated in a Mec1-dependent manner also in response to HU (Figure 3A). Notably, the improved resolution of Exo1 phospho-forms by Phos-tag SDS-PAGE [32] allowed us visualizing the complete pattern of Exo1 phosphorylation in response to replicative stress (Figure 3A and 3B).
Next, we asked whether 14-3-3 proteins might be involved in the regulation of Exo1 phosphorylation and stability. Western blot analysis showed that in 14-3-3-deficient cells total Exo1 levels were unchanged (Figure S2), but Exo1 was not phosphorylated to the same stoichiometry observed in wild type cells (Figure 3B, 90 min). Moreover, the rate of Exo1 dephosphorylation upon recovery from HU was considerably reduced in mutant cells, with Exo1 being completely dephosphorylated in wild type but not in 14-3-3-deficient cells (Figure 3B, 120 min). Defective Exo1 phosphorylation in HU-treated 14-3-3-deficient cells is not an indirect consequence of defective checkpoint activation, as under these conditions Rad53, another Mec1-dependent checkpoint target, is promptly phosphorylated (Figure 2C and 2D). Since phosphorylation restrains yeast Exo1 activity [17], we propose that 14-3-3 proteins play an important role in the dynamic control of Exo1 activity upon DNA replication stress and may act as platform for the control of Exo1 phosphorylation. In this respect, attempts to assess the phosphorylation status of Bmh-bound Exo1 were unfortunately inconclusive, as - differently from TCA extracts (Figure 3) - the extracts used for immunoprecipitation fail to be resolved in discrete bands by Phos-tag SDS-PAGE (data not shown). Given that 14-3-3 proteins bind ligands in phospho-dependent and -independent manner [21], it will be important to overcome these technical limitations to address the role of 14-3-3 proteins in controlling the phosphorylation of Exo1 and, possibly, additional targets in the DNA damage response (see below).
As Exo1 activity and Rad53 phosphorylation have been linked to the processing of stalled DNA replication forks, we decided to assess whether defective Rad53 and Exo1 phosphorylation in 14-3-3-deficient cells could reflect changes in the fine architecture of stalled forks. To answer this question, we synchronized the cells in G1, released them for 1 h in YPD medium containing 0.2 M HU and examined RIs by electron microscopy (EM) under non-denaturing conditions [33]. For each strain, about 100 RIs were analyzed in duplicate. 14-3-3-deficient cells showed a dramatic accumulation of ssDNA gaps behind the replication fork (Figure 4A). Statistical analysis indicated that approximately 50% of all RIs analyzed contained one or more ssDNA gaps (Figure 4B). Interestingly, deletion of EXO1 in the bmh1bmh2 background completely suppressed this phenotype, leading to a reduction of the ssDNA gaps behind the fork to a level similar to wild type or exo1Δ cells (Figure 4B). The comparison of ssDNA gaps length scored by EM evidenced a striking difference: whereas bmh1bmh2 cells displayed a significant number of large size gaps (>0.5 Kb), the latter were absent in bmh1bmh2exo1Δ cells (Figure 4C). The resolution limit of 50–70 nucleotides may have impaired detection of nicks/small gaps in this as well as in previous EM studies with HU [34]. Such structures, however, become visible in 14-3-3-deficient cells, where the unleashed Exo1 activity would enlarge gaps above the detection limit.
These data suggest that 14-3-3 proteins are required to prevent unscheduled Exo1 activity behind stalled replication forks in a checkpoint-proficient background. The implications of these observations are of great significance. A loose control of Exo1 activity may render DNA synthesis more discontinuous in conditions of replicative stress, either promoting additional uncoupling/repriming events or enlarging existing ssDNA gaps via 5′-3′ exonucleolytic processing. Although additional work is required to directly address this point, it is conceivable that continuous polymerase stall due to insufficient deoxynucleotide levels might per se lead to increased repriming events, thus raising the number of 5′-ends available for processing by Exo1. In this setting, a strict control of Exo1 activity would be needed to limit damage. We observed no bias for the presence of gaps in leading vs. lagging strands - whenever these could be identified [35] - and we could occasionally detect gaps on opposite strands within the same molecule (Figure S3), suggesting that unscheduled Exo1 activity in 14-3-3 defective cells is not restricted to leading or lagging strand.
Replication recovery defects have been previously described and usually reflect replication fork collapse detectable by 2D gel analysis [3]. On the contrary, stalled replication forks in 14-3-3 deficient cells, albeit unable to restart DNA synthesis and abnormally processed by Exo1 activity, upon prolonged HU treatment show a 2D gel pattern indistinguishable from that of wild type cells. We thus decided to investigate in more detail the structure and progression of these forks, performing 2D gel analysis at different time points after HU addition. To this end, cells synchronized in G1 by α-factor were released into medium containing HU and RIs were examined by 2D gels. Figure 5B shows the probes designed to visualize replication fork progression in a region of Chromosome III that contains, besides the early active origin ARS305 [36], a contiguous passively replicated region (Part A) and a region including the dormant origin ARS301 (Part D) [3]. As compared to wild type, bmh1bmh2 cells showed the same kinetics of origin firing, albeit with slightly lower efficiency as revealed by the intensity of the bubble arc at 30 min (Figure 5C). Progression of the forks in HU from ARS305 across the region of Part A (∼5 Kb to the left of ARS305) was completed after 2–3 h in wild type cells, with the peak of intermediates detectable after ∼1 h. In bmh1bmh2 cells the first intermediates appeared on this region with 30 min delay, whereas the peak of intermediates was delayed of ∼2 h as compared to wild type cells (Figure 5C), indicative of a significant decrease in the rate of the replication fork progression in HU. Slow RIs disappearance from the origin and delayed invasion of adjacent chromosomal regions may in principle result also from asynchronous firing of ARS305 during the HU arrest. However we consider this alternative interpretation unlikely for the following reasons: a) by budding experiments, 14-3-3 defective cells do not display asynchronous entrance in S-phase (data not shown); b) the comparable intensity of the Y arc on fragment A in the wt (60 min) and in the 14-3-3 mutant (180 min) suggests that forks progress synchronously but slower from the early origin ARS305; c) accordingly, the progressive accumulation of the Y signal on fragment A in 14-3-3 defective cells strictly correlates with the disappearance of the bubble ark on the ARS305 fragment, further suggesting slow but synchronous progression of replication forks on the ARS305 replicon.
It was previously shown that yeast 14-3-3 proteins bind to the checkpoint kinase Rad53 and directly influence its DNA damage-dependent functions [27]. Therefore, we asked whether the slow fork progression in bmh1bmh2 cells might be solely due to checkpoint defects. To address this issue, we used checkpoint defective Rad53-mutant cells (rad53-K227A). The latter displayed striking differences when compared to bmh1bmh2 cells. Both the destabilization of RIs (ARS305 and Part A) and the uncontrolled firing of dormant origins displayed by rad53-K227A cells (Part D) [3], were absent in bmh1bmh2 cells (Figure 5C). Furthermore, EM did not display any fork reversal or accumulation of ssDNA at replication forks, typical of HU-treated rad53 cells [34] (data not shown). Finally, 2D gel analysis (Figure S4C, S4F and S4H) and drop assays (Figure S5) revealed synergistic effects of 14-3-3 and Rad53 on both fork stability and survival. Overall, these data indicate that the phenotype observed in 14-3-3 deficient cells reflects a genuine role of 14-3-3 proteins at replication forks and that 14-3-3 and Rad53 have crucial but distinct roles at HU-challenged forks.
Deletion of EXO1 partially rescued the HU-sensitivity of rad53-K227A cells, but not of a bmh1bmh2 strain (Figure S5). Furthermore, in contrast to checkpoint defective cells, where stability of RIs could be rescued by EXO1 deletion [4], fork progression defects of bmh1bmh2 cells were not rescued by loss of EXO1 (Figure S4G). Thus, while the processing defect that leads to accumulation of ssDNA gaps in 14-3-3-deficient cells was completely suppressed by EXO1 deletion, this did not reflect in suppression of HU sensitivity nor of defective fork progression in HU-treated 14-3-3 deficient cells. Altogether this evidence suggests that 14-3-3 proteins might regulate additional targets during replication stress, possibly through modulation of their phosphorylation. This is not unexpected, given the role of 14-3-3 as integrators of signaling pathways [19] and considering the multiplicity of 14-3-3 targets [37], [38]. Our data implicate 14-3-3 proteins as possible central regulator of the checkpoint response. In analogy with previously reported cases [22] and according to structural data on the dynamic nature of 14-3-3 dimers [39], one may envisage a role for 14-3-3 proteins as docking clamp tethering Exo1 - and other unknown targets - with the kinase controlling its/their activity. Notably, 14-3-3 proteins were reported to bind Rad53 [27], one of the candidate checkpoint kinases required for Exo1 phosphorylation [17].
In conclusion, this work sheds further light on processes occurring at stalled replication forks, proposing 14-3-3 proteins as central integrators of signals that regulate fork stability and processing. Challenges lying ahead consist in the identification of components of the replisome, or proteins controlling them, that may be 14-3-3 targets, as well as in the elucidation of the exact mechanism by which 14-3-3 modulate Exo1 phosphorylation and activity.
The antibodies used in this study were: goat polyclonal anti-Omni-probe (M21, sc-499, Santa Cruz Biotechnology); rabbit polyclonal anti-pan 14-3-3 (SA-483, Biomol); mouse monoclonal anti-HA (12CA5, Sigma) and anti-Myc (9E10, Santa Cruz Biotechnology); rabbit polyclonal anti-Rad53 (a kind gift from C. Santocanale, Galway, Ireland); mouse monoclonal F9 to phosphorylated Rad53 [30] (a kind gift of M. Foiani, Milano, Italy).
The chemicals and peptides used in this study were: Hydroxyurea (Sigma); α1-Mating Factor (Sigma).
The yeast strains used in this study are isogenic to W303-1A (wild type) [40] and are listed in Table 1. All strains have been obtained by one step replacement using the indicated markers and tags that have been generated by PCR. The isolated clones have been verified by colony PCR and Southern Blot or Western Blot.
All deletion (Δ) strains lack the entire coding sequence. All strains containing the bmh1-280 mutation have been generated from strain YLL1090 [26]. The strain KE17 has been generated from DMP4644/4A (M.P. Longhese, unpublished) and is a derivative of YLL1090 [26].
The yeast two-hybrid screening was performed with ΔN-EXO1 (EXO1366–846) as bait on a cDNA library generated from human peripheral blood mRNA (a kind gift of I. Stagljar, Toronto, Canada) as described previously [41] and using THY AP4 as reporter strain.
Western blot analysis of yeast proteins was carried out upon TCA extraction [42]. To visualize Exo1, an optimized Phos-tag system (50 µM Phos-tag reagent) was employed according to [32]. Proteins were transferred to nitrocellulose (porablot NCP, 0.45 µm pore size, Machery-Nagel) overnight at room temperature applying constant amperage (200 mA). Far Western blotting [43] was performed using purified recombinant GST-Bmh1 [44] to probe Exo1 that was immunoprecipitated from control or HU-treated yeast cells.
HEK293T cells were grown and lysed as described [14] and protein concentration was determined using the Bio-Rad Protein Assay Reagent (Bio-Rad). Immunoprecipitation of Omni-EXO1 or 14-3-3 proteins from 2.5 mg total cell extracts with specific antibodies was performed as previously described [14].
For immunoprecipitation of yeast proteins, cells were lysed using ice-cold lysis buffer (25 mM Tris-HCl pH 7.4, 15 mM NaCl, 15 mM EGTA, 1 mM NaF, 1 mM Na orthovanadate, 4 mM p-Nitro-Phenyl-Phosphate (pNPP), 0.1% Triton X-100, 1 mM PMSF, complete protease inhibitors cocktail (Roche)). 14-3-3-HA was immunoprecipitated from 10 mg total yeast cell extracts using the monoclonal HA antibody.
DNA extraction with the CTAB method and neutral-neutral two-dimensional gel electrophoresis were performed as described [45]. EM analysis was performed as described [33].
|
10.1371/journal.pgen.1005636 | Single Strand Annealing Plays a Major Role in RecA-Independent Recombination between Repeated Sequences in the Radioresistant Deinococcus radiodurans Bacterium | The bacterium Deinococcus radiodurans is one of the most radioresistant organisms known. It is able to reconstruct a functional genome from hundreds of radiation-induced chromosomal fragments. Our work aims to highlight the genes involved in recombination between 438 bp direct repeats separated by intervening sequences of various lengths ranging from 1,479 bp to 10,500 bp to restore a functional tetA gene in the presence or absence of radiation-induced DNA double strand breaks. The frequency of spontaneous deletion events between the chromosomal direct repeats were the same in recA+ and in ΔrecA, ΔrecF, and ΔrecO bacteria, whereas recombination between chromosomal and plasmid DNA was shown to be strictly dependent on the RecA and RecF proteins. The presence of mutations in one of the repeated sequence reduced, in a MutS-dependent manner, the frequency of the deletion events. The distance between the repeats did not influence the frequencies of deletion events in recA+ as well in ΔrecA bacteria. The absence of the UvrD protein stimulated the recombination between the direct repeats whereas the absence of the DdrB protein, previously shown to be involved in DNA double strand break repair through a single strand annealing (SSA) pathway, strongly reduces the frequency of RecA- (and RecO-) independent deletions events. The absence of the DdrB protein also increased the lethal sectoring of cells devoid of RecA or RecO protein. γ-irradiation of recA+ cells increased about 10-fold the frequencies of the deletion events, but at a lesser extend in cells devoid of the DdrB protein. Altogether, our results suggest a major role of single strand annealing in DNA repeat deletion events in bacteria devoid of the RecA protein, and also in recA+ bacteria exposed to ionizing radiation.
| Deinococcus radiodurans is known for its exceptional ability to tolerate exposure to DNA damaging agents and, in particular, to very high doses of ionizing radiation. This exceptional radioresistance results from many features including efficient DNA double strand break repair. Here, we examine genome stability in D. radiodurans before and after exposure to ionizing radiation. Rearrangements between repeated sequences are a major source of genome instability and can be deleterious to the organism. Thus, we measured the frequency of recombination between direct repeats separated by intervening sequences of various lengths in the presence or absence of radiation-induced DNA double strand breaks. Strikingly, we showed that the frequency of deletions was as high in strains devoid of the RecA, RecF or RecO proteins as in wild type bacteria, suggesting a very efficient RecA-independent process able to generate genome rearrangements. Our results suggest that single strand annealing may play a major role in genome instability in the absence of homologous recombination.
| The extreme resistance of the bacterium D. radiodurans to DNA-fragmenting treatments, such as ionizing radiation or desiccation, is correlated with the ability to reconstruct a functional genome from hundreds of chromosomal fragments. The rapid reconstitution of an intact genome is thought to occur through an extended synthesis-dependent strand annealing process (ESDSA) followed by DNA recombination [1,2]. During ESDSA, chromosomal fragments with overlapping regions are used both as primers and templates for a massive synthesis of single-stranded DNA extensions. Newly synthesized complementary single stranded DNA extensions appear to anneal so that contiguous DNA fragments are joined together forming long linear intermediates. These intermediates require RecA-dependent homologous recombination to mature into reconstituted circular chromosomes representing DNA patchworks of numerous double-stranded DNA blocks synthesized before irradiation connected by DNA blocks synthesized after irradiation.
We have recently shown that the Deinococcal RecF, RecO, RecR proteins, by their ability to load RecA onto its single-stranded DNA substrate, play a crucial role in DNA double strand break repair via ESDSA and recombinational repair pathways [3]. Mutant ΔuvrD bacteria showed a markedly decreased radioresistance, an increased latent period in the kinetics of DNA double strand break repair and a slow rate of fragment assembly correlated with a slow rate of DNA synthesis, suggesting that UvrD helicase might be involved in the processing of double stranded DNA ends and/or in the DNA synthesis step of ESDSA [3].
More recently, it was proposed that a single strand annealing (SSA) process participates in an early step of DNA double strand break repair by facilitating the accurate assembly of small fragments to generate suitable substrates for subsequent ESDSA-promoted genome reconstitution [4]. The DdrB protein was previously shown to exhibit in vitro properties akin to those of SSB protein [5] but also to promote annealing of single stranded DNA [6]. The DdrB protein, recruited early after irradiation into the nucleoid, was also shown to be involved in the slow DNA double strand break repair observed in cells devoid of the RecA protein, and thus to play a major role in RecA-independent DNA double strand break repair through SSA [4,6].
Rearrangements between repeated sequences are a major source of genome instability and can be deleterious to the organism. These rearrangements can result in deletion or duplication of genetic material flanked by direct repeats. In Escherichia coli, recombination between directly repeated sequences occurs via both RecA-independent and RecA-dependent mechanisms, depending on the size of the repeats and of the intervening sequences between the repeated sequences [7–9]. Insertion of a sizable DNA sequence in between the repeated sequences substantially increased the RecA dependence, suggesting that increasing the distance separating the homologous regions preferentially inhibits the RecA-independent recombination in E. coli [9,10]. In E. coli, RecA-independent rearrangements between short repeats, such as deletions, are stimulated by mutations that affect the DNA polymerase or other proteins involved in DNA replication [11–13] leading to the proposal that these events occur during DNA replication by a mechanism involving mispairing of the newly synthetized DNA strand with an alternative complementary template sequence located nearby [7,12] (for review, see [14,15]). An alternate mechanism for RecA-independent deletion events involves DNA breakage, exonucleolytic erosion of the DNA ends and single strand annealing (SSA) of exposed complementary single stranded DNA [14]. A single strand annealing mechanism has also been proposed for RecA-independent deletions associated with a restart of collapsed replication forks [7,11,12,14,16]. Here, we measured the frequency of recombination between direct repeats separated by intervening sequences of various lengths restoring a functional tetA gene in the presence or absence of radiation-induced DNA double strand breaks in D. radiodurans. We also assessed the involvement of the RecA, RecO, RecF, UvrD and DdrB proteins in the deletion process. The role of these proteins in the progression of replication forks was also discussed.
To investigate the role of recombination proteins in the occurrence of repeat-mediated deletion events in D. radiodurans, we constructed a mutated tetA allele bearing an internal duplication and a spc cassette inserted between the duplicated regions. The engineered tetA allele was inserted into the dispensable amyE locus of chromosome 1 and provided a recombination substate in which the direct repeats (438 bp long) were separated by a 1,479 bp spacer (Fig 1A). Precise deletion of one of the direct repeats and the intervening sequence restores the wild type tetA allele. The presence of a functional tetA gene on one copy of chromosome 1 suffices to confer tetracycline resistance to the cells, although D. radiodurans bacteria contain 4 to 10 genome equivalents per cell. In contrast, when the deletion of a gene generates a loss-of-function mutant, all the copies of the gene must be eliminated to detect the mutant phenotype, and failure to obtain a homozygote provides a first indication that the gene might encode a function essential for cell viability [3,17].
The frequency of spontaneous deletion events in a population was estimated by measuring the frequency of [TetR] bacteria. As shown in Fig 1B, the median of the frequencies of [TetR] in the wild type bacteria was 6.5x10-4 and did not decrease in ΔrecA bacteria (median value: 8x10-4). Does the apparent RecA-independent high frequency of deletion events result from a functional redondancy of RecA activities in the cells? We tested the involvement of the RadA protein, a RecA-related protein, and showed that wild type frequencies of deletions were found in ΔradA and in ΔrecA ΔradA bacteria (Fig 1B), suggesting that the RadA protein was not required in RecA-independent recombination to compensate for the absence of the RecA protein. We also observed that the frequency of [TetR] bacteria was not reduced in cells devoid of RecF or RecO proteins required for loading RecA onto its single-stranded DNA substrate (Fig 1B). Altogether, our results suggest an important contribution of RecA-independent mechanisms in the generation of deletions between repeated DNA sequences in D. radiodurans.
To test whether the same 438 bp homologous fragments can undergo efficient RecA-dependent recombination, we constructed a plasmid-by-chromosome recombination assay in which the recombining tetA fragments were placed in a different configuration, one being located at the chromosomal amyE locus and the other on a resident plasmid (Fig 2A). In this assay, the reconstitution of a functional tetA gene resulted from the integration of the plasmid into the chromosomal DNA as verified by PCR analysis of few [TetR] colonies (S1 Fig). The integration event is well tolerated by the cell, since we used a low copy number plasmid p15002, a derivative of plasmid pI8 maintained at 4 to 10 copies per cell in D. radiodurans [18]. As can be seen in Fig 2B, the frequency of [TetR] bacteria dropped from a median value of 2x10-5 in the wild type to less than 5 x 10−7 in cells devoid of the RecA or the RecF proteins, indicating that the 438 bp homologous fragments recombine through a classical RecA-promoted strand exchange mechanism. In contrast, loss of the RadA protein did not impair recombination efficiency. In E. coli, the RadA protein has been involved in processing of branched recombination intermediates. However single radA mutants have a modest effect on recombination and DNA survival while they show a strong synergistic effect in combination with mutations in the recG or the ruvAB Holliday junction proteins [19,20].
The presence of mutations in one of the 438 bp repeats (Fig 3A) reduced the frequency of the deletion events between the chromosomal repeats. A single mutation sufficed to significantly decrease the frequency of the deletion events as shown by median values that decreased by a factor of 4.8 in recA+ bacteria and 4.6 in ΔrecA bacteria, when compared to fully homologous repeats. The decrease was greater when 3 mutations were present in one of the repeats, yielding reduction factors of 10.0 and 19.0 in recA+ and ΔrecA bacteria, respectively (Fig 3B). A plot of the frequency of [TetR] bacteria as a function of the number of differences shows a linear decrease in the deletion frequency with similar regression slopes in recA+ and ΔrecA bacteria (S2 Fig).
A similar analysis was performed in recA+ ΔmutS and ΔrecA ΔmutS bacteria devoid of the MutS protein, the key enzyme involved in mismatch recognition. The data (Fig 3B and S2 Fig) show that in this case the differences between the repeats did not significantly affect the deletion frequency. These results suggest that, as in homologous recombination intermediates, a heteroduplex DNA is formed during RecA-independent processes leading to the reconstitution of a functional tet gene and that an efficient mismatch repair aborts recombination between the DNA repeats in ΔrecA as well as in recA+ bacteria.
It was previously shown that uvrD mutations stimulate RecA-dependent recombination [21–23] and enhance tandem repeat deletions in E. coli [23]. Here, we show that the absence of UvrD enhanced the efficiency of RecA-dependent recombination between chromosomal and plasmid DNA by a factor of 21.2 (Fig 2B), suggesting that deinococcal UvrD protein possesses an anti-RecA activity as previously shown for the E. coli UvrD protein [24,25]. The absence of UvrD also enhanced the frequency of deletions between the chromosomal direct repeats by a factor of 7.1 (Fig 1B). This increase may be due to the anti-RecA activity of UvrD protein that can possibly inhibit RecA-dependent recombination between the repeated sequences in a recA+ uvrD+ background. However, the absence of UvrD might also disturb DNA replication, and thus increase genome instability. A clue to understand how the absence of the UvrD protein might be involved, independently of its anti-RecA activity, in the stimulation of deletion events requires an analysis of its effects in a recombination-deficient background. Unfortunately, we were unable to obtain homozygotes for recA, recF or recO deletion in combination with a uvrD deletion, even after extensive purification steps (S3 Fig), suggesting that uvrD deletion is colethal with a recA, recF or recO deletion. We propose that the UvrD protein, by displacing obstacles downstream of the replisome, plays an important role in the progression of replication forks (see discussion).
Previous in vitro and in vivo results suggest that the DdrB protein plays a major role in a single strand annealing process (SSA) that operates early in genome reconstitution after DNA damage [4,6]. Single strand annealing is the only activity of DdrB known besides binding to single strand DNA [5,6]. Thus, to analyse the involvement of SSA in generating deletions via a RecA-independent pathway, we decided to construct double mutants devoid of DdrB and RecA or RecO proteins. Homogenotization of ΔddrB ΔrecA and ΔddrB ΔrecO double mutants was difficult, requiring 7 steps of purification, suggesting growth inhibition of the mutated cells (S3 Fig). Thus, we compared the growth rate and the plating efficiency of the double mutants with those of the single ΔddrB, ΔrecA and ΔrecO mutants and of the parental wild type strain. Wild type and ΔddrB bacteria exhibited a generation time of 105 min. The recombination deficient bacteria grew more slowly, as ΔrecA and ΔrecO bacteria during exponential growth showed a generation time of 285 min whereas ΔddrB ΔrecO and ΔddrB ΔrecA exhibited a generation time of 370 min. Moreover, during exponential growth phase, the single ΔrecA and ΔrecO mutants had a 15 fold decreased plating efficiency as compared with the wild type, whereas the ΔddrB ΔrecA and ΔddrB ΔrecO double mutants had a 35 fold decreased plating efficiency (Fig 4). These results suggest that the DdrB protein may be involved in management of blocked replication forks in the absence of the RecA or RecO proteins. Another striking result was the increased lethality of the recombination deficient mutants in late stationary phase. Indeed, after reaching a plateau after 6 hours of incubation for the wild type and ΔddrB bacteria, the number of CFU did not decrease during 70 additional hours of incubation. In contrast, the single ΔrecA and ΔrecO recombination deficient mutants and the double ΔddrB ΔrecA and ΔddrB ΔrecO mutants reached a plateau after 18 to 20 hours of incubation and the number of CFU decreased 2–3 orders of magnitude after 30 hours of incubation (Fig 4) suggesting that DNA lesions are generated during prolonged stationary phase and require recombination functions to be repaired.
We found that the absence of DdrB had a strong negative effect on the frequencies of deletions events between the chromosomal repeats generated via a RecA-independent pathway. Indeed, when the repeats are separated by 1,479 bp, the median values of the deletion frequencies in ΔddrB ΔrecA and ΔddrB ΔrecO bacteria decreased by a factor of 4.9 and 5.1 respectively, as compared to their ΔrecA and ΔrecO counterparts (Fig 1B). These results indicate that 80% of the [TetR] bacteria generated in the absence of RecA or RecO proteins were formed in a DdrB-dependent manner, suggesting a major role of single strand annealing in RecA-independent recombination between the direct repeats. We can also hypothesize that the DdrB protein might be involved in the stabilization of DNA polymerase template switching intermediates.
In E. coli, in which both RecA-dependent and RecA-independent mechanisms can contribute to recombination between direct repeats, deletion events become increasingly RecA-dependent as the distance between the repeated sequences increases [9]. To verify if this also applies in D. radiodurans, we modified the test deletion construct shown in Fig 1 by replacing the 1,479 bp spacer with sequences of increasing length to analyse the impact of the distance between the repeats on the incidence of deletions and their genetic control (Fig 5A).
We found that the increase of the distance between the repeats from 1,479 bp (Fig 1A) up to 10,500 bp (Fig 5A) had no effect on the deletion frequency in recA+ as well as in ΔrecA or ΔrecF hosts (compare the [TetR] frequencies in Figs 1B, 5B, 5C and 5D). Likewise, the distance between the repeats had no effect on the stimulation of deletion events by the absence of the UvrD protein (Figs 1B, 5B, 5C and 5D).
In contrast, the involvement of DdrB in the deletion events became more apparent when the distance between the repeats increased. Indeed, while a DdrB deficiency had no effect on the frequency of deletions in a recA+ background when the spacer between the repeats was 1,479 bp long, it produced a 2 to 3-fold decrease in the deletion frequency when the spacer length increased (Figs 1B, 5B, 5C and 5D). When the ddrB deletion was associated with a recA deletion, the reduction factors were found to be between 18- and 20-fold if the length of the intervening sequences was ≥ 3,500bp as compared to their single ΔrecA counterparts (Fig 5B, 5C and 5D). A similar effect of a ddrB deletion was also observed in cells devoid of the RecO protein (Fig 5B, 5C and 5D). These results suggest that, in the absence of the RecA-promoted homologous recombination, approximately 95% of the recombination events were dependent on the DdrB protein and may be related to an SSA pathway.
We used our deletion assay to analyze the impact of the presence of repeated sequence on the stability of the genome in γ-irradiated cells during the process of genome reconstitution. We showed that the frequency of repeat-induced deletions restoring a functional tetA gene increased as a function of the dose of γ-irradiation used (Fig 6A). Thus, we further exposed the cells to 5 kGy γ-irradiation, a dose producing hundreds of DNA double strand breaks [26]. The repeats in the tested cells were separated either by 1,479 bp (“short” spacer), 3,500 bp, 6,500 bp, or by 10,500 bp (“long” spacer) sequences. The deletion analysis was performed only in a recA+ background, since the extreme radio-sensitivity of ΔrecA or ΔrecF bacteria (cell survival was less than 10−5 after exposure to 5 kGy) precluded the inclusion of these cells in the genetic assay. The deletions were induced by exposure to γ-irradiation independently of the length of the spacers (Fig 6B).
As shown in Fig 6, in the wild type bacteria, the deletion frequency increased by 11.9 fold and 5.9 fold after irradiation when the repeats are separated by the”short” spacer and the “long” spacer, respectively (compare left panels in Fig 6C and 6D). In cells devoid of the UvrD protein, the frequency of deletions moderately increased after irradiation (induction factors of 1.45 and 1.7 for the “short” and “long” spacer, respectively), likely because these cells already have an elevated spontaneous level of recombination in the absence of irradiation (compare right panels in Fig 6C and 6D). In contrast, cells devoid of the DdrB protein showed a marked reduction in the induced levels of deletion events (compare middle panels in Fig 6C and 6D), suggesting that single strand annealing might play an important role in the generation of the [TetR] recombinants during genome reconstitution after irradiation.
Repeated sequences are targeted by recombination processes leading to amplifications, deletions, and other rearrangements of the genetic material. These events play an important role in genome plasticity and rapid adaptation to environmental challenges, but are also potential source of genome instability and can be deleterious to an organism (see for review [27]). D. radiodurans contains an enhanced number of repetitive sequences as compared to other bacteria, including insertion sequences, small non-coding repeats, and intragenic repeats (see for review [28]). This bacterium is also known for its capacity to reconstitute an entire genome from a myriad of fragments after exposure to elevated γ-irradiation doses. Genome reconstitution occurs through extended synthesis dependent single strand annealing (ESDSA) followed by classical homologous recombination. In heavily irradiated cells, a RecA-independent single strand annealing (SSA) process takes place before ESDSA allowing the assembly of small fragments into substrates that can be further processed through ESDSA. The single strand annealing activity of the DdrB protein plays a major role in this early step of DNA double strand break repair [4].
Here, we show that DdrB also plays a key role in RecA-independent recombination between direct repeats, leading to the occurrence of deletions, and substantially contributes to the induction of deletions in γ-irradiated recA+ bacteria. Our results points to an important contribution of a DdrB-dependent pathway in D. radiodurans genome plasticity.
Here, we determined the frequencies of spontaneous and radiation-induced recombination between chromosomal direct repeats and investigated the role of RecA and of other key recombination and repair proteins in the occurrence of these events in D. radiodurans. We found that recombination events restoring a functional tetA gene occurred at a very high frequency. In a recA+ background, the median frequency of [TetR] cells was equal to 6.5 x 10−4 (Figs 1, 5B, 5C and 5D), values more than 10-fold higher than those measured in a study that used a similar substrate (chromosomal 358 bp direct repeats separated by an intervening sequence of 850 bp) to measure recombination in Helicobacter pylori, a bacterium known for its high recombination proficiency [29]. Moreover, inactivation of recA resulted in a 10-fold decrease in recombination between the direct repeats in H. pylori [29]. In contrast, introduction of a ΔrecA, ΔrecF or ΔrecO mutation in the D. radiodurans tester strains did not change the elevated recombination frequencies between the repeats (Figs 1, 5B, 5C and 5D). In E. coli and in B. subtilis, the distance between the repeats plays a key role in determining the mechanisms involved in the recombination processes, the efficiency of RecA-independent recombination decreasing sharply when the distance between the repeats increases [9, 10, 30]. No such proximity effect was observed in D. radiodurans. Indeed, the frequency of appearance of the recombinants in our assay remained elevated in ΔrecA (and ΔrecF or ΔrecO) bacteria as in the parental rec+ bacteria when the spacer between the repeats increased from 1,479 up to 10,500 bp (Fig 5).
Our results suggest that, in the absence of RecA (or in the absence of “facilitator” proteins required for loading RecA onto its single-stranded DNA substrate), alternate pathways ensure recombination between repeated sequences in D. radiodurans. These RecA-independent pathways do not necessarily predominate in recA+ bacteria, although we observed a similar frequency of recombinants in recA+ and ΔrecA bacteria. Indeed, ΔrecA (or ΔrecF or ΔrecO) bacteria had a low plating efficiency with less than 10% of cells able to form colonies.
In E. coli, mutations in components of the DNA Pol III holoenzyme result in elevated levels of tandem repeat rearrangements, supporting the idea that RecA-independent recombination occurs during the process of chromosome replication [14]. A replication slipped misalignment model [14] proposed that a pause in DNA synthesis and dissociation of the polymerase from its template allows the nascent strand to translocate to a new pairing position. Slipped misalignment is thought to occur on single-stranded DNA and thus more frequently during lagging-strand synthesis [14]. The availability of single-stranded DNA on the lagging-strand template, and thus, the length of the Okazaki fragments, might constitute parameters that govern the efficiency of the deletion events and might explain the strong dependence of the deletion frequencies on the proximity of the repeated sequences [14]. Our findings that the distance between the repeats in the 1–11 kb range has no influence on the deletion frequency raise the question as to whether these events were generated through a slipped misalignment mechanism, thus implying the presence of very large single stranded DNA regions on the lagging strand template in D. radiodurans. A response to this question awaits better knowledge of the replication machinery in D. radiodurans and a determination of the average size of Okazaki fragments in this bacterium.
We found that the frequencies of the deletion events in ΔrecA ΔddrB (or ΔrecO ΔddrB) bacteria were reduced 5-fold and 18- to 20-fold as compared with those measured in the single ΔrecA (or ΔrecO) mutant counterparts (Figs 1B and 5) in strains containing repeats separated by 1479 bp and 3,500 to 10,500 bp, respectively. These results strongly suggest that the DdrB protein strongly stimulates RecA-independent recombination in D. radiodurans. The DdrB protein was shown to bind single stranded DNA [5] and to mediate in vitro fast annealing of complementary oligomers [6]. In vivo, the single strand annealing activity of DdrB is supported by its involvement in plasmid establishment during natural transformation [4]. Thus, we propose that RecA-independent recombination between direct repeats occurs mainly through a DdrB-dependent single strand annealing (SSA) pathway. SSA was first proposed to explain circularization of linear duplex phage DNA containing terminal repetitions by annealing complementary terminal single overhangs [31]. The SSA model was postulated later to take place in eukaryotic cells [32,33] where it is facilitated by RPA and RAD52 in a RAD51-independent manner [34,35]. SSA involves an initial DNA double strand break in the sequence between the duplications followed by the action of a 5’ to 3’ exonuclease to expose single stranded regions in both repeats that are subsequently aligned and annealed by the RAD52-RPA-ssDNA ternary complex. Annealed intermediates are then processed by digestion of the displaced single stranded DNA, polymerase filling-in and ligation to generate the final recombination product (For review, see [36]).
Although DdrB does not share sequence similarity with the eukaryotic RAD52 protein, it might act as its functional equivalent [37,38]. The activity of the RAD52 protein is strongly stimulated by the presence of RPA [39,40]. In contrast, the single strand annealing activity of DdrB is not stimulated but rather inhibited by inclusion of the SSB protein in the in vitro annealing reaction [4,6]. The deinococcal SSB protein is an essential protein and the DdrB protein is unable, even when overexpressed, to replace SSB for cell viability [41]. SSB is crucial for all aspects of DNA metabolism [42] while DdrB seems to have a more specialized role in DNA repair and plasmid transformation by stimulating the SSA pathway. Both DdrB and SSB bind to 3’ single stranded tails of resecting ends [5]. The ends can be engaged in two alternative pathways: annealing to complementary ssDNA in the SSA pathway, or, depending on the formation of a RecA nucleofilament, invasion of homologous dsDNA to promote strand exchange in homologous recombination or to prime DNA synthesis in ESDSA. Polymerization of RecA on ssDNA requires the displacement of SSB or DdrB from ssDNA. The SSB protein can be efficiently displaced through the action of RecO and RecR proteins [43,44]. DdrB protein binds more tightly than SSB to ssDNA [5] and might be displaced with more difficulty from ssDNA. We propose that, in D. radiodurans, homologous recombination and SSA might also compete for substrate in making deletions between direct repeats. Reams et al. [45] proposed that, in Salmonella enterica, a single-strand annealing pathway might also be activated to generate duplication between tandem copies of the ribosomal RNA genes (rrn) when two single-stranded DNA ends are provided and neither strand is coated with inhibitory RecA protein. Under these conditions, the activation of single-strand annealing might compensate the loss of homologous recombination [45].
We found that ΔrecA (or ΔrecO) bacteria had a 15 fold decreased plating efficiency as compared with the wild type during exponential growth phase (Fig 4). This important lethal sectoring suggests that problems resulting in the arrest of replication fork (see for review [46]) occur at high frequency in D. radiodurans, and that RecA-mediated recombination plays a key role in the recovery of stalled replication forks in this bacterium. A further (2-fold) decreased plating efficiency was observed when a DdrB deficiency was combined with a RecA (or RecO) deficiency (Fig 4). These results are consistent with a single strand annealing model but also with any model that envisions annealing of complementary DNA strands, for example misannealing of the direct repeats during recovery from replication fork collapse in cells devoid of the RecA protein. D. radiodurans also seems to be very sensitive to prolonged stationary phase, with a rapid loss of cell viability when proteins involved in homologous recombination were absent, suggesting that DNA double strand breaks were generated in “old” cells and could not be repaired in the absence of RecA or RecO proteins.
Another important feature was the increased recombination frequency between repeated sequences measured in a D. radiodurans mutant devoid of the UvrD protein at a level almost equivalent to those measured after irradiation of wild type bacteria (Fig 6). It was previously shown that uvrD mutations enhance tandem repeat deletion in the E. coli chromosome [23,47] and stimulate RecA-dependent recombination [21,48,49]. In E. coli, mutations in uvrD induce the SOS response, a common phenotype in cells with replication defects [50]. The obstacles possibly encountered by replication forks during their progression are multiple, such as tightly bound proteins, nicks or DNA lesions. The Rep helicase acts by dislodging proteins in front of replication forks [51–53] and its absence results in a marked slowing down of replication progression [54], suggesting increased fork arrest. Simultaneous inactivation of Rep and UvrD helicases is lethal in E. coli [23] suggesting that UvrD might partially substitute for the Rep protein in ensuring replication progression [55]. In favour of this hypothesis, it was recently shown that UvrD displaces the obstacles downstream of the replisome in vitro [52] and plays a major role to displace transcription complexes [56]. Moreover, it was proposed that UvrD acts at blocked replication forks by clearing RecA, facilitating replication fork reversal [57, 58], a hypothesis supported by the ability of UvrD to directly remove RecA nucleoprotein filaments in vitro [25].
In D. radiodurans, we previously showed that inactivation of uvrD results in a marked slowing down of replication progression in un-irradiated cells [3]. During post-irradiation recovery through ESDSA, the absence of UvrD results in a delayed kinetics of DNA double strand break repair that coincided with delayed and less extensive DNA synthesis than that observed in the wild type cells [3]. D. radiodurans bacteria are naturally devoid of the RecB and RecC proteins, and it was suggested that UvrD, in association with the RecJ exonuclease, might play an important role in the processing of DNA double strand ends required for priming of DNA synthesis, but also may act in the DNA synthesis elongation step of ESDSA and more generally may play an important role for the progression of replication forks [3]. It is important to notice that we were unable to obtain mutants devoid of the RecJ protein [3], and recJ mutants constructed by Hua and his collaborators were shown to grow very slowly and to be thermosensitive [59].
In D. radiodurans, we were unable to delete the recA gene when bacteria were devoid of the UvrD protein, suggesting colethality of uvrD and recA deficiencies. These results are reminiscent of phenotypes observed in particular rad3 mutants of Saccharomyces cerevisiae. The RAD3 gene, a homolog of the human gene XPD, encodes a helicase which is a component of the NER apparatus as part of the transcription factor TFIIH. Interestingly rad3-101 and rad3-102 mutants accumulate DNA double strand breaks and are lethal when in combination with mutations in recombinational repair genes, strongly suggesting that Rad3 protein influences either the generation of DNA double strand breaks or their processing by homologous recombination [60].
We propose that the absence of UvrD in D. radiodurans may disturb the progression of the replication fork, and thus might, as RAD3 in S. cerevisiae, influence the generation of DNA double strand break, favouring recombination and also single strand annealing between DNA repeats.
We used our assay to analyze the impact of the presence of repeated sequence on the stability of the genome in γ-irradiated cells. We found that exposure to a dose of 5 kGy γ-irradiation increased the recombination level about 10-fold in the wild type but to a lesser extent in cells devoid of the DdrB protein, suggesting that SSA might play an important role in recombination between the duplicated sequences during the process of genome reconstitution.
In D. radiodurans, interplasmidic recombination between homologous regions was previously shown to be induced by exposure to γ-radiation [61]. Moreover, when two TetS alleles were inserted on the same chromosome into two randomly distant sites, 2% of TetR bacteria were found among the surviving cells exposed to 17.5 kGy, whereas TetR isolates were only very rarely found without irradiation [62]. Interestingly, when two slightly different E. coli plasmids were inserted in the D.radiodurans genome generating adjacent duplication insertions, circular derivatives of the tandemly integrated plasmids were formed in the first 1.5 h postirradiation before the onset of recA-dependent repair in cells exposed to 17.5 kGy γ-irradiation. These circular derivatives had structures consistent with the hypothesis that DNA repair occurred immediately postirradiation by a recA-independent single strand annealing process [63]. These authors proposed that SSA may be a preparatory step for further DNA repair in wild-type D. radiodurans, a hypothesis in accordance with our recent results, suggesting that DdrB-dependent single-strand annealing might facilitate the assembly of the myriad of small fragments generated by extreme radiation exposure to generate suitable substrates for subsequent ESDSA-promoted genome reconstitution [4]. Genome reassembly in irradiated D. radiodurans cells was considered for a long time as an error-free process since no genome rearrangements were detected after post-irradiation DNA repair. Gross chromosomal rearrangements were detected for the first time in recA+ D. radiodurans cells exposed to extremely high γ-doses (25 kGy) and in recA mutant cells that survived 5 kGy γ-radiation [64]. The recA mutants were also shown to be prone to spontaneous DNA rearrangements during normal exponential growth [64]. These authors presumed that SSA, by pairing ectopic repetitive sequences, may be the main source of these chromosomal rearrangements, a hypothesis reinforced by our results suggesting an important role of SSA in recombination between repeated sequences (this work), in DNA double strand break repair in cells devoid of the RecA protein [4,6], and in early reassembly of small DNA fragments when cells were exposed to high γ-doses [4].
Altogether, these results suggest that SSA plays a major role in RecA-independent recombination between repeated sequences in the radioresistant D. radiodurans bacterium. In un-irradiated wild type bacteria, the deletion events might result, as proposed by Susan Lovett in E. coli, from RecA-dependent intermolecular unequal crossing over or intramolecular recombination between the overlapping 5’ and 3’ regions of the tetA gene, and from RecA-independent processes such as replication slippage or template switching [10] or single strand annealing [10,14]. Difficulties in replication can lead to breakage of the fork when replication forks are halted by obstacles or DNA damage in virtually every cell and every cell generation [65,66]. If this occurs in the context of repeated DNA sequences, single-stranded DNA substrates might be generated by resection of the DNA ends, and genetic rearrangements can result through strand-invasion of the broken chromosome with its sister or through SSA at the repeats. Replication of damaged DNA templates can further elevate the probability of fork breakage [67,68]. Moreover, when D. radiodurans cells were exposed to a dose of 5 kGy γ-irradiation, generating hundreds DNA double strand breaks, DdrB-dependent SSA and RecA-dependent ESDSA processes involved in DNA double strand break repair increased the opportunities to generate deletion events when DNA repeats are present in the DNA fragments.
D. radiodurans strains were grown at 30°C, 150 rpm in TGY2X (1% tryptone, 0.2% dextrose, 0.6% yeast extract) or plated on TGY1X containing 1.5% agar. E. coli strains were grown at 37°C, 150 rpm in Lysogeny Broth (LB). When necessary, media were supplemented with the appropriate antibiotics used at the following final concentrations: kanamycin, 6 μg/mL; chloramphenicol, 3.5 μg/mL; hygromycin, 50 μg/mL; spectinomycin, 75 μg/mL; tetracycline 2.5 μg/mL for D. radiodurans and kanamycin, 25 μg/mL or spectinomycin 40 μg/mL for E. coli.
The bacterial strains and plasmids used in this study are listed in Table 1. The E. coli strains used were DH5α as the general cloning host, and SCS110, a dam dcm mutant strain, to propagate plasmids prior introduction into D. radiodurans via transformation [69]. Transformation of D. radiodurans with genomic DNA, PCR products, or plasmid DNA was performed as described [70]. All D. radiodurans strains were derivatives of strain R1 ATCC 13939. The genetic structure and the purity of the mutant strains were checked by PCR. Oligonucleotides used for strain constructions and diagnostic PCR will be provided on request.
Cells were plated on TGY agar and incubated at 30°C for 3 days, or 5 days for ΔrecA, ΔrecO and ΔrecF mutant bacteria and 7 days for double mutant ΔddrB ΔrecA and ΔddrB ΔrecO bacteria. Three to six colonies per strain were inoculated in 3 mL of TGY2X and incubated at 30°C, 150 rpm. Appropriate dilutions of the bacterial cultures, grown to OD650nm = 1.5 were plated on TGY and TGY + tetracycline 2.5 μg/mL. Colonies were counted after 4 to 7 days of incubation at 30°C. The experiments were repeated at least three times, using, when possible, strain isolates obtained independently during the strain constructions.
To measure recombination between repeated sequences after γ-irradiation, bacterial strains were treated as previously described upstream, except that bacterial cultures were grown to an OD650nm = 0.5 before being concentrated by centrifugation in TGY2X to an OD650nm = 10 and irradiated on ice at a dose of 5 kGy with a 60Co irradiation system (LABRA, CEA, Saclay) at a dose rate of 100 Gy/min. Following irradiation, samples of 100 μL were inoculated in 4.9 mL of TGY2X and incubated at 30°C, 150 rpm. After 20 hours of post-irradiation incubation, appropriate dilutions of bacterial culture were plated on TGY and TGY + tetracycline 2.5 μg/mL. Colonies were counted after 4-7days of incubation at 30°C. Unirradiated controls were treated as irradiated cells, except that they were maintained on ice without irradiation during the period when the irradiated cells were exposed to γ-rays.
Cells containing plasmid p15002 were plated on TGY agar + spectinomycin (75 μg/mL) and incubated at 30°C during 3 days (or 5 days for ΔrecA and ΔrecF bacteria). Three colonies per strain were inoculated in 3 mL of TGY2X + spectinomycin (75 μg/mL) and incubated at 30°C, 150 rpm. Appropriate dilutions of the bacterial cultures grown to OD650nm = 1.5 were plated on TGY and TGY+ tetracycline 2.5 μg/mL. Colonies were counted after 4 to 7 days of incubation at 30°C.
Strains were streaked on TGY plates supplemented with the appropriate antibiotics. Independent colonies were inoculated in 3 mL TGY2X supplemented with the appropriate antibiotics (only kanamycin for double mutants) and grown at 30°C to an OD650nm = 1.5. Cultures were then diluted 200 to 5,000 fold and grown overnight at 30°C to an OD650nm = 0.1 (time 0 of the growth curves). Then, the OD650nm was measured and appropriate dilutions were plated on TGY plates at different times during 80 h of incubation at 30°C with agitation (150 rpm). Colonies were counted after 3 (WT or ΔddrB bacteria), 5 (ΔrecA or ΔrecO bacteria) or 7 days (ΔddrB ΔrecA or ΔddrB ΔrecO bacteria) of incubation at 30°C.
In Figs 1 and 3 and 5, in order to establish the statistical differences between the [TetR] frequencies measured in mutant and WT strains, non parametric Dunn’s multiple comparison test [75] were used, taking into account the p-value correction and performed with the GraphPad Prism6 software. All the comparisons were bi-sided.
Linear regressions and the slope significances observed in S2 Fig were estimated using the GraphPad Prism6 software.
In Fig 6, statistically significant differences between the irradiated and the non-irradiated conditions were calculated by non-parametric Mann-Withney tests performed in GraphPad Prism 6 software.
|
10.1371/journal.pcbi.1000223 | Quantification of Local Morphodynamics and Local GTPase Activity by Edge Evolution Tracking | Advances in time-lapse fluorescence microscopy have enabled us to directly observe dynamic cellular phenomena. Although the techniques themselves have promoted the understanding of dynamic cellular functions, the vast number of images acquired has generated a need for automated processing tools to extract statistical information. A problem underlying the analysis of time-lapse cell images is the lack of rigorous methods to extract morphodynamic properties. Here, we propose an algorithm called edge evolution tracking (EET) to quantify the relationship between local morphological changes and local fluorescence intensities around a cell edge using time-lapse microscopy images. This algorithm enables us to trace the local edge extension and contraction by defining subdivided edges and their corresponding positions in successive frames. Thus, this algorithm enables the investigation of cross-correlations between local morphological changes and local intensity of fluorescent signals by considering the time shifts. By applying EET to fluorescence resonance energy transfer images of the Rho-family GTPases Rac1, Cdc42, and RhoA, we examined the cross-correlation between the local area difference and GTPase activity. The calculated correlations changed with time-shifts as expected, but surprisingly, the peak of the correlation coefficients appeared with a 6–8 min time shift of morphological changes and preceded the Rac1 or Cdc42 activities. Our method enables the quantification of the dynamics of local morphological change and local protein activity and statistical investigation of the relationship between them by considering time shifts in the relationship. Thus, this algorithm extends the value of time-lapse imaging data to better understand dynamics of cellular function.
| Morphological change is a key indicator of various cellular functions such as migration and construction of specific structures. Time-lapse image microscopy permits the visualization of changes in morphology and spatio-temporal protein activity related to dynamic cellular functions. However, an unsolved problem is the development of an automated analytical method to handle the vast amount of associated image data. This article describes a novel approach for analysis of time-lapse microscopy data. We automated the quantification of morphological change and cell edge protein activity and then performed statistical analysis to explore the relationship between local morphological change and spatio-temporal protein activity. Our results reveal that morphological change precedes specific protein activity by 6–8 min, which prompts a new hypothesis for cellular morphodynamics regulated by molecular signaling. Use of our method thus allows for detailed analysis of time-lapse images emphasizing the value of computer-assisted high-throughput analysis for time-lapse microscopy images and statistical analysis of morphological properties.
| Cell morphological change is a key process in the development and homeostasis of multicellular organisms [1],[2]. Various types of morphological change appear during migration and differentiation; essential events occurring as part of these processes usually accompany morphologically different phenotypes. Therefore, cell morphology has been used as a key indicator of cell state [3]. High-throughput analyses of cell morphodynamic properties have been used recently to discover new functions of specific proteins [4]. Moreover, the outcomes of morphological change such as the intricate shape of neuronal dendrites, remind us that morphogenesis itself plays a role in the emergence of cellular function [5].
Quantitative approaches are helping to unveil cellular morphodynamic systems, and they are generating new technical requirements. Because cellular morphological change is highly dynamic, time-lapse imaging is necessary to understand the mechanism of cell morphology regulation. Progress in the development of fluorescent probes has enabled the direct observation of cell morphological changes and/or the localization and activity of specific proteins [6]–[8], but time-lapse imaging has highlighted the difficulty of extracting characteristic information from an immense number of images. Nevertheless, several approaches in the context of quantitative analysis have appeared recently. A series of studies using quantitative fluorescent speckle microscopy, for instance, revealed the power of computer-assisted high-throughput analysis for time-lapse microscopy images: analysis of the number of moving and blinking speckles suggested distinct regulation of actin reorganization dynamics in different intracellular regions [9],[10].
Indeed, computational methods have been used to determine the properties of morphological dynamics, protein activity and gene expression [11]–[14]. There are two major approaches for the detailed analysis of local morphological changes of cells. One is the kymograph, which is a widely used method to describe motion with a time-position map of the morphology time course. The time course of change in intensity could also be monitored by arranging sequential images of a specific region of interest (ROI) [15]. Although there are drawbacks to this approach, such as restriction of the analyzed area to a narrow ROI and the need to manually define the ROI, recent studies have avoided these limitations by using polar coordinates to explore the motility dynamics of the entire peripheral region of round cells. Indeed, the polar coordinate-based approach showed isotropic and anisotropic cell expansion, and examined stochastic, transient extension periods (named STEP) or periodic contractions [12],[16]. The second approach is to track cellular edge boundaries by tracing virtually defined markers. Kass and Terzopoulos introduced an active contour model known as SNAKES [17], which is widely used to analyze moving video images in applications including biomedicine. For example, Dormann et al. used SNAKES to quantify cell motility and analyze the specific translocation of PH domain-containing proteins into the leading edge [14]. Marker-based tracking has advantages in quantifying highly motile cell morphology, because it does not require a fixed axis, which is necessary in the kymograph approach. Recently, Machacek and Danuser developed an elegant framework to trace a moving edge, using marker tracking modified by the level set method to elucidate morphodynamic modes of various motile cells such as fibroblasts, epithelial cells, and keratocytes [18].
Although previous methodologies have successfully described the specific aspects of cellular morphodynamics, there remain challenges to quantify the relationship between morphodynamics and signaling events. One representative problem is the association between regions in different frames. To scrutinize the dynamic relationship between morphological change and molecular signaling, we need to cross-correlate them in a time-dependent manner (Figure 1A). A polar coordinate system does not ensure the association of time-shifted local domains (Figure 1B), and is unsuitable for non-circular cell shapes. The virtual marker tracking method satisfies this requirement for cells with broadly consistent shapes, but its fixed number of markers causes unequal distribution when a dramatic shape change such as the persistent growth of neurites in neurons, occurs (Figure 1C). Taking these problems into account, we perceive the need for a novel quantification method to better understand the mechanisms of morphodynamic regulation by molecular signaling.
We focused on the Rho-family small GTPases, or Rho GTPases, as signaling molecules associated with cell morphodynamics. Rho GTPases, which act as binary switches by cycling between inactive and active states (Figure 2), play key roles in linking biochemical signaling with biophysical cellular behaviors [19],[20] mainly through reorganization of the actin and microtubule cytoskeleton [21]. It is well known that RhoA, Rac1, and Cdc42 have unique abilities to induce specific filamentous actin structures, i.e., stress fibers, lamellipodia, and filopodia, respectively [19]. Considerable evidence, mainly obtained using constitutively-active or dominant-negative mutants, supports a promotional role of Rac1 and Cdc42 and an inhibitory role of RhoA in cell protrusion [19],[21]. Although some researchers have challenged this widely-accepted notion in a variety of cell contexts [22]–[24], our current study has been motivated by this predominant view.
The objective of this study was to uncover the relationship between spatio-temporal activities of Rho GTPases and morphological changes of the cells. To achieve this, we needed a data analysis tool to assess the link between biochemical signaling and biophysical phenomena. However, we do not focus on unveiling the orchestration of the complete signaling pathways that regulate cell morphology. In addition, we elucidated how Rho GTPases regulate “two-dimensional” morphological changes of cells, rather than “three-dimensional” changes. These findings will however be meaningful because the results can be compared with earlier findings [25]–[28]. Therefore, we first present an algorithm called edge evolution tracking (EET) to quantify local morphological change. The main features of our method are that (1) identification of a local morphological change is based on an area difference between two consecutive frames; (2) cell edge is not characterized by point markers, but by line segments, which are defined by the area difference; and (3) past history and future evolution of each segment can be evaluated by connecting segments between consecutive frames. Therefore, this method enables us to trace complex cell edge extension and contraction while maintaining the consistency of the ROI during the analysis. Second, applying EET to fluorescence resonance energy transfer (FRET) time-lapse images of three Rho GTPases (Rac1/Cdc42/RhoA), we found a significant time-shifted cross-correlation between morphological change and GTPase activity. Our study reveals the utility of detailed cellular morphodynamic profiling and spatio-temporal signal profiling to measure the time-shifted relationship between morphodynamics and protein activity.
The EET algorithm describes the time course of local cell morphological changes based on area differences of sequential images. We focused on the local area change, rather than the local structural change as a morphological property; therefore, EET analysis did not make clear distinctions between filopodia and lamellipodia. Subdivided regions along the cell edge boundaries are connected to the corresponding subdivided regions in the next frame, and movements of the subdivided regions are then defined by these connected subregions. Thus, the subdivided regions called “segments” are basic units in EET for quantification of morphological changes. EET describes the time course of local protrusion and retraction as follows:
These connected anchor points indicate the spatial associations between neighboring time frames, and allow us to trace the corresponding regions along the time course by means of the graph structure, which represents the lineage of the segments along the time course. A flow chart of the EET procedure above is shown in Figure 3D.
It should be noted that EET defines how the ancestral segments of a certain segment at a certain time behave along the time course (Figure 3E). Because the definition of segments depends on area differences, if a cell becomes transiently immobile the subdivided regions fuse into a single, and hence integrated, edge. In such a case, integration can be avoided by an exceptional operation that maintains the anchor points during the period of immobility. This procedure keeps the spatial resolution (number of segments) of EET without artificial bias as far as used for immobile anchor points, because the average activity of a single segment and that of its divided segments are the same, and the area differences are always 0. Generally, however, continuous fluctuation is observed along the whole edge, and it is therefore possible to extract a sufficient number of subdivided regions to be analyzed. Actually, this exceptional operation is not used when analyzing the data in this manuscript. Although threshold parameters for the binarization in preprocessing affect the extraction of cell boundaries and area differences, the results of EET are consistent once the threshold parameters have been determined, even if cells show highly fluctuating behavior.
Local activity along a cell boundary is defined as the mean FRET ratio inside a circle, which has its center on the cell boundary and radius r. This is equivalent to using a smoothing filter with a kernel size of r. In EET, the representative activity of a segment is defined as the mean of local activities in the segment. We thus obtain a vector of activity a, composed of the representative activities within each segment in time-lapse images. In polar coordinate-based and marker-tracking based analyses, on the other hand, local activity denotes the mean activity inside a circle, whose center is located at the intersection of a cell boundary and a radial axis or a marker position, and whose radius is r. Therefore, local activity is defined in EET in a manner that is conceptually similar to that in the polar coordinate-based and the marker-tracking-based methods; however, the EET analysis is performed segment-by-segment, which is statistically more stable than the polar coordinate-based and marker-tracking-based methods. The activity profile at time N, obtained from N images of time-lapse activity data, is denoted by aN.
We calculated cross-correlation coefficients between local area changes and activities based on the defined segments. Vector data {a(t)|t = 1, …, N−1} and {d(t)|t = 1, …, N} denote the activities and area differences of the segments extracted from the first to N−1 and the first to N frames of the same image sequence, respectively. a(t) and d(t) represent local activities at time t and local area differences between times t and t+1, respectively. According to Pearson's product-moment correlation coefficient, the correlation function R({a(t)},{d(t)},N) is defined aswhere i and t are indices for segments denoting positions along a cell boundary and time (frames), respectively, and Mt denotes the number of segments at frame t. Note that our EET defines the activity in a segment-wise manner, and therefore a(t) and d(t+1) have the same dimensionality Mt. Because the histogram of the activities in the segments was found to be approximated as a normal (Gaussian) distribution but with a heavy tail in some samples, samples whose activity exceeded 3σ (where σ is the standard deviation) were removed to avoid disproportionate influences of outliers on the correlation coefficients. When the data distributions diverged from the Gaussian, we also calculated Spearman's rank correlation coefficient, which is independent of the shape of the sample distributions, to verify the results of the Pearson's correlation coefficient. Spearman's rank correlation function Rs({a(t)}, {d(t)}, N) is defined aswhere , pj is the number of rank j samples of {d(t)|t = 1, …, N}, nd the number of ranks in {d(t)|t = 1, …, N}, pa the number of rank k samples of {a(t)|t = 1, …, N−1}, and na the number of ranks in {a(t)|t = 1, …, N−1}. Because the EET calculates the cross-correlations based on segments, it is insensitive to the physical size of segments; that is, the cross-correlation coefficients indicate event-wise correlations between molecular activities and morphological changes over the whole cell edge.
We investigated τ time-shifted cross-correlation between activities and area differences to incorporate the time lag between molecular events and morphodynamics. Because the ancestry relationship between a single segment in a focused frame and segments in another frame is not one-to-one (Figure 3C), we defined the transition matrix At,t+τ so that the τ-shifted area difference d(t,t+τ) could be defined. Because the graph structure was obtained under the basic assumption that each local event is defined in terms of ‘segment’, a morphological property, we calculate the τ-shifted values only for the area differences. A series of the corresponding area differences by sequential τ, for example, d(t,t+1), d(t,t+2), d(t,t+3),…, denotes the time course of edge evolution; d(t,t+τ) is defined below. The transition between Mt segments at time t and Mt+1 segments at time t+1 is represented by an Mt×Mt+1 matrix At,t+1, which consists of 0 and 1 denoting unconnected and connected segments, respectively, in the ancestry graph (Figure 3C). Because the column dimensionality of the transition matrix at time t and the row dimensionality of the transition matrix at time t+1 are the same as the number of segments between time t and time t−1, the transition matrix between time t and t+u can be calculated algebraically asThis means that each component is substituted by one if the matrix calculation results in a positive value. Corresponding area changes from time t to time t+τ are then expressed as:The i-th element of d(t,t+τ) denotes the summation of area differences among the segments at t+τ, which are ancestral to the i-th segment at t, according to the ancestry graph. In Figure 3C, for example, d(T) = {la, ab, br} and d(T+1) = {lc, cd, de, ef, fr}, where each element in the sets denotes an area difference (typically, a number of pixels). The transition matrix is given by:Then, d(T+1,T) = (AT,T+1 (d(T+1))’)’ = {lc+cd+de, de+ef+fr, fr}, where the addition is applied to the area difference values. Based on these time-shifted corresponding area differences, a one-to-one relationship between the segments in different frames is constructed. The cross-correlation coefficient with a time-shift of τ is thus obtained by calculating R ({a(t)}, {d(t, t+τ)}, N−τ).
In this study, cell boundaries and area differences were all extracted from fluorescence time-lapse images. To emphasize the cell edges, the images were filtered with an unsharp mask (implemented by the image-processing software MetaMorph [Universal Imaging, Sunnyvale, CA]), which subtracts a low-pass filtered and scaled image from its original image. The Gradient Anisotropic Diffusion filter [29],[30] was then applied to smooth edge boundaries for complex cell shapes. After the filtering step, the intracellular and extracellular regions were segmented using the global threshold determined for the first frame. The cell boundary was extracted directly from the outline of the thresholded images. Typically, the extracted cell boundaries were distorted when edge extraction was applied to threshold regions with one-pixel width, such as thin spikes. To avoid this, each pixel in a thresholded image was divided into sub-pixels before extraction of boundaries. Boundary extraction was then executed for each binary image at a sub-pixel resolution. We did not apply spline fitting in EET or polar coordinate-based analysis to avoid spoiling steep edge structures with filopodium-like thin shapes. Area differences were also extracted from the thresholded images. Increased areas were determined by subtracting the current frame from its next frame, while decreased areas were determined by subtracting the next frame from the current one. Most of these procedures, including EET and cross-correlation analysis, were implemented by Matlab (The MathWorks, Natick, MA).
For this study, we used neurite outgrowth of rat pheochromocytoma PC12 cells as an example of cells displaying complex morphological dynamics, while random migration of human fibrosarcoma HT1080 cells was used for analysis of the cross-correlation between morphological changes and Rho GTPase activity. PC12 cells were plated on polyethyleneimine- and laminin-coated 35-mm glass-base dishes (Asahi Techno Glass, Chiba, Japan), and then transfected with pRaichu-1011x encoding Rac1 FRET probe. One day after transfection, the cells were stimulated with 50 ng/ml NGF in phenol red-free Dulbecco's modified Eagle's medium/F12 containing 0.1% bovine serum albumin for 48 h to induce neurite outgrowth. HT1080 cells were transfected with pRaichu-1011x, pRaichu-1054x encoding a Cdc42 FRET probe, or pRaichu-1294x encoding RhoA FRET probe and, after 24 h, cells were plated on collagen-coated 35-mm glass-base dishes. The medium was then changed to phenol red-free Dulbecco's modified Eagle's medium/F12 containing 10% fetal bovine serum, overlaid with mineral oil to prevent evaporation, and image acquisition was started. The cells were imaged with an inverted microscope (IX81 or IX71; Olympus, Tokyo, Japan) equipped with a cooled charge-coupled device camera (Cool SNAP-K4 or Cool SNAP-HQ; Roper Scientific, Duluth, GA), and a laser-based auto-focusing system at 37°C. The filters used for the dual-emission imaging were purchased from Omega Optical (Brattleboro, VT): an XF1071 (440AF21) excitation filter, an XF2034 (455DRLP) dichroic mirror, and two emission filters (XF3075 [480AF30] for CFP and XF3079 [535AF26] for FRET). The cells were illuminated with a 75-W xenon lamp through a 6%, 10% or 12% ND filter and viewed through a 60× oil-immersion objective lens (PlanApo 60×/1.4). The exposure times for 2×2 or 3×3 binning were 400 or 500 ms for CFP and FRET images. After background subtraction, FRET/CFP ratio images were created with MetaMorph software, and the images were used to represent FRET efficiency. Further details of microscopy and sample preparation can be found in previous reports [26],[27].
We executed a permutation test between positive (6 min), negative (−6 min) and non time-shifted correlations according to the following procedure. Letters/numbers in bold fonts represent vectors.
For example, if we have CP = [0.6 0.4 0.6] and CN = [0.3 0.5 0.4], then D_P_N = [0.3 −0.1 0.2]. Each permutated difference vector is an element of the set of possible ones: D_P_Nper∈{[0.3 0.1 0.2], [0.3 0.1 0.2], [0.3 0.1 −0.2], [−0.3 0.1 0.2], [−0.3 −0.1 0.2], [−0.3 0.1 −0.2], [0.3 −0.1 −0.2], [−0.3 −0.1 −0.2]}. Owing to the independence assumption of a sign-change between elements, the permutation (null) distribution is simply obtained by arranging all the possible sequences, whose number is 23 = 8 in the above example, with uniform probability. The permuted difference vectors whose mean is larger than that of the original difference vector are thus {[0.3 0.1 0.2], [0.3 −0.1 0.2]} and number 2. In this particular example, the permutation p-value is then given as p = 2/8 = 0.25. If this p-value is smaller than a specified significance level (usually 5%), the difference between CP and CN is said to be significant. In the case of two-sided permutation test, the significance level is simply divided by 2.
We applied EET to branching PC12 cells to validate its usefulness for quantifying complex cell morphological changes. As shown in Figure 4A, the PC12 cells extended their neurites with branches after treatment with NGF. A time-lapse series (1-min intervals) of the images was trimmed to help maintain visual correspondence with EET profiles because large image sizes may make the visual inspection difficult. We chose the branching region to verify the utility of EET for the complex cell shape. Next, following the EET procedure, we determined the profiles of edge boundary states, as depicted in Figure 4C, in which red, blue and green colors denote protrusive, retractile and pausing states of the cell edge boundary, respectively. Black lines connect the anchor points (see Materials and Methods), and represent the corresponding segments and subdivided regions. Small fragments of the segments show spatially independent and transient behaviors of the edge evolution and contraction, while long segments represent simultaneous occurrence of edge evolution and contraction in neighboring regions during the time lapse. We also monitored global changes in cell morphology using total area and complexity (bottom of Figure 4C), together with the state profiles, because the state profile by itself does not illustrate the global characteristics of cellular morphodynamics. The monitored total areas and complexity represent the balance between the length of the cell edge boundary and the total area. These values will help to identify rough images of morphological changes. To visualize the dynamics of local area differences by EET, an area difference map was constructed as shown in Figure 4D. Despite the complex morphological changes, EET was successful in quantifying detailed local area changes and preserving the positional correspondence among the subdivided edges. For example, the white squared area in Figure 4A showed a slight extension until 20 min and then retraction between 30–50 min; this corresponds to the region in the state profile starting from 60–80 mm (ordinate) at 0 min (abscissa) (Figure 4C). This quantification and visualization method reduces the difficulty in dealing with time-lapse image data by summarizing the morphodynamic characteristics into two-dimensional state profiles.
Because previous studies have shown the localization of GTPase activities at peripheral regions [26], we applied EET to motile HT1080 cells to further quantify the relationship between local morphological changes and local GTPase activity. First, we imaged motile HT1080 cells with a 1-min time-lapse. Figure 5A shows a series of FRET/CFP ratio images of a single motile HT1080 cell expressing Raichu-1011x (Rac1 probe), and the FRET efficiency is shown in pseudo colors. Based on a previous study indicating a correlation between FRET efficiency and Rac1 activity [26], we assumed that the Rac1 activity should be well represented by the FRET efficiency. The time-lapse images reveal the wandering behavior of the HT1080 cell and a spatio-temporal activity pattern of Rac1 within the cell. To emphasize the protruded and retracted areas in consecutive frames, each image was first transformed into a binary image by extraction of the cell and background regions. The consecutive subtracted images were then obtained frame by frame, and the protrusion and retraction areas were colored in red and blue, respectively (Figure 5B). As reported previously, the coincidence of morphological changes with increases in Rac1 activity was seen by comparing the FRET and subtracted binary images (Figure 5A and 5B).
Next, we applied EET to precisely examine the spatio-temporal relationships between morphological changes and GTPase activities in motile HT1080 cells. As with PC12 cells (Figure 4C and 4D), the state profile and local area difference map were acquired (Figure 5C and 5D). Simultaneously, we acquired the local activity map (Figure 5F) based on segment-wise local activity (Figure 5E, Materials and Methods). This time-position map of the local GTPase activity corresponds to both the state profile and the local area difference map (Figure 5C, 5D, and 5F). There appeared to be similar patterns between the local area difference map and the local activity map. The area difference map revealed chunks of persistently protruding or retracting regions at the cell periphery, while the activity profile revealed spatially and temporally associated activity patches at the cell boundary, suggesting that their dynamics correlated with each other.
Visual inspection of the local area difference map (Figure 5D) and local activity map (Figure 5F) helped us to detect patterns of cell morphology and GTPase activity. The upper left area of Figure 5D shows that formation of large lamellipodia (between 6–20 min) was preceded by the local retraction of the cell edge, and this retraction-extension pattern was also identified in other cell types (data not shown). Cell edge retraction has the potential to induce tension-dependent development of molecular activities involving Rho GTPase signaling [31]. Our data are consistent with this mechanosensory function and provides a possible mechanism for interactions between morphological changes and molecular signaling. On the other hand, the large retraction between 12 and 18 min (Figure 5D) was preceded by a local decrease in Rac1 activity (blue zone in Figure 5F at 10–12 min) and similar patterns were also observed in other cells (data not shown). Potentially, the local decline in Rac1 activity may contribute to the subsequent cell-edge retraction. In addition, in contrast with the morphological changes, the local activity map revealed that the GTPase activity changed moderately at the same position. This moderate change may help maintain the stability of the polarity.
We further investigated this spatio-temporal cross-correlation between morphological changes and Rho-family GTPase activity. First, we summarized their statistical characteristics to examine the cross-correlation. Figure 6A shows a scatter plot of the local activity and the local area difference for all identified segments. Because there were no non-linear relationships in this plot, we considered that common statistical analyses could be applied to these data. Next, we examined the histograms of the activity and area difference and found that the activities had a Gaussian distribution (Figure 6B); heavy tails were observed in some samples, but not in the area differences (Figure 6C). Although the activity histograms of a few samples exhibited one or two minor peaks in addition to the major peaks (data not shown), we assumed that they could still be approximated by Gaussian distribution for simplicity; in subsequent analyses, we used both Pearson's product-moment correlation coefficient and Spearman's rank correlation to confirm the cross-correlation data.
We next examined the effects of time-shifts on cross-correlation between the activity and the area difference. The graphical structures of EET profiles display local area differences in the corresponding time-shifted segments. The middle panels of Figure 6D show time-shifted local area difference maps with various time-shift values. Different patterns appeared on the area difference map depending on the time-shifts, showing that the correlation changes depend on the time-shift values. The scatter plots of activity without time-shift against time-shifted area differences show a linear relationship for negative values of the time-shift (Figure 6D upper).
We calculated time-shifted cross-correlations between the local activities of Cdc42/Rac1/RhoA and local morphological changes, as shown in Figure 7. As expected, there were strong correlations between Cdc42/Rac1 activities and morphological changes, but the peaks of the correlation coefficients were slightly time-shifted. Moreover, and surprisingly, the peaks indicated that the local morphological changes preceded changes in local activity, which can be seen in Figure 6D. We confirmed statistical significance of the difference between negative (−6 min), zero and positive (+6 min) time-shifts by performing permutation tests (see Table S1). The number of samples used to calculate the cross-correlations was sufficiently large (see Figure S2 and Figure S3). Although there are some conspicuous morphological events seen in the EET profile (Figure 5C), such as the protrusion around 6–16 min and the retraction around 12–18 min, the cross-correlation based on the EET analysis was designed to be robust against such local events arising in limited sites in the cell. In this specific case of Rac1 activity in HT1080 cell, our finding that the cross-correlation profile is highly correlated with minus time-shift values is unchangeable, even when these conspicuous morphological events are replaced by normal morphological events (see Figure S4). Note that the Spearman's rank correlation also reduces the bias effect of large values (events) on statistical values.
The results do not appear to be intuitive with regard to the causal relationship between morphological changes and molecular signaling; upstream molecular signaling should control downstream morphological changes, for example via actin reorganization, adhesion and/or retrograde flow. In the cases of both Rac1 and Cdc42, the time-shifted correlations showed that morphological change preceded local GTPase activity. Cdc42 activity, in particular, showed large deviations when the preceding time-shifts were short, and the correlation decayed steeply when the time-shifts were longer. Rac1 activity, on the other hand, elicited small deviations and the decay of the correlation was less steep when the preceding time-shifts were longer. It should be noted that the time-shifted correlation generally approaches zero over long time-shifts owing to an increase in the number of connections between the original segment and time-shifted segments (see Figure 3E). This reflects a weakened relationship, i.e., not one-to-one but one-to-multi relationship between the original region (segment) and its time-shifted regions (connected segments). However, this weakened relationship does not imply a decrease in the reliability of calculations of time-shifted coefficients by making vague relationships between time-shifted segments, but instead represents the natural dilution of the correspondence between an original region and its time-shifted regions.
We further examined the spatial property of the relationship between GTPase activity and morphology change by comparing the original EET profile with rotated (see Figure S5A) and permutated (see Figure S5B) segments of EET profiles. EET profiles of rotated segments showed a decreased correlation with increased rotation (see Figure S5C). Because the segments have a range of lengths along the cell edge, EET did not directly show an exact proximity. However, it showed the significance of the locality of morphodynamic regulation signal. The signal locality dependency was also shown by a lack of correlation of the permutated segments profile with EET.
We compared EET analysis to polar coordinate-based analysis to further prove the utility of EET. We first performed polar coordinate-based analysis to the cell in Figure 5 for direct comparison with EET (Figure 8A). The polar coordinate-based analysis produced time-position maps of local activities and local morphological changes that were similar to the activity map and area difference map of EET (see Figure S6). As for EET, local activity was determined as a mean value within an ROI, which was a circle of radius r. We used the same r value for EET and the polar coordinate-based method. Both analyses produced similar maps (see Figure S6), and time-shifted cross-correlations were then calculated (Figure 8B). Both of the time-shifted cross-correlations showed similar patterns for the timing between local morphological changes and GTPase activities (Rac1), i.e., a high correlation with negative time-shifts and a low correlation with positive time-shifts. However, the EET analysis showed a higher correlation than that with the polar coordinate-based analysis at the time-shifts of −3 to −20 min.
A similar tendency was observed when a population of the cells in Figure 7 was analyzed by the polar coordinate-based method (Figure 8C). The averaged peaks of cross-correlations obtained by the polar coordinate-based analysis were substantially lower than those obtained with EET, particularly for Cdc42 and Rac1 (Figures 7 and 8C). Permutation tests revealed significant differences between the time-shifted cross-correlations by the polar coordinate-based analysis (see Table S1). This might be due to the relatively large correlation values at the time-shift of zero. However, the variances were small, and the correlations prominently decreased when the time-shift value was far from zero. Statistical tests generally showed significant differences between two groups when the variance of each group was small. Here, the small variances in the correlations are likely to be obtained by averaging a large number of samples with small values, and the small values may be due to inconsistency in position alignment between different frames. Note that the polar coordinate-based analysis acquired a large number of samples at 1-degree intervals (i.e., 360 samples in each image) from a single cellular edge and that adjacent samples were likely to have similar values because of physical edge continuity. Our EET implemented the sensitivity to detect correlations between activities and morphological changes by maintaining a consistent position between consecutive frames in terms of segments. Thus, we believe that the correlation peak at the time-shift of zero, obtained by the polar coordinate-based analysis, could be an artifact stemming from position misalignment.
We also compared EET analysis with simple implementation of marker-tracking-based analysis. In this marker-tracking-based analysis, virtually defined markers were aligned uniformly along the spline-fitted cellular edge in the first frame of time-lapse FRET images. Then, the movements of markers in the direction perpendicular to the cellular edge during a single time-frame were measured according to the current marker position and the intersection of the perpendicular axes of the current cellular edge and the next cellular edge (Figure 9A and 9B). Figure 9A and 9B show time-lapse cellular edges of the same cell as in Figure 5, colored from blue (6 min) to red (11 min), with virtually defined markers (black dots) and movements of the markers (black lines). Topological violations of the markers (crossing the black lines) are indicated in Figure 9B, which is probably due to the highly complex morphological changes in the edges. Such complex changes could affect the marker movement maps (although the map obtained by the marker-tracking-based method was comparable to that obtained by EET and by polar coordinate-based analysis; see Figure S6), but our statistical analysis was not affected. Instead, the changes in marker distribution from a uniform (black dots on the blue line in Figure 9B) to a non-uniform alignment (black dots on the red line in Figure 9B) would have non-negligible influences on the time-shifted statistical analysis (e.g., Figure 1C). As with EET and the polar coordinate-based method, the local activity was determined as a mean value within an ROI, which was a circle of radius r. We used the same r value in EET, the polar coordinate-based and the marker-tracking-based methods. All analyses produced similar maps (see Figure S6), and time-shifted cross-correlations were then calculated (Figure 9C).
The time-shifted cross-correlations in Figure 9C show lower correlations at the negative time-shifts compared with EET. The marker-tracking-based analysis produced similar patterns of time-shifted cross-correlations for Cdc42 and Rac1 (Figure 9D) and the permutation tests revealed significant differences between the correlations at zero, and the negative and positive time-shifts (see Table S1). Similar to the polar coordinate-based method, the marker-tracking-based analysis revealed weaker characteristics in the time-shifted cross-correlations. This seems to result from biased sampling by the non-uniform marker distribution caused by morphological changes, which can be seen in Figure 9B. Thus, we suggest that the marker-tracking-based analysis has undesired affects on the statistical analysis, particularly when the cellular edge has a persistent deforming property.
We have developed an algorithm called EET, which describes changes in cell morphology using time-lapse live cell imaging. Spatio-temporal area difference maps revealed morphodynamic properties as patterns of extension and retraction, and the correspondence between time-shifted segments, achieved using anchor points, ensured that the related subdivided edges were connected between time-shifted frames. Therefore, EET effectively accounts for complex morphodynamics that include persistent extension or retraction, and arborization. This property is realized by the graphical representation of edge evolution, and ensures EET is suitable for depicting changes in cell shape, such as the branching that occurs during neural development. Application of EET to the extending neurites of PC12 cells provided a clear evidence of its utility by precisely revealing the persistent protrusion and retraction patterns. Besides, a second application to motile HT1080 cells illuminated distributions of local area differences and corresponding local activity of GTPases. Although the graph structure itself potentially generates biases when correlating the one-to-multi segments between temporally distant frames, we confirmed that our results were consistent even when we obtained our result differently by associating the area change in each segment with the average molecular activities over the corresponding segments (see Figure S7). Because cellular morphological changes have probabilistic characteristics [32], the statistical analysis approach used here is a powerful tool for exploring the nature of dynamic processes in cellular behaviors.
It has been established that Rho-family GTPases (Rac1, Cdc42 and RhoA) play key roles in morphological changes through cytoskeletal reorganization [19], [33]–[35]. Furthermore, previous FRET imaging studies have shown that these GTPases are exquisitely regulated spatio-temporally [25],[26],[28],[36]. In this study, we obtained additional results with EET analysis. In particular, the activities of Rac1 and Cdc42 were localized around the peripheral regions and strongly correlated with the preceding changes in the local area, while the local activity of RhoA was only weakly correlated with changes in the local area. The activity of Cdc42 immediately preceding to the activity of Rac1 is consistent with earlier finding, suggesting that Rac1 is activated by active Cdc42 [37], while the difference in time-shifted cross-correlations between RhoA and Cdc42/Rac1 (Figure 7) would supports the existence of feedback loops common to Rac1 and Cdc42. However, the relationship between RhoA activity and morphology remains controversial [25],[38]. Quantitative analyses in different experimental conditions will clarify this issue. Our results, however, should prompt further investigation of the role of GTPase in regulation of morphodynamics, because this challenges the hypothesis that Rac1 and Cdc42 promote extension of lamellipodia or filopodia, respectively.
The precise mechanism by which local area changes precede local activity around the cell boundary remains unclear from our current analysis. However, we speculate four possible mechanisms based on our results. The first explanation is the existence of upstream signaling molecules that regulate extension in parallel with GTPase activity. If the reactions of the signaling cascades involved with extension are faster than those linked to GTPase activation, extension could precede GTPase activity. In this respect, it would be interesting to conduct a study similar to the current one for PI3K, which activates many signaling molecules including Rac1 activators [39].
The second explanation is that protrusion site-specific stimulation activates the GTPases. There are several mechanisms by which physical force can be converted into biochemical responses [40], and a theoretical study has suggested that signaling activity might be affected by cell shape [41]. In addition, we have shown that there is a positive feedback loop from actin polymerization to Rac1/Cdc42 activation via PI3K [39]. Therefore, it is possible that the detected increase in Rac1/Cdc42 activation was, in fact, secondary to actin polymerization at the protruding regions.
The third possibility is that signaling crosstalk regulates the timing of extension and retraction [42]. If the GTPase activity induces extension and also activates factors that promote edge retraction, the peak GTPase activity appears to be delayed with morphological changes by balancing with activated retraction promoter.
The fourth possibility is the existence of different mechanisms for cell edge extension. EGF-stimulated initial protrusion in MTLn3 rat adenocarcinoma cells is caused by cofilin activation and severing of F-actin, which is coincident with actin polymerization and formation of lamellipodia [43]. On the other hand, Rac1-dependent edge expansion is followed by stabilization of the protrusions [44]. Further investigations will enable us to determine which hypothesis (including coexistence) is most likely with the observed phenomena. In addition, the effects of the dynamics in the perpendicular axis such as changes in cell thickness and volume should be determines, because our results are restricted to the horizontal dynamics. Probe-related mechanisms should also be considered carefully; for example, the difference in the expression levels between the FRET probes and the endogenous Rho GTPases might affect the timing and dynamics of activation of GTPases.
Quantitative analysis of live cell microscopy images is invaluable for better understanding of the dynamic properties of processes such as chemotaxis and development. Such quantitative data can go beyond descriptions of the dynamic features of cellular behavior to serve as a scaffold for theoretical study and to enhance system-level understanding. Based on quantitative data acquired by polar coordinate-based analysis of neurons, for example, Betz et al. discussed a bistable stochastic process derived from velocity histograms and calculated potential distribution [32]. Therefore, connecting modeling studies with quantitative experimental studies has the potential to yield breakthroughs in system-level understanding of cellular functions [45]–[47]. The EET method allows us to quantify details of morphological dynamics of cells. Moreover, it also enables to investigate the spatio-temporal relationship between morphological dynamics and local molecular signaling dynamics. Further application of EET to other signals, e.g., different species of GTPases such as Ras and upstream signals of Rho GTPases such as PI3K, and also to localization of actin should shed light on some of the dynamic and complex properties of regulation of the morphological/migratory systems in cells.
|
10.1371/journal.pntd.0007343 | Zika viruses of African and Asian lineages cause fetal harm in a mouse model of vertical transmission | Congenital Zika virus (ZIKV) infection was first linked to birth defects during the American outbreak in 2015/2016. It has been proposed that mutations unique to the Asian/American-genotype explain, at least in part, the ability of Asian/American ZIKV to cause congenital Zika syndrome (CZS). Recent studies identified mutations in ZIKV infecting humans that arose coincident with the outbreak in French Polynesia and were stably maintained during subsequent spread to the Americas. Here we show that African ZIKV can infect and harm fetuses and that the S139N substitution that has been associated with the American outbreak is not essential for fetal harm. Our findings, in a vertical transmission mouse model, suggest that ZIKV will remain a threat to pregnant women for the foreseeable future, including in Africa, Southeast Asia, and the Americas. Additional research is needed to better understand the risks associated with ZIKV infection during pregnancy, both in areas where the virus is newly endemic and where it has been circulating for decades.
| Zika virus (ZIKV) was first discovered in Uganda in 1947, and is thought to have spread from Africa through equatorial Asia in the middle of the 20th century. Along the way the virus diversified, so that now two genetic lineages, African and Asian/American, are recognized. Congenital Zika syndrome (CZS), the set of fetal and neonatal complications associated with ZIKV infection in pregnancy, was noted during the recent outbreak in the Americas. But the origins of CZS remain a mystery. In particular, it is unclear whether ZIKV recently acquired the ability to cause CZS, perhaps as Asian-lineage viruses spread to the Americas, or whether African-lineage viruses can also cause CZS. To address this question, we used a mouse model of vertical ZIKV transmission to assess pathogenic potential to the fetus of African and Asian/American ZIKV. Our data show that ZIKV of both African and Asian/American lineages can cause fetal harm in the mouse pregnancy model, and that this capacity does not require asparagine at amino acid residue 139, which recently emerged in Asian-lineage viruses and has been suggested to increase ZIKV’s pathogenic potential for fetuses. Our results, therefore imply that ZIKV infection during pregnancy poses a risk for fetal harm in all regions where the virus is endemic.
| Zika virus causes adverse pregnancy outcomes including fetal loss, developmental abnormalities, and neurological damage, collectively termed congenital Zika syndrome (CZS) [1–4]. Why does CZS seem like a new complication when ZIKV has been circulating in Africa and Asia for decades? A provocative explanation for the recent appearance of CZS is that, during their geographic spread from Asia to the Americas, contemporary ZIKV strains acquired mutations that enhance neurovirulence and/or transplacental transmission. In several arboviruses, simple point mutations are known to result in changes in host range and/or the efficiency of infection and replication in key amplification hosts or vectors (see [5] for review). Accordingly, a single serine-to-asparagine substitution in the premembrane (prM) protein of ZIKV (S139N) that is unique to the Asian/American lineage viruses has been postulated to increase neurovirulence and contribute significantly to the microcephaly phenotype [6].
Yuan et al. [6] recently demonstrated that S139N substantially increased ZIKV infectivity in both human (in vitro) and mouse (in vivo) neural progenitor cells (NPCs), leading to restricted brain growth in an ex vivo embryonic mouse brain model, as well as higher mortality rates in neonatal mice following intracranial (i.c.) inoculation. Zhang et al. [7] reported similar findings, suggesting that American strains of ZIKV have a greater capacity for neurovirulence. However, accumulating data suggest that in historically endemic areas in Africa and Southeast Asia, ZIKV has always been teratogenic [8–12]. The degree to which the capacity to cause fetal harm is an emergent property unique to ZIKV circulating in the Americas, as well as the extent to which neurovirulence in laboratory models correlates with risk of fetal harm, remain open questions.
We aimed to better understand the hypothesized recent emergence of CZS by investigating whether both Asian- and African-lineage strains have the capacity to cause CZS. Using a vertical transmission model in mice, we assessed fetal outcomes after infection at embryonic day E7.5 by African- and Asian-lineage ZIKV strains, as well as the impact of the S139N substitution on the severity of gestational infection. We found that all ZIKV strains in our model caused adverse fetal outcomes, suggesting that the capacity of ZIKV to cause CZS does not map principally to polyprotein residue 139 and is not a newly acquired property.
This study was approved by the University of Wisconsin-Madison and University of Minnesota, Twin Cities Institutional Animal Care and Use Committees (Animal Care and Use Protocol Numbers V5519 (UW) and 1804–35828 (UMN)).
African Green Monkey kidney cells (Vero; ATCC #CCL-81) were maintained in Dulbecco’s modified Eagle medium (DMEM) supplemented with 10% fetal bovine serum (FBS; Hyclone, Logan, UT), 2 mM L-glutamine, 1.5 g/L sodium bicarbonate, 100 U/ml penicillin, 100 μg/ml of streptomycin, and incubated at 37°C in 5% CO2. Aedes albopictus mosquito cells (C6/36; ATCC #CRL-1660) were maintained in DMEM supplemented with 10% fetal bovine serum (FBS; Hyclone, Logan, UT), 2 mM L-glutamine, 1.5 g/L sodium bicarbonate, 100 U/ml penicillin, 100 μg/ml of streptomycin, and incubated at 28°C in 5% CO2. The cell lines were obtained from the American Type Culture Collection, were not further authenticated, and were not specifically tested for mycoplasma. ZIKV strain PRVABC59 (ZIKV-PR; GenBank:KU501215), originally isolated from a traveler to Puerto Rico in 2015 with three rounds of amplification on Vero cells, was obtained from Brandy Russell (CDC, Ft. Collins, CO). ZIKV-PR served as the backbone for the reverse genetic platform developed by Weger-Lucarelli et al. [13] upon which the single-amino acid substitution- N139S- was introduced. ZIKV strain DAK AR 41524 (ZIKV-DAK; GenBank:KX601166) was originally isolated from Aedes luteocephalus mosquitoes in Senegal in 1984, with a round of amplification on Aedes pseudocutellaris cells, followed by amplification on C6/36 cells, followed by two rounds of amplification on Vero cells, was obtained from BEI Resources (Manassas, VA). Virus stocks were prepared by inoculation onto a confluent monolayer of C6/36 mosquito cells. ZIKV strain FSS 13025 (ZIKV-CAM; GenBank:JN860885), originally isolated from a child in Cambodia in 2010 with three rounds of amplification on Vero cells, was obtained by Brandy Russell (CDC, Ft. Collins, CO). Virus stocks were prepared by inoculation onto a confluent monolayer of Vero cells.
An infectious clone for ZIKV-PR was constructed as previously described [13]. Infectious-clone derived virus (ZIKV-PR-IC) was recovered following electroporation of in vitro transcribed RNA into Vero cells. To engineer the N139S substitution into the ZIKV genome, the corresponding single-amino acid substitution was introduced into the ZIKV-PR-IC using the in vitro assembly cloning method [14]. The infectious clone plasmids were linearized by restriction endonuclease digestion, PCR purified, and ligated with T4 DNA ligase. From the assembled fragments, capped T7 RNA transcripts were generated, and the resulting RNA was electroporated into Vero cells. Infectious virus was harvested when 50–75% cytopathic effects were observed (6 days post transfection; ZIKV-N139S). Viral supernatant then was clarified by centrifugation and supplemented to a final concentration of 20% fetal bovine serum and 10 mM HEPES prior to freezing and storage as single use aliquots. Titer was measured by plaque assay on Vero cells as described in a subsequent section. We deep sequenced all of our challenge stocks (both wildtype and infectious clone-derived viruses) to verify the expected origin and amino acid at residue 139 (see details in a section below). All ZIKV stocks had the expected amino acid at residue 139: ZIKV-PR-IC (N), ZIKV-DAK (S), ZIKV-CAM (S), ZIKV-PR-N139S (S). Importantly, no single nucleotide polymorphisms were detected at residue 139 at a frequency greater than 1%, nor did we detect evidence of Dezidougou virus, an insect-specific Negevirus present in some ZIKV DAK AR 41524 stocks.
All ZIKV screens from mouse tissue and titrations for virus quantification from virus stocks were completed by plaque assay on Vero cell cultures. Duplicate wells were infected with 0.1 ml aliquots from serial 10-fold dilutions in growth media and virus was adsorbed for one hour. Following incubation, the inoculum was removed, and monolayers were overlaid with 3 ml containing a 1:1 mixture of 1.2% oxoid agar and 2X DMEM (Gibco, Carlsbad, CA) with 10% (vol/vol) FBS and 2% (vol/vol) penicillin/streptomycin. Cells were incubated at 37 °C in 5% CO2for four days for plaque development. Cell monolayers then were stained with 3 ml of overlay containing a 1:1 mixture of 1.2% oxoid agar and 2X DMEM with 2% (vol/vol) FBS, 2% (vol/vol) penicillin/streptomycin, and 0.33% neutral red (Gibco). Cells were incubated overnight at 37 °C and plaques were counted.
Viral RNA was extracted from sera using the Viral Total Nucleic Acid Kit (Promega, Madison, WI) on a Maxwell 48 RSC instrument (Promega, Madison, WI). Viral RNA was isolated from homogenized tissues using the Maxwell 48 RSC Viral Total Nucleic Acid Purification Kit (Promega, Madison, WI) on a Maxwell 48 RSC instrument. Each tissue was homogenized using PBS supplemented with 20% FBS and penicillin/streptomycin and a tissue tearor variable speed homogenizer. Supernatant was clarified by centrifugation and the isolation was continued according to the Maxwell 48 RSC Viral Total Nucleic Acid Purification Kit protocol, and samples were eluted into 50 μl RNase free water. RNA was then quantified using quantitative RT-PCR. Viral load data from serum are expressed as vRNA copies/mL. Viral load data from tissues are expressed as vRNA copies/tissue.
For ZIKV-PR, vRNA from serum and tissues was quantified by QRT-PCR using primers with a slight modification to those described by Lanciotti et al. to accommodate African lineage ZIKV sequences [15]. The modified primer sequences are: forward 5’-CGYTGCCCAACACAAGG-3’, reverse 5’-CACYAAYGTTCTTTTGCABACAT-3’, and probe 5’-6fam-AGCCTACCTTGAYAAGCARTCAGACACYCAA-BHQ1-3’. IUPAC nucleotide codes are as follows: Y: C or T; B: C or G or T; R: A or G. The RT-PCR was performed using the SuperScript III Platinum One-Step Quantitative RT-PCR system (Invitrogen, Carlsbad, CA) on a LightCycler 480 instrument (Roche Diagnostics, Indianapolis, IN). As a one-step assay, this assay uses the reverse primer to prime both reverse transcription and PCR and therefore detects positive-sense viral RNA. The primers and probe were used at final concentrations of 600 nm and 100 nm respectively, along with 150 ng random primers (Promega, Madison, WI). Cycling conditions were as follows: 37°C for 15 min, 50°C for 30 min and 95°C for 2 min, followed by 50 cycles of 95°C for 15 sec and 60°C for 1 min. Viral RNA concentration was determined by interpolation onto an internal standard curve composed of seven 10-fold serial dilutions of a synthetic ZIKV RNA fragment based on a ZIKV strain derived from French Polynesia that shares >99% similarity at the nucleotide level to the Puerto Rican strain used in the infections described in this manuscript.
Six-well plates containing confluent monolayers of Vero cells were infected with virus (ZIKV-PR-IC or ZIKV-PR-N139S), in triplicate, at a multiplicity of infection (MOI) of 0.01 PFU/cell. After one hour of adsorption at 37°C, the inoculum was removed and the cultures were washed three times. Fresh media were added and Vero cell cultures were incubated for 5 days at 37°C, with aliquots removed daily, diluted 1:10 in culture media, and stored at −80°C. Viral titers at each time point were determined by plaque titration on Vero cells and viral loads were determined by QRT-PCR.
Female Ifnar1-/- mice on the C57BL/6 background were bred in the specific pathogen-free animal facilities of the University of Wisconsin-Madison Mouse Breeding Core within the School of Medicine and Public Health. Male C57BL/6 mice were purchased from Jackson Laboratories. Timed matings between female Ifnar1-/- mice and male C57BL/6 mice resulted in Ifnar1-/+ progeny. Untimed, pregnant BALB/c mice were purchased from Charles River.
All pregnant dams were between six and eight weeks of age. Littermates were randomly assigned to infected and control groups. Matings between female Ifnar1-/- dams and wildtype sires were timed by checking for the presence of a vaginal plug, indicating a gestational age E0.5. At embryonic day 7.5 (E7.5), dams were inoculated in the left hind foot pad with 103 PFU of ZIKV in 25 μl of sterile PBS or with 25 μl of sterile PBS alone to serve as experimental controls. All animals were closely monitored by laboratory staff for adverse reactions and signs of disease. A single sub-mandibular blood draw was performed 2 days post inoculation and serum was collected to verify viremia. Mice were humanely euthanized and necropsied at E14.5.
Following inoculation with ZIKV or PBS, mice were sacrificed at E14.5. Tissues were carefully dissected using sterile instruments that were changed between each mouse to minimize possible cross contamination. For all mice, each organ/neonate was evaluated grossly in situ, removed with sterile instruments, placed in a sterile culture dish, photographed, and further processed to assess viral burden and tissue distribution or banked for future assays. Briefly, uterus was first removed, photographed, and then dissected to remove each individual conceptus (i.e, fetus and placenta when possible). Fetuses and placentas were either collected in PBS supplemented with 20% FBS and penicillin/streptomycin (for plaque assays) or fixed in 4% PFA for imaging. Crown-rump length was measured by tracing distance from the crown of the head to the end of the tail using ImageJ. Infection-induced resorbed fetuses (~61%) were excluded from measurement analyses because they would not survive if the pregnancy was allowed to progress to term. We characterized an embryo as in the resorption process if it met the following criteria: significant growth retardation compared to litter mates and controls accompanied by clearly evident developmental delay, i.e., morphology was ill defined; or visualization of a macroscopic plaque in the uterus [16].
Tissues were fixed in 4% paraformaldehyde for 24 hours and transferred into cold, sterile DPBS until alcohol processed and embedded in paraffin. Paraffin sections (5 μm) were stained with hematoxylin and eosin (H&E). Pathologists were blinded to gross pathological findings when tissue sections were evaluated microscopically. The degree of pathology at the maternal-fetal interface was rated on a scale of 0–4: 0 –no lesions (normal); 1 –mild changes; 2 –mild to moderate changes; 3 –moderate to severe changes; 4 –severe. The final scores were determined as a consensus score of three independent pathologists. Two patterns of injury were identified ranging from small focal or multifocal lesions to larger geographic areas of pathology. Each zone was assigned a quantitative score based on the following findings by each pathologist: 0—no lesions; 1 (mild) - 1–2 focal lesions or 5–10% of zone involved by pathology; 2 (mild to moderate) - 3–4 focal lesions or 10–15% of zone involved by pathology; 3 (moderate to severe) - 4–6 focal lesions or 15–25% of zone; and 4—greater than 6 focal lesions or >25% of zone involved by pathology. The majority of the placentas classified as severe (score of 4) usually had larger geographic areas of pathology exceeding 25%. The scoring of the 3 pathologists was within +/- 1 category for all placentas examined. For each zone in the placenta (myometrium, decidua, junctional zone, labyrinth, and chorionic plate/membranes) a ‘General’ overall score was determined, a score for the amount of ‘Inflammation’, and a score for direct ‘Vascular Injury’. The ‘General’ score was based on an interpretation of the overall histopathologic findings in each placenta, which included features of necrosis, infarction, hemorrhage, mineralization, vascular injury, and inflammation. The ‘Inflammation’ score quantified the amount of inflammation in that layer. The ‘Vascular Injury’ score assessed vascular wall injury (fibrinoid necrosis, endothelial swelling), dilatation of the vessels or spaces, and intraluminal thrombi. The myometrial layer representing the uterine wall and the chorionic plate/membranes were often not present in histologic sections and therefore meaningful comparisons between strains could not be assessed. The decidual layer (maternal in origin), the junctional zone composed of fetal giant cells and spongiotrophoblast, and the labyrinth layer (the critical layer for gas and nutrient exchange between the fetal and maternal vascular systems) were scored. Photomicrographs were obtained using a bright light microscope Olympus BX43 and Olympus BX46 (Olympus Inc., Center Valley, PA) with attached Olympus DP72 digital camera (Olympus Inc.) and Spot Flex 152 64 Mp camera (Spot Imaging), and captured using commercially available image-analysis software (cellSens DimensionR, Olympus Inc. and spot software 5.2).
To test ZIKV strain neurovirulence, one-day-old BALB/c mice were intracranially (i.c.) inoculated at the lambda point with 10 PFU of virus or PBS alone. Each strain was tested in at least three litters. Following i.c. inoculation, mice were monitored twice daily for 28 days. Average survival time and percent mortality were calculated. Mice that succumbed within 24 hours of i.c. inoculation were excluded from further analyses. Mice that survived to 28 days were weighed and humanely euthanized. Survival curves are representative of two independent experiments, data were combined, and study site was included as a stratification factor in the analyses.
Virus populations replicating in mouse sera were sequenced in duplicate using a method adapted from Quick et. al. [17]. Viral RNA was isolated from mouse sera using the Maxwell 16 Total Viral Nucleic Acid Purification kit, according to manufacturer’s protocol. Viral RNA then was subjected to RT-PCR using the SuperScript IV Reverse Transcriptase enzyme (Invitrogen, Carlsbad, CA). Input viral RNA was 106 viral RNA templates per cDNA reaction. For sera from mice infected with ZIKV-PR-IC and ZIKV-PR-N139S, the cDNA was then split into two multi-plex PCR reactions using the PCR primers described in Quick et. al with the Q5 High-Fidelity DNA Polymerase enzyme (New England Biolabs, Inc., Ipswich, MA). For sera from mice infected with ZIKV-DAK, the cDNA was amplified in a PCR reaction for sequencing of a single amplicon with ZIKV-DAK specific primers (forward 5’-ACCTTGCTGCCATGTTGAGA-3’, reverse 5’CCGTACACAACCCAAGTCGA-3’) using Q5 High-Fidelity DNA Polymerase (New England Biolabs, Inc., Ipswich, MA). PCR products were tagged with the Illumina TruSeq Nano HT kit and sequenced with a 2 x 250 kit on an Illumina MiSeq.
A vial of the viral stocks used for primary challenge (ZIKV-PR-IC, ZIKV-PR-N139S, ZIKV-DAK, ZIKV-CAM), were each deep sequenced by preparing libraries of fragmented double-stranded cDNA using methods similar to those previously described [18]. Briefly, the sample was centrifuged at 5000 rcf for five minutes. The supernatant was then filtered through a 0.45-μm filter. Viral RNA was isolated using the QIAamp MinElute Virus Spin Kit (Qiagen, Germantown, MD), omitting carrier RNA. Eluted vRNA was then treated with DNAse I. Double-stranded DNA was prepared with the Superscript Double-Stranded cDNA Synthesis kit (Invitrogen, Carlsbad, CA) and priming with random hexamers. Agencourt Ampure XP beads (Beckman Coulter, Indianapolis, IN) were used to purify double-stranded DNA. The purified DNA was fragmented with the Nextera XT kit (Illumina, Madison, WI), tagged with Illumina-compatible primers, and then purified with Agencourt Ampure XP beads. Purified libraries were then sequenced with 2 x 300 bp kits on an Illumina MiSeq.
Amplicon data were analyzed using a workflow we term “Zequencer 2017” (https://bitbucket.org/dhoconno/zequencer/src). Briefly, R1 and R2 fastq files from the paired-read Illumina miSeq dataset were merged, trimmed, and normalized using the bbtools package (http://jgi.doe.gov/data-and-tools/bbtools) and Seqtk (https://github.com/lh3/seqtk). Bbmerge.sh was used to merge reads, and to trim primer sequences by setting the forcetrimleft parameter 22. All other parameters were set to default values. These reads were then mapped to the reference amplicon sequences with BBmap.sh. Reads substantially shorter than the amplicon were filtered out by reformat.sh (the minlength parameter was set to the length of the amplicon minus 60). Seqtk was used to subsample to 1000 reads per amplicon. Quality trimming was performed on the fastq file of normalized reads by bbmap’s reformat.sh (qtrim parameter set to ‘lr’, all other parameters set to default). Novoalign (http://www.novocraft.com/products/novoalign/) was used to map each read to the appropriate ZIKV reference sequence: ZIKV-PRVABC59 KU501215, ZIKV DAK AR 41524 KX601166, ZIKV FSS13025 JN860885. Novoalign’s soft clipping feature was turned off by specifying the parameter “-o FullNW”. Approximate fragment length was set to 300bp, with a standard deviation of 50. We used Samtools to map, sort, and create an mpileup of our reads (http://samtools.sourceforge.net/). Samtools’ base alignment quality (BAQ) computation was turned off; otherwise, default settings were used. SNP calling was performed with VarScan’s mpileupcns function (http://varscan.sourceforge.net/). The minimum average quality was set to 30; otherwise, default settings were used. VCF files were annotated using SnpEff [19]. Accurate calling of end-of-read SNPs are a known weakness of current alignment algorithms [20]; in particular, Samtools’ BAQ computation feature is known to be a source of error when using VarScan (http://varscan.sourceforge.net/germline-calling.html). Therefore, both Novoalign’s soft clipping feature and Samtools’ BAQ were turned off to increase the accuracy of SNP calling for SNPs occurring at the end of a read.
Viral stock sequences were analyzed using a modified version of the viral-ngs workflow developed by the Broad Institute (http://viral-ngs.readthedocs.io/en/latest/description.html) implemented in DNANexus and using bbmap local alignment in Geneious Pro (Biomatters, Ltd., Auckland, New Zealand). Briefly, using the viral-ngs workflow, host-derived reads that map to a human sequence database and putative PCR duplicates were removed. The remaining reads were loaded into Geneious Pro and mapped to NCBI Genbank Zika virus reference sequences using bbmap local alignment. Mapped reads were aligned using Geneious global alignment and the consensus sequence was used for intra sample variant calling. Variants were called that fit the following conditions: have a minimum p-value of 10e-60, a minimum strand bias of 10e-5 when exceeding 65% bias, and were nonsynonymous.
All analyses, except for deep sequencing analysis, were performed using GraphPad Prism. For survival analysis, Kaplan-Meier survival curves were analyzed by the stratified (by study site) log-rank test. Unpaired Student’s t-test was used to determine significant differences in crown-rump length, and viral loads of fetuses versus placentas. Fisher’s exact test was used to determine differences in rates of normal vs. abnormal concepti. One-way ANOVA with Tukey’s multiple comparison test was conducted to compare virus titers in maternal serum and to compare viral loads in placentas, fetuses, and concepti.
To characterize the range of pathogenic outcomes of congenital ZIKV infection and to assess the role of S139N on an alternate genetic background, we engineered the reverse amino acid substitution (asparagine reverted to serine at residue 139 in the viral polyprotein) into the Puerto Rican ZIKV isolate PRVABC59 (ZIKV-PR-N139S). Prior to use in mice, we assessed viral infectivity and replication of ZIKV-PR-N139S in vitro using Vero cells. ZIKV-PR-N139S and a control virus derived from an infectious clone bearing the wild type ZIKV-PRVABC59 consensus sequence (ZIKV-PR-IC; N at residue 139) gave similar growth curves (Fig 1A). These results suggest that the “reverse substitution” N139S did not have a significant effect on either infectivity or replicative capacity in vitro. Next, to assess whether N139S, in the context of the PRVABC59 genome, decreased mortality in the neonatal mouse model, we inoculated one-day-old BALB/c mice i.c. with 10 PFU of either ZIKV-PR-IC; ZIKV-PR-N139S; a ZIKV strain isolated in Cambodia in 2010 (ZIKV-CAM; FSS 13025; S at residue 139); a low-passage African ZIKV strain isolated in Senegal in 1984 (ZIKV-DAK; DAK AR 41524; S at residue 139); or, as a control, phosphate-buffered saline (PBS). Surprisingly, and in contrast to the results described by Yuan et al., i.c. inoculation of ZIKV-DAK and ZIKV-CAM resulted in 100% and 64% mortality, respectively, whereas 42% of mice succumbed to ZIKV-PR-IC and 33% to ZIKV-PR-N139S by 28 days post inoculation (dpi; Fig 1B). ZIKV-PR-IC and ZIKV-PR-N139S survival curves did not significantly differ (stratified log-rank test p-value = 0.1753), and all strains caused significant mortality (Fisher’s exact test) by 28 dpi as compared to the PBS-inoculated controls (ZIKV-CAM, ZIKV-DAK:p-value < 0.0001; ZIKV-PR-IC: p-value = 0.002; ZIKV-PR-N139S: p-value = 0.018). Additionally, pups from each treatment group that survived to 28 dpi were weighed to assess evidence for growth restriction (Fig 1C). Weight followed parallel trends observed with mortality, and pups in all ZIKV groups were too small to be weaned from dams at the standard 21-day timepoint. Weights from all ZIKV groups were significantly lower (Student’s t-test) than the PBS group (ZIKV-PR-IC: p-value < 0.0001, t-value = 4.93, df = 21; ZIKV-PR-N139S: p-value < 0.0001, t-value = 5.81, df = 24; ZIKV-CAM: p-value < 0.0001, t-value = 12.97, df = 20), ZIKV-PR-IC and ZIKV-PR-N139S did not differ significantly (p-value = 0.252, t-value = 1.17, df = 27), and the weights of surviving ZIKV-CAM pups were significantly lower than ZIKV-PR-IC (p-value < 0.0027, t-value = 3.36, df = 23) but not ZIKV-PR-N139S (p-value = 0.073, t-value = 1.87, df = 26). No ZIKV-DAK pups survived to 28 days to be included in these analyses. It is possible that the differences in survival in our study relative to Yuan et al. may be due to the use of a specific strain (PRVABC59 vs. GZ01). That is, the impact of a single amino acid substitution like S139N may vary in different strain backgrounds. However, these results are consistent with other studies in pregnant animal models that have provided evidence of neurovirulence and fetal demise caused by ZIKV strains isolated before the American outbreak [21–24]. Another possible explanation for the disparity in outcomes with ZIKV-CAM relative to other studies may be that previous studies only tested mutations at codon 139 in isolation on an otherwise isogenic background [6]. Likewise, others have used clone-derived viruses to understand overall ZIKV-CAM (strain FSS13025) pathogenesis [7,25,26]. Instead we directly compared pathogenic potential of a natural isolate of ZIKV-CAM. For some RNA viruses, swarm diversity can impact pathogenesis through cooperative interactions in the viral population [27] and this may not occur with clone-derived viruses or to the same degree as natural isolates.
The previous experiments establish the ability of each ZIKV strain tested to cause lethal infections in neonates, but direct intracranial inoculation does not fully recapitulate the events of a natural congenital infection. To better compare the abilities of these ZIKV strains to induce birth defects following vertical transmission, we used a previously established murine pregnancy model for ZIKV [22,28], in which dams lacking type I interferon signaling (Ifnar1-/-) were crossed with wildtype sires to produce heterozygous offspring. Because they have one intact Ifnar1 allele, these offspring more closely resemble the immune status of human fetuses. Time-mated dams were inoculated subcutaneously in the footpad with 103 PFU of ZIKV-PR-IC, ZIKV-PR-N139S, or ZIKV-DAK on E7.5, corresponding to the mid-to-late first trimester in humans [29]. We omitted ZIKV-CAM in the vertical transmission experiment due to previous experiments demonstrating its ability to cause fetal harm in Ifnar1-/- [22]. However, it should be noted that those experiments used intravaginal inoculation of ZIKV-CAM to examine the effects of gestational ZIKV infection, which precludes direct, quantitative comparisons to the results we describe here using subcutaneous inoculations. We collected serum samples from dams at 2 dpi to confirm infection and to sequence viral populations replicating in vivo. All dams were productively infected and one-way ANOVA with Tukey’s multiple comparisons test was conducted to compare titer between treatment groups. Maternal viremia was not significantly different in ZIKV-PR-IC- and ZIKV-PR-N139S-inoculated animals (p-value = 0.845), whereas ZIKV-DAK replicated to significantly higher titers on 2 dpi (ZIKV-PR-IC vs. ZIKV-DAK: p-value = 0.0002; ZIKV-PR-N139S vs. ZIKV-DAK: p-value < 0.0001). Deep sequencing of virus populations replicating in maternal serum confirmed that the N139S mutation was stably maintained in vivo (Table 1). Dams were monitored daily for clinical signs until time of necropsy. Overt clinical signs were only evident in ZIKV-DAK-inoculated dams and included hunched posture, ruffled fur, and hind limb paralysis indicative of neurotropism. All ZIKV-DAK-infected dams met euthanasia criteria at time of necropsy on E14.5.
Next, to assess fetal outcomes, ZIKV-inoculated dams were sacrificed at E14.5. Gross examination of each conceptus (both fetus and placenta, when possible) revealed overt differences among fetuses within pregnancies and with uninfected counterparts. In general, fetuses appeared either grossly normal or abnormal, defined as being prone for embryo resorption (Fig 2A–2C) [16]. At time of necropsy, we observed high rates of resorption in both ZIKV-PR-IC- and ZIKV-PR-N139S-infected pregnancies. The proportion of abnormal fetuses for the two strains did not differ significantly (53.2% vs. 67.3%, Fisher’s exact test p-value = 0.21). In contrast, ZIKV-DAK-infected pregnancies resulted in 100% resorption of fetuses (Fig 2A). Only fetuses that appeared grossly normal were included for measurement of crown-rump length (CRL) to provide evidence for intrauterine growth restriction (IUGR). There was a modest reduction in size in grossly normal ZIKV-PR-IC fetuses. Mean CRL did not differ significantly (Student’s t-test) between fetuses of ZIKV-PR-IC- or PBS-inoculated dams (p-value = 0.22, t-value = 1.23, df = 54), whereas there was a statistically significant (Student’s t-test) reduction in mean CRL between fetuses whose dams were inoculated with ZIKV-PR-IC vs. ZIKV-PR-N139S (p-value < 0.0001, t-value = 5.42, df = 34; Fig 2B). This lack of apparent IUGR for ZIKV-PR-IC is contrary to other studies using Asian-lineage ZIKVs in which fetuses developed severe IUGR [22,28,30]. Again, this disparity may be the result of differences in timing of challenge and necropsy [31], subtle phenotypic differences in virus strain, dose and/or route of inoculation, or metrics for defining grossly normal fetuses compared to those undergoing resorption at a later embryonic age. Critically, the rates of resorption between ZIKV-PR-IC and ZIKV-PR-N139S were not significantly different and are consistent with the rates reported by Miner et al. [28]. These data also are consistent with a recent report suggesting that fetal demise may be a more common outcome of ZIKV infection than previously recognized [24]. Because our experiments only investigated the S139N mutation on the ZIKV-PRVABC59 background we cannot exclude the possibility that other SNPs, singly or in combination, may play functionally redundant roles with S139N in promoting neurovirulence. It is important to note that all studies seeking to link viral genotype to phenotype using reverse genetics can be limited by the genetic context of the cloned virus(es) chosen for analysis. In our case, we do not observe a significant difference in pathogenic potential of ZIKV encoding serine or asparagine at polyprotein position 139. But we examined this particular point mutation in a single genetic background. It is possible that S139N may act in concert with other polymorphisms to affect ZIKV neurotropism, neurovirulence, and/or other factors, such that S139N mutations examined in the context of different ZIKV strains might have different phenotypic effects. Nonetheless, our results, together with others, clearly show that ZIKV of multiple genetic lineages can cause fetal harm irrespective of the amino acid residue present at position 139.
To confirm vertical transmission of ZIKV to the developing conceptus, viral loads were measured from representative placentas and fetuses from each litter of all treatment groups by quantitative RT-PCR (Fig 2D–2F). vRNA was detected in all fetuses and placentas that were tested. Viral loads were significantly higher (Student’s t-test) in placentas than in fetuses (p-value < 0.0001, t-value = 5.04, df = 95; Fig 2D and 2E), whereas placental viral loads were not significantly different between groups infected with different viruses nor among littermates within the same litter (ZIKV-PR-IC vs. ZIKV-PR-N139S: p-value = 0.063, df = 47; ZIKV-PR-IC vs. ZIKV-DAK: p-value = 0.43, df = 47; ZIKV-PR-N139S vs. ZIKV-DAK: p-value = 0.64, df = 47; one-way ANOVA with Tukey’s multiple comparisons). In contrast, ZIKV-DAK infected fetuses had significantly higher viral loads when compared to ZIKV-PR-IC and ZIKV-PR-N139S (ZIKV-PR-IC vs. ZIKV-DAK: p-value = 0.04, df = 44; ZIKV-PR-N139S vs. ZIKV-DAK: p-value = 0.04, df = 44; one-way ANOVA with Tukey’s multiple comparisons). Although day 2 viremia was significantly higher in ZIKV-DAK-infected dams, the fact that fetal and placental viral loads are broadly similar, both within and across treatment groups indicates that it is not simply a matter of enhanced replication of the African virus in the dam causing more severe fetal outcomes. Still, we cannot exclude the possibility that the more severe fetal outcomes observed with the African virus are the result of IFNAR-dependent fetal demise [22] singly or in combination with IFNAR-independent causes of birth defects like poor maternal health, direct pathogenic effects of the virus infection (as indicated by a significant, albeit moderate, increase in fetus viral loads), or a bystander effect associated with immune responses unrelated to type 1 interferons. Additionally, comparable viral loads were measured from both grossly normal and abnormal fetuses and placentas. Detection of ZIKV RNA in grossly normal fetuses does not preclude the possibility that pathology may develop later in pregnancy or even postnatally, consistent with reports from humans that the effects of in utero exposure may not be evident at birth [32].
To better understand the impact of in utero ZIKV infection, tissues of the developing placenta and decidua were evaluated microscopically. In PBS-inoculated dams, we observed normal decidua, junctional zone, and labyrinth with normal maternal and fetal blood spaces (Fig 3A–3C). In contrast, ZIKV-inoculated dams displayed varying degrees of placental pathology, including vascular injury involving maternal and/or fetal vascular spaces, infarction (obstructed blood flow), necrosis, inflammation, and hemorrhage (Fig 3D–3F). There also were clear strain-specific differences in the amount of placental pathology, with ZIKV-DAK displaying the most severe histologic phenotype, consistent with gross observations (Fig 3G–3I). These data are consistent with some reports suggesting that African-lineage ZIKVs might have greater systemic virulence in mice than Asian ones (reviewed in [33]). The underlying mechanism responsible for enhanced virulence remains unknown, but it is important to note that many of the available African isolates have undergone passage in suckling mouse brain, whereas ZIKV-DAK used in our experiments has not. Variations in passage history could result in small, but biologically important, genotypic and phenotypic differences. Still, reports with low-passage African isolates also suggest that these African strains display increased virulence in numerous mouse models when compared to Asian isolates (e.g., [34–36]). As a result, maternal illness could be a potential confounder and the phenotype reported here may not necessarily reflect a pregnancy-specific effect. Ultimately, it will be important to determine whether these results can be recapitulated in a translational model that provides a closer representation of the human morphological, developmental, and immune environment at the maternal-fetal interface, e.g., macaque monkeys.
Importantly, we assessed outcomes following only a single infection time point at E7.5 and outcomes may differ depending on the timing of infection. Critically, the placenta acts as a barrier against infections but does not have a definitive structure in mice until the midpoint of gestation: ~E10.5–11.5 [37]. It is possible then that inoculations later during gestation, when there is a more fully developed placental barrier, may exhibit greater resistance to infection and result in less severe pathologic outcomes [38].
Together our data show that infection with ZIKV isolates of either the African or Asian lineages during pregnancy can lead to fetal harm, with varying levels of damage to maternal, placental, and fetal tissues, frequently including death of the developing fetus. Likewise, intracranial inoculation of neonatal mice confirmed a similar neurovirulence phenotype across ZIKV lineages. A recent study identified the potential capacity of an ancestral Asian strain to cause fetal brain infection after maternal infection in mice [39]; that study used a cloned virus created from reverse genetics that corresponded to a publicly available consensus ZIKV genome sequence. To our knowledge, our study is the first to assess a low-passage African isolate, or the S139N substitution, in a vertical transmission model. The observation that a low-passage African ZIKV isolate can cause severe fetal harm suggests that, for decades, ZIKV could have been causing pregnancy loss and birth defects, which were either undiagnosed or attributed to other causes. If this hypothesis is correct, CZS is not a new syndrome caused by a recently emerged ZIKV variant, but rather an old entity that was only recognized in the large-scale American ZIKV outbreak that began in 2014–15. These results provide compelling motivation to re-evaluate hypotheses explaining the emergence of CZS. A lack of thorough surveillance, together with myriad co-circulating pathogens causing febrile illnesses, make understanding both the past and current prevalence of gestational ZIKV infection and any resulting fetal outcomes in Africa challenging [40–42]. Recent seroprevalence studies have now identified low, but consistent circulation of ZIKV in several African [43–46] and southeast Asian [47] countries, indicating a large population potentially at risk. Accurate assessment of the risk posed by ZIKV infection to pregnant women and their babies in both Africa and southeast Asia should be a priority.
|
10.1371/journal.ppat.1002792 | Hyperthermia Stimulates HIV-1 Replication | HIV-infected individuals may experience fever episodes. Fever is an elevation of the body temperature accompanied by inflammation. It is usually beneficial for the host through enhancement of immunological defenses. In cultures, transient non-physiological heat shock (42–45°C) and Heat Shock Proteins (HSPs) modulate HIV-1 replication, through poorly defined mechanisms. The effect of physiological hyperthermia (38–40°C) on HIV-1 infection has not been extensively investigated. Here, we show that culturing primary CD4+ T lymphocytes and cell lines at a fever-like temperature (39.5°C) increased the efficiency of HIV-1 replication by 2 to 7 fold. Hyperthermia did not facilitate viral entry nor reverse transcription, but increased Tat transactivation of the LTR viral promoter. Hyperthermia also boosted HIV-1 reactivation in a model of latently-infected cells. By imaging HIV-1 transcription, we further show that Hsp90 co-localized with actively transcribing provirus, and this phenomenon was enhanced at 39.5°C. The Hsp90 inhibitor 17-AAG abrogated the increase of HIV-1 replication in hyperthermic cells. Altogether, our results indicate that fever may directly stimulate HIV-1 replication, in a process involving Hsp90 and facilitation of Tat-mediated LTR activity.
| Fever is a complex reaction triggered in response to pathogen infection. It induces diverse effects on the human body and especially on the immune system. The functions of immune cells are positively affected by fever, helping them to fight infection. Fever consists in a physiological elevation of temperature and in inflammation. While the role of inflammatory molecules on HIV-1 replication has been widely studied, little is known about the direct effect of temperature on viral replication. Here, we report that hyperthermia (39.5°C) boosts HIV-1 replication in CD4+ T cells. In single-cycle infection experiments, hyperthermia increased HIV-1 infection up to 7-fold. This effect was mediated in part by an increased activation of the HIV-1 promoter by the viral protein Tat. Our results also indicate that hyperthermia may help HIV-1 to reactivate from latency. We also show that the Heat Shock Protein Hsp90, which levels are increased at 39.5°C, mediates in a large part the positive effect of hyperthermia on HIV-1 infection. Our work suggests that in HIV-1-infected patients, fever episodes may facilitate viral replication.
| Fever is a physiological process induced by endogenous pyretics (IL-6, IL-1β, TNFα) in response to stresses such as pathogen infection. It consists in hyperthermia, an elevation of the body temperature to 38–40°C, associated with an inflammatory state. Fever is generally beneficial for the host, triggering multiple events that lead to the strengthening of immunological defenses. For instance, hyperthermia increases dendritic cells (DC) maturation, migration and antigen presentation [1]. Hyperthermia positively impacts cytokine and antibody production by lymphocytes, and enhances their migration to lymph nodes [2], [3]. Hyperthermia also intensifies cytotoxic activity of Natural Killer cells and phagocytosis by macrophages [4], [5]. Together, these events explain why fever is often associated with better disease outcome [2].
Temperature has various consequences on viral replication. Infection at 41°C inhibits the replication of some human viruses such as Poliovirus, Herpes Simplex Virus type 1 and Measles Virus [6]. Heat shock inhibits Vesicular Stomatitis Virus and Mayaro Virus replication [7], [8]. In contrast, hyperthermia promotes infection by Rotavirus, Dengue Virus, Epstein-Barr Virus, Human Cytomegalovirus and plant viruses [9], [10], [11], [12], [13].
HIV-1-infected patients can experience fever at various stages of the disease. During acute infection, HIV-1 replication is intense, viral loads reach very high levels, and patients are subjected to fever and strong inflammation. Opportunistic infections, which are frequent at the final stages of AIDS, can also induce fever. They directly impact HIV-1 replication, and treating them significantly reduces viral loads [14]. Several millions of HIV-1-positive patients, the majority of which not receiving any treatment, also suffer from tuberculosis or malaria [15], [16]. The two causative pathogens induce fever episodes, and are associated with increased HIV-1 viral loads [17], [18], [19], [20], [21]. Fever may thus modify the environment for HIV-1 replication, either in a positive or a negative way. The relative contribution of direct effects of co-infecting pathogens, inflammation, and elevated temperature to this process is not clearly understood. The role of inflammation on HIV-1 pathogenesis has been widely documented [22], [23], [24], [25]. Inflammation and immune activation represent a driving force for CD4+ T cell depletion, facilitation of viral replication, and AIDS progression [23], [24], [25]. Immune activation also likely impacts the establishment of viral persistence [26]. In culture, pro-inflammatory cytokines such as IL-1β, IL-6 and Tumor Necrosis Factor α (TNFα) favor HIV-1 replication [27], [28], [29], [30]. Knowledge about the role of temperature on HIV-1 replication remains fragmented. Previous research has mainly been focused on heat shock, a transient and non-physiological treatment (a few minutes to a few hours) at 40–45°C, rather than on hyperthermia, an incubation at 38–40°C for up to a few days. Heat shock stimulates HIV-1 production in latently infected cell lines [31], [32] and in Peripheral Blood Mononucleated Cells (PBMCs) [33]. An increased temperature also enhances plasma membrane fluidity and might facilitate viral entry [34]. A precedent work, aimed at characterizing thermosensitive HIV-1 integrase mutants, did not report any viral increase at 7 days post infection (p.i.), at 39.5°C, using a high Multiplicity of Infection (MOI), in CEM cells [35].
Heat Shock Proteins (HSPs) are induced in response to stresses and act as chaperones, helping the folding of proteins and preventing their aggregation. HSPs are overexpressed in many types of tumoral cells, like multiple myeloma, breast or prostate cancer cells [36], [37], [38]. HSPs also have important roles in both innate and adaptive immunity [39]. Extracellular HSPs stimulate cytokine production by DCs. HSP-bound peptides can be endocytosed by DCs and other Antigen Presenting Cells through interaction with the CD91 receptor, and participate to antigen cross-presentation [39]. These properties led to the use of HSPs as adjuvants in vaccine development [40]. Several studies indicated that HIV-1 induces the synthesis of some HSPs such as Hsp27 or Hsp70 [41], [42]. Hsp70 is found in virions and might interfere with Vpr functions [42], [43], [44], [45], [46]. Hsp40 and Hsp70 have been suggested to regulate Nef activity [47]. Recently, it has been reported that Hsp90 and the transcription factor HSF-1 can both increase HIV-1 transcription [48], [49]. Hsp90 was also shown to rescue the impaired replication of ritonavir-resistant viruses [50]. In these studies, the role of HSPs in HIV-1 replication was mainly investigated through over-expression or depletion experiments, whereas the direct impact of temperature on HSPs levels and HIV-1 infection was not characterized. Overall, the importance of HSPs for HIV-1 replication is still not fully elucidated, and the precise role of temperature remains unknown.
We examined here the effect of physiological, fever-like temperature (39.5°C) on several steps of HIV-1 life cycle. We show that hyperthermia enhanced HIV-1 replication in primary CD4+ lymphocytes and cell lines. Transactivation of the Long Terminal Repeat (LTR) promoter by Tat, and viral reactivation in a model of latency, were both more potent at 39.5°C than at 37°C. Hyperthermia increased Hsp90 co-localization with actively transcribing HIV-1 provirus, suggesting a direct effect of this cellular protein on viral gene expression. Our study suggests that fever episodes may promote HIV-1 replication in infected individuals.
We first analyzed the effect of hyperthermia on HIV-1 replication in Jurkat lymphoid cells and in primary CD4+ T lymphocytes. After 2 hours of infection at 37°C, cells were cultivated either at 37°C or 39.5°C (Fig. 1A). Viral spread was then followed by measuring the appearance of Gag+ cells by flow-cytometry at different time points. A representative experiment in Jurkat cells, using different MOI (0.1 and 1 ng Gag p24/ml/106 cells), indicates that HIV-1 replication was more rapid and efficient at 39.5°C than at 37°C (Fig. 1B). In a compilation of independent experiments, hyperthermia increased HIV-1 replication by 4 fold (Fig. 1C). In some experiments, when a higher MOI was used, the differences between the two temperatures were attenuated, probably because viral replication reached saturation levels at both 37°C and 39.5°C (not shown). Similar results were obtained upon infection of primary CD4+ T cells (Fig. 1D), even if the effect of hyperthermia was less marked (2.5 fold increase, Fig. 1E). Depending on the experiments, the peak of infection was either higher (Fig. 1D, donor 1) or occurred earlier (Fig. 1D, donor 2) under hyperthermic conditions. Thus, hyperthermia increases HIV-1 replication in both Jurkat and primary CD4+ T cells.
We then examined how cells responded to hyperthermia in the absence of infection. The growth and viability of primary CD4+ T cells and Jurkat cells were not detectably influenced by hyperthermia (Fig. 2A and not shown). Hyperthermia significantly increased the amount of Hsp70 and Hsp90 in CD4+ T cells, as assessed by Western Blot (Fig. 2B). The surface levels of viral receptors (CD4 and CXCR4), Major Histocompatibility Complex class I (MHC-I) and HLA-A2, the activation marker CD69, and adhesion molecules ICAM-1 and LFA-1 (CD11a and CD18 chains), were similar at 37°C and 39.5°C (Fig. 2C). This suggests that hyperthermia does not profoundly alter the behavior of CD4+ T lymphocytes, with non-specific consequences on HIV-1 replication.
We then asked if hyperthermia affected infection with single-cycle HIV-1. Jurkat cells were exposed to increasing doses of Δenv(VSV), a VSV-G-pseudotyped, env-deleted HIV-1 strain. Gag levels were assessed 24 hours later by flow-cytometry (Fig. 3A). At various viral inputs, the fraction of Gag+ cells was about 3 times greater at 39.5°C than at 37°C. Similar results were observed in P4C5 cells, a reporter HeLa cell line expressing CD4, CCR5, and harboring an HIV-1 LTR-lacZ cassette. Hyperthermia increased single-cycle infection with Δenv(VSV), assessed by measuring β-Galalactosidase activity (Fig. 3B) or Gag levels (not shown). We then tested a panel of viral strains with different tropisms. Hyperthermia increased infection by either the ×4 strain NL4-3, the R5 strain NLAD8, or the dual-tropic primary isolate 132W, by 4 to 7 fold in P4C5 cells (Fig. 3C). Therefore, the effect of hyperthermia is independent of the route of entry and of co-receptor usage. We then performed a temperature “titration” experiment. We infected P4C5 cells at temperatures ranging from 37°C to 40°C (Fig. 3D). Infection began to increase at 38°C (almost 2 fold), and was further enhanced at 39°C, 39.5°C and 40°C (6–7 fold).
In contrast, hyperthermia did not augment transduction of P4C5 and Jurkat cells with an HIV-1 lentivector (LV-GFP) (Fig. 3E). With this lentivector, GFP expression is driven by the CMV promoter, independently of Tat, suggesting that hyperthermia may induce its effects during or after HIV-1 transcription.
To gain further insight into the underlying mechanism of hyperthermia-mediated increased infection, we examined several steps of HIV-1 replication cycle. We first asked whether hyperthermia affects the half-life of cell-free viral particles after release. We incubated virions (produced at 37°C) at 37°C or 39.5°C for various periods of time (15 min to 24 hours). The infectivity of the viral preparations was then measured on P4C5 reporter cells. Incubation at 37°C led to a rapid drop of infectivity, with a half-life of about 2.5 hours. This half-life was slightly, but not significantly, decreased at 39.5°C (Fig. S1A).
A previous study suggested that a higher temperature of infection enhances membrane fluidity and might thus facilitate viral entry [34]. To quantify HIV-1 entry in Jurkat cells, we used a virion fusion assay, which allows to discriminate cytoplasmic access of viral cores from endosomal capture [51], [52], [53]. This assay consists in the use of viruses containing a β-lactamase-Vpr (Blam-Vpr) protein chimera. The successful cytoplasmic access of Blam-Vpr as a result of fusion, after 2 hours of infection, is monitored by the enzymatic cleavage of CCF2-AM, a fluorogenic substrate of β-lactamase loaded in target cells. This assay was previously validated as being linear and able to detect differences in the 2-fold range [51], [52]. A typical experiment, with two different MOI, is represented in Fig. 4A. Using this system, we did not detect any significant effect of hyperthermia on viral fusion and access to the cytoplasm (Fig. 4B).
We then asked if HIV-1 reverse transcription was influenced by hyperthermia. Using an enzymatic in vitro cell-free assay, we monitored the reverse transcriptase (RT) activity of HIV-1 virions. Elevation of the temperature from 37°C to 39.5°C did not affect the enzymatic activity of RT (Fig. 4C), which is consistent with earlier studies [54], [55], [35]. To quantify reverse transcription in the context of infected cells, Jurkat cells were exposed to single-cycle virus, at two MOI. The levels of “early” and “late” RT products were determined by quantitative PCR at 8 hours p.i.. As expected, levels of viral DNA products correlated with the viral input, and the addition of Nevirapine (NVP), an RT inhibitor, almost completely blocked viral DNA synthesis (Fig. 4D). Hyperthermia had no effect on the levels of early or late RT products in infected cells, at 8 hours p.i. and other time points (Fig. 4D and not shown). Together, these results strongly suggest that hyperthermia acts after viral access to the cytoplasm and reverse transcription.
The lack of effect of a high temperature on LV-GFP transduction (Fig. 3) suggested a possible impact of hyperthermia on HIV-1 transcription. To test this hypothesis, we transfected HeLa cells carrying an integrated LTR-Luciferase cassette (HeLa LTR-Luc cells), with increasing amounts of a Tat-expressing plasmid (pcDNA-Tat-Flag). HeLa LTRΔTAR-Luc cells, unable to bind Tat, were used as a negative control. After 4 hours of incubation at 37°C, cells were washed to remove excess plasmids and transfection reagents, and grown at either 37°C or 39.5°C for 48 hours. We verified by Western Blot that Tat-Flag was expressed in equal amounts at 37°C and 39.5°C (Fig. 5A). Transactivation of the LTR was then assessed by measuring luciferase activity in cell lysates. Activity of the LTRΔTAR promoter was very low, with or without Tat, and hyperthermia had no effect on this residual activity (Fig. 5B). Tat efficiently stimulated the full length LTR, and this transactivation was significantly higher at 39.5°C than at 37°C (2 fold increase, Fig. 5B). The greater transactivation of the LTR at 39.5°C was not caused by a higher production of transfected Tat (Fig. 5A), by changes in Tat nuclear localization (not shown) nor by a trans-effect of secreted Tat by neighboring cells (Fig. S1B). In contrast to its action on the LTR promoter, hyperthermia did not significantly impact the activity of a CMV promoter, upon transfection of a pCMV-β-Galactosidase reporter plasmid (not shown). Therefore, a more potent activation of the LTR by Tat likely facilitates HIV-1 replication at 39.5°C.
The existence of latent HIV-1 reservoirs is a long-standing issue in the treatment of AIDS. The formation of these reservoirs is not prevented by Highly Active Antiretroviral Therapy (HAART) and drives viral re-emergence if therapy is stopped [56], [57], [58], [59], [60]. Reactivation from latency is a highly regulated process. Earlier studies outlined the impact of the site of integration, the role of pro-inflammatory cytokines, transcription factors like NFκB, and epigenetic modifications such as cytosine methylation [61], [62], [63], [64], [65]. We asked whether hyperthermia could trigger HIV-1 reactivation from latency, directly or in synergy with other stimuli. To this aim, we used the J-Lat 10.6 model [66]. Briefly, J-Lat 10.6 are Jurkat cells, carrying a latent, integrated provirus, where env is deleted and nef replaced by gfp. Without activation, GFP is not produced, but treatment with TNFα, PMA, or other molecules, induces HIV-1 reactivation and GFP expression. As outlined Fig. 6A, J-Lat 10.6 cells were exposed to various stimuli, and grown at 37°C or 39.5°C for 24 hours. Viral reactivation was then followed by measuring the appearance of GFP+ cells by flow-cytometry.
A representative experiment is shown Fig. 6B, and the mean of all experiments in Fig. 6C. Alone, hyperthermia was not sufficient to trigger viral reactivation (<2% of GFP+ cells at both temperatures). As a positive control, we used TNFα, which induced viral reactivation in up to 80% of J-Lat 10.6 cells (Fig. 6). Even used at sub-optimal concentrations, TNFα and PMA induced a similar reactivation at 37°C or 39.5°C (Fig. S2 and not shown), suggesting that these stimulators may be too strong to evidence any difference. We then used a more physiological stimulation, by exposing J-Lat 10.6 cells to conditioned medium from PBMCs. Supernatants from non stimulated PBMCs (but treated with IL-2 to avoid massive cell death), as well as from PHA-activated PBMCs, induced GFP expression in J-Lat 10.6 cells at 37°C (5% and 22% of GFP+ cells, respectively). Noteworthy, supernatants from PHA-activated PBMCs were more potent than that from non-stimulated PBMCs. This suggests that reactivation in J-Lat 10.6 cells is mediated by cytokines or other molecules that are up-regulated in activated PBMCs. Interestingly, PBMCs supernatants were more potent when J-Lat 10.6 cells were incubated at 39.5°C, with a significant 2-fold increase in the levels of GFP-expressing cells. We then directly co-cultivated J-Lat 10.6 cells with PBMCs. Co-culture with IL-2-treated PBMCs efficiently reactivated J-Lat cells, with up to 17% of GFP+ cells (Fig. 6B). Hyperthermia further increased this effect by 2-fold. Co-culture with PHA-activated PBMCs resulted in high levels of cell death (not shown).
There are different J-Lat clones, in which the extent of viral reactivation varies according to the viral integration site [62]. To check that the effect of hyperthermia on viral reactivation was not restricted to the 10.6 clone, we used J-Lat 6.3, J-Lat 8.4 and J-Lat 9.2 cells. These clones are less susceptible to reactivation than the 10.6 clone [62]. Indeed, we did not detect viral reactivation following exposure to PBMCs conditioned medium nor co-culture with PBMCs (not shown). However, stimulation with various doses of TNFα resulted in viral reactivation in J-Lat 6.3, J-Lat 8.4 and J-Lat 9.2 cells, at a weaker level than in J-Lat 10.6 (Fig. S2). Hyperthermia increased modestly but significantly the effect of TNFα in J-Lat 6.3, J-Lat 8.4 and J-Lat 9.2 cells. To achieve similar levels of GFP expression, the concentration of TNFα required at 37°C was two fold greater than at 39.5°C.
Altogether, these results show that PBMCs conditioned medium, as well as a direct co-culture with PBMCs, induce viral reactivation in the J-Lat model. Hyperthermia significantly enhances this phenomenon.
Hsp90 may play a role during HIV-1 infection [48]. Chromatin immunoprecipitation experiments recently revealed that Hsp90 associates with the viral LTR and may regulate gene expression [48]. The underlying mechanisms are only partly understood, and may involve an effect of Hsp90 on chromatin remodeling, to facilitate transcription [48]. Moreover, the chaperone activity of Hsp90 has been reported to promote formation of a functional P-TEFb/Tat/TAR complex [67]. This prompted us to examine whether Hsp90 co-localizes with actively transcribing provirus in HIV-1-expressing cells. We used an original approach to directly visualize viral RNA in living cells. The technique takes advantage of an HIV-1 strain encoding an RNA which includes 24 binding sites for the phage MS2 protein (HIV_Exo_24×MS2) [68], [69], [70]. U2OS cells carrying an integrated HIV_Exo_24×MS2 genome (U2OS HIVexo) allow the visualization of nascent RNA from a single chromatin location [68], [69], [70]. Upon Tat expression, this RNA is synthesized and can be detected by specific high-affinity interaction with the YFP-MS2nls reporter protein. U2OS HIVexo cells were transfected with the YFP-MS2nls reporter, with or without a Tat-expressing plasmid. After an overnight incubation, the transcribing provirus and endogenous Hsp90 were both visualized by immunofluorescence. In the absence of Tat, YFP-MS2nls displayed a diffuse nuclear staining, whereas the Hsp90 signal was mostly detected in the cytoplasm with very little, if any, nuclear localization (not shown). With Tat, the nascent HIV RNA was detected as a single bright spot of YFP-MS2nls within the nucleus (Fig. 7A). Previous studies demonstrated that these spots represent true sites of viral transcription, rather than sites of HIV RNA sequestration [68]. At 37°C, Hsp90 co-localized with YFP-MS2nls in about 27% of the cells in which a nuclear YFP-MS2nls bright spot was visible (Fig. 7B), indicating Hsp90 can be recruited to the viral transcription site in HIV-infected cells. Hsp90 co-localization with HIV transcripts was significantly increased when cells were incubated overnight at 39.5°C, reaching 70% of the cells in which a viral transcription site was visible (Fig. 7B).
We determined further the role of Hsp90 during hyperthermia. We sought to silence Hsp90, but the extent of silencing achieved with various siRNAs or shRNAs was partial, precluding further analysis (not shown). We thus used a well-characterized pharmacological inhibitor of Hsp90, 17-AAG (also known as tanespimycin) [71]. This compound is a geldanamycin-derived molecule, inhibiting the ATPase activity of Hsp90 and blocking various functions of the heat shock protein [71], [72], [73]. Interestingly, 17-AAG is known to reduce HIV-1 replication, but its effects were studied so far at 37°C [48], [50]. We thus examined whether 17-AAG reversed the stimulating effect of hyperthermia on HIV-1 replication. To this aim, P4C5 cells were exposed to increasing doses of 17-AAG (20–250 nM) and infected with the single-cycle Δenv(VSV) virus. These 17-AAG concentrations were chosen because they did not induce obvious cytotoxicity (not shown). This is in agreement with a previous report indicating that the toxic concentration (CC50) is about 2 µM [50]. At 37°C, 17-AAG decreased infection, but this effect was modest, requiring 100–250 nM concentrations of the inhibitor (Fig. 7C). Interestingly, the compound was more active in hyperthermic cells, starting to decrease the infectivity enhancement at 20 nM, and abrogating this positive effect at 100 nM (Fig. 7C).
Therefore, Hsp90 is recruited to HIV-1 transcription sites, and this process occurs more efficiently at 39.5°C. The Hsp90 inhibitor 17-AAG significantly decreases the stimulating effect of hyperthermia on HIV-1 infection.
We report here a positive impact of hyperthermia on HIV-1 replication. Hyperthermia is known to enhance the functions of immune cells and to confer protection against pathogen infection [1], [2], [4], [5], Previous studies on temperature and HIV-1 mostly focused on chronically infected cell lines [31], [32] or used non-physiological heat shock treatment to study viral reactivation from latency (a few minutes at 42–45°C [33]). Here, we report that elevation of temperature to fever-like levels (39.5°C) stimulates HIV-1 replication in primary CD4+ T lymphocytes as well as in Hela and Jurkat cell lines. In single-cycle infection assays, hyperthermia increased HIV-1 infection by 2 to 7 fold. This stimulation was apparently not due to unspecific alterations of cellular metabolism, since cell growth, viability, or surface levels of various molecules were not significantly affected by hyperthermia.
To get insight into how hyperthermia stimulates HIV-1 replication, we compared the efficiency of various steps of the viral life cycle at 37°C and 39.5°C. Viral entry and fusion, measured by the Vpr-β-lactamase assay, were similar at the two temperatures. Enzymes have a range of conditions of pH, salt concentration, and temperature, in which they display optimal activity. We did not observe an effect of hyperthermia on reverse transcriptase, as both in vitro catalytic activity of the enzyme and the levels of viral DNA synthesis in infected cells were unchanged by temperature. We then examined the influence of temperature on the viral transcription step. Hyperthermia did not induce basal LTR activity without Tat. However, in the presence of Tat, hyperthermia lead to a significantly better transactivation of the LTR. This is in line with earlier reports, demonstrating that a transient heat shock at 42°C increases HIV-1 transcription in monocytic cells lines [31], [32]. Noteworthy, the activity of the CMV promoter was not increased at 39.5°C (not shown), suggesting that hyperthermia does not trigger a global increase of cellular transcription. Accordingly, the steady state levels of several cellular proteins (actin, CD4, ICAM-1, MHC-I, etc.) were apparently similar at normal and elevated temperatures.
To characterize the molecular mechanism by which hyperthermia up-regulates HIV-1 infection and transcription, we examined the role of Hsp90. This protein exerts diverse functions in normal and stressed cells, through its ATPase activity and its protein binding domain [72], [73]. It acts as a chaperone for many cellular proteins. Hsp90 assists folding, assembly, intracellular transport, maintenance and degradation of proteins, and regulates cell signaling and cell cycle [74], [75]. Hsp90 is involved in HIV-1 infection at 37°C, regulating viral gene expression [48]. Hsp90 also impacts the replication of other viral species, such as Human Cytomegalovirus, Influenza Virus, Flock House Virus and Hepatitis C Virus [76], [77], [78], [79]. We show here that the levels of Hsp90 are augmented at 39.5°C, in primary lymphocytes and other cells (Fig. 2 and not shown). By using an immunofluorescence technique allowing the visualization of nascent viral RNA in living cells, we demonstrate that, in presence of Tat, Hsp90 can be found in the nucleus, at HIV-1 transcription sites. This localization was rather infrequent at 37°C, but was significantly increased at 39.5°C (27% and 70% co-localization, respectively). Furthermore, 17-AAG, a pharmacological inhibitor of Hsp90, reversed the stimulating effect of hyperthermia on single-cycle infection in P4C5 cells. Altogether, these results point out for a previously uncharacterized role of Hsp90, facilitating HIV-1 transcription and replication at 39.5°C. It will be worth further dissecting how Hsp90 acts on viral transcription at this temperature. One can speculate that the chaperone protein may bind more efficiently to the P-TEFb/Tat/TAR transcription complex [67] and thus increase its activity, and/or may enhance chromatin modeling and accessibility to the viral promoter [48].
Mechanisms regulating HIV-1 gene expression are also involved in viral reactivation from latency [80]. We show here that the conditioned medium from PBMCs induced viral reactivation, in the J-Lat 10.6 model of latently infected T cells. Strikingly, reactivation was more pronounced at 39.5°C than at 37°C. Futures studies will help understanding which cytokines or other molecules produced by PBMCs mediate this effect. For instance, heat shock at 42°C is known to act in synergy with IL-6 to induce viral reactivation in a latently infected monocytic cell-line [31]. It will be of interest to compare the stimulating effect of IL-6 and other cytokines, at normal and fever-like temperatures, not only in J-Lat cells, but also in other models of viral latency (PBMCs from HAART-treated patients, or latently-infected, resting primary CD4+ T cells [81]).
In this study, we have focused our analysis of the effect of temperature on a few key steps of the viral life cycle. We demonstrate that hyperthermia globally facilitates viral replication. At 39.5°C, viral entry, fusion and reverse transcription occur normally, whereas Tat-mediated transactivation of the LTR is significantly more efficient. It has been previously reported that the activity of HIV-1 integrase and protease is not increased at 39.5°C [35]. This does not rule out the possibility that other steps of HIV-1 infection (nuclear import, selection of integration sites in the cellular genome, viral translation, assembly, release, etc.) might be positively or negatively modified at a fever-like temperature.
What is the physiological relevance of our observations? Patients treated with HAART and with controlled viremia can experience transient bursts of HIV-1 replication termed viral blips [82]. Furthermore, co-infections are frequent in HIV-1-positive individuals and are often associated with fever and acute illnesses [83]. For instance, Plasmodium falciparum, the causative agent of malaria, induces recurrent, strong episodes of fever lasting 2–3 days, which correlates with increased viral loads [84]. The origin of these viral blips, or of other more pronounced viral rebounds is likely multi-factorial. Our results suggest that fever may directly stimulate viral replication or reactivation from latent reservoirs, in association with other inflammatory or immunological events.
Jurkat (clone 20), MT4C5, J-Lat (clone 6.3, 8.4, 9.2 and 10.6), PBMCs and primary CD4+ T cells were grown in RPMI 1640 with Glutamax, supplemented with 10% heat-inactivated Fetal Bovine Serum (FBS) and antibiotics. HEK-293T, U2OS HIVexo [68], HeLa, HeLa Tat, and P4C5 cells were grown in Dulbecco modified Eagle medium (DMEM) supplemented with 10% heat-inactivated FBS and antibiotics. For P4C5 cells, a HeLa-derived cell line expressing CD4, CCR5, and harboring a LTR-lacZ reporter cassette, G418 (500 µg.mL−1, Sigma) and Hygromycin B (50 µg.mL−1, PAA) were added to the culture medium. HeLa expressing Tat were grown in presence of Methotrexate (2 µM, Sigma). Primary CD4+ T cells were purified from human peripheral blood by Ficoll centrifugation, followed by immunomagnetic selection (Miltenyi Biotec). The blood was provided by the EFS (Etablissement Français du Sang, the French Official Blood Bank). About 98% of cells were CD4+ CD3+. For activation, primary CD4+ T cells were treated with phytohemagglutinin (PHA, 1 µg.ml−1, Remel Europe LTD) for 24 hours and then cultured in interleukin 2 (IL-2)-containing medium (50 U.mL−1, Abcys). If not stated different, cells were grown at 37°C, 5% CO2. Cells grown at 39.5°C were cultivated in a distinct incubator. Temperature was monitored by 2 different thermometers. HIV-1 strains, including NL4-3 and NLAD8, were produced by calcium-phosphate transfection of HEK-293T cells. The primary isolate 132W [85] was produced by infection of MT4C5 cells. Vesicular stomatitis virus type G (VSV-G) pseudotyped viruses were generated by co-transfection of HEK-293T cells with the NL4-3 provirus and VSV-G expression plasmid (8∶1 ratio). For the production of Blam-Vpr containing viruses, HEK-293T cells were co-transfected with NL4-3 or NL4-3-F522Y proviruses along with a Blam-Vpr expression plasmid (3∶1 ratio), kindly provided by Warner Greene [52], [51]. NL4-3-F552Y provirus encodes a non-fusogenic gp120/g41 Env complex [86]. LV-GFP were produced by co-transfection of HEK-293T cells by the packaging plasmid (R8-2), the GFP plasmid (pTrip-GFP) and VSV-G expression plasmid (5∶5∶1 ratio). Nevirapine (NIH Catalog, n°4666, batch 01990) was used at 25 nM. 17-AAG (Enzo Life Sciences) was diluted in DMSO and used at the indicated concentrations.
Cell surface stainings were performed at 4°C for 30 min with the following monoclonal antibodies (mAbs): ICAM-1 (clone 1H4, Immunotools, dilution 1/20), CD11a (Immunotools, 1/10), CD18 (Immunotools, 1/10), CD69 (clone FN50, BD, 1/30), MHC-I (clone W632, Sigma, 1/100), HLA-A2 (clone BB7.2, BD, 1/50), CD4 (clone SK3, BD, 1/30), CXCR4 (clone 12G5, NIH AIDS Research and Reference reagent Program, 1/100), Goat anti-Mouse Alexa 647 (Invitrogen, 1/400). Isotype-matched mAbs were used as negative controls. Cells were fixed in PBS-paraformaldehyde (PFA, Sigma) 4% for 5 min. For quantification of Gag levels, cells were harvested at indicated time points and fixed in PBS-PFA 4% for 5 min. Cells were washed in PBS and stained for 30 min in PBS containing 1% BSA (Sigma), 50 µg.mL−1 saponin (Sigma), and the anti-Gag antibody (clone KC57, 1/500, Beckman Coulter). Fluorescence was assessed by flow cytometry on a FacsCanto II (BD).
Cells transfected with 20 ng of Tat-Flag were lysed with passive lysis buffer (Promega) and probed by Western Blot for Tubulin (T9026, Sigma, 1/5000) and Tat (Flag antibody, F3165, Sigma, 1/5000). Cells cultured for 8 hours at 37°C or 39.5°C were lysed in PBS Triton 1% in presence of a proteases inhibitors cocktail (Roche) and probed by Western Blot for Actin (clone AC-15, Sigma, 1/5000), Hsp70 (clone W27, Santa Cruz, 1/200) and Hsp90 (clone 68, BD, 1/1000, kindly provided by Jean-Michel Heard).
One day before infection, 8×104 P4C5 cells were plated in 96-well plates. Cells were infected in triplicate with 1 or 5 ng Gag p24 per well. Cells were lysed in PBS 0.1% NP40 5 mM MgCl2 36 hours post-infection, and incubated at room temperature with Chlorophenol Red-β-D-Galactopyranoside (CPRG, Roche) at a final concentration of 3.65 mg.mL−1. 570 nm OD was measured every 15 min. Jurkat cells were infected with 5, 20 or 100 ng Gag p24/mL/106 at 37°C or 39.5°C for 2 hours. Cells were washed in PBS and grown at either 37°C or 39.5°C. Gag or GFP levels were assessed by flow cytometry 24 hours p.i.
Jurkat or primary CD4+ T cells (1×106) were infected with the indicated doses of NL4-3 for 2 hours at 37°C or 39.5°C, washed once in PBS, and grown at 37°C or 39.5°C. Medium was changed every 2–3 days and samples were harvested at the indicated time points. Gag levels were assessed by flow cytometry. To combine results from independent experiments, we have measured the area under the curve at 37°C and 39.5°C. In Jurkat cells, there is no real peak of viral replication, since most of the cells (>80%) may get infected, and then die. In primary CD4+ T cells, a peak of infected cells may be detected, probably because a fraction of the cells is not sensitive to infection. We have thus calculated the area under the curve for all time points in Jurkat, and for time points occurring until the appearance of the peak at 39.5°C in primary CD4+ T cells.
HeLa LTR-Luc cells contain a single copy of integrated HIV-1 LTR-luciferase reporter construct [87] and were transfected with 4 ng or 20 ng of pcDNA-Tat-Flag completed to 2 µg pcDNA3.1, or pcDNA3.1 alone, using JetPEI reagent (Polyplus) in 6 well plates. After 4 hours at 37°C, cells were washed and grown at 37°C or 39.5°C. 48 hours after incubation, cells were lysed using Passive Lysis Buffer (Promega) and luciferase activity was measured according to the manufacturer's protocol (Promega). Luciferase activity was normalized to protein concentration using Bradford assay (Biorad).
Viral entry was assessed by a test adapted from Cavrois et al. [51], [52]. Briefly, 5 or 100 ng Gag p24 of ultracentrifuged virions containing the Blam-Vpr fusion protein were used to infect 1.5×105 Jurkat cells at 37°C or 39.5°C in a minimal volume (100 µL), in presence of 10 µM Hepes and 2 µg.mL−1 DEAE Dextran (Sigma). After 2 hours, cells were washed in cold CO2-independent medium (Invitrogen), without FBS, resuspended in cold CO2-independent medium supplemented with 10% FBS and incubated with the CCF2-AM substrate (CCF2-AM kit, Invitrogen), in the presence of 1.8 mM Probenecid (Sigma), for 2 hours. Cells were extensively washed in cold CO2-independent medium and fixed in PBS-PFA 4% for 5 min. The cleaved CCF2-AM fluorescence (excitation at 405 nm, emission at 450 nm) was then immediately measured by flow cytometry using the DAPI channel, on a FacsCanto II (BD).
NL4-3 virions were lysed in PBS NP40 0.08% and incubated with Retrosys RT Activity Kit (Innovagen) substrates for 1 or 3 hours, according to the manufacturer's indications. RT activity was measured with a alcaline phosphatase readout (405 nm OD).
Jurkat cells were infected with 50 or 250 ng Gag p24 at 37°C or 39.5°C with NL4-3 Δenv(VSV). After 2 hours, cells were washed and grown at 37°C or 39.5°C. Cells were harvested 8 hours p.i. and lysed in AL Buffer (Qiagen) in presence of Proteinase K (Qiagen) for 1 hour at 56°C. Total DNA was extracted from by phenol-chloroform extraction and ethanol precipitation. To remove all traces of plasmidic DNA, samples were treated with DpnI Fast Digest (Fermentas) for 15 min at 37°C. To avoid inhibition of the PCR reaction by DpnI reaction buffer, samples were diluted 40 times in pure water (Gibco). Early RT products (amplicon length: 183 bp) were quantified by real-time PCR with the primers M667 (GGC TAA CTA GGG AAC CCA CTG) and AASM (GCT AGA GAT TTT CCA CAC TGA CTA A) using the the following program: 35 cycles 10 s 95°C, 10 s 57°C, 15 s 72°C. For late RT products (amplicon length: 200 bp), we used the primers M667 and M661 (CCT GCG TCG AGA GAG CTC CTC TGG), with a slightly different PCR program (35 cycles, 10 s 95°C, 15 s 57°C, 15 s 72°C). The number of cells was calculated by quantification of genomic GAPDH (amplicon length: 303 bp), using the same program as late RT products and the following primers: forward (GGG AAA CTG TGG CGT GAT) and reverse (GGA GGA GTG GGT GTC GTT).
J-Lat clones 6.3, 8.4, 9.2 and 10.6, kindly provided by Stéphane Emiliani, have been previously described [66]. Briefly, J-Lat are Jurkat cells with a latent, integrated, and env-deleted provirus, encoding gfp instead of nef. Stimulation of J-Lat cells with recombinant TNFα (Peprotech) or other molecules such as PMA triggers HIV-1 reactivation. PBMCs were isolated from peripheral blood, PHA-activated or treated with IL-2 for 3 days. Supernatants were then collected and used to stimulate J-Lat cells. Serial dilutions of supernatants were tested, as raw supernatants were sometimes toxic. For co-cultures experiments, PBMCs were isolated from peripheral blood, treated for 3 days with IL-2, and co-cultivated with Far-Red labelled J-Lat 10.6 cells at a 1∶4 ratio in IL-2-containing medium. 48 hours after stimulation, cells were harvested, fixed in PBS-PFA 4%. Viral reactivation was followed by appearance of GFP+ J-Lat cells by flow-cytometry.
U2OS HIVexo cells were transfected with a Tat-expressing plasmid to stimulate viral gene expression. The transcribing provirus was detected by MS2 tagging, Hsp90 was detected by immunofluorescence with a monoclonal antibody. YFP-MS2nls, yellow fluorescent protein fused to MS2 with a nuclear localizing signal. U2OS HIVexo cells and plasmids for Tat and EYFP-MS2nls expression were described previously [68]. 2.5×105 cells were plated on coverslips and grown overnight at 37°C. Next day, cells were transfected (Lipofectamine LTX, Invitrogen) with plasmids expressing Tat (50 ng) and YFP-MS2nls (300 ng). After 4 hours cells were washed and shifted to an incubator at 39.5°C in medium containing 25 mM Hepes for 20 hours. Control cells were kept at 37°C also with 25 mM Hepes in sealed flasks. After treatment, cells were fixed and treated for immunofluorescence essentially as described [88] with a mouse mAb (H90-10, Abcam, 1/500). Cells showing a transcription spot were scored for their co-localization with Hsp90 on a confocal microscope Zeiss 510 META.
Statistical tests were performed with GraphPad Prism 5.
|
10.1371/journal.pntd.0000982 | Evaluation of Cost-Effective Strategies for Rabies Post-Exposure Vaccination in Low-Income Countries | Prompt post-exposure prophylaxis (PEP) is essential in preventing the fatal onset of disease in persons exposed to rabies. Unfortunately, life-saving rabies vaccines and biologicals are often neither accessible nor affordable, particularly to the poorest sectors of society who are most at risk and upon whom the largest burden of rabies falls. Increasing accessibility, reducing costs and preventing delays in delivery of PEP should therefore be prioritized.
We analyzed different PEP vaccination regimens and evaluated their relative costs and benefits to bite victims and healthcare providers. We found PEP vaccination to be an extremely cost-effective intervention (from $200 to less than $60/death averted). Switching from intramuscular (IM) administration of PEP to equally efficacious intradermal (ID) regimens was shown to result in significant savings in the volume of vaccine required to treat the same number of patients, which could mitigate vaccine shortages, and would dramatically reduce the costs of implementing PEP. We present financing mechanisms that would make PEP more affordable and accessible, could help subsidize the cost for those most in need, and could even support new and existing rabies control and prevention programs.
We conclude that a universal switch to ID delivery would improve the affordability and accessibility of PEP for bite victims, leading to a likely reduction in human rabies deaths, as well as being economical for healthcare providers.
| Rapid delivery of post-exposure vaccination is essential for preventing the fatal onset of rabies in persons bitten by rabid animals. But for communities most at risk of exposure to rabies (in resource poor countries where domestic dog rabies is still common), post-exposure vaccines are often not affordable and are only available in limited quantities. Several safe and effective regimens for delivery of these vaccines are recommended by WHO, but these are inconsistently implemented and there are no clear recommendations as to which is the best regimen for specific settings. We developed a framework for comparing the cost-effectiveness of different vaccination regimens, including existing approved regimens and new candidates subject to approval, in terms of costs per death averted. We demonstrate that post-exposure vaccination is an extremely cost-effective public health intervention and that changing delivery from intramuscular to intradermal vaccination has considerable benefits. Large savings in the volume of vaccine required to treat the same number of patients could potentially both mitigate vaccine shortages and reduce delays in delivery, enabling wider vaccine distribution, and thus improving the accessibility and affordability of these life-saving vaccines. We also present financing mechanisms that could help subsidize the cost for those most in need, and even support new and existing rabies control and prevention programs, without compromising existing healthcare budgets.
| Rabies is invariably fatal once clinical signs appear but can be readily prevented after exposure with prompt and appropriate post-exposure prophylaxis (PEP) [1]. PEP is therefore the most critical life-saving intervention essential for the prevention of rabies in humans after exposure [2]. In reality, most of the estimated 7 million people exposed to rabies each year live in resource poor countries where life-saving rabies vaccines are not always available or easily affordable [3], [4], [5], [6]. This stark situation contributes to the fact that almost all of the estimated 55,000 annual human rabies deaths occur in Africa and Asia where the virus circulates endemically in domestic dog populations [7].
The WHO-recommended PEP protocol includes immediate wound washing, expeditious administration of rabies vaccine, and for severe categories of exposure, infiltration of purified rabies immunoglobulin (RIG) in and around the wound [8]. RIG is rarely administered in low-income countries because it is expensive (from USD$25 to over USD$200 depending on whether it is of equine or human origin)[7], [9] and in short supply (see the following examples: [3], [4], [5], [10]). Therefore, it is usually only post-exposure vaccination (without RIG) that is administered to protect a bite victim from succumbing to rabies [3]. Several factors affect the likelihood of promptly obtaining and completing PEP vaccination. Vaccine vials cost from USD$7–20 in most low-income countries and multiple vials are required per patient contingent upon the PEP regimen used. In some countries, governments provide vaccine free-of-charge or subsidize its cost, but budgets allocated for this are often insufficient, resulting in shortages or leaving only a few centres with a reliable supply. Alternatively, victims pay for vaccine, but charges are often prohibitive [3]. Costs of travel and or accommodation accumulate according to the number of clinic visits that a patient and, in many cases, an accompanying family member, makes to complete PEP. These considerable indirect costs [7] are affected by vaccine availability, and rise during shortages when patients and families are forced to travel further (often to multiple clinics), wasting time and money [3]. Delays caused by shortages also reduce compliance. All too often victims fail to promptly obtain or complete PEP, which in the worst cases results in rabies deaths [3], [4], [5].
Following WHO approval of intradermal (ID) administration of PEP vaccines [11], there has been significant discussion of the value of ID versus intramuscular (IM) delivery [2], [12], [13], [14], [15], [16], [17]. The main argument is that ID vaccination is more economical because smaller volumes of vaccine can be used to elicit an equivalent immune response (0.1 mL for each ID injection versus one 0.5 or 1 mL vial for each IM injection). The caveat of ID regimens is that vaccine remaining in partially used vials must be discarded within 6 to 8 hours to minimize risks of bacterial contamination (current vaccines do not contain preservatives) [2], [18], which may be perceived as waste. Moreover, vial sharing amongst patients may lead to practical difficulties in health provider budgeting. All WHO-recommended PEP regimens for WHO pre-qualified vaccines are safe, immunogenic, and efficacious. Thus policy should aim to prevent failures in PEP delivery by preventing vaccine shortages, reducing costs for victims and healthcare providers and promoting patient compliance to ensure PEP efficacy. The variety of WHO-recommended PEP regimens allows flexibility, but can lead to confusion regarding which regimen best meets the needs for a specific setting. Additionally, new regimens are continually being developed that require evaluation prior to implementation. Here, we provide a framework for comparing cost-effectiveness of different PEP regimens (including existing approved regimens and new candidates subject to approval) from the perspective of both healthcare providers and bite victims under a range of scenarios (from low to high throughput clinics) and under realistic constraints (poor patient compliance and some vaccine wastage). We discuss the implications for PEP affordability, availability and accessibility and offer recommendations for policy formulation and vaccine research.
We developed a simulation framework for evaluating vaccine use under different PEP regimens. We compared different vaccination regimens (detailed in Table 1) that are currently approved by WHO and the Advisory Committee on Immunization Practices (ACIP), together with two additional candidate regimens (the 4-site and 1-week ID PEP regimens), that have recently undergone clinical trials and could potentially be used in future conditional upon review and approval by WHO. The algorithm for our simulations is shown in Figure 1 with an example scenario.
We used cost data associated with rabies reported from previous studies (Table 2). These include direct (medical) costs corresponding to rabies vaccines and their administration and indirect (non-medical) costs including transport to and from clinics and loss of income while receiving PEP. We assume that the time taken to vaccinate a patient is equivalent for both ID and IM administration and we did not include costs of RIG because most bite victims in Africa and Asia do not receive RIG [3], [4], [5].
We explored vaccine use according to different model inputs that are detailed in Table 3 and defined as follows:
We analyzed outcomes in terms of savings in vaccine use, human rabies cases averted and incremental cost-effectiveness ratios. We evaluated incremental cost-effectiveness ratios (ICER) in terms of dollars per rabies death averted: costsPEP/(EffectivenessPEP-EffectivenessNo PEP), where subscripts refer to whether or not PEP vaccination was administered. We calculated cost-effectiveness from the perspective of the health provider and included only direct medical costs. We also modify this calculation of cost-effectiveness under poor compliance and according to hypothesized protective efficacy of incomplete vaccination (Table 3).
We compiled data on clinic throughputs in different settings and calculated the annual costs of PEP vaccination using different regimens for these settings. In practice, many clinics operate on a cost-recovery basis and charge for PEP. However, the number of new and returning animal-bite patients expected at a clinic on a daily basis cannot be precisely predicted, making it difficult to determine appropriate charges for ID administration. We compared four pricing strategies: 1) charging patients per injection according to the amount of vaccine used; 2) charging patients per injection at rates that are marginally higher than the price of the amount of vaccine used (illustrated with patients paying per injection at a rate that is 25% or 30% of vial costs); 3) charging patients a set price on their first visit (illustrated with a fee equivalent to 1.5 vials), but providing all subsequent doses without payment; 4) charging patients the price of one vial for each of their first and second hospital visits, but providing vaccine for free on subsequent visits. For all the strategies we assume that patients pay the costs of materials for vaccination and a consultation fee, which is equivalent to the price of overhead for a clinic visit in addition to the vaccination costs described above. We explored implications for cost recuperation (based on costs of PEP delivery shown in Table 2) by estimating annual savings of the different pricing strategies dependent upon throughput and the regimen in use. We present the net gains under these pricing strategies for the updated TRC regimen (the only currently WHO approved ID regimen) for clinics for which throughput data was compiled.
We compare the costs of PEP for bite-victims, depending upon the pricing strategies described above and including the provision of PEP free-of-charge and under different assumptions about indirect costs (based on the range of indirect costs in Table 2). Specifically we assume that bite victims travel further to reach a clinic in rural rather than urban settings and incur correspondingly higher costs. We also calculate the likely risks for patients of poor PEP compliance according to assumptions about vaccine efficacy (Table 2) and explore how costs may affect compliance with PEP regimens and implications for the risk of developing rabies.
All analyses were performed using the statistical programming language R. Scripts implementing our simulations are available upon request.
For IM vaccination, both the reduced 4-dose Essen and the Zagreb regimens are more economical than the 5-dose Essen because they use only 4 vaccine vials in comparison to 5 vials for the 5-dose schedule, i.e. 80% of the total volume of vaccine (Table 1). All ID regimens use less vaccine than IM regimens and are cost less per rabies death averted (Figure 2). Clinic throughput generally increases the cost-effectiveness of ID vaccination, with high throughput clinics most cost-effective and low throughput clinics least cost-effective. The only exception to this is the situation where some wastage is assumed and 0.5 mL vials are used, in which case the 1-week ID regimen does not increase in cost-effectiveness with throughput (Figure S1), but is still considerably more cost-effective than IM regimens.
The updated TRC and the 4-site ID regimens were the most cost-effective in high throughput settings (between $100 and $55 per life saved depending on throughput and vial size). While in very low throughput clinics, the 1-week ID regimen was the most cost-effective ($160–150 per life saved, Figure 1). In high throughput clinics the updated TRC and 4-site ID regimens use just 40% of the volume of vaccine in comparison to preferred IM regimens (Essen 4-dose and Zagreb) when 0.5 mL vials are used and 20% of the volume when 1 mL vials are used (Table 1, Figure 1). The cost-effectiveness of all ID regimens increases considerably when 1 mL vials are used instead of 0.5 mL vials (Figure 1). The estimated costs of PEP vaccination to health providers are shown for different regimens in Table 4 for a variety of throughput settings and illustrate how provision of PEP using ID regimens is considerably more economic than provision of PEP using IM regimens.
When health providers charge for PEP vaccination according to the strategies described, substantial costs are recovered when using ID regimens and in most cases savings are made. The extent of savings and how these vary with clinic throughput for different ID regimens are shown in Figure 3. Using 1 mL vials rather than 0.5 mL vials increases savings for all pricing strategies. Charging patients for exactly the amount of vaccine administered using the ID route results in a net loss for healthcare providers except in high throughput clinics. Charging patients per ID injection at rates slightly greater than the price of the vaccine used results in net savings in most locations (Figure 3 shows savings for charging $2.5 or $3/injection assuming vials cost $10). Other strategies such as charging higher rates but for the primary presentation only, or for primary and secondary presentations only, also result in significant savings for high throughput clinics and losses occur only in very low throughput clinics when 1 mL vials are used. For example, when using 1 mL vials, charging the price of a vial for each of the first 2 presentations, results in savings in clinics that receive over 15 new patients per month with all the ID regimens, whereas losses are always incurred when using 0.5 mL vials in lower throughput clinics (Figure 3). Similar savings are made in high throughput clinics charging a fixed price for a full PEP course when 1 mL vials are used (illustrated by a $15 set rate, assuming a vial costs $10, in Figure 3), however, charging $15 for a full ID course (all regimens) is not sufficient to recuperate costs when using 0.5 mL vials. Extrapolations assuming use of the $15 full course of the updated TRC regimen with 1 mL vials (the only ID regimen currently recommended by WHO) suggest that even in countries with mainly low throughput clinics (e.g. Tanzania), savings recuperated from urban centres would ensure sustainability, and in the highest throughput settings annual savings could exceed $100,000 in a single clinic (Table 4).
Where PEP vaccination is provided free-of-charge, the Zagreb IM and the recently proposed 1-week ID regimens are most preferable for patients, who incur only indirect costs (Table 5). This is because only 3 hospital visits are required as compared to the Essen IM and the updated TRC and 4-site ID regimens, which all require 4 visits (Table 1). When patients are required to pay for PEP vaccination, the most preferable regimen for bite victims varies depending on pricing strategies and relative travel costs (Table 5). However, in terms of price, ID regimens are always preferable over IM regimens. When travel costs are low and PEP is charged per injection, the updated TRC and the 4-site ID regimens are preferable. The 1-week ID regimen is preferable when travel costs are high, and particularly when flat rates are charged for the full PEP course, rather than per injection (Table 5).
We assume that high costs reduce patient compliance, which in turn reduces the effectiveness of PEP in preventing rabies and thus the cost-effectiveness of PEP. Specifically, we assume 100% compliance when patients pay $10 or less for PEP, and that for every dollar increase there is a 0.05% reduction in compliance (Table 3), thus the most expensive regimen ($98.4 per course for the Essen 4-dose IM for patients with high travel costs, Table 5), has only 42.8% compliance. The cost per rabies death averted decreases as the efficacy of the regimen increases, and therefore cost-effectiveness increases and a greater proportion of preventable deaths are averted (Figure 4A & B). Cost-effectiveness is lowest at low levels of compliance. The proportion of deaths prevented also increases with vaccine efficacy (Figure 4B). At low levels of vaccine efficacy, regimens that require 3 clinic visits (Zagreb, 4-site ID, 1 week ID) prevent a greater percentage of deaths than regimens that require 4 clinic visits (Essen 4-dose, updated TRC). Overall, the risk of death increases with the costs of PEP as patients become less likely to comply with regimens (Figure 4C). For all pricing strategies that we present, patients who live further from clinics have reduced compliance and heightened risks. When PEP vaccination is free of charge, risks are minimized, and risks are maximized when charging for IM regimens (Figure 4C).
Rabies post-exposure vaccination is essential for preventing this fatal disease but can be out of the financial reach of many bite victims. Vaccine shortages are common in developing countries and due to limited availability bite victims often need to travel long distances to obtain vaccine. Thus, patients often incur substantial costs and face dangerous delays in securing PEP and avoidable human rabies deaths occur as a direct result of poor access to affordable PEP [3], [4], [5]. We examined the costs of IM versus ID administration of PEP vaccine in different settings and under realistic constraints such as poor compliance. We demonstrate that ID delivery of PEP is considerably more cost-effective than IM delivery in terms of averting rabies cases and saving lives. Clinic throughput affects the capacity for vial sharing, and therefore the cost-effectiveness of ID administration relative to IM. As throughput increases, ID regimens become increasingly cost-effective, using up to 80% less vaccine (Figure 1). Yet, even clinics with relatively low throughput (∼10 new patients/month) would reduce vial use by 25% by switching from IM to ID administration of PEP. Increased use of ID regimens could therefore prevent vaccine shortages and enable wider vaccine distribution, both increasing the number of patients that can be treated and the overall accessibility of PEP. Concurrent changes in PEP costs to patients could also improve affordability, while providing incentives for compliance without compromising existing health budgets. These issues should be considered in the design of PEP policy because they could reduce the burden of rabies by increasing the availability of vaccines for the rural poor who bear the brunt of rabies in most developing countries.
Our principal finding that ID administration of PEP is more cost-effective than IM administration and reduces the amount of vaccine used is important given the frequency with which PEP vaccine shortages occur at clinics in many developing countries. Savings in vaccine use are substantially larger when using equivalently priced 1 mL rather than 0.5 mL vials, especially in high throughput clinics because of greater vial sharing. In this situation there is no advantage to stocking a mixture of vials (100% of 1 mL vials is always most cost-effective), but should pricing change (so that 1 mL and 0.5 mL vials differ in price), optimal stocking strategies should be evaluated as a priority. For safety reasons (potential for contamination) vial sharing is only possible on the day of vaccine reconstitution, even though potency remains high when properly stored [23]. Research into methods of preserving rabies vaccines and preventing contamination could therefore enable more economical use of vaccines, including production in larger volume vials.
Despite policies to provide PEP free-of-charge, many bite victims need to pay to promptly obtain PEP. In the light of this, a switch to ID administration could reduce costs to bite victims. But, there are many ways to charge for PEP. Only in high throughput locations, where vials can be shared completely, could patients be charged exactly for vaccine used without clinics operating at a loss. Rates could be set proportional to clinic throughput to prevent losses and ensure cost-recovery, but this would result in inequities (with higher throughput clinics providing cheaper PEP) that would disadvantage patients attending lower throughput clinics, i.e. the rural poor. More equitably, patients could be charged set rates that are much lower than for IM PEP (Table 5), whilst ensuring cost recovery (Figure 3). Savings (see Table 4) from higher throughput clinics could subsidize either lower throughput clinics that might operate at a loss, or the poorest patients who are unable to afford PEP (e.g. as part of a rolling fund or an insurance system) or even other rabies control and prevention activities. Innovative financing mechanisms could provide more affordable PEP and generate potentially high returns from high throughput clinics, but effective monitoring would be critical.
Health policy aims to reduce the burden of disease, but conflicts inevitably arise between the individual interests of patients and the population-level interests of healthcare providers. Choices about which regimens are preferable depend upon whether indirect or direct costs are a greater obstacle to bite victims. When PEP is provided free-of-charge, the recently developed 1-week ID regimen and the Zagreb IM are most advantageous for patients, because they entail fewer clinic visits. But both have drawbacks as the 1-week ID regimen is not yet approved by WHO and the Zagreb regimen, which is approved by WHO, uses more vaccine. If budgets and therefore vaccine supply are limited, the 4-site, if eventually approved by WHO, and updated TRC ID regimens are most preferable. Policies need to balance these issues to reduce costs for bite victims and prevent shortages.
A further consideration in PEP delivery is how to promote compliance and therefore improve the effectiveness of PEP. We assume that affordability of PEP will improve compliance and provision of PEP free-of-charge is therefore the ideal solution (Figure 4). However, when charging for PEP, flat rates that are more affordable than IM regimes (e.g. $15 for a full course or $20 for the first two visits, see Table 2) might incentivise compliance, but would recuperate costs only when 1 mL vials are used. Alternatively wider distribution of vaccine (even when charging for PEP) could reduce indirect costs for bite-victims and improve compliance. Staying in the vicinity of a clinic rather than travelling back and forth for each scheduled vaccination might also be cheaper for bite-victims, which would apply particularly for the 1-week ID regimen (we do not currently explore such complexities but data could inform model inputs for future analyses). Although the 1-week ID uses more vaccine than other ID regimens, reduced indirect costs could make it more affordable for bite-victims. This may facilitate compliance and has the added benefit of earlier complete protection reducing anxiety for bite-victims. In contrast, for the 4-dose ID regimen, the last dose of vaccine is not administered until day 90, which could reduce compliance in comparison to other ID regimens. Further study is therefore warranted to better quantify indirect costs of obtaining PEP and to understand the major constraints to PEP access and compliance for those most in need. Incomplete and late PEP is less effective in preventing the onset of disease, but there are no data available to quantitatively compare risks to full compliance (we were only able to explore hypothetical changes in PEP effectiveness with poor compliance). Contact tracing could potentially reveal more about these issues, and longitudinal serology studies could provide a useful proxy measure for immunogenicity that could be used to inform PEP policy.
Despite being more economical, misperceptions about ID, the lack of strong recommendations and a profusion of complex schedules have deterred their widespread adoption. Yet our analyses show that switching from IM to ID administered PEP has benefits to patients and healthcare providers. The updated TRC is the only currently WHO approved ID regimen, but the 4-site ID regimen is also highly cost-effective and the 1-week ID has other benefits for bite-victims. Their further evaluation by WHO is clearly warranted. More generally, since ID procedures involve delivery of only small amounts of vaccine, in order to apply our findings to settings where non pre-qualified vaccines are used, rigorous evaluation of the product including manufacturing standards, safety, immunogenicity and efficacy must be prioritized.
The absolute cost-effectiveness of PEP depends upon the regimen, clinic throughput, clinic overhead and costs of materials for vaccine delivery and vaccine vials. Nonetheless, our estimates suggest that PEP is more cost-effective in averting deaths than childhood immunization through the Expanded Program on Immunization (USD$205/death averted in sub-Saharan Africa and South Asia [24]), which is considered one of the most cost-effective health interventions available [25]. Even considering vaccine waste, the worst-case scenario for PEP cost-effectiveness is around $200/death averted (for IM regimens) and in high throughput clinics use of ID regimens can reduce costs to just $60/death averted. Cost-effectiveness will decline if PEP is administered to patients who are bitten by non-rabid animals. We do not currently factor this into our calculations, but positive predictive values obtained from field data in Cambodia and in Tanzania [4], [26] suggest that PEP is largely administered to genuine rabid bite victims and that cost-effectiveness will remain high even with liberal provision of PEP [19]. Thus effective PEP delivery should be considered an extremely cost-effective investment for public health, given the current poor availability of this life saving intervention. However, rabies can only be eliminated through intervention in the animal reservoir [27], and this is likely to be the most cost-effective way of averting human rabies deaths in the long-term [28].
Our model has several simplifications, which could be elaborated on in future. We assume that the day of the week does not affect the likelihood of presenting for PEP vaccination. But patients may be less likely to present on Sundays (in many countries clinics providing PEP are not open on Sundays) and/or more likely to present on Mondays or other days of the week (e.g. after pay day), which may affect vial sharing. We also do not include pre-exposure vaccination. Livestock officers and extension workers involved in animal vaccinations and more at risk of animal bites should be pre-vaccinated. Pre-exposure vaccinations would likely make PEP more cost-effective (as less vaccine is required in the event of an exposure) and preliminary vaccinations could be coordinated to ensure effective vial sharing (e.g. prior to dog vaccination campaigns). But, in general in low-income countries, such pre-vaccinated persons are rare relative to non-vaccinated bite victims. In some high-risk settings pre-vaccination of children is under consideration [29], and our simulation framework could be useful for their further evaluation.
The availability and affordability of PEP is critical in determining the burden of rabies. Incidence in resource poor countries is directly affected by the inability of bite victims to obtain PEP and obtain it promptly. Reducing the cost of PEP and preventing administration delays is therefore particularly important in resource-limited settings. The variety of PEP regimens, vial sizes, and routes of administration has also made the delivery of these life-saving vaccines unnecessarily complicated. Our results provide evidence to show that a simplification to universal ID delivery of PEP could have massive advantages in low-income countries: streamlining guidelines, reducing the volume of vaccine use, mitigating vaccine shortages and making PEP more affordable to the most vulnerable. Health workers routinely deliver childhood immunizations intradermally, so there should be no technical difficulty in switching to ID administration. ID vaccination is as safe and efficacious as IM vaccination and is well-tolerated [30]. The immense advantages of ID PEP delivery should be specifically highlighted in outbreak situations, such as those recently reported from Bali [31] and in areas where vaccine supply is limited, as considerably more bite victims can be protected using the same volume of vaccine.
|
10.1371/journal.pgen.1004484 | Common Transcriptional Mechanisms for Visual Photoreceptor Cell Differentiation among Pancrustaceans | A hallmark of visual rhabdomeric photoreceptors is the expression of a rhabdomeric opsin and uniquely associated phototransduction molecules, which are incorporated into a specialized expanded apical membrane, the rhabdomere. Given the extensive utilization of rhabdomeric photoreceptors in the eyes of protostomes, here we address whether a common transcriptional mechanism exists for the differentiation of rhabdomeric photoreceptors. In Drosophila, the transcription factors Pph13 and Orthodenticle (Otd) direct both aspects of differentiation: rhabdomeric opsin transcription and rhabdomere morphogenesis. We demonstrate that the orthologs of both proteins are expressed in the visual systems of the distantly related arthropod species Tribolium castaneum and Daphnia magna and that their functional roles are similar in these species. In particular, we establish that the Pph13 homologs have the ability to bind a subset of Rhodopsin core sequence I sites and that these sites are present in key phototransduction genes of both Tribolium and Daphnia. Furthermore, Pph13 and Otd orthologs are capable of executing deeply conserved functions of photoreceptor differentiation as evidenced by the ability to rescue their respective Drosophila mutant phenotypes. Pph13 homologs are equivalent in their ability to direct both rhabdomere morphogenesis and opsin expression within Drosophila, whereas Otd paralogs demonstrate differential abilities to regulate photoreceptor differentiation. Finally, loss-of-function analyses in Tribolium confirm the conserved requirement of Pph13 and Otd in regulating both rhabdomeric opsin transcription and rhabdomere morphogenesis. Taken together, our data identify components of a regulatory framework for rhabdomeric photoreceptor differentiation in Pancrustaceans, providing a foundation for defining ancestral regulatory modules of rhabdomeric photoreceptor differentiation.
| Visual systems are populated by one of two fundamental types of photoreceptors, ciliary and rhabdomeric. Each photoreceptor type is defined by the opsin molecule expressed and the final morphological form adapted to house the phototransduction machinery. Here we address whether a common transcriptional mechanisms exists for the differentiation of rhabdomeric photoreceptors. We demonstrate that orthologs of two Drosophila (fruit fly) transcription factors, Pph13 and Orthodenticle, are expressed in photoreceptors of Pancrustaceans, Tribolium (red flour beetle) and Daphnia (water flea), and are capable of executing conserved functions of rhabdomeric photoreceptor differentiation. In particular, Tribolium and Daphnia orthologs are capable of substituting and rescuing the photoreceptor differentiation defects observed in their corresponding Drosophila mutants. Furthermore, loss of function analysis in Tribolium of both Pph13 and orthodenticle genes demonstrate they regulate opsin transcription and morphogenesis of the photoreceptor apical membrane. Our data illuminate a framework for rhabdomeric photoreceptor differentiation and provide the foundation for defining the ancestral regulatory modules for rhabdomeric differentiation and potential modifications that underlie the functional diversity observed in rhabdomeric photoreceptors.
| Rhabdomeric (r) photoreceptors are one of two fundamental types of photoreceptors that have been described [1]. Typically, r- photoreceptors populate the visual systems of protostomes including insects, crustaceans, and annelids (reviewed in [2]). This wide phylogenetic distribution and the presence of both types of photoreceptors in many species imply that r- photoreceptors, like their deuterostome counterparts, ciliary photoreceptors, were present before the split of bilaterian animals [3]–[6]. Despite this wide utilization of r- photoreceptors, knowledge about r- photoreceptor differentiation has been virtually exclusively defined from studies in the Drosophila (fruit fly) visual system (reviewed in [7], [8]). Generally, two features characterize r- photoreceptor differentiation. The first is the expression of an r- opsin for light detection, which upon the absorption of a photon leads to the activation of a Phospholipase C cascade and the depolarization of the photoreceptor. This phototransduction cascade permits the amplification of responses to single photons of light [9]. The second program concerns the generation of the rhabdomere, an expansion of the photoreceptor apical membrane to house the phototransduction machinery. This adaptation is necessary for increasing the accuracy of measuring light intensity required for vision [10]. Therefore, understanding the development and evolution of rhabdomeric photoreceptor differentiation requires clarification of how these processes are transcriptionally regulated and whether this regulation is conserved within all rhabdomeric photoreceptor types.
In Drosophila, two homeodomain proteins have been identified that are critical for regulating r- photoreceptor differentiation. The first, Orthodenticle (Otd), is the Drosophila ortholog of a conserved family of Otd/Otx homeodomain transcription factors, which are essential for head and brain development across species [11], [12]. In the r- photoreceptors of Drosophila eyes, otd is required for both aspects of differentiation [13]. Otd promotes the proper morphogenesis of rhabdomeres and directs multiple aspects of the differential expression of r- opsin paralogs, which characterizes the complex visual organization of the Drosophila retinas (for a review see [14]). In particular, Otd is required for the expression of rh3, an ultra-violet (UV) sensitive r- opsin, and rh5, a blue (B) sensitive r- opsin in the two inner photoreceptors of Drosophila ommatidia [15], [16]. In addition, Otd is critical for repressing rh6, the Drosophila ancestral long-wave (LW) opsin [17] in the six outer photoreceptors of the Drosophila ommatidium [16].
The second critical transcription factor is PvuII-PstI homology 13 (Pph13), a paired-class homeobox protein that is similar to the vertebrate Aristaless-related homeodomain (Arx) proteins [18], [19]. Like Otd, the loss of Pph13 results in defects in rhabdomere morphogenesis in the Drosophila eye [19]. Interestingly, the concurrent removal of Pph13 and Otd results in complete elimination of the rhabdomeres in Drosophila, suggesting that the two proteins cooperate and have overlapping functions with respect to photoreceptor morphology [20]. Pph13 is also essential for the expression of r- opsins rh6 and rh2 [20]. In contrast to otd mutants, phototransduction is abolished in Pph13 mutants due to the loss and reduced transcription of several key components of the phototransduction machinery [19], [20]. Lastly, Pph13 regulates photoreceptor differentiation by binding to a subset of Rhodopsin core sequence I (RCSI) elements [20], [21], which are conserved elements present in Drosophila Rhodopsin promoters [22]–[24]. Consistent with this, Drosophila Pph13 is necessary and sufficient for driving photoreceptor specific reporter gene expression from the artificial 3XP3 promoter [20], which has been assembled from a Pax6 homeodomain binding site [25]. Given the central role of both Pph13 and Otd in r- photoreceptor differentiation in Drosophila, the question we address here is whether Pph13 and Otd functions represent a common regulatory pathway of arthropod r-visual photoreceptor differentiation.
To examine whether Pph13 and Otd could represent a common set of transcription factors required for r- visual photoreceptor differentiation, we chose to investigate their orthologs from two key nodal species, Tribolium castaneum (red flour beetle), a second insect, and Daphnia (water flea), a crustacean. Together, insects and crustaceans define the superclade Pancrustacea within the Arthropoda [26], [27] and any similarities between Daphnia, Tribolium, and Drosophila r- photoreceptor differentiation would indicate a pathway common to the ancestor that generated both lineages, at least 500 million years ago [28], [29]. First, we demonstrate that Otd and Pph13 orthologs are present and expressed in the visual systems of both species. Consistent with conservation of Pph13 mediated r- opsin regulation, the Pph13 RCSI binding site is conserved in the promoters of the r-opsin genes of both Tribolium and Daphnia and found only within LW r- opsins. Further, the Tribolium and Daphnia Pph13 homologs have retained similar DNA binding capabilities to their respective endogenous RCSI sites and we confirmed their functional equivalency to direct photoreceptor differentiation in Drosophila photoreceptors by transgenic rescue. The Otd paralogs of Tribolium and Daphnia are comparable in their ability to direct rhabdomere morphogenesis but exhibit differential abilities with respect to r- opsin regulation in Drosophila. Lastly, functional analyses in Tribolium reveal that both Pph13 and Otd homologs are essential for both aspects of photoreceptor differentiation, rhabdomere creation and r-opsin expression. In particular, Pph13 is a critical factor for LW r-opsin expression and Otd2 is necessary for the transcription of UV sensitive r-opsin. In summary, our data identify common components for rhabdomeric photoreceptor differentiation among Pancrustaceans, providing a foundation for defining the ancestral transcriptional mechanisms for rhabdomeric photoreceptor differentiation throughout Bilateria.
As a first step towards examining whether the role of Pph13 and Otd in Drosophila r- visual photoreceptor differentiation was conserved, we investigated the conservation of orthologs in the genome sequences of the distantly related arthropod species, Tribolium castaneum and Daphnia pulex [30], [31]. Tribolium had been previously shown to possess two paralogs of Otd: Otd1 and Otd2 [32]. The same state was described in the Crustacean Parhyale hawaiensis [33]. However, the relationships of the crustacean and coleopteran Otd homologs to the singleton homolog of Drosophila were previously considered unresolved due to the low level of sequence conservation outside the homeodomain; within Diptera there has been a reduction to only one otd paralog (Figure 1A, S1 and [33]). As in Parhyale, our search in Daphnia pulex as well as Daphnia magna identified two Otd homologs. Protein sequence alignment of an expanded set of Otd homologs (Figure 1A) revealed a highly conserved leucine (L) residue at the C-terminal end of the Otd1 homeodomain, which was unique for Paired-class homeodomain proteins in general [34], distinguished the insect representatives of the Otd1 subfamily, including all dipteran homologs. This finding established Drosophila Otd as a member of the insect Otd1 subfamily, implying the loss of insect Otd2 in the evolutionary lineage to Diptera. This conclusion was tentatively supported in a molecular phylogenetic analysis of the relationships between Otd homologs (Figure S1). The latter approach and amino acid residues in the homeodomain that were unique to each of the Parhyale and Daphnia Otd sequences further suggested that the latter duplicates represented the results of independent gene duplications in crustacean and insect lineages. Thus, the use of the previously introduced acronyms Otd1 and Otd2 for Parhyale and Daphnia paralogs do not imply specific orthology to insect Otd1 and Otd2.
In contrast to the well-characterized deep conservation of the Otd/Otx gene family, homologs of Pph13 have thus far been identified only in a limited number of insects. Previous efforts identified Pph13 homologs in the Tribolium [17] and honey bee genomes [35] but not in the Daphnia pulex genome [36]. Closer inspection of candidate Daphnia pulex orthologs revealed one annotated locus encoding a 5′ truncated Pph13-related homeodomain, suggesting that the annotation for this locus (JGI_V11_8835 –wfleabase.org) was incorrect. Further examination of upstream genomic regions revealed that the DNA encoding the 5′ portion of the homeodomain was present. This conclusion was confirmed by RT-PCR (data not shown) and in the genome draft of a second related species: Daphnia magna. The complete Daphnia pulex Pph13 cDNA sequence thus included sequence from previously annotated loci JGI_V11_8835 and JGI_V11_313449. Sequence conservation between Daphnia, Tribolium, and Drosophila homologs was confined to the homeodomain (Figure 1B). However, examination of a larger sample of Pph13 homeodomain sequences in combination with that of members of the Aristaless (Al/Arx) gene family revealed amino acid residues that defined each subfamily and added further support of their hypothesized common descent (Figure 1B and Figure S2) [35]. A tyrosine (Y) residue at the fourth homeodomain position, otherwise not observed in the Drosophila Paired-class homeodomain proteins [34], characterized all members of the Arx gene family (Figure 1B). Furthermore, a phenylalanine (F) residue at homeodomain position 30 specifically marked the Pph13 orthologs, in contrast to the threonine (T) residue in the large majority of Al and Arx orthologs. While this variation was in line with the overall variability at homeodomain position 30, we noted the singularity of the Pph13 characteristic phenylalanine (F) among all Drosophila homeodomain proteins [34]. Based on these clues, we also identified putative orthologs of Pph13 and Al in the mollusk Aplysia californica. Taken together, these alignment data indicated that a gene duplication at least predating the origin of the Pancrustacea gave rise to Pph13. Finally, we noted the presence of a third Al-related Arx gene family member in the Tribolium genome (Tcas A12) that may be of similar ancient origin based on conservation in non-arthropod invertebrates (Figure 1B).
The presumed functional conservation of the Pph13 and Otd homologs predicted their expression in photoreceptors of Tribolium and Daphnia. We therefore assayed the spatial expression patterns of Pph13, otd1 and otd2 during adult eye development in both Tribolium and Daphnia. Like Drosophila, the Tribolium adult compound eye consists of individual ommatidia each of which contains six outer and two inner photoreceptors [37]. Approximately 30–40 hours after pupation we detected a signal from antisense probes to each transcript in all eight photoreceptors of a single ommatidium (Figure 2), whereas the corresponding sense probes did not generate any specific pattern (Figure S3). Pph13 and otd1 appeared to have similar expression patterns, with equal levels of transcript in each photoreceptor (Figure 2A and B). An antibody raised against Tribolium Pph13 confirmed its expression in the nucleus of each photoreceptor (Figure 2D–F). Together the Pph13 RNA in situ pattern and immunofluorescence staining suggested Pph13 is expressed in every photoreceptor of the eye as observed in Drosophila. We also detected otd2 in all eight photoreceptors but its expression in the two central photoreceptors, R7 and R8, was considerably stronger than in the outer photoreceptors (Figure 2C).
In Daphnia, the adult eye consists of a single bilaterally symmetrical compound eye, containing a total of twenty-two ommatidia with eight photoreceptors in each ommatidium [38]. Each Daphnia eye is generated by the fusion of two lateral groups of ommatidia along the midline late in embryogenesis. Due to the lack of molecular markers, the exact biogenesis of the photoreceptors has not been described. However, previous transmission electron microscopy (TEM) studies of the development of the axonal photoreceptor connections with lamina neurons predict a model in which the photoreceptors begin to differentiate at the midline and move laterally as they mature [39]–[41]. To confirm and differentiate the developing photoreceptors we first examined the expression of a limited set of r-opsins in Daphnia magna. In Daphnia pulex, there are 27 annotated r-opsin paralogs [30]. Like Drosophila, Daphnia pulex contain representatives of UV, LW and B- light sensitive r-opsins. This diversity includes 23 LW opsins that split between the LOPA and LOPB clades [30] (Figure S4). The r-opsin family has not been defined for Daphnia magna but for our examination we assayed the expression of a pool of three putative representatives from LOPA (Figure 3A) and LOPB (Figure 3B) and the putative UV opsin (Figure 3C). Besides observing a differential display of expression between all three groups, the RNA in situ hybridization patterns confirmed and delineated the embryonic tissue that gives rise to the photoreceptors of the eye and ocellus (Figure 3 and Figure S5A–F). Our expression analysis of Otd1, Otd2, and Pph13 in Daphnia magna also yielded results that were consistent with this model of eye formation (Figure 4 and Figure S5G–J). The expression of Otd1 was limited to the midline region of the embryo and not necessarily associated with visual photoreceptors (Figure 4A,C). In contrast to Otd1, Otd2 was expressed in an increasing number of cells in two lateral symmetrical regions of the head during embryogenesis (Figure 4B–F). By 48 hours after egg deposition (AED), each lateral cluster contained approximately 75–78 Otd2 positive cells (Movie S1). Considering that the adult eye consists of two lateral clusters of eleven ommatidia, this suggested that there were seven Otd2 positive photoreceptors per ommatidium. Furthermore, the additional Otd2 staining present in the central portion of the embryo corresponded to the region where the ocellus develops (Figure 4A–F and Figure 3B). Similar and consistent expression patterns were detected for Daphnia magna homolog of Pph13, which was expressed in two symmetrical lateral clusters of cells, in the cells of the presumptive ocellus (Figure 4G–I) potentially colocalizing with Otd2 protein expression (Figure 4H,I).
To further explore the possibility of conserved regulatory roles of Pph13 and Otd in visual photoreceptor differentiation, we probed for the conservation of the RCSI site in candidate target genes of Pph13 in Tribolium and Daphnia. Previous work has shown that Pph13 binds a subset of RCSI sites and that this binding site is essential for transcriptional activation [19], [20]. The same studies defined a Pph13 RCSI site as a palindromic sequence of TAAT spaced by three nucleotides with one half site matching the consensus sequence of 5′-CTAATTG-3′ [20]. In Drosophila, these Pph13-specific RCSI sites are present in the 5′ cis-regulatory DNA of r- opsin genes and in several other key phototransduction proteins, including the heterotrimeric G-protein β subunit (Gβ76C) [19], [20]. Tribolium contains two r- opsins, one of which belongs to the LW opsin subfamily and one of which belongs into the UV sensitive subfamily [42]. Scanning their upstream regions for the RCSI motif, we found a potential RCSI site in both of them. Furthermore, in the upstream region of the Tribolium homolog of Drosophila visual Gβ (Gβ76C), LOC662674 (beetlebase.org), we also detected a RCSI site (Table S1 and Figure S6). Examining the immediate upstream regions of Daphnia LW, UV, and B opsins revealed putative RCSI motifs only in the LOPB clade of LW opsins (Figure S4 and Table S1). In addition, the closest homolog to both Drosophila and Tribolium visual Gβ subunit in Daphnia (JGI_V11_210534 -wfleabase.org) contained a potential RCSI site (Figure S6 and Table S1).
The presence of potential RCSI sites in photoreceptor-expressed genes of all three species suggested that these sites could serve as Pph13 binding sites. To test this possibility, we investigated whether the Pph13 homologs could bind the putative endogenous RCSI sequences with electrophoretic mobility shift assays (EMSAs). These experiments revealed that Tribolium and Daphnia Pph13 have similar binding abilities to a consensus RCSI site (P3) [24], [43] as well as specific Drosophila RCSI sites (Figure 5 and data not shown). Furthermore, each has the capability to bind to their endogenous RCSI sites (Figure 5 and Figure S7). Interestingly, in Tribolium like Drosophila [20], we observed a differential affinity of Pph13 to the identified RCSI sites of UV and LW r-opsins. Tribolium Pph13 bound efficiently to the LW opsin RCSI site but binding was barely detectable on the UV opsin RCSI element, suggesting that the simple presence of a correctly spaced palindromic sequence of TAAT was not sufficient to bind Pph13.
The in vivo expression patterns and in vitro binding assays provided strong evidence that Pph13 and Otd regulate r- visual photoreceptor cell differentiation in Drosophila, Tribolium and Daphnia. To test for functional equivalency among the orthologs and, more importantly, the ability to direct photoreceptor differentiation, we examined whether Daphnia and Tribolium orthologs were capable of rescuing the photoreceptor defects observed in Drosophila Pph13 and otd mutants.
Drosophila Pph13 mutants have two distinct characteristics. First, Pph13 is necessary for expression of opsin rh6 (Figure 6A,B) in 70% of the R8 photoreceptor cells [44]. Second, Pph13 mutants have severe defects in rhabdomere morphology (Figure 7A,B). The morphological defects are acute enough to hamper the detection and accumulation of other r-opsins [20], [21] (Figure 6B). For rescue experiments, each homolog was placed under the control of GAL4 transcription [45] and inserted into the identical locus in the Drosophila genome. To drive expression, we generated a GAL4 driver under the control of the endogenous Drosophila Pph13 cis-regulatory region – Pph13-Gal4. In testing the Pph13 homologs of Daphnia and Tribolium, we evaluated both the restoration of Rh6 opsin expression and rhabdomere morphogenesis as compared to rescue with Drosophila Pph13. We found that all Pph13 homologs were capable of restoring Rh6 expression. In addition, we also detected a mosaic expression pattern of Rh6 in the R8 photoreceptors (Figure 6C–E). Of note, this result also demonstrated the specific rescue of rhabdomere morphology in the R8 photoreceptors that express Rh5 opsin in a Pph13 independent manner. To assay rhabdomere morphology directly, TEM analysis of each rescue condition was performed. We observed wild-type rescue of rhabdomere morphology with all three Pph13 homologs (Figure 7). However, the rescue was not fully penetrant with Tribolium and Daphnia Pph13. In particular, we observed photoreceptors missing rhabdomeres with Tribolium Pph13 (Figure 7D′) and with Daphnia Pph13, the rhabdomeres did not maintain their position and morphology along the proximodistal axis of the photoreceptor (compare Figure 7E and E′).
For examination of functional equivalency among Otd orthologs, we used a previously established rescue paradigm [46], [47]. Otd is required for rh3 opsin expression in the distally located inner photoreceptor, the activation of rh5 in the proximal inner photoreceptor, repression of rh6 opsin in the outer photoreceptors, and correct rhabdomere morphology in every photoreceptor [13], [15], [16]. In our rescue experiments, we assayed for all these functions.
With respect to rhabdomere morphology (Figure 8), TEM analysis demonstrated that both Tribolium Otd paralogs (Figure 8D,E) and Daphnia Otd1 (Figure 8F) could direct rhabdomere morphogenesis in Drosophila. In these three rescue experiments, we observed the return of symmetrical wild-type pattern of rhabdomere shape and size. However, the rescue was not fully penetrant with Tribolium Otd2 and Daphnia Otd1. For example, some ommatidia have photoreceptors that lack a detectable rhabdomere. The expression of Daphnia Otd2 resulted in an adult eye that contained a mosaic of intact and dead tissue, regardless of the presence of endogenous Otd (Figure S8); this phenotype was not observed with any of the other Otd orthologs. TEM analysis confirmed the lack of photoreceptors in the discolored regions (data not shown). In the normal pigmented regions we observed the presence of rhabdomeres that were not characteristic of the otd mutant or normal rhabdomeres in Drosophila suggesting these defects resulted from the misexpression of the non-endogenous Otd (Figure 8G), which prevented further analysis.
As for rescue of r- opsin regulation (Figure 9 and Figures S9 and S10), our results exposed differential effects among the Otd orthologs. For example, Tribolium Otd1 could activate rh3 expression (Figure 9D) but failed to repress rh6 expression (Figure S9D). Tribolium Otd2 in contrast was capable of both (Figure 9E and S9E). On the other hand, Daphnia Otd1 could activate rh3 expression and repress rh6 expression (Figure 9F and S9F), even though Daphnia Otd1 does not appear to be expressed in photoreceptors. Despite the associated cell death and irregular rhabdomere morphology with expression of Daphnia magna Otd2, we observed that the expression of Rh6 appeared to be limited to a single photoreceptor of each ommatidium suggesting Daphnia magna Otd2 can execute the rh6 repression function (Figure S9G). However, we did not detect any expression of Rh3, suggesting that Daphnia magna Otd2 failed to execute the rh3 activation function (Figure 9G). In Drosophila, the expression of opsin rh5 in a subset of R8 photoreceptors is dependent on direct activation by Otd. Moreover, Otd also regulates the feedback loop responsible for generating the correct ratio of Rh5 and Rh6 expressing R8 photoreceptors [16], [48]. In agreement with previous results utilizing this rescue paradigm Drosophila Otd is relatively insufficient in activating rh5 expression [47] and only detected Rh5 expression upon the rescue with Tribolium Otd2 (Figure S10).
Taken together, our expression and rescue assays supported a common role of Pph13 and Otd in r- visual photoreceptor differentiation among Pancrustaceans. For further functional verification, we took advantage of the effective RNAi protocol in Tribolium [49]. Thus to examine the in vivo role of the Tribolium Pph13, otd1, and otd2 genes, we generated double stranded RNA (DsRNA) against each corresponding mRNA for injection into Tribolium larvae. None of the DsRNAs affected developmental timing, viability, or external morphological structures as compared to mock injections; scanning electron microscopy of the adult eye did not reveal any major effects on the external organization of the compound eyes (Figure 10A–E). To investigate the possible role of Pph13 and Otd in rhabdomere morphogenesis, RNAi knockdown adults were prepared for TEM analysis within twelve hours after eclosion to minimize any potential later disruptions of photoreceptor morphology as a result of long-term degeneration. In these specimens, we found that the knockdown of Tribolium Pph13 resulted in the complete absence of rhabdomeres (Figure 10G). The knockdown of either Tribolium otd1 or otd2 caused a completely distinct set of defects. In otd1 knockdowns, the rhabdomeres were present but reduced compared to wild-type controls (Figure 10H). The ordered array of microvillar projections was normal in each of the rhabdomeres but the microvilli were smaller but no evidence of photoreceptor degeneration could be detected. In otd2 knockdown animals, the rhabdomeres were in a state of disarray suggesting degeneration (Figure 10I and Figure S11). We detected whole rhabdomeres that however appeared to be unraveling; as suggested by the presence of large membrane protrusions into the photoreceptor cell body and enlarged distances between microvillar projections. In addition some photoreceptors completely lacked rhabdomeres. Lastly, the combinatorial removal of both otd paralogs resulted in the absence of all rhabdomere structures (Figure 10J).
As a first approximation of whether Pph13 and Otd homologs may be required for the second step defining r- visual photoreceptor differentiation, i.e. the transcription of the phototransduction machinery, we asked if 3XP3-RFP, an in vivo transcriptional reporter for Pph13 activity in Drosophila [20], was disturbed upon reduction of Pph13, otd1 or otd2 (Figure 11). DsRNAs against each transcript were injected into Tribolium m26 larvae, which express RFP from a 3XP3-RFP reporter transgene. In these experiments, only the Pph13 knockdown resulted in the elimination of the photoreceptor specific expression of RFP (Figure 11B and Table S2). Single as well as combinatorial injection of the Tribolium otd1 and otd2 dsRNAs did not affect the expression of 3XP3 reporter (Figure 11C–E).
To further explore whether Tribolium Pph13 and Otd were required for r- opsin expression in Tribolium, we assayed the transcription of the Tribolium LW and UV opsins by RT-PCR (Figure 11F) and by RNA-seq (Figure 11G,H and Table S5 and S6) in the RNAi knockdown conditions. Based on our DNA binding assays and the requirement of Pph13 for 3XP3 expression, we predicted that the knockdown of Pph13 should affect only LW r- opsin transcription. Indeed, the knockdown of Tribolium Pph13 was associated with the reduction of LW transcription but not UV opsin transcription (Figure 11F,G). There was a 3.27 Log2 fold decrease in LW opsin expression in Pph13 knockdown animals as compared to mock injections. The knockdown of Otd2, however, resulted in the virtual absence of UV opsin transcription (Figure 11F,H), as indicated by 5.73 Log2 fold decrease compared to control animals. Finally, DsRNA directed against both Tribolium otd paralogs had no discernible effect on LW opsin transcription (Figure 11F,G). Although Tribolium otd1 is expressed in all photoreceptors, we did not detect any significant effect on UV expression alone or enhancement in combination with knockdown of otd2 (Figure 11F,H).
Our results demonstrate a key role of both Pph13 and Otd for visual r- photoreceptor differentiation among Pancrustaceans. While the conservation of Otx transcription factors in metazoan eye development has been documented by numerous studies [47], [50]–[53], our data provide the first evidence of a deeper evolutionary conservation of Pph13. In combination with the initial failure to detect orthologs in other species beyond Drosophila, the specificity of Pph13 expression and function to a single developmental context, terminal photoreceptor differentiation and maintenance [19], [20], raised the possibility that Pph13 represented a more recently evolved regulator in insect retinal development; our findings here refute this scenario. Moreover, our comparative sequence analyses consolidate that Pph13 arose by duplication of an ancestral singleton member of the Arx gene family of homeodomain transcription factors, as previously hypothesized [18], [19]. Further identification of orthologs will be required to date the exact time point of this gene duplication. In our preliminary analyses we have also identified putative Pph13 and Al orthologs in the mollusk Aplysia califonica. However, searches in genome and transcriptome data from other invertebrates as well as non-Pancrustacean arthropods recovered only orthologs of Al/Arx at this point.
Here, in this study, we have deployed a combination of assays to assess the functional conservation of Pph13 over hundreds of millions of years of Pancrustacean evolution. Our expression analyses revealed that Pph13 is specifically expressed in the visual systems of Daphnia and Tribolium. Furthermore, this study and previous work now identifies the binding of Pph13 to the RCSI site as the driver of default activation of LW-opsins in both Drosophila and Tribolium. Our findings demonstrate the conserved ability of Pph13 homologs to discriminate between RCSI sites, preferably binding the RCSI sites in LW r-opsins [20]. Moreover, the functional analysis in both Drosophila as well as Tribolium reveals that Pph13 is required for transcription of only the LW r- opsins. We therefore believe that activation through Pph13 at the RCSI site was a module in the cis-regulatory control of the ancestral singleton LW-opsin. Consistent with the conservation of this evolutionarily conserved target sequence, all three homologs are capable of rescuing both aspects of r- photoreceptor differentiation in Drosophila Pph13 mutants: rhabdomere biogenesis, and opsin regulation. The observed decrease in Pph13 rescue efficiency with evolutionary distance may be the result of a failure to interact with the required cofactors in Drosophila or to activate transcription through the Drosophila RCSI sites. Given the conserved binding activity of all Pph13 homologs, it is most likely that the lack of sequence conservation outside the homeodomain compromises interactions with cofactors in the across-species rescue experiments.
Our data also demonstrate conserved critical roles of Otd in both aspects of rhabdomeric photoreceptor differentiation. However, the presence of two paralogs in Tribolium and Daphnia, and their differential expression patterns and functional abilities complicate defining the ancestral role of Otd in visual r- photoreceptor differentiation. In all three species at least one Otd ortholog is expressed in developing photoreceptors. Further, the downregulation of Otd orthologs leads to a disruption of rhabdomere formation in both Drosophila and Tribolium. Moreover, with the exception of Daphnia magna Otd2, each Otd homolog that we tested has maintained the ability to direct rhabdomere morphogenesis in Drosophila. The expression of Daphnia magna Otd2 in Drosophila resulted in cell death and as such precluded assessment of its ability to rescue the Drosophila otd mutant. Taken together, these data suggest that the requirement of Otd in rhabdomere morphogenesis is ancestral. In agreement with this, previous studies demonstrated that all three vertebrate OTX paralogs were capable of rescuing rhabdomere morphogenesis when expressed in Drosophila otd mutant photoreceptors [47].
Consistent with the evidence of independent duplication events in the insect and crustacean lineages, we find that there is no simple correlation between the expression profiles of otd paralogs and their ability to direct r- opsin expression in our data set. Most conspicuously, Daphnia magna Otd1, which is not detected in photoreceptor cells within Daphnia magna, has the ability to execute both the repression of rh6 and activation of rh3 r- opsins in the Drosophila Otd rescue paradigm. Daphnia magna Otd2, on the other hand, which is endogenously expressed in photoreceptor cells, was only able to rescue the appropriate repression of Drosophila rh6, which however works in conjunction with defective proventriculus (Dve) [48]. It is also noteworthy that while we could not establish that Daphnia Otd1 is orthologous to Drosophila Otd or Tribolium Otd1, it has the ability to rescue both rh6 repression and rh3 activation whereas Tribolium Otd1 the ortholog of Drosophila Otd rescued only the activation of rh3. Moreover, even though Tribolium Otd1 is expressed in all photoreceptors and has the ability to activate r- opsin transcription in Drosophila, this ortholog is apparently dispensable for r- opsin expression within Tribolium. Interestingly, the three vertebrate Otx homologs also exhibited differential rescuing activities of rh3, rh5 and rh6 regulation [47]. The sum of these data indicates that the paralogs, which originated through independent duplications of the otd locus in insect and crustacean lineages contributed to different subfunctionalization trajectories. While the functional diversification that resulted from this appears bewildering, it implies continued evolutionary interchangeability, which may have been key to consolidating all Otd-related functions onto a single homolog during the loss of otd2 in the lineage to dipteran species.
The activation of structural gene batteries forms the endpoint in the gene regulatory network control of cell differentiation [54]. The synergistic activation of the structural genes by both Otd and Pph13 in the Drosophila eye is a good example of this paradigm. Interestingly, our functional analysis in Tribolium reveals differences in how Pph13 and Otd are employed in directing rhabdomere morphogenesis compared to Drosophila. First, within Tribolium, the reduction of either Otd1 or Otd2 generates non-overlapping defects in rhabdomere morphogenesis while the simultaneous knockdown of both genes leads to complete failure of rhabdomere formation. This outcome could result from incomplete subfunctionalization, leaving a limited degree of genetic redundancy in place. Alternatively, the two paralogs may have limited capacities to compensate for the downregulation of the sister paralog via expression level increase. Second, in Drosophila, both Pph13 and Otd are providing independent and overlapping functions to generate the rhabdomere [20]. As a result, the removal of both Otd and Pph13 is required to generate photoreceptors that lack rhabdomeres. In Tribolium, the knockdown of either Pph13 alone or the knockdown of both otd paralogs is sufficient to eliminate rhabdomeres. Thus, there appears to be significantly less redundant control of rhabdomere formation in Tribolium in contrast to Drosophila. Assuming the general conservation of rhabdomere structure target genes, this finding implies evolutionary turnover of Otd and Pph13 dedicated target genes in the context of rhabdomere formation. Given the well defined binding properties of Pph13 and Otd and the conservation of the Pph13 binding sites between Drosophila and Tribolium, it should be feasible to elucidate the diverged distribution of Otd versus Pph13 targets in Tribolium versus Drosophila. Such studies will expand our understanding of the evolution of gene regulatory networks by specifically testing the proposal that downstream network components enjoy greater degrees of evolutionary freedom than intermediate upstream components [55].
Ever since Darwin pondered about the evolution of the eye [56], the process has and continues to be a challenge to investigators [57]. To date, with respect to photoreceptors, it is now accepted that the two fundamental types, ciliary and rhabdomeric, were present before the split of bilaterian animals and share a common ancestor [3]–[6]. However, compounding the study of the evolution of photoreceptors is the fact that the photoreceptors are utilized in various visual and non-visual light detection systems and obtain diverged morphological states in both protostomes and deuterostomes. Therefore, a critical component to provide clarity for defining homologous photoreceptors [3] is to identify the conserved regulatory proteins and switches required for the differentiation of the various classes of animal photoreceptors. Our studies have now identified two common regulators of one type of photoreceptor: rhabdomeric visual photoreceptors. Nonetheless, a key for a complete understanding will be to not only document their presence in but also confirm their functional roles in the visual systems of emerging model systems; to date our attempts with RNAi against Daphnia Pph13 and otd2 have not been successful. Fortunately, with the advent of TALEN and CRISPR technology [58], [59] functional studies may no longer be the limiting step. Thus future work will seek to define the ancestral mode of rhabdomeric photoreceptor differentiation among protostomes and in addition determine how the regulatory cascade is modified to generate diversity. For example, it will be interesting to explore how the origin of Pph13 relates to the diversification of ciliary and rhabdomeric photoreceptors during early metazoan evolution and whether Pph13 or Otd is required for the differentiation of non-visual rhabdomeric photoreceptors, as exemplified by intrinsic photosensitive retinal ganglion cells (ipRGC) [60], [61]; ipRGCs express an r- opsin but do not develop the characteristic membrane folds of a rhabdomere. With respect to Pph13, we have not observed a vertebrate ortholog. While, Arx transcription factors are known to carry out patterning functions in the developing forebrain that could affect the visual system [62], [63] no Arx functions have been reported that relate to the terminal differentiation of photoreceptors. Furthermore, the identification of Otd orthologs as critical components in both rhabdomeric and ciliary photoreceptor cell differentiation [50]–[53] raises the question whether Otd represents the ancestral mechanism for regulating photoreceptor differentiation in both fundamental types of photoreceptors. Overall, these answers will come from comprehending how the differences in phototransduction and morphology between the two fundamental types of photoreceptors are transcriptionally regulated, whether the regulation is conserved within each photoreceptor type, and how transcriptional regulation is modified dependent upon whether the photoreceptor is incorporated into a visual or non-visual system.
cDNAs representing Tribolium Pph13 and otd2 were constructed from RT-PCR reactions from total RNA isolated from beetle heads. The otd1 cDNA was a gift from Dr. G. Bucher. cDNAs representing Daphnia magna Pph13 and otd2 were constructed from RT-PCR reactions from total RNA isolated from whole adults. The Daphnia magna cDNA of otd1 was isolated by screening an embryonic cDNA library [64]. For transgenics, all cDNAs were cloned into pUASTattB vector and integrated into genome position 65B2 (Rainbow Transgenic Flies, Inc.). Pph13-GAL4 was generated by inserting the immediate upstream 1.6 kb of genomic DNA extending from the first coding Methionine of the Pph13 locus into pCHS-GAL4. Drosophila strains utilized: cn bw, cn bw Pph13hzy [19], otduvi [65], [66], y w; Sp/Cyo; UAS-Flag-otd/TM2 [46], [47], and otduvi; otd-GAL4, Pwiz6/Cyo; TM2/TM6B [46], [47]. For otd rescue experiments females of otduvi; otd-GAL4, Pwiz6/Cyo; TM2/TM6B were crossed to w; +; UAS – otd X transgenic lines and only non CyO;TM6B male progeny were examined. For Pph13 rescue experiments w; cn bw Pph13/CyO; Pph13-GAL4/TM6B homozygous flies were crossed to w; cn bw Pph13/CyO; UAS- Pph13 X/TM6B and only non CyO; TM6B progeny were examined.
All cDNAs were cloned into pCDNA 3.1(Invitrogen) and EMSAs were performed as described in [19], [67]. Proteins were generated in reticulocyte lysates (Promega). The sequences of probes utilized are listed in Table S1.
To generate RNAi knockdown animals, 1 ug/ul of total probe, DsRNA was injected into pu11, m26, or vw late stage larvae; pu11 and m26 contain a 3XP3-GFP and a 3XP3-RFP reporter, respectively [68], [69]. Two independent DsRNAs for each gene were tested and eyes from at least five different subjects were examined to confirm phenotypes. The regions listed for each gene are listed in Table S3. For all the results reported here a mixture of the two DsRNAs for each gene were utilized. The mixtures of DsRNAs contained equal amounts of each individual DsRNA and were used and a final concentration of 1 ug/ul was injected. Dark-reared, newly eclosed (<12 hours old) beetles were collected, scored and utilized for various procedures.
Total RNA was isolated using Trizol and DNAase treated. First strand synthesis was accomplished with the Superscript III (Invitrogen) kit. Each reaction contained 2 ug of total RNA and oligo-dT as primers. PCR amplification was performed with 1 ul (1/20th) of RT reaction and samples were collected at 25 or 30 cycles. All reactions were repeated three times with two independent sets of total RNA. Equal amounts of PCR reactions were analyzed by gel electrophoresis. The list of PCR primers used can be found in Table S4.
For each condition, duplicate sets of total RNA from entire animals (<12 hours old) were isolated. Stranded RNA sequencing libraries were constructed using the TruSeq Stranded mRNA Sample Preparation Kit (Illumina, San Diego, CA) according to manufacturer's instructions. Libraries were quantified using the KAPA SYBR FAST Roche LightCycler 480 2X qPCR Master Mix (Roche, Indianapolis, IN), pooled in equal molar amounts, and sequenced on a HiSeq2000 instrument (Illumina, San Diego, CA) using a 100 bp paired-end run. HISeq read sequences were cleaned using Trimmomatic version 0.30 [70] to remove adapter sequences and perform quality trimming. Trimmomatic was run with the following parameters, “3:30:10 LEADING:3 TRAILING:3SLIDINGWINDOW:4:20 MINLEN:75”. The resulting reads were mapped against the Tribolium release 3.0 gene models (http://www.Beetlebase.org/) using TopHat2 version 2.0.9 [71] with the parameters “–b2-very-sensitive–read-edit-dist 2 –max-multihits 100 –library-type fr-firststrand”. Read counting was done for each gene using htseq-count from the HTSeq package version 0.5.4p3 [72] with the “–stranded = reverse” parameter. For Tribolium, read counts were normalized across samples using the DESeq package (version 1.12) in R/Bioconductor [73]. DESeq [72] was further used to detect statistically significant differences in expression between two conditions using the binomial test with a .05 adjusted p-value cutoff. The complete data set will be presented and discussed elsewhere but total read data and reads specific to Tribolium LW and UV opsin are listed in Tables S5 and Tables S6.
Drosophila and Tribolium heads were prepared as previously described [20], [74]. All samples were from newly emerged adults and for each genotype at least three retinas from three different heads were examined. Samples were observed under transmission electron microscope (TEM), operated at 60 KV and digital images were captured and imported into Adobe Photoshop.
The Tribolium Pph13 polyclonal antibody, 171, was created by injecting rats with a GST-fusion protein representing amino acids 104–220 of the protein. The Daphnia magna Otd2 polyclonal antibody was prepared in guinea pigs against a bacterially expressed C-terminal portion of the protein (amino acids 235–395). The Daphnia magna Otd1 polyclonal antibody was prepared in rats against a bacterially expressed portion of the protein (amino acids 220–351).
Whole mount RNA in situ hybridization in Tribolium and Daphnia was performed as previously described [42], [75]. The regions of the sense and ant-sense probes are listed in Table S3. The following primary antibodies were used: rabbit anti-Rh6 (1∶2500; Dr. C. Desplan), rat anti- Tribolium Pph13 (1∶20), rat anti- Daphnia Otd1 (1∶500; Dr. Y. Shiga), guinea pig anti- Daphnia Otd2 (1∶500; Dr. Y. Shiga) mouse anti-Rh3 (1∶50; Dr. Steve Britt), mouse anti-Rh5 (1∶50; Dr. Steve Britt) and rabbit anti-Rh6 (1∶1000; Dr. Claude Desplan). For Otd2, in Daphnia, signals were amplified with the Tyramide Signal Amplification (TSA) system (PerkinElmer). FITC and Rhodamine conjugated secondary antibodies were utilized (Jackson ImmunoResearch). Immunofluoresence studies were performed as previously described [47], [76]. All Drosophila samples were from newly emerged adults and for each genotype at least three retinas from three different heads were examined.
Many of the sequences utilized in this study can be found in Table S7.
|
10.1371/journal.pntd.0004211 | Socio-economic and Climate Factors Associated with Dengue Fever Spatial Heterogeneity: A Worked Example in New Caledonia | Understanding the factors underlying the spatio-temporal distribution of infectious diseases provides useful information regarding their prevention and control. Dengue fever spatio-temporal patterns result from complex interactions between the virus, the host, and the vector. These interactions can be influenced by environmental conditions. Our objectives were to analyse dengue fever spatial distribution over New Caledonia during epidemic years, to identify some of the main underlying factors, and to predict the spatial evolution of dengue fever under changing climatic conditions, at the 2100 horizon.
We used principal component analysis and support vector machines to analyse and model the influence of climate and socio-economic variables on the mean spatial distribution of 24,272 dengue cases reported from 1995 to 2012 in thirty-three communes of New Caledonia. We then modelled and estimated the future evolution of dengue incidence rates using a regional downscaling of future climate projections.
The spatial distribution of dengue fever cases is highly heterogeneous. The variables most associated with this observed heterogeneity are the mean temperature, the mean number of people per premise, and the mean percentage of unemployed people, a variable highly correlated with people's way of life. Rainfall does not seem to play an important role in the spatial distribution of dengue cases during epidemics. By the end of the 21st century, if temperature increases by approximately 3°C, mean incidence rates during epidemics could double.
In New Caledonia, a subtropical insular environment, both temperature and socio-economic conditions are influencing the spatial spread of dengue fever. Extension of this study to other countries worldwide should improve the knowledge about climate influence on dengue burden and about the complex interplay between different factors. This study presents a methodology that can be used as a step by step guide to model dengue spatial heterogeneity in other countries.
| Dengue fever is the most important viral arthropod-borne disease worldwide and its geographical expansion during the past decades has been of growing concern for scientists and public health authorities because of its heavy sanitary burden and economic impacts. In the absence of an effective vaccine, control is currently limited to vector-control measures. In this context, understanding the sociologic, entomologic and environmental factors underlying dengue dynamics is essential and can provide public health authorities with sound information about control measures to be implemented. In this study, we analyse socio-economic, climatic and epidemiological data to understand the impact of the studied factors on the spatial distribution of dengue cases during epidemic years in New Caledonia, a French island located in the South Pacific. We identify at risk areas, and find that temperature and people’s way of life are key factors determining the level of viral circulation in New Caledonia. Hence, communication campaigns fostering individual protection measures against mosquito bites could help reduce dengue burden in New Caledonia. Using projections of temperature under different scenarios of climate change, we find that dengue incidence rates during epidemics could double by the end of the century, with areas at low risk of dengue fever being highly affected in the future.
| Dengue fever is the most important mosquito-borne viral disease, with an estimated 50 million people being infected each year and 2.5 billion people living in areas at risk of dengue worldwide [1]. The true burden of clinically apparent dengue could be twice as high, and the total burden of dengue fever infections could reach 390 million people when including asymptomatic cases [2]. Whereas only nine countries were affected by dengue epidemics in the 1970's, more than a hundred countries are now reporting dengue outbreaks on a regular basis, making dengue fever the most rapidly spreading mosquito-borne viral disease in the world [1,2]. This rapid global spatial spread over the past 40 years probably results from recent socio-economic changes such as global population growth and uncontrolled urbanisation. Lack of effective mosquito control in endemic areas, increased international air traffic or decay in public health infra-structure in developing countries are also important factors that could explain the rapid regional spread of the disease [3–6]. However, in a given country where there are sufficient numbers of susceptible hosts, these factors need to be associated with suitable climate conditions before dengue fever can establish, since it is transmitted by mosquito species whose life cycle is influenced by temperature, humidity and rainfall [7–9]. Indeed, several studies have pointed out that the current geographic distribution of dengue fever or its vector worldwide could be predicted accurately based on climate variables using either statistical models [10] or deterministic models [11–13]. Other studies have pointed out that climate change could have profound consequences on the epidemiology of dengue fever, because increased temperature and rainfall could facilitate viral transmission and could lead to the geographic expansion of the mosquito species responsible for its transmission [11,14–17].
The complex interplay and relative importance of each factor in the occurrence and spread of dengue fever epidemics might differ from one country to another, depending on the specific climate conditions, cultural and socio-economic environment the virus circulates in [18]. Identifying the factors limiting dengue fever spatial spread at a national level could help understanding the worldwide pattern of dengue disease, could help predicting its future spatial distribution, and could provide national decisions-makers with useful information regarding the appropriate control measures to be implemented.
Most studies trying to identify dengue risk factors spatially were performed at a city scale or a local scale (< 80 km) [19–34]. Among these studies, some have identified risk factors for the presence of Aedes mosquito species, such as socio-economic factors [29,30], proximity to specific plantations [28,32], proximity of potential breeding sites [22,28,29,32], or human behaviour [30,32]. Some studies highlighted the importance of human movement [23,26,31,33] or population immunity [34] in shaping the spatial transmission of dengue fever at small spatial scales. High dengue incidence rates have also been found in neighbourhoods with low social income [20,26], difficult access to piped water [21,26], or no implementation of mosquito protection measures [21,27].
Spatial analyses at a country or territorial scale (> 200 km) are scarce. Some of these studies focused on the spatio-temporal dynamics of the disease only [35–37] and proposed hypotheses about the underlying processes, but did not include analyses of potential explicative factors. To our knowledge, there are only five studies to date identifying and quantifying spatial risk factors for dengue at a “national” scale > ~200 km and < ~1000 km. Four studies identified temperature as having a major influence on the spatial distribution of locally acquired dengue cases [38–41], the last one did not assess the role of climate factors [42]. The role of other factors in the spatial distribution of dengue cases varied from one place to another. For example, in Australia (Queensland) [38] and Taiwan [40], rainfall seemed to play a minor role whereas in Brazil, rainfall played a major role [39]. In Taiwan [40] and Argentina [41], urbanization level was a key factor in dengue fever spatial distribution, and in one province of Thailand [42], the main factor identified was the proximity to major urban centres. No association was found with socio-economic covariates in Argentina [41] or Australia [38].
New Caledonia, where the present study takes place, has a unique situation: it is a developed insular territory located in the inter-tropical area of the South Pacific where the access to high quality data and the lack of terrestrial borders with other countries make it a natural laboratory to study dengue dynamics. A gross average of ten imported cases is detected each year by the Public Health authorities. However, large dengue epidemics develop only every three to five years, sometimes causing the circulation of the same serotype during two consecutive years [43]. A recent study analysing the temporal relationship between dengue epidemic occurrence and climate variables at an inter-annual scale showed that the development of an epidemic in New Caledonia needs precise climate conditions relying on both temperature and relative humidity [43].
The objectives of the present study were i) to characterise the spatial distribution of dengue cases in New Caledonia once an epidemic spreads over the territory; ii) to determine which of possible covariates are shaping the observed distribution; iii) to explore the potential spatial distribution of dengue cases under future climate projections.
We present a complete methodology, from data collection, data transformation, variable selection, and application to future climate projections. We address a number of methodological issues such as spatial autocorrelation, correlation between explanatory variables, or potential non-linearity between epidemiological data and explanatory covariates.
All analyses and figures were performed using R software version 3.2.0 [44], except S3 Fig.
New Caledonia is a French territory, located in the Pacific Ocean 1,500 km East of Australia. It is divided into 33 communes covering 18,576 km2. Out of the 245,344 inhabitants (2009), around 58% (147,365 people) live in Noumea, the main city, and its surroundings. The rest of the population is scattered in small towns of about 2,000 people, or live in rural areas, including traditional Melanesian settlements locally called “tribes” (Fig 1). Although the average population density outside Noumea is very low (5.3 inhabitants per km2), local densities can be high as people gather in small settlements.
New Caledonia is located at the limit of the tropical zone between latitudes 19° and 23° South. The East coast and the West coast are separated by a mountain range culminating at 1629 m. Climate is heterogeneous: the East coast and the southern tip of the main island get more rain than the West coast, as mountains provide a vertical lift to the warm and humid air brought by trade winds. Average rainfall range from 800 mm/year in some western weather stations to 3200 mm/year in the East. Temperature can drop below 10°C during the cool season on clear nights and sometimes rise above 35°C due to the influence of tropical air masses [45].
From an oversimplified point of view, there are three population groups, having different cultural and social habits: Melanesian people, people of French descent who migrated two hundred years ago, and people from various origins who migrated recently. Although the three groups are spatially partially mixed, Melanesian people live mostly on the East coast, whereas the second group live mostly on the West coast and the third group live mainly in Noumea.
In New Caledonia, dengue represents a major public health problem with large epidemics affecting the territory every three to five years and involving a succession of all four serotypes [43,46–48]. Co-circulation of different dengue virus serotypes (DENV1-4) during major epidemics is rare, and has been observed only once (2009). Before 2003, vector control measures consisted in systematic chemical control of adult mosquitoes covering large areas during the warm season, independently from the occurrence of dengue cases. Since 2003, systematic spreading of adulticide has been stopped, and vector control measures include continuous large communication and prevention campaigns fostering source reduction aimed at all citizens, as well as focal chemical control of adult mosquitoes 100 m around declared cases within 24 h of notification. Public Health infrastructure is reliable, and the surveillance system for dengue fever has been efficient for many years. All people have access to medical care, even though people living in remote areas might have more difficulties to reach local health centres.
As described below in the results section, temperature is a key factor determining dengue spatial variability over New Caledonia. We thus decided to explore the evolution of dengue average annual incidence rates during epidemics under changing climate conditions (considering all others variables as remaining constant), by applying the best explicative multivariable model with inputs from maps of temperature for the future (see methods/data/climate covariates: assessing the trends of future mean temperature in New Caledonia). Because the use of kernels in non–linear SVM models impairs predictions outside the observed range of explanatory variables, we built a linear approximation of the best SVM model on present observed data. The linear approximation consists of a simple linear model linking the two best explanatory variables to observed dengue age-standardised average (across epidemic years) annual incidence rates as the response variable. Normality and homoscedasticity of residuals were confirmed by the Shapiro-Wilks' test and the Bartlett's test respectively [67]. To evaluate the error in incidence rates predictions due to the inter-GCM variability of mean temperature increase projections, we calculated, for each time-period and each scenario, the average annual incidence rates during epidemics as predicted by each GCM, and then calculated a standard deviation of predicted annual incidence rates across the different models.
Fig 3 shows that once an epidemic spreads over the territory, dengue cases are distributed heterogeneously. Mean annual age-standardised incidence rates across epidemic years range from 22 to 375 cases per 10,000 people per year, with a mean across communes of 168 cases per 10,000 people per year and a standard deviation across communes of 83 cases per 10,000 people per year. On average the East coast is more affected than the West coast. We can also see that the North-eastern corner of New Caledonia is heavily affected, with dengue incidence rates two to three times higher than in the rest of the territory.
By definition, the average across epidemic years of age-standardised annual incidence rates reflects mainly the spatial pattern of severe epidemics, i.e. epidemics of years 1995, 1998, 2003 and 2009. During years 1995, 1998 and 2003, the North-eastern corner was the most affected. During the 2009 epidemic, the most affected communes were Voh and Koné, on the West coast (see Fig 3 for the location of these communes), but the North Eastern corner was still severely affected [72].
The semi-variograms of dengue incidence rates did not reveal any significant spatial autocorrelation, whether they were calculated for each epidemic year separately or on the average incidence rates across years. This suggests that the local spread of dengue viruses around a case imported in a commune do not exceed the mean radius of a commune in New Caledonia, e.g. approximately 13 kilometres. Hence, we did not incorporate any spatial structure into the subsequent models.
Table 1 shows Pearson's correlation coefficients (rho) between each explanatory variable and dengue age-standardised annual incidence rates averaged across epidemic years. Dengue is spatially positively correlated with variables related to temperature and precipitation, but is negatively correlated with variables reflecting mean thermal range or extreme thermal conditions (see "Isothermality", "Temp range" or the number of days when maximum temperature exceeds 32°C in January, February and March in Table 1). This suggests that, in a given commune, marked temporal variations of temperature is a factor limiting viral circulation. Based on the linear dependence measure of correlation, dengue is also more strongly associated with temperature than with precipitation. Socio-economic variables are highly spatially correlated to dengue average (across epidemic years) annual incidence rates. Variables reflecting people's way of life (e.g. place of birth), local human density (e.g. mean number of people per household, percentage of premises under 40 m2), or human movement are more correlated with dengue average (across epidemic years) annual incidence rates than variables related to the housing type (e.g. premises with inside toilets) (absolute value of rho up to 0.75 for the former and 0.58 for the latter). In particular, the fact that the place where people were born is spatially significantly associated with dengue fever incidence rates (correlation coefficient around 0.5 for people born in New Caledonia and– 0.5 for people born elsewhere) whereas the type of premise is not (absolute correlation coefficient lower than 0.3 for variables describing access to water or electricity) suggests that individual behaviours have a stronger influence on incidence rates than local housing conditions.
Fig 4 shows the PCA results. For clarity reasons, we only show the results of PCA performed on the variables most spatially correlated with dengue average (across epidemic years) annual incidence rates, with an absolute Pearson correlation coefficient over 0.6 for socio-economic variables, and over 0.4 for climate variables (these thresholds were selected after verifying that they did not modify the variable pre-selection results).
PCA of climate variables (Fig 4A) shows that in New Caledonia, temperature is the factor accounting for most of the spatial climatic variability among communes. Temperature is highly correlated with the first PCA axis which represents 68% of the total climatic variance. Temperature and rainfall are not spatially correlated at the commune level. In each group of temperature or rainfall variables, the variables most spatially correlated with dengue average (across epidemic years) annual incidence rates were the average mean temperature (Mean temp) and the mean daily rainfall during the wettest quarter of the year (Wettest quarter) (see Table 1). In addition to these two variables, we decided to keep a third variable, the average daily rainfall, for further statistical modelling, as this variable is more easily available in other countries or climate model simulations.
Fig 4B shows that the spatial variability of socio-economic factors mainly reflects the spatial distribution of people with different cultural habits. Communes where a high proportion of inhabitants live in a tribal way, in small premises, with few means of transportation and a high percentage of unemployment are opposed to communes where many people live a western way of life, in permanent buildings, using air conditioning and getting around using cars. Even though the number of people per premise seems to be correlated with the proportion of people living in tribes, we kept this variable as it stands out of the cluster of variables representing the way of life. We thus decided to keep the percentage of unemployed people and the mean number of inhabitants per housing as representative of socio-economic factors for further statistical modelling.
Table 2 shows the RMSE of the optimised models built on all possible combinations of one, two or three of the five selected explanatory variables (Mean temperature, daily rainfall averaged over the wettest quarter, average daily rainfall, number of people per household and fraction of unemployed people).
When looking at univariable non-linear SVM models, the best variable explaining the spatial heterogeneity of dengue average (across epidemic years) annual incidence rates is the percentage of unemployed people per commune. The second most important explanatory variable is the mean temperature. Rainfall is the least explanatory variable of those selected for multivariable regression modelling. Moreover, the RMSE of models based on observed rainfall almost equal the initial standard deviation (across the territory) of dengue average annual incidence rates, which means that rainfall are poor predictors of dengue average annual incidence rates during epidemic years. The relationship between dengue average annual incidence rates and each of the explanatory variables is linear, except for the fraction of unemployed people (S4 Fig). When looking at the spatial structure of dengue average annual incidence rates predicted by SVM models based only on one of the selected variables (S5 Fig), we see that temperature captures mainly the South to North gradient of increasing incidence rates (S5B Fig) whereas socio-economic variables captures the spatial heterogeneity between the West coast and the East coast (S5E and S5F Fig). Temperature seems to have no influence in communes located below 21°S (S5B Fig).
All models based on two explanatory variables and including at least one variable related to rainfall (best RMSE of ~58 cases per 10,000 people per year) performed worse than the best univariable model (RMSE of ~53 cases per 10,000 people per year). This suggests that in New Caledonia, rainfall has little influence on the spatial variability of dengue viral circulation at the commune level. Models combining two explanatory variables (excluding rainfall) performed better than models based on only one variable. The addition of a third explanatory variable did not improve significantly model performances. Hence we focused our attention on models combining two explanatory variables.
The best explicative model is a model predicting increasing average annual incidence rates during epidemics in communes where the mean temperature and the mean number of people per premise increase (see Fig 5). The influence of these two variables on the spatial structure of dengue incidence rates is close to linear as shown by almost parallel contour lines on Fig 5A. This model accurately predicts the sharp mean increase in incidence in the three communes of the North East of New Caledonia (Hienghène, Ouégoa and Pouébo). The maximal error of the model is observed for Farino (West coast), which is the only commune where all inhabitants live at an altitude higher than 200 m above sea level.
Fig 6 shows the spatial structure of observed average (across epidemic years) annual incidence rates (Fig 6A) and of average annual incidence rates as predicted by the best SVM model based on two explanatory variables (Fig 6B). This model captures the observed spatial heterogeneity in average annual incidence rates between the East coast and the West coast, as well as the sharp increase in the three communes of the North East.
Table 3 shows, for both climate change scenarios, the average increase of mean temperature for the two selected 20-year periods compared to the 1980–1999 historical simulations. All models predict that the mean temperature will increase over time, with projections being more pessimistic for RCP 8.5 simulations. The CMIP5-AR4 inter-model variability in temperature increase is presented in Table 3 and S3 Fig. According to the RCP 8.5 scenario, temperature could increase by more than 3°C by the end of the next century, with a standard deviation across models of only 0.6°C, showing the strong coherency in different model projections.
Fig 6 shows a comparison of the average (across epidemic years) annual dengue incidence rates predicted by the SVM model (panel 6B) or the linear model (panel 6C). The SVM model performs slightly better than the linear one: the correlation coefficient between observed and predicted incidence rates are 0.89 (SVM) and 0.85 (linear), and the RMSE are 42 and 43 cases/10,000 people/year respectively for the SVM and the linear model. The low RMSE of the linear model (~43 cases per 10,000 people per year) shows that the linear model based on the two best explanatory variables is suitable. The Shapiro-Wilks and the Bartlett's test confirmed the normality and homoscedasticity of residuals.
Fig 6D and 6E show the potential future spatial distribution of dengue incidence rates during epidemics according to the RCP 4.5 and RCP 8.5 emission scenarios. By the end of the century, dengue incidence rates during epidemic years could reach a maximum of 378 cases per 10,000 people per year in the most affected commune under the RCP 4.5 scenario (Fig 6D), and 454 cases per 10,000 people per year in the most affected commune under the RCP 8.5 scenario (Fig 6E). Under the RCP 8.5 scenario, communes at low risk now might experience a sharp increase in dengue incidence rates during epidemic years from 64 to more than 200 cases per 10,000 people per year. According to RCP 8.5 climate projections, the average (across communes) dengue mean annual incidence rates during epidemic years could raise by 29 cases per 10,000 people per year for the 2010–2029 period, and by 149 cases per 10,000 people per year for the 2080–2099 period, almost doubling dengue burden in New Caledonia by the end of the century (Table 3).
The spatial association found between temperature and dengue incidence rates during epidemics in New Caledonia can be explained by the influence of temperature on the life cycle of the mosquito transmitting the virus in New Caledonia, Aedes aegypti. High temperatures increase the productivity of the breeding sites through an acceleration of the metabolism of the mosquito, and a faster development of the micro-organisms the larvae feed on, resulting in a higher vector density even with the same number of breeding sites [7,73–75]. High temperatures also speed up the extrinsic incubation period [7,76,77], with the effect that an increased proportion of females Ae. aegypti can reach the infectious stage before dying. Finally, warmer temperatures accelerate the mosquito gonotrophic cycle, and make females Ae. aegypti more aggressive [7,74,78–80], increasing the biting rate and the frequency of potential transmission of viral particles to susceptible hosts. Regarding the effect of increasing temperatures on the mortality of Ae. aegypti adult mosquitoes, a review of 50 field mark-release-recapture studies has shown that in the field, unless temperatures become extreme (over 35°C or less than 5°C), temperature has little effect on daily mortality rate [81], highlighting the central importance of the length of the extrinsic incubation period in the ability of adult mosquitoes to transmit dengue viruses.
In Noumea, the main city, precise climate variables and important thresholds values have been identified as necessary conditions to trigger an epidemic (e.g. number of days when maximal temperature exceeds 32°C in January/February/March, and number of days when maximal relative humidity exceeds 95% during January [43]). At the scale of the entire territory, we found that the spatial distribution of dengue cases during epidemic years is strongly influenced by the average mean temperature. These results suggest that temperature has a major role in dengue dynamics in an insular territory characterised by climate seasonality. However, we did not find a strong association between the spatial distribution of dengue cases during epidemics and average rainfall or with the number of days when maximal temperature exceeds 32°C. The variables influencing either the triggering of an epidemic [43] or its spatial distribution are not the same. Our findings highlight the complexity of studying and understanding dengue dynamics, the importance of well separating the two epidemiological processes of epidemic triggering in a susceptible population, and its intensity once it has started by clearly defining the modelling target (incidence rates for epidemic intensity, or dummy variables for epidemic triggering), and the importance of well defining the scale of study (temporal evolution, or spatial distribution).
The positive association found between the mean temperature and dengue incidence rates is consistent with the one found in previous studies having analysed the spatial distribution of dengue cases at spatial scale > 200 km [38–42]. In these studies as well as in ours, all regions were located between 10° and 25° of latitude, at the fringe of the tropical area, except Argentina, where the region studied extends to 35° South. In the 10° ˗ 25° latitudinal band, annual mean temperature lies in a range of temperature where the life cycle of the mosquito is very sensitive to temperature changes [7].
Some Aedes species, including Ae. aegypti, are able to breed in very small amounts of water, e.g. snails’ shells. Rainfall can play a role in dengue transmission cycle by filling up potential breeding sites [7], thus influencing the vector density. Rainfall also increases the relative humidity, which extends the mosquitoes’ lifespan and therefore the likelihood of those who had an infectious blood meal to reach the infectious stage. However, our study suggests that in New Caledonia, there is no strong association between rainfall and the spatial distribution of cases during epidemics. A plausible explanation can be the multi-factorial nature of dengue fever, and the relative influence each factor plays on dengue dynamics: despite suitable rainfall conditions, dengue might not circulate well if other factors are limiting dengue viral circulation, such as some human behaviour influencing the contact between vector and host. This aspect has been highlighted very clearly in the United States [27]. Some studies have found that the effect of rainfall on vector density can be modulated by human activities such as water storage practices [82]. However, in New Caledonia, we are not aware of specific practices to store water that could explain the lack of association between rainfall and virus circulation intensity. Another potential explanation could be that in dry areas, breeding sites are filled up by other non-climatic mechanisms, such as automatic irrigation or plant watering.
Worldwide, the spatial association between rainfall and the spatial distribution of dengue cases at a “national” scale (> 200 km) is not as clear as the one for temperature: one spatial study did not find any association between dengue incidence rates and rainfall [40], whereas two others did [38,39]. Other factors influencing dengue transmission (e.g. anthropogenic factors influencing the availability of filled breeding sites) and not included in the different studies might blur the rainfall signal.
It would be interesting to perform the same kind of multi-factorial spatial analysis in areas of epidemic or endemic transmission located closer to the equator, where the mean temperature is higher, to see what climatic factors impact the spatial distribution of cases. This kind of study could help understand better the complex interplay between the different factors (climate, socio-economic, immunologic, viral, entomologic…) associated with dengue fever transmission.
Regarding the link between socio-economic variables and dengue incidence rates during epidemics, a limitation of this study is the absence of historical time series of socio-economic variables. We then had to assume that the data retrieved from the 2009 census is representative of the mean socio-economic spatial pattern over the epidemic years of the 1995–2012 period. As there has been no major historical event leading to population migration in New Caledonia during this time period and as socio-economic variables represent mainly people’s way of life, we think this assumption is realistic.
Our results are consistent with previous studies that have pointed out the importance of socio-economic factors on the spatial distribution of dengue cases, whatever the spatial scale studied: national (>200 km) [39–42] or local (<10 km) [20,26,29,83]. The spatial association between the percentage of unemployed people and dengue in New Caledonia cannot be interpreted in terms of lack of economic activity only, as shown by the PCA on socio-economic factors. This variable we selected as input for the models is highly correlated with other variables reflecting the way of life, socio-economic and cultural differences existing in New Caledonia, which are in turn highly correlated to housing type. Therefore, at this spatial scale in New Caledonia, it is difficult to statistically differentiate the role played by human behaviour, human activity or housing type in dengue fever transmission. However, those three factors influence the contact rate between viraemic patients or susceptible hosts on one hand, and mosquitoes on the other hand. This highlights the importance of limiting the contact between humans and vectors and should lead local authorities to strengthen communication campaigns about personal protection measures towards populations at risk.
Regarding the spatial association found between the fraction of unemployed people (i.e. people’s way of life) and dengue incidence rates during epidemics in New Caledonia, it is interesting to point out that on the East coast, a larger fraction of inhabitants are Melanesian people living in tribes, whereas on the West coast, the majority of people are people from French descent having a western way of life. It would be interesting to perform sociologic studies to precisely identify which human behaviour leads to an increased risk of catching dengue fever. Such information would be useful to define communication messages towards at risk populations.
The spatial association between the number of people per household and dengue incidence rates can be explained by the short flight range of Ae. aegypti mosquitoes. These mosquitoes are often captured in the very house where they emerged or in the neighbouring houses, flying an average of 40 to 80 m during their life [84–87]. Hence, dengue outbreaks involving Ae. aegypti as the main vector are known to be highly spatially focal, with dengue cases usually clustering within 200 m to 800 m of each other [23,33,34,88–94]. Our results suggest that, in New Caledonia, dengue cases probably cluster within houses. Sick people should protect themselves until they are no longer vireamic to avoid human to mosquito transmission, and people living around a case should protect themselves to avoid getting infected while infectious mosquitoes are still active in the neighbourhood. Taking such individual actions could reduce the intensity of dengue transmission and reduce dengue burden over the territory. This message could be strengthened in the recommendations given by the authorities.
The results about climate change must be interpreted keeping in mind that they represent a climate risk only, and that the spatial association between dengue incidence rates during epidemics and temperature might change over time depending on socio-demographic changes, or changes in dengue control strategy. Assuming all other factors remain constant in time, our results suggest that Public Health authorities can expect the dengue burden to raise significantly during the next century over the territory, and can expect the dengue spatial range to increase. As the GCM projections are spatially homogeneous over the territory, and as the model used to predict dengue incidence rates in the future is linear and is based on only one climate variable, the predicted absolute increase in dengue incidence rates is currently the same for all communes. This highlights the need for spatially downscaling GCM projections to gain a better understanding of the impact of climate change in the future.
Communes that are already severely affected by dengue epidemics will have to prepare to face higher burden of dengue fever. For communes that are at low risk now, we can see that in the future they might be affected as severely as communes at high risk now. These communes might not be prepared now to face severe epidemics of dengue fever, and they will probably need support for adaptation.
As said earlier, the positive association found between temperature and dengue incidence rates during epidemics for mean temperatures ranging from 22°C to 25°C can be explained by the effect of temperature on the mosquito life cycle and duration of extrinsic incubation period. Here, by applying a statistical model built using current observed temperature to future projections, we make the assumption that the biological effect of temperature on the mosquito life cycle and on the extrinsic incubation will remain the same under the range of temperature that might be observed in the future. For most parameters influencing transmission, this assumption is reasonable. For example, we know that in Thaïland, for DENV-2, the extrinsic incubation period is reduced from 15 days at 30°C to 7 days at 32–35°C [77], which is in support of increasing temperatures inducing an increase in dengue incidence rates under future climate. However, because we used a statistical model, we were not able to incorporate the known negative effect that an increase in temperature might have on dengue transmission when temperature reaches extremes. For example, a review of fifty mark-release-recapture studies has shown that the survival and longevity of Ae. Aegypti mosquitoes is highly reduced when temperatures exceed a threshold, which might be around 35°C [81]. It would be interesting to develop models that are able to integrate these negative effects in the future in order to gain a better understanding of the effects of climate change on dengue transmission.
Here we used data collected routinely by the Direction of Sanitary and Social Affairs. As any surveillance system, it is highly probable that not all dengue cases have been recorded. However, in New Caledonia, the data is of high quality, and the spatial standardisation of the surveillance system (i.e. all the actions taken to be able to compare data collected by different people, at different places [95]) is good, which means that the proportion of cases that are not recorded by the surveillance system are probably comparable from one commune to another. Hence, maps of incidence rates calculated from routinely collected data can be used to study the spatial variability of true incidence rates.
To calculate mean incidence rates, we have used the consultation date of cases. Consultation can occur 1 to 5 days after the onset of symptoms, and the incubation period lasts 4 to7 days on average [3], which means that the consultation date can differ from one to two weeks from the date of infection. This loss of temporal precision is not important here to calculate maps of incidence rates, as for each commune, we have averaged incidence rates across many years.
As in any epidemiological spatial study using routinely collected surveillance data, it is possible that some spatial bias has been introduced due to the fact that the spatial data recorded is the commune where people live, which can sometimes differ from the commune where they got infected.
Another limitation of using routinely collected data is that only clinically apparent cases are recorded, dismissing clinically inapparent cases, whose proportion can vary in time and space [96,97]. A seroprevalence survey is currently undertaken by the Public health authorities. It will be interesting to compare the spatial distribution of seroprevalence to the spatial distribution of average incidence.
The analysis of the spatial pattern of infectious diseases, in relation with environmental or socio-economic factors raises a number of methodological issues, such as the presence of spatial auto-correlation, the spatial scale of aggregation of the data, the existence of possible non-linear links between the response and the explanatory variables, or the presence of multi-collinearity between the response variables. Most issues have already been addressed in the past, and solutions already exist to handle them. For example, the reviews by Dormann et al. deal with multi-collinearity [66] or spatial-autocorrelation [98]. The main issue we have been confronted with in our study was the spatial upscaling of meteorological data observed at precise locations to the same spatial level of aggregation as the epidemiological data. In existing spatial studies of dengue fever at a national scale, authors have geo-spatially interpolated climate variables on regular grids using kriging methods, and have averaged gridded values over a given administrative division [38–41]. This approach has two drawbacks in New Caledonia. Simple kriging models do not take into account the potential elevation between two given points, leading to biased estimates of temperature in mountainous regions. Moreover, the traditional approach used in climatology, which consist in aggregating temperature over grid points taken uniformly over the whole aggregative area makes the implicit assumption that people at risk are distributed homogeneously over the aggregative area. This is particularly problematic in New Caledonia where large areas are not inhabited. In our approach, as epidemiological data are collected at the individual level, we tried to estimate the climate conditions for each individual (and therefore for the mosquitoes surrounding each individual). However, the algorithm used introduces some noise, due to the fact that the weather stations are sometimes kilometres away from some towns or tribes. High spatial resolution climate data obtained from high resolution modelling of atmospheric conditions could be used, but some noise will be introduced by the modelling error compared to the observed data. This issue needs further attention in the future to increase the quality of spatial epidemiological and environmental studies.
Some factors that could influence the spatial distribution of dengue cases during epidemics have not been taken into account in this study: the location of the first cases introduced each year, the spatial variability in population immunity, viral factors such as the serotype circulating, or factors associated to the mosquito such as the spatial variability in vector competence, or dengue vector control measures. We decided not to include the serotype, as we performed the analysis on averages over several years, and as, except in 2009, there was no co-circulation of different serotypes over the territory. Therefore, spatial differences in the level of viral circulation cannot be associated with genetic differences between serotypes. A territorial seroprevalence survey to assess population immunity has been implemented recently in New Caledonia, but data are not available yet. It could be interesting to include environmental variables derived from GIS data or remote-sensing in this kind of study. For example, GIS data about built areas could be used to create indicators of the proximity of houses to reflect the fragmentation of the Ae. aegypti habitat per commune, given that fragmentation of this habitat could potentially slow down viral circulation. The amount of vector control effort implemented in each commune is heterogeneous on the territory, as this activity falls within the commune’s authority, and each commune is free to implement or not the territorial guidelines. The vector control effort in each commune is thus difficult to quantify, and data were not available yet at the time of analysis. As soon as these data will be collected by local authorities, they could be incorporated in the modelling process to assess the efficiency of vector control measures.
The study we present here is about the spatial heterogeneity of dengue incidence rates across epidemic years, independently of the inter-annual variability of dengue incidence rates from one epidemic year to another. The mean spatial pattern studied is very robust to changes in the definition of an epidemic year, as severe epidemics will always be considered in the calculation of the mean, whatever the threshold used to distinguish epidemic from non-epidemic years. It would be interesting to know whether the spatial association found here between the severity of dengue epidemics, temperature, local people’s density and people’s way of life is consistent through time or not, and to identify the factors associated to the temporal variability of spatial patterns.
Two other viruses transmitted by Ae. aegypti like dengue virus caused outbreaks recently in New Caledonia. Chikungunya virus has been introduced on four occasions since 2011 but in each case, the outbreaks were limited to a few cases in Noumea and surroundings. Conversely, Zika virus caused large epidemics over the territory in 2014 and 2015, with more than 1,500 confirmed cases and more than 11,000 estimated cases. Although these viruses are transmitted by the same mosquito as dengue fever, no sufficient data are available to know if the socio-economic and climatic factors driving epidemics are the same. It is likely that local vector competence and population immunity represent major limiting factors. Although dengue has caused major outbreaks in NC in 2013, chikungunya viruses have only caused a limited number of cases for reasons that remain unexplained today and despite the competence of local Ae. aegypti for chikungunya virus transmission [99]. It is likely that climatic factors and interactions between viruses circulating together between human-hosts and mosquito-vectors influence the epidemiology of arboviruses in New Caledonia. A comparative analysis of the spatio-temporal distribution of these three arboviruses in an insular territory accommodating only Ae. aegypti represents an important issue to understand and predict outbreaks.
|
10.1371/journal.ppat.1007614 | Interleukin 21 collaborates with interferon-γ for the optimal expression of interferon-stimulated genes and enhances protection against enteric microbial infection | The mucosal surface of the intestinal tract represents a major entry route for many microbes. Despite recent progress in the understanding of the IL-21/IL-21R signaling axis in the generation of germinal center B cells, the roles played by this signaling pathway in the context of enteric microbial infections is not well-understood. Here, we demonstrate that Il21r-/- mice are more susceptible to colonic microbial infection, and in the process discovered that the IL-21/IL-21R signaling axis surprisingly collaborates with the IFN-γ/IFN-γR signaling pathway to enhance the expression of interferon-stimulated genes (ISGs) required for protection, via amplifying activation of STAT1 in mucosal CD4+ T cells in a murine model of Citrobacter rodentium colitis. As expected, conditional deletion of STAT3 in CD4+ T cells indicated that STAT3 also contributed importantly to host defense against C. rodentium infection in the colon. However, the collaboration between IL-21 and IFN-γ to enhance the phosphorylation of STAT1 and upregulate ISGs was independent of STAT3. Unveiling this previously unreported crosstalk between these two cytokine networks and their downstream genes induced will provide insight into the development of novel therapeutic targets for colonic infections, inflammatory bowel disease, and promotion of mucosal vaccine efficacy.
| Diarrheal diseases still remain the second leading cause of mortality in children younger than 5 years old worldwide, leading to 1.3 million deaths per annum. The diarrheagenic Escherichia coli (DEC) pathotypes are considered NIAID Biodefense Category B agents. Human infections with enteropathogenic and enterohemorrhagic Escherichia coli (EPEC and EHEC, respectively) are associated with human disease. EPEC is a common cause of infantile diarrhea in the developing and underdeveloped world, and EHEC is considered an emerging zoonotic infection. These enteric pathogens cause a wide range of clinical symptoms, varying from mild diarrhea to more complicated clinical presentations, including hemolytic-uremic syndrome (HUS) and hemorrhagic colitis. Using a murine model of Citrobacter rodentium infection, we found the requirement of a functional IL-21/IL-21R signaling axis in the control of enteric microbial infections via augmenting activation of STAT1 in mucosal CD4+ T cells in a murine model of Citrobacter rodentium colitis. Understanding how the IL-21/IL-21R signaling pathway contributes to the host immunity in the colon will further provide insights into the development of novel preventive and therapeutic targets for human subjects with enteric microbial infections and other inflammatory conditions, including inflammatory bowel disease (IBD) and celiac disease.
| Several microbial pathogens elicit the type I (e.g. IFN-α, IFN-β), type II (IFN-γ) or type III (e.g. IFN-λ) interferons, leading to the transcription of several hundred interferon-stimulated genes (ISGs) [1, 2]. The activation of each of these interferon systems induces a distinct but partially overlapping set of signature ISGs which have been reported to contribute important roles to host defense against a wide variety of pathogens, including viruses, bacteria, and parasites, by directly targeting genes conferring resistance to infection [3, 4]. The essential requirement of ISGs in host immunity has been shown by the fact that mice deficient in one or more ISGs are more susceptible to infections with multiple viral and bacterial pathogens [5–7]. The role of ISGs in host defense at mucosal surfaces of the gut is not fully understood. In an attempt to elucidate the mechanisms by which infection is controlled at mucosal surfaces of the small intestine, the cooperation between two other cytokine networks, interferon lambda (IFN-λ) (not IFN-γ) and IL-22, was shown to be required for optimal control of viral replication in a mouse model of rotavirus infection [6]. However, the role of the IL-21/IL-21R signaling axis in colitis of any origin and in collaboration with IFN-γ is not known.
Infection of mice with the murine enteric pathogen, Citrobacter rodentium, is considered a robust model to study host immune response to minimally-invasive gut pathogens at the mucosal surfaces of the large intestine [8–10]. This extracellular bacterial pathogen shares virulence features with the closely related human enteropathogenic Escherichia coli (EPEC) and enterohemorrhagic Escherichia coli (EHEC), making it an ideal surrogate model to study the human disease [8, 10]. Several lines of evidence suggest that the immune response to the human EPEC and EHEC relies on both innate [11–13] and adaptive [14–17] arms of the immune system. The C. rodentium colonization elicits a robust TH1 response, characterized by the predominant production of IFN-γ, IL-12 [14] and TNF-α [18], as well as a TH17 response producing IL-17 [19]. The absolute requirement of TH1 responses in host defense against C. rodentium has been reinforced by the observations that CD4+ T cell-deficient mice (CD4-/-) (but not CD8+ T cell-deficient mice) IFN-γ knockout (Ifng-/-) mice and IL-12-deficient mice (IL-12p40-/-) showed greater susceptibility to enteric infection, increased systemic dissemination of the bacterium and enhanced pathology after C. rodentium infection [16, 17, 20].
IL-21 binding to its cognate receptor, IL-21R, results in the activation of Janus kinase 1 (JAK1) and JAK3 and the subsequent phosphorylation of signal transducer and activator of transcription (STAT) proteins, mainly STAT3 but also STAT1 and STAT5 [21, 22]. The activated phospho-STAT proteins dimerize and translocate into the nucleus, bind to the interferon (IFN)-γ-activated sequence (GAS) motif and initiate a transcription program that includes some IL-21 target genes [23]. Although the cascade of events occurring after the IL-21-activation of STAT3 is well-studied, it is still not fully understood how the activation of STAT1 via IL-21 influences downstream target genes [23]. However, it is interesting that IL-21 via STAT1 can augment expression of both Tbx21 and Ifng gene expression as well as expression of certain interferon-regulated genes Ifit1 and Ifit2, that are also IL-21 targets, and that STAT3 activation diminishes these effects, either in mice or humans [24].
Despite the critical requirement of the IL-21/IL-21R signaling axis in the generation of germinal center B cells [25], the roles played by this signaling pathway in the context of enteric microbial infection is not well-characterized. Based on this, we investigated the role of IL-21/IL-21R axis, in collaboration with the type II interferon, IFN-γ, in protection against enteric microbial infection with C. rodentium in the colon. We found that an intact IL-21/IL-21R axis was required for resistance against and clearance of enteric infection with this pathogen and that activated CD4+ T cells were the exclusive expressors of IL-21 following infection with C. rodentium in the distal colon. The CD4+ T cell-derived IL-21 curtailed enteric infection and also contributed to C. rodentium-induced inflammation and pathology at the mucosal surfaces of the colon. We also found that IL-21 acted in concert with IFN-γ to optimally activate STAT1 in CD4+ T cells and to promote subsequent optimal expression of ISGs in the distal colon. These events were independent of STAT3. Our findings revealed a previously unknown effector function for the IL-21/IL-21R signaling axis in amplification of ISG expression and the modulation of host response to microbial infection at the mucosal surfaces of the gut. The understanding of the mechanisms by which the IL-21/IL-21R signaling axis regulates intestinal epithelial integrity and host immunity after infection with minimally-invasive gut pathogens (e.g. Escherichia coli) will provide insights into novel preventive and therapeutic targets for the control of human infections with enteric bacterial pathogens. Such collaboration between IL-21 and IFN-γ also provides mechanisms by which the IL-21/IL-21R signaling axis regulates inflammation in the colon and provides insights into novel preventive and therapeutic targets for inflammatory conditions in humans, including inflammatory bowel disease, as well as inflammation-induced cancers.
Using the murine intestinal pathogen, C. rodentium, we investigated the requirement of the IL-21/IL-21R signaling axis in protection against mucosal microbial infection in the gut. Our findings indicated that Il21r-/- mice had significantly higher (2 logs) bacterial burden in the feces as compared with WT controls, both early and late in the infection, although the peak bacterial load was comparable (Fig 1A and 1B). While the WT controls were able to control the infection with C. rodentium by day 21 p.i., Il21r-/- mice had an impaired ability to clear the enteric infection with this pathogen (Fig 1A and 1B). The differences in fecal bacterial burdens were noticeable as early as day 2 p.i. in Il21r-/-mice, suggesting an important roleplayed by this signaling pathway in early host protection events to this enteric pathogen. Although the WT and Il21r-/- mice were cohoused for 2 weeks to equilibrate microbiota, we also bred matched WT heterozygotes and homozygous Il21r-/- mice from the same set of homozygous Il21r-/- females by crossing with WT or Il21r-/- males, and then kept the litters nursing together until weaned, so that they would obtain the same microbiota from their mothers intrapartum and during nursing/foster nursing. Consistently, mice homozygous for the targeted mutation of IL-21R (Il21r-/-) showed significantly higher bacterial burdens and delayed clearance of C. rodentium infection as compared with their heterozygous littermate controls (Il21-/+) that were bred together (S1 Fig), confirming that the difference in susceptibility to C. rodentium infection was not due to distinct microbiomes. Furthermore, comparable infection kinetics were observed following infection with an OVA-expressing C. rodentium, indicating that overexpression of a plasmid carrying the chicken ovalbumin did not alter the infectivity, or the ability to colonize the mucosal surfaces, of OVA-Citrobacter as compared with WT-Citrobacter (S2 Fig). Collectively, these findings were consistent with one previous report indicating a role played by the IL-21/IL-21R axis in protection against C. rodentium [26].
Despite effective bacterial replication and significantly higher bacterial burden in Il21r-/- mice, histology demonstrated only a moderate increase in inflammatory cell recruitment, predominantly in the mucosa, and mild hyperplasia accompanied with loss of goblet cells in the distal colons of Il21r-/- mice 9 days after infection with C. rodentium, the time of peak infection when bacterial loads are not significantly different between WT and Il21r-/- mice (Fig 1A–1C). In WT controls increased numbers of inflammatory cells were observed in the mucosa extending into the submucosa of the distal colon. Epithelial cell erosion and ulcerations, and submucosa edema, were more prominent in WT mice than in their Il21r-/- counterparts at day 9 p.i. (Fig 1C). Crypt hyperplasia and loss of goblet cells accompanied by significant submucosal edema with significantly higher perivascular inflammatory cells were more noticeable in the distal colon of WT mice 9 days after infection than in the Il21r-/- mice (Fig 1D and 1E). However, the percentage and the absolute numbers of those cells were comparable in whole colons between the two genotypes at days 3 and 9 after C. rodentium infection (Fig 1F and S3A–S3G Fig). Overall, despite having higher bacterial burden, the Il21r-/- mice appeared to have less inflammation, indicating that IL-21 response was a critical factor in the inflammatory response and at least part of that inflammatory response may be necessary for control of the bacterial load.
Using an ex-vivo organ culture system, we examined the kinetics of IL-21 production by ELISA in the distal colon of WT mice following C. rodentium infection. As shown in Fig 2A, the peak of IL-21 production in the distal colon of WT mice infected with C. rodentium occurred 9 days p.i., when the infection was at its peak. Likewise, similar kinetics were observed for other cytokines important for protection against C. rodentium infection, including IFN-γ, IL-17A and IL-22 in the distal colon of WT (Fig 2A). To further investigate the extent to which immune or non-immune cell types in the distal colon contributed to the expression of IL-21 after C. rodentium infection, we sorted cells in the colon into hematopoietic (CD45+EpCAM-) or non-hematopoietic (CD45-EpCAM+) cells and sorted for immune cell types. Using the Nanostring method, we observed that IL-21 was almost exclusively expressed by mucosal CD4+ T cells (CD45+EpCAM-CD3+CD4+) in WT mice 9 days after C. rodentium infection, while other cells of hematopoietic origin, such as natural killer (NK) cells, dendritic cells (DCs), neutrophils, and macrophages, did not express significant levels of IL-21 transcripts following infection (Fig 2B). Furthermore, mucosal CD4+ T cells expressed higher levels of transcripts for IL-21R as compared with other immune and non-immune cells after infection with C. rodentium (Fig 2C). However, there was some IL-21R expression by innate cells such as NK cells, neutrophils, macrophages and DCs that may contribute to the difference in early control of C. rodentium at day 2–3 pi. (Fig 1A and 1B). Interestingly, the colonic intestinal epithelial cells (IECs) expressed neither detectable transcripts for IL-21 (Fig 2B) nor mRNA for IL-21R following infection (Fig 2C). The latter observations were consistent with lack of IL-21R expression in a C57BL/6 colon carcinoma cell line (MC-38) model of colonic IECs (S4 Fig).
We performed principal component analysis (PCA) of gene expression profiles in the whole distal colon of WT and Il21r-/- mice using Nanostring. Our analysis showed four distinct clusters between uninfected and infected WT and Il21r-/- mice 9 days after infection with C. rodentium (Fig 3A). While the uninfected WT and Il21r-/- mice clustered closed to each other, the infected WT and Il21r-/- mice formed two distinct clusters that were separated far from each other, indicating differences in their gene expression profiles.
Several lines of evidence suggest that colonization with C. rodentium elicits a robust, highly polarized TH1 response in the colon, as shown by increased expression of IFN-γ, tumor necrosis factor (TNF)-α and IL-12 [14]. The critical requirement of IFN-γ during C. rodentium infection has been highlighted by the observations that IFN-γ knockout (Ifnγ-/-) mice had an impaired ability to clear infection [14, 17]. In this model, IFN-γ produced by antigen-experienced CD4+ T cells mediates the mucosal immune response to C. rodentium and its subsequent eradication [20]. Our findings demonstrated that type I- and type II-specific, but not type III-specific, ISGs were induced after C. rodentium infection in the distal colons of WT mice (Fig 3B, 3C and 3D and S5 Fig). Considering the key requirement of IFN-γ and downstream ISGs in host defense against C. rodentium infection, we next investigated the contribution of IL-21/IL-21R signaling axis to the optimal expression of ISGs in the colon of Il21r-/- mice under homeostatic conditions as well as 9 days after enteric infection. Our findings demonstrated that most genes impaired in the whole distal colon of Il21r-/- mice infected with C. rodentium were known ISGs (Fig 3C and 3D). In each case, the fold increase during infection compared to uninfected controls was substantially less in the Il21r-/- mice than in the WT controls (Fig 3C), and in most cases the increase in Il21r-/- mice was not significant, and the increase in the WT was significantly greater than the increase in Il21r-/- mice. Interestingly, the expression of both the type I- and type II-specific ISGs were impaired in the whole colon of Il21r-/- mice 9 days after C. rodentium infection (Fig 3D and 3E).
To further explore the extent of impaired expression of ISGs in the absence of an intact IL-21/IL-21R axis specifically in CD4+ T cells, FACS-sorted CD4+ T cells (the main producers of IL-21 and main expressors of IL-21R, Fig 2B and 2C) were isolated from the distal colon lamina propria (LP) of Il21r-/- mice and WT controls 9 days after C. rodentium infection and analyzed by Nanostring. Consistently, our findings indicated that the majority of genes impaired in CD4+ T cells isolated from the LP of Il21r-/- mice were known ISGs (Fig 3F and 3G), indicated by red bars (Fig 3G). Gene ontology analysis of processes enriched in mucosal CD4+ T cells impaired from the distal colons of Il21r-/- mice identified genes with a wide range of functions (Fig 3H). Likewise, the expression of type I- and type II-specific, but not type III-specific, ISGs by CD4+ T cells was impaired in Il21r-/- mice as compared with WT controls (Fig 3I).
In view of these results, we hypothesized that IL-21 and IFN-γ produced in response to C. rodentium infection may act in concert for the optimal expression of ISGs in the colon and are required for the control of C. rodentium infection in vivo. To experimentally test the hypothesis, we treated naïve splenocytes with recombinant murine IL-21 or IFN-γ alone or in combination for 24 hr and the expression of representative ISGs LAG3 and granzyme A by CD4+ T cells was measured by flow cytometry. When cells were treated with a combination of recombinant murine IL-21 or IFN-γ the expression of representative ISGs by CD4+ T cells isolated from WT mice was significantly upregulated compared to cells treated with either IL-21 or IFN-γ alone (Fig 3J for LAG-3 and S6 Fig for granzyme A). Interestingly, almost all CD4+ T cells positive for IFN-γR were also positive for IL-21R as well (S7 Fig), so the same cells could respond to both cytokines, not a sum of some cells that could respond to IFN-γ and some to IL-21. To further address whether differences in the gene expression profiles between WT and Il21r-/- mice did not merely reflect the severity of inflammation, we investigated the expression of a representative ISG, LAG-3, expressed by CD4+ T cells isolated from the colonic LP of naïve (uninfected) WT and Il21r-/- mice. Remarkably, Il21r-/- mice expressed significantly lower surface expression of LAG-3 (p = 0.001; n = 7 animals) even in the absence of colonic inflammation (S8 Fig). These findings indicate that the differences in the gene expression profiles between WT and Il21r-/- mice are not likely a consequence of the inflammation severity. Collectively, these findings indicate previously unrecognized collaboration between IL-21 and the IFN-γ signaling pathway to optimally express ISGs and a requirement for an intact IL-21/IL-21R signaling axis for the optimal expression of ISGs by CD4+ T cells.
The absolute requirement of CD4+ T cell responses during C. rodentium infection has been reinforced by the observations that CD4+ T cell-deficient mice (CD4-/-), but not CD8+ T cell-deficient mice (β2m-/-), showed greater susceptibility to C. rodentium-induced colitis and increased systemic dissemination of the bacterium to extra-intestinal sites [16]. Consistent with our findings that IFN-γ, but not IFN-α, was the dominant interferon expressed in the colon after infection with C. rodentium (Fig 4A), significantly lower concentrations of IFN-α than IFN-γ were detected in the colon of both infected WT and Il21r-/- mice (Fig 4B). Collectively, these findings suggested that IFN-γ is the predominant interferon produced in the colon in response to C. rodentium and that the lack of an intact IL-21/IL-21R signaling axis does not negatively affect the IFN-γ expression and production in response to infection in Il21r-/- mice.
Considering that both type I- and type II-specific ISGs were impaired in infected Il21r-/- mice, we investigated which type of interferon was required for the control of infection. Type I interferons have been studied mostly in the context of viral infections [7]. The roles played by type I interferons in non-viral infections, including bacterial infections have recently been investigated [7, 27, 28]. We further experimentally determined roles played by type I interferons in host protection following C. rodentium infection. Consistent with these results, mice deficient in interferon-α/β receptor (Ifnar-/-) had bacterial burdens comparable to those of their WT controls and were able to efficiently clear infection after oral challenge (Fig 4C). However, mice deficient in IFN-γ (Ifng-/-) showed a delayed clearance similar to that seen in Il21r-/- mice and exhibited significant weight loss early in the course of C. rodentium infection (Fig 4D and 4E). These findings suggested that type I IFNs played minimal, if any, roles in protection against C. rodentium infection in mice and that the type II IFN (i.e. IFN-γ) contributed significantly to host protection.
The earliest step at which IL-21 could influence production of ISGs is in the production of IFN-γ itself, so we addressed these levels in the Il21r-/- mice. It has been shown by other investigators that Il21r-/- mice express significantly higher levels of IFN-γ in the colon LP compared with WT counterparts during dextran sulfate sodium (DSS)-induced colitis [29]. Our findings demonstrated that IFN-γ was the dominant interferon, mainly expressed by CD4+ T cells. However, dendritic cells, to a much lesser extent, expressed IFN-β, but not IFN-α or IFN-λ or much IFN-γ (Fig 4A). Consistent with these findings, by using an ex-vivo organ culture system we found that the lack of the IL-21/IL-21R signaling axis did not negatively affect the production of IFN-γ in the distal colon of Il21r-/- mice, as evidenced by significantly higher levels of IFN-γ production in the distal colon as compared with WT controls 9 days after C. rodentium infection (Fig 4B). Although the expression of IFN-γ by gut-associated CD4+ T cells isolated from Il21r-/- mice was impaired (~3 fold) as compared with WT controls, higher expression of the cytokine by NK cells (~3 fold) could explain higher levels of IFN-γ observed in the whole distal colon of the Il21r-/- mice (S9 Fig). Likewise, significantly higher levels of IL-17A were noted in the distal colon of Il21r-/- mice as compared with WT controls at that time (Fig 4B). This contrasts with the poorer control of the infection, which should benefit from increased IFN-γ and IL-17. No significant differences were observed in the intracellular expression of IFN-γ by CD4+ T cells isolated from the colonic LP of either naïve Il21r-/- mice or naïve WT controls (Fig 4F). Intracellular staining for IFN-γ expression by ovalbumin-specific mucosal CD4+ T cells in WT mice after OVA-Citrobacter infection demonstrated that CD4+ T cells are a major source of Citrobacter-induced IFN-γ following infection in the intestine (Fig 4G). Because IFN-γ production is higher, not lower, in the colons of Il21r-/- mice, the reduction in ISGs cannot be due to a decrease in IFN-γ itself or to the need for IL-21 signaling to optimally induce IFN-γ. Thus, the effect on ISGs must be downstream of levels of IFN-γ itself. We therefore asked whether the effect on ISGs was due to a decrease in IFN-γ receptor expression, the next step that might account for the lower levels of ISGs. The lack of the IL-21/IL-21R signaling axis did not impair the expression of IFN-γR1 and IFN-γR2 (CD119) by CD4+ T cells isolated from the LP of Il21r-/- mice 9 days after C. rodentium infection as compared with WT controls (Fig 4H–4K). Indeed, CD4+ T cells isolated from the LP of Il21r-/- mice expressed similar levels of IFN-γR1 and IFN-γR2 as WT controls (Fig 4H–4K). Collectively, these findings demonstrated that the IL-21/IL-21R axis was not required for optimal expression and production of IFN-γR β-chain and IFN-γ and that the increased bacterial burden in Il21r-/- mice and reduced expression of ISGs were not due to impairment of IFN-γ or its receptor in these mice, but must be further downstream in the IFN-γ signaling pathway.
If the IL-21/IL-21R axis is not necessary for expression of IFN-γ or its receptor, we asked whether it affected the next step in signal transduction downstream of the IFN-γR, STAT1 phosphorylation. It is known that IL-21 signals mainly via STAT3 but also via STAT1 and STAT5 [21, 30, 31]. We therefore hypothesized that IL-21 acts in concert with IFN-γ to facilitate the activation of STAT1 in CD4+ T cells and that might explain why Il21r-/- mice failed to optimally express multiple ISGs in the distal colon after C. rodentium infection. To address this hypothesis, we stimulated total splenocytes with either recombinant mIL-21 or mIFN-γ alone or in combination and analyzed the activation of STAT proteins by CD4+ T cells. Interestingly, a combination of mIL-21 and mIFN-γ (20 ng/ml of each cytokine) led to significantly enhanced phosphorylation of STAT1 in CD4+ T cells, as compared with CD4+ T cells stimulated with either mIL-21 or mIFN-γ alone (Fig 5A, top row, Fig 5B). As expected, the stimulation of CD4+ T cells isolated from Il21r-/- mice with a combination of mIL-21 and mIFN-γ did not enhance the activation of STAT1, as compared with cells stimulated with mIFN-γ alone (Fig 5A, bottom row, Fig 5B). However, no significant differences were observed between the activation of STAT1 in CD4+ T cells stimulated by mIFN-γ alone in the two genotypes, indicating that the IFN-γ axis is functional in the absence of IL-21/IL-21 R signaling in Il21r-/- mice. However, the treatment of splenocytes with a combination of IFN-γ with IL-17A or IL-22 did not result in enhanced activation of STAT1 in CD4+ T cells isolated from WT or Il21r-/- mice (S10A–S10D Fig). These findings suggest that IL-21, in collaboration with IFN-γ, enhances the expression of ISGs by CD4+ T cells via enhanced phosphorylation of STAT1, but that IL-21/IL-21R signaling axis is not necessary for IFN-γ to induce the phosphorylation of STAT1. In addition, we next investigated other possible complementary mechanisms besides a direct effect of IL-21 on STAT1.
It is known that the IL-21 binding to its cognate receptor, IL-21R, leads to the activation of STAT3 protein and subsequently activates a transcription program that includes some IL-21 target genes [31]. Remarkably, some of the IL-21 target genes (i.e. Gzma, Gzmb, Il10) are known ISGs with non-redundant critical roles in host protection against a wide variety of microbial pathogens, including enteric infections [30]. Based on this, we analyzed the activation of STAT3 following stimulation with a mIL-21 or mIFN-γ alone or in combination. Although phosphorylation of STAT3 was induced in CD4+ T cells by the IL-21 stimulation alone, the combined application of IL-21 and IFN-γ did not further enhance STAT3 phosphorylation (Fig 5C, Fig 5D, top row, and S11 Fig). As expected, CD4+ T cells isolated from Il21r-/- mice failed to induce the STAT3 activation upon treatment with IL-21 alone or in combination with IFN-γ (Fig 5C and Fig 5D, bottom row, Fig 5E). To test whether STAT3 played any role in the collaboration observed between IL-21 and IFN-γ in ISG induction, we generated conditional knockout mice.
The biological functions of IL-21 are mediated mainly via the activation of the STAT3 signaling axis downstream of the IL-21R in a wide variety of hematopoietic cells, although IL-21 is also known to exert some biological effects via the activation of STAT1 and STAT5 [23]. It has been shown that STAT3 is activated in intestinal epithelial cells following C. rodentium infection in vivo and that mice conditionally deficient in STAT3 in epithelial cells (Stat3ΔIEC) were highly susceptible to infection and developed severe colitis after infection with C. rodentium [32]. Our findings suggested that IL-21 was exclusively expressed by mucosal CD4+ T cells and that mucosal CD4+ T cells expressed higher levels of transcripts for IL-21R than other mucosal cells tested, although several immune cell types express this receptor (Fig 2B and 2C). We bred CD4-conditional STAT3-/- mice by crossing CD4-Cre mice with STAT3flox/flox mice (See Materials and Methods). The abrogation of the STAT3 signaling pathway in these mice was confirmed by the lack of STAT3 activation upon IL-6 treatment of CD4+ T cells isolated from CD4stat3-/- mice (Fig 6A). To address the role played by the IL-21/IL-21R signaling axis in CD4+ T cells in host protection after C. rodentium infection, we assessed the bacterial burden, the infection kinetics and survival rates in conditional knockout mice with a CD4+ T cells-specific deletion of STAT3 activity (Fig 6B–6D). The CD4stat3-/- conditional deficient mice showed increased bacterial burden at early time points (days 3 and 7), a higher peak bacterial load at day 9 p.i., and impaired clearance at days 14, 17, 21 and 29. Thus, they were even more impaired in their ability to handle C. rodentium infection than the Il21r-/- mice. That may be because STAT3 is critical not only for IL-21 signaling, but also for IL-17, and induction of TH17 cells, which are another key mediator of C. rodentium clearance. Survival of these mice after C. rodentium infection was also significantly impaired (Fig 6C). These finding are consistent with previous reports demonstrating that the STAT3 activation in TH17 and TH22 CD4+ T cells is important for protection against C. rodentium [33]. This confirms that a deficiency in these cytokines in CD4+ T cells alone is sufficient to seriously impair their ability to handle this bacterial colonic infection. At necropsy, the conditional deletion of STAT3 signaling in CD4+ T cells resulted in watery stool as well as hematomas along the lengths of the distal colons of CD4stat3-/- mice 9 days after C. rodentium infection (Fig 6E). Histological examinations of the distal colons of mice 9 days after infection demonstrated significantly lower crypt hyperplasia scores, a hallmark of pathology during C. rodentium infection, and considerably shorter crypt lengths in of CD4stat3-/- mice (Fig 6F–6H). Collectively, these findings indicated that C. rodentium infection induced moderate pathological changes in CD4stat3-/- mice as compared with littermate STAT3flox/flox control mice, despite significantly higher bacterial burdens in the distal colons of those mice.
It is known that IL-21 exerts some of its downstream effects via the activation of STAT1, STAT3 and STAT5 [21], whereas the induction of IFN-γ-target genes exclusively requires the activation of STAT1 signaling pathway [34]. We already have shown that the activation of STAT3 in CD4+ T cells did not increase when stimulated with a combination of IL-21 and IFN-γ compared with cells treated with IL-21 alone (Fig 5C and 5D). Hypothetically, the increased STAT1 phosphorylation when IL-21 was combined with IFN-γ could have been due to a direct effect of IL-21 on STAT1, or to an indirect effect of some events downstream of IL-21’s principal signaling pathway through STAT3. Thus, we sought to determine whether the optimal expression of ISGs requires intact STAT3 signaling in CD4+ T cells or is occurring in a STAT3-independent manner. To specifically target STAT3 in CD4+ T cells and to avoid the off-target effects of the global STAT3 deletion and its possible indirect effects of these on CD4+ T cells, we stimulated conditional Stat3-/- CD4+ T cells (lacking STAT3 only in CD4+ T cells) or control CD4+ T cells with IFN-γ or IL-21 or a combination of both and determined the expression of ISGs following the single or dual cytokine treatment. In particular, we sought to determine whether the optimal expression of ISGs by CD4+ T cells requires an intact STAT3 signaling or that the enhanced expression of ISGs in CD4+ T cells following treatment with both IFN-γ and IL-21 occurs in a STAT3-independent manner. Further analyses demonstrated that Stat3-/- CD4+ T cells upregulated an ISG profile in response to exogenous IFN-γ or IFN-γ+IL-21 in a fashion similar to CD4+ T cells from STAT3flox/flox controls (Fig 7B, top and middle rows). Accordingly, Stat3-/- CD4+ T upregulated a gene signature profile very similar to the one induced in control animals in response to a combination of rmIFN-γ and IL-21 (Fig 7B, bottom row, and Fig 7C). These findings show that the lack of a functional STAT3 in CD4+ T cells does not impair the expression of ISGs in Stat3-/- CD4+ T cells and that both Stat3-/- CD4+ T cells and cells from STAT3flox/flox littermates upregulated analogous gene profiles in response to IFN-γ and IL-21. Furthermore, the lower panels comparing the cytokine combination with IFN-γ alone show that many ISGs are upregulated more by the combination than by IFN-γ alone in both the intact and STAT3-deficient CD4+ T cells. We conclude that the enhanced induction of most ISGs when IL-21 is added to IFN-γ does not depend on the former’s signaling through STAT3.
To more mechanistically delineate the roles played by IL-21 in enhanced expression of ISGs via the facilitated activation of STAT1, we particularly asked whether the enhancement of STAT1 phosphorylation by combining IL-21 with IFN-γ was a direct effect of IL-21 on STAT1 or an indirect effect dependent on STAT3. To address this question, we measured STAT1 activation in Stat3-/- CD4+ and Stat3+/+ CD4+ T cells in response to IFN-γ or IL-21 alone or in combination. Our findings indicated that Stat3-/- CD4+ T cells and control Stat3+/+ CD4+ T cells significantly and comparably phosphorylated STAT1 more in response to a combination of IFN-γ and IL-21 than to the single-cytokine-treated CD4+ T cells (Fig 7D and 7E). These findings suggest that IL-21-induced enhancement of STAT1 activation was independent of STAT3. Interestingly, the IL-21 treatment of CD4+ T cells isolated from mice deficient in IFN-γ/IFN-γR signaling (Ifngr-/-) induced the activation of STAT1 in these cells, indicating that the endogenous IFN-γ signaling pathway is not required for the STAT1 activation by IL-21 (S12A Fig). Moreover, the expression of LAG-3 in CD4+ T cells isolated from Ifngr-/- mice was not induced by IFN-γ treatment alone, and was not higher with combined IL-21/IFN-γ than with IL-21 alone, indicating the expression of LAG-3 is not enhanced by IL-21 in the absence of an intact IFN-γ/IFN-γR signaling pathway (S12B Fig).
In this study we discovered a previously unknown collaboration between IL-21 and IFN-γ inducing interferon-stimulated genes (ISGs), in the course of studies on the role of the IL-21/IL-21R signaling pathway in resistance to and clearance of infection with a minimally-invasive murine intestinal pathogen C. rodentium. Deficiency in this pathway leads to attenuated inflammation in the colon following infection with this pathogen and impaired clearance of the pathogen. These effects appear to be partially dependent on ISGs, even though IL-21-induced STAT3 activation could play a role in IL-17- or IL-22-mediated protection against C. rodentium. Nanostring analysis identified that the majority of genes substantially impaired (≥ 2-fold) in the whole distal colon of Il21r-/- mice as well as in CD4+ T cells isolated from those animals after infection with C. rodentium were ISGs. We showed that IFN-γ, but not IFN-α/β, mediated resistance to and clearance of C. rodentium. Importantly, we have discovered unexpectedly that IFN-γ collaboratively interacts with IL-21 for the optimal activation of STAT1 and subsequent induction of ISGs. Thus, the collaboration occurs at the level of STAT1 downstream of the IFN-γR, not in the expression of IFN-γ itself or its receptor. We further showed the absolute requirement of the STAT3 signaling pathway in CD4+ T cells for host defense against C. rodentium, by a separate mechanism because the enhanced activation of STAT1 and the subsequent induction of ISGs in CD4+ T cells by the combination of IL-21 and IFN-γ occurred in a STAT3-independent manner. STAT3 is known to be critical for IL-17 function which is important for host defense against extracellular pathogens [33].
The expression of ISGs is tightly regulated by the immune system to avoid excessive and persistent induction causing inflammation and tissue damage [35]. Excessive expression of ISGs has been linked to several inflammatory conditions [34]. Multiple microbial pathogens trigger type-specific interferons, resulting in the transcription of a wide range of downstream gene signatures with distinct or overlapping functions. Some of these genes have roles in the regulation of immunity and inflammation in different immune compartments, including the colon [36–38]. Several ISGs play important roles in host defense against viral, bacterial, and parasitic infections by directly targeting genes conferring protection against infection [39]. Mice deficient in Isg15 (Isg15-/-) are more susceptible to infections with several viral [40, 41] and bacterial infections [42]. As such, studies in vivo showed that IFN-β produced by Legionella-infected macrophages promoted host defense via the upregulation of ISGs and this induction was required for host defense against L. pneumophila [43].
In addition to protective roles, ISGs can mediate inflammatory responses in the colon, predisposing the host to colitis-associated colon cancer [44]. Similarly, IL-21 was highly expressed in the colons of C57BL/6 mice with dextran sulfate sodium (DSS)-induced colitis and Il21r-/- mice manifested milder DSS-induced colitis as compared with their WT counterparts [29]. These findings suggest that the IL-21/IL-21R signaling axis could be part of a positive feedback loop that amplifies an inflammatory response in the gut. Our data support this interpretation, because the Il21r-/- mice had less inflammation and edema in the distal colons after C. rodentium infection despite having a higher bacterial burden (Fig 1C).
A network of cytokines signals through the STAT3 pathway with overlapping or opposing pro- or anti-inflammatory properties, including IL-6, IL-10, IL-17, IL-21 and IL-22. These cytokines activate the STAT3 signaling cascade by phosphorylation and subsequent STAT3 dimerization and translocation into the nucleus [45, 46]. In a mouse model of intestinal inflammation (DSS-induced colitis), for example, IL-22 was excessively secreted and the antibody blockade of IL-22 led to exacerbated inflammation in the colon, whereas the IL-22 overexpression resulted in attenuated inflammation [45]. These results suggest anti-inflammatory roles for IL-22 in the colon following intestinal inflammation. IL-6 is historically considered a pro-inflammatory cytokine and is known to promote inflammation in several models of inflammatory disease [46]. Conversely, the genetic loss of IL-6 or the antibody blockade of IL-6 results in attenuated colitis following intestinal inflammation [47]. While STAT3 plays an important role in protection against C. rodentium infection as we have seen, we found that it is not necessary for the collaboration between IL-21 and IFN-γ.
Minimally-invasive enteric microbial pathogens adhere very closely to the intestinal epithelial surfaces and induce drastic physiological and cytoskeletal changes and structural reorganization in underlying epithelial cells [48]. Several defensive mechanisms have evolved to control enteric microbial infections at the intestinal epithelial barrier [49]. A cooperation between innate lymphoid cell (ILC)-derived IL-22 and IFN-λ was required for the optimal expression of STAT1 in IECs, leading to enhanced expression of ISGs by these cells and subsequent control of rotavirus infection in vivo [6]. It has been suggested that an evolutionary collaboration between two distinct but related cytokine signaling pathways facilitates the control of infection in the intestine. Our finding demonstrated that the receptor for IL-21 was more highly expressed by mucosal CD4+ T cells during intestinal infection than by other immune subsets in the colon LP (i.e. dendritic cells, macrophages, neutrophils, NK cells) and that mucosal CD4+ T cells expressed significant levels of transcripts for the IFN-γ receptor. Concurrent engagement of both receptors on CD4+ T cells was essential for optimal induction of ISGs in those cells. Our findings also indicated that colonic IECs did not express the receptor for IL-21 in response to intestinal microbial infection, whereas IFN-γ receptor was expressed by these cells. The lack of the expression of the IL-21 receptor by IECs excludes these cells as targets for the collaborative effects of IL-21 and subsequent modulation of ISG expression during C. rodentium infection as shown in other models of intestinal infection, in which IECs were the main targets of IL-22 and IFN-λ cooperation [6].
The indispensable role of CD4+ T cells in protection against intestinal infection with C. rodentium has been established in several studies [16, 17]. Mice deficient in CD4+ T cells (but not CD8+ T cells) are extremely susceptible to intestinal infection with C. rodentium as well as to the systemic spread of the bacterium to extra-intestinal sites, including the mesenteric lymph nodes (MLNs), spleen and the liver [15]. Intestinal infection with this pathogen elicits a TH1-biased immune response, characterized by the induction of IFN-γ and TNF-α (17). CD4+ T cells play a central role in immune response to intestinal infection with C. rodentium via the production of cytokines required for host resistance to this pathogen, including IL-17, IL-22 and IFN-γ, and they have been shown to be a main source of antigen-specific induction of IFN-γ during intestinal infection with this bacterium [17]. Our observations show dual roles for mucosal CD4+ T cells following infection with C. rodentium in the colon: First, IL-21 was exclusively expressed by mucosal CD4+ T cells of the colon LP and second, mucosal CD4+ T cells express significant levels of transcripts for IL-21R and IFN-γR and are targets for the collaborative activity of IFN-γ and IL-21. These finding identify a previously unidentified collaboration between two distinct signaling pathways in the optimal expression of ISGs in CD4+ T cells, and protection against C. rodentium, and may further explain the absolute requirement of CD4+ T cells in the regulation of mucosal immune response to this pathogen in the colon.
The type I IFNs have emerged as key players in host defense against both extracellular and intracellular bacterial pathogens in recent years [7, 27]. We observed that both type I- and type-II specific ISGs were induced in the distal colon following infection with C. rodentium. Further analysis demonstrated that mice deficient in IFN-α/β signaling (Ifnar-/-) managed to clear infection with the bacterium in kinetics similar to WT controls. However, mice lacking functional IFN-γ (Ifng-/-) failed to clear the infection, indicating that type II-specific, but not type I-specific, ISGs were required for host defense in the colon.
Other effects of IL-21 receptor deficiency on intestinal host defense have been reported very recently involving defective IgA responses to atypical commensal bacteria, such as segmented filamentous bacteria (SFB) and Helicobacter species, indirectly affecting C. rodentium-induced immunopathology [50]. However, in this situation, the effect was mostly on inflammation and the effect on bacterial burden was minor. This is clearly distinct from our model in which it is clear that IFN-γ plays a critical role and the role of IL-21 is mainly to amplify the IFN-γ signal in CD4+ T cells. Furthermore, we have cohoused mice for at least 2 weeks in all experiments to equalize intestinal microbiota, as mice, which are coprophagic, rapidly equilibrate their microbiomes when cohoused for 2 weeks [51–54], and have replicated the same results with Il21r-/- mice vs. heterozygous controls that were bred from the same group of Il21r-/- mothers, to further ensure that the microbiota were equivalent (S1 Fig).
The collaborative link we established here between IFN-γ and IL-21 signaling pathways provides mechanistic elucidation to the enduring conundrum as to why the lack of an intact IL-21/IL-21R signaling axis renders host susceptible to a wide range of pathogens [55, 56]. It also provides insight into why the lack of an intact IL-21/IL-21R pathway leads to attenuated inflammation in the colon following insults during infectious and non-infectious insults [29, 44]. Considering that IECs, as the first line of defense against minimally-invasive intestinal pathogens including C. rodentium, do not express the receptor for IL-21, our findings suggest the collaborative effect we describe between IFN-γ and IL-21/IL-21R signaling axes acts deep in the lamina propria of the colon as a second defensive layer against mucosal pathogens. Understanding the interactions of these cytokine networks and their signaling pathways should allow development of novel therapeutic targets for colonic infections, inflammatory bowel disease, and promotion of mucosal vaccine efficacy.
Six- to eight-week old sex- and age-matched female mice were used in all experiments. The mice were bred in-house or purchased from the Jackson Laboratory (Bar Harbor, ME, USA). In order to exclude the effects of differing microbiome compositions between mouse genotypes on the experimental designs, the knockout and WT mice (both on a C57BL/6 background) were cohoused for 2 weeks at a 1:1 ratio before all experiments as described before [51–54]. In addition, to further exclude microbiome differences, we bred WT or Il21r-/- males to the same group of Il21r-/- females to produce matched pups that were either heterozygotes (WT phenotype) or homozygous Il21r-/- and that were then foster nursed together, so the microbiome obtained from their mothers and nursing mothers would be identical (S1 Fig). C57BL/6 were purchased from Charles River Laboratories (Wilmington, MA, USA). B6N.129-Il21rtm1Kopf/J (019115) were purchased from the Jackson Laboratory. Ifnar-/- and Ifngr-/- mice on a C57BL/6 background were kindly provided by Dr. Howard Young (NCI/NIH).
We generated mice in which STAT3 protein is conditionally deleted in CD4+ T cells (CD4stat3-/-) by breeding CD4-Cre mice (Tg(Cd4-cre)1Cwi/BfluJ; 017336) to STAT3flox/flox (B6.129S1-Stat3tm1Xyfu/J mice; 016923 (both strains from the Jackson Laboratory) as described earlier [57]. Mice were genotyped using DNA isolated from tail snips. STAT3flox/flox littermates were used as wild-type (WT) controls.
All experiments were carried out in accordance with guidelines and protocols approved by the National Cancer Institute Animal Care and Use Committee in compliance with the National Institutes of Health Guidelines (VB-014).
Spleens were aseptically removed from naïve WT, Il21r-/-, STAT3flox/flox or CD4stat3-/- mice and cells were mechanically disrupted through a 100 μm cell strainer (BD Biosciences, San Jose, CA) using the plunger of a 6 ml syringe. RBCs were lysed in ACK lysing buffer (Lonza, Walkersville, MD), and the remaining cells were washed twice with ice-cold PBS. A total of 5×106 splenocytes were cultured in duplicate in 1 ml of RPMI 1640 (Gibco) supplemented with 10% FBS and 100 μg/ml penicillin/streptomycin (all from Gibco). Cells were allowed to rest for 2 additional hours at 37°C and subsequently were stimulated in the presence of recombinant murine IFN-γ, IL-17A, IL-21 and IL-22 (20 ng/ml; Peprotech, Rocky Hill, New Jersey, USA) alone or in combination (IFN-γ+IL-17A; IFN-γ+IL-21; IFN-γ+IL-22; 20 ng/ml each). Cells receiving no cytokine treatments were used as controls. Cells were harvested at different intervals following the addition of cytokines and were stained with indicated antibodies.
C. rodentium strain DBS100 (ATCC 51459) was propagated in Luria-Bertani (LB) broth at 37°C, harvested by centrifugation, and resuspended in PBS at a concentration of 5×109 colony forming units (CFU)/mL. In some experiments, an ovalbumin-expressing C. rodentium (OVA-Citrobacter) under a kanamycin-resistance gene was used for infection. Mice infected with OVA-Citrobacter were given kanamycin (1g/L) in drinking water ad libitum, starting 4 days before infection and during the entire course of infection to prevent loss of OVA expression. Mice were infected with 100 μl of the bacterial suspension containing 5×108 CFU of C. rodentium/mouse by oral gavage as described previously [58, 59]. For bacterial quantification, fecal pellets (50–100 mg) were weighed, homogenized in 2 mL of sterile PBS, serially diluted, and plated onto MacConkey agar as described before [58, 59]. The detection limit of the culture method was 103 CFU/g feces.
The colons were cleaned, rolled into a Swiss roll configuration, fixed in 10% buffered formalin overnight, followed by fixation in 70% ethanol and subsequently embedded in paraffin. Tissue sections (5-μm thick) were stained with hematoxylin and eosin (H&E) and digitized with Aperio ScanScope (Aperio, Vista CA) and were analyzed using Aperio ImageScope software. The severity of colitis was assessed by an unbiased (blinded) observer using a scoring system developed previously [60].
An ex vivo organ culture system was used to determine the kinetics of cytokine production in the colon of mice after C. rodentium infection as described before [26]. The distal colons from infected mice at different time-points or uninfected controls were removed, washed briefly and opened longitudinally and were cultured in RPMI1640 culture medium (Gibco) supplemented with 10% FBS, and 100 μg/ml penicillin/streptomycin (Gibco). Culture supernatants were collected at different intervals post-culture and were examined for cytokine concentrations by ELISA. Mouse ELISA kits for IFN-α, IFN-γ, IL-17A, IL-21 and IL-22 (all from eBioscience) were used. Results were expressed as picograms per g tissue (pg/g).
The colonic LPLs were isolated as described elsewhere [59]. Briefly, the distal colons were removed, cut longitudinally, and washed with ice-cold PBS. The colons then were cut into small pieces and incubated for 20 min in pre-digestion solution (PBS, pH7.4, containing 30 mM EDTA and 1 mM dithiothreitol) at 37°C with agitation. After incubation, the samples were vortexed for 20 seconds and the supernatants were discarded. Subsequently, the tissues were washed twice with ice-cold PBS, minced, and digested in RPMI1640 (Gibco), containing 420 μg/ml Liberase TL (Roche, Indianapolis, IN, USA), and 0.1 mg/ml DNase (Roche), for 45 min at 37°C. Digesting tissues were further mechanically dissociated by vortexing vigorously every 10 min. Digested tissues were vortexed for an additional 20 seconds and passed through a 70-μm cell strainer (DB Falcon, San Jose, CA, USA). Isolated cells were washed twice with ice-cold PBS, counted, and stained with the indicated antibodies.
Isolated LPLs were incubated with 1 μg/106 cells anti-mouse CD16/CD32 (clone 93; Biolegend, San Diego, CA, USA) in FACS buffer (PBS supplemented with 3% FBS) for 20 min on ice to block Fc receptor binding, followed by live/dead cell labeling using a LIVE/DEAD Fixable Aqua Dead Cell Stain Kit (Life Technologies, Eugene, OR, USA) for 20 min at 4°C in the dark per the manufacturer’s instructions. For surface marker staining, cells were stained in duplicate with a cocktail of the following conjugated antibodies in FACS buffer: anti-CD3-BV421 (clone 17A2), anti-CD4-Alexa Fluor 488 (clone GK1.5), anti-CD8-Alexa Fluor 700 (clone 53–6.7), anti-NK1.1-APC (clone PK136), anti-CD45 (clone 30-F11), anti-CD11b-APC (clone M1/70), anti-MHC class II (I-A/I-E)-Pacific Blue (clone M5/114.15.2), anti-F4/80-Brilliant Violet 421 (clone BM8), anti-LAG-3-PerCP/Cy5.5 (clone C9B7W), anti-EpCAM-PE-Cy7 (clone G8.8), anti-IFN-γR-β-PE (clone MOB-47; all from Biolegend), anti-CD11c-Texas Red (clone MCD11C17; Invitrogen) and LAG-3-PE (clone C9B7W; eBioscience). For measuring the surface expression of IL-21R, LPLs were stained using a biotinylated anti-IL-21R Ab (eBio4A9; eBioscience), followed by staining with PE-streptavidin (Biolegend). The immune cell number in the colon was quantified by flow cytometry using CountBright absolute counting beads according to the manufacturer’s instructions (Molecular Probes, Invitrogen). For intracellular staining, the LPLs isolated from the colon were stained without stimulation or were stimulated ex vivo for 8 h in the presence of PMA (50 ng/ml) and ionomycin (350 ng/ml) or chicken egg ovalbumin (OVA) (100 μg/ml; Sigma), adding brefeldin A at 10 μg/ml (all from Sigma-Aldrich) for the last 6 h. The cells then were fixed, permeabilized, and stained with anti-IFN-γ-PE (clone XMG1.2; Biolegend). Data were acquired using an LSRII flow cytometer (BD Biosciences) and were analyzed using FlowJo software (Tree Star, Inc., San Carlos, CA, USA).
We applied phospho-flow cytometry analysis to investigate the phosphorylation of STAT1 or STAT3 proteins following the treatment of naïve splenocytes with a combination of recombinant murine IFN-γ, IL-17A, IL-21, IL-22 or IL-6. Following the treatment of cells (5×106/well) with IFN-γ, IL-17A, IL-21 or IL-22 (20 ng/ml each) alone or in combination (IFN-γ+IL-17A; IFN-γ+IL-21; IFN-γ+IL-22; 20 ng/ml each), the cells were fixed and permeabilized followed by staining with either anti-phospho-STAT1 (pY701; clone 4a; BD Biosciences), anti-STAT3 (pY705; clone 13A3-1; Biolegend) or mouse IgG1κ isotype control (clone P3.6.2.8.1; eBioscience). Data were acquired using an LSRII flow cytometer as described earlier.
LPLs were isolated as described above from distal colons of naïve mice or from mice infected with C. rodentium 9 days p.i. The colonic LPL were FACS-sorted into CD4+ T cells (EpCAM-CD45+CD3+CD4+) by using a FACSAria cell sorter (Becton Dickinson). Splenic CD4+ T cells were purified (≥95% purity) using negative selection with a mouse CD4+ T cell isolation kit (Miltenyi Biotec, Auburn, CA) and total RNA was isolated from CD4+ T cells using a Qiagen RNeasy Plus Micro Kit (Qiagen, Hilden, Germany). In some experiments total RNA from whole distal colons was isolated using a Qiagen RNeasy Plus Mini Kit (Qiagen, Hilden, Germany) and used as the source of RNA for further analysis. Total RNA (100 ng) was used as samples for probe-based NanoString system (nCounter XT Code Set; Seattle, WA, USA). The raw data for each gene was compile and normalized against the spike-in positive (6 genes) and negative (8 genes) internal reference genes and were expressed as normalized counts by using the nSolver analysis software version 3.0 (NanoString Technologies). The gene expression data were further normalized to the geometric mean of the expression of internal reference genes and presented as normalized counts/gene/biological sample.
The status of impaired genes in the whole distal colon or CD4+ T cells isolated from the distal colon was tallied against the Interferome database (http://interferome.org) to establish whether the expression of a given gene is affected by type I, type II or type III interferons in the previously published databases [61]. The gene classification based on biological processes and gene ontology (GO) were performed using the Database for Annotation, Visualization and Integrated Discovery (DAVID, version 6.8; http://david.abcc.ncifcrf.gov) Bioinformatics Resources [62, 63]. The normalized Nanostring values were used to generate heatmaps by using the web-based open software Morpheus (http://software.broadinstitute.org/morpheus).
Data were analyzed using GraphPad Prism, version 7.03, software (GraphPad, San Diego, CA, USA) and expressed as the Mean ± SEM. For statistical analyses, a 2-tailed Mann-Whitney U test or a one-way ANOVA followed by Bonferroni post-hoc adjustment test for multiple comparison were employed. *p < 0.05 was considered statistically significant.
|
10.1371/journal.pgen.1000263 | Aging Predisposes Oocytes to Meiotic Nondisjunction When the Cohesin Subunit SMC1 Is Reduced | In humans, meiotic chromosome segregation errors increase dramatically as women age, but the molecular defects responsible are largely unknown. Cohesion along the arms of meiotic sister chromatids provides an evolutionarily conserved mechanism to keep recombinant chromosomes associated until anaphase I. One attractive hypothesis to explain age-dependent nondisjunction (NDJ) is that loss of cohesion over time causes recombinant homologues to dissociate prematurely and segregate randomly during the first meiotic division. Using Drosophila as a model system, we have tested this hypothesis and observe a significant increase in meiosis I NDJ in experimentally aged Drosophila oocytes when the cohesin protein SMC1 is reduced. Our finding that missegregation of recombinant homologues increases with age supports the model that chiasmata are destabilized by gradual loss of cohesion over time. Moreover, the stage at which Drosophila oocytes are most vulnerable to age-related defects is analogous to that at which human oocytes remain arrested for decades. Our data provide the first demonstration in any organism that, when meiotic cohesion begins intact, the aging process can weaken it sufficiently and cause missegregation of recombinant chromosomes. One major advantage of these studies is that we have reduced but not eliminated the SMC1 subunit. Therefore, we have been able to investigate how aging affects normal meiotic cohesion. Our findings that recombinant chromosomes are at highest risk for loss of chiasmata during diplotene argue that human oocytes are most vulnerable to age-induced loss of meiotic cohesion at the stage at which they remain arrested for several years.
| In humans, chromosome segregation errors during meiosis are the leading cause of birth defects and miscarriages. Moreover, as women age, these errors increase dramatically. For accurate segregation during the first meiotic division, homologous chromosomes must remain physically associated until anaphase I. Normally, attachments along the arms of sister chromatids keep the recombinant homologues together. Human oocytes complete meiotic recombination during fetal development and arrest until ovulation. Therefore, accurate segregation of homologous chromosomes during the first meiotic division requires that recombinant chromosomes remain associated for decades. One hypothesis to explain why segregation errors increase as women age is that the connections between sister chromatids deteriorate over time and allow recombinant homologues to dissociate prematurely. Here, we address this hypothesis using Drosophila as a model system. We find that when Drosophila oocytes undergo experimentally induced aging, recombinant homologues missegregate during meiosis I. Furthermore, the meiotic stage at which Drosophila oocytes are most vulnerable to age-induced errors is analogous to the stage at which human oocytes remain arrested for decades. Together, our data argue that aging does cause premature loss of the connections between meiotic chromosomes and that this is a major determinant of segregation errors in both Drosophila and human oocytes.
| In humans, meiotic chromosome segregation errors that give rise to aneuploid gametes are the leading cause of fetal loss and birth defects [1]. Approximately 30% of miscarriages result from aneuploidy and at least 5% of all clinically recognized pregnancies and 0.3% of live-borns are aneuploid [1]. Female meiosis in humans is especially error-prone and the majority of segregation errors in oocytes originate during meiosis I [1],[2].
The link between increased maternal age and meiotic segregation defects in humans is well established. At maternal age 25, the risk of a trisomic pregnancy is approximately 2% but increases to approximately 35% for a woman at age 42 [3]. Despite its clinical importance, the specific mechanisms that give rise to age-dependent meiotic nondisjunction (NDJ) are not understood. The prevailing theory is that segregation errors in older oocytes arise in large part because of the protracted prophase I arrest at which human oocytes remain suspended for decades [1],[4].
Accurate segregation during meiosis I requires that homologous chromosomes undergo recombination and remain physically attached to one another until they segregate to opposite poles during anaphase I. In the absence of a stable connection, homologues will segregate randomly resulting in meiosis I NDJ. Cohesion between the arms of sister chromatids provides an evolutionarily conserved mechanism for maintaining the association of recombinant homologues (Figure 1). Normally, the release of arm cohesion at anaphase I allows recombinant homologues to segregrate to opposite poles [5],[6]. In the absence of cohesion, chiasmata are not maintained and homologous chromosomes missegregate during meiosis I [5],[7],[8].
Given that human oocytes undergo meiotic recombination during fetal development and remain suspended in a prolonged dictyate (diplotene) arrest until ovulation, the continuous association of homologous chromosomes demands that meiotic sister-chromatid cohesion remain intact for decades. One attractive hypothesis to explain age-dependent NDJ is that deterioration of cohesion with age causes recombinant homologues to dissociate prematurely and segregate randomly during the first meiotic division (Figure 1). However, testing this hypothesis in humans presents several insurmountable challenges.
Using Drosophila as a model system, we have developed an experimental regimen to study the mechanisms that contribute to increased levels of meiotic NDJ in oocytes as a result of aging [9]. The Drosophila ovary is composed of several ovarioles, each of which contains a linear array of oocytes at progressive stages of development (Figure 2A). Throughout the lifetime of the female, germline stem cells at the tip of each ovariole continuously generate a steady stream of newly formed oocytes that enter meiotic prophase and grow and develop as they move posteriorly; as mature oocytes pass through the oviduct they complete meiosis and are fertilized. Under normal conditions (continuous egg laying), Drosophila oogenesis is an uninterrupted process with only a brief arrest at metaphase I before ovulation. However, when egg laying is suppressed, the majority of Drosophila oocytes within each ovariole are halted in developmental progression and “age” within the abdomen of the female (Figure 2B). Such experimentally induced aging of Drosophila oocytes can be used to mimic the normal aging process that human oocytes undergo within the ovary during a female's lifespan [9]. We have used this aging regimen to test the hypothesis that meiotic cohesion deteriorates as the oocyte ages and increases the frequency at which recombinant homologues missegregate. Although maternal age does not dictate the age of the oocyte in fruit flies as it does in humans, for simplicity we will refer to increased NDJ in experimentally aged Drosophila oocytes as “age-dependent NDJ.”
The cohesin complex consists of four subunits and is required for sister-chromatid cohesion during mitosis and meiosis [6],[10]. Although wild-type Drosophila oocytes subjected to our standard four-day aging regimen do not exhibit increased levels of NDJ [9], we reasoned that this time period may be insufficient for cohesion to deteriorate enough to detect a loss in chiasma maintenance and that reduction of functional cohesin might render oocytes more vulnerable to aging effects. Unlike other eukaryotes, meiosis-specific cohesin subunits have not been uncovered in the Drosophila genome [11]. Therefore, to weaken but not eliminate meiotic cohesion, we reduced the dosage of the cohesin subunit SMC1 by utilizing flies heterozygous for a deletion that removes the smc1 gene [12]. Immunoblot analysis confirmed that the level of SMC1 protein is reduced approximately two-fold in ovaries from smc1+/− females (Figure 3A).
In order to assay loss of chiasma maintenance with age, we needed to verify that chiasma formation was not severely disrupted in smc1+/− oocytes. Cohesion between sister chromatids is required for normal levels of crossovers during meiosis [13]–[17]. Therefore, we measured the frequency and distribution of exchange in smc1+/− heterozygous females (Figure 3B). Although reduction of SMC1 protein has a semi-dominant effect on the number and distribution of meiotic crossovers during female meiosis, high levels of homologous exchange were still observed (76% of the wild-type control, Figure 3B). The tetrad exchange rank for Drosophila oocyte bivalents can be estimated from analysis of recombinant and non-recombinant meiotic products [18]. Normally, the X chromosome is achiasmate (non-recombinant) in 6–12% of Drosophila oocytes [19]. Our recombination data indicates that although crossovers were reduced in smc1+/− oocytes, the majority of tetrads (74%) had at least one exchange event (Figure 3B). In addition, viability and fertility of smc1+/− females is normal and they do not exhibit meiotic segregation defects in the absence of an aging regimen (Table S1). Therefore, smc1+/− oocytes provide an excellent sensitized system that can be used to assay whether cohesion and chiasma maintenance decline with age.
To examine the effect of SMC1 dosage on age-dependent NDJ, we subjected smc1+/− females to the aging regimen and assayed for chromosome missegregation (see Figure 2B, 2C). However, meiotic NDJ did not increase when smc1+/− oocytes were aged (Table S2). Because Drosophila females harbor a robust mechanism that directs the segregation of achiasmate chromosomes [20], it is possible that this achiasmate mechanism also ensures accurate disjunction of recombinant chromosomes that fail to maintain chiasmata (Figure 4). Therefore, we reasoned that in order to detect missegregation of recombinant chromosomes in Drosophila oocytes that have lost chiasmata, the achiasmate system must be compromised (Figure 4). Hawley and colleagues have shown that P-element disruption of one copy of matrimony (mtrm[KG08051]/+) disrupts the segregation of achiasmate chromosomes in Drosophila oocytes [21]. However, meiotic cohesion, synaptonemal complex assembly and crossover frequency appear to be unaffected in mtrm+/− oocytes [21],[22]. Therefore, to monitor whether exchange bivalents lose chiasmata with age, we compromised the achiasmate pathway by reducing the dosage of mtrm in smc1+/− females.
We observed increased segregation errors in smc1+/− oocytes following the four-day aging regimen when the achiasmate pathway was also compromised (mtrm+/−) (Figure 5A). Significant levels of age-dependent nondisjunction were observed for broods 1 and 2, which represent oocytes that are developmentally most mature during the aging regimen (see Figure 2A). We have performed multiple experiments to compare NDJ in smc1+/− mtrm+/− aged and non-aged oocytes and have also assayed missegregation of different X chromosomes (data not shown). Although the absolute level of NDJ may vary between experiments, we have repeatedly observed higher levels of NDJ in smc1+/− mtrm+/− aged oocytes in the first two 24 hour broods (see Tables S3A and S3B for examples of raw data from two independent experiments). In contrast, mtrm+/− oocytes with normal levels of SMC1 do not exhibit age-dependent NDJ in the first brood (Table S4); we did detect a significant increase in brood 2, but the absolute level of NDJ was relatively low (Table S4). The finding that increased meiotic chromosome missegregation in aged Drosophila oocytes occurs when the dosage of a cohesin subunit is reduced is consistent with deterioration of cohesion and loss of chiasma maintenance as an underlying cause for age-dependent NDJ.
To ask whether deterioration of chiasmata contributed to the increase in age-dependent nondisjuntion in smc1+/− mtrm+/− oocytes, we assayed the recombinational history of the missegregating chromosomes. For this analysis, we collected “Diplo-X” female progeny that received two maternally contributed X chromosomes due to meiotic missegregation (see Figure 2C) and performed an additional cross to genotype the two X chromosomes (Figure 6). Using this strategy, we determined whether missegregating chromosomes had undergone one or more crossovers (Table S5). In addition, a centromere-proximal marker (car) allowed us to assess whether NDJ events involved homologous chromosomes (MI NDJ) or sister chromatids (MII NDJ). The genotype of the Diplo-X females suggested that the majority of NDJ events (70/72) occurred during meiosis I (Table S5). Approximately one-third of all Diplo-X females arose from nondisjunction of exchange tetrads (R) in both aged and non-aged oocytes (Figure 5B). Notably, recombinant bivalents missegregated 1.8 times more frequently in the aged oocytes than in non-aged oocytes (compare R/N for aged and non-aged, Figure 5B). In addition, the number of recombinant bivalents that missegregate is under-represented in our assay because it is possible for a Diplo-X female to inherit two non-recombinant chromatids from a recombinant tetrad (see Figure 6). These data argue that upon reduction of SMC1, recombinant bivalents become more vulnerable to missegregation when oocytes age.
Although the achiasmate pathway is compromised in mtrm+/− mutants, our data indicate that it is not completely disengaged. An estimated 26% of the tetrads are achiasmate in smc1+/− oocytes (Figure 3B) but we only observe 4–9% NDJ in non-aged oocytes when the achiasmate pathway is compromised (smc1+/− mtrm+/−, Table S3). Moreover, segregation of an X chromosome (In(1)dl-49) harboring an inversion that severely suppresses crossovers is not completely random in mtrm+/− oocytes (Table S1). The data in Figure 5B indicate that when the level of SMC1 protein is reduced, aging further increases the frequency at which achiasmate chromosomes missegregate in mtrm+/− oocytes (compare NR/N for aged and non-aged). These data implicate the cohesin complex in accurate disjunction of achiasmate as well as chiasmate bivalents. However, the low number of MII NDJ events that we observe (as determined by the centromere proximal car marker) argues that the increased missegregation of non-recombinant chromosomes is not due to loss of centromeric cohesion between sister chromatids.
Our analysis of age-dependent NDJ in smc1+/− mtrm+/− oocytes indicates that broods 1 and 2 are most susceptible to age effects. Each Drosophila ovariole consists of multiple oocytes at progressive stages of development. At the posterior end of the ovariole (stages 13/14), the mature oocytes undergo nuclear envelope breakdown and spindle assembly. Stage 14 oocytes remain arrested at metaphase I until passage through the oviduct triggers resumption of the meiotic divisions. After an aging regimen, the most mature oocytes are laid first and therefore will contribute to the progeny in brood 1 (see Figure 2A).
In mature oocytes, the chromosomes have already made stable connections with the meiotic spindle microtubules and premature loss of chiasmata during the aging regimen should not affect their segregation. However, if the mature oocytes develop gross defects in the spindle machinery as they age, recombinant chromosomes could missegregate by a mechanism unrelated to chiasma maintenance. Conversely, oocytes that undergo aging at a stage prior to meiotic spindle assembly would depend on chiasma maintenance for proper segregation. Therefore, it was important to determine what fraction of broods 1 and 2 correspond to metaphase I arrested oocytes.
To compare the relative distribution of oocytes at different developmental stages in ovaries from smc1+/− mtrm+/− control females and those subjected to the aging regimen, we examined ovaries at 8-hour intervals after completion of the four-day regimen. One striking difference was the increased number of mature oocytes (stages 13/14) following aging (Figure 7 and Figure S1). Previous reports by other investigators also have documented that mature oocytes accumulate when egg laying is suppressed [23],[24]. Quantitative analysis indicated that our aging regimen resulted in a ∼2.5-fold increase in mature oocytes compared to the ovarioles of control females (Figure 7). However, after 16 hours of egg laying, this excess of mature oocytes was no longer observed. These results argue that in our NDJ tests, at least some fraction of progeny in brood 1 (0–24 hours) arise from oocytes that have not yet assembled meiotic spindles.
Evaluation of the distribution of different developmental stages in fixed ovaries also indicated that following the aging regimen, the increase in mature oocytes occurred primarily at the expense of stage 9–12 oocytes (Figure 7) while oocytes at earlier stages (stages 1–8) remained abundant. Significantly, we found that 16 hours after the aging regimen ceased and egg laying commenced, the distribution of late stages in the aged group resembled that in control ovaries, although the total number of oocytes between stages 8 and 14 was slightly less in the aged ovaries. Consistent with our findings, King [23] reported that stages 9–12 are significantly under-represented in ovaries of 4 day old virgins. Together, these data indicate that oocytes at stages1–8 halt in development and age during the four-day aging regimen, while oocytes at later stages (stages 9–13) continue to mature and do not “age” until they arrest at metaphase I (stage 14).
Given the altered distribution of oocytes at different developmental stages following the aging regimen, we carried out NDJ tests to determine which stages during oocyte development are the most susceptible to age-dependent segregation errors. We subjected smc1+/− mtrm+/− females to our standard four-day aging regimen and set up crosses to assay meiotic NDJ (see Figure 2). However, this time, we transferred the parents to new vials every 8 hours (instead of every 24 hours) to generate 8-hour “sub-broods” of progeny. This allowed us to measure age-related segregation errors in “sub-broods” of progeny that represented snapshots of oocytes at different meiotic stages (pachytene, diplotene-like and metaphase I).
Following the aging regimen, age-dependent NDJ was observed in sub-broods 3 and 4, which arose from oocytes that were fertilized 16–32 hours after the aging regimen was completed (Figure 8A, Table S6). Our cytological analysis (Figure 7 and Figure S1) and the published time-frame for oogenesis progression [19],[25], indicate that these sub-broods correspond to oocytes at stages 7 and 8 which have already disassembled their synaptonemal complex before initiation of the aging regimen [26],[27]. In contrast, a significant increase in meiotic segregation errors was not observed during the first 16 hours in aged smc1+/− mtrm+/− oocytes (Figure 8A, Table S6). These results argue that metaphase I arrested smc1+/− mtrm+/− oocytes (stage 14) are not vulnerable to age-related defects. Similarly, oocytes that undergo aging during pachytene (sub-broods 5 and 6) also appear to be refractory to age effects. Therefore, only oocytes that age during a very specific window of oogenesis (post-pachytene/pre-metaphase I) exhibit age dependent NDJ.
To determine what fraction of missegregating chromosomes were recombinant in the aged versus non-aged oocytes, we genotyped the Diplo-X progeny for a subset of sub-broods obtained from the NDJ analysis shown in Figure 8A. In addition to sub-broods 3 and 4 which exhibited age-dependent nondisjunction, we also analyzed the recombinational history of missegregating-X chromosomes in sub-broods 1 and 6, which represent opposite ends of the developmental spectrum examined in this study (Tables S7, S8). Again, we found that both recombinant and non-recombinant chromosomes missegregated in aged and non-aged oocytes (Tables S7, S8). In addition, a striking trend emerged for the aged oocytes when we calculated the frequency at which recombinant (R) chromosomes missegregated in each sub-brood (R/N; Figure 8B). NDJ of recombinant chromosomes in sub-brood 3 aged oocytes was significantly higher than that for aged oocytes in sub-brood 1 (P = 0.0029) or sub-brood 6 (P = 0.0179). In addition, this trend holds true when the NDJ of recombinant chromosomes in sub-brood 4 aged oocytes is compared to that for sub-broods 1 or 6 aged oocytes, although the difference is not statistically significant. These data contrast strongly with those for non-aged oocytes, in which recombinant chromosomes missegregate at similar frequencies in the four sub-broods analyzed.
The data shown in Figure 8 provide compelling evidence that only oocytes at a specific meiotic stage are vulnerable to age-dependent segregation errors even when all oocytes within the ovary are subjected to the aging regimen. The increased nondisjunction observed in specific sub-broods arises in part because segregation of recombinant bivalents becomes more error-prone (Figure 8B). When one takes into account the well-documented time-line for oocyte development while evaluating the sub-brood NDJ results, a striking pattern emerges (Figure 8C). When pachytene oocytes undergo the aging regimen, recombinant chromosomes do not exhibit an age-dependent increase in nondisjunction (sub-broods 5 and 6). In contrast, sub-broods 3 and 4, which display the greatest age-related defects, are composed primarily of oocytes at stages 7 and 8 of oocyte development. Both electron microscopy and immunofluorescence analyses have verified that disassembly of the synaptonemal complex in Drosophila oocytes initiates prior to these stages [26],[27]. Although Drosophila chromosomes do not exhibit typical diplotene/diakinesis morphology in prophase I oocytes, stages 7 and 8 of oogenesis correspond to these classical meiotic stages [19]. Transit through each of these stages takes approximately 5–8 hours during normal oogenesis, but when their developmental progression is halted by our aging regimen, meiotic chromosomes remain suspended in this state up to 19 times longer than usual. Of special note is the finding that recombinant chromosomes are most susceptible to segregation errors if Drosophila oocytes undergo aging during diplotene (Figure 8) because this is the meiotic stage at which human oocytes remain arrested for decades.
Our sub-brood analysis also suggests that non-recombinant chromosomes in smc1+/− mtrm+/− oocytes missegregate more frequently when aging is induced in oocytes prior to metaphase I (Figure S2), although the difference between sub-brood 1 and the other sub-broods is not statistically significant (0.10<P<0.27). However, non-recombinant chromosomes in sub-brood 3 are significantly more vulnerable to missegregation after aging (P = 0.029). These results support the model that the connections that keep achiasmate bivalents together depend at least in part on meiotic cohesion proteins [9].
Together, our data support the hypothesis that weakening of meiotic cohesion with age is a significant determinant of age-dependent nondisjunction in both fly and human oocytes. One major advantage of these studies is that we have lowered but not eliminated the level of the cohesin subunit SMC1. Although smc1+/− oocytes are sensitized to the effects of aging, meiotic cohesion is initially intact. Therefore our regimen allows us to observe the effect of aging upon the normal cohesin complex that functions during Drosophila meiosis. In this respect our analysis differs markedly from previous studies using knock-out mice in which the meiosis-specific subunit SMC1-β was eliminated [8].
Here we demonstrate that deterioration of normal meiotic cohesion during the aging process causes loss of chiasma maintenance. Loss of arm cohesion in aged oocytes accounts for the increased frequency with which recombinant chromosomes missegregate during meiosis I. In contrast, low levels of MII NDJ indicate that centromeric cohesion remains intact, at least for the duration of our aging regimen. Recombinant chromosomes in Drosophila oocytes that undergo aging after they have disassembled their synaptonemal complex (diplotene-like) are most vulnerable to age effects and these oocytes bear a striking resemblance to the stage at which human oocytes remain arrested for decades. Interestingly, Drosophila oocytes that are subjected to aging during pachytene do not exhibit significant age-dependent nondisjunction. Using epifluorescence microscopy to compare smc1+/− mtrm+/− aged and non-aged oocytes, we have not observed any obvious differences in SC morphology, or the timing of its assembly or disassembly (VVS and SEB, unpublished results). Therefore, we speculate that the synaptonemal complex may play a role in protecting sister-chromatid cohesion from age-induced deterioration. The differential response of pachytene and diplotene-like Drosophila oocytes to the aging process has profound implications regarding the vulnerability of human oocytes to age-induced loss of meiotic cohesion. Our data argue that the meiotic stage (dictyate) at which human oocytes remain suspended as women age is also the stage during which recombinant chromosomes are at highest risk for loss of chiasmata.
Because the achiasmate pathway in Drosophila oocytes will ensure the accurate segregation of recombinant chromosomes that lose chiasmata due to deterioration of arm cohesion during the aging process, we needed to dismantle this system for our analysis. Therefore, we used a P-element insertion in matrimony (mtrmKG08051) as a genetic tool to disrupt the achiasmate pathway. However, we do not think that heterozygosity for mtrm contributes to age-dependent loss of arm cohesion. Recent work has demonstrated that Mtrm protein physically interacts with and inhibits the activity of Polo kinase in Drosophila oocytes from stage 11 until nuclear envelope breakdown (NEB) at stage 13 [22]. When Mtrm protein is reduced (mtrm+/−), NEB can occur prematurely (at stage 12) and the spindle assembles around a karyosome that is less compact than in wild type. Although this may account for the eventual missegregation of achiasmate chromosomes in mtrm+/− heterozygotes, sister chromatid cohesion remains intact in these oocytes as evidenced by the striking chiasmata visible in the individualized bivalents [22]. In addition, premature NEB in mtrm+/− oocytes still occurs much later than the stages that exhibit sensitivity to aging in our experiments (stages 7–8). Therefore, we conclude that reducing the dosage of Mtrm is not causing the increased missegregation of recombinant chormosomes that we observe in aged oocytes.
Because the achiasmate segregation pathway is not completely disrupted in mtrmKG08051 heterozygotes, we also have been able to assess the effect of aging on the segregation of non-recombinant chromosomes. Interestingly, our data demonstrate that when the dosage of SMC1 is reduced, MI missegregation of non-recombinant chromosomes also increases with age. This result is reminiscent of our previous findings that NDJ of achiasmate chromosomes increases with age when activity of the meiotic cohesion protein ORD is compromised by a hypomorphic missense mutation [9]. In Drosophila oocytes, accurate segregation of achiasmate chromosomes relies on homologous pairing of centromere proximal heterochromatin [28]–[30] and both ORD and the cohesin complex are highly enriched at the percentric heterochromatin in oocytes [11],[17]. In addition, our recent data indicate that heterochromatin-mediated pairing is moderately disrupted by ord mutations, even in the absence of aging (VVS and SEB, unpublished results). Together, these data suggest that in addition to holding sister chromatids together, cohesion proteins play an important role in maintaining the heterochromatic pairing of achiasmate homologues in Drosophila oocytes. Whether cohesion proteins function directly to “glue” homologues together or play a more indirect role, such as influencing the structure of heterochromatin, remains to be determined. Regardless of the mechanism, loss of heterochromatin-associated cohesion proteins with age could account for the increased missegregation of non-recombinant chromosomes that we observe.
Our data support the hypothesis that age-dependent NDJ in women is caused at least in part by progressive loss of sister-chromatid cohesion over time. One unresolved issue regarding meiotic cohesion in human oocytes is whether the original cohesin molecules used to establish cohesion are maintained for decades or continually replaced as oocytes age. Addressing this question will be essential in understanding the specific mechanisms that lead to loss of meiotic cohesion and chiasma maintenance during the aging process. Early experiments in yeast led to the widely accepted model that cohesion between sister chromatids can only be established during DNA replication [31],[32]. However, recent evidence in mammalian tissue culture cells indicates that cohesin association with chromatin is much more dynamic than originally predicted [33]. In addition, experiments in yeast have recently demonstrated that cohesion can be re-established on a genome-wide scale during G2/M in a DNA damage-dependent manner [34],[35] and support the model that non-canonical mechanisms can establish cohesion after S phase is completed.
The power of genetic manipulation combined with the ability to age oocytes makes Drosophila an ideal model system to address whether re-establishment of meiotic cohesion occurs during prophase I. These future studies will be pivotal in understanding how cohesion dynamics during meiosis govern chiasma maintenance and ultimately, why and how loss of cohesion occurs as oocytes age.
Flies were reared at 25°C on standard cornmeal molasses media. The smc1ex46 allele is a deletion that removes the gene [12] and is denoted as smc1−. The mtrm[KG08051] allele results from a P-element disruption of the gene [21] and is denoted as mtrm−. Descriptions of the other genetic markers and chromosomes used in this study can be found at http://www.flybase.org.
Although several proteins (Axs, Ald/Mps1, Nod and Mtrm) are required for the accurate segregation of achiasmate chromosomes in Drosophila oocytes, we considered a mtrm allele to be the ideal choice for our studies [21],[22]. In contrast to nod, achiasmate segregation is severely affected in the mtrm heterozygote and chromosome loss is not observed [36]. In addition, unlike Axs oocytes, spindle defects and premature onset of anaphase I have not been observed in mtrm heterozygotes [37]–[39]. We thought it unwise to compromise achiasmate segregation using ald mutations given the role of this protein in the spindle assembly checkpoint; ald oocytes also exhibit low levels of NDJ of the exchange bivalents [40]–[42]. The specificity of the mutant phenotype in mtrm+/− oocytes combined with the observation that sister-chromatid cohesion appears to be normal convinced us that this was the best approach to disrupt the achiasmate segregation pathway.
To assay X-chromosome crossover frequency and distribution in smc1+/− oocytes, 5–7 females were crossed to 3 yw males per vial. Crossover frequency and distribution was measured by assaying sc, cv, v, f, car markers in the male progeny. The recombination data was used to estimate tetrad exchange ranks [18].
To age Drosophila oocytes, the aging regimen described by Jeffreys et al. was modified [9] such that the glucose agar media was prepared without the addition of fungal inhibitors (methyl paraben and ethyl acetate). The glucose agar media contained 2% agar (Fisher) and 5% dextrose (Fisher) and was prepared with milli-Q grade water. Yeast paste was prepared by dissolving 30 g of active dry yeast (Red Star) in 50 mL milli-Q grade water.
A schematic of the aging regimen is shown in Figure 2B. Approximately 200 virgin females of the desired genotype were collected within an 8-hour period and the females were fed yeast overnight in vials with cornmeal molasses media to promote vitellogenesis. Overnight incubation of females with yeast allows yolk deposition and maturation of oocytes so that the ovaries contain a complete complement of oocytes at the different stages. The following day, females were split into two groups and placed in separate plexi-glass laying bottles containing glucose agar plate with a smear of yeast paste. The control group of females was supplied with an equal number of male flies and laid their eggs continuously. Their oocytes were “non-aged”. The experimental group of females were deprived of males. Because oviposition was suppressed in these females and the developmental progression of oogenesis was halted, oocytes “aged” within their abdomens. This experimentally induced aging of oocytes mimics the aging of oocytes in human females. Control and experimental flies were held in the laying bottles for four days with fresh yeast paste/glucose-agar plates supplied each day.
At the end of the four-day aging regimen, the experimental females (with aged oocytes) and the control females (with non-aged oocytes) were crossed to X^Y, v f B males to measure meiotic nondisjunction in the oocytes (see Figure 2C). To generate 24-hour broods, 7 female flies were mated with 3–5 X^Y, v f B males (per vial). The parents were transferred to new vials every 24-hours and three broods of progeny were analyzed for NDJ.
For simplicity, we have used the term “sub-broods” to differentiate 8-hour broods of progeny from the 24-hour broods of progeny. To generate sub-broods of progeny at the end of the aging regimen, 21 experimental or control females were crossed to 10 X∧Y, v f B males (per vial). The parents were transferred to new vials every 8-hours for a total of 48 hours and six sub-broods of progeny were analyzed for NDJ.
Because Drosophila can tolerate certain sex chromosome aneuploidies, segregation errors during meiosis can be monitored in the viable progeny by using differentially marked sex chromosomes (see Figure 2C). To compensate for the fact that only half of the exceptional progeny survive (see Figure 2C), total NDJ was adjusted according to the following formula: [2*Exceptional Progeny/(2*Exceptional Progeny+Normal Progeny)]*100.
To assess whether recombinant bivalents missegregated in oocytes, the Diplo-X female progeny obtained from NDJ tests were genotyped. Each Diplo-X female was crossed to two yw males and the genotype of the X chromosomes of the Diplo-X female was inferred from the sc, cv, v, f, car markers in her male progeny (see Figure 6). Because some fraction of Diplo-X females either died before they could be genotyped or were sterile/sub-fertile, the number of Diplo-X females genotyped was lower than the number of Diplo-X females recovered from the NDJ test.
After female flies were subjected to the four-day aging regimen, ovaries from experimental and control females were hand-dissected in modified Robb's buffer [36] at 8 hour intervals and fixed in 4% formaldehyde, PBS (130 mM NaCl, 7 mM Na2HPO4 and 3 mM NaH2PO4) for 10 min. The ovaries were rinsed in PBS containing 0.01% Tween-20. Images were captured using a SMZ1500 stereo-microscope (Nikon) equipped with a Pixelink camera (Diagnostic Instruments).
To quantify the relative number of ooctyes at specific developmental stages, a single fixed ovary from three females was teased to generate individual ovarioles. Oocyte stages were tabulated and averaged for aged and non-aged ovaries at each time point. Oocytes were staged based on the standard morphological criteria described by Mahowald and Kambysellis [43]. Briefly, oocyte stages were distinguished based on the size of the egg chamber, the relative size of the oocyte and the presence of yolk within the oocyte. Yolk deposition begins during stage 8 and provides a convenient marker. Mature (stage 13–14) oocytes were distinguished by the presence of chorionic appendages.
Newly eclosed wild-type and smc1 heterozygous females were fattened with yeast for two days. 30 sets of ovaries from each genotype were dissected in modified Robb's buffer [36] and frozen in liquid N2. Each frozen tissue sample was homogenized in 240 µL of buffer containing 8 M Urea, 2% SDS, 100 mM Tris HCl pH 6.8 and 5% Ficoll containing 4 mM AEBSF (protease inhibitor, Sigma). The extract was cleared by centrifugation at 13,000 rpm for 10 min at room temperature and the supernatant was aliquoted and frozen in liquid N2. The protein concentration for each extract was determined using a Bradford assay (Biorad). For each genotype, 10 and 20 µg of total protein was separated by SDS-PAGE on a 7.5% gel and transferred to PVDF membrane. The blot was cut horizontally and anti-SMC1 guinea pig serum [11] was used at 1∶1000 dilution for the top half of the blot and monoclonal DM1A (Sigma) was used at 1∶15,000 to detect the α-tubulin loading control on the bottom half of the blot.
To compare the relative frequency of missegregation occurring in aged and non-aged oocytes, an odds ratio (OR) was calculated. The odds ratio provides a method for determining whether the frequency of a certain event (missegregation) is equal for two different treatments (aged and non-aged). In our analysis, it is calculated as: where Ae = # of exceptional progeny from aged oocytes, An = # of normal progeny from aged oocytes, NAe = # of exceptional progeny from non-aged oocytes, and NAn = # of normal progeny from non-aged oocytes [44]. If the treatments result in equal frequencies of the event, then the odds ratio will be 1. In our analysis, an OR >1 implies that the frequency of missegregation is higher when oocytes undergo aging than when they do not undergo aging.
Standard errors (SE) are calculated on the logarithm of the odds ratio according to: . In Figures 5 and 8, error bars represent ±1 SElogOR retransformed back into non-logarithmic units.
For the NDJ tests, P values were calculated using a 2×2 χ2 Contingency Test (two-tailed). For these calculations, raw data values (not adjusted values) for a specific brood or sub-brood were compared for aged versus non-aged treatments (# of exceptions and # of normal progeny for aged and non-aged). A 2×2 χ2 Contingency Test also was used to analyze the recombinational history of missegregating chromosomes recovered in Diplo-X progeny. For all tests, a P value of <0.05 was considered statistically significant (rejection of the null hypothesis that the two groups are the same).
|
10.1371/journal.pgen.1003476 | The Hourglass and the Early Conservation Models—Co-Existing Patterns of Developmental Constraints in Vertebrates | Developmental constraints have been postulated to limit the space of feasible phenotypes and thus shape animal evolution. These constraints have been suggested to be the strongest during either early or mid-embryogenesis, which corresponds to the early conservation model or the hourglass model, respectively. Conflicting results have been reported, but in recent studies of animal transcriptomes the hourglass model has been favored. Studies usually report descriptive statistics calculated for all genes over all developmental time points. This introduces dependencies between the sets of compared genes and may lead to biased results. Here we overcome this problem using an alternative modular analysis. We used the Iterative Signature Algorithm to identify distinct modules of genes co-expressed specifically in consecutive stages of zebrafish development. We then performed a detailed comparison of several gene properties between modules, allowing for a less biased and more powerful analysis. Notably, our analysis corroborated the hourglass pattern at the regulatory level, with sequences of regulatory regions being most conserved for genes expressed in mid-development but not at the level of gene sequence, age, or expression, in contrast to some previous studies. The early conservation model was supported with gene duplication and birth that were the most rare for genes expressed in early development. Finally, for all gene properties, we observed the least conservation for genes expressed in late development or adult, consistent with both models. Overall, with the modular approach, we showed that different levels of molecular evolution follow different patterns of developmental constraints. Thus both models are valid, but with respect to different genomic features.
| During development, vertebrate embryos pass through a “phylotypic” stage, during which their morphology is most similar between different species. This gave rise to the hourglass model, which predicts the highest developmental constraints during mid-embryogenesis. In the last decade, a large effort has been made to uncover the relation between developmental constraints and the evolution of genome. Several studies reported gene characteristics that change according to the hourglass model, e.g. sequence conservation, age, or expression. Here, we first show that some of the previous conclusions do not hold out under detailed analysis of the data. Then, we discuss the disadvantages of the standard evo-devo approach, i.e. comparing descriptive statistics of all genes across development. Results of such analysis are biased by genes expressed constantly during development (housekeeping genes). To overcome this limitation, we use a modularization approach, which reduces the complexity of the data and assures independency between the sets of genes which are compared. We identified distinct sets of genes (modules) with time-specific expression in zebrafish development and analyzed their conservation of sequence, gene expression, and regulatory elements, as well as their age and orthology relationships. Interestingly, we found different patterns of developmental constraints for different gene properties. Only conserved regulatory regions follow an hourglass pattern.
| Developmental constraints have been suggested to play an important role in shaping the evolution of embryonic development in animals. Briefly, the concept of developmental constraints assumes that the scope of developmental mechanisms limits the set of phenotypes that may evolve. Thus, morphological similarities between embryos of different species could reflect these underlying constraints [1]. Two main models of embryonic developmental constraints have been put forward. The early conservation model predicts that the highest developmental constraints occur at the beginning of embryogenesis. This corresponds to von Baer's third law [2], postulating that embryos of different species progressively diverge from one another during ontogeny. However, in modern times, the highest morphological similarity between embryos of different species was observed in the phylotypic stage (i.e., mid-embryogenesis) [3]–[5]. Consequently, Duboule [6] and Raff [7] proposed the so-called hourglass model, which has since become widely accepted (see, e.g., [8], [9]). It predicts the highest developmental constraints during mid-embryogenesis.
At the genomic level, the hourglass model was originally linked to the expression of Hox genes in animals [6]. More recently, the emphasis has shifted to the relation, if any, between developmental constraints and the evolution and function of the genome (reviewed in [9]). Different studies have reported several characteristics supporting the hourglass model in animals on the genomic level. Hazkani-Covo et al. [10] reported the highest protein sequence similarity between mouse and human for genes expressed in mid-development. In two influential papers, Domazet-Lošo and Tautz [11] reported that the genes expressed in mid-development of zebrafish are older than genes expressed early or late, while Kalinka et al. [12] showed that genes expressed in mid-development of fruit flies have the highest expression conservation. Similarly, Irie and Kuratani [13] reported the highest expression conservation between zebrafish, frog, chicken and mouse, for genes expressed in mid-development. Very recently, the hourglass model was argued to hold also for plants embryogenesis with respect to gene age and sequence conservation [14]. However, some of these results do not hold out under detailed analyses (see Box 1 and Text S1). For example, applying a standard log-transformation [15], [16] to microarray signal intensities used in [11] changes the reported pattern such that it no longer supports the hourglass model (Figure 1). Moreover, other studies have also found genetic patterns supporting an early conservation model [17], [18].
In most of the studies of developmental constraints the authors compared descriptive statistics of all genes across all developmental time-points (e.g., median expression [17], weighted mean age [11], mean expression correlation [13]). Such an approach introduces dependencies between the sets of genes which are compared, and consequently can produce results biased by genes expressed at many time-points. For example, housekeeping genes contribute to the average gene expression at all time points, and hence dilute trends. To overcome this essential problem, we have used a modularization approach, which we applied to the recently published transcriptome data of zebrafish development [11]. We decomposed the genes into independent sets, i.e., modules, that contained genes overexpressed solely in one of seven developmental stages: cleavage/blastula, gastrula, segmentation, pharyngula, larva, juvenile and adult. This decomposition allowed us to compare only sets of genes that have specific functions during embryonic development. For each of the seven modules, we studied five properties of its genes: 1) gene sequence conservation, 2) gene age, 3) gene expression conservation, 4) gene orthology relationships, and 5) regulatory elements conservation.
Here, we show that different levels of molecular evolution follow different patterns of developmental constraints. First, the regulatory elements are most conserved for transcription factors expressed at mid-development, consistent with the hourglass model. Contrary to what has been reported previously [10], [11], [13], we did not detect the hourglass pattern for gene sequence, age and expression. Second, constraints on gene duplication and on new gene introduction are the strongest in early development, supporting the early conservation model (consistent with [17]). Finally, all gene properties displayed the least conservation in late development and adult, which is in agreement with both models of developmental constraints.
Our goal was to analyze the developmental constraints acting on different gene properties. To this end we identified and analyzed groups of genes co-expressed during distinct developmental stages. We applied the Iterative Signature Algorithm (ISA) [19], [20] to the zebrafish expression data published by Domazet-Lošo and Tautz [11], which measured the dynamics of the transcriptome during development with a resolution of 60 time points. The ISA is a modularization algorithm that finds genes with similar expression profiles and groups them into so-called transcription modules. In order to detect modules of genes with specific expression during the zebrafish development, we initialized the ISA with seven idealized expression profiles that corresponded to successive developmental stages (see Text S1 and Figure S10).
We obtained seven modules, each containing genes overexpressed during one of the following developmental stages: cleavage/blastula, gastrula, segmentation, pharyngula, larva, juvenile and adult (Figure 2). Overall, the modules covered the entire development. The phylotypic stage in which the hourglass model predicts the highest evolutionary constraint corresponds to the segmentation and pharyngula modules. We will refer to these two modules as phylotypic modules. The cleavage/blastula and gastrula modules will be referred to as early modules, and larva, juvenile and adult modules as late modules.
The adjacent modules partially overlapped in their gene content. In order to allow for unbiased cross-module comparisons, genes belonging to two modules were kept in the one with the highest ISA gene score (see Methods); this concerned 534 genes in total. The seven modules, i.e., cleavage/blastula, gastrula, pharyngula, segmentation, larva, juvenile and adult, contained 444, 820, 487, 414, 415, 290 and 207 genes, respectively (see Table S3 for the lists of the genes). Overall, 3077 different genes were present in these modules, which implies a significant reduction of the number of genes being analyzed in comparison to the original data (14293 genes on the microarray). In particular, the ISA removed the bias related to the genes expressed uniformly across development (like housekeeping genes).
We verified the function of genes in modules detected by the ISA by comparing them to relevant known lists of genes. We found that the cleavage/blastula module was significantly enriched in maternal genes identified in [21] (36 genes vs. 19 expected by chance; hypergeometric test, ), and the gastrula module was highly significantly enriched in post-midblastula transition (post-MBT) genes identified in [21] (78 genes vs. 25 expected by chance; hypergeometric test, ). We confirmed the relevance of the segmentation and pharyngula modules by verifying that they were enriched in Hox genes (24 and 7 genes vs. 1 expected by chance, respectively; hypergeometric test, and , respectively), which is consistent with their role in mid-development [22]. We did not have any gold standard for genes expressed at the late stages of development. However, since the early and phylotypic modules were enriched in genes with relevant functions, we are confident that the same is true for the late modules.
Moreover, gene ontology (GO) enrichment analysis confirmed that genes from the modules were enriched in functions relevant to the respective developmental stages. For example, the cleavage/blastula module was enriched in genes involved in protein phosphorylation and dephosphorylation processes, which is consistent with kinase-dependent control of cell cycle and regulation of mid-blastula transition (MBT) in vertebrates [23], [24]. The pharyngula module was enriched in genes associated with cell differentiation, and anatomical structure development. Finally, the adult module was enriched in genes involved in responses to environment, although not significantly (Table S2).
We checked whether the sequences of genes from different modules evolved under different selective pressure. To this end, we calculated the non-synonymous to synonymous substitution ratios () for genes in the modules and asked if the ratio was significantly lower for any of them. With the early conservation model, we would expect the lowest values for genes from early modules. Whereas with the hourglass model, we would expect the lowest values for genes from the phylotypic modules. In the cleavage/blastula module the median was not different from the median for all genes (equal to 0.15). In the other four modules covering embryonic development the median was lower than the median for all genes (Figure 3A), and the difference was significant for all but the segmentation module (randomization test, for the gastrula, pharyngula and larva modules). In the juvenile module, the median was significantly higher than the median for all genes (randomization test, ). In the adult module, the median was also higher than the median for all genes, but the difference was not significant. When analyzing separately sites under purifying selection or evolving neutrally, we also find weaker purifying selection during post-embryonic stages (see Text S1 and Figure S11).
These results were consistent with the study by Roux and Robinson-Rechavi [17], who also reported equally low values during the entire zebrafish embryogenesis, and a small increase in mid-larva, juvenile and adult. In contrast, Hazkani-Covo et al. [10] reported an hourglass pattern for protein distance between mouse and human genes expressed during development. However, the trend was not significant. In [17] some evidence for early conservation was reported in mouse. Projecting the genes from zebrafish modules to mouse-human orthologs, we found equal conservation across development (Figure S12). Overall, data analyses support similar evolutionary constraints on sequences of genes expressed during whole embryogenesis of zebrafish, while for mouse more developmental data is needed to be conclusive.
The differences in age of genes expressed during different stages of the development have been suggested to be a good indicator of evolutionary constraints [11], [25]. Thus, we investigated the age of genes belonging to different modules. We dated each gene by its first appearance in the phylogeny and assigned it to one of the five age groups: 1) Fungi/Metazoa, 2) Bilateria, 3) Coelomata+Chordata, 4) Euteleostomi and 5) Clupeocephala+Danio rerio. Next, for each module we calculated the age distribution of its genes, i.e., the number of genes belonging to each age group, and compared it with the age distribution of all genes.
For all but the cleavage/blastula module we detected significant age variations which differed across modules (Figure 3B; chi-square goodness of fit test, all ). The oldest genes which belong to the Fungi/Metazoa class were overrepresented in the gastrula module (36.7% of genes in the module vs. 25.7% of all genes). The younger Bilateria genes were overrepresented in the phylotypic modules (45.5% and 52.1% of genes in the segmentation and pharyngula modules, respectively, vs. 34.4% of all genes). The youngest genes were overrepresented in the late modules (e.g., for Euteleostomi genes: 25.7%, 35.1% and 35.6% of such genes in larva, juvenile and adult modules, respectively, vs. 18% of all genes). In contrast, Domazet-Lošo and Tautz [11] reported that genes expressed in early and late development tend to be younger than genes expressed in mid-development, supporting the hourglass model. Yet, that result does not hold for log-transformed gene expression levels (Box 1), and is not recovered with measures of gene age other than the transcriptome age index (see Text S1 and Figure S6). With the modular approach we observed that the age of expressed genes decreased throughout ontogeny. This pattern suggests that the oldest evolutionary stages tend to express the oldest genes.
Both gene duplication and gene loss can impact phenotypic evolution [26]–[30]. The outcome of these events can be summarized by the resulting gene family size. Consequently, constrained developmental stages should display less changes in gene family size than other stages. To test this hypothesis, for each zebrafish module we calculated the number of its genes that were in 1) one-to-one, 2) one-to-many, 3) many-to-many, and 4) no orthology relation to mouse genes (i.e., no ortholog detectable by the criteria used in Ensembl Compara [31]).
We compared the observed distributions with the distribution of the ortholog relationships for all genes. We detected significant variations of the ortholog relationship for the cleavage/blastula module and for all three late modules (chi-square goodness of fit test, all ). Moreover, the pattern of variation itself differed across different modules. The number of one-to-one orthologs decreased throughout development (Figure 3C). It was significantly higher than expected only in the cleavage/blastula module (54.6% of genes in the module vs. 45.4% of all genes). In contrast, the number of genes with no orthologous relationship increased throughout development (Figure 3C). It was significantly higher than expected only in the juvenile and adult modules (38.2% and 38.4% of genes in the two modules, respectively, vs. 20.4% of all genes), consistent with the excess of “young” genes. A similar pattern was observed for many-to-many orthologs (10.4% and 7.8% of genes in the two modules, respectively, vs. 3.9% of all genes). Finally, the number of one-to-many orthologs was higher than expected only in the larva module (45.6% of genes in the module vs. 30.3% of all genes), and did not differ from expectation in all other modules.
These results were consistent with [17] in which the genes retained in duplicates after the teleost-specific whole genome duplication were reported to have low expression early in the development. Here, we recovered an analogous pattern with the modular approach, showing that the genes expressed early in the development are retained in duplicates less often than genes expressed later. Note that our observation is not limited to whole genome duplication. In addition, we detected the highest number of novel genes amongst genes expressed late in the development.
Changes in gene expression are one of the main sources of morphological variation [32]–[34]. The developmental constraints on gene expression might differ from those on the gene sequence [35]–[37]. Thus, for each module, we compared the mean expression profile of its genes with the mean expression profile of their one-to-one orthologs in mouse. We used two different data sets [13], [38] with expression values of mouse genes during the development. The use of two data sets was necessary, because there does not exist a single experiment covering the entire mouse development. The incompatibility of the two microarrays impaired the statistical strength of the analysis. For this reasons the results reported here should be regarded rather as qualitative than quantitative.
Since homology cannot be defined for individual developmental stages between zebrafish and mouse, we first mapped every time point to its broad metastage defined in Bgee database [39] (Figure 4). Next, we calculated the mean expression level in every metastage. This resulted in six expression values for each gene during the development of mouse and zebrafish: zygote, cleavage, blastula, neurula, organogenesis, and post-embryonic stage. Note that the mouse microarrays did not cover the gastrula stage at all. For each module we calculated the Pearson's correlation between the mean expression of its genes and their mouse orthologs across the six metastages. For the cleavage/blastula module no correlation was detected, probably due to the incompatibility of the two mouse microarrays. Nevertheless, there exists a plausible, biological interpretation of the differences in gene expression between the early stages of zebrafish and mouse development. Zebrafish and mouse form two different embryological structures during blastulation, a blastula and a blastocyst, respectively. The blastocyst is a mammalian innovation that consists of an embryoblast (that develop into structures of the fetus) and a trophoblast (that form the extraembryonic tissue). In contrast, there is no extraembryonic tissue in zebrafish. Overall, the lack of correlation between gene expression for the early stages of mouse and zebrafish development could be explained by these structural differences. For other modules the correlation was positive (Figure 3D), however due to the low number of data points in the analysis, no correlation values were significant (all ).
These results stood in contrast with the report by Irie and Kuratani [13] who showed the highest conservation of gene expression in mid-development. However, a re-analysis of their data suggested that this observation was not significant (see Text S1 and Figure S9). Also, both their and our studies shared problems related to the use of two data sets from different sources to cover mouse development. This and the lack of a straightforward homology between ontogenies of different species made it difficult to conclude on the conservation of gene expression during vertebrate development.
The cis-regulatory hypothesis asserts that most morphological evolution is due to changes in cis-regulatory sequences [40]–[42]. A reasonable prediction of this hypothesis is slower cis-element turnover in morphologically conserved developmental periods. We examined the presence of highly conserved non-coding elements (HCNEs) [43] and of transposon-free regions (TFRs) [44] in the proximity of genes from each module. In the analysis of HCNEs, we counted their number between zebrafish and mouse (detected with 70% identity) in regions of 500 base pairs upstream from the transcription start site. We found that only genes from the phylotypic modules were significantly enriched in HCNEs (hypergeometric test, , and for segmentation and pharyngula modules, respectively). We tested the sensitivity of the results by changing the analyzed regions' length to 200 and 1000 base pairs upstream from the transcription start site, by looking for HCNEs in introns, and using HCNEs detected with identity of 90%. In all cases, we obtained similar results (see Table S1). In the analysis of TFRs, we counted the number of genes from each module that have been associated with TFRs in zebrafish. Importantly, these TFRs were reported to be conserved between vertebrates as distant as zebrafish and human. We found that only genes from the pharyngula module were significantly enriched in TFRs (hypergeometric test, ).
The highly conserved non-coding elements and transposon-free regions are often associated with developmental regulatory genes, and with transcription factors (TFs) in particular [43]–[47]. In order to confirm this association, we calculated the fractions of genes with HCNEs or with TFRs in their proximity. We observed that for both features this fraction was higher for TFs than for all genes. Importantly, we observed that only the phylotypic modules were enriched in TFs (Figure 3E). This partially explained the enrichment in HCNEs and TFRs for genes expressed in mid-development. In addition, HCNEs were more often present in the proximity of TFs from the pharyngula module than in the proximity of TFs in general (Figure 3E; 8.8% of TFs from the pharyngula module had at least one HCNE in their proximity, and only 3.7% of all TFs had at least one HCNEs in their proximity). Also TFRs were more often present in the proximity of TFs from the phylotypic modules than in the proximity of TFs in general (Figure 3E; 31% and 45% of TFs from the segmentation and pharyngula modules, respectively, had TFRs in their proximity, and only 26% of all TFs had TFRs in their proximity). Consequently, the enrichment in HCNEs and TFRs for genes expressed in the phylotypic stage seems to be related to the regulation of developmental processes. Interestingly, only few Hox genes from phylotypic modules were associated with HCNEs (four Hox genes from segmentation module), and with TFRs (six Hox genes from segmentation module, and one Hox gene from pharyngula module).
In addition, we checked for genes that preserved their specific ancestral order in the genome across metazoans (so called conserved ancestral microsyntenic pairs, [48]) and are known to be involved in the regulation of development. We found that they were slightly overrepresented in the segmentation module, but only at the limit of statistical significance (see Text S1).
Finally, we checked for core developmental genes in each module (see [47] for the list of genes). These genes are known to be involved in the regulation of development, and to have highly conserved regulatory regions within different taxa, including, nematodes, insects and vertebrates [47]. We detected a significant enrichment in these genes only in the pharyngula module (20 core genes; hypergeometric test, ), supporting the hourglass model.
Our goal was to study developmental constraints acting on various gene properties in vertebrates. Overall, we analyzed and compared five gene characteristics, namely the conservation of gene sequence, gene expression, and regulatory elements, as well as age and orthology relationships. To this end we identified distinct sets of genes with time-specific expression in zebrafish development, i.e., genes over-expressed in one of the seven consecutive stages: cleavage/blastula, gastrula, segmentation, pharyngula, larva, juvenile and adult. We believe that the change in expression level is a reliable indicator of gene involvement in different stages, although genes might also play a role outside the stages of their highest expression. Moreover, the modules contained genes overexpressed in relation to other stages, regardless of the absolute values of their expression. Thus, lowly expressed genes were also considered by the modularization algorithm, as long as they displayed some variance in expression levels over developmental time.
Several features do not show any significant pattern over embryonic development, often in contradiction to previous reports. There is notably no evidence for change in selective pressure acting on sequences of protein-coding genes (i.e., ) over development (in contrast to [10]). Unfortunately, the available data does not allow a strong conclusion concerning the conservation of expression (in contrast to [13]), despite the probable importance of this feature in the evolution of development. In this respect, the situation in vertebrates stands in contrast to the relatively clear results in flies [12], where the evolution of expression has been shown to be most constrained in mid-development.
Gene orthology relations support the early conservation model. We show that early stages are less prone to tolerate both gene duplication (consistent with [17]) and gene introduction. The deficit in duplication in early development could also be due to a lack of opportunities for neo- or sub-functionalization in the anatomically simpler stages, which is not exclusive with strong purifying selection. The interpretation of transcriptome age is less straightforward. Our observations suggest that the oldest evolutionary stages tend to express of the oldest genes. It is possible that early stages are evolutionarily oldest, and that this is why they are enriched in oldest genes. Consequently, it is the presence of young genes in a module that would mark relaxed developmental constraints during the corresponding stage. However, neither early nor phylotypic modules are enriched in young genes (Euteleostomi and Clupeocephala+Danio rerio), which suggests similar developmental constraints in early and mid-ontogeny. In any case, we do not find any support for the hypothesis that the phylotypic stage would be characterized by the oldest transcriptome (in contrast to [11]).
While the modularization approach does not support several previous hypotheses of genomic traces of the phylotypic period, it allows us to distinguish a strong signal of conservation of gene regulation in mid-development. While this had not yet been reported in genomic studies, it is consistent with early descriptions of the phylotypic stage as characterized by Hox genes body patterning activity [6]. Of note, the patterns that we observe are robust to the removal of Hox genes (data not shown), so they are more general than this original observation. We observed an excess of HCNEs only for genes expressed in the pharyngula module, and an excess of TFRs only for genes expressed in the phylotypic modules. The enrichment in HCNEs and TFRs has been related to developmental regulatory genes, and to transcription factors in particular [43], [45]–[47]. Indeed, we observed that more TFs were expressed in mid-development than in other stages. Also, we showed that a significant proportion of TFs expressed in mid-development had conserved regulatory regions (i.e., HCNEs and TFRs), in contrast to TFs expressed early or late. Consequently, the enrichment in HCNEs and TFRs for genes expressed in mid-development can be explained by both a higher number of TFs and a higher number of HCNEs and TFRs for these TFs, than for genes expressed earlier or later. Moreover, the pharyngula module was associated with core developmental genes. Overall, these results suggest that mid-developmental processes have extremely high conservation of regulation. This conservation could translate into observed common traits of the phylum expressed at the phenotypic level during mid-development. In addition, core developmental genes are known to be present in different taxa (e.g., nematodes, insects and vertebrates), in each of which they have a conserved regulation that evolved in parallel [47]. This could explain why the phylotypic stage is observed not only in vertebrates [49], but also in other phyla, e.g., in arthropods [4], [12].
Finally, for all of the features which we have considered there is at least some trend towards weaker evolutionary constraints in the latest stages: is higher in post-embryonic stages and there are less sites under purifying selection (Figure S11); correlation of expression is lowest for maternal, larval and adult genes; young genes and genes with duplications in fishes or other vertebrates are overrepresented in late modules; and genes expressed in juveniles and adults have the less HCNEs and TFRs. Although not all of these trends are significant, no feature shows stronger conservation in late development or adult. Thus, while different aspects of gene evolution show constraints at different times of development, there appears to be a generally faster evolution of all aspects of larval, juvenile and adult genes. Whether this is due to lower constraints (i.e., less purifying selection) or to stronger involvement in adaptation (i.e., more diversifying selection), remains an open question.
In summary, we studied evidence for, or against, any particular pattern of developmental constraints by considering sets of genes with time-specific expression patterns. Comparing such independent sets of genes with a clear function during embryogenesis resulted in cleaner and more fine-grained characterization of evolutionary patterns than previously reported. Notably, we showed that different levels of molecular evolution follow different patterns of developmental constraints. The sequence of regulatory regions is most conserved for genes expressed in mid-development, consistent with the hourglass model. Gene duplication and new gene introduction is most constrained during early development, supporting the early conservation model. Whereas, all gene properties coherently show the least conservation for the latest stages, consistent with both the early conservation and the hourglass models.
Microarray data of zebrafish development were downloaded from NCBI's Gene Expression Omnibus [50] (GSE24616). This study was performed on the Agilent Zebrafish (V2) Gene Expression Microarray. In total, expression profiles for 60 developmental stages (from unfertilized egg to adults stages) were measured. The last ten stages (55 days–1 year 6 months) were measured separately for male and female. Two replicates were made per time point, resulting in microarrays in total. For each microarray, values of gProccessedSignal were log10 transformed and normalized as follows. Separately for each replicate, we equalized the expression signals between microarrays using the spike-ins reference, to account for different amounts of RNA present throughout development. To this aim, we first quantile normalized the expression signal of all spike-ins from all microarrays. Then, for each spike-in level we took the median value of expression signal before and after quantile normalization. This resulted in 10 pairs of expression signals (original signal vs. normalized signal). With linear interpolation between these points, we obtained a piecewise linear curve that defined a mapping from original to normalized expression signals, which we used to equalize the expression signals from all microarrays. This was done by projecting each expression signal onto the piecewise linear curve and calculating the corresponding normalized value. Finally, we quantile normalized the data within replicates and computed the mean value for each gene within replicates. Expression values measured separately for males and females were averaged for each time point.
Microarray data of mouse development were downloaded from Array Express (E-MEXP-51 and E-MTAB-368). The E-MEXP-51 study was performed on () F1 mice using Affymetrix GeneChip Murine Genome U74Av2. In total, expression profiles for 10 early developmental stages (zygote, early 2-cell, mid 2-cell, late 2-cell, 4 cell, 8 cell, 16 cell, early blastocyst, mid-blastocyst, late blastocyst) were measured. 2–4 replicates were made per time point. The data were normalized using gcRMA package.
The E-MTAB-368 study was performed on C57BL/6 mice using Affymetrix GeneChip Mouse Genome 430 2.0. In total, expression profiles for 8 mid and late developmental stages (E7.5, E8.5, E9.5, E10.5, E12.5, E14.5, E16.5, E18.5) were measured. 2–3 replicates were made per time point. The data were normalized using gcRMA package.
Agilent probe sets were mapped to their corresponding zebrafish genes (Ensembl release 63 [51]) using BioMart [52]. Probe sets which did not map unambiguously to an Ensembl gene were excluded from the analysis. A total of 19049 probe sets corresponding to 14293 zebrafish genes were taken into account in our analysis.
Affymetrix probe sets were mapped to their corresponding mouse genes (Ensembl release 63 [51]) using BioMart [52]. Probe sets which did not map unambiguously to an Ensembl gene were excluded from the analysis. For genes that were mapped by several probe sets we used the signal averaged across the probe sets. A total of 2883 mouse genes mapped by probe sets present on both mouse microarrays were taken into account in the gene expression analysis.
The ISA identifies modules by an iterative procedure. A detailed description of the algorithm in the general case is given in [19] (see also http://www2.unil.ch/cbg/homepage/downloads/ISA_tutorial.pdf). In this specific study, the algorithm was initialized with seven candidate seeds, each consisting of one artificial expression profile corresponding to one of the zebrafish developmental stages (see Text S1 for details). Next, these seeds were refined through iterations by adding or removing genes and developmental time points until the processes converge to stable sets, which are referred to as (transcription) modules. Each developmental time point and gene received a score indicating their membership (if non-zero) and contribution to a given module. The closest the score for a gene or developmental time point was to one, the stronger the association between the gene/developmental time point and the rest of the module.
The ISA was run twice with the following sets of thresholds: 1) and , and 2) and , for genes and developmental time points, respectively. We obtained the pharyngula module only in the case of , and all other modules with both and . All the modules contained their corresponding idealized profile. For further analysis, we kept a single module per developmental stage. From the pair of modules, we chose the one in which the idealized profile had a higher gene score. Overall, segmentation, pharyngula and juvenile modules were obtained with , and cleavage/blastula, gastrula, larva, and adult modules were obtained with .
Gene ontology (GO) association for all genes mapped by zebrafish probe sets were downloaded from Ensembl release 63 [51], using BioMart [52]. GO enrichment was tested by Fisher's exact test, using the Bioconductor package topGO [53] version 2.2.0. The reference set consisted of all Ensembl genes mapped by probe sets of the microarray used. The “elim” algorithm of topGO was used to eliminate the (tree-like) hierarchical dependency of the GO terms. To correct for multiple testing the Bonferroni correction was applied. For every module GO categories with corrected P-value lower than 0.01 were reported, if less then ten GO categories were significant we reported the top ten (see Table S2).
Ensembl Perl API release 70 [54] was used to extract all Ensembl Compara gene trees (and alignments) with a Clupeocephala (bony fishes) root. Sequences with too many gaps, or undefined nucleotides, were removed from the tree and alignment by MaxAlign (version 1.1) [55]. Only trees without duplication (one-to-one orthologs) and with at least six leaves were kept. This resulted in 6769 trees.
The site model from codeml [56] (PAML package release 4.6; models M1a and M2a in codeml) was used to predict sites-specific selection in these trees. Finally, 916 trees were removed due to the lack of zebrafish genes, and 81 were removed due to lack of expression data on the zebrafish microarray. This resulted in 5772 trees. For every gene tree we calculated its mean value ().
For every module we calculated the median ratio of its genes, where was the number of genes belonging to one of the 5772 trees. Next, we generated 10000 sets of randomly chosen genes. For each set we calculated the median ratio. Thus, we constructed a sampling distribution of the median values for a set of genes. Then we calculated the probability that the median of the original module was sampled from the constructed distribution. This allowed us to assess if the observed median ratio was significantly different from the expected median value. To correct for multiple testing we applied the Bonferroni correction. We used 0.01 as a significance level.
To study the age of genes belonging to different modules we dated the genes by their first appearance in the phylogeny. This consisted of retrieving the age of the oldest node of their Gene tree in Ensembl release 63 [51]. Genes' age was described with one of the following categories: Fungi/Metazoa, Bilateria, Coelomata, Chordata, Euteleostomi, Clupeocephala, and Danio rerio. To fit the chi-square test requirements (more than 5 elements in a group) we merged the genes into five age categories: Fungi/Metazoa, Bilateria, Coelomata+Chordata, Euteleostomi, Clupeocephala+Danio rerio. Next, for every module we calculated the age distribution of its genes. We performed chi-square goodness of fit test to compare the observed and expected distributions of age classes in the modules. The expected distribution was estimated by classifying all zebrafish genes into one of the five age categories. To correct for multiple testing we applied the Bonferroni correction. We used 0.01 as a significance level.
Homology information of zebrafish and mouse genes was retrieved from Ensembl release 63 [51], using BioMart [52]. A total of 17482 pairs of zebrafish-mouse orthologous genes had expression information in the zebrafish microarray data (14293 zebrafish genes and 11322 mouse genes). Among them there were 6441 one-to-one orthologous pairs, 5048 one-to-many orthologous pairs, and 2993 many-to-many orthologous pairs. 2901 zebrafish genes showed no orthology relationship with mouse genome. From further analysis we excluded 99 “apparent-one-to-one” gene pairs. For every module we calculated the number of genes that were in one-to-one, one-to-many, many-to-many and no orthology relation to mouse genes. Next, we performed chi-square goodness of fit test to compare the observed and expected distributions of orthology classes in the modules. The expected distribution was estimated by classifying all zebrafish genes into one of the four orthology categories. To correct for multiple testing we applied the Bonferroni correction. We used 0.01 as a significance level.
To study expression conservation between zebrafish genes assigned to the modules and their mouse one-to-one orthologs, we used gene expression data for 2883 orthologous gene pairs (the limiting factor being the mapping to both mouse microarrays). For genes that were mapped by several probe sets we averaged their signal across the probe sets for both species. In order to compare gene expression between two species, we first calculated the mean expression for zebrafish genes present in the modules and their one-to-one mouse orthologs. Due to the incompatibility of two mouse microarray data used it was difficult to provide a meaningful comparison of expression for the two species. To calculate the correlation between expression profiles between zebrafish and mouse we reduced their expression profiles to six metastages: zygote, cleavage, blastula, neurula, organogenesis, and post-embryonic stage (see [39] for detailed definition of metastage). For every module and every metastage we calculated the mean expression level for zebrafish genes and their mouse one-to-one orthologs, and next we calculated the Pearson's correlation coefficient between them.
Location data for highly conserved non-coding elements (HCNE) between zebrafish and mouse (70% of identity) was retrieved from Ancora [43] (http://ancora.genereg.net/downloads/danRer7/vs_mouse). The file HCNE_danRer7_mm9_70pc_50col.bed.gz was downloaded and used in the analysis. For each of the 14293 Ensembl genes considered in our analysis, we calculated the number of HCNE in regions of 500 base pairs upstream from the transcription start site. Next, for every module we performed a hypergeometric test to assess if they were significantly enriched in genes with HCNE. To correct for multiple testing we applied the Bonferroni correction. We used 0.01 as a significance level. In additional analyses, we calculated the number of HCNE in regions of 200 and 1000 base pairs upstream from the transcription start site, as well as in introns. Also, we repeated the analysis with HCNEs of 90% identity (see Text S1).
Location data for transposon-free regions (TFRs) in zebrafish was retrieved from [44] (http://www.biomedcentral.com/content/supplementary/1471-2164-8-470-S1.txt). First, each TFR was associated with Ensembl ID [51] of its closest transcript from genome assembly Zv6. Then for each Ensembl transcript ID we retrieved an Ensembl gene ID from genome assembly Zv7. For every module we performed a hypergeometric test to assess if they were significantly enriched in genes with TFRs in their proximity. To correct for multiple testing we applied the Bonferroni correction. We used 0.01 as a significance level.
The set of transcription factors (TFs) was defined based on GO category annotation: GO: 0006355, regulation of transcription, DNA-dependent. Among 14293 Ensembl genes, 957 were annotated as transcription factors. For every module we performed a hypergeometric test to assess if they were significantly enriched in TFs. Next, we performed a hypergeometric test to assess if the TFs present in the modules were enriched in HCNEs and TFRs. To correct for multiple testing we applied the Bonferroni correction. We used 0.01 as a significance level.
|
10.1371/journal.pntd.0004799 | Inhibition of the Hantavirus Fusion Process by Predicted Domain III and Stem Peptides from Glycoprotein Gc | Hantaviruses can cause hantavirus pulmonary syndrome or hemorrhagic fever with renal syndrome in humans. To enter cells, hantaviruses fuse their envelope membrane with host cell membranes. Previously, we have shown that the Gc envelope glycoprotein is the viral fusion protein sharing characteristics with class II fusion proteins. The ectodomain of class II fusion proteins is composed of three domains connected by a stem region to a transmembrane anchor in the viral envelope. These fusion proteins can be inhibited through exogenous fusion protein fragments spanning domain III (DIII) and the stem region. Such fragments are thought to interact with the core of the fusion protein trimer during the transition from its pre-fusion to its post-fusion conformation. Based on our previous homology model structure for Gc from Andes hantavirus (ANDV), here we predicted and generated recombinant DIII and stem peptides to test whether these fragments inhibit hantavirus membrane fusion and cell entry. Recombinant ANDV DIII was soluble, presented disulfide bridges and beta-sheet secondary structure, supporting the in silico model. Using DIII and the C-terminal part of the stem region, the infection of cells by ANDV was blocked up to 60% when fusion of ANDV occurred within the endosomal route, and up to 95% when fusion occurred with the plasma membrane. Furthermore, the fragments impaired ANDV glycoprotein-mediated cell-cell fusion, and cross-inhibited the fusion mediated by the glycoproteins from Puumala virus (PUUV). The Gc fragments interfered in ANDV cell entry by preventing membrane hemifusion and pore formation, retaining Gc in a non-resistant homotrimer stage, as described for DIII and stem peptide inhibitors of class II fusion proteins. Collectively, our results demonstrate that hantavirus Gc shares not only structural, but also mechanistic similarity with class II viral fusion proteins, and will hopefully help in developing novel therapeutic strategies against hantaviruses.
| The infection of cells by enveloped viruses involves the fusion of membranes between viruses and cells. This process is mediated by viral fusion proteins that have been grouped into at least three structural classes. Membrane-enveloped hantaviruses are worldwide spread pathogens that can cause human disease with mortality rates reaching up to 50%, however, neither a therapeutic drug nor preventive measures are currently available. Here we show that the entrance of Andes hantavirus into target cells can be blocked by fragments derived from the Gc fusion protein that are analogous to inhibitory fragments of class II fusion proteins. The Gc fragments acted directly over the viral fusion process, preventing its late stages. Together, our data demonstrate that the hantavirus Gc protein shares not only structural, but also mechanistic similarity with class II fusion proteins, suggesting its evolution from a common or related ancestral fusion protein. Furthermore, the results outline novel approaches for therapeutic intervention.
| The genus Hantavirus of the family Bunyaviridae comprises diverse viruses that are highly pathogenic to humans. In Asia and Europe the Hantaan, Seoul and PUUV viruses cause hemorrhagic fever with renal syndrome, while in America ANDV and Sin Nombre virus can lead to hantavirus pulmonary syndrome with mortality rates above 30% [1–5]. Hantaviruses are currently the most lethal human pathogenic viruses known to occur in America and, due to the lack of Food and Drug Administration (FDA)-approved preventive or therapeutic measures [6, 7], they have been classified as category A pathogens. Like other members of the Bunyaviridae family, hantaviruses have a tri-segmented single strand RNA genome of negative polarity packaged by the nucleoprotein into viral ribonucleocapsids, which are also associated to the viral RNA-dependent RNA polymerase [8]. A lipid membrane, which further envelopes the viral ribonucleocapsids, anchors the viral Gn and Gc glycoproteins. This viral envelope is derived from a host cell membrane during the budding process of the virus [9, 10]. To infect new cells, hantaviruses bind to cell surface receptors [11–14], and are subsequently taken up by endocytosis [15, 16]. A crucial step in viral cell entry is the fusion of the virus with an endosomal membrane of the host, escaping from its degradation in lysosomes. Yet, little is known about the fusion process of hantaviruses; however, our recent data show that the low pH of endosomes triggers a non-reversible fusion process, in which the Gc protein inserts into target membranes and forms a highly stable post-fusion homotrimer [17].
In general, virus-cell membrane fusion is thought to be accomplished by multiple steps [reviewed in 18, 19]. After the activation, viral fusion proteins insert a fusion peptide or fusion loop into a target membrane. At this intermediate stage, the fusion peptide is located at one end of the fusion protein while the opposite end is anchored to the viral envelope membrane by a transmembrane region, thereby bridging the viral and cellular membranes. By undergoing additional conformational changes, the fusion protein is thought to pull both anchors together, until it reaches a hairpin-like conformation in which both membrane-inserted domains are located at the same end of the protein. Once the opposed membranes have been brought into a close distance by the introduction of local membrane curvature, the fusion of the outer leaflets of the membranes produces a hemifusion intermediate, followed by the full fusion of the membranes. The fusion culminates in the formation of a fusion pore through which the virus can deliver its ribonucleocapsids into the cell cytosol to initiate replication.
Viral fusion proteins are currently grouped into at least three different classes based on their molecular structures: class I fusion proteins have a high alpha helical content, class II proteins consist principally of beta sheets, while class III fusion proteins include characteristics from the first two classes [reviewed in 19–21]. Early in silico and in vitro analyses suggested that the Gc glycoprotein of hantaviruses shares structural similarity with class II fusion proteins [22, 23]. This notion has also been proposed for other members of the Bunyaviridae [24–28], and the crystal structure of Gc from Rift Valley Fever virus (RVFV) ultimately confirmed this notion for phleboviruses [29].
Class II fusion proteins are composed of three domains (I-III) and a stem region that connects the ectodomain to the transmembrane anchor [30–34]. To adopt a hairpin-like structure, DIII moves towards the fusion loop [35, 36], while the stem region, which is connected to the transmembrane anchor, is thought to follow the movement of DIII by folding against the trimeric core formed by the fusion protein [37–39]. The extensive conformational changes that occur during the fusion process offer opportunities to disrupt the fusion cascade, thereby blocking viral infection. Ligands that bind selectively to an intermediate form of the fusion protein preceding its post-fusion conformation can delay or inhibit viral entry. In the case of human immunodeficiency virus 1, which has a fusion protein with a class I fold, there is a licensed drug based on a 20-residue peptide [reviewed in 40, 41]. This peptide comprises a partial sequence of the outer layer of the trimeric post-fusion hairpin conformation of gp41 and binds to the trimeric core of the fusion protein in its extended intermediate conformation, preventing the foldback reaction [reviewed in 42, 43]. Among class II proteins, DIII and the stem region form the outer layer of the trimeric post-fusion conformation [35–37]. Liao & Kielian (2005) showed that the addition of soluble DIII with or without the stem region of Semliki Forest virus E1 protein and soluble DIII of Dengue virus type 2 (DV2) E protein inhibit the entry of the respective virus into cells and confirmed a common inhibitory mechanism of class I and class II fusion proteins [44]. Other studies have shown that peptides derived from the stem region of the fusion protein of flavi- and phleboviruses also inhibit viral cell entry [45, 46]. The binding of stem peptides to fusion proteins is thought to prevent the post-fusion conformation as in the case of DIII; however, their amphipathic characteristics seem to allow binding to the virus before attachment to the cell, and are hence thought to be carried on the virus into endosomes [47, 48]. This characteristic provides an advantage for the delivery of the inhibitor to the site of virus-cell membrane fusion when this process occurs in a closed endosomal compartment. Here, we hypothesized that if hantavirus Gc shares mechanistic similarity with class II fusion proteins, then it should be inhibited with strategies used for other class II fusion proteins. To test this hypothesis, we predicted and produced DIII and the stem region of ANDV Gc and assessed them for ANDV inhibition. Our results show that indeed both, recombinant DIII and synthetic stem peptides, interfered with the ANDV infection, acting at late stages of the ANDV fusion process.
For the PCR amplification of the predicted DIII and DIIIS sequences we used the cloned cDNA of the M segment from ANDV isolate CHI-7913 [49] and from PUUV strain K27 cloned into pWRG/PUUV-M(s2) expression plasmid [50] (kindly provided by Jay Hooper, USAMRIID, USA), GenBank accession numbers AAO86638 and L08754, respectively. The PCR products were cloned into pET28a, which gave rise to fusion proteins with an N-terminal tag of 34 residues including polyhistidine (His-tag). For the preparation of DIII without the His-tag, the PCR product of ANDV DIII was cloned into pGST-Parallel-1 [51]. The expression product of this plasmid contained an N-terminal tag of 314 residues including a Glutathione S transferase (GST) affinity tag followed by a cleavage site for the tobacco etch virus (TEV) protease. After cleavage with TEV protease, 7 residues from the GST-fusion protein remained fused to the N-terminal of DIII, corresponding to the sequence GAMDPEF.
The expression of recombinant DIII proteins was performed as established before [52]. In brief, His-tagged fusion proteins were expressed in Escherichia coli (E.coli) BL21, which was transformed with the different pET28a/DIII plasmids. The isopropyl-beta-D-thiogalactopyranoside-induced bacteria were lysed by sonication in buffer containing 20 mM Tris, 50 mM NaCl, and 0.2 mM phenylmethanesulfonylfluoride, pH 8. Lysis was performed by sonication on ice and soluble proteins were purified through the ion exchange resin Bio-Rex 70 (Bio-Rad Laboratories) followed by tandem ultrafiltration on devices with 30 kDa and 3 kDa of molecular weight cut-off. Recombinant, untagged DIII was produced following a similar purification procedure described elsewhere [53], with some modifications. Briefly, E. coli Origami 2(DE3) (Novagen), were transformed with pGST-Parallel-1/DIII, and expression was induced, and bacteria were pelleted and lysed as described above. The clarified supernatant was purified on glutathione-Sepharose 4B (GE Healthcare), and the bound protein was eluted with 20 mM reduced glutathione (Sigma-Aldrich). His6-tagged TEV protease was used to cleave off the GST moiety, and the GST moiety and the TEV protease were removed by sequential passage through glutathione-sepharose 4B and Ni-NTA (Qiagen) resins. Eventually, DIII was concentrated using an ultrafiltration device with 3 kDa cut-off, and further purified on a HiLoad 16/60 Superdex 200 prep grade column (GE Healthcare).
Proteins were analyzed by standard Tris-glycine SDS-polyacrylamide gel electrophoresis using 4–12% or 15% gels. For irreversible reduction, 8 mM dithiothreitol was added, the proteins were heated to 95°C for 3 min, and subsequently alkylated by incubation with 24 mM iodoacetamide for 15 min at 37°C. Circular dichroism measurements were performed using a spectropolarimeter (Jasco-J600), and 5 scans were recorded at room temperature between 190 and 260 nm, using a 1 mm optical pass cuvette. Measured values of ellipticity were converted into ellipticity per amino acid residue. For deconvolution of the spectra, different databases of CONTIN and CDSSTR servers [54–56] were used.
Peptides spanning the predicted stem region of ANDV Gc were synthesized (New England Peptides). The N-terminal R1 peptide comprised the sequence HLERVTGFNQIDSDKVYDDGAPP, and the C-terminal R2 peptide comprised the sequence TFKCWFTKSGEWLLGILNGN. Two additional peptides were used, R2.1 and R2.2, comprising either the first ten residues or the last ten residues of the R2 peptide, respectively. To avoid the introduction of additional charges, the C-terminal of R2.1 was amide-modified, and the N-terminal of R.2.2 was acid-modified, comprising the sequences TFKCWFTKSG and EWLLGILNGN, respectively. The NN peptide comprising the sequence QLVTARQKLKDAEKAVEVDPDDVNKSTLQRRAAVSTLETKLG, derived from the ANDV nucleoprotein, was used as control. Such nucleoprotein-based peptide was chosen to limit the possibility of inter- or intramolecular interactions with the fusion protein that may occur during its conformational changes in the fusion process. This NN peptide of 42 residues was used as control of both short R peptides and longer DIII fragments. All peptides were soluble in HNE buffer (10 mM HEPES, 130 mM NaCl, 0,1 mM EDTA, pH 7.4), except peptides R2 and R2.2, which were dissolved in 10 mM borate buffer (pH 9).
ANDV isolate CHI-7913 (kindly provided by Héctor Galeno, Instituto de Salud Pública, Chile) was propagated in Vero E6 cells (ATTC) as described before [57]. All work involving the infectious virus was performed under biosafety level 3 conditions (Centro de Investigaciones Médicas, Pontificia Universidad Católica de Chile, Chile). 293FT cells (Invitrogen) were propagated in DMEM supplemented with 10% fetal calf serum (FCS). CHO-K1 cells (ATTC) were grown in F12-K medium containing 10% FCS.
The infection of Vero E6 cells by ANDV (multiplicity of infection (MOI) = 0.1) was quantified by flow cytometry as formerly established [57]. Briefly, cells were incubated with ANDV for 1 h at 37°C in the presence and absence of protein or peptide inhibitor candidates. Subsequently, cells were washed and infection was allowed to proceed for 16 h. Cells were next fixed with 2% (w/v) paraformaldehyde for virus inactivation, and permeabilized with 0.1% Triton X-100. For immunofluorescence staining, cells were incubated for 45 min with anti-ANDV N monoclonal antibody (mAb) 7B3/F6 [58] and then the primary antibody was detected with goat anti-mouse immunoglobulin conjugated to Alexa 555 (Life Technologies). ≥10.000 cells of each condition were analyzed using a flow cytometer (FACS CAN II, Becton Dickinson). The standard deviation of at least n = 3 experiments is indicated as the error bar of each value.
SIV vectors bearing the VSV glycoprotein were prepared as indicated elsewhere [59]. Briefly, 293FT cells were transfected with 8 μg of plasmid coding for SIV Gag-Pol (pSIV3+), 8 μg of plasmid encoding GFP as an RNA minigenome (pGAE1.0) [60], and 4 μg of plasmid coding for the envelope protein G of VSV (pVSV-G, Clontech). Alternatively, the plasmid pI.18/GPC coding for ANDV glycoproteins was used to generate SIV vectors pseudotyped with ANDV glycoproteins. At 48 h post-transfection, supernatants containing pseudotyped particles were concentrated by centrifugation at 100,000 g for 75 min. Different dilutions of VSV-G pseudotyped SIV vectors were incubated for 1 h with Vero E6 cells in the presence and absence of protein or peptide inhibitor candidates. Three days later, the expression of GFP in transduced cells was analyzed by flow cytometry (FACScan, Becton Dickinson). ≥10.000 cells were counted for each experimental condition.
A cell proliferation assay was used to assess cytotoxicity of the Gc domains and peptides as described by the manufacturer (CellTiter96, Promega). Briefly, Gc domains and peptides were incubated with Vero E6 cells for 18 h at 37°C, and the conversion of tetrazolium to formazan was assessed by measuring the absorbance at 570 nm in a microplate reader (Synergy 4, BioTek).
A three-color based cell-cell fusion assay was performed as established before [22]. Briefly, 48 h after the transfection of Vero E6 cells with pI.18/GPC [59] or pWRG/PUUV-M(s2) [50] coding for the glycoproteins of ANDV and PUUV, respectively, the cells were incubated with low pH media (E-MEM, 20 mM sodium succinate, pH 5.5) for 5 min at 37°C. Three hours later, the cell cytoplasm was stained with 5-chloromethylfluorescein diacetate (CellTracker Green CMFDA, Invitrogen), cell nuclei with DAPI (Life Technologies), and Gc was labeled by anti-Gc mAb 2H4/F6 [61], and anti-mouse immunoglobulin conjugated to Alexa555 (Invitrogen). The fusion index of Gc-expressing cells was calculated by the formula:
Fusionindex=1−[numberofcellsnumberofnuclei]
(1)
For each sample, approximately 200 nuclei per field were counted (20x magnification), and the mean fusion index of five fields was calculated (n = 3).
Vero E6 cells were pre-chilled on ice for 10 min with 20 mM NH4Cl. The adsorption of ANDV (MOI = 0.2) was performed at 4°C for 1h. Next, cells were washed, and the fusion of the virus with the plasma membrane was triggered by incubation in low pH media (E-MEM, 20 mM sodium succinate, pH 5.5) for 5 min at 37°C in the presence and absence of inhibitor candidates. Next, cells were washed, and infection was followed by incubation for 16 h at 37°C in the presence of 20 mM NH4Cl. Subsequently, viral infection was assessed as described above.
The multimerization state of Gc was assessed by sucrose gradient centrifugation as previously established [17]. Briefly, ANDV was treated at pH 5.5 to allow for the rearrangement of glycoproteins on the viral envelope. Where indicated, DIII or R2 were added to the virus during its low pH incubation during 30 min at 37°C. Subsequently, viral glycoproteins were extracted by 1% Triton X-100 and separated in a gradient of 7–15% (w/v) sucrose by centrifugation at 150,000 g for 16 h. Fractions were collected, and the presence of Gc in each fraction was assessed by western blot analysis using anti-Gc mAb 2H4/F6 [61]. The molecular mass of each fraction was assessed by the Coomassie staining of a molecular marker (Gel filtration standard, Bio-Rad) that was applied to the same sedimentation gradient. The experimental molecular mass of the marker was next plotted against the log of its theoretical molecular mass in the panel above the western blots of the gradient. The stability of the Gc homotrimer was further tested by trypsin digestion as indicated before [17]. Briefly, well-characterized VLPs projecting ANDV glycoproteins were prepared as described elsewhere [62] and were incubated for 30 min at pH 5.5. After the acidification, VLPs were incubated with TCPK trypsin (Sigma) for the indicated times. Finally, digestion was stopped by adding sample buffer and heating to 95°C for 10 min. The digestion of Gc was tested by western blot analysis, using anti-Gc mAb as described above.
In this assay, the transfer of monosialotetrahexosylganglioside (GM1) from an effector cell to a target cell was measured as described before for SNARE proteins [63]. To adapt this assay to measure the ANDV glycoprotein-mediated GM1 transfer, 293FT cells (GM1+) were transfected with pI.18/GPC [59] using Lipofectamine 2000 (Invitrogen), as indicated by the manufacturer. Forty-five h post-transfection, the cells were detached from the plates, and resuspended in supplemented DMEM. At the same time, target CHO-K1 cells (GM1-) were trypsinized and labeled with 40 μM 7-amino-4-chloromethylcoumarin (CellTracker Blue CMAC, Invitrogen) in supplemented 12-K medium for 40 min at 37°C. The excess dye was then removed by replacing the dye-containing media with supplemented F12-K medium and subsequent incubation for 30 min. After washing with PBS, target cells were resuspended in supplemented DMEM. Next, effector cells (transfected 293FT cells) and CMAC-labeled target cells (CHO-K1 cells) were combined in a ratio of 1:4 (effector:target) for 3 h. Then, the fusion protein was activated by incubating the cells in low pH media (DMEM, sodium succinate 20 mM, 10% FCS, pH 5.5) for 5 min at 37°C. This medium was replaced with supplemented DMEM, pH 7.4 and after 30 min of incubation the cells were fixed with 4% paraformaldehyde. For fluorescence staining, cells were incubated for 30 min with 5 μg/ml of cholera toxin β-subunit (CTX) conjugated to Alexa Fluor 488 (Invitrogen) at 37°C, washed with PBS, and mounted with DABCO. Cells were examined by confocal microscopy (Fluorview FV1000, Olympus and single plane images from 10 different microscopic fields were taken in each condition. The percentage of GM1 transfer was calculated using the formula:
%GM1transfer=NaNb⋅100
(2)
where Na corresponds to the number of cells positive for CMAC and GM1, while Nb corresponds to the number of cells positive for GM1 that are furthermore in contact with at least one target cell. Cells were considered positive for GM1 transfer when the fluorescent label was detected at the full circumference of the cells. The standard deviation of at least n = 3 experiments was calculated and was indicated as an error bar for each value.
A Student´s t-test was performed for statistical evaluation of n≥3 independent epxeriments: ***, P < 0.00025; **, P < 0.0025; *, P < 0.025.
For the hantavirus Gc protein, we predicted DIII and the stem region by sequence alignments with known class II fusion proteins and subsequent model derivation. These proteins included among others the more recently crystallized RVFV Gc [29]. None of the new sequence alignments achieved a higher sequence identity, greater cysteine match or model validation scores compared to the alignment used for the original Gc model structure [23], which is based on the pre-fusion structure of the tick-borne encephalitis virus (TBEV) E protein (PDBid: 1SVB) [31]. For this reason, we used the alignment from the original ANDV Gc model [23] to identify a putative DIII in ANDV Gc (Fig 1A and 1B). The sequence that was derived from this model (residues Asp315-Leu414) was termed ANDV DIII, and served as template to predict a putative DIII in Gc of other hantaviruses such as PUUV. We subsequently defined the putative stem region in ANDV Gc as the sequence encompassing residues Leu414-Asn456, which corresponded to the region between the C-terminal end of the predicted DIII and the predicted Gc transmembrane region obtained by the TMpred server [64] (Fig 1C).
The production of DIII from different class II fusion proteins, including those of flaviviruses and alphaviruses, has been previously established in E. coli [52, 67–71]. The feasibility of preparing soluble DIII from flavi- and alphaviruses in a prokaryotic expression system retaining the native structure may be related to its globular IgG-like fold and lack of glycosylation [44]. The purification of DIII is generally achieved from inclusion bodies followed by refolding [44, 71], or from the supernatant of the cell lysate [52, 67–70]. Here, we prepared three recombinant DIII proteins; DIII derived from ANDV Gc with or without N-terminal His-tag (ANDV hDIII, ANDV DIII), and DIII from PUUV Gc with N-terminal His-tag (PUUV hDIII). The proteins were obtained from the supernatant of E. coli BL21 or Origami 2(DE3) cells, and eluted after their purification as a single peak detected by absorbance at 280 nm. Fig 2A shows an elution profile example of purified ANDV DIII from the last size exclusion chromatography column together with the homogeneity of the preparation assessed by SDS-PAGE. Because the predicted ANDV and PUUV DIII sequences contain eight highly conserved cysteine residues, the DIII proteins were next characterized for the presence of disulfide bridges by reduction and subsequent alkylation. An increase in the electrophoretic mobility could be detected for reduced ANDV DIII, ANDV hDIII and PUUV hDIII compared to its unreduced control (Fig 2B), indicating that the cysteines seemed to be arranged in disulfide bridges. We next explored the presence of secondary structure elements in DIII from hantaviruses by circular dichroism. The spectra showed a unique negative maximum at 209 nm (Fig 2C), confirming the presence of secondary structure. Deconvolution of the circular dichroism spectra into four components by different servers [54–56] indicated that DIII contained 40–41% β-sheets, ~60% random coils and turns with an α-helical content close to zero. This composition coincides with the high content of β-sheets and turns observed in DIII of class II fusion proteins [31]. Taking these data together, the monomeric form of recombinant ANDV DIII in solution, the presence of disulfide bridges, the secondary structure content, and the solubility of the recombinant protein (>20 mg/ml) indicate that DIII was folded.
In addition, we also produced a protein spanning the predicted DIII and stem region of ANDV Gc, including an N-terminal His-tag (ANDV hDIIIS). However, we did not obtain enough ANDV hDIIIS for its purification, presumably due to the poor solubility of this protein. To overcome this difficulty, we generated the predicted Gc stem region separately in two synthetic peptides, one spanning the N-terminal part (R1 peptide) and the other spanning the C-terminal part (R2 peptide) (Fig 1C). We also used a negative control peptide comprising a region of the ANDV nucleoprotein (NN peptide).
Once the predicted DIII and stem fragments were synthesized, purified, and characterized, we measured their inhibitory activity against ANDV during virus cell entry via the native, endosomal infection route. For this purpose, ANDV was incubated with Vero E6 cells in the presence of the Gc fragments. After 1 h incubation, the cells were washed and the infection monitored after 16 h based on an earlier established protocol [57]. ANDV DIII and ANDV hDIII reduced ANDV infection up to 60%, at 3–4 μM (Fig 3A). The N-terminal His-tag did not further improve inhibition of viral infection, as observed for the alpha- and flavivirus DIII proteins [44]. Interestingly, PUUV hDIII did not show any cross-inhibition of ANDV. This result was unexpected since cross-reactivity by DIII has been reported within the alphavirus and flavivirus genera [44, 72]. It is likely that the absence of cross-inhibition was due to the presence of the N-terminal His-tag in PUUV DIII (see results below; Fig 4C).
Next, we tested whether the ANDV stem peptides also impair ANDV infection. The R2 peptide, comprising the C-terminal part of the predicted Gc stem region, blocked the infection of cells by ANDV up to 55% at 20 μM (Fig 3B). In contrast, the R1 peptide spanning the N-terminal part of the predicted Gc stem, did not show any reduction of ANDV infection, similar to the negative control NN peptide (Fig 3B). To test whether a specific region of R2 was required for inhibition, we further tested two different parts of R2 (see Fig 1C). The N-terminal R2.1 and C-terminal R2.2 peptides both showed similar reducing effects on viral infection (Fig 3B).
To further explore the specificity of the Gc fragments on viral inhibition, we used as a model the unrelated vesicular stomatitis virus (VSV). While VSV enters cells by endocytosis and low pH-triggered fusion, this virus achieves fusion by a different class of fusion protein (class III). To analyze VSV-mediated entry, SIV vectors [24] were pseudotyped with the envelope glycoprotein G of VSV and the transduction of cells by this vector was evaluated by the expression of the GFP reporter gene. Neither ANDV DIII nor the ANDV stem peptides altered cell transduction at any tested concentrations up to 6 μM and 60 μM, respectively (Fig 3C). In contrast, when the pH of the endocytic route was neutralized with the weak base ammonium chloride, VSV-G mediated transduction was blocked up to 80% (Fig 3C). On the other hand, when SIV vectors were pseudotyped with ANDV glycoproteins, the DIII and R2 fragments produced ~40% and ~30% of inhibition of these vectors at 6 μM and 20 μM, respectively (Fig 3D), corroborating that the inhibitory fragments were active in this system.
None of the Gc-derived peptides and recombinant domains abrogated the viability of cells when they were incubated with Vero E6 cells (Fig 3E), indicating that the reduction of ANDV infection was not due to a cytopathic effect of the Gc-derived fragments.
It was previously reported that flavivirus stem peptides bind to the virus before it enters the cell, helping its delivery into endosomal compartments [47]. Based on this observation, we compared the inhibition of ANDV by (a) pre-incubation of the virus with the R2 peptide, or (b) co-incubating the R2 peptide with the virus during its adsorption to cells. Comparing the results of both experimental designs, a similar dose-dependent inhibition could be observed (Fig 3F), coinciding with the results obtained for flaviviruses (47). Based on this result, it seems plausible that binding of the R2 peptide to ANDV may occur very fast, making longer incubation times unnecessary.
Taken together, our data demonstrate that the exogenous fragments derived from ANDV Gc, comprising the predicted DIII or the C-terminal part of the predicted stem region, abrogate the entry of ANDV into cells.
The inhibitory potential of the Gc fragments was next tested in a cell-cell fusion assay driven by the ANDV glycoproteins [22]. Therefore, we transfected Vero E6 cells with a plasmid coding for the ANDV glycoproteins, and incubated the cells 48 h post-transfection with low pH media for 5 min at 37°C, as described in Materials and Methods. The acidification of the media allows the activation of the viral Gc fusion protein on the plasma membrane of cells, thereby triggering the formation of syncytia. When the ANDV Gc fragments were incorporated in the low pH incubation step, ANDV DIII and ANDV hDIII diminished the glycoprotein-mediated fusion activity by ~70% at 25 μM and 50 μM, respectively (Fig 4A). In concordance with the results obtained with infectious ANDV (Fig 3A), the PUUV hDIII did not achieve cross-inhibition of ANDV glycoproteins (Fig 4A). On the other hand, the ANDV R2 stem peptides impaired the ANDV glycoprotein fusion activity up to 75% at 20 μM (Fig 4B).
The ANDV-derived Gc fragments were likewise tested for their potential to cross-inhibit other hantaviruses such as PUUV. Therefore, we performed a cell-cell fusion assay driven by the PUUV glycoproteins. In this assay, PUUV hDIII blocked the PUUV glycoprotein-mediated fusion process, reducing fusion up to 65% at 25 μM (Fig 4C). This result coincides with the concentration range in which the ANDV DIII proteins block the ANDV glycoprotein fusion activity and confirms the activity of PUUV hDIII. When we tested the ANDV DIII proteins for cross-inhibition, we found that ANDV DIII, but not ANDV hDIII, blocked PUUV glycoprotein-mediated fusion up to 50% (Fig 4C). These results, together with the data on absent cross-reaction of PUUV hDIII with ANDV, suggest that the hantavirus DIII proteins without the N-terminal His-tag have cross-inhibiting activity; however the N-terminal His-tag of these domains seems to interfere in the inhibition. Thus so far, this notion remains to be corroborated with PUUV DIII lacking the N-terminal His-tag.
When we tested the ANDV stem peptides to block PUUV-mediated fusion, we found that they also had a cross-inhibition function (Fig 4D). Among them, the R2.1 peptide reached the highest cross-reduction result of ~70% at 20 μM, in line with the observation that this peptide also achieved the highest inhibition value of ANDV-mediated cell-cell fusion. Collectively, these results on hantavirus cross-inhibition suggest that residues in DIII and stem fragments are involved in intramolecular interactions. Some of these residues seem to be conserved between ANDV and PUUV (Fig 1D), allowing for the cross-interaction with exogenous DIII and stem fragments from a different hantavirus.
ANDV cell entry can be blocked at different steps such as receptor binding and membrane fusion. For hantaviruses, the envelope glycoprotein or the specific domain involved in binding to receptors has not yet been identified. To discard a possible effect of the ANDV inhibitors in steps preceding virus-cell membrane fusion, we incubated the cells with ANDV DIII or the R2 peptide for 1 h before the addition of the virus. Unbound fragments were subsequently washed out, and cells were then infected with ANDV. The addition of neither DIII nor R2 at 10 and 20 μM, respectively, before the cells were incubated with ANDV led to a significant decrease in virus infection (Fig 5A). These data emphasize that the pre-incubation of DIII and stem fragments did not abrogate early steps in virus infection such as receptor binding or cellular signaling pathways. Furthermore, the results coincide with data obtained with stem peptides derived from DV2, where the pre-incubation of cells with these peptides also did not affect infection by DV2 [47].
Next, we asked whether the DIII and stem fragments interfered directly in the virus-cell fusion process. To test this, we assessed inhibition in a fusion infection assay, fusing ANDV with the plasma membrane. Therefore, ANDV was pre-bound to cells at 4°C and then ANDV DIII or R2 were added during the 5 min low pH pulse that triggers fusion. The blockade of this viral entry pathway was highly efficient in the case of the R2 peptide, reaching over 95% of inhibition at 20 μM (Fig 5B). Recombinant DIII led to a lower inhibition efficiency of 70% at 20 μM, result that in comparison to that obtained with R2 could be explained by the short incubation time in this experimental design. Compared to inhibition of the normal entry route of the virus, higher concentrations of ANDV DIII might have been necessary for inhibition since more input virus was used to reach similar levels of infection (MOI = 0.2). Taking these data together, our results confirm that the exogenous DIII and stem fragments function specifically during the viral fusion process.
Since the fusion process involves multiple steps, we next assessed at which specific stage the Gc fragments interfere in this process. To this end, we started analyzing the trimerization of Gc using an earlier protocol [17]. ANDV was therefore incubated at neutral or low pH with or without DIII. Subsequently, the viral glycoproteins were extracted from the virus and applied to a sucrose gradient (7–15%) to evaluate their molecular mass. After ultracentrifugation, each fraction was examined by western blot analysis for the presence of Gc. Gradient sedimentation at pH 7.4 led to the detection of Gc in fractions corresponding to the molecular mass of Gc monomers of ~50 kDa (fractions 5–7), in the presence or absence of ANDV DIII (Fig 6A). Two Gc migration bands could be observed in the reducing electrophoretic system, which may correspond to different oxidation forms of Gc as described earlier [73]. Only the lower molecular mass band was previously found to shift from Gc monomers at neutral pH to Gc trimers at low pH [17]. When we incubated ANDV at pH 5.5, in the presence or absence of the DIII inhibitor, the lower molecular mass band of Gc shifted from the fractions corresponding to monomers and was found in fractions corresponding to Gc homotrimers (fractions 11–12; Fig 6A). This result indicates that ANDV DIII abrogated neither Gc fusion activation, nor Gc trimerization.
The post-fusion conformation of ANDV Gc corresponds to a highly stable homotrimer [17]. In this context, we next explored the stability of the Gc homotrimer formed in the presence of exogenous DIII and stem fragments by assessing its resistance to protease digestion. For this experimental approach we used ANDV-like particles (VLPs) [62], since they can be purified to higher concentrations than the virus. These VLPs resemble the viral envelope by exposing the Gn and Gc glycoproteins. To test their trypsin resistance, the VLPs were first incubated at neutral or low pH in the presence or absence of the inhibitors, and subsequently subjected to digestion with trypsin for different incubation times. As expected, the neutral pH form of Gc was fully degraded within 30 min, while the low pH form of Gc was largely resistant to digestion (Fig 6B). In contrast, Gc lost its trypsin resistance and digestion could be observed when ANDV DIII or the R2 peptide was co-incubated with VLPs during the low pH incubation step, indicating that they interfered in the formation of a stable Gc homotrimer (Fig 6B). Some residual Gc could be detected, most likely because the DIII or stem fragments did not block all Gc molecules, which coincides with the inhibitory results (Figs 3A and 3B and 5B). The formation of a stable homotrimer was however not prevented when the control peptide NN was added before acidification, nor when either the NN peptide, DIII or the R2 peptide was added after the low pH incubation, confirming the specificity of the assay. Together, these data suggest that the exogenous DIII and stem fragments prevented the formation of a stable post-fusion hairpin structure, presumably by direct binding to Gc in its extended intermediate conformation.
The presence of exogenous Gc fragments interfered neither with the activation of ANDV Gc nor with its homotrimerization; however it did not allow the formation of a stable post-fusion structure. The hemifusion intermediate is a stage that occurs between these fusion steps, in which the outer membranes of the virus and cell have fused, while the inner leaflets still remain apart. In order to test if the inhibition of ANDV by exogenous DIII occurs before or after the hemifusion intermediate state, we developed a hemifusion assay for the ANDV glycoproteins. For this purpose, we took advantage of a previously developed cell-cell fusion assay between 293FT cells expressing the ANDV glycoproteins (effector cells) and CHO-K1 cells (target cells) [17]. In this assay, the low pH-triggered transfer of GM1 from effector cells to target cells was analyzed by confocal microscopy, which allows detection of GM1 at the cell surface in a single focal plane, with minimal sample intervention (Fig 7A). To allow unambiguous identification of GM1 transfer from effector cells to target cells, CHO-K1 cells were only defined as GM1+ when the label was detected on their full circumference (Fig 7A, red arrows). Conversely, target cells in contact with effector cells, but showing only partial or no GM1 stain, were defined as GM1- (Fig 7A, yellow arrows). Although this assay allows for the detection of lipid mixing between cells, it does not discriminate between full fusion and hemifusion of membranes. To obtain a quantitative measure of GM1 transfer from effector to target cells, we quantified the percentage of transfer in each condition. At pH 7, a background level of transfer around 20% was observed in all conditions (Fig 7B). When glycoproteins were activated at pH 5.5, GM1-transfer from effector cells to target cells was detected in ~70% of Mock control treatment (Fig 7B). However, when ANDV DIII was incorporated in the low pH incubation, the transfer of GM1 appeared to be less frequent, decreasing to below 40% (Fig 7B). This value is still higher than the background level observed for GM1 transfer at neutral pH, indicating that this domain reduced lipid mixing in an incomplete manner. Most probably, exogenous ANDV DIII did not make contact with all fusion proteins during the short low pH incubation, which coincides with the results for blocking cell-cell fusion (see Fig 4A). The impairment of GM1-transfer was highly specific, since the incorporation of PUUV hDIII, which has no cross-inhibition activity against ANDV (see Figs 3 and 4A), did not abrogate the acid-induced GM1 transfer of ~70% (Fig 7A and 7B). In summary, our results show that ANDV DIII arrests the fusion process after Gc trimerization, but before reaching a hemifusion intermediate. In this context, it seems likely that Gc fragments prevent the movement of the endogenous DIII or the stem region towards the core of the Gc trimer as described for other class II fusion proteins [39, 44].
The fusion of the viral membrane with a host cell membrane is a crucial step in the entry of enveloped viruses into cells. In the present study we demonstrated that predicted DIII and stem fragments blocked acid-induced fusion of ANDV within the endosomal entry pathway and with the cell surface. Fusion was allowed to proceed until Gc trimerization, but prevented membrane hemifusion and fusion pore formation. These results not only provide novel information about inhibitory strategies against ANDV and other hantaviruses, but also provide a proof of concept that Gc shares structural similarity with the overall fold of class II fusion proteins.
Comparing the inhibition of hantaviruses by exogenous DIII with that of other class II fusion proteins, ANDV and PUUV hantaviruses were blocked by ANDV DIII without additional N-terminal His-tag or C-terminal residues, although containing seven N-terminal residues derived from the GST-tag. While the addition of an N-terminal His-tag achieved a 100-fold improvement in blocking fusion of Semliki Forest virus [44], ANDV DIII ‒ with or without N-terminal His-tag ‒ achieved similar inhibitory results. For SFV DIII it has been proposed that this N-terminal tag may mimic the domain I-DIII linker region, thereby stabilizing the interaction with the fusion protein core [44]. On the other hand, it has been reported that the presence of C-terminal residues derived from the stem region is necessary for inhibition by exogenous DIII of DV2 and chikungunya virus E proteins [44, 72]. More specifically, for chikungunya virus it has been shown that nine residues from the E stem region are required for DIII to bind to the fusion protein domain I-domain II core. Hence, the potency of inhibiting the ANDV fusion process through ANDV and PUUV DIII may be further improved in future studies by adding N- or C-terminal residues.
The ANDV DIII, including the stem region, was largely insoluble and therefore this larger Gc fragment could not be tested in the ANDV entry assays. The R1 peptide spanning the first 17 N-terminal residues of the 44-residue stem region did not affect ANDV entry, while the R2 peptide spanning the last 20 C-terminal residues blocked ANDV infection to a similar extent as DIII. When we tested the two peptides R2.1 and R2.2, comprising either the N- or C-terminal half of the ANDV R2 stem peptide, the inhibitory activity against ANDV was largely retained by both peptides. Hence, the R2 peptide seems to contain residues in these two regions that participate similarly during inhibition. Similar to the C-terminal region of the ANDV stem region, the C-terminal region, but not the N-terminal, of DV2 E protein binds and blocks the fusion protein [47]. Interestingly, inhibitory peptides derived from membrane proximal regions (or stem regions) of diverse fusion proteins such as those of RVFV, DV2, SARS-coronavirus, Influenza virus, and Hepatitis C virus have been found likely to interact with membrane interfaces by a hydropathy segment [74], which is predicted by the Wimley-White interfacial hydrophobicity scale (WWIHS) for the transfer of a peptide from an aqueous environment to a palmitoyl-oleoyl-phosphatidyl choline interface [75, 76]. A need for such a membrane-binding property for inhibitory activity coincides with studies on peptides derived from the stem region of alphavirus fusion proteins; these peptides generate WWIHS below values of 1 (S1 Table), indicative of weak membrane binding [74], and fail to block alphavirus fusion activity [44, 72]. In the case of the DV2 E protein, such a membrane-binding sequence is found at the C-terminal end of the peptide [48] with a high WWIHS value (S1 Table). In fact, the stem peptides of the DV2 E protein have been shown to interfere with the fusion of the virus in the endosomal compartment by a two-step mechanism: first by binding to the viral membrane outside the cell, and next by binding against the E core trimer once fusion has been triggered in the endocytic compartment (47). Using the Wimley-White scale [77], an interfacial hydropathy segment with a WWIHS vale >5 was predicted for the ANDV Gc stem region that coincides with the sequence of the R2 peptide (S1 Table), indicative for moderate membrane partitioning [74]. If the ANDV R2 peptide would interact with membranes by a hydropathy segment, then such an interaction did not favor ANDV inhibition sufficiently in the endosomal route, since fusion impairment was most efficient when directly present in the fusion compartment; when applied at the same concentration, the R2 stem peptide blocked 45% of ANDV infection when entry occurred via the endocytic pathway, while R2 reached over 95% inhibition of ANDV infection when fusion occurred at the plasma membrane. In this sense, modifications to the R2 peptide that favor membrane interaction, such as those introduced to peptides derived from the West Nile virus E stem region [48] or the membrane proximal region of Influenza virus hemagglutinin and HIV gp41 [78, 79], may help in the future to improve its inhibitory activity and to direct it towards the closed environment of endosomes.
The ANDV DIII and stem peptides blocked not only fusion mediated by ANDV glycoproteins, but also the fusion activity of the glycoproteins of another hantavirus (PUUV). This result is in accordance with the high sequence identity of 72% between DIII of these viruses and also with reports on the cross-inhibition activity of DIII within the genus Alphavirus [72], where DIII conservation is as high as 50%. However, the N-terminal His-tag of ANDV hDIII seemed to prevent the cross-inhibition of PUUV fusion activity, indicating that this tag may interfere in specific binding to Gc from heterologous species. It is likely that the histidines of the tag become positively charged in the low pH environment, which in turn may induce repulsion with positively charged residues in the Gc of hantaviruses. Such repulsion may be overcome by a higher binding affinity of DIII to Gc from the same hantavirus, but not to heterologous viruses. In addition to the DIII of ANDV, peptides derived from the stem region of ANDV also cross-inhibited the PUUV fusion activity, which further corroborates the presence of conserved residues among hantavirus Gc proteins that are involved in the likely binding of this peptide. Cross-inhibition of fusion proteins by stem peptides has been previously reported for viruses from the genus Flavivirus; Dengue virus stem peptides blocked different Dengue virus serotypes but not other flaviviruses. [48]. The absence of cross-inhibition in that case was related not to a poor interaction with the respective E protein, but rather to a poor interaction with the viral membrane [48]. Finally, stem peptides derived from the RVFV Gc protein have been reported to block the three different classes of viral fusion proteins [46], acting as a broad-spectrum fusion inhibitor [80]. For the exogenous stem peptides from ANDV we did not observe cross-inhibition of other fusion proteins such as that of VSV at concentrations up to 60 μM. Therefore, it is more likely that the ANDV stem fragments may be applied to inhibit similar viruses within the same genus, but not other viral fusion machineries.
Taken as a whole, our results demonstrate that strategies employed against class II fusion proteins allow for the inhibition of hantaviruses such as ANDV and PUUV. Although targeting the endosomal site of virus fusion has not yet been optimized, it was possible to block fusion and infection under physiological virus entry conditions. Hopefully, the novel inhibitory strategy based on ANDV DIII and stem peptides will help in the future development of therapeutic strategies against different hantaviruses.
|
10.1371/journal.pgen.1002805 | A Genome-Wide Association Meta-Analysis of Circulating Sex Hormone–Binding Globulin Reveals Multiple Loci Implicated in Sex Steroid Hormone Regulation | Sex hormone-binding globulin (SHBG) is a glycoprotein responsible for the transport and biologic availability of sex steroid hormones, primarily testosterone and estradiol. SHBG has been associated with chronic diseases including type 2 diabetes (T2D) and with hormone-sensitive cancers such as breast and prostate cancer. We performed a genome-wide association study (GWAS) meta-analysis of 21,791 individuals from 10 epidemiologic studies and validated these findings in 7,046 individuals in an additional six studies. We identified twelve genomic regions (SNPs) associated with circulating SHBG concentrations. Loci near the identified SNPs included SHBG (rs12150660, 17p13.1, p = 1.8×10−106), PRMT6 (rs17496332, 1p13.3, p = 1.4×10−11), GCKR (rs780093, 2p23.3, p = 2.2×10−16), ZBTB10 (rs440837, 8q21.13, p = 3.4×10−09), JMJD1C (rs7910927, 10q21.3, p = 6.1×10−35), SLCO1B1 (rs4149056, 12p12.1, p = 1.9×10−08), NR2F2 (rs8023580, 15q26.2, p = 8.3×10−12), ZNF652 (rs2411984, 17q21.32, p = 3.5×10−14), TDGF3 (rs1573036, Xq22.3, p = 4.1×10−14), LHCGR (rs10454142, 2p16.3, p = 1.3×10−07), BAIAP2L1 (rs3779195, 7q21.3, p = 2.7×10−08), and UGT2B15 (rs293428, 4q13.2, p = 5.5×10−06). These genes encompass multiple biologic pathways, including hepatic function, lipid metabolism, carbohydrate metabolism and T2D, androgen and estrogen receptor function, epigenetic effects, and the biology of sex steroid hormone-responsive cancers including breast and prostate cancer. We found evidence of sex-differentiated genetic influences on SHBG. In a sex-specific GWAS, the loci 4q13.2-UGT2B15 was significant in men only (men p = 2.5×10−08, women p = 0.66, heterogeneity p = 0.003). Additionally, three loci showed strong sex-differentiated effects: 17p13.1-SHBG and Xq22.3-TDGF3 were stronger in men, whereas 8q21.12-ZBTB10 was stronger in women. Conditional analyses identified additional signals at the SHBG gene that together almost double the proportion of variance explained at the locus. Using an independent study of 1,129 individuals, all SNPs identified in the overall or sex-differentiated or conditional analyses explained ∼15.6% and ∼8.4% of the genetic variation of SHBG concentrations in men and women, respectively. The evidence for sex-differentiated effects and allelic heterogeneity highlight the importance of considering these features when estimating complex trait variance.
| Sex hormone-binding globulin (SHBG) is the key protein responsible for binding and transporting the sex steroid hormones, testosterone and estradiol, in the circulatory system. SHBG regulates their bioavailability and therefore their effects in the body. SHBG has been linked to chronic diseases including type 2 diabetes and to hormone-sensitive cancers such as breast and prostate cancer. SHBG concentrations are approximately 50% heritable in family studies, suggesting SHBG concentrations are under significant genetic control; yet, little is known about the specific genes that influence SHBG. We conducted a large study of the association of SHBG concentrations with markers in the human genome in ∼22,000 white men and women to determine which loci influence SHBG concentrations. Genes near the identified genomic markers in addition to the SHBG protein coding gene included PRMT6, GCKR, ZBTB10, JMJD1C, SLCO1B1, NR2F2, ZNF652, TDGF3, LHCGR, BAIAP2L1, and UGT2B15. These genes represent a wide range of biologic pathways that may relate to SHBG function and sex steroid hormone biology, including liver function, lipid metabolism, carbohydrate metabolism and type 2 diabetes, and the development and progression of sex steroid hormone-responsive cancers.
| Sex hormone-binding globulin (SHBG) is a protein secreted mainly by the liver that binds to the sex steroids, testosterone, dihydrotestosterone, and estradiol, transports them in the circulation, and influences their action in target tissues by regulating their bioavailability. SHBG thereby influences the expression of sex hormone sensitive phenotypes including sexual characteristics and reproductive function in men and women [1]. In addition to regulating sex steroid hormone effects, SHBG may exert independent effects through its own receptor [2]. Variation in SHBG concentration has also been associated with various chronic diseases including cancers [3], polycystic ovary syndrome (PCOS) [4], [5] and type 2 diabetes (T2D) [6], [7]. Although SHBG is estimated to have a heritable component (∼50%) [8], little is known about the genetic regulation of SHBG. Polymorphisms at the SHBG gene locus have been associated with SHBG concentrations [9], [10], but much remains unknown about specific genetic variants that may determine circulating SHBG concentrations. Identifying genetic factors that influence SHBG may provide insights into the biology of sex steroid hormone regulation, metabolism and tissue effects that underlie their relationship with chronic diseases such as T2D as well as hormone-sensitive cancers such as breast and prostate cancer.
We identified nine loci associated with SHBG concentrations at the genome-wide significance threshold of p = 5×10−8 (Table 1 and Figure 1) in a genome-wide association study (GWAS) meta-analysis of circulating SHBG concentrations in 21,791 men and women from 10 studies (Table S1). All nine lead SNPs at these loci had effects in the same direction (seven with p<0.05) in the validation dataset of 7,046 men and women from six additional studies (Table S2). The strongest association was within the SHBG locus (rs12150660, p = 2×10−106). Together, these nine lead SNPs explained 7.2% of the genetic variance (assuming 50% heritability) in SHBG concentrations.
We next performed a series of additional analyses to explain more of the phenotypic variance (Figure 2). First, we hypothesized that genetic effects may be different in men and women, as SHBG concentrations are >50% higher in females than males, and may be differentially regulated between sexes. In a sex stratified analysis, three of the nine loci showed evidence of sex-differentiated effects at p<0.02 when we would not expect any signals to have reached this level of significance by chance. The associations at the 17p13.1-SHBG and Xq22.3 loci were stronger in males whereas the association at the 8q21.13 locus was stronger in females. To investigate the apparent differential sex effect for the X chromosome further we ran a recessive regression model for the X chromosome SNP rs1573036 in women in the Framingham Heart Study and found no association with SHBG suggesting the sex-differentiated effect is not the result of a recessive inheritance pattern. Sex stratified GWAS identified one novel signal in men, which showed no association in women (4q13.2: men p = 2.5×10−8, women p = 0.66, heterogeneity p = 0.003).
A series of conditional analyses were performed to identify statistically independent signals. At the SHBG locus, three apparently independent additional signals separate from the main index SNP were observed, based on low (r2<0.05) pairwise correlations in HapMap (rs6258 p = 2.7×10−46, rs1625895 p = 1.2×10−14 and rs3853894 p = 2.5×10−11). A series of iterative conditional analyses (Table 2) involving SNPs at the SHBG locus generated a final regression model including six statistically independent SHBG SNPs. Four of these SNPs (#1–4 Table 2) retained GWS when conditioned against the other five, and two were nominally associated (SNP#5 p = 0.0001, SNP#6 p = 0.01). Re-running the GWAS meta-analysis adjusting for these six SNPs revealed evidence for three additional statistically independent (pairwise HapMap r2<0.01) signals at the SHBG locus (SNP#7 p = 1.5×10−7, SNP#8 p = 4.6×10−5, SNP#9 p = 9.9×10−6) (Figure 3). There were also two additional trans signals located at 2p16.3 and 7q21.3 (Table 1). Although the 2p16.3 signal dropped below GWS when combined with follow-up samples (p = 1×10−7), the index SNP at 2p16.3 is ∼300 kb away from a strong candidate gene, the luteinizing hormone receptor gene (LHCGR).
The majority of pair-wise correlations for the nine SHBG locus SNPs highlighted by our conditional analyses showed very low HapMap r2 values. However, the pairwise D′ values are often high (Table S3) indicating that no or few recombination events have occurred between some SNPs, and that combinations of SNPs may be tagging un-typed variants on a common haplotype. To investigate this possibility, we performed more extensive analyses in a single study (NFBC1966, n = 4467). We used a denser set of SNPs imputed from the June 2011 version of the 1000 Genomes data and performed model selection analyses. Model selection identifies a set of SNPs that best explain phenotypic variation, while simultaneously penalizing each SNP included in this set, and therefore correlated SNPs tend to be excluded from the final model. These analyses consistently included at least seven SNPs in the model, although it is hard to estimate the false-negative rate of using the reduced sample size. While we are underpowered to accurately pinpoint the exact number of independent signals, these analyses support the results of the conditional analysis and suggest that multiple variants at the SHBG locus are independently associated with SHBG concentrations.
Data from an independent study, the InCHIANTI study, was used to calculate the proportion of genetic variance in SHBG concentrations explained when accounting for sex specific effects, the multiple signals of association at the SHBG locus, and the additional trans signals identified post conditional analysis. In men and women we explained ∼15.6% and ∼8.4% of the heritable component respectively. The SHBG locus accounted for ∼10% and ∼6.6% of the genetic variation in men and women respectively with the lead SNP in isolation accounting for ∼7.8% and ∼3.3% of the variation in men and women, respectively.
We identified genes near the associated SNPs and explored their biologic relevance to SHBG. The genes associated with identified SNPs included the SHBG locus (rs12150660, 17p13.1, p = 1.8×10−106), PRMT6 (rs17496332, 1p13.3, p = 1.4×10−11), GCKR (rs780093, 2p23.3, p = 2.2×10−16), ZBTB10 (rs440837, 8q21.13, p = 3.4×10−09), JMJD1C (rs7910927, 10q21.3, p = 6.1×10−35), SLCO1B1 (rs4149056, 12p12.1, p = 1.9×10−08), NR2F2 (rs8023580, 15q26.2, p = 8.3×10−12), ZNF652 (rs2411984, 17q21.32, p = 3.5×10−14), TDGF3 (rs1573036, Xq22.3, p = 4.1×10−14), LHCGR (rs10454142, 2p16.3, p = 1.3×10−07), BAIAP2L1 (rs3779195, 7q21.3, p = 2.7×10−08), and UGT2B15 (rs293428, 4q13.2, p = 5.5×10−06) (Figure 1).
We used the online tool STRING (www.string-db.org) to perform pathway analyses to explore possible interactions between the SHBG gene and the proteins encoded by the 11 most plausible genes nearest the 11 SNPs listed above. There was an interaction noted between GCKR and JMJD1C which were associated with the lipoprotein fractions VLDL and HDL, respectively [11]. In an expanded analysis, we assessed protein interactions among SHBG and 67 genes within 500 kb of our 11 identified SNPs and uncovered additional protein interaction pathways. An interaction between two proteins encoded by GTF2A1L and STON1 was found; these proteins are co-expressed in testicular germ cells in the mouse [12]. An interaction between LHCGR and BRI3 encoded proteins that are associated with the G-protein coupled receptor complex in the human luteinizing hormone receptor was also identified [13]. Finally, an interaction between LHCGR and IAPP (amylin) proteins which are components of a ligand/G-protein receptor/G-protein alpha subunit complex was found (database: www.reactome.com).
Targeted analysis of two strong candidate genes, hepatocyte nuclear factor-4α (HNF4α) and peroxisome-proliferating receptor γ (PPARγ) did not identify any SNPs at HNF4α but did identify one SNP, rs2920502, at PPARγ that reached statistical significance (p = 9.9×10−5) and a second SNP at PPARγ, rs13081389, that reached nominal significance (p = 0.01).
In total, we identified 12 genomic regions associated with circulating SHBG concentrations, including extensive allelic heterogeneity at the SHBG locus itself. Conditional meta-analyses carried out at the SHBG locus, identified nine genome-wide significant SNPs with low correlation (r2<0.01) between them. Two of these signals (rs6258 [10] and rs6259) are missense variants and two are low frequency variants (MAF ∼2%). Furthermore, rs12150660 is highly correlated (r2>0.95) [10] with a pentanucleotide repeat, which affects SHBG expression in-vitro [14]. To our knowledge, the magnitude of secondary signals observed at this locus are the largest seen for any complex trait.
The proportion of genetic variance in SHBG serum concentrations explained when accounting for sex specific effects, the multiple signals of association at the SHBG locus, and the additional trans signals identified post conditional analysis was ∼15.6% in men and ∼8.4% in women. The SHBG locus accounted for ∼10% and ∼6.6% of the genetic variance in men and women, respectively, with the lead SNP explaining most of the genetic variation at ∼7.8% for men and ∼3.3% for women. Thus additional signals at the SHBG locus identified through conditional analyses approximately doubled the variance of the trait explained. While we provide evidence for multiple variants associated with SHBG concentrations, further studies are needed to pinpoint the causal loci and functional variants. For the 11 regions outside the SHBG locus, most have biologically plausible related genes within 300 kb.
Several genes near the identified SNPs regulate sex steroid production and function. The NR2F2 locus (15q26.2) encodes a nuclear receptor important in testicular Leydig cell function, the primary source of gonadal testosterone production [15], and has been linked to male infertility [16]. NR2F2 has also been associated with estrogen receptor alpha (ERα) signaling and may influence hormone responsivity in breast cancer [17]. PRMT6 (1p13.3) also encodes a nuclear receptor regulatory protein that mediates estrogen signaling as a co-activator of the estrogen receptor [18]. LHCGR (2p16.3) encodes the luteinizing hormone receptor which was associated with polycystic ovary syndrome (PCOS) in a recent GWAS [19], [20]. PCOS is both a reproductive and metabolic disorder characterized by higher testosterone serum concentrations as well as an increased prevalence of obesity, insulin resistance, and T2D in women. Inappropriate secretion of luteinizing hormone leads to increased ovarian production of testosterone. Coincident lower SHBG concentrations contribute to increased bioavailable testosterone concentrations and the expression of both reproductive and metabolic phenotypes in PCOS [21], [22], [23].
The SLCO1B1 locus encodes a liver-specific transporter of thyroid hormone as well as estrogens which impact liver production of SHBG [24]. JMJD1C (10q21.3), also known as TRIP 8 (thyroid hormone receptor interactor protein 8 [25]), may impact SHBG concentrations via thyroid hormone effects on liver protein production. Thyroid hormone may alter SHBG production through effects on HNF4α which is known to regulate SHBG transcription [26], [27].
Many of the genes identified are involved in carbohydrate and lipid metabolism and liver function. The GCKR locus (2p23.3) encodes a protein that regulates glucokinase activity and has been associated with T2D in several ethnic populations [28], [29], [30], [31]. GCKR has been associated with metabolic and inflammatory traits including triglyceride concentrations and other lipid fractions [30], [32], fasting plasma glucose [33], [34], insulin concentrations, uric acid, c-reactive protein (CRP), and non-alcoholic fatty liver disease which are all characteristic of the metabolic syndrome and T2D [28], [35], [36], [37], [38], [39], [40], [41], [42]. The SLCO1B1 locus (12p12.1) codes for a protein, hepatocyte protein anion-transporting polypeptide 1B1, involved in liver metabolism of both endogenous and exogenous compounds [43]. Consistent with SLCO1B1's role in liver metabolism, the same SNP (rs4149056) has been associated with circulating bilirubin concentrations in previous GWAS [44]. BAIAP2L1 (7q21.3) encodes a protein important in cytoskeleton organization [45] that has been associated with the inflammatory marker CRP in patients with arthritis [46]. BAIAP2L1 is also known as IRTKS (insulin receptor tyrosine kinase substrate) which is involved in insulin receptor signaling [47] and may relate to insulin resistant states including obesity and T2D [48], [49], [50], [51], [52], [53], [54]. We conducted a targeted analysis of PPARγ, a gene that influences SHBG gene expression in the liver [1], [55] and is associated with T2D [56], [57]. Our analysis identified one significant SNP (rs2920502, p = 9.9×10−5) and a second nominally significant SNP (rs13081389, p = 0.01) at PPARγ. Some of the identified genes involved in hepatic metabolism of lipids and carbohydrates may be affect SHBG concentrations indirectly through effects on the SHBG transcription regulator HNF4α although HNF4α itself was not identified in this meta-analyses [27], [58], [59], [60].
The UGT2B15 locus (4q13.2) was significantly associated with SHBG concentrations in men but not women in this meta-analysis. UGT2B15 belongs to a family of genes (the UGT2B gene family) that code for enzymes involved in the metabolism of sex hormones through glucuronidation which allows for excretion of sex steroids through the kidney and the gut via bile excretion [61], [62], primary clearance mechanisms for sex steroids [63]. UGT2B15 is involved in the conjugation and inactivation of testosterone [64]. An association between rs293428 in the UGT2B15 locus and circulating SHBG concentrations in men is supported by a previous study demonstrating that a non-synonymous SNP in UGT2B15 (rs1902023; D85Y) is associated with serum SHBG concentrations in younger adult men [65]. UGT2B15 is thought to play a significant role in local tissue inactivation of androgens in androgen dependent prostate cancer [66], [67]. The mechanism behind the influence of genetic variants in UGT2B15 on SHBG concentrations is unknown, but one may speculate that UGT2B15 affects the local androgenic environment in selected tissues, which in turn results in regulation of SHBG concentrations.
In addition to UGT2B15, three other genes near the identified SNPs are associated with carcinogenesis, particularly in the prostate and breast. ZBTB10 (8q21.13), has been linked to breast cancer [68]. In breast cancer cell lines ZBTB10 is suppressed by ROS-microRNA27a thereby enhancing ERα alpha expression and mediating estrogen effects [17]. The ZNF652 (17q21.32) locus codes for a DNA binding protein thought to act as a tumor suppressor gene in breast cancer [69], [70], [71] that is also co-expressed with the androgen receptor in prostate cancer [72]. TDGF3, teratocarcinoma derived growth factor 3, is the only significant region identified on the X chromosome ((Xq22.3). TDGF3 is a pseudogene of TDGF1 located on chromosome 3p23-p21 that has been associated with testicular germ cell tumors [73].
This GWAS meta-analysis incorporated data from approximately 22,000 men and women from 16 epidemiologic cohorts. The overall size of the study yields power but the meta-analysis of data from different epidemiologic studies requires the inclusion of different laboratory methods. The different studies used a variety of assay methodologies to measure serum SHBG concentrations although the vast majority were immunoassays (Tables S1 and S2, Text S1) with similar methodologies. Variation introduced by the use of different SHBG assays would result in loss of statistical power and likely bias toward the null. Additionally, the majority of women were post-menopausal as ascertained by self-report in all studies (Table S1). SHBG concentrations, like testosterone, decline only slightly across the menopause [74] so adjustment for menopause status is not necessary. SHBG may also increase with ovulation and be slightly higher in the luteal versus the follicular phase of the menstrual cycle in premenopausal women, but most studies did not collect data on menstrual phase at the time of SHBG measurement so adjustment for menstrual phase was not possible [75]. Finally, individuals were not excluded based on health status, therefore some individuals with chronic conditions that may affect hepatic production of or clearance of proteins including SHBG such as liver disease, renal disease, or severe malnutrition, may have been included in this analysis.
SHBG synthesis in the liver is known to be affected directly or indirectly by estrogens, androgens and thyroid hormones and has been observed to be inversely associated with the higher insulin concentrations characteristic of insulin resistant states such as T2D [1], [6]. In summary, the results of this GWAS reflect these influences. Three regions map to proteins related to hepatic function (12p12.1-SLCO1B1 [76], 2p23.3-GCKR [77] and 10q21.3-JMJD1C [77]). In addition, 2p23.3-GCKR and 7q21.3-BAIAP2L1 [alias insulin receptor tyrosine kinase substrate (IRTKS)] are involved in susceptibility to T2D [48] and insulin signaling [47], respectively. Two signals also mapped to loci involved in thyroid hormone regulation (10q21.3-JMJD1C and 12p12.1-SLCO1B1). One signal mapped to the receptor for luteinizing hormone 2p16.3-LHCGR [20], the hormone that stimulates testosterone production. Five regions mapped to genes previously implicated in androgen and estrogen signaling (1p13.3-PRMT6 [18], 8q21.13-ZBTB10 [17], 12p12.1-SLCO1B1 [76], 15q26.2-NR2F2 [78], 4q13.2-UGT2B15 [63]).
We have combined a conventional GWAS approach with detailed additional analyses, including sex stratification, conditional analysis and imputation from 1000 Genomes. Our results demonstrate that these approaches can lead to an appreciable gain in heritable variance explained. It does however highlight the complexity of elucidating individual variant causality through statistical approaches. In addition to the extensive allelic heterogeneity at the SHBG locus, our data identify loci with a role in sex steroid hormone metabolism, which may help elucidate the role of sex steroid hormones in disease, particularly T2D and hormone-sensitive cancers.
We performed a genome wide association study (GWAS) meta-analysis of 21,791 individuals (Table S1: 9,390 women, 12,401 men) from ten observational studies. Data from an additional six studies totaling 7,046 individuals (Table S2: 4,509 women; 2,537 men) were used for validation. The proportion of variance explained was estimated in an independent study (InCHIANTI, n = 1,129). The individual study protocols were approved by their respective institution's ethics committee/institutional review board and all participants provided informed consent prior to participation. Individuals known to be taking hormonal contraceptives or hormone replacement therapy at time of SHBG measurement were excluded from analysis. Age, sex and body mass index (BMI) were included as covariates. After applying standard quality control measures, imputed genotypes were available for approximately 2.5 M SNPs. See Figure 2 for an overview of the analytic plan and the Text S1 for further information for individual studies included in this meta-analysis.
We performed a sensitivity analysis using samples from the 1966 Northern Finland Birth Cohort (NFBC1966) study to further investigate allelic heterogeneity at the SHBG locus (Text S1). The conditional meta-analysis showed evidence for up to nine signals at the SHBG locus, but it is possible that these signals could be explaining a much smaller number of causal variants in the region. Since 1000 Genomes imputation allows us to assess the genetic variation associated with a phenotype across a much denser set of markers, it increases our power to detect allelic heterogeneity within a region. Therefore, 1000 Genomes imputation was carried out on all the samples in the NFBC1966 study and forward selection was used to identify the set of SNPs that best explain the variation in the SHBG phenotype. 1000 Genomes imputation was carried out using IMPUTE2. The mean genotype probabilities for each SNP were calculated and used in the model selection step. Only SNPs 250 kb upstream and 250 kb downstream from the SHBG locus (7283453–7786700 bp) were used in the analysis. All SNPs with MAF <0.1% or an imputation quality score less than 0.4 were excluded from the analysis. In total, 1978 SHBG region SNPs measured or imputed in 4467 samples from the NFBC1966 study were used in the sensitivity analysis. Forward selection was implemented in R (version 2.13.0) using the stepAIC package to estimate the Akaikie Information Criterion (AIC), an inclusion parameter. Given the high degree of correlation between the SNPs in this region, we increased the penalty (k) on the number of terms included in the model to 12 (where it is usually two), to minimize possible over fitting. The final model included seven SNPs, adjusted for sex and BMI.
We examined potential interactions among the proteins encoded by the SHBG locus and the proteins encoded by the 11 genes (ZBT10, TDGF1, ZNF652, PRMT6, JMJD1C, GCKR, BAIAP2L1, LHCGR, SLCO1B1, UGT2B15, NR2F2) closest to the 11 identified SNPs using pathway analysis with Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) Pathways Analysis (www.string-db.org). The interactions explored by STRING include direct (physical) and indirect (functional) associations. We then expanded the analysis to examine protein interactions among the SHBG gene and the proteins encoded by 67 genes within 500 kb of the 11 identified SNPs.
We conducted targeted analysis of two strong candidate genes, hepatocyte nuclear factor-4α (HNF4α) and peroxisome-proliferating receptor γ (PPARγ). Statistical significance thresholds were set correcting for the number of SNPs tested in each gene region (±100 kb).
|
10.1371/journal.pcbi.1000917 | Differences in Nutrient Requirements Imply a Non-Linear Emergence of Leaders in Animal Groups | Collective decision making and especially leadership in groups are among the most studied topics in natural, social, and political sciences. Previous studies have shown that some individuals are more likely to be leaders because of their social power or the pertinent information they possess. One challenge for all group members, however, is to satisfy their needs. In many situations, we do not yet know how individuals within groups distribute leadership decisions between themselves in order to satisfy time-varying individual requirements. To gain insight into this problem, we build a dynamic model where group members have to satisfy different needs but are not aware of each other's needs. Data about needs of animals come from real data observed in macaques. Several studies showed that a collective movement may be initiated by a single individual. This individual may be the dominant one, the oldest one, but also the one having the highest physiological needs. In our model, the individual with the lowest reserve initiates movements and decides for all its conspecifics. This simple rule leads to a viable decision-making system where all individuals may lead the group at one moment and thus suit their requirements. However, a single individual becomes the leader in 38% to 95% of cases and the leadership is unequally (according to an exponential law) distributed according to the heterogeneity of needs in the group. The results showed that this non-linearity emerges when one group member reaches physiological requirements, mainly the nutrient ones – protein, energy and water depending on weight - superior to those of its conspecifics. This amplification may explain why some leaders could appear in animal groups without any despotism, complex signalling, or developed cognitive ability.
| Making decisions together to reach a consensus is one of the most important challenges of any society. In some communities, however, some leaders have more weight in the decisions than the other individuals. Similar rules exist in animal societies. Studies on animal groups have shown that some individuals are more likely to be leaders because of their social power or the pertinent information they possess. This leader may be the dominant one, the oldest one, but also the one having the highest physiological need. However, how may other group members have their needs satisfied if always the same individual decides? To gain insight into this problem, we build an agent-based model where group members have to satisfy different needs but the individual with the lowest reserve decides when and where to move for all its conspecifics. This simple rule leads to a viable decision-making system that satisfies all individuals and suits their requirements. However, a single individual, the one with the highest needs, becomes the leader in 38% to 95% of cases according to the heterogeneity of needs in the group.
| Social animals have to coordinate their activities in order to maintain the advantages of group living [1]–[3]. This coordination constitutes one of the major challenges of any animal society, including human beings, and arouses the interest of scientists, sociologists, and politicians [4]–[9]. Whatever the group size or the level of communication – global or local [8], [9] – several categories of group decision making have been described: a leadership process where one individual will propose or impose a decision that other group members will follow [10]–[15], and a voting process in which each individual indicates a direction, for instance, and after which the group will move in the direction of the majority [4], [16]–[18]. A group leader is usually defined as the individual initiating group movements but also as the individual coordinating individual during the group progression, and then mainly at the front of the progression [4], [8]–[13]. In different species of animals, leadership is not necessarily homogeneously distributed among group members [8]–[15]. Some individuals are more likely to become leaders thanks to specific internal or social traits increasing their probability of initiating a movement [10]–[12]. Studies of elephants [19], ravens [20], or fishes [21] have reported that some individuals may have a greater knowledge about their environment – which is the best site to eat or to drink – and these individuals have been observed leading the group more often than their conspecifics. In other species, individuals having a high social status, in terms of dominance or affiliation, also have a greater likelihood of being leaders. Probably the best known examples come from wolves and gorillas [10], [22] where the dominant male or couple is described as always deciding for the entire group. In Tonkean macaques, however, the most affiliated individuals – who are not necessarily the most dominant ones – seem to have a greater influence than their conspecifics in collective decision making [23].
However, one of the major factors influencing leadership should be the different physiological requirements of group members [10]–[12]. Such heterogeneity implies conflicts of interests between individuals that must be resolved in order to maintain group cohesion. Leaving the leadership to highly motivated individuals seems to be one compromise. Indeed, the moving decision seems to be taken by those with highest needs in fishes, zebras, and primates [8], [9], [14], [24]. Nevertheless, we still lack data on the way leaders emerge and the viability of the decision-making system concerning the entire group's satisfaction. Using a modelling approach, Rands et al. [12], [25] and Conradt et al. [26] showed that individuals with the highest nutrient requirements can be more prone to lead the group if there is an advantage to foraging together. Their studies were however restricted to pairs of individuals or to situations in which individuals faced two mutually exclusive target destinations only.
Here, we use a state-dependent dynamic model [27], [28] to determine how nutrient and social requirements can determine the synchronization of a group of n individuals, their activity budget, and the emergence of leaders. This kind of models allows us to understand how simple rules based on nutrient requirements and social factors explain synchronization between group members [27] but also segregation as shown in ungulates [28], where each individual has some requirements to satisfy (nutrient requirements such as protein, energy and water but also other social requirements and resting). We assume that if there is an advantage to being in a group, then the group members should synchronize their activities in order to stay cohesive. We assume that individuals do not know the requirements of their conspecifics (and further show that such ability may not be necessary for effective group coordination). Each individual requirement combines a reserve and a motivation that we call probability to lead. When the reserve decreases, the probability increases. At one moment, the individual with the lowest reserve – among its own needs and in comparison to other individuals – will have the highest probability to lead (these individual probabilities are compared at each time step) and will decide for the entire group on changing activity in order to fulfil its respective reserve [8], [9], [12], [25], [26]. In the next step, when the need of the previous leader is satisfied, a new leader will emerge and decide for the whole group. We applied this condition in our model in order to assess if this simple hypothesis “leading according to needs and deciding for all the group” is viable and if so, how the leadership will be distributed among group members less or more heterogeneous in their needs.
We first tested a group of two individuals, A and B, with two needs, x1 and x2. We set three conditions: 1) needs are equal (a1 = a2 = b1 = b2), 2) needs of each individual are different but their sums are equal between individuals (a1>a2; b1>b2; a1+a2 = b1+b2) and 3) the sum of needs for individual A are always superior to the one of individual B (a1+a2>b1+b2). Values of needs for each tested group of each condition are detailed in table 1. We tested 10 groups for each condition. Results show that the decision system is viable, no individual dies, i.e., no individual has needs not met (reserves go to 0), whatever the tested condition. When sums of needs are equal between individuals (conditions 1 and 2), the leadership (proportion of decisions, i.e., initiations per individual) is equally distributed between the two individuals (Fig. 1A), even if, at each time step, one individual is the leader and the other one is the follower according to the reserves' difference. This result is similar to the one of the paper of Rands et al. [25] where individuals are identical. However, when the sum of needs is superior for one individual (Fig. 1B), this one becomes the leader of the pairs of individuals and the other individual becomes a follower almost all the time (Kolmogorov-Smirnov test, P<0.0001). The leadership difference between individuals increases with their relative difference of needs in a logarithmic way (curve estimation test: R2 = 0.96, P<0.00001; Fig. 1C). This result is similar to the one of Rands and colleagues [12]: leaders emerge when individual reserves differ.
In a second step, we used data coming from animals in order to validate our model and to study emergence of leaders in larger groups.. According to needs of macaques, animals were divided into five categories: adult males, adult cycling females, lactating females, subadults, and juveniles. An individual has five requirements to satisfy: water, protein, energy, resting, and socializing [29]–[34]. The nutrient requirements of an individual (water, protein and energy) depend on its body mass whilst social and resting needs did not [31]–[34]. We chose to include social activity in the model because many social species spend time maintaining their relationships and group cohesion [31]–[35]. Group composition (table 2) and individual characteristics (table 3) are based on data on macaques and are detailed in the method section. We tested 10 groups of 5, 10 and 20 individuals with same needs (individuals of the same category and with the same body mass) and 10 ones with different needs (individuals of both different categories and different body masses).
Simulations showed that the system – leadership by those in need – is sustainable in groups of 5, 10 and 20 individuals. All individual requirements are satisfied at the end of simulations, whatever the group composition. Moreover, the group activity budget is fairly similar to the activity budget of wild primate groups (27.3±1.7% of time devoted to moving, 33.8±1.7% to foraging, 21.7±0.7% to resting, and 17.2±3.1% to socializing). All individuals could become leaders but the distribution of the leadership proportion is not the same according to the requirements' heterogeneity (equal or different needs; Kruskal-Wallis test, P<0.001). In groups with similar needs, the proportion of leadership differs weakly between individuals and is about 10% per individual. The relation between leadership proportion and rank (i.e., individuals were ranked from the most frequent leader to the less frequent one) is linear (linear curve estimation test: R2 = 0.92, F1,8 = 93.05, P<0.00001, y = −0.0006x+0.1339). The leadership is about 14% for the individual that decides the most and 7.6% for the individual that decide the least (Fig. 2A). This result corresponds to the equiprobability of being leader per individual (proportion divided by the number of individuals per group). On the other hand, the leadership is not equally distributed in heterogeneous groups. The relation between the proportion of leadership and individuals is exponential (exponential curve estimation test: R2 = 0.83, F1,8 = 38.07, P = 0.0002, y = 3.5727e−4.602x, Fig. 2B), with one individual being responsible for 38% to 95% of decisions per group, while some individuals decide only in 0.0003% to 0.0007% of cases per group. We obtained the same relationship with groups of 5 (exponential curve estimation test; R2 = 0.97, F1,3 = 12.81, P<0.00001, y = 0.9825x−3.207, Fig. 3A) and 20 individuals (exponential curve estimation test; R2 = 0.96, F1,18 = 498.95, P<0.00001, y = 11.48x−4.86, Fig. 3B).
We compared this unequally distributed leadership to the requirements of individuals in order to understand how so many differences can emerge in heterogeneous groups. We calculated the relative difference in requirements (corresponding to the highest probability to lead) between each leader and other individuals. The relationship between the leadership and this difference in requirements follows a sigmoid curve ( with a threshold S of 1.37 and a minimal n value of 30; curve estimation test: R2 = 0.71, F1,108 = 269.72, P<0.00001; Fig. 4A). The n value represents the sensitivity of the process. The higher the n value is, the more sensitive the process is (quick and sudden transition between the two states). In our context, this means that one individual becomes the most frequent and prominent leader of a group as soon as one of its requirements exceed about 137% of those of one of its conspecifics. This transition between equally distributed leadership and one exclusive leader is highly non-linear, given the n value we observed. The same sigmoid law is observed between the proportion of leadership and the body mass of individuals (sigmoid curve estimation test: R2 = 0.66, F1,108 = 205.73, P<0.00001; Fig. 4B). When the mass of an individual is more than 170% (S = 1.7, n = 30) of those of its conspecifics, this individual is the main group leader. Except for lactating females, requirements and then leadership are related to body mass in about 60% of cases. The rest of the decisions concern resting and socializing and are not related to mass. We obtain similar results for groups of 5 (sigmoid curve estimation test: R2 = 0.71, F1,53 = 111.52, P<0.00001, , Fig. 5A) and 20 individuals (sigmoid curve estimation test: R2 = 0.16, F1,218 = 42.13, P<0.00001, , Fig. 5B). For 8 out of 10 groups of 20 individuals, 4.6±2.2 group members were never leader. They were satisfied by following their conspecifics.
Leading by those highest in need resembles the results obtained by Rands et al. [25], where the individual with the lower reserve spontaneously becomes the leader. Moreover, a recent study by Conradt et al. [26] showed that a small minority of individuals with strong needs are more prone to lead the group than a larger majority of individuals with few needs. However, it is the first time that a threshold [2], [18], [36] has been demonstrated concerning the emergence of leadership. The decision-making system implies high differences in leadership proportion whilst relatively small differences are observed in the requirements of individuals. The threshold we obtained in this study is probably dependent on 1) the group structure of primates (one or a small number of males compared to the other categories) [35] and 2) to the physiology of primates [31]–[34]. Indeed, in primates, and especially in macaques, a sexual dimorphism exists and males may reach a mass 150 to 200% superior to the one of females.
Several authors suggested that dominant individuals are the only leaders in several species [9]–[12], [14], [22], [23]. The dominance is however strongly correlated to the body mass and then to the nutrient requirements of animals [10]. This indirect effect of dominance on leadership, through the needs and then the probability to initiate a movement, needs to be taken into account in subsequent studies testing dominance effects. For instance, two field studies in baboons showed that the main leader – the individual initiating most of movements – was the dominant male. However this male is also certainly the biggest individual in the group. In the study of Stueckle and Zinner [36], the four males of the group, bigger than females, are the ones initiating the most of movements (Fig. 6). Moreover, the distribution of leadership also follows an exponential as the one in the study model. We may suggest that the slope of this exponential distribution of leadership will be less or more important according to the group composition. This slope would be around 0 when the group is homogeneous and increases with group heterogeneity.
The non-linear differences in leadership among group members eventually emerge from two simple rules: individuals need to remain cohesive and the individual with the lowest reserve at one moment decides for the group [2], [3], [24]–[26]. Mechanisms of coordination and cohesion do not need complex signalling or complex cognitive ability [2], [3], [13], [24]. The emergence of a unique leader may also occur when decisions are not necessarily imposed on other group members but because other individuals do not express the necessity to move or to make a decision. An individual becomes a leader because its conspecifics decide to follow it [8], [9]. This outcome may make important contributions to our understanding of decision making in animal and human societies.
The model was developed in Netlogo 3.15 [37]. The model and model's procedures can be found in the supplementary material “Dataset S1”. One time-step in the simulation represents one minute. We defined the probability to lead α for the requirement A and the individual i as:The probability to lead for the individual i is:In this way, the probability to lead can vary between 1 (highest probability to lead, weakest reserve) and 0 (weakest probability to lead, highest reserve). Each reserve is bounded by a maximum above which each group member cannot gain further reserves and a minimum at which each group member is assumed to die if it is reached. At each timestep (equal to one minute), each reserve of each individual decreases (i.e., expenditure) depending on the individual category and the current activity. This reserve decrease will increase the individual probability to lead. In order to fulfil this reserve, the individual should have to carry out the corresponding activity (i.e., intake). This gain may be done by becoming a leader or by following the leader. We implement optimal foraging decisions in the model: when an individual decides to forage, it will forage until its reserve has been fulfilled. After the end of each activity period, the individual with the highest probability to lead Pi becomes the new leader. Individuals have a walking speed of 0.4m.s−1.
The group environment is two-dimensional environment of 96×96 connected cells. Each cell represent one meter. Each cell has four immediate neighbours and the sides of the arena were joined to form a torus. The number of areas where animals fulfil their reserves is two for the first model with two individuals having two needs and four for the model from 5 to 20 individuals having five needs (see details below). At the start of a simulation, individuals are at the same distance of each area (i.e., at the middle of the torus). According to the distribution of areas inside the torus, groups have a travel distance between two areas ranging from a minimum of 25 meters to a maximum of 75 meters. This range fits with travel distances in primate species of similar body mass and similar group size [4], [38]–[41]. Positions of areas were fixed in our model but this does not affect results since variability among needs – what is the highest need and the weakest one – is much more important between individuals and groups. This means that the areas corresponding for instance to the two highest needs for an individual are not always the closest ones. There is no intragroup competition in this model: all individuals can occupy the same area. We run 1000 simulations for each group. A simulation stops when one reserve of one individual reaches 0 or after 90 days.
The two individuals have two needs and thus two daily requirements. Values of these requirements for each condition and each individual are described in table 1. We tested ten different groups for each condition. Expenditures of each reserve are 0.07±0.035 units.min−1. Intakes are 10 units.min−1. The environment is composed of two areas, one for each requirement. Individuals have to move to the respective area to fulfil each reserve.
According to data in macaques, the daily protein requirement is estimated to 2.54g.day−1.kg−1, daily energy requirement to 351.7Kcal.day−1.kg−1, and daily water requirement to 0.24ml.KJ−1, except for lactating females for which these requirements are higher than the ones of non lactating females (about 125% for proteins and 200% for energy and water of requirements of non lactating females) [31]–[34]. Social and resting times are not dependent on body mass. Individual expenditure per need and activity is described in table 4. Details about individual intake rate per need are in table 5.
The environment is composed of four areas: one area for foraging for proteins, one area for foraging for energy, one waterhole, and one resting site [42]. When individuals need energy, proteins, or water, they have to move toward the respective areas. Until the group is in a specific activity among the five ones (eating proteins, eating energy, drinking water, resting or socializing), each individual gains a certain amount of the requirement according to its category (table 5). Concerning resting, individuals need to go to the resting site for the night (at the 720th time-step and for 720 time-steps), but during the day they can rest in any area. The same rule applies to socializing. Concerning resting and socializing activity, we fixed a minimal period of 5 minutes for doing these activities.
Differences in leadership between individuals were tested using a Kolmogorov-Smirnov test for groups of 2 individuals and a Kruskall-Wallis test for groups from 5 to 20 individuals. The relations between the proportion of leadership and differences in needs or mass were determined through a curve estimation test. We compared observed curves to exponential, linear and sigmoid ones. Only theoretical curves best fitting with observed data are indicate in results. Analyses were performed in SPSS 10.00. α was set at 0.05. Means were ± S.E.M.
|
10.1371/journal.pntd.0001035 | Triatoma dimidiata Infestation in Chagas Disease Endemic Regions of Guatemala: Comparison of Random and Targeted Cross-Sectional Surveys | Guatemala is presently engaged in the Central America Initiative to interrupt Chagas disease transmission by reducing intradomiciliary prevalence of Triatoma dimidiata, using targeted cross-sectional surveys to direct control measures to villages exceeding the 5% control threshold. The use of targeted surveys to guide disease control programs has not been evaluated. Here, we compare the findings from the targeted surveys to concurrent random cross-sectional surveys in two primary foci of Chagas disease transmission in central and southeastern Guatemala.
Survey prevalences of T. dimidiata intradomiciliary infestation by village and region were compared. Univariate logistic regression was used to assess the use of risk factors to target surveys and to evaluate indicators associated with village level intradomiciliary prevalences >5% by survey and region. Multivariate logistic regression models were developed to assess the ability of random and targeted surveys to target villages with intradomiciliary prevalence exceeding the control threshold within each region. Regional prevalences did not vary by survey; however, village prevalences were significantly greater in random surveys in central (13.0% versus 8.7%) and southeastern (22.7% versus 6.9%) Guatemala. The number of significant risk factors detected did not vary by survey in central Guatemala but differed considerably in the southeast with a greater number of significant risk factors in the random survey (e.g. land surface temperature, relative humidity, cropland, grassland, tile flooring, and stick and mud and palm and straw walls). Differences in the direction of risk factor associations were observed between regions in both survey types. The overall discriminative capacity was significantly greater in the random surveys in central and southeastern Guatemala, with an area under the receiver-operator curve (AUC) of 0.84 in the random surveys and approximately 0.64 in the targeted surveys in both regions. Sensitivity did not differ between surveys, but the positive predictive value was significantly greater in the random surveys.
Surprisingly, targeted surveys were not more effective at determining T. dimidiata prevalence or at directing control to high risk villages in comparison to random surveys. We recommend that random surveys should be selected over targeted surveys whenever possible, particularly when the focus is on directing disease control and elimination and when risk factor association has not been evaluated for all regions under investigation.
| Chagas disease is a vector-borne parasitic zoonosis endemic throughout South and Central America and Mexico. Guatemala is engaged in the Central America Initiative to interrupt Chagas disease transmission. A major strategy is the reduction of Triatoma dimidiata domiciliary infestations through indoor application of residual insecticides. Successful control of T. dimidiata will depend on accurate identification of areas at greatest risk for infestation. Initial efforts focused primarily on targeted surveys of presumed risk factors and suspected infestation to define intervention areas. This policy has not been evaluated and might not maximize the effectiveness of limited resources if high prevalence villages are missed or low prevalence villages are visited unnecessarily. We compare findings from the targeted surveys to concurrent random surveys in two primary foci of Chagas disease transmission in Guatemala to evaluate the performance of the targeted surveys. Our results indicate that random surveys performed better than targeted surveys and should be considered over targeted surveys when reliability of risk factors has not been evaluated, identify useful environmental factors to predict infestation, and indicate that infestation risk varies locally. These findings are useful for decision-makers at national Chagas Disease control programs in Central America, institutions supporting development efforts, and funding agencies.
| In Guatemala, nearly 4 million individuals are projected to be at risk for infection with Trypanosoma cruzi, the causative agent of Chagas disease, with approximately 30,000 new cases a year and a prevalence of 730,000 [1], [2]. The estimated prevalence and annual incidence is more than double any other country in Central America and is substantially greater than that observed in Mexico [1], [2]. Based on the results of the national survey of triatomine populations conducted from 1995–8, the principal focus of transmission is considered to be in the southeastern and central departments of the country where the prevalence of triatomine vectors [3], the estimated human population at risk for Trypanosoma cruzi infection [3], and the infection rate of triatomine vectors with T. cruzi [4] is greatest[1]. This is also the region where the vector Triatoma dimidiata (Latreille 1811) is most abundant [3], [4], [5].
The Guatemalan National Ministry of Health (GNMH) is engaged in the Central America Initiative to interrupt Chagas Disease transmission (IPCA) [6], [7], [8], and Guatemala is the country with the most progress to date [9]. All available information indicates that Rhodnius prolixus has been eliminated (GNMH communication) and populations of the indigenous T. dimidiata have been reduced in the domestic environment three to nine fold [10], [11]. However, since T. dimidiata is a native species also occurring in the peridomestic and sylvatic environments, elimination is virtually impossible [2], [12], [13], [14]. Therefore, the goal is to reduce and maintain T. dimidiata village level intradomiciliary prevalence and colonization (nymphal intradomiciliary prevalence) below 5% [1], [2], [6], [7], [8], [11], [15].
Vector control relies primarily on the intradomiciliary application of residual insecticides [16]. For the current control program, third-generation synthetic pyrethroids, including beta-cyfluthrin (12.5% suspension concentrate [s.c.], at 25% active ingredient [a.i.]/m2), cyfluthrin (10% wettable powder [w.p.], at 50 mg a.i./m2), delatamethrin (10% s.c. or 5% w.p. at 25 mg a.i./m2), and lambda-cyhalothrin (10% w.p. at 30 mg a.i./m2) (GNMH communication), were used based on market availability [17]. The current policy for selecting villages to spray entails a 5% intradomiciliary prevalence threshold but relies on targeted surveys of presumed risk factors and suspected infestation [11], [15], namely “villages suspected of being infested with R. prolixus or T. dimidiata, where infestation was reported or in rural areas where the majority of the houses are constructed with mud walls and/or thatched roofs” [15]. However, if villages with low prevalences are visited unnecessarily, or villages with high prevalences are missed, such a policy may not necessarily maximize the effectiveness of limited resources. In a resource limited setting, developing a rational control program to sustain T. dimidiata village intradomiciliary prevalence below 5%, will depend upon ensuring that control efforts are targeted to villages with the highest risk of infestation.
From 2000–3, GNMH, the Japanese International Cooperation Agency (JICA), and the Universidad del Valle de Guatemala (UVG) with other collaborating instituions undertook a series of targeted and random surveys to assess T. dimidiata prevalance prior to vector control [1], [11], [15], [18]. This study makes use of the data gathered in the central department of Baja Verapaz and southeastern department of Jutiapa to compare the effectiveness of random and targeted surveys in determining villages at high risk for T. dimidiata infestation in these two regions. Specifically, our objective was to evaluate the capability of the random and targeted survey methods in directing control to villages at greatest risk of infestation by comparing the ability of environmental and/or domiciliary risk factors to predict intradomestic prevalence >5% by survey and department.
Triatoma dimidiata intradomiciliary prevalence data at the village level for the departments1 of Baja Verapaz and Jutiapa from 2000–3 were obtained from randomized cross-sectional pre-spray surveys implemented by UVG [10] and from targeted cross-sectional pre-spray surveys performed by GNMH [11], [18]. These departments are positioned within two principal regions of T. dimidiata infestation. Baja Verapaz is located in the temperate and subtropical dry forests [19] of central Guatemala, 89.93°–90.81°W and 13.74°–14.56°N, encompassing an area of 2864 km2. Jutiapa is positioned in the subtropical moist forest [19] in the southeast, 89.50°–90.30°W and 13.74°–14.56°N, covering an area of 3318 km2. The geographic distribution of villages surveyed by department and study is illustrated in Figure 1. Here, department prevalence refers to the proportion of villages within each department that are intradomiciliary infested with T. dimidiata, and village prevalence refers to the proportion of infested domiciles within each surveyed village.
Specific details of data collection and survey design have been previously published [1], . In brief, both surveys analyzed here are subsets of larger studies aimed at determining triatomine prevalence in central and southeastern Guatemala prior to a vector control campaign. Baja Verapaz and Jutiapa were selected here due to similarities in the broad geographic coverage of sampled villages, the presence of significant T. dimidiata infestation with limited R. prolixus infestation [3], [4], [18], [20], and the locations of the departments in two different regions, central and southeastern Guatemala. Moreover, the departments were analyzed separately as the vector surveys were administered at the department level [1] and due to the location of the departments in two different Holdridge Life zones. Baja Verapaz occurs in the subtropical and warm temperate dry forest and Jutiapa occurs in the subtropical moist forest [21].
The random data set was derived from a cross-sectional survey supported by the Tropical Disease Research and Training program (TDR), World Health Organization no. 990545 and the Centers for Disease Control and Prevention CoAg U50 CCU021236 by UVG in collaboration with GNMH from 2000–3 [10]. In each municipality, villages and domiciles were selected randomly [10]. All eight municipalities in Baja Verapaz and 16 of 18 municipalities in Jutiapa were surveyed. Within these municipalities, georeferenced data was obtained from 79 villages and 1021 domiciles in Baja Verapaz and 162 villages and 2215 domiciles in Jutiapa. Entomological evaluation was conducted using an abbreviated man-hour collection method [3]. For each domicile selected, the intradomestic and surrounding peridomestic environments were surveyed manually for triatomines by two entomology technicians for 15–30 minutes, as determined subjectively by the size of the house [10].
The targeted data set was derived from cross-sectional entomological surveys carried out by GNMH in collaboration with JICA from 2000–3 [1], [11], [18]. Domiciles were selected from a sampling frame that excluded villages sampled in the random survey. Within Baja Verapaz, all eight municipalities were surveyed while 14 of 18 were examined in Jutiapa. In contrast to the random survey, study villages were targeted in rural areas on the basis of anecdotal surveys, suspected infestation, previous infestation, or presumed risk factors, e.g., domiciles with walls made of mud and/or roofs constructed of thatch [11], [15]. Georeferenced data were obtained from 262 villages and 5306 domiciles in Baja Verapaz and 244 villages and 2954 domiciles in Jutiapa. Entomological evaluation was also conducted by an abbreviated man-hour collection method [3]. The intradomestic and peridomestic environments of selected domiciles were searched manually for triatomines for 30 minutes by one entomology technician and for 15 minutes by two technicians [11]. These findings were later used by GNMH to target pyrethroid spraying to domiciles and peridomestic annexes in villages with intradomiciliary prevalences >5% [1], [10], [11], [15], [18].
Environmental and socioeconomic data were obtained from multiple sources and are described in Table 1. Covariate and georeferenced infestation data were imported into the GIS TNTmips 2008:74 (Microimages, Lincoln, NE). Layers were processed and linked geographically. With the exception of land cover, environmental covariate values were defined using the geographic coordinates for each village. For land cover, the proportion of each land cover class (forest, grassland, cropland, wetland, and settlement) within a 2 km buffer of each village was determined. Domiciliary construction data were then summarized by calculating the proportion of each domicile construction material per village. All data were then extracted by village and exported for statistical analysis. Data were displayed and mapped using ArcView GIS v. 9.2 (Environmental Systems Research Institute, Inc., Redlands, CA).
Analysis of T. dimidiata pre-spray prevalence data was limited to those villages where at least five domiciles were surveyed. Similarities in the geographic distribution of villages between the two studies were maximized by excluding villages from one study when their distance to the closest village in the opposing study exceeded five kilometers. For the remaining villages, descriptive statistics of T. dimidiata village and department level prevalences were summarized by study and department.
Analyses of risk factors associated with T. dimidiata intradomiciliary prevalence at the village level for each department and study were carried out using univariate and multivariate logistic regression. First, univariate logistic regression models for grouped data were fitted to each of the grouped climatic variables (land surface temperature, normalized difference vegetation index, middle infrared reflectance, and relative humidity) to identify covariates in each category that best discriminated village prevalence. For ease of interpretation and direct comparison of climate characteristics between studies, variables were selected from the analyses of the UVG random data set in each department. Variables with a Wald's P>0.05 were excluded from further analyses due to the large number of significant covariates. The best fit model for each category was then selected on the basis of its Akaike weight (wi). Although the number of parameters for each model was the same in this investigation, the statistic provided a simple and easily interpretable measure for model comparison [22], [23].
The environmental and domiciliary risk factors associated with T. dimidiata village prevalence >5% was investigated by univariate logistic regression for each study by department. The outcome variable was defined by village as T. dimidiata intradomiciliary prevalence ≤ or >5%. Explanatory variables included climate variables selected from the discriminative univariate analyses, the remaining environmental covariates (elevation, precipitation, and land cover), and all domiciliary construction covariates. A logistic regression model was fitted to each covariate to define the odds of infestation associated with each potential risk factor.
Predictions of the probability of village prevalence >5% were then made by fitting a series of multivariate logistic regression models using a jackknife procedure, whereby a single village was excluded and an estimate of its predictive probability was made using the remaining data [24], [25]. This method maximizes the data used to estimate a villages predictive probability and allows for model validation using independent data [24]. All significant covariates from the logistic regression models were used to fit multivariate models. Predictive models of environmental and domiciliary covariates for each study by department were generated individually and together. Diagnostic statistics were generated to compare model accuracy. The area under the receiver-operator curve (AUC) was calculated to compare overall model performance and kappa (κ), sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated across the range of predicted probability thresholds. All statistical analyses were conducted in Stata/IC version 10 (Stata Corporation, College Station, TX, USA).
The geographic distribution and intradomiciliary prevalences of villages selected for analysis are shown by department and study in Figure 1. In Baja Verapaz, villages in all eight municipalities were incorporated into the analysis of both surveys and included 894 domiciles in 72 villages from the random study and 4403 domiciles in 212 villages from the targeted study, representing 16.86% and 49.65% of villages, respectively, and 66.51% of villages overall (n = 427). Village prevalence of T. dimidiata was highest in the northwest and southern regions of the department. Department prevalence was highest in the random survey at 51.4% (95% CI 39.9–62.9), but not significantly different from the targeted survey with a prevalence of 39.2% (95% CI 32.6–45.7). In contrast, village prevalence was significantly higher in the random survey (13.0%, 95% CI 10.9–15.2) than in the targeted survey (8.7%, 95% CI 7.8–9.5).
T. dimidiata was distributed throughout Jutiapa with village prevalences highest in the central and southern regions. In the random study, 1919 domiciles in 138 villages and 16 municipalities were used for analyses, while in the targeted study, 2243 domiciles in 108 villages and 14 municipalities were used for analyses, representing 17.95% and 14.04% of villages, respectively, and 31.99% of villages overall (n = 769). Again, department prevalence was not significantly different between the random (68.8%, 95% CI 61.1–76.6) and targeted (62.0%, 95% CI 52.9–71.2) surveys, but village prevalences were significantly higher in the random (22.7%, 95% CI 20.9–24.6) than in the targeted (6.9%, 95% CI 5.9–8.0) surveys.
The grouped climate variables that best explained T. dimidiata village prevalence are presented in Table 2. These covariates were used in all subsequent analyses.
Table 3 shows the significant results of the environmental risk factor analyses for village prevalence exceeding the 5% control threshold for each survey and department. For Baja Verapaz, the significant environmental risk factors were the same for both survey types and similarly describe the odds of infestation. The magnitude of the observed effect of each covariate with the exception of annual precipitation (equal impact) was greatest in the random study. An increase in the average daytime LST, average MIR, and proportion of cropland within a 2 km buffer of villages were associated with an increase in the risk of infestation. In contrast, minimum NDVI, minimum RH, and the proportion of evergreen forest within a 2 km buffer were associated with a decrease in the risk of infestation. Annual precipitation and elevation had weak negative effects.
In Jutiapa, fewer environmental risk factors were significant in the targeted study than in the random study. The direction of the relationships of similar significant risk factors in both studies was the same. As with the relationships of the covariates in the Baja Verapaz studies, the magnitude of the observed effects was greatest in the random study but not significantly different as the confidence intervals overlapped. For both studies, the average NDVI had a substantial positive effect on the risk of infestation, while the odds of infestation were negatively associated with the average MIR. In addition, the proportion of grassland with in a 2 km buffer of an infested village and the maximum RH were associated with an increased risk of infestation in the random study. Moreover, the average daytime temperature, and proportion of cropland and settlements within a 2 km buffer of infested villages were associated with a decreased risk of infestation.
Significant domicile construction risk factors associated with village prevalence >5% are shown in Table 4. Fewer villages contained data on domicile construction materials than environmental covariates in each study and department. In Baja Verapaz, 64 of 72 villages in the random survey and 160 of 212 villages in the targeted survey had corresponding construction data, while in Jutiapa 123 of 138 villages in the random survey and 89 of 108 villages in the targeted survey had data on domicile construction covariates. The effect of similar domicile construction materials in both departments was consistent among studies. Risk was higher in adobe walled domiciles and lower in aluminum roofed domiciles in Baja Verapaz. In Jutiapa, risk was higher in domiciles with dirt floors and roofs of aluminum or tile and lower in domiciles with floors made of clay tile or cement.
In both departments, village prevalence in each study was often associated with different risk factors. For example, the targeted survey in Baja Verapaz found an increased risk associated with tile roofed domiciles that was not detected in the random study. Moreover, the random survey in Jutiapa detected a series of associations with wall materials not observed in the targeted survey. In particular, walls constructed of stick and mud or palm and straw were associated with considerable increases in the risk of infestation. Brick and block walls had marked protective effects. Interestingly, the direction of the effect of similarly significant materials such as aluminum and tile roofs contrasted between departments.
A summary of the performance of the multivariate logistic regression models ability to predict village prevalence >5% is presented in Table 5. Models were constructed using villages with data for both environmental and domicile construction covariates to allow for direct comparison. The area under receiver-operator curve (AUC) is the best measure of a model's overall discriminative ability [25], [26]. With the exception of the domicile construction material model in Baja Verapaz, the random models for both departments had reasonably good discriminative capacity and performed significantly better than the corresponding targeted models. All targeted models had poor discriminative capacity. Moreover, the environmental and combination models in the Baja Verapaz random surveys had similar predictive power and performed significantly better than the domicile construction material model. In the Jutiapa random surveys, no significant difference in predictive performance was detected between models.
κ, sensitivity, specificity, PPV, and NPV all vary with the selection of the predicted probability threshold. The maximum κ obtained for each model is reported in Table 5 and the remaining accuracy measures are calculated using the corresponding threshold. All models from random surveys, with the exception of the domicile construction material model in Baja Verapaz, performed significantly better than chance alone. With regard to the random surveys, predictions based on environmental covariates had the greatest accuracy in Baja Verapaz, and environmental and combination models had similar and greater accuracy than domicile construction covariates in Jutiapa.
Sustained control of T. dimidiata depends on the accurate identification of areas at greatest risk of infestation in order to efficiently target limited resources. In their efforts to eliminate Chagas disease from Guatemala, vector control initiatives have relied on targeted surveys of villages with presumed risk factors or suspected infestation [11], [15], [16], however, their performance has not been evaluated. The data sets analyzed here afforded a unique opportunity to compare the abilities of random and targeted baseline cross-sectional surveys of T. dimidiata village prevalence conducted concurrently in time and space and resulted in several important findings relevant to T. dimidiata vector control: 1) random surveys performed just as well if not better than targeted surveys at defining the risk of T. dimidiata infestation, 2) intradomiciliary and environmental risk factor associations with T. dimidiata prevalence >5% varied with geographic location, 3) environmental risk factors provide additional insight into the intradomiciliary risk of T. dimidiata prevalence exceeding the control threshold, and 4) predictive modeling has a role to play in directing T. dimidiata control in Guatemala if data sets are appropriately defined and expectations realistic. To our knowledge, this is the first study to compare targeted and random surveys for T. dimidiata and has implications for T. dimidiata control in Guatemala and Central America.
The failure of the targeted surveys to detect higher department and village prevalences than random surveys was surprising. These findings illustrate that the methods used to focus targeted surveys were not any better than random sampling at determining villages at greatest risk for T. dimidiata infestation. Therefore, presuming risk factors and infestation was inadequate and when initiating a program, efforts should favor risk factor evaluation and validation prior to targeting surveys or favor random sampling, as the results could reflect insufficiently defined risk factors and/or the assumption of geographic similarity in risk factor effect. Although, the findings could also be attributed to greater experience and expertise among UVG surveyors who conducted the random surveys [10], [18].
The analysis of the intradomiciliary and environmental risk factors further supports the notion that the poor performance of the targeted surveys resulted at least in part from insufficiently defined risk factors and geographic heterogeneity in their effect. The limited ability of the presumed risk factors is illustrated by our ability to detect further robust relationships with additional indicators in the analysis of the targeted survey data. Moreover, many of the risk factors contrasted in their significance and the direction of their effect between departments. Even the presumed risk factors contrasted in their significance between departments. Walls of adobe had strong positive association with T. dimidiata village prevalence exceeding the control threshold in Baja Verapaz only, while walls of stick and mud were significant in Jutiapa only. The lack of a significant association with thatch roofs in both surveys and departments likely reflects the inclusion of this risk factor to aide in the targeting of R. prolixus [2].
Particularly interesting was the contrasting relationship between tile roofs and infestation exceeding the control threshold in the departments. Tile roofs had a protective effect in Jutiapa but were associated with increased risk in infestation in Baja Verapaz. A similar increased risk was detected in Costa Rica where it was suggested that the presence of spare roofing tiles in the peridomestic environment provided suitable habitat for T. dimidiata [27]. Peridomestic surveys associated with the targeted study in Baja Verapaz found established T. dimidiata populations, although specific peridomestic environments were not reported [18]. These findings suggest the potential for roofing tiles to play a similar role in Baja Verapaz. Peridomestic populations are also present in Jutiapa [10] but were not reported here. The protective effect could indicate tile roofs in this region are associated with improved living conditions, thus, limiting intradomestic populations. In addition, previous studies in Jutiapa found no direct association between intradomestic and peridomestic infestation [10], indicating that spare roofing tiles in the peridomestic environment may be of little significance to intradomestic T. dimidiata populations in Jutiapa. More detailed studies are needed to clarify the variation of risk factors in different ecological settings.
Moreover, the analysis of the environmental covariates also illustrated the geographic heterogeneity in risk factor association with T. dimidiata infestation >5% and indicated their potential value as indicators of infestation exceeding the control threshold. For example, villages with higher temperatures, increasingly barren landscapes, and more cropland were associated with increases in prevalence above the threshold in Baja Verapaz, while in Jutiapa an increase in vegetated landscapes, the proportion of grassland, and maximum RH were associated with increased risk of infestation. Future surveys should evaluate the inclusion of environmental risk factors as an aide in focusing control efforts. Furthermore, the observed geographic heterogeneity of both domiciliary and environmental risk factors illustrates the need to evaluate risk factors prior to use in a particular geographic location and the risk in extrapolating findings beyond the geographic limits for which they were defined. This observed heterogeneity is even more important in light of recent molecular studies suggesting that T. dimidiata in Guatemala represents a geographically diverse species complex [28], [29] with one study elevating a member to specific status [28].
The findings from the predictive models indicate the potential for this type of analysis and risk mapping to aide in directing T. dimidiata control to regions at greatest risk as well as support the findings discussed above with regard to the abilities of the random surveys and potential value of environmental covariates. The reasonably high sensitivities and PPV's among the best performing models from the random surveys in both departments indicate marginal resource loss when applying control measures. Similarly, the respectable specificity and NPV's suggest that the number of positive villages missed would be moderately low. Moreover, the performance of the targeted surveys suggest that they might have a limited role to play in generating predictive models if risk factors are adequately defined first and sensitivity and PPV are reasonably good in targeting high risk villages. Although, one would have to accept a significant number of positive villages would be excluded from control due to the expected low specificity and NPV.
Also notable among the results were the performance of the environmental covariates in predicting risk of T. dimidiata prevalence >5%. The predictive performance of environmental models was just as good if not significantly better than domicile construction material models. As mentioned previously, this could relate to insufficiently defined risk factors and/or geographic heterogeneity in their effect. In addition, it could be that the association with environmental covariates is related to the peridomestic populations in these regions, implying that peridomestic populations give rise to intradomestic populations or are in constant movement from one environment to the other. However, it might also be that the environmental conditions that are present in a region dictate the domicile construction materials used and represent confounding relationships with existing covariates and subsequently the type of construction defines the temperature and relative humidity inside the domicile. This could explain why the predictive models combining both environmental and domicile construction risk factors failed to improve overall model performance. Future models might be improved by the inclusion of intradomiciliary physical variables such as temperature and relative humidity.
As with any study, it is important to point out the limitations that exist. First, the targeted sample is biased by the exclusion of villages sampled in the random survey. Differences in our results could reflect differences in the villages sampled, although, we tried to account for significant variation by comparing geographically similar villages. Secondly, the findings are relevant to surveys conducted by the man-hour collection method, which is labor intensive with small reward and likely varies with expertise and experience [30]. Other collection methods could be less biased by experience, more consistent and efficient, and better able to define risk factors. Thus, the lack of the results could reflect variation in the ability to adequately detect bugs and not the absence of bugs and their associations with the risk factors. In addition, neither study was designed with our analysis in mind and therefore doesn't allow for optimal comparison. Future studies could control for this by selecting villages from the same sample frame, choosing the same number of domiciles in each village to survey, and conducting surveys with similarly experienced technicians. In addition, a true comparison of survey effectiveness should balance scientific abilities against their cost, with decisions made accordingly.
Nonetheless, the findings from our study lead us to several recommendations for T. dimidiata control in Guatemala and Central America. First, a priori knowledge, a prerequisite for targeted surveys, was not reliable for T. dimidiata surveys in Guatemala. Random surveys performed just as well if not better than targeted surveys, and have the additional benefit of risk factor detection, resulting from increased sample heterogeneity. Therefore, random surveys should be considered over targeted surveys if the reliability of the risk factors used to target surveys has not been evaluated. Secondly, risk factors for T. dimidiata infestation should be characterized for a particular geographic location through proper epidemiological investigation. One should keep in mind that the risk of extrapolation error increases as the distance from the source from which it was defined increases [31]. Furthermore, the role of environmental risk factors should be considered in addition to traditional intradomiciliary construction risk factors when investigating the risk of T. dimidiata infestation. Finally, our results indicate that predictive modeling has a role to play in targeting T. dimidiata control as long as the surveillance data is appropriately defined and/or model error is acceptable. It should be stressed that random surveys are not simply a luxury but an investment in programs. Future surveys should weigh their benefits as well cost when initiating a vector control program.
In conclusion, sustained control of T. dimidiata will depend on accurate and thorough epidemiological investigation. It is essential that the sample surveys on which decision making is based are evaluated to ensure that policy is not formed blindly and resources are not wasted. Here we show that a priori knowledge was not reliable in defining T. dimidiata risk in Guatemala. The random survey performed just as well if not better than the targeted survey. Moreover, our findings illustrate the blanket application of “presumed risk factors” should be applied with caution and based on initial scientific evaluation to ensure geographic extrapolation is appropriate. Future targeting of T. dimidiata surveys should also include environmental risk factors as they performed just as well if not better than domicile construction covariates at detecting infestation exceeding the control threshold. Random surveys were generally more successful at detecting risk factors and predicting infestation than targeted surveys and should be applied over targeted surveys when risk factor identification, predictive modeling and extrapolation to the general populations is the goal. These findings illustrate the need for studies that are well defined, geographically specific, and based on reliable epidemiological investigation.
|
10.1371/journal.pntd.0000192 | Infections with Immunogenic Trypanosomes Reduce Tsetse Reproductive Fitness: Potential Impact of Different Parasite Strains on Vector Population Structure | The parasite Trypanosoma brucei rhodesiense and its insect vector Glossina morsitans morsitans were used to evaluate the effect of parasite clearance (resistance) as well as the cost of midgut infections on tsetse host fitness. Tsetse flies are viviparous and have a low reproductive capacity, giving birth to only 6–8 progeny during their lifetime. Thus, small perturbations to their reproductive fitness can have a major impact on population densities. We measured the fecundity (number of larval progeny deposited) and mortality in parasite-resistant tsetse females and untreated controls and found no differences. There was, however, a typanosome-specific impact on midgut infections. Infections with an immunogenic parasite line that resulted in prolonged activation of the tsetse immune system delayed intrauterine larval development resulting in the production of fewer progeny over the fly's lifetime. In contrast, parasitism with a second line that failed to activate the immune system did not impose a fecundity cost. Coinfections favored the establishment of the immunogenic parasites in the midgut. We show that a decrease in the synthesis of Glossina Milk gland protein (GmmMgp), a major female accessory gland protein associated with larvagenesis, likely contributed to the reproductive lag observed in infected flies. Mathematical analysis of our empirical results indicated that infection with the immunogenic trypanosomes reduced tsetse fecundity by 30% relative to infections with the non-immunogenic strain. We estimate that a moderate infection prevalence of about 26% with immunogenic parasites has the potential to reduce tsetse populations. Potential repercussions for vector population growth, parasite–host coevolution, and disease prevalence are discussed.
| In many cases, parasites adapt to their hosts' biology over time and the extent of their harmful effects gradually diminishes. Insect-transmitted parasites such as African trypanosomes, however, are unusually pathogenic for their mammalian hosts because they rely on their invertebrate hosts for transmission to the next mammalian host. To ensure their maximum transmission, it is essential that parasite infections do not compromise insect host's fitness traits, including longevity and host-finding ability. Our results in tsetse indicate that, as theory predicts, trypanosome infections do not reduce host longevity. Instead, they divert host resources from reproduction and can reduce reproductive output by as much as 30%. Such loss of reproductive fitness occurs as a result of the induction of tsetse's immune responses. A closely related non-immunogenic parasite line does not induce host responses and does not compromise host fecundity. It is possible that host immune responses are needed in the case of the immunogenic line to control the parasite density to prevent excessive host damage. Because tsetse are viviparous and each adult female typically gives rise to only few progeny during their lifetime, even modest costs on reproduction can have a significant impact on host abundance. Our model predicts that if the prevalence of immunogenic parasite infections in tsetse populations reaches over 26%, they begin to have a negative impact on population growth rate. Infection rates as high as 30% have been reported with trypanosomes in the field. Our laboratory findings coupled with our modeling studies now provide a framework to investigate the status of co-infections, host immune activation processes, fecundity outcomes, transmission dynamics, and host virulence phenotypes in natural tsetse–trypanosome populations.
| Insect vectors are essential for the transmission of malaria and African sleeping sickness, among many other diseases. Despite the high disease incidence in mammalian hosts, infection prevalence in insect vectors is typically low. For example, with the tsetse vectors of trypanosomes that cause African sleeping sickness, often only 1–3% of flies are infected in field populations (reviewed in [1]). This is also reflected in laboratory experiments where although all flies are subjected to an infectious bloodmeal, only a few show established midgut infections [2],[3]. Successful parasite infection of vectors likely reflects a balance between the effectiveness of the vector insect's immune response and the ability of the parasite to evade this response. The ability to resist parasitism has been shown to carry a fitness cost [4]–[7]. In several cases, a decrease in reproductive output has been shown to result from activation of the host's immune responses [8]–[10]. In a recent study, transgenic mosquitoes which expressed molecules that conferred parasite resistance were found to be more fit than wild type insects when exposed to malaria parasites [11]. Presumably the transgenic insects were able to eliminate infections prior to the activation of costly natural immune responses. In parasitized insects, the parasites likely compete for the insect's restricted nutritional resources. Here a balance also probably exists, since reductions in host survival or fitness would diminish the likelihood of parasite transmission. To compensate for nutrient competition while maintaining longevity, many insect vectors exhibit reduced or delayed reproduction as a trade-off during parasite infection (reviewed in [10]). Understanding the effect of parasitism on vector fitness is fundamental to predicting the future trajectory of coevolution between parasites and their hosts and eventual disease transmission dynamics.
Little has been reported on the impact of trypanosome infection outcomes on tsetse. Unlike most insects that are oviparous, tsetse females are viviparous producing one larva at a time, a process that results in few offspring produced over a lifespan. The pregnant females nurture their single larva in utero via specialized accessory gland (milk gland) secretions. It is likely that tsetse's viviparous character would result in different outcomes on lifetime fitness traits when compared to other insects where reproductive output is greatest and most sensitive to adverse occurrences in early adulthood [12]. In addition to reproductive fitness, vector longevity is especially relevant as African trypanosomes undergo multiple stages of differentiation in tsetse and require extensive developmental periods before transmission to their next mammalian host.
During blood-feeding, adult tsetse can acquire bloodstream form (BSF) trypanosomes from mammalian hosts. During the early course of infection, BSF, which differentiate and replicate as procyclic forms in the midgut, undergo a massive attrition. In a small percentage of flies, parasites continue to proliferate and establish midgut infections. The mechanisms involved in parasite elimination are likely based on tsetse immune responses [13] and may include lectin agglutination [14],[15], other lectin-like activities [16],[17], antioxidant activity [18]–[20] and killing by antimicrobial peptides (AMPs) [2],[18]. Both AMP transcripts and their encoded products are produced during parasite attrition in the midgut [2],[18],[21]. Additional support for a role of AMPs in resistance comes from studies where knock down of AMP expression in tsetse resulted in increased parasite prevalence in vivo [3] and where recombinant Attacin has been shown to exhibit trypanolytic activity in vivo [22].
In this study, we examined the cost to tsetse of resistance to trypanosome infections. We examined whether T. b. rhodesiense midgut infections can increase mortality and reduce reproductive fitness and cause delayed life-history effects on future progeny. We evaluated the cost of infections using two parasite lines, which differ in their ability to activate tsetse immune responses. We identified and discuss one putative mechanism by which tsetse reproductive physiology might be compromised upon parasite infection. Using a mathematical model, we then estimated the impact of fecundity cost on tsetse populations and discuss the putative effect of this cost on disease transmission.
BSF of the YTat1.1 parasites were derived from T. brucei rhodesiense stabilate TREU 164 (TREU = Trypanosomiasis Research Edinburgh University). TREU 164 represents passage 21 in mice from parasites originating from a capsule of bovine blood on which Glossina pallidipes captured at Lugala, Busoga, Uganda, in 1960, were allowed to feed. The pedigree of TREU 164 has been published [23]. The ETat3 variant derived from TREU 164 was triply cloned at Yale and referred to as YTat1. ETat3, and similarly, YTat1 clone1 (YTat1.1) are both highly virulent to the mammalian host, in that mice, even if infected with only a single organism, invariably die during the first parasitic wave. Procyclic culture forms (PCF) of YTat1.1WT cells were derived from BSF taken from infected rat blood and maintained axenically at 28°C in SDM-79 medium supplemented with 10% heat inactivated fetal bovine serum and penicillin-streptomycin antibiotic cocktail [24].
For YTat1.1EP selection, YTat1.1WT PCF were grown to about 8×107 cells/ml and diluted with fresh medium (1:20) once a week. The selection of the YTat1.1EP phenotype of trypanosomes was reproducible in three different experiments and was observed after maintaining cells under high-density culture conditions for at least two months. For maintenance of YTat1.1WT and the selected YTat1.1EP line, parasite cultures were split three times per week, after they typically reached 6–8×106 cells/ml. The YTat1.1WT cells (PCFs or the BSFs) are not able to establish salivary gland infections in the fly.
Glossina morsitans morsitans (Westwood) flies were maintained in the insectary at Yale University [25]. To initiate infections, newly emerged teneral flies were given a blood meal containing PCFs (105 cells/ml). Parasite infection prevalence in flies was confirmed by midgut dissection and microscopic viewing at designated times. Three independent experiments were performed. No significant difference was seen by arcsine transformation analysis, which allowed pooling of the groups. Chi-square analysis was used on the pooled data.
Fat body was dissected from individual trypanosome-infected or uninfected flies and homogenized in Trizol (Invitrogen, Carlsbad, CA) to extract total RNA following the manufacturer's instructions. Total RNA (10 µg per sample) was analyzed on a 1.5% agarose gel and UV crosslinked onto nylon membrane (Hybond N+, Amersham Pharmacia Biotech, NY). The membrane was hybridized with P32-labeled GmmAttA1 or GmmDef probes overnight as described [2]. The GAPDH housekeeping gene was used to normalize RNA input. The abundance of mRNAs was determined using a Phosphorimager (PSI-Molecular Dynamics). One representative data set is shown from six independent replicates.
In vitro maintained YTat1.1WT or YTat1.1EP PCF were used for immunoblotting. Parasites (5×103) were boiled in Laemmli sample loading buffer and their proteins were separated by SDS-PAGE. After transfer to polyvinylidene difluoride membranes, procyclins were detected with EP specific mAb 274 [26] and GPEET specific mAb 5H3 [27], respectively. For analysis of procyclins in vivo, infections were initiated with YTat1.1EP and YTat1.1WT PCF and midguts from microscopically positive flies were dissected. Parasites were eluted from dissected midgut and proventriculus tissues by gentle homogenization and washed 5 times in serum-free SDM-79 medium. Cell numbers were determined with a hemocytometer and 2×103 cells were used for immunoblot analysis.
The YTat1.1WT and YTat1.1EP PCF were transfected with pHD1034-GFP or pHD1034-RFP plasmid, respectively and the transformants were selected with 1 µg/ml puromycin [28]. The in vitro growth rate of each parasite line and parasite numbers in mixed cultures was measured over time by fluorescence microscopy using a hemocytometer. Parasite numbers in dissected and homogenized midgut extracts were similarly determined 14 days post acquisition. Infections were initiated either singly (106 cells/ml) or as mixed infections (5×105 cells/ml of each parasite strain mixed). Although multiple experiments were performed the results from two representative experiments are shown.
Female adults, 48 hours post emergence, received a single blood meal containing YTat1.1WT or YTat1.1EP parasites (105 cells/ml) and those that fed were maintained on a normal blood meal diet. Mortality rates were determined for these two groups and an uninfected control group over a period of sixty days. To determine the effect of parasite infections on tsetse fecundity, all females were mated 96 hours post emergence and maintained in individual cages and monitored daily for larval deposition. Fly midguts were dissected and microscopically analyzed for infection status at the completion of the experiment.
Data were analyzed using SAS system v. 8.02 for Windows [29]. Student's t-tests or Mann Whitney U-tests were used to determine whether larval deposition time intervals significantly differed between trypanosome infected (YTat1.1WT or YTat1.1EP) and control (both resistant and non-challenged) females. Student's t-tests were also employed to determine whether pupal weight and wing traits (length and width) significantly differed in the progeny. F-tests were applied to assess the homogeneity of variances.
Total RNA from eight YTat1.1WT and YTat1.1EP parasite infected 24 day old female flies and their age-matched normal controls were prepared. Following treatment with RNase-free Turbo DNase I (Ambion) the absence of DNA was confirmed by PCR amplification in a PCT-200 Peltier Thermal Cycler. One µg of total RNA was used for each sample for cDNA synthesis (Superscript II reverse transcriptase kit, Invitrogen, Carlsbad, CA). For qRT-PCR standard construction, inserts were cloned into the pGEM-T easy vector system (Promega, Madison, WI) [30]. Transcript quantification was performed on an iCycler Real-time detection system (Bio-Rad) and data were analyzed using software version 3.1. qRT-PCR for triplicate samples was performed using primers GmmMGPF 5′-CTGGACTCTTGACCCGTGAAC-3′ and GmmMGPR: 5′-GGGGAAGTGATGTTCCTTGA-3′ for 36 cycles at 58°C. For normalization, G. m. morsitans tubulin was amplified using the primer pair GmmTubF: 5′-GACCATGACGTGGATCACAG-3′ and GmmTubR: 5′-CCATTCCCACGTCTTCACTT-3′ for 36 cycles at 58°C.
Exposure of flies to wild type T. b. rhodesiense (Yale Trypanozoon antigenic type 1.1; YTat1.1WT) cells results in the upregulation of tsetse immune responses. For this analysis, we evaluated as immune markers the expression of two AMPs, attacin and defensin, which we had previously described from trypanosome-infected tsetse flies [2]. Infections with YTat1.1WT trypanosomes activated the host immune system and induced the expression of attacin and defensin as early as 3 days post acquisition (Figure 1, lane 1). This heightened response persisted (when analyzed at day 10) in susceptible insects harboring midgut parasite infections (lane 2). The immune response was also evident in those resistant flies that lacked microscopically detectable parasite infections at day 10 (lane 3). Only later, at day 30, had the immune response of resistant flies subsided to normal levels, indicating the long lasting impact of parasite recognition on host immune stimulation (data not shown, [2]). By passaging procyclic culture forms at late log-phase, a cell line was selected and designated (YTat1.1EP). This parasite line did not induce tsetse AMP expression when analyzed at day 3 and day 10, following parasite acquisition (lanes 5–6, respectively). The phenotype of this trypanosome strain is described below.
We first examined the expression of the cell surface procyclins on the immunogenic wild type parasites and the non-immunogenic strain that we had accidently selected in culture. We did this in order to understand the basis of the differential host immune activation, since the procyclic trypanosome surface coats are possible candidates for interaction with the fly's immune system. In T. brucei sspp., procyclins exist in several forms that are defined by the amino acid sequences of their C-terminal repeat domains carrying extensive glutamic acid-proline (EP) dipeptide repeats of differing lengths or pentapeptide repeats (GPEET). However, the expression of the procyclins is not static. It has been observed that the ratio of EP to GPEET procyclin expression in vitro can vary considerably between different parasite lines [27] or between different passages of the same culture [31] and that the composition of the coat may change in response to extracellular signals in vitro and during development in vivo [32]. Down-regulation of GPEET expression in trypanosomes has also been shown to be accelerated in vitro by hypoxia or be prevented by exogenous glycerol [32]. Procyclic culture forms of the non-immunogenic YTat1.1EP parasites expressed EP and not GPEET procyclins, whereas the immunogenic YTat1.1WT cells expressed both EP and GPEET procyclins, as determined using specific monoclonal antibodies (mAbs) by immunoblotting (Figure 2, panels A and B) and by flow cytometry (Figure 2, panels C and D). We do not know how our specific culture conditions caused the selection of the phenotype that exhibited decreased GPEET expression. However, the procyclin repertoire of the trypanosome lines we investigated was remarkably stable and did not change over several months of culture. Indeed, after being frozen for more than 6 months, thawed parasites grown in log-phase for a month exhibited the same phenotype (Figure 2C and D). Procyclic midgut trypanosomes purified from infected flies 12 days post parasite acquisition primarily expressed the EP procyclins with both parasite strains (Figure 2, D and F). Attempts to evaluate procyclin expression in epimastigotes in the proventriculus failed, likely due to the small number of parasites that could be obtained (Lane 3, Figure 2D and F). For these experiments, it is especially important to obtain pure parasite preparations given the cross reactivity EP monoclonal antibody exhibits with the unrelated tsetse EP proteins [33]. Nevertheless, these results suggest that in vivo YTat1.1WT parasite undergo a similar differentiation process as YTat1.1EP with respect to the expression of procyclins. It is of interest that the non-immunogenic trypanosomes originally fed to tsetse did not express GPEET whereas the immunogenic wild type parasites did.
Both YTat1.1WT and YTat1.1EP parasites had similar growth rates with a doubling time of 12±0.15 hours under in vitro cultivation conditions. The parasite lines were labeled with green fluorescent protein (GFP-YTat1.1WT) or red fluorescent protein (RFP-YTat1.1EP), respectively, to allow visual discrimination in mixed infection experiments. There was no statistically significant difference between the numbers of red and green trypanosomes in in vitro cultures that were initiated with single phenotype or mixed parasites (Figure 3A).
Prevalence of midgut infections with trypanosomes YTat1.1WT and YTat1.1EP were compared (Table 1), but showed no significant differences (p = 0.726. No significant difference in midgut infection intensity (parasite numbers) was found two weeks post infection acquisition with either parasite line (Figure 3B). However, when fly infections were established by feeding equal numbers of a mixture of the tagged parasites, a significantly higher number of YTat1.1WT cells were seen in established midgut infections (Figure 3B). Prior studies have shown that YTat1.1 cells exhibit high virulence in the mammalian host [34]. Since the YTat1.1 cells that we used have lost the ability to establish salivary gland infections in the fly, we were unable to compare the transmission potential of the two parasite lines and the potential virulence of YTat1.1EP for mammalian hosts. We thus limited our analysis to the impact of midgut parasite infections on tsetse fecundity and resulting population structure.
We compared the mortality and fecundity of flies infected with each parasite line for up to 60 days. There was no significant difference in mortality rates between the YTat1.1WT parasite exposed group and the age-matched, unchallenged control group (30.6% versus 30%, respectively). We next compared the fecundity of fertile females infected with YTat1.1WT and YTat1.1EP trypanosomes to age matched, uninfected controls (Table 2). Females were monitored daily for larval deposition through their initial three reproductive cycles. The infection status of the females was microscopically confirmed by examination of dissected midguts at the conclusion of the experiment. Our experiments with YTat1.1WT and YTat1.1EP parasites were conducted during 2002 and 2006, respectively. The difference in the larval deposition periods observed between these experiments can reflect environmental variations such as temperature and humidity conditions in the insectary. Hence, each experimental group was compared to its age-matched control group subjected to the same environmental conditions. For final analysis, the control group consisted of the nonchallenged and challenged but uninfected (resistant) females because of the lack of any significant differences between these two groups (P>0.05). All three larval deposition periods of flies infected with YTat1.1WT parasites were found to be significantly longer than those of the corresponding uninfected controls, while YTat1.1EP infected females had similar larval deposition periods to those of their control group.
We also evaluated the expression of tsetse larvagenesis associated protein, (milk gland protein; GmmMGP). GmmMGP is an abundant milk protein synthesized by the female accessory gland tissue [35] and supplied to the developing intrauterine progeny in the “milk” secretion [36]. Fertile females infected with the immunogenic YTat1.1WT parasite strain exhibited significantly decreased expression levels of GmmMGP in comparison to uninfected age-matched control females (Figure 4A). A similar reduction in GmmMGP was not observed in flies infected with YTat1.1EP parasite strain (Figure 4B), suggesting that the delayed larvagenesis process in immune stimulated flies may result from the decreased expression of the GmmMGP protein, which is necessary for larval growth.
In addition to the cost on direct life-history traits, we obtained morphological data reflective of adult fitness from three sequential progeny produced by mothers infected with YTat1.1WT parasites and compared these to an uninfected cohort (Table S1). There was no difference in pupal weight or in hatch rate (percent pupae that successfully emerged) between the two groups of progeny. In addition, there were no significant differences in wing width and length of the hatched progeny between the two groups. Based on morphological findings, delayed larvagenesis does not apparently result in loss of fitness of future progeny despite reducing the reproductive output of the mother.
We performed a mathematical analysis to translate the empirically measured times between larva depositions (Table 2) into the relative differences in fecundity of tsetse infected with the immunogenic versus the non-immunogenic trypanosome strains. In our analysis, we assumed that a proportion sP of deposited larva survive pupation to become adults, while the expected duration of the pupation is lP. We let m be the proportion of offspring that are female. We also assumed a constant adult death rate, μA, resulting in the expected duration of the adult stage . The time to first deposition was denoted t1 and the time between subsequent depositions t2. Thus, the second deposition occurs at time t1+t2, the third deposition at t1+2t2, and so on. The expected total number of female offspring of a female parent over her lifetime is then(1)which, when divided by the expected life span, lP+lA, gives the reproductive rate(2)
Reproductive rates in the absence of trypanosome infection, rU, and in the presence of infection, rI, are related by(3)so that ε is the relative fecundity cost of infection. Then(4)
Taking t1 to be the time to first deposition from our empirical measurements (Table 2), t2 to be given by the mean of the times to the second and third depositions (Table 2), sP = 0.82, lP = 31.4 days, and μA = 0.0253 day−1 [37] gives ε = 0.3. That is, trypanosome infection causes a reduction in fecundity of approximately 30%.
We next converted the fecundity differences calculated at the level of the individual fly into differences for tsetse population growth depending on whether or not the tsetse are infected. The population growth rate, r, is the root of the Euler–Lotka equation [38],[39].(5)For the parameter values above and 50% of offspring being female (m = 1/2), equation (5) gives rU = 0.0008 day−1 for uninfected tsetse and rI = −0.0025 day−1 for a population composed entirely of infected tsetse. Thus, for uninfected tsetse, rU is slightly positive, indicative of the slow viviparous reproduction of tsetse, while rI is negative, which would lead to population collapse if all tsetse were infected.
If the probability of surviving the pupal period is decreased to sP = 0.82 ([37] for G. palpalis gambiensis), the growth rate is negative, rU = −0.0012 day−1, while for infected tsetse, rI = −0.0044. With sP = 1, an increased death rate of μA = 0.030 day−1 has an even stronger effect on the basic reproduction number and growth rate, rU = −0.00046 day−1 and rI = −0.0082 day−1.
For a varying level of infection, if we assume no vertical transmission, the population growth rate for the overall population is given by(6)where p is the prevalence of antigenic virulent trypanosome infection in tsetse. With sP = 1 and μA = 0.022 day−1, at around p = 0.26, r = 0. For infection prevalence below 26%, the tsetse population continues to grow, while prevalence above 26% gives declining tsetse populations (Figure 5). The relationship between population growth rate and infection prevalence is monotone but not linear. The prevalence of trypanosomiasis would be expected to be reduced with a decline in the vector tsetse population. These modeling results are not intended to be precise quantitative predictions, but indicate likely qualitative dynamics of the interaction between tsetse and trypanosomes. Thus, the main conclusion that infection with the immunogenic parasite strain has the potential to suppress the tsetse population should apply broadly. However, the precise prevalence of trypanosomiasis that result in negative tsetse population growth depends on the exact parameter values, which may vary seasonally and spatially.
We studied the tsetse-trypanosome system to evaluate the molecular and physiological aspects of host-parasite interactions and the consequences of parasite infections on host fecundity. Two trypanosome strains that differentially activate host immunity were employed to assess the cost of the ability to clear parasite infections (resistance) and the cost of midgut parasite infections in susceptible flies. In contrast to the dogma that expression of parasite resistance traits incurs a fitness cost to the insect host, we observed no significant fecundity cost associated with trypanosome resistance in laboratory reared tsetse. However, activation of tsetse immune responses by infecting flies with immunogenic trypanosomes reduced host reproductive output while infections with non-immunogenic trypanosomes did not.
It is unusual that we did not detect a fecundity cost associated with parasite resistance in tsetse. This observation, that differs from what is seen in mosquitoes, may reflect the viviparous reproductive biology of tsetse where investments in reproductive output begin significantly later and continue throughout the lifetime of adult female flies. Female tsetse develop a single oocyte per gonotrophic cycle. Following ovulation and fertilization, the embryo develops within the uterus and hatches into a first instar larva, which molts through two more instars before being deposited as a fully developed larva which quickly pupates in the soil. From this point on, there is a continuous investment in nurturing the single larva produced one at a time, approximately every ten days, through the remainder of the female lifespan. This differs dramatically from mosquitoes, which have a high reproductive output as young adults and hence may be more sensitive to perturbations early in life. We cannot rule out however that our in vitro experiments conducted under uniform environmental conditions and ample nutritional supply may have skewed any potential fecundity cost that flies may experience upon expression of parasite resistance in the wild.
Reduced host fecundity has been observed in various systems with parasitized insects (reviewed in Hurd, 2003). Our result is the first demonstration of a significant loss of fecundity in parasitized tsetse but only when infected with an obviously immunogenic trypanosome line. Loss of fecundity in tsetse may arise from the cost of prolonged immune activation since a similar fitness cost was not observed in flies infected with the non-immunogenic parasites. Our results suggest that the decreased expression of milk gland protein GmmMGP, the most abundant protein product in the “milk” secretion of modified accessory glands, may cause the observed delayed larvagenesis process [36],[40]. We recently noted that in addition to GmmMGP, transferrin, which is also expressed in the female milk gland organ and is transported into the developing larva is down regulated by YTat1.1WT infections [41]. A similar reduction in the abundance of vitellogenin (Vg) mRNA and the titer of circulating Vg in the hemolymph has been reported in plasmodium infected mosquitoes during early oocyst development [42]. Later in the infection process, changes in the ovarian follicular epithelium have also been associated with a decrease in Vg uptake by the ovary and may also result in reduced hormone ecdysone production, which is needed for the transcriptional regulation of Vg expression [43]. Our results suggest several avenues for future research. It remains to be seen whether the regulation of host gene expression could result from an adaptive strategy that has evolved in response to parasitism, or that it reflects the manipulation of the host insect by the parasites. The regulation of milk-gland protein and transferrin transcription remains to be described, but may similarly be subject to hormonal regulation, which in turn may be influenced by parasite infections.
One potential difference we identified between the trypanosome strains used to infect tsetse was in their EP and GPEET procyclin compositions. Under our culture conditions, the immunogenic YTATWT line expressed both major forms of procyclin, while the YTATEP line only expressed EP procyclins. Procyclin epitopes analyzed from parasites obtained from midgut infections indicated that both lines expressed the EP isoforms later in the infection process, similar to previous reports [44]. It is possible that the highly phosphorylated GPEET molecules present on YTATWT procyclic culture form parasites, upon acquisition in the blood meal, are recognized by the host's immune system early in the infection process and elicit the immune activation we observed in our study. Supporting this idea, when beads carrying different surface charges were introduced into mosquitoes, different physiological networks were induced resulting in varied vector cost outcomes [45]. It is possible that as a consequence of host-parasite co-evolutionary dynamics, trypanosomes have evolved to down regulate the expression of GPEET in order to avoid inducing host resistance mechanisms. It is entirely possible however, that variations in parasite molecules other than procyclins may be responsible for the observed host immune outcome. The expression of GPEET alone in the early infection process cannot explain the sustained induction of host immune effectors in the case of infections with the immunogenic line since this line also switches to express predominantly the EP procyclins in late midgut infections. As molecular data on tsetse immunity are accumulating, it would be of interest to conduct a global expression analysis of both RNA and protein using the two parasite lines to gain a broader appreciation of the host-parasite interactions in the tsetse system [46].
Multiple vector immune components are likely to play a role in parasite transmission and resistance phenotype. Although our previous results have identified an important role for antimicrobial peptides in parasite establishment in tsetse, lack of AMP expression does not result in greater parasite infection prevalence for YTATEP. It is possible that other trypanocidal effectors, such as lectins [14]–[17] and antioxidant activity in the midgut milieu at the time of parasite acquisition that are necessary for resistance. Alternatively, cascades in immunity pathways could result in the synthesis of different trypanolytic immune effectors, which are uncharacterized at the present time. The similar midgut infection intensities observed with both trypanosome lines may also suggest general regulatory processes whereby homeostasis of parasite density is maintained to prevent excessive harm on host physiology. Our mixed infection experiments in vivo however showed that the immunogenic parasites outcompeted the non-immunogenic parasites in midguts. It is plausible that tsetse immune products, in particular antimicrobial peptides, may have varying trypanolytic effect on parasites with initial differences in surface protein composition. It is then possible that infections with the immunogenic parasites, invoking host immune responses, could prevent the transmission of other strains that may be more sensitive to the host immune products.
The prevalence of tsetse infections reported with T. brucei sspp. in the wild is low, typically in the order of 1–3% [47]–[49], suggesting the potential for only a small impact on the whole tsetse population. However, the prevalence of infections with other trypanosome species such as T. congolense and T. vivax, can be significantly higher, for example 28% reported in Burkina Faso [50] and 24% in Tanzania [51]. Furthermore infections with mixed trypanosome species are common, comprising over 30% of infections reported in natural populations [52]–[54]. Our modeling results suggest that these higher infection rates could have substantial effect on the tsetse population dynamics if the presence of these parasite species were to cause similar reproductive delays. Such reproductive delays caused by immunogenic trypanosome strains could lead to a reduction in tsetse population size and consequently to less transmission of parasites to vertebrate hosts. In turn, the resulting reduced trypanosome prevalence in vertebrates would lead to a lower prevalence in tsetse and a corresponding rebound of the tsetse population until an equilibrium of trypanosome prevalence in tsetse and vertebrate hosts is reached at lower levels than would occur with an avirulent strain. The mechanisms of tsetse population regulation are not well understood [55], so the magnitude of the impact of trypanosome-induced fecundity delays on trypanosome infections in vertebrates is difficult to predict.
Assuming that both parasite lines we investigated have equal probability for maturation in the salivary glands for mammalian transmission, our co-infection dynamics would favor the spread of the immunogenic phenotype. Reducing its tsetse host's fitness is evolutionarily disadvantageous to the trypanosome, but only weakly compared to the possible benefits of increased virulence. Interestingly, YTat1.1WT is known to be highly virulent in the mammalian host, but we do not have a virulence phenotype for YTat1.1EP since we cannot obtain salivary gland mammalian transmissible infections with this line. Trade-offs between virulence and parasite transmissibility can maintain virulence, as can competition within a host between multiple strains of a parasite [56]. Co-evolution between pathogen and host could lead to more complicated dynamics, from the emergence of mutualism to an escalating cycle of virulence and host resistance. In fact, a field study demonstrated that the resistance of Anopheline mosquitoes to Plasmodium parasite development varied considerably between different combinations of parasite and vector isolates, suggesting that there are specific compatibilities between the insect and parasite genotypes [57]. If there is no cost of resistance against virulent trypanosomes, we would expect resistance to increase until the “virulent” trypanosome is driven extinct. In contrast, the “avirulent” trypanosome would not impose selection on its tsetse host, and would persist. If there is a cost of resistance however, there would be an equilibrium proportion of resistance genotype at which the cost of resistance is balanced by the cost and prevalence of virulent infection. In this case, if the avirulent trypanosome strain without any fecundity cost emerged and spread, genetically susceptible tsetse would be expected to completely replace resistant tsetse. Simultaneously, disease prevalence in both tsetse and mammalian hosts would rise. Future experiments will need to be conducted with flies in the field to understand the impact of natural parasite infections on host immunity, and the potential burden of host immune activation on fecundity in order to relate to disease epidemiology. Our laboratory findings coupled with our modeling studies now provide a framework to investigate the status of co-infections, host immune activation processes, fecundity outcomes, transmission dynamics and host virulence phenotypes in natural tsetse-trypanosome populations. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.