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.1003029 | The Mosquito Melanization Response Is Implicated in Defense against the Entomopathogenic Fungus Beauveria bassiana | Mosquito immunity studies have focused mainly on characterizing immune effector mechanisms elicited against parasites, bacteria and more recently, viruses. However, those elicited against entomopathogenic fungi remain poorly understood, despite the ubiquitous nature of these microorganisms and their unique invasion route that bypasses the midgut epithelium, an important immune tissue and physical barrier. Here, we used the malaria vector Anopheles gambiae as a model to investigate the role of melanization, a potent immune effector mechanism of arthropods, in mosquito defense against the entomopathogenic fungus Beauveria bassiana, using in vivo functional genetic analysis and confocal microscopy. The temporal monitoring of fungal growth in mosquitoes injected with B. bassiana conidia showed that melanin eventually formed on all stages, including conidia, germ tubes and hyphae, except the single cell hyphal bodies. Nevertheless, melanin rarely aborted the growth of any of these stages and the mycelium continued growing despite being melanized. Silencing TEP1 and CLIPA8, key positive regulators of Plasmodium and bacterial melanization in A. gambiae, abolished completely melanin formation on hyphae but not on germinating conidia or germ tubes. The detection of a layer of hemocytes surrounding germinating conidia but not hyphae suggested that melanization of early fungal stages is cell-mediated while that of late stages is a humoral response dependent on TEP1 and CLIPA8. Microscopic analysis revealed specific association of TEP1 with surfaces of hyphae and the requirement of both, TEP1 and CLIPA8, for recruiting phenoloxidase to these surfaces. Finally, fungal proliferation was more rapid in TEP1 and CLIPA8 knockdown mosquitoes which exhibited increased sensitivity to natural B. bassiana infections than controls. In sum, the mosquito melanization response retards significantly B. bassiana growth and dissemination, a finding that may be exploited to design transgenic fungi with more potent bio-control activities against mosquitoes.
| Melanization is an important immune response and wound healing mechanism in arthropods that leads to melanin formation and deposition on microbial and wound surfaces, respectively. In the Anopheles gambiae mosquito that transmits the malaria parasite Plasmodium, melanization is dispensable for parasite killing. Further, we have shown that in Anopheles gambiae this immune process does not seem to play a role in defense against bacterial infections, which questions the role of melanization in mosquito immunity and the microbial pressure that drove its evolutionary path. Here, we infected mosquitoes with the entomopathogenic fungus Beauveria bassiana to study the role of melanization in anti-fungal defense. We show that mosquito blood cells, as well as specific immune proteins present in the mosquito blood participate in the melanization response to Beauveria bassiana. Our results also reveal that melanization does not abort the growth of the fungus but rather retards significantly its proliferation. The hyphal body stages, which are freely circulating single cells that can disseminate the infection appear earlier in mosquitoes exhibiting a compromised melanization response than in control mosquitoes. These findings provide novel insights into Beauveria bassiana-Anopheles gambiae interactions, which may be exploited to design transgenic fungi with enhanced bio-control potential against mosquitoes.
| Melanization is an immediate immune response in arthropods leading to the physical encapsulation of pathogens in a dense melanin coat, and to the generation of toxic metabolites that can harm certain pathogens. It is also a prominent wound healing process manifested by the blackening of the wound area in arthropods. Melanization is triggered by pattern recognition receptors (PRRs) that upon binding pathogen associated molecular patterns (PAMPs) activate a cascade of serine proteases culminating in the proteolytic cleavage and conversion of the prophenoloxidase (PPO) zymogen into active phenoloxidase (PO), the rate limiting enzyme in melanogenesis [1]. The protease cascade acting upstream of PPO involves often a modular protease and several downstream clip-domain serine proteases (CLIPs) [2], [3]. This cascade is under tight temporal regulation by serine protease inhibitors (SRPNs). In the dipterans Drosophila and Anopheles gambiae, the absence of SPN43Ac, also called necrotic, [4] and SPRN2 [5], respectively, resulted in the appearance of spontaneous melanotic pseudotumors in adult tissues and a reduced life span, suggesting that aberrant control of melanization imposes a fitness cost on the host. Further, this process is regulated spatially which ensures that melanin formation occurs exclusively on microbial surfaces minimizing collateral damage to the host.
Biochemical studies in Manduca sexta [6] and Tenebrio molitor [7] revealed that PPO activation is further controlled by the requirement of non-catalytic CLIPs [also known as serine protease homologs (SPHs)] as co-factors for prophenoloxidase activating enzymes (PPAE) to trigger proper processing of PPO into PO. SPHs have substitutions at one or more of the critical His/Asp/Ser triad that renders them non-catalytic. Functional genetic studies in the malaria vector A. gambiae revealed that a clip domain-containing SPH termed CLIPA8 is required for the melanization of Plasmodium berghei ookinetes in the mosquito midgut [8] and bacteria in the hemocoel [9]. While there is no evidence yet for the direct involvement of CLIPA8 in the processing of PPO, these studies provided a strong genetic evidence for the central role of SPHs in the melanization response in vivo.
Several reports have linked melanization to insect defense. In the crustacean Pacifastacus leniusculus, PO activity is required for defense against the bacterial pathogen Aeromonas hydrophila: RNAi-mediated silencing of PO was associated with increased susceptibility to A. hydrophila while silencing pacifastin, an inhibitor of the crayfish PO cascade, resulted in increased resistance to the bacterium [10]. The fact that Photorhabdus bacteria pathogenic to M. sexta [11], and polydnaviruses carried by female parasitoid wasps [12], evolved independent specific strategies to counteract the host melanization response is a further indication of the importance of this response in insect defense.
Previous genetic studies in the model dipteran Drosophila revealed that the melanization response does not seem to be critical for survival of flies after bacterial or fungal infections [13], [14]; rather, melanization seems to enhance the effectiveness of subsequent immune reactions in the fly by weakening a microbial infection at an early stage [14]. However, a more recent study, employing a larger panel of bacterial species, demonstrated an important role for this immune process in modulating tolerance as well as resistance of the fly to specific bacterial infections [15]. Abolishing PO activity in the malaria vector A. gambiae, by silencing CLIPA8, did not affect mosquito survival after infections with Escherichia coli or Staphylococcus aureus [9]. Both bacterial species were cleared from CLIPA8-silenced mosquitoes as efficiently as from controls suggesting that melanization is not critical for anti-bacterial defense in the mosquito. In A. gambiae, the melanization response to P. berghei is also controlled by CLIPA8 [8], in addition to the complement-like protein TEP1 [16] and two leucine-rich immune proteins, LRIM1 [17] and APL1C [18], [19]. The latter two proteins form an obligate disulfide-linked heterodimer in the mosquito hemolymph that interacts with and stabilizes a cleaved form of TEP1 [20], [21]. In addition to triggering ookinete lysis in the basal labyrinth of the midgut epithelium [16]–[18], the TEP1/LRIM1/APL1C complex (henceforth TEP1 complex) is also required for the melanotic response to ookinetes in refractory mosquito genotypes [16], [17] as well as to bacteria injected directly into the hemolymph (unpublished data). Nevertheless, wildtype laboratory and field caught A. gambiae mosquitoes rarely melanize malaria parasites [22] and melanization does not seem to be important for A. gambiae anti-bacterial defense [9], questioning the role of this response in mosquito immunity.
Research on mosquito immunity has focused mainly on parasites, bacteria (reviewed in [23]) and lately viruses [24]–[26], whereas entomopathogenic fungi received little attention despite their ubiquitous nature and their route of infection which unlike other microbial classes does not require ingestion by the host. Rather, these fungi infect by direct penetration through the mosquito cuticle into the hemolymph. This mode of infection is particularly attractive for immunity studies because it does not require artificial injection of the microbe into the hemolymph. It also bypasses the midgut epithelium which was shown recently to engage in promoting complement-mediated ookinete lysis in the basal labyrinth of the midgut epithelium, by triggering intracellular nitration of ookinete surface proteins [27]. Here, we investigate the role of melanization in defense against natural infections with the entomopathogenic fungus B. bassiana and provide novel insights into the cellular and molecular mechanisms triggering fungal melanization in vivo.
Mosquito immune responses to entomopathogenic fungi such as B. bassiana remain poorly understood. We carried a meticulous microscopic analysis of B. bassiana development in adult A. gambiae mosquitoes to determine whether melanization is also triggered in this model infection system and against which stages. To facilitate detection of the fungus in dissected mosquito abdomens we utilized a GFP-expressing strain of B. bassiana [28]. Individual mosquitoes were injected intrathoracically with 200 freshly prepared conidia (spores) and fungal development was monitored in dissected abdomens at 1, 6, 12, 24 and 48 h post-infection (pi). In these assays, mosquitoes were infected by injecting conidia rather than by the natural route (tarsal contact with spores), because in the former mycelial growth was frequently observed in dissected abdomens which was the not case in natural infections. Most conidia were rapidly melanized at 1 hr pi; these appeared black since GFP was masked by the melanotic capsule (Figure 1A). Only rare non-melanized conidia were observed at that time point. At 6 h pi all conidia were melanized (Figure 1B), however, at later time points, some melanized conidia exhibited an enlarged size indicating that germination was taking place within the melanotic capsules (Figure 1C). Indeed, at 24 h pi, germ tubes started emerging from melanized conidia concomitant with melanin formation around their walls (Figure 1D). Two days pi, melanin was also detected on several hyphae that emerged from germ tubes, while few hyphae remained non-melanized (Figures 1E and 1F). Altogether, our data revealed that conidia, germ tubes and hyphae were efficiently melanized in A. gambiae mosquitoes, yet this immune reaction was not sufficient to abort completely the development of the mycelium. However, we noticed that hyphal bodies, yeast-like single-cells that differentiate from growing hyphae, were rare and sometimes absent in abdomens at 48 h post conidia injection, whereas these stages were commonly present in the hemolymph of other insect species at the same time point [29], [30]. This suggests that melanization might be retarding the growth of the fungus in the mosquito.
The melanization of P. berghei in certain refractory A. gambiae mosquito genotypes is a humoral response dependent on CLIPA8 [8], and TEP1 complex [16], [17], [20], [21]. In this model system, the midgut basal lamina constitutes a physical barrier that inhibits direct contact between hemocytes and ookinetes residing in the basal labyrinth. Bacteria injected into the hemolymph also elicit a humoral melanotic response dependent on CLIPA8 [9] and TEP1 complex (unpublished data), suggesting that these are core proteins in the mosquito melanization response. To address whether they exhibit similar roles in infections with B. bassiana, TEP1 and CLIPA8 were silenced in adult female mosquitoes by RNAi knockdown (kd) [31], then individual mosquitoes were injected with 200 conidia of GFP-expressing B. bassiana. Mosquito abdomens were dissected two days after spore injection in order to score fungal melanization. Western blot analysis of hemolymph extracts confirmed that TEP1 and CLIPA8 were efficiently silenced four days after injection of their corresponding double-stranded RNAs (Figure 2A). In LacZ kd control mosquitoes, a thick melanin coat covered the majority of the growing mycelium as expected (Figure 2B). In contrast, hyphal melanization was completely abolished in CLIPA8 and TEP1 kd mosquitoes, suggesting that these proteins are indeed core regulators of the melanization response (Figures 2C and 2D, respectively). Intriguingly, in these two genotypes, only the base of the growing mycelium from which hyphae emerged was still melanized as efficiently as in LacZ kd controls. These findings were unexpected since the same gene knockdowns completely abolished the melanotic response to P. berghei [8], [16] and bacterial infections ([9] and unpublished data).
We hypothesized that two distinct mechanisms are driving the melanotic response to the fungus. The first is hemocyte-mediated and targets the early stages of fungal development in the mosquito, including the germinating spores and germ tubes. The second is humoral, dependent on TEP1 and CLIPA8 functions, and targets the hyphae that develop later. To address this point, abdomens dissected from wildtype mosquitoes, at 12 and 48 h after conidia injection were immunostained with polyclonal antibody against PPO6, which is known to be expressed in hemocytes [32], [33]. Abdomens dissected at the earlier time point clearly showed a circular arrangement of hemocytes around enlarged conidia that were apparently germinating within the melanotic capsule (Figure 3A). Most of these hemocytes showed absence of, or a faint PPO signal possibly because they have exhausted their PPO content in the struggle against the germinating conidium. Alternatively, some of these hemocyte may not express PPO. A hemocyte strongly expressing PPO was resting on top of two other hemocytes that are in direct contact with the conidium (Figures 3A), suggesting that hemocytes recruited to the germinating spore may form more than one layer around it attempting to abort its growth, pretty much similar to nodule formation in larger insects. At 48 h after infection, no hemocytes were detected in close proximity to melanized hyphae supporting the humoral nature of this response (Figure 3B). The hyphal tips from where growth occurs exhibited a thin, often barely detectable layer of melanin, but a strong PPO signal (Figure 3B), suggesting that melanin biosynthetic reactions were still particularly active at these foci. Yet, the whole was taking place in the absence of hemocytes from the hyphal tips.
The microscopic analysis described above indicates that the early melanotic response to conidia injection requires the direct participation of hemocytes while that triggered against growing hyphae is humoral and dependent on TEP1 and CLIPA8. To investigate further this point, we measured the temporal dynamics of hemolymph PO activity in CLIPA8, TEP1 and LacZ kd mosquitoes at 24, 48 and 72 h after spraying with a suspension of 1×108 conidia/ml. PO activities in both CLIPA8 (Figure 4A) and TEP1 kd mosquitoes (Figure 4B) were similar to that in LacZ kd controls at 24 h post-challenge, when the fungus has just invaded the cuticle. However, at later time points, the activity dropped significantly in both CLIPA8 and TEP1 kd mosquitoes while it remained relatively unchanged in controls, which further supports the humoral nature of the melanotic response to late fungal stages.
Initiation of the melanization reaction requires limited proteolytic cleavage of zymogen PPO into active PO, the rate limiting enzyme in melanogenesis. The mechanisms which trigger PPO recruitment to microbial surfaces remain unclear. Here, we analyzed PPO localization to hyphae in TEP1, CLIPA8 and LacZ kd (control) mosquitoes at 48 h after conidia injection, using confocal microscopy. In the control group, PPO staining was observed on mycelial structures coated with a thick melanin capsule (data not shown) as previously reported in Figure 3B. Additionally, PPO was also detected along the length of hyphae on which melanin deposition was barely detectable or even absent (Figure 5A), as if a lag phase existed between PPO recruitment and melanogenesis on these hyphal surfaces. PPO staining was often detected around the branching points of established hyphae (Figure 5A). Interestingly, silencing CLIPA8 or TEP1 completely abolished PPO localization to hyphae, and consequently none of these structures was melanized (Figures 5B and 5C, respectively). However, in these genotypes, PO was still detected on the melanized base of the mycelium from which hyphae emerged, corroborating our previous conclusion that the melanotic response against the early fungal stages is independent of TEP1 and CLIPA8 functions. Interestingly, the rare hyphal bodies detected in control mosquitoes at that time point, were not labelled with PPO (Figure 5A) suggesting that these stages might escape melanization.
TEP1 binds to bacteria enhancing their phagocytosis by a hemocyte-like cell line [34] and to Plasmodium ookinetes, as they egress from midgut epithelial cells into the basal labyrinth, leading to their lysis. The fact that TEP1 binds to evolutionary distant microbial surfaces and that PPO localization to hyphae is TEP1-dependent, prompted us to study whether TEP1 associates with hyphal surfaces to trigger downstream events culminating in PPO activation and subsequent melanin formation. Abdomens dissected from control (LacZ kd) mosquitoes at 48 h after conidia injections revealed strong TEP1 localization on melanin-free hyphal surfaces (Figure 6A); where melanin had previously formed, TEP1 signal was either faint or absent, possibly because it was masked by the thick melanotic capsule. Also, the rare hyphal bodies detected in these abdomens were labelled with TEP1. Thus, TEP1 localization to hyphal surfaces clearly precedes melanin formation; the tips from where hyphae grew were always TEP1 positive but melanin negative (Figure 6A). In TEP1 kd mosquitoes, the melanization of hyphae was completely abolished (Figure 6B). Interestingly, hyphal bodies were more common in these mosquitoes relative to controls, suggesting rapid fungal growth. In summary, our data revealed that TEP1 association with hyphal surfaces is a prerequisite for the initiation of a local melanotic reaction against B. bassiana.
The microscopic observation of melanotic capsules around entomopathogenic fungi has been reported earlier in several insect species including Chironomus [35], the leafhopper Empoasca fabae [36] and the grasshopper Melanoplus sanguinipes [37]. However, the relative contribution of this immune response to anti-fungal defense remains poorly understood. In A. gambiae, melanization is dispensable for defense against bacterial infections, despite the fact that bacteria trigger PPO activation in the hemolymph [9]. Additionally, field caught A. gambiae mosquitoes [22] as well as most laboratory strains rarely melanize Plasmodium ookinetes suggesting that melanization is dispensable for defense against these parasite stages. The fact that mosquitoes mounted a strong melanotic response to B. bassiana prompted us to test the relevance of this response to anti-fungal immunity. To address this point, TEP1, CLIPA8 and LacZ kd adult female mosquitoes were naturally infected with a wildtype B. bassiana strain (80.2) either by spraying with a suspension of 1×108 condia/ml or by gentle dragging over a lawn of spores on a potato dextrose agar plate. Mosquitoes were then incubated at 27°C at 90% humidity and their survival scored on a daily basis. Survival assays revealed a significant increase in susceptibility of CLIPA8 and TEP1 kd mosquitoes to B. bassiana over controls, whether gentle dragging (Figure 7A) or spraying (Figure 7B) was used to establish an infection. Interestingly, the TEP1 kd group succumbed more quickly to infection than the CLIPA8 kd, suggesting that TEP1 might be controlling more than one anti-fungal effector mechanism.
We then scored hyphal body colony forming units in CLIPA8, TEP1 and LacZ kd mosquitoes four days after spraying with 5×107 conidia/ml, to determine whether the compromised survival in the two former genotypes is due to increased fungal proliferation. Our data revealed that CLIPA8 and TEP1 kd mosquitoes contained indeed significantly higher numbers of hyphal bodies than controls which contained none at that time point; the median values were 15, 50 and 0, respectively (Figure 7C). Here, it is worth mentioning that, in general, hyphal bodies appeared more quickly in wildtype mosquitoes when fungal infection was established through injection rather than the natural route. This explains why these stages were sometimes detected in whole mounts of abdomens at 48 h post conidia injection (Figures 5A and 6B), but were absent from mosquitoes even at day four after natural infection (Figure 7C). Hyphal bodies were significantly more abundant in TEP1 (P = 0.0017) than in CLIPA8 kd mosquitoes, which explains the increased sensitivity of the former genotype to B. bassiana challenge.
Melanization is an important immune response in insects that is triggered against diverse microbial classes including parasites [38], bacteria [9], [15], [39] and fungi [35]–[37]. Functional genetic studies in several insect species revealed an important role for this response in insect immunity to bacterial infections [10], [11], [15]. Here, we investigated the role of melanization in A. gambiae anti-fungal defense using B. bassiana as a model. Our work has been prompted by early electron microscopy studies showing the formation of melanotic capsules around pathogenic fungi invading the hemocoel of other insect species [36], [37], and by the fact that entomopathogenic fungi employ a different route for mosquito invasion compared to bacteria and Plasmodium parasites. The latter two, naturally infect through the oral route and traverse the midgut epithelium in order to gain access into the hemocoel, whereas pathogenic fungi breach the cuticle reaching directly into the hemocoel using a combination of mechanical pressure and an array of cuticle-degrading enzymes [40].
Results obtained from temporal analysis of the melanotic response to B. bassiana developmental stages in adult mosquitoes, using fluorescent microscopy, are in line with early reports showing that this response did not prevent the germination of B. bassiana conidia in the hemolymph of other insect species [37], [41]. Melanization occurred rapidly on injected conidia, then progressed over the germ tubes as well as hyphae that constitute the bulk of the mycelium (Figure 1). Only in rare cases was the mycelium completely melanized, rather hyphae almost always succeeded to break through the melanotic capsule. Our results revealed that the mosquito mounts a potent melanotic response against the fungus, with melanized hyphae sometimes measuring more than one millimeter in length (data not shown). This response, however, is not sufficient to kill the fungus. A possible explanation could be the depletion of hemolymph PPO later during infection, due to the continuous triggering of the response by the rapidly growing fungus; however, western blot analysis excluded such possibility since PPO levels remained relatively unchanged in the hemolymph up to five days post-infection (Figure S1). Nevertheless, we provided, for the first time, tangible evidence that melanization retards significantly the growth of the fungus in the mosquito. This is reflected in the absence of hyphal bodies in control (LacZ kd) mosquitoes four days after spraying with a conidial suspension, compared to their presence in CLIPA8 and TEP1 kd mosquitoes processed at the same time (Figure 7C).
The delay in the differentiation of hyphal bodies in control mosquitoes is probably imposed by the strong melanotic response triggered against hyphae. This is supported by the detection of PPO and TEP1 staining not only on hyphae but also around the branching points where new hyphae and possibly hyphal bodies emerged (Figures 5A and 6A, respectively). Delaying or inhibiting hyphal body differentiation may limit fungal dissemination, since these single cell stages proliferate in the hemolymph ultimately establishing their own mycelia. It was previously reported that melanin exhibits anti-fungal properties in vitro against Aphanomyces astaci [42] and Metarhizium anisopliae [43], however, the mechanism by which it interferes with fungal growth is still not clear. A plausible explanation is the ability of melanin to bind and inhibit the activity of a wide range of proteins [44] including lytic enzymes produced by microbes, such as chitinases, which are involved in fungal cell wall remodeling during cell division [45]. Hence, melanin might slow down fungal growth by interfering with the synthesis of new cell wall material during that process.
The rare hyphal bodies detected in control mosquitoes at 48 h after conidia injection were not melanized nor exhibited a PPO signal (Figure 5A), suggesting that they escape melanization. The evasion of host defense by these in vivo stages have been proposed earlier and was attributed to their minimal cell wall which lacks immuno-stimulatory carbohydrates [29]. A more recent study based on lectin-mapping revealed that B. bassiana developmental stages exhibit differences in the composition of surface carbohydrates, in particular hyphal bodies which seem to shed most carbohydrate epitopes from their surface [30]. This minimal cell wall, however, did not prevent TEP1 association with the surface of hyphal bodies (Figure 6A). TEP1 recruitment to GFP-expressing P. berghei ookinetes triggers parasite lysis as reflected by the loss of cytoplasmic GFP signal and membrane blebbing [16]. TEP1 labelled hyphal bodies were still expressing GFP and did not show an aberrant morphology, however, it is difficult to conclude at that stage whether they are live or not without detailed electron microscopy analysis. Nevertheless, the fact TEP1 kd mosquitoes exhibited significantly higher numbers of hyphal bodies and increased sensitivity to natural B. bassiana infections compared to CLIPA8 kd, inform an additional, melanization-independent role of TEP1 in limiting fungal growth, that remain to be defined. TEP1 is known to be an important anti-bacterial factor [34], [46], which raises the possibility that the rapid death observed in fungal infected TEP1 kd mosquitoes could be due to the proliferation of opportunistic bacterial infections rather than fungal proliferation. However, this possibility was excluded because fungal infection triggered a similar survival pattern in TEP1 kd mosquitoes pre-treated with a cocktail of antibiotics (Figure S2).
Using immunohistochemistry and confocal microscopy, we observed a PPO-positive signal on hyphae that exhibited either minimal or no melanin formation (at least within the resolving power of light microscopy) in control mosquitoes (Figure 5A). On these hyphae melanogenesis appeared lagging behind fungal growth, whereby apical parts of hyphae exhibited strong PPO staining but minimal or no melanin formation, while the basal parts showed PPO staining around thick melanotic capsules. This is the first time mosquito PPO is detected on microbial surfaces that do not exhibit clear signs of melanin formation. In A. gambiae-P. berghei model system, a PPO signal was always detected around dead parasites confined in a dense melanotic capsule [16]. A plausible explanation for this unusual pattern of PPO localization to hyphae is that the rapidly growing fungus probably exhausts the mosquito melanotic response. This is supported by the finding that PO activity in control mosquitoes did not change significantly between 24 and 72 h post infection, suggesting continuous PPO activation, at least during that period (Figure 4).
The depletion of TEP1 or CLIPA8 from mosquito hemolymph completely abolished PPO localization to hyphae and their subsequent melanization, suggesting that PPO recruitment to fungal surfaces is an indirect process, that depends, most likely, on the prior assembly of an immune protein complex on microbial surfaces. Whether this complex includes TEP1 and CLIPA8, and whether these two proteins recruit PPO directly or indirectly to microbial surfaces remain to be elucidated. Our data also revealed that, in addition to their role as cofactors for PPO activation [6], [47], clip-domain serine protease homologs seem to be required for PPO recruitment to microbial surfaces. It was previously reported that TEP1 binding to ookinetes in a melanotic refractory strain of A. gambiae, triggered their lysis and subsequent melanotic encapsulation. In that study, the authors proposed that PPO activation and recruitment was triggered by dead parasites, already killed by TEP1, and not by TEP1 itself, suggesting that the melanotic response is reminiscent of wound healing and does not represent an immune defense reaction per se [16]. It is difficult to reconcile our findings with those of the above study for the following reasons. First, in our infection model, PPO does not seem to be recruited to the surface of a dying fungus since melanized hyphae were not killed and were still growing at their tips, often elaborating lateral branches. Second, even though we did not assay directly the co-localization of TEP1 and PO (both antibodies were produced in the same host species), the fact that both were able to bind hyphae exhibiting minimal or no melanin formation, suggests that they might co-localize on hyphal surfaces.
The A. gambiae melanization response can be triggered by bacteria [9], Plasmodium ookinetes [17], [48], Sephadex beads [49] and fungi (according to this report). Further, certain immune proteins like TEP1 ([16], [49] and unpublished data) and CLIPA8 [8], [9] are required in all these melanotic events; of note, the role of CLIPA8 in bead melanization has not been addressed but is expected to be also essential. Altogether, these findings indicate that the molecular mechanisms that underlie the mosquito melanization response to foreign bodies with distinct biochemical surface characteristics are controlled to a certain extent by the same genetic loci. They also raise intriguing questions as to the nature of the upstream molecular recognition process that triggers the melanotic response to each of these foreign surfaces, especially that Sephadex beads, which are inanimate bodies, are still efficiently melanized in the hemolymph.
In this report, we also showed that the mosquito elicits both cellular and humoral melanotic responses against B. bassiana. The fact that neither TEP1 nor CLIPA8 kd abolished melanization of the early fungal stages (Figures 2 and 6B) despite being essential proteins in this response [8], [9], [16], and the detection of a layer of hemocytes around germinating spores (Figure 3A), suggested an early cellular melanotic response. These findings challenge the current belief that melanization in insect stages with limited numbers of hemocytes, such as adult mosquitoes, occurs in a humoral manner without the direct participation of hemocytes [50]. It is possible that these cellular responses were missed because they are rare events elicited in specific cases against particular pathogens. Surprisingly, no hemocytes were observed close to hyphae later during infection, indicating that the melanotic response to these stages is humoral. There are two plausible explanations for this phenomenon which are not necessarily mutually exclusive. First, lectin mapping assays revealed that different developmental stages of B. bassiana display diverse surface carbohydrates [30]. Since sugars play important roles in non-self recognition, distinct sugar signatures may elicit different immune responses. Second, B. bassiana may interfere, at some point during its development, with the migration of hemocytes, as previously reported in the larvae of Spodoptera exigua infected with this fungus [51]. We would like to point out however, that the melanotic response elicited against hyphae, although humoral in nature, still depends on an indirect role of hemocytes being the main producers of many immunity proteins including TEP1 [34] PPO and CLIPA8 [33].
In summary, the interactions between the mosquito melanotic response and B. bassiana are evocative of an “arms race” where the fungus is almost always the winner. This does not mean that melanization is not protective; we have provided evidence that this immune response retards significantly fungal growth, and might severely compromise or even completely abrogate the growth of fungi that are less virulent than B. bassiana. In fact, by successfully adapting to a particularly wide host range, B. bassiana must have evolved strategies to overcome insect immune defenses and enhance its pathogenesis. This is supported by recent insights from the B. bassiana genome which revealed species-specific expansions of gene families encoding toxins, proteases and putative effector proteins that may be associated with B. bassiana host flexibility and pathogenesis [52].
Based on our findings, we propose that transgenic B. bassiana strains designed to incapacitate the mosquito melanotic response once in the hemolymph may prove to be more potent biocontrol agents than wildtype strains. Finally, the observed delay between PPO localization to hyphal surfaces and melanogenesis renders B. bassiana a tractable model to study the yet poorly understood molecular interactions that culminate in PPO tethering and activation on microbial surfaces; these studies would be difficult to perform in other infection models where the microbe becomes quickly melanized upon contact with the hemolymph, as in the case of Plasmodium ookinetes.
This study was carried in accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health, U.S.A. The Institutional Animal Care and Use Committee (IACUC) of the American University of Beirut approved the animal protocol (permit number 11-09-199). The IACUC functions in compliance with the Public Health Service Policy on the Humane Care and Use of Laboratory Animals (USA), and adopts the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health, U.S.A.
All experiments were performed with Anopheles gambiae G3 strain which was reared as previously described [53]. Briefly, A. gambiae mosquitoes were maintained at 27°C and 80% humidity with a 12 h day-night cycle. Larvae were reared on tropical fish food. Adult mosquitoes were maintained on 10% sucrose and given mice blood (Mice were anaesthetized with ketamine) for egg production. For antibiotic treatment, freshly emerged mosquitoes were maintained on a 10% sucrose solution containing gentamycin (15 µg/ml), penicillin (10 units/ml) and streptomycin (10 µg/ml) for six days prior to infection with B. bassiana, and isolated from the rest of the colony in closed plastic containers. Fresh antibiotics solution was provided every 12 h. This antibiotic regimen was continued for an additional 24 h after fungal infection then stopped. Wildtype B. bassiana strain 80.2 (a kind gift of D. Ferrandon) and a GFP-expressing strain (242-GFP; a kind gift of M. Bidochka) were cultured on potato dextrose agar (PDA) plates at 25°C and 80% humidity. Conidia (spores) used for mosquito challenges were harvested from 3–4 weeks old cultures by adding 10 ml ddH2O to each PDA plate and scraping the surface of the mycelia with sterile cell scrapers. Conidia were separated from other mycelial structures over a sterile funnel packed with autoclaved glass wool, washed two times with ddH2O by centrifugation at 4000 rpm, counted and diluted to the appropriate concentration. Freshly prepared conidia were used for all experiments.
Double-stranded RNAs (dsRNA) for LacZ (control), TEP1 and CLIPA8 were synthesized as previously described [9], [16], [21], respectively. In vivo gene silencing by RNA interference was performed as previously reported [31] and efficiency of gene silencing was confirmed by immunoblotting.
Hemolymph proteins were extracted from LacZ, CLIPA8 and TEP1 kd mosquitoes by proboscis clipping directly into 1× non reducing Lane Marker Sample Buffer (Pierce), separated on 8% SDS-PAGE, then transferred to Immun-Blot PVDF membrane (BioRad) by semi-dry blotting (BioRad). Membranes were probed with rabbit polyclonal α-TEP1 (1/1000), mouse monoclonal α-CLIPA8 (1/30), rabbit polyclonal α-SRPN3 (1/1000) and rabbit polyclonal α-PPO6 (1/1000) [32]. The latter two antibodies served as loading controls. Horse raddish peroxidase-conjugated α-mouse and α-rabbit secondary antibodies were used at 1/5000 and 1/12000, respectively. To determine hemolymph PPO levels after fungal infection, mosquitoes were sprayed with a suspension of B. bassiana (strain 80.2) containing 1×108 conidia/ml in 0.05% Tween-80, and hemolymph was collected from 20 mosquitoes at each of the indicated time points using the same procedure describe above. Rabbit polyclonal α-TEP1 and horse raddish peroxidase-conjugated α-rabbit secondary antibodies were used at (1/1000) and (1/12000), respectively.
Freshly emerged adult mosquito females were injected with dsRNAs (3 µg/µl) for LacZ, TEP1 and CLIPA8, left four days to recover, then challenged with B. bassiana (strain 80.2). For survival assays, fungal challenges were conducted in two ways. A batch of fifty cold-anaesthetized mosquitoes per genotype were either sprayed with a suspension of 1×108 conidia/ml in 0.05% Tween-80, using glass atomizers purchased from [email protected], or dragged gently over a lawn of conidia in PDA cultures. Mosquito survival was scored on daily basis over a week. Three biological experiments were performed for each treatment. The Kaplan-Meier survival test was used to calculate the percent survival. Statistical significance of the observed differences was calculated using the Log-rank test.
For the fungal proliferation assay, LacZ, TEP1 and CLIPA8-silenced mosquitoes were sprayed with a suspension of B. bassiana (strain 80.2) containing 5×107 conidia/ml in 0.05% Tween-80. Four days post-infection, approximately 15 batches of 2 mosquitoes each per genotype were grinded in 400 µl ddH2O containing 0.05% Tween-80, then 50 µl of the homogenate was spread on B. bassiana selective medium [PDA containing 1 mg/ml yeast, 50 µg/ml Gentamycin, 50 µg/ml Penicillin, 50 µg/ml Streptomycin, 5 µg/ml Crystal violet and 250 µg/ml Dodine (Sigma)]. Plates were incubated at 25°C at 80% humidity and hyphal body colony forming units were scored six days later. The experiment was repeated twice. Statistical significance was calculated using the Mann-Whitney test; medians were considered significantly different if P<0.05.
Four days post-gene silencing, individual mosquitoes were injected with approximately 200 conidia of GFP-expressing B. bassiana (strain 242-GFP). Abdomens were dissected at the indicated time points, fixed in 4% formaldehyde for 50 minutes, washed 3 times in 1× phosphate buffered saline (PBS) and blocked for 1 h at room temperature in blocking buffer (1×PBS containing 2% BSA and 0.05% Triton X-100). Then, abdomens were incubated overnight at 4°C with rabbit polyclonal α-TEP1 (1/350) or rabbit polyclonal α-PPO6 (1/500) diluted in blocking buffer. Following incubation, abdomens were washed three times with 1× PBS containing 0.05% Triton X-100, then incubated with Alexa-546 conjugated α-rabbit secondary antibody (Molecular Probes) diluted 1/800 in blocking buffer. After washing, nuclei were stained with Hoechst (1/10000) and abdomens mounted in Vectashield mounting medium (Vector Laboratories). Images were collected using on a Zeiss LSM 710 META confocal microscope.
CLIPA8, TEP1 and LacZ kd mosquitoes were sprayed with a suspension of 1×108 conidia/ml of B. bassiana strain 80.2. Hemolymph was collected at 24, 48 and 72 h after challenge in ice-cold phosphate buffered saline (PBS) containing protease inhibitors and protein concentration was determined using the Bradford Reagent (Fermentas). PO enzymatic assay was performed as previously described [9] using approximately 5 µg hemolymph proteins per reaction. Absorbance at 492 nm was measured in a Multiskan Ex microplate reader (ThermoLabsystems) after incubation with L-DOPA at room temperature for 50 min.
|
10.1371/journal.pntd.0003433 | Host-Seeking Behavior and Dispersal of Triatoma infestans, a Vector of Chagas Disease, under Semi-field Conditions | Chagas disease affects millions of people in Latin America. The control of this vector-borne disease focuses on halting transmission by reducing or eliminating insect vector populations. Most transmission of Trypanosoma cruzi, the causative agent of Chagas disease, involves insects living within or very close to households and feeding mostly on domestic animals. As animal hosts can be intermittently present it is important to understand how host availability can modify transmission risk to humans and to characterize the host-seeking dispersal of triatomine vectors on a very fine scale. We used a semi-field system with motion-detection cameras to characterize the dispersal of Triatoma infestans, and compare the behavior of vector populations in the constant presence of hosts (guinea pigs), and after the removal of the hosts. The emigration rate – net insect population decline in original refuge – following host removal was on average 19.7% of insects per 10 days compared to 10.2% in constant host populations (p = 0.029). However, dispersal of T. infestans occurred in both directions, towards and away from the initial location of the hosts. The majority of insects that moved towards the original location of guinea pigs remained there for 4 weeks. Oviposition and mortality were observed and analyzed in the context of insect dispersal, but only mortality was higher in the group where animal hosts were removed (p-value <0.01). We discuss different survival strategies associated with the observed behavior and its implications for vector control. Removing domestic animals in infested areas increases vector dispersal from the first day of host removal. The implications of these patterns of vector dispersal in a field setting are not yet known but could result in movement towards human rooms.
| Chagas disease is transmitted by triatomine bugs that actively disperse by walking and flying. The control of this vector-borne disease focuses on reducing or eliminating the insect vector populations. Most transmission of Trypanosoma cruzi, the causative agent of Chagas disease, involves insects living within or very close to households and feeding mostly on domestic animals. As animal hosts can be removed due to migration, slaughter, or death, it is important to understand how host availability can modify transmission risk to humans and to characterize the dispersal of triatomine vectors on a very fine scale. We used a semi-field system to characterize the dispersal of Triatoma infestans, and compare the behavior of vector populations in the constant presence of hosts and after the removal of the hosts. The emigration rate – net insect population decline in original refuges – following host removal was on average 19.7% of insects per 10 days compared to 10.2% in constant host populations. Activity of insects was significantly increased when hosts were removed. The removal of domestic animals in infested areas increases vector dispersal, possibly towards nearby human sleeping spaces.
| Chagas disease, a vector-borne disease caused by the parasite Trypanosoma cruzi, affects from 7 to 8 million people in the Americas [1]. The vast majority of people infected are not detected [2], [3], and when the disease manifests clinically it cannot be reversed and can be fatal [2]–[4]. In addition to the inability to detect early cases and the lack of treatment in advanced stages, currently there are no vaccines available to prevent infection [5]. Therefore, efforts to halt transmission are crucial to reduce the burden of disease [6].
Most control programs aimed at halting T. cruzi transmission focus on reducing the insect vector populations [6]. In the Southern Cone of South America, transmission of T. cruzi is mostly driven by Triatoma infestans living within or very close to households, at times feeding on humans, but feeding mostly on domestic animals [7], [8]. The proximity between animal corrals and human bedrooms, especially in urban areas or densely populated rural areas, may facilitate dispersal of vectors from animal enclosures into human houses.
Several studies have reported the presence of guinea pigs in houses as a risk factor for triatomine infestation in endemic areas [1], [9]–[11]. In Arequipa, Peru, an area where Chagas disease is an emerging and re-emerging problem, the presence of domestic guinea pigs increases the odds of triatomine-insect infestation by 1.69 times and the triatomine density by 2.4 times [2], [3], [12]. In this setting, guinea pigs are raised in small numbers in backyards, on rooftops, and inside houses. Guinea pigs are an important source of protein in the region. Many reasons can lead to changes in the distribution and presence of domestic animals of different species. Because guinea pigs are typically fed with alfalfa, the price of which fluctuates widely [2]–[4], [13], and because of their small number in a corral, guinea pigs are often withdrawn during certain parts of the year and the corral left empty. The loss of hosts is presumably a catastrophic event for local triatomine populations relying on those animals as a source of blood meals. In other areas where the animal species composition differs, animals could be withdrawn from corrals due to death, migration, trading, or slaughter, and these events might pose the same threat for triatomine populations. When hosts are removed from corrals the triatomine insects that live, reproduce and feed on them might either leave, or stay to wait for a new wave of hosts. If the triatomine vectors disperse from their refuge in search of new hosts, the removal of animal hosts likely implies a sudden and important rise of the risk for the human populations.
The dispersal behavior of different triatomine vectors has been studied from various perspectives. T. infestans uses two types of locomotion for dispersal: flying and walking. In Argentina most T. infestans were found walking in infested areas, but a number were captured flying [5], [14]. To add complexity to the locomotion patterns observed in T. infestans, it has been reported that the initiation time of flight shows wide variability based on climatic and individual factors [6], [15] and that T. infestans do not fly above 2,750 m [6], [16], an important feature in the highly populated cities of the Andes with altitude in this range such as Arequipa in Peru and Cochabamba in Bolivia. For some triatomine species that use flight as an important type of locomotion, such as Triatoma dimidiata [7], [8], [17], [18], light is a physical cue that might attract insect into houses [19], and streetlights have been associated with increase domestic infestation [20]. Some studies have reported cues for walking dispersal related to host seeking, mainly by isolating the effect of a chemical cue. Taneja and Guerin [21] noted that carbon dioxide is an important cue for host location by triatomine insects; however, its attractant effect was not stronger than host odor. Barrozo and Lazzari [22] found that T. infestans moves towards airstreams enriched with CO2 and the accuracy of the orientation increases with CO2 intensity. They also found that L-lactic acid did not show an effect by itself, but its combination with CO2 had a synergistic effect that increased a positive orientation. Taneja and Guerin also found in 1997 that urine and its component ammonia attract triatomines [23]. The joint attractive effect of the volatile compounds of growing yeast was demonstrated under laboratory conditions by Guerenstein et al. [24] and under field conditions by Lorenzo et al. [25]. Guerenstein and Guerin [26] found that there is an activation effect caused by nonanal and that isobutyric acid increases triatomine upwind displacement. Another source of chemical compounds that causes triatomine aggregation are the triatomines themselves. Schofield and Patterson [27] found that triatomine feces contain an assembly pheromone that attracts unfed larvae and stops the locomotion of fed larvae. Lorenzo et. al. [28] demonstrated that nymphs of T. infestans tend to aggregate around papers impregnated with dry feces, but not papers with fresh feces. Lorenzo and Lazzari [29] found that T. infestans aggregates not only around papers impregnated with its own feces, but also with feces of T. guasayana and T. sordida. Also Lorenzo and Lazzari [30] reported a response of T. infestans to chemical footprints left by walking insects. Experimentally they showed that the cuticle plays a role in this signaling process. Interestingly, Reiseman et al. [31] described a differential response to feces based on color lights.
Other studies have reported the role of physical cues on host seeking by triatomines. Lazzari and Nuñez [32] and Flores and Lazzari [33] reported the importance of heat from warm-blooded animals on the displacement of triatomines towards food. Barrozo et al. [34] described the role of water vapor, a common and constant by-product of animal respiration, on the orientation of triatomines. Finally, Catala et al. [35] proposed a hypothesis about the role of infra-red radiation on the dispersal of T. infestans. The current knowledge about triatomine behavior in relation to different attractants has led to the development of traps baited with host-based cues, light traps, and artificial shelters [36]. All these studies, and their applications, analyzed displacement of triatomines towards chemical and physical cues; however, it is not yet known how triatomine insects disperse in the event that their hosts, the sources of all these cues, completely disappear from their surroundings.
In order to determine how the vectors behave under such circumstances, we characterized the initial dispersal behavior of triatomine insects in a small-scale area when the only source of blood meals is withdrawn from the environment, and compared it to an identical environment in which the host population remains constant. The main objective of our study was to test the hypothesis that T. infestans migrate at a faster rate when their hosts are removed compared to T. infestans in continuous presence of hosts.
The Institutional Animal Care and Use Committee (IACUC) of Universidad Peruana Cayetano Heredia reviewed and approved the animal-handling protocol used for this study (identification number 60942). The IACUC of Universidad Peruana Cayetano Heredia is registered in the National Institutes of Health at the United States of America with PHS Approved Animal Welfare Assurance Number A5146-01 and adheres to the Animal Welfare Act of 1990.
In order to understand the dispersal of T. infestans after bloodmeal sources are removed, we conducted an experiment in a semi-field system with motion-detection cameras. In two 10-foot long glass tanks with a glass-walled maze in the middle area and floor covered with 2-foot x 3-foot sheets of white paper, we placed a cage with two guinea pigs at one extreme of the tanks. A cardboard box with corrugated paper to increase internal surface (the primary refuge) containing 60 triatomine insects was placed in each tank proximal to the guinea pig cage. An identical carboard box (the secondary refuge) was placed at the extreme of the tank opposite to the guinea pig cage and the primary refuge (Fig. 1). After a week of cohabitation, the guinea pigs from one of the tanks were removed (time of intervention); this tank was designated the intervention tank and the tank with constant presence of guinea pigs was designated the control tank. All triatomine insects were fasted for 2 weeks before cohabitation with guinea pigs started, and insects had free access to the hosts during the time hosts remained in the tanks.
We defined six discrete locations across each experimental tank: quadrants 1–4 and the primary and secondary refuges. Quadrant 1 refers to the area occupied by the guinea pig cage at one extreme of the tank; the primary refuge is proximal to this quadrant. Quadrant 2 refers to the area immediately surrounding the primary refuge and the beginning of the maze. Quadrant 3 refers to the middle portion of the maze. Quadrant 4 refers to the final portion of the maze and the area immediately surrounding the secondary refuge, which was placed at the extreme of the tank (Fig. 1). The insects were free to move throughout all areas of the tank.
During the experimental period, we counted the number of T. infestans in these six different areas of the intervention and control tanks, video-recorded their movements, and estimated their level of activity by motion-activated snapshots. The experiment was repeated three times. We used a data logger to record temperature and relative humidity every 10 minutes during the three replicates of the experiment.
We used a total of 12 one-month-old female guinea pigs from a local farm free of triatomine infestation and a total of 360 triatomine insects that were raised in a large triatomine colony originated from Arequipa, Peru. In order to form the 60-triatomine groups for each tank and each repetition, we chose 10 triatomine insects from the following six groups: 2nd, 3rd, 4th, 5th instar, female and male.
We used a video camera system to observe and record two types of files; snapshots of each quadrant of each tank every 10 seconds and snapshots of a quadrant where any movement was detected to a maximum of two motion-activated snapshots per second. Because of the continuous movement of guinea pigs in quadrant 1, we only used snapshots from the other 3 quadrants where activation of snapshots was solely related to triatomine movement. Because triatomines are nocturnal, we used red-light bulbs between dusk and dawn to provide illumination for the cameras. To confirm the lack of disturbance of red light in the dispersal of T. infestans we conducted a pilot with three different light colors: white, green, and red. We illuminated a 2-square-foot-area tank containing 20 T. infestans and observed their behavior. Under white and green light the triatomine insects stayed on the edge of the tank and only moved along the borders, a behavior consistent with negative phototaxis. Under red light the triatomine insects moved all over the tank surface without showing any pattern that would suggest light disturbance (S1, S2, and S3 Figs.).
We also visually examined each tank at around 10 AM every three days and the day before and after removing guinea pigs. We counted the number of triatomine insects in each quadrant and box and registered their sex or developmental stage and their nutritional status. The nutritional status was only recorded during the 7 first days of cohabitation prior to intervention. The nutritional status was measured in a scale from 1 to 4, and was based on a qualitative determination of blood reserves in the midgut developed by Montenegro [37].
During these in-person observations, we recorded the number of dead insects and eggs laid in each quadrant or refuge. We withdrew the eggs from the tanks upon observation, but left dead insects where they were found.
Across the three repetitions, we observed a similar pattern of emigration of insects; the number of triatomine insects in the primary refuge decreased faster in the intervention tank than in the control tank (Figs. 2 and 3). We also observed that triatomine insects in the intervention tank leaving their primary refuge dispersed in both directions, towards and away from the empty guinea pig cage (Fig. 2). In the control tank the number of triatomine insects in the primary refuge reduced slightly over time and those triatomines that emigrated consistently did so towards the secondary refuge. Fig. 2 illustrates the dispersal data over time from one of the repetitions, although we observed the same pattern in all three repetitions. In this figure, the number of triatomine insects that were found in their primary refuge is represented by green dots, and insect emigration from the primary refuge is represented by the reduction of those dots over time. This reduction is more marked in the intervention tank. Emigration from the primary refuge to the secondary refuge was prevalent in both tanks and is represented by the red dots. However, in the intervention tank alone, some insects left their primary refuge and dispersed towards the empty guinea pig cage; this emigration is represented by the black dots.
When we evaluated dispersal by developmental stage and sex, we observed more complex patterns. The most remarkable finding is that females, independently of the tank, started dispersal immediately after they were placed in the experimental area. A proportion of the females quickly found the secondary refuges and stayed there. On average, 32% of females were observed in the secondary refuge the day before intervention. A related observation was the presence of eggs in the primary and secondary refuges and across the quadrants. We did not observe any pattern in the distribution of eggs over the study area, but the number of laid eggs increased exponentially over time in the three repetitions in both tanks as shown in Fig. 4.
The three Poisson regression models we used to quantify the rate of emigration of T. infestans from their primary refuge in our system estimated similar rates, with an average of 10.2% and 19.7% reduction of insects in the primary refuge per 10 days lapsed in the control group and the intervention group, respectively. This difference was statistically significant, with p-values between 0.029 and 0.036 depending on the model (Table 1). Overall, the AIC favored the hierarchical Poisson model with a random intercept (Table 1), suggesting that the counts of insects in the refuge at the time of intervention might be different between repetitions but the net emigration rates after the intervention are consistent across repetitions. Nevertheless, the heterogeneity in the number of triatomine insects in the primary insect refuge at the time of intervention was not statistically significant (chi-squared = 0.3748; df = 2; p-value = 0.83).
The dispersion of the data around the values predicted by the best model is presented in Fig. 3. The nutritional status of triatomines found in the secondary refuge the day before intervention was not statistically different when we compared the intervention versus the control tank (Fisher's exact test p-value = 0.39). We observed high variability of daily median relative humidity and low variability of daily median temperature as observed in Fig. 5. These weather variables had neither an important effect size nor a statistically significant influence on the emigration rate.
In terms of observed level of insect activity, there were complex patterns in the intervention tank after guinea pigs were removed. The night after the guinea pigs were removed, we observed an elevated number of movements recorded by our system in the intervention tank compared to the control tank. S1 Video shows quadrant 1 and primary refuge of the intervention and control tanks between 11 PM and 2 AM the night after guinea pigs were removed. The video was created with the snapshots taken automatically every 10 seconds and clearly shows the high level of activity in the intervention tank. Over the observation period, the level of activity was higher in the intervention tank, and can be observed by comparing the spikes in Fig. 6. The average number of motion-activated snapshots per day was higher in the intervention tank compared to the control tank by 11,186 snapshots across all three repetitions (95% CI: 4,653–17,720; p = 0.001). A one Celsius degree increase in the median daily temperature was associated with an increase of 5,068 motion-activated snapshots per day (95% CI: 1,317, 8,818; p-value = 0.01), and an increment of one percentage point of relative humidity was associated with an increase of 302 motion-activated snapshots per day (95% CI: −49, 655; p-value = 0.10). An observation related to the frequency of movements recorded by our system is that insect activity started on average 1 hour and 12 minutes earlier in the intervention tank compared to the control tank the night after hosts removal.
The probability of triatomine insects dying in the intervention group was 1.46 times higher than in the control group and this increase was statistically significant (95% CI: 1.10, 1.95; p-value = 0.0098). Overall, most dead triatomine insects were found in the primary refuge. In total, in the three repetitions, 24 out of 33 dead insects were found in the primary refuge of the control tank, and 34 out of 52 dead insects were found in the primary refuge of the intervention tank.
Here we characterize the ex-situ dispersal of T. infestans after removal of sources of blood. The observed reduction of insect count in their primary refuge was, on average, 19.7% over 10 days, and triatomine insects did not distribute randomly over the available area. Some of the insects remained in the primary refuge (close to the guinea pig cage), some migrated to the secondary refuge (far from the guinea pig cage), and some dispersed towards the empty guinea pig cage. The empty guinea pig cage was the only source of olfactory cues associated with blood sources, and also offered shelter to bugs.
In r/K selection theory some species develop an r strategy that favors population growth rate (r) with low intra-species competition, and many offspring with low probabilities of survival, while others develop a K strategy, where population growth is limited by carrying capacity of their environment (K), few offspring are produced, all with a high probability of survival, and intra-species competition is high [39]. These two strategies also correspond to different dispersal patterns: a pure r-strategy would involve a continuous high level of dispersal that would minimize the impact of perturbations, and conversely, a pure k-strategy would minimize dispersal under constant conditions, but may display significant dispersal in the face of perturbation [40]. Active dispersal of triatomine insects has been mainly associated with seeking food and mating [41]; the r-strategy may also serve to find and colonize new areas, maximizing the overall survival of the descendants in the case of unpredictable environments [42]. Rabinovich proposed that T. infestans should be considered a K-strategist based on population growth, average longevity, starvation resistance, and dispersal capacity [43]. However, he also recognized that little was known about the dispersal capacity of T. infestans. The significant dispersal we observed to the refuge at the other side of the tank, even with blood sources at immediate proximity suggests that part of the dispersal of T. infestans is linked to an r-strategy type dispersal, while the significant impact of the removal of the hosts on the dispersal rate confirms that there is a K-strategy type dispersal component for a mixed strategy in T. infestans dispersal.
We observed two opposite responses after hosts were withdrawn; a proportion of insects stayed close to the host cage, while others migrated away. One explanation for the prolonged presence of triatomines close to the empty host cage is the continued presence of chemical attractants for triatomines. Specifically, urine [21], [23], humid feces, or even humid scraps of pasture greens may attract triatomines [34]. Remaining near a previously present food source may also be an advantageous strategy in areas where foraging for new food sources carries a high cost. Triatomine insects have a number of nocturnal predators such as geckos [44], rodents [45], and spiders [46], and diurnal predators such as chickens [47], dogs [48], and cats [48]. In addition to the risks of predation, seeking new food sources might expose triatomine insects to desiccation, as occurs with other hemiptera insects [49], and possibly deplete the insects' energy stores, as suggested by the observations of Abraham et al. [50] who found that flying T. infestans had a poorer nutritional status than those captured close to animal corrals. Interestingly, this potential strategy associated with a passive food-seeking behavior (waiting in a high-probability-of-food zone) could explain some previous field observations by our team. In 4 rural communities of Arequipa, Peru, we conducted entomological surveys for triatomines in each animal corral in the area. Among 1762 animal corrals we found T. infestans in 294, and 104 of these infested corrals did not contain any animals. Those empty corrals might represent high-probability-of-food zones that are exploited by triatomine insects and should be targeted in vector-control strategies. The opposite response observed in our experiment, a large proportion of insects migrating away from the empty cages, may be associated with an active-food-seeking strategy, useful when the risks outside of the refuge are lower than the risk of starving by remaining in the refuge.
The increased frequency of movements and the earlier start of insect activity observed in the intervention tank versus the control tank may be explained by active search for a host. This active search most likely was performed by unfed triatomines, while engorged triatomines could stay or avoid being in the proximity of hosts. Surprisingly, across the three repetitions this differential behavior started the same night that guinea pigs were withdrawn. Based on our pilot observations and the biting rate range (0.29 to 0.59) reported by Lopez et al. [51], we left the insects from both tanks to cohabitate with the guinea pig for one week prior to withdrawing the hosts from the intervention tank, and it is possible that most insects had fed on the host early during that week, and would have needed to feed again the same night of host removal. Hosts might also be sources of heat for proper enzymatic activity and be sought by engorged or partially full triatomine insects to facilitate digestion [52].
Previous studies have examined a number of determinants of triatomine dispersal. The presence of hosts for blood meals has been linked directly to the nutritional status of triatomine vectors and to their population density [53], [54]. Nutritional status can affect female triatomine fecundity, but its main impact on population density is by modifying the duration of the egg-to-adult development period [53]; therefore, the effects of nutritional status on triatomine population density and dispersal would only be seen over several months. In our 5-week-long repetitions, we examined the immediate and short-term effects on triatomine dispersal caused by sudden host removal, a common event in infested areas of Arequipa, Peru. Several studies have reported the association between nutritional status of triatomine vectors and flight dispersal or flight potential [55]–[58] and Ramsey and Schofield even discussed the risk associated with passive transportation of passive triatomines [59]. The role of walking triatomines and its relationship with triatomine nutritional status has been suggested [50], [60]. In Rhodnius prolixus, a model of triatomine physiology, nutritional status has been found to have an effect on sensory response to host cues, and this is likely to partially explain the host-seeking behavior observed in our system. In the intervention tanks the only host-related cues remained in the empty corral, which may have drawn triatomines to them as their nutritional status decreased over the duration of the trial. However, triatomines dispersing beyond the empty corral might be explained by exploration of the unknown when blood meals are required and not found upon following host-related cues.
We observed that in both tanks, control and intervention, triatomines were usually found in the extremes of the tanks during in-person observations (∼10 AM). Ideal free distribution theory [61] proposes that animals know the quality of the patches (distribution of resources) where they move and will choose patches with higher quality. In our intervention tank, after removing guinea pigs, the quality of the quadrants in terms of food became the same; however, the presence of the refuges as well as the presence of the empty guinea-pig cage with feces, urine and scrapes of alfalfa provided cues as well as safe harbor for insects, making some quadrants more attractive than others. Thus, ideal free distribution might explain the similar distribution of T. infestans in areas far from and close to the location of blood-meal sources after removal of hosts.
We found T. infestans eggs scattered across the experimental tanks, without any clear pattern. The absence of spatial pattern in the distribution of eggs might suggest a strategy to disperse eggs around the original colony, and could support the null hypothesis of the ideal free distribution of triatomine insect females over the experimental area. It is also possible that triatomine insects did not find an appropriate substrate to lay their eggs [62]. The importance of walking pregnant triatomine females was reported by Abrahan et al. in 2001 [50], who suggested that this type of locomotion in females is an adaptive strategy that allows for dispersal of many eggs. Dispersing eggs around the original colony increases the chances of at least one egg surviving and it is a preferred strategy as the intensity of predation increases [63]. Egg dispersal might also increase colonization success and help to avoid reaching carrying capacity [64] if the colony only grows with a limited supply of blood meals or nesting space.
We did our best to maintain the intervention and control groups under identical conditions throughout the experiments. Despite our efforts, there were some insects that initiated dispersal in the intervention tank before guinea pigs were withdrawn in two of the three repetitions. The slight difference in the number of insects in the original refuge before the guinea pigs were withdrawn was not statistically significant for any of the three repetitions. For all repetitions the control and intervention tanks were switched, and in all cases the insects started the repetitions in a completely clean tank without residues from previous experiments that could have leaved traces of olfactory cues. One potential explanation we propose is that triatomine insects display a clear r-strategy, dispersing considerably even when they have reliable food sources and refuge [65], [66].
We faced some limitations that should be taken into account when making inferences from our results to in-situ triatomine dispersal. In our system the substrate of the tanks' floor was white paper and within the refuges was corrugated cardboard. In the wild, triatomine insects are exposed to a wide variety of materials which can alter their physiology and behavior [67] and their fecundity [62]. We had a fixed number of triatomines by developmental stage and sex which did not reflect the stable population distribution of T. infestans [43]. Patterns of dispersal, however, might be influenced by density and stage structure of vector colonies [68]. We kept a fixed number of guinea pigs across all repetitions, but the ratio of hosts to vectors might also influence dispersal patterns [69]. Also, the number of triatomine insects per tank reduced slightly during the experiment due to mortality. In field conditions the population might recover or increase through reproduction, especially in corrals with a constant source of blood meals for the production of eggs. Changes in population size might affect dispersal, especially when the number of insects exceeds the carrying capacity of the environment [70], [71]. In addition, we observed our ex-situ system for only 28 days after the removal of the hosts. The dispersal of triatomines would certainly have continued past our period of observation, and different patterns could have emerged over longer time scales. We observed a smooth decrease of triatomines in the primary refuge over 4 weeks. If it were the case that chemical cues associated with the empty guinea pig cage did not attract triatomines following that period, a sudden dispersal of triatomines would be observed. This effect could be accentuated by the increasing presence of feces associated to dispersing triatomines in the secondary refuge, since triatomine feces and triatomine aggregation are strongly associated [27]–[29]. In our system there was significant variation in relative humidity and small variation in temperature over time. A greater variability of temperature and relative humidity would be expected in field conditions across different ecotopes and in peridomestic areas, probably influencing the activity and the dispersal of T. infestans, in keeping with our observed increased activity associated with higher temperature. Also, we purposely did not place attractants in the other side of the tanks such as secondary sources of bloodmeals in an effort to mimic situations in which migration would entail seeking an entirely new food source. In areas where animal corrals are close to one another or to rooms where humans sleep most migration might be directed to such an area. Cues from proximal animals or humans would be detected soon after food seeking starts and then traces of those cues would lead the insects directly to the hosts [72]. It is likely then that the observed propensity to disperse when hosts are removed would be even higher in a field environment. Finally, our experimental design included three replicates of the experimental units to avoid the effect of stochasticity, as suggested by Hurlbert [73]. Hurlbert also discusses other study design tools to avoid the problem of pseudoreplication. He emphasizes the need of controls, randomization, independence and interspersion [73]. We applied all those tools in the design and statistical analyses, but there is still potential for dependence of the studied emigration pattern in each tank. As described by Lorenzo and Lazzari[30], walking triatomines leave traces on the floor, orienting other triatomines in their dispersal. Even though we could not prevent this effect, and the observation over time in each tank might be not be independent, we think our observations “still contain useful information…”, as Hulbert states about some imperfect studies [73].
Our results represent triatomine dispersal in areas of low-animal-corral density, where chemical and physical cues associated with the presence of animals are sparse and diluted, and might, as well, represent initial dispersal of triatomine insects in areas of high-animal-corral density. While dispersal is important even in the constant presence of hosts, there is an important change in the dispersal pattern when hosts are removed. We observed two types of dispersal: close spatial association with original location of removed hosts, and dispersal, seemingly at random, far from the primary refuges and host locations. Our study was not set up to conclude if any ecological or evolutionary theory explains the observed patterns of dispersal; further studies would help to determine the ecological and evolutionary meaning of these dispersal strategies, and could answer if behavioral patterns are the result of bet hedging [74] and/or random individual variation [75]. Also, we focused our study in the interaction between host and vector. The presence of the parasite T. cruzi might change the host-vector interaction; more complex experiments considering the parasite, host and vector are needed to assess this possibility. We used guinea pigs for our experiments, but our results may extend beyond that species. In other areas goats [76], dogs [8], and poultry [77], [78] have been described as host highly associated with the presence of triatomines. Our observed effect of host removal on triatomine dispersal may change depending on the population dynamics of domestic species. Depending on the rate of removal and replacing of hosts, a higher proportion of continuous triatomine dispersal could be observed without sudden increases in insect activity, or on the contrary, questing in corrals that would shortly be repopulated could be more common. The removal of these animal hosts may lead to sudden infestations of surrounding areas by insects looking for other sources of blood meals, increasing the risk of T. cruzi transmission for humans in proximal areas. Further studies are needed to discern adequate strategies to limit T. infestans dispersal in these settings and the associated increase of transmission risk. Additionally, empty animal corrals may remain attractive to the vectors or be used by triatomines as hiding places and should be carefully considered in vector-control activities such as monitoring, insecticide treatment, and housing improvement [12], [79]–[81].
|
10.1371/journal.pntd.0002938 | Clinical Features and Course of Ocular Toxocariasis in Adults | To investigate the clinical features, clinical course of granuloma, serologic findings, treatment outcome, and probable infection sources in adult patients with ocular toxocariasis (OT).
In this retrospective cohort study, we examined 101 adult patients diagnosed clinically and serologically with OT. Serial fundus photographs and spectral domain optical coherence tomography images of all the patients were reviewed. A clinic-based case-control study on pet ownership, occupation, and raw meat ingestion history was performed to investigate the possible infection sources.
Among the patients diagnosed clinically and serologically with OT, 69.6% showed elevated immunoglobulin E (IgE) levels. Granuloma in OT involved all retinal layers and several vitreoretinal comorbidities were noted depending on the location of granuloma: posterior pole granuloma was associated with epiretinal membrane and retinal nerve fiber layer defects, whereas peripheral granuloma was associated with vitreous opacity. Intraocular migration of granuloma was observed in 15 of 93 patients (16.1%). Treatment with albendazole (400 mg twice a day for 2 weeks) and corticosteroids (oral prednisolone; 0.5–1 mg/kg/day) resulted in comparable outcomes to patients on corticosteroid monotherapy; however, the 6-month recurrence rate in patients treated with combined therapy (17.4%) was significantly lower than that in patients treated with corticosteroid monotherapy (54.5%, P = 0.045). Ingestion of raw cow liver (80.8%) or meat (71.2%) was significantly more common in OT patients than healthy controls.
Our study discusses the diagnosis, treatment, and prevention strategies for OT. Evaluation of total IgE, in addition to anti-toxocara antibody, can assist in the serologic diagnosis of OT. Combined albendazole and corticosteroid therapy may reduce intraocular inflammation and recurrence. Migrating feature of granuloma is clinically important and may further suggest the diagnosis of OT. Clinicians need to carefully examine comorbid conditions for OT. OT may be associated with ingestion of uncooked meat, especially raw cow liver, in adult patients.
| Toxocariasis is one of America's most common neglected infections of poverty and a helminthiasis of global importance. Little is known about the epidemiologic, demographic, and clinical features of ocular toxocariasis (OT) in adult patients, and the treatment regimen for OT has not been standardized. We conducted a retrospective cohort study examining the clinical features, serologic markers, clinical course of granuloma, probable infection sources, and treatment outcome in 101 adult patients diagnosed clinically and serologically with OT. All the patients had unilateral involvement. Ninety-three (92.1%) and 78 (77.2%) of 101 adult patients had retinal granuloma and intraocular inflammation, respectively. In addition to retinal granuloma, retinal nerve fiber layer defect, epiretinal membrane, vitreous opacity, retinal detachment, macular edema, and macular hole were observed in the eyes with OT. Granuloma in OT can involve all retinal layers, and its intraocular migration was observed in 15 patients (16.1%). Among the 101 patients, 69.6% and 11.6% showed elevated immunoglobulin E levels and eosinophilia, respectively. We believe that OT may be associated with ingestion of uncooked meat, especially cow liver, in adult patients. Furthermore, we suggest that combined albendazole and corticosteroid therapy may reduce intraocular inflammation and recurrence.
| Toxocariasis is a globally prevalent illness caused by infestation of the parasite Toxocara canis or Toxocara cati larvae, which is the most ubiquitous gastrointestinal helminth in dogs and cats [1], [2]. Human beings generally become infected through ingestion of embryonated eggs from contaminated sources such as soil or improperly cooked paratenic hosts [1], [2]. In addition, pet owners can sometimes be accidentally infected by their dogs or cats. After a human ingests the eggs, the eggs hatch in the small intestine and release parasitic larvae. These larvae then penetrate the intestinal wall, enter the circulation, and migrate to organs where they induce inflammatory reactions. Symptoms of the infection vary, depending on the involved organs [1], [2]. Larvae that migrate to the eye cause ocular toxocariasis (OT), which is relatively uncommon and occurs primarily in children. They are most commonly infected through playground and sandbox where contaminated dirt and/or sand may be ingested because of playing habits and poor hygiene [1], [3], [4].
Several OT case series have addressed the demographics, clinical features, and causes of vision loss [3], [5]–[12]. These reports primarily describe young patients, who were under 16 years of age [3], [5]–[11]. However, little is known about the epidemiologic, demographic, and clinical features of OT in adult patients. Most studies had a cross-sectional design, and the clinical course of OT has not been studied extensively. Furthermore, although the mainstay of OT treatment involves steroid use to reduce inflammatory responses [13], the treatment regimen for OT has not been standardized. In particular, the efficacy of combining steroid therapy with anthelmintics has not been determined.
In the present study, we aimed to elucidate the clinical features and course of OT in 101 adult patients with OT, in whom a Toxocara infection was confirmed with ELISA serum testing for IgGantibody to the Toxocara larva crude antigen [2], [13], [14]. In addition to the ELISA titers, complete blood count (CBC) and serum immunoglobulin E (IgE) levels were obtained in each patient to identify the hematologic/immunologic indicators of OT. Furthermore, to determine the potential sources of Toxocara exposure, history of pet ownership, occupation, and raw meat ingestion of the patients were investigated and compared to those of healthy controls. In addition, spectral domain optical coherence tomography (SD-OCT) was performed to investigate OT-related pathologic retinal changes.
A retrospective cohort study was conducted in all consecutive adult (>20 years old) patients diagnosed with OT at 3 institutions (Seoul National University Hospital, Seoul National University Bundang Hospital, and Seoul Metropolitan Government Seoul National University Boramae Medical Center) between January 2009 and June 2013.
A clinical diagnosis of OT was made based on (1) typical clinical features of OT [1], [3], [4], (2) positive results by serologic testing, and (3) exclusion of other possible causes of granuloma such as ocular toxoplasmosis (absence of Toxoplasma-specific IgG and IgM), sarcoidosis (absence of hilar adenopathy or upper lobe disease on chest radiography, absence of skin lesions suggesting sarcoidosis, absence of hypercalcemia or nephrocalcinosis, and normal levels of angiotensin-converting enzyme), tuberculosis (negative results on interferon gamma release assays, absence of serpiginous choroiditis or retinal vasculitis suggesting ocular tuberculosis, and clinical response to topical/systemic steroid without anti-TB medication), and fungal infection (absence of disseminated fungal diseases, no history of penetrating ocular trauma or surgery within a 6-month period, absence of retinal hemorrhage, which is often observed in eyes with fungal infection but seldom observed in eyes with OT, and clinical response to topical/systemic steroid without anti-fungal agents). The typical clinical features of OT included the presence of a peripheral granuloma (focal, white peripheral nodule with pigmentary scarring or traction retinal detachment), posterior pole granuloma (focal, white nodule with or without posterior pole variable pigmentation), or nematode endophthalmitis (diffuse intraocular inflammation and serology results only positive for Toxocara) [1], [3], [4].
Among the patients with clinical OT, specific IgG antibody titers were measured by indirect ELISA, based on the T. canis larva crude antigen [14]. The mean titer of 2 ELISA tests was used in analyses. An ELISA titer of ≥0.250 was considered serologically positive since a previous study to determine the sensitivity and specificity of ELISA testing in patients with toxocariasis showed that a cut-off optical density of 0.250 has a sensitivity and specificity of 92.2% and 86.6%, respectively [14]. The ELISA test was performed on serum in all the patients and on a 1-ml undiluted vitreous sample (obtained during vitreous surgery) in 9 patients who were treated with vitreoretinal surgery. Additionally, a CBC was performed and serum total immunoglobulin-E (IgE) was examined to evaluate any serologic/immunologic abnormalities. The results of abdominal computed tomography (CT) and chest CT were analyzed in this study, if they were performed within 6 months of OT diagnosis. From these images, the prevalence of granulomas or abscesses in other organs such as the lung or liver, as determined by a trained radiologist, were determined.
Best-corrected visual acuity (BCVA), intraocular pressure (IOP), slit-lamp biomicroscopy findings, and dilated fundus examination findings were reviewed in all the patients. Inflammation in the anterior chamber and vitreous chamber was graded based on the number of cells in a 1×1 mm slit beam under maximal light intensity and magnification [15]. Briefly, grade 0 indicated <1 cell; grade 0.5+, 1 to 5 cells; grade 1+, 6 to 15 cells; grade 2+, 16 to 25 cells; grade 3+, 26 to 50 cells; and grade 4+, >50 cells. Funduscopic findings were photographed with a Kowa VX-10 fundus camera (Kowa Co Ltd, Tokyo, Japan). Changes in granuloma size and location were evaluated using photographs from each follow-up visit. SD-OCT (Spectralis, Heidelberg engineering, Heidelberg, Germany) was performed to sectionally image the retina and view pathologic changes in eyes with granuloma and other vitreoretinal complications.
Patients with OT were treated with drugs or surgery based on symptom severity, inflammation, and retinal comorbidities. Drug therapy involved corticosteroids when intraocular inflammation was present. Systemic (oral prednisolone; 0.5–1 mg/kg/day loading dose and tapering) and topical (prednisolone acetate 1% four times a day) corticosteroids were used depending on the site of inflammation. Patients with eosinophilia or elevated serum IgE level were treated with albendazole (400 mg twice a day for two weeks). For patients with retinal comorbidities requiring surgery such as visually significant epiretinal membrane, vitreous opacity obscuring visual axis, and tractional or rhegmatogenous retinal detachment, pars plana vitrectomy was performed. Patients were separated into 4 groups, based on the medical treatment they received (i.e., combined steroid and albendazole, albendazole only, steroids only, or no treatment). For patients with a ≥3-month follow-up period, treatment response was evaluated based on clinical characteristics observed at the follow-up visits. These included BCVA, intraocular inflammation grades, symptom improvements, and recurrence rates. Recurrence was defined as returning intraocular inflammation or new granuloma development. Treatment outcomes were assessed in each treatment group.
For investigation of the probable infection sources of OT, we conducted a standardized interview and ensured complete responses through improved understanding of the participants to enhance the validity of the interview. During a face-to-face interview, a trained interviewer (medical doctor) used standardized interview procedures to collect data concerning history of eating raw animal tissues and contact with animals and soil during the period between January 2011 and June 2013. Both patients and controls were asked the same set of questions, including puppy/kitten exposure; history of ingestion of raw animal liver, raw meat, and raw animal blood; and occupation-associated contact with animals or soil. The data obtained also included the time of ingestion and the species of animals. For the interview, the questions and uniform nonverbal signals were presented in exactly the same way by one trained interviewer to avoid introducing biases into the responses.
Patients who visited our clinics during January 2011 and June 2013 and showed no abnormal ocular findings on complete ophthalmic examination were selected as controls. Among 59 control subjects who responded to our interview, 50 were matched for age (within 3 years) and sex with the 52 patients who responded to our interview. Between the included controls (n = 50) and the others (n = 9), there were no significant differences in demographic features and probable infection sources, except age (51.0±11.4 in the included controls and 28.3±19.6 in the excluded, P<0.001).
Descriptive statistical analyses were performed on the demographic data, clinical features, funduscopic and OCT findings, serologic markers, systemic involvement, and granuloma clinical course. Snellen BCVA measurements were converted into logarithmic minimum angle of resolution (logMAR) equivalent values for statistical analysis. The association between funduscopic and OCT findings and granuloma location was assessed by a chi-square test, which compared the frequencies of funduscopic findings in patients with posterior pole granulomas to those with peripheral granulomas. Treatment outcomes were compared between and within groups using the Wilcoxon signed-rank test and Mann–Whitney test, respectively. Continuous and interval data are reported as mean ± standard deviation. Statistical analyses were performed using SPSS for Windows (Ver. 18.0, Statistical Package for the Social Sciences, SPSS Inc., Chicago, IL), and a P value<0.05 was considered statistically significant.
The Institutional Review Board approved this study (Approval #: B-1101/120-102) and all patient data were anonymized for the analysis. The study adhered to the tenets of the Declaration of Helsinki.
The demographic and clinical features of the patients are presented in Table 1. Most OT patients were men (76 of 101, 75.2%), and the mean presentation age was 51.7±12.6 years (range: 21–77 years). Three of the 101 patients (3.0%) were known to have been infected with Toxocara before they were diagnosed with OT. In the other 98 patients, OT was the first symptom of toxocariasis. Systemic involvement of the Toxocara granulomas is also summarized in Table 1. CT images revealed liver (17.6%) or lung (42.9%) granulomas in a significant proportion of our patients.
Ninety-three of 101 patients (92.1%) were diagnosed with OT based on the presence of a retinal granuloma. Eight patients (7.9%) were diagnosed based on diffuse intraocular inflammation and a positive Toxocara antibody (ELISA). Of the 8 patients with nematode endophthalmitis, OT was confirmed in 2 patients, based on the presence of a peripheral granuloma, which became visible only after vitreous opacities had been cleared by vitrectomy. Depending on the location of the granuloma, Toxocara granuloma was classified as posterior pole (47 eyes, 50.5%), peripheral (41 eyes, 44.1%), or combined (both posterior pole and peripheral granulomas, 5 eyes, 5.4%). Intraocular inflammation was observed in 78 eyes (77.2%), most of which had intermediate uveitis (53 eyes, 67.9%). All OT cases were unilateral. Mean BCVA changed from 0.51 (20/64 Snellen equivalent) ±0.65 (range: no light perception [NLP] to 30/20) at baseline to 0.45 (20/56 Snellen equivalent) ±0.65 (range: NLP to 30/20) at the final visit. Seventeen (16.8%) and 14 (13.9%) patients had severe vision loss (BCVA<20/200 Snellen equivalent) at baseline and the final visit, respectively. Possible causes of vision loss in patients with OT include retinal damage caused by granuloma itself, comorbidities of OT (Table 1), and intraocular inflammation. Eosinophilia (>500 eosinophils/µl peripheral blood or ≥10% of total white blood cell count [16]) at the time of diagnosis was noted in 10 of 86 patients (11.6%) in whom CBC results were available. Increased serum IgE level (>100 unit/ml) was noted in 39 of 56 patients (69.6%). Mean ELISA titer for serum Toxocara IgG was 0.398±0.115 (range: 0.254–0.737).
A few photographs demonstrating retinal and vitreous findings are shown in Figure 1. In addition to retinal granuloma, patients with OT showed retinal nerve fiber layer (RNFL) defect (Figure 1A) in 32 of 101 eyes (31.7%), epiretinal membrane (ERM, Figure 2B and 3C) in 27 eyes (26.7%), vitreous opacity (Figure 1F) in 22 eyes (21.8%), retinal detachment (Figure 1E) in 13 eyes (12.9%), macular edema in 4 eyes (4.0%), and macular hole in 2 eyes (2.0%), as summarized in Table 1. Table 2 shows the association between these vitreoretinal comorbidities and the location of granuloma (posterior pole or peripheral retina). For example, eyes with a posterior pole granuloma had more frequent RNFL defects (53.2% vs. 7.3%, P<0.001) and ERMs (40.4% vs. 14.6%, P = 0.007) than eyes with a peripheral granuloma. In addition, vitreous opacity was observed more often in eyes with a peripheral granuloma than in those with a posterior pole granuloma (31.7% vs. 12.8%, P = 0.031).
Pathologic retinal changes were visible on SD-OCT images, which showed a moderately hyper-reflective round mass that sometimes had posterior shadowing (Figure 2). Granulomas were observed in almost all retinal layers. Secondary ERM was also commonly seen, and Figure 3 shows the course of retinal damage that leads to vision loss from granuloma and other OT-associated retinal pathologies.
Three patterns were noted in the clinical course of Toxocara granuloma: complete/partial granuloma resolution, persistent granuloma, and granuloma migration (Table 3). Granulomas completely or partially resolved in 36 of 93 patients (38.7%), with approximately half of these resulting in pigmentary scarring (Figure 4). Forty-two patients (45.2%) showed no significant changes in the size, number, or location of granulomas. Continuous or discontinuous granuloma migration within the eye was observed in 15 patients (16.1%; Figure 5). In continuous migration (12 eyes, 12.9%), the Toxocara granuloma migrated but remained adjacent to the originally observed location. However, in discontinuous migration (4 eyes, 4.3%), the granulomas moved discontinuously (relocated far from the originally observed location), increasing the total number of granulomas. One patient showed both types of intraocular migration.
Table 4 shows treatment responses to the various OT medical therapies. Four types of treatments—combined corticosteroid and albendazole use, corticosteroid use only, albendazole use only, and no treatment—were performed for our patients. The use of corticosteroids significantly decreased the degree of intraocular inflammation (Table 4), but there was no significant improvement in BCVA 3 months after the initiation of drug therapy in any group. In patients with active intraocular inflammation, there was no significant difference in changes in inflammation grade (P = 0.619), BCVA (P = 0.445), or symptomatic improvement (P = 0.274) between patients treated with only corticosteroids and those receiving a combination of corticosteroid and albendazole (Figure S1). In those without active intraocular inflammation, no significant difference was noted in the changes in inflammation grade (P = 1.00), BCVA (P = 0.855), and symptoms (P = 0.206) between patients treated with (Albendazole only group) and without albendazole (No treatment group). However, the 6-month rate of recurrence was significantly lower in the combination treatment group (17.4%) than in the steroid only group (54.5%, P = 0.045). In eyes without active inflammation, however, no recurrences were observed in both the Albendazole only and No treatment groups.
Thirty-two of 101 patients (31.7%) were surgically treated due to ERM (n = 19), vitreous opacity (n = 9), and/or retinal detachment (n = 2). The surgical outcomes (i.e., BCVA, anatomic success, symptomatic improvement, and recurrence) in each surgical indication are summarized in Table S1. Anatomic success, defined as complete removal of the ERM, vitreous opacity, or retinal reattachment, was achieved in 13 (68.4%), 8 (88.9%), and 2 (50%) patients with ERM, vitreous opacity, and retinal detachment, respectively.
Table 5 lists the probable sources of infections in adult patients with OT. In demographic features, no significant differences were observed between the patient and control groups in the mean age (51.6±13.0 in the patient group and 51.0±11.4 in the control group, P = 0.81) and sex distribution (men∶women = 38∶12 and 39∶13 in the patient and control groups, respectively, P = 0.92). Compared to healthy controls, there were no significant differences in the proportion of patients who had ingested raw animal blood, were exposed to puppies or kittens, or had occupation-associated contact with animals and/or soil. However, raw animal meat (71.2% vs. 52%, odds ratio [OR] = 2.28, P = 0.047) or cow liver (80.8% vs. 22.0%, OR = 14.9, P<0.001) was ingested significantly more often in OT patients than in normal controls.
Despite being the most prevalent human helminth infection in industrialized countries [17], toxocariasis remains relatively unknown to the public [2]. This study describes the pathologic changes caused by OT-associated retinal granulomas using SD-OCT images. The clinical course of granuloma was also examined, and we showed that intraocular granuloma migration is an important and distinguishing clinical feature of OT. In addition, our study showed an association between OT and ingestion of raw cow liver or uncooked meat. This information may also be helpful in diagnosing OT in adult patients.
Systemic and ocular manifestations of toxocariasis have rarely been reported in the same group of patients, and only a few such cases have been described in the literature [18]. In our study, both ocular larva migrans (OLM) and visceral larva migrans (VLM) were assessed in the same group of patients who had undergone ocular examination and liver or chest CT, although it was not proven that granulomas on CT images were caused by Toxocara infection. A significant proportion of our patients had liver or lung granulomas on CT images, and further study is needed to better understand the association between VLM and OLM. Given that the vast majority of patients (98 of 101 patients, 97.0%) with toxocariasis were diagnosed first with OT, a thorough ophthalmologic examination is important for the detection of human toxocariasis.
The present study identified several diagnostic serologic markers of Toxocara infection. The standard, current test for diagnosing human toxocariasis is detection of serum anti-toxocara IgG using an indirect ELISA based on the Toxocara larva antigen [2], [13]. Testing of intraocular fluids for Toxocara antibodies has been helpful in diagnosing toxocariasis in a few previous studies [19]–[21] and in some patients of our study. However, the low positive rates (33%) obtained by using the same cut-off value with serum antibodies (0.250) for vitreous antibodies, may not be acceptable for OT detection. Because there is no consensus on the cut-off titers for vitreous antibodies, further research on the diagnostic capabilities of vitreous ELISA is needed. Our study also revealed that serum IgE level is elevated in about 70% of OT patients. Eosinophilia was not as helpful as serum anti-toxocara IgG (ELISA) or total IgE evaluations, although it may indicate the presence of VLM, as shown in previous reports [2], [13].
Toxocara-associated RNFL defects, ERMs, and vitreous opacities are common comorbidities of OT. These features were associated with granuloma location, which suggests that careful examination for these associated complications is necessary in patients with retinal granulomas. Vitreous opacities and ERMs were common causes of vision loss in our OT patients, for which surgical treatment was needed. Secondary ERMs in eyes with OT progressed rapidly, resulting in severe retinal distortion and vision deterioration. SD-OCT revealed retinal granulomas and secondary complications, including ERM and tractional membranes, and thus played a critical role in the clinician's decision to perform surgery. Therefore, in patients with OT, SD-OCT may be an important imaging modality for diagnosis and decision making in clinical settings.
The migration of Toxocara larvae from the circulatory system into the posterior segment of the eye causes OT. Although migration is a key feature of Toxocara larvae, migration within the eye and its clinical significance have not been studied. In this report, we demonstrated a continuous and a discontinuous pattern of intraocular migration. These patterns have been individually reported in different case reports [22], [23], but their incidences have not been determined in longitudinal studies with a large number of patients. Our study showed that 12.9% and 4.3% of patients had continuous and discontinuous migration, respectively. In addition, our findings show clinical implications of the intraocular migration of Toxocara larvae. Continuous migration widened RNFL defects and discontinuous migration increased the number of RNFL defects since localized RNFL defects followed granuloma formation. In one case, discontinuous migration resulted in a macular granuloma, which resulted in significant macular destruction and subsequent vision loss, as well as a secondary ERM (Figure 3).
Although the mainstay treatment for OT is the use of corticosteroids to reduce ocular inflammation, this treatment has not yet been standardized. In our study, compared to steroid monotherapy, a combination of albendazole and corticosteroids resulted in a lower rate of recurrence over 6 months. The efficacy of this combination therapy, i.e., no ocular inflammation recurrence and visual acuity improvement, has been shown previously in one case [24]. Although our results showed no significant difference in BCVA before and after treatment, the reduced risk of OT recurrence favors the use of corticosteroids with albendazole in patients with severe OT. Therefore, we recommend using both corticosteroids and albendazole to minimize severe, recurrent inflammation and associated retinal damage.
Our demographic analyses revealed that OT predominantly occurred in men, and disease transmission from pets was less frequent than that reported previously [3]. Male predominance has been previously reported in Japanese [25] (male∶female ratio = 2.5∶1) and Korean [26] (4∶1) populations. In these studies, the mean age was >30 years, which is much higher than that that in studies conducted in Western countries, in which patients were generally <20 years old. Together, these results suggest that East Asian men may have a toxocariasis-related behavior, for example, the ingestion of raw cow liver, which is served in some restaurants in East Asian countries and is believed to be nutritionally beneficial for the middle-aged. Indeed, most reported cases from East Asia involved middle-aged men with a history of ingesting uncooked meat from infected animals [19], [25]–[28]. Contact with a puppy or kitten, which was reported in 82% of patients in a previous OT study [3], has been considered a major infection source, but was only reported in approximately 20% of our OT patients. This indicates that the infection source may differ based on geographic and behavioral differences [2], and clinicians should consider the local cultural context (e.g., food habits) when identifying the probable infection source in adult patients with OT.
Increasing public awareness about toxocariasis is the first step in reducing human exposure to Toxocara. Proper handwashing, limiting children's outdoor activity in sandboxes, appropriate disposal of dog and cat feces, and controlling infections in dogs and cats through deworming are the recommended prevention strategies, especially in children [29]. We suggest that OT may be prevented in many adults by avoiding uncooked meat ingestion, especially raw cow liver. Educating the targeted population (middle-aged, East Asian men) as well as clinicians regarding this disease may be effective in preventing the disease [30]. The Japanese government recently banned restaurants from serving raw cow liver. Although intended to prevent infection with a virulent strain of Escherichia coli, the action may also reduce the incidence of toxocariasis, especially if applied in East Asia.
Some limitations of our study should be considered. First, although our study included a relatively large cohort of patients compared to previous studies, it is a retrospective study, with intrinsic drawbacks that may introduce bias. The patients presented at >20 years of age, but it does not necessarily mean adult presentation. Additionally, our results on adult infection source may not necessarily extrapolate to the rest of the world as regional food habits vary widely. The serum IgE level was measured only in 55% of the included patients since we started our investigation of immunologic indicators of OT in October 2010. Although the decision regarding whether a patient with OT would be tested for IgE or not was not made by the clinician, selection bias may not be neglected.
Additionally, comparability issues exist in the comparison of treatment outcomes. The method of treatment was determined based on clinical presentation. In OT patients with active intraocular inflammation but without eosinophilia or elevated IgE levels, the current standard treatment for OT, i.e., corticosteroid, was administered. However, in the OT patients with eosinophilia or elevated IgE levels, anthelmintic treatment combined with corticosteroid was preferred due to the possibility of VLM. In this setting, it may not be feasible to compare treatment outcomes between combination therapy and steroid monotherapy groups; similarly, in patients without active inflammation, it may not be feasible to compare albendazole monotherapy and no treatment groups. However, as there were no differences in the baseline ocular characteristics between the two treatment groups, the ocular treatment outcomes could be compared despite the limitation. Further prospective randomized trials are required to compare treatment outcomes between the groups. Despite these limitations, our analyses may be valuable since an optimal treatment for OT has not yet been determined and the role of anthelmintics in OT has not been evaluated.
Another limitation of our study is related to the method of interview used for the investigation of probable infection sources of OT. We tried to avoid introducing biases in the interview procedures between patients and controls by using a standardized interview protocol. However, clinical diagnosis of the participants was not completely blind to the interviewer, possibly leading to bias.
In conclusion, the present study showed that intraocular granuloma migration is an important and distinguishing clinical feature of OT. Among the serologic markers, total IgE, in addition to anti-toxocara IgG antibody (ELISA), may be useful for the diagnosis of OT. In cases showing severe inflammation, combined corticosteroid and albendazole therapy may reduce inflammation and recurrence of OT. The risk of OT in adults may be reduced by avoiding ingestion of uncooked meat, particularly cow liver; however, further studies are required to validate this suggestion.
|
10.1371/journal.pgen.1007277 | Novel function of HATs and HDACs in homologous recombination through acetylation of human RAD52 at double-strand break sites | The p300 and CBP histone acetyltransferases are recruited to DNA double-strand break (DSB) sites where they induce histone acetylation, thereby influencing the chromatin structure and DNA repair process. Whether p300/CBP at DSB sites also acetylate non-histone proteins, and how their acetylation affects DSB repair, remain unknown. Here we show that p300/CBP acetylate RAD52, a human homologous recombination (HR) DNA repair protein, at DSB sites. Using in vitro acetylated RAD52, we identified 13 potential acetylation sites in RAD52 by a mass spectrometry analysis. An immunofluorescence microscopy analysis revealed that RAD52 acetylation at DSBs sites is counteracted by SIRT2- and SIRT3-mediated deacetylation, and that non-acetylated RAD52 initially accumulates at DSB sites, but dissociates prematurely from them. In the absence of RAD52 acetylation, RAD51, which plays a central role in HR, also dissociates prematurely from DSB sites, and hence HR is impaired. Furthermore, inhibition of ataxia telangiectasia mutated (ATM) protein by siRNA or inhibitor treatment demonstrated that the acetylation of RAD52 at DSB sites is dependent on the ATM protein kinase activity, through the formation of RAD52, p300/CBP, SIRT2, and SIRT3 foci at DSB sites. Our findings clarify the importance of RAD52 acetylation in HR and its underlying mechanism.
| DNA double strand breaks (DSBs) are the most dangerous type of DNA damage in cells. Homologous recombination (HR) is a DSB repair system in which a central player, RAD51, functions with several proteins, including RAD52. DSBs activate the DNA damage response signaling network, in which the ataxia telangiectasia mutated (ATM) protein plays a chief role, by phosphorylating numerous target proteins. As compared to phosphorylated proteins, relatively few acetylated proteins have been functionally characterized in DNA repair. In addition, beyond the roles in phosphorylation signaling, much less is known about whether ATM functions are linked with other protein modifications, such as acetylation. Here, we found that RAD52 at DSB sites is acetylated by p300/CBP acetyltransferases and then deacetylated by SIRT2/SIRT3 deacetylases. RAD52 acetylation is required for sustained RAD51 colocalization at DSB sites, and is therefore essential in HR. ATM is required for the recruitment of RAD52, p300/CBP and SIRT2/SIRT3 to DSB sites, and therefore is essential for RAD52 acetylation. Thus, the RAD52-acetylation state is critical for HR, and its regulation is linked to ATM signaling. Our work demonstrates the importance of the regulation of RAD52 acetylation in HR and its underlying mechanism.
| Ionizing radiation (IR) induces deleterious DNA lesions, such as DNA double-strand breaks (DSB). In response to DSBs, DNA damage response (DDR) signaling is induced. Ataxia telangiectasia mutated (ATM) protein kinase is one of the central players for phosphorylation-mediated DDR signaling, which is activated at DSB sites and phosphorylates numerous proteins, including the histone variant H2AX, and cell cycle checkpoint and DNA repair proteins [1]. Homologous recombination (HR) is an important mechanism for the repair of DSBs [2]. HR repairs DSBs through DNA strand invasion and exchange, in which the damaged DNA strand retrieves genetic information from an undamaged homologous DNA strand. After DSB formation, HR is initiated by a 5' to 3' end resection generating 3' single-stranded (ss) DNA overhangs. In mammalian cells, DSB end resection is mediated by the MRE11-RAD50-NBS1 (MRN)-CtIP complex and the EXO1 protein [3,4,5]. Afterwards, replication protein A (RPA) rapidly coats the 3'-overhang ssDNA regions, thereby removing secondary structures that form on the ssDNA region. Subsequently, the RPA coating the ssDNA regions is displaced by the RAD51 recombinase, to form a right-handed nucleoprotein filament. The RAD51 nucleoprotein filament then catalyzes DNA strand invasion and exchange between ssDNA and the homologous sequence within double-stranded (ds) DNA.
The replacement of RPA with RAD51 requires additional proteins, such as recombination mediators, because prior binding of RPA to ssDNA inhibits the nucleation of RAD51 on ssDNA. Biochemical studies using recombinant proteins demonstrated that the yeast Rad52 protein stimulates the Rad51-mediated displacement of RPA from ssDNA regions [6]. In the mouse, the targeted inactivation of RAD52 reduces HR and may be involved in certain types of DSB repair processes [7]. However, this mediator function of human RAD52 for the loading of RAD51 onto the RPA-coated ssDNA region has never been demonstrated, despite extensive biochemical analyses [2,8]. Instead, biochemical studies have revealed that the human BRCA2 protein, which does not have a yeast homologue, promotes the RAD51 nucleoprotein filament formation on RPA-covered ssDNA in vitro [9,10]. Therefore, instead of RAD52, BRCA2 is thought to mediate RAD51-dependent HR in human cells. This is supported by the fact that a knockdown of BRCA2 in human cells decreases the efficiency of IR-induced RAD51 foci formation. Interestingly, a RAD52 knockdown in BRCA2-knockdown or BRCA2-deficient cells almost completely inhibits IR-induced RAD51 foci formation [11], which suggests that human RAD52 could act as a RAD51 mediator or complement the RAD51-dependent pathway in HR. Since most previous biochemical studies of RAD52 have utilized an unmodified, recombinant RAD52 protein expressed in Escherichia coli, it is possible that a recombination mediator activity of RAD52 might only be revealed upon post-translational modifications, as discussed by San Filippo et al. [2].
RAD52 preferentially binds ssDNA [12] rapidly and tightly, by wrapping the ssDNA around itself [13]. In contrast, RAD52 binds dsDNA slowly and weakly, but changes the dsDNA mechanics probably by intercalating into the DNA helix [13]. RAD52 also interacts with RPA and RAD51 [14,15]. Both yeast and human RAD52 exhibit ssDNA annealing activity [12,16], which may be required in the steps following strand invasion mediated by RAD51 [17,18], as well as in the RAD51-independent single-strand annealing (SSA) pathway [2]. Human RAD52 also has a D-loop formation activity [19]. Both human and yeast RAD52 are multimeric proteins [6,20]. Three-dimensional reconstitution from electron microscopy images revealed that full-length human RAD52 exists as a heptameric ring [21]. The crystal structure of the amino (N)-terminal half of RAD52 revealed an undecameric ring with a highly positively-charged groove outside the ring [22,23]. The N-terminal half of human RAD52 encompasses the catalytic domain for homologous pairing. Structure-based alanine scan mutagenesis of the N-terminal half of RAD52 revealed that several lysine (K) residues within the positively-charged groove are essential for DNA binding [22,24]. The carboxyl (C)-terminal region of human RAD52 contains domains that interact with RAD51 and RPA.
Post-translational modifications, such as phosphorylation, ubiquitylation, small ubiquitin-like modifier (SUMO)ylation, and acetylation, regulate biological processes by controlling a wide variety of protein functions. Previously, some post-translational modifications of Rad52 were identified. Yeast Rad52 is modified by SUMO at the K10, K11, and K220 sites, and the SUMOylation is induced by a treatment with DNA-damaging agents [25]. SUMOylation of yeast Rad52 protects it from proteasomal degradation. Human RAD52 is also modified by SUMO, but the SUMOylation sites of human RAD52 differ from those of yeast Rad52. The in vitro SUMOylation sites of human RAD52 are K411, K412, and K414, which are located within the putative nuclear localization signal near the C-terminus (Saito et al., 2010). SUMOylation does not affect the biochemical activities of human RAD52, but mutations at SUMOylation sites inhibit RAD52 nuclear localization [26]. Nuclear phosphatase and tensin homolog on chromosome 10 (PTEN) was recently found to be involved in regulating RAD52 SUMOylation [27]. PTEN is also modified by SUMOylation, which is involved in the exclusion of the protein from the nucleus [28]. The function of the SUMOylation of human RAD52, however, remains poorly understood. Human RAD52 is also phosphorylated at tyrosine (Y) 104 by c-ABL tyrosine kinase upon exposure to IR, and the phosphorylation deficiency inhibits the IR-induced foci formation of RAD52 [29]. Phosphorylation at Y104 enhances the ssDNA annealing activity of RAD52 by increasing the binding specificity for ssDNA [30]. No other post-translational modifications of RAD52 have yet been identified.
Among the several post-translational modifications, acetylation occurs on specific lysine residues and is catalyzed by histone acetyltransferases (HATs). Histones are well-known target proteins for acetylation. Histone acetylation influences chromatin structure, thereby regulating a wide variety of DNA transaction processes, such as transcription [31], DNA replication [32], DNA recombination [33], and DNA repair [34]. HATs can also acetylate non-histone proteins, including some DNA repair proteins [35,36]. During HR, CtIP is deacetylated by SIRT6 histone deacetylase (HDAC), and deacetylation is required for DNA end-resection, although the specific acetyltransferase for CtIP has not yet been identified [37]. The HATs, p300 and CBP, accumulate at laser microirradiation- or I-SceI-induced DSB sites, and promote histone acetylation at DSB sites [34]. Whether the accumulated p300 and CBP at DSB sites also induce the acetylation of non-histone proteins involved in DSB repair, however, is unclear.
Here, we provide evidence for human RAD52 acetylation by p300/CBP upon DSB induction, and its involvement in RAD51 localization at DSB sites during HR repair. We show how HR is regulated via RAD52 acetylation and reveal the link between the acetylation event and the ATM-dependent phosphorylation.
The acetylation of non-histone DNA repair proteins has attracted recent attention [35,36,37]. We searched for new HAT substrates among human DNA repair proteins, and found that human RAD52 interacted with CBP, one of the well-known HATs (Fig 1A). FLAG-tagged CBP coimmunoprecipitated with RAD52, but not with the glutathione S-transferase (GST) control. RAD52 also specifically interacted with p300, which is structurally and functionally similar to CBP (S1A Fig). An in vitro acetylation assay was performed to examine whether RAD52 is acetylated by either CBP or p300. DNA polymerase β was used as a positive acetylation control substrate [35]. Strikingly, the incubation of human RAD52 with p300 or CBP, in the presence of acetyl CoA, promoted RAD52 acetylation, which was detected by immunoblotting using an anti-acetyl lysine antibody (Fig 1B). RAD52 acetylation was also confirmed by an in vitro acetylation assay using 14C-labeled acetyl CoA; the 14C-labeled acetyl group was transferred onto the ε-amino group of the lysine residue. Acetylation of RAD52 was specifically detected when RAD52 was incubated with CBP (Fig 1C; lane 6) or p300 (S1B Fig; lane 6) in the presence of 14C-labeled acetyl CoA. Notably, RAD52 was more efficiently acetylated than the control substrate, DNA polymerase β (lane 3 in Fig 1C and S1B Fig). By contrast, neither RAD51 nor DNA polymerase κ [38], which are key factors in homologous recombination, was acetylated by CBP or p300 in vitro (S1C–S1E Fig).
To map the acetylated sites in RAD52, we performed an in vitro acetylation assay with the N- or C-terminal half of RAD52 (Fig 1D and S1F Fig). Both RAD52 fragments were acetylated by CBP or p300, although the C-terminal fragment of RAD52 (209–418) was more efficiently acetylated. To identify the acetylated residues, we performed a liquid chromatography mass spectrometry (LC-MS) analysis using in vitro acetylated full-length, N-terminal, and C-terminal RAD52 fragments. We identified 11 acetylation sites in RAD52 (FL), and two additional acetylation sites (K133, K177) in the C-terminally truncated RAD52 (N) (S1–S4 Tables and Fig 2A). We next purified two full-length RAD52 mutants, one containing arginine substitutions at the 11 acetylation sites (11xR), and the other containing arginine substitutions at all 13 identified acetylation sites (13xR). Using 11xR and 13xR, we confirmed that the p300/CBP-mediated acetylation of RAD52 was diminished when 11 or all 13 of the identified lysine residues were mutated to arginine (Fig 2B). The identified sites are well conserved among different species (S2 Fig). Lysines 411, 412, and 414 were previously identified as SUMOylation sites, and overlap with the nuclear localization signal (NLS) [26]. Mutations at these SUMOylation sites inhibited RAD52 nuclear localization. Another notable acetylation site is lysine 133, which is an important site for DNA binding [24]. To study the RAD52 acetylation in detail, we used the acetylation-deficient, lysine-to-arginine substituted mutants in in vivo studies (S3 Fig). The 11xR mutant and the unmodified RAD52 (Wt) displayed similar ssDNA binding activities, suggesting that the multiple lysine to arginine substitutions do not affect the RAD52 activity (S4A and S4B Fig).
To examine whether RAD52 is acetylated in human cells, we expressed an N-terminally FLAG-tagged RAD52 in human embryonic kidney 293 (HEK293) cells, and immunoprecipitated RAD52 using anti-FLAG antibody-conjugated agarose. Overexpression of CBP induced RAD52 acetylation, based on immunoblotting using an anti-acetyl lysine antibody (S5A Fig).
To evaluate the acetylation status and localization of RAD52 in vivo, we produced anti-acetyl RAD52 antibodies against the acetylated lysine residues 274 or 323. The antibody specificity for acetylated RAD52 was confirmed by comparison with the in vitro acetylated or non-acetylated RAD52 protein (Fig 3A and 3B). Immunoblotting with each antibody revealed a positive reaction to the acetylated RAD52 protein (Fig 3A and 3B). We then used these anti-acetyl RAD52 antibodies to examine whether the induction of DSBs changes the acetylation status of RAD52 in cells. We used the chemical DSB inducer doxorubicin, because it constitutively produces DSBs in the cells, and thus we thought it may induce a stronger DSB signal. Doxorubicin induced RAD52 acetylation in repair-proficient mesenchymal stem cells (MSCs; Fig 3C) [39]. A band shift of in vivo acetylated RAD52 was observed following doxorubicin treatment (Fig 3D). The migration distance of the in vitro acetylated RAD52 was indistinguishable from that of the non-acetylated RAD52 in sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) (Fig 1B). This was not the case for the acetylated RAD52 produced in vivo (Fig 3A), raising the possibility that the in vivo acetylation induces additional modifications of RAD52. Furthermore, we expressed a RAD52 construct in which 10 of the acetylation sites, except for lysine 411, 412, and 414 involved in the nuclear localization of RAD52 [26], were substituted with arginine (10xR; Fig 3E and S3 Fig), and performed immunoblotting analyses to examine whether the mutations affect the interaction between RAD52 and the anti-acetyl RAD52 antibody. We did not achieve a positive signal with the anti-acetyl RAD52 antibody (K323) in the presence of the 10xR mutant (Fig 3E) or in the absence of RAD52-HA expression by the vector (S6 Fig). These findings indicate that the anti-acetyl RAD52 antibody (K323) specifically detects and precisely evaluates the cellular acetylation status of RAD52. The anti-acetyl RAD52 antibody (K274) also specifically detected the acetylation of RAD52, but the specificity of the anti-acetyl RAD52 antibody (K323) against acetylated RAD52 was superior to that of the anti-acetyl RAD52 antibody (K274) (Fig 3C and 3E).
To examine the RAD52 acetylation at DSB sites, MSCs expressing RAD52 (Wt) or the RAD52 (10xR) mutant were irradiated with 8 Gy γ-rays and examined by an immunofluorescence approach. IR treatment transiently induces DSBs in cells, and thus IR is better suited for examining the level of RAD52 acetylation as a function of time. RAD52 foci, which colocalize with phosphorylated H2AX (γH2AX), were detected with both wild-type and mutant RAD52 (Fig 3F). In contrast, radiation-induced acetylated RAD52 (assessed using the anti-acetyl RAD52 antibody) was only detected in cells expressing RAD52 (Wt) protein at 1 h after IR. Acetylation of RAD52 (Wt) was not detected at 6 h after IR, suggesting that deacetylation of RAD52 occurs between 1–6 h at DSB sites. Significant acetylated RAD52 signals were not detected in cells expressing the RAD52 (10xR) protein throughout the time period studied, indicating that the anti-acetyl RAD52 antibody specifically recognized acetylated RAD52 by immunofluorescence. We also investigated the distribution of p300 and CBP in these irradiated cells, and observed p300 and CBP foci, which colocalized with γH2AX, at 1 h and 6 h after IR in human fetal lung fibroblast (MRC5) cells (S5B and S5C Fig). This finding is consistent with a previous report that describes the recruitment of p300/CBP to laser-induced DSB sites [34]. The p300 and CBP foci also colocalized with the RAD52 foci, as observed at 1 h or 6 h after IR in HEK293 cells (S5D and S5E Fig). In addition, the specific interaction of p300 or CBP with RAD52, but not GST, was detected by dithiobis succinimidyl propionate (DSP)-mediated in vivo cross-linking experiments in exogenously p300- or CBP-overexpressing HEK293 cells (S5F and S5G Fig). The interaction of p300 with RAD52 was increased upon doxorubicin treatment (S5F Fig). Finally, the knockdown of both p300 and CBP inhibited the DSB-induced acetylation of RAD52 (Fig 3G and 3H; S5H and S5I Fig). Taken together, our findings indicate that IR induces p300/CBP foci formation and acetylation of RAD52 at DSB sites.
By analogy with yeast Rad52, human RAD52 is reported to bind to RPA-coated ssDNA followig resection. RAD52 is also thought to bind to both ssDNA and dsDNA during the DNA strand exchange reaction. The N-terminal half of RAD52 contains a DNA binding region, whereas the C-terminal half contains an RPA binding region. Therefore, using an in vitro acetylation assay, we examined whether RAD52 binding to DNA or RPA affects the level of its acetylation. The addition of linear (L) ssDNA to the reaction mixture for the in vitro acetylation of RAD52 (FL) decreased the level of RAD52 acetylation (Fig 4A). The addition of circular (C) or linear dsDNA also decreased the acetylation of RAD52 (FL), but less than that of the linear ssDNA. Auto-acetylation of CBP was not affected by the addition of these DNA substrates. The N-terminal half of RAD52, RAD52 (1–212), contains the DNA-binding domain (Fig 2A). The acetylation of RAD52 (1–212) was completely inhibited by the addition of any one of the DNA substrates (Fig 4B), whereas the acetylation of the C-terminal half of RAD52, RAD52 (209–418), was not affected by the addition of the DNA substrates (Fig 4C). RAD52 (209–418) contains the RPA-binding domain (Fig 2A), and RAD52 (209–418) acetylation was inhibited by the addition of RPA in a dose-dependent manner (Fig 4D). Therefore, these results suggest that at least some of the acetylation sites of RAD52 are located at the DNA and RPA interacting surfaces.
HATs and HDACs regulate protein acetylation levels. Therefore, we next examined which HDACs are involved in the deacetylation of acetylated RAD52. After RAD52 acetylation by CBP in vitro, linear ssDNA was added to the reaction mixture to inhibit further RAD52 acetylation. The reaction mixture was divided, and each sample was subjected to an in vitro deacetylation assay using recombinant HDAC proteins. Among the HDAC proteins examined, the recombinant HDAC3/NCOR2 complex, SIRT2, and SIRT3 proteins deacetylated the acetylated RAD52 (Fig 5).
Since the acetylated RAD52 appeared to be deacetylated at DSB sites after IR (Fig 3F), we examined whether the identified HDACs for RAD52 were recruited to DSB sites after IR. We observed that SIRT2 and SIRT3, but not HDAC3, colocalized with γH2AX and/or RAD52 at 1 h after IR (Fig 6A–6F). The SIRT2 and SIRT3 foci colocalized with γH2AX and RAD52 even at 6 h after IR (Fig 6A, 6B, 6E and 6F), suggesting that SIRT2 and/or SIRT3 are involved in the deacetylation of acetylated RAD52 at DSB sites. Therefore, we examined whether the knockdown of Sirt2 or Sirt3 affects the decrease of RAD52 acetylation observed at 6 h after IR. The expression of SIRT2 and SIRT3 was effectively reduced by a small interfering (si)RNA treatment (Fig 7A and 7B, S7A and S7B Fig). Importantly, acetylated RAD52 was maintained at DSB sites even at 6 h after IR following SIRT2 or SIRT3 depletion (Fig 7C). Therefore, our results strongly suggest that SIRT2 and SIRT3 deacetylate RAD52 at DSB sites.
The RAD52 protein localized to DSB sites after IR (Figs 3F and 8A). Therefore, we examined whether lysine to arginine substitutions at the acetylation sites influence the accumulation of RAD52 at DSB sites. The RAD52 (13xR) mutant (with arginine substituted for lysine at the 13 identified acetylated lysine sites, including 3 substitutions in the C-terminal NLS sequence; S3 Fig) localized to the cytoplasm rather than the nucleus, independently of IR (Fig 8B). Interestingly, the RAD52 (13xR) foci colocalized with γ-tubulin, indicating that RAD52 can localize to the centrosome.
To examine the acetylation-deficient effect of the RAD52 (13xR) mutant in the nucleus, we constructed a RAD52 (13xR) mutant with an N-terminally fused NLS (NLS-RAD52 [13xR]; S3 Fig). NLS-RAD52 (13xR) was detected in the nucleus, but did not colocalize with γH2AX in MRC5 cells at 6 h after IR (Fig 8C). We next examined the kinetics of the colocalization of RAD52 (Wt) and RAD52 acetylation-deficient mutants with γH2AX in irradiated MSCs. The control NLS-RAD52 (Wt) foci colocalized with γH2AX for 6 h after IR. In contrast, the NLS-RAD52 (13xR) foci initially colocalized with γH2AX, but then dissociated from γH2AX within 2 h after IR (Fig 9A and 9C). The RAD52 C-terminal region, including the acetylation sites K411, 412, and 414, is known to be essential for the nuclear localization of RAD52. Therefore, we next used the RAD52 (10xR) mutant harboring a normal NLS sequence for analyzing the acetylation effects of RAD52 on IR-induced foci-formation at DSB sites in cells. The RAD52 (10xR) mutant foci similarly showed an initial colocalization with γH2AX, followed by an early dissociation (Fig 9B and 9D). Strikingly, during the first 30 min or 1 h after γ-irradiation, both NLS-RAD52 (13xR) and RAD52 (10xR) showed clear colocalization with γH2AX (Fig 9A–9D).
In order to confirm the effect of the 10xR mutation on the cellular localization of RAD52 expressed by the native promoter, we generated RAD52 (Wt or 10xR) knock-in cells by using CRISPR/Cas9-mediated genome editing. The DNA donor plasmid shown in S9 Fig was constructed by the Multisite Gateway method, and was co-transfected with CRISPR Nuclease Vector plasmids into HeLa pDR-GFP or HEK293 cells. The expression of the HA-tagged RAD52 proteins from the native promoter was confirmed by an immunoblotting analysis (S9F and S9G Fig). We then used the knock-in cells for an immunostaining analysis. In order to deplete the RAD52 protein expressed from the untargeted allele, the knock-in cells were treated with an siRNA targeted to the 3'UTR region of RAD52 (S7C Fig). Consistent with the results described above, the colocalization of RAD52 with γH2AX foci was decreased in the 10xR mutant at 6h after irradiation, whereas both the Wt and 10xR RAD52 proteins colocalized with γH2AX at 1h after irradiation (S10A and S10B Fig).
To clarify the critical acetylation sites involved in this colocalization defect, we further examined various RAD52 proteins mutated in several functional domains, such as the highly conserved region (K133R, K133/K177R), the RPA binding region (K262R), and the RAD51 binding region (K323R), and also mutated residues outside the domains (190/192R). We also used the RAD52 (8xR) mutant containing multiple mutations, except in the highly conserved region and the C-terminal NLS region (S3 Fig). In contrast to the results obtained with the NLS-RAD52 (13xR) and RAD52 (10xR) mutants, the colocalization of these RAD52 mutants (K133R, K133/177R, K190/192R, K262R, K323R, and 8xR) with γH2AX was not affected after γ-irradiation (Fig 10A and 10B), indicating that this defect is associated with all of the acetylation sites except for those in the C-terminal NLS region. These findings demonstrate that RAD52 acetylation is required for its sustained retention at DSB sites.
ATM is a pivotal mediator of signal transduction in response to DSBs. As RAD52 acetylation was induced by DSBs, we examined whether its acetylation was triggered by ATM. KU55933 is a potent and specific ATM kinase inhibitor (ATMi). To verify the functionality of the inhibitor, we observed that the ATM-mediated phosphorylation of CHK2 on threonine 68 (T68) was inhibited in the KU55933-treated cells used in our present study. We found that the doxorubicin-induced acetylation of RAD52 was inhibited by the KU55933 (ATMi) treatment (Fig 11A). Consistent with this result, the knockdown of ATM by siRNA treatment decreased the doxorubicin-induced acetylation of RAD52 (Fig 11B, 11C and 11D).
The IR-induced acetylation of RAD52 was also inhibited by the treatment with KU55933 (ATMi) (Fig 11E). Notably, the ATM inhibition with KU55933 (ATMi) did not completely diminish the γH2AX foci formation, consistent with the previous findings that the IR-induced phosphorylation of H2AX can be redundantly affected by ATM or DNA-PK [40]. These γH2AX foci did not colocalize with the RAD52 foci, indicating that the treatment with KU55933 (ATMi) inhibited the accumulation of RAD52 at DSB sites (Fig 11E and 11F). The IR-induced accumulation of both p300 and CBP at DSB sites was inhibited by the treatment with KU55933 (ATMi) in MRC5 cells (Fig 12A–12D). ATM is involved in the DNA damage-induced phosphorylation of p300 at serine 106 (S106) [41]. We used p300 mutant proteins in which S106 was substituted with alanine (S106A) or aspartic acid (S106D), which mimics the phosphorylated state (Fig 12E). Both mutant proteins colocalized with γH2AX after IR, suggesting that the ATM-mediated phosphorylation of p300 at the S106 site is not involved in its colocalization at DSB sites. The colocalization of these two mutant proteins with γH2AX was inhibited by KU55933 (ATMi). Interestingly, the radiation-induced accumulation of SIRT2 and SIRT3 at DSB sites was also inhibited by the treatment with KU55933 (ATMi) (Fig 13A–13E). These ATM-associated events were not caused by non-specific effects, because the colocalization of 53BP1 with γH2AX was not inhibited by KU55933 (ATMi) (Fig 13F). The number of foci of RAD52, p300, CBP, SIRT2, and SIRT3 decreased in the presence of KU55933 (ATMi) (Figs 11F, 12C, 12D, 13C and 13D), whereas the cellular levels of these proteins did not significantly change (S8A–S8E Fig). Thus, ATM inhibition prevents the accumulation of RAD52, p300/CBP, and SIRT2/SIRT3 at DSB sites, thereby causing the intra-cellular diffusion of these proteins, even after IR treatment. Therefore, our results suggest that the interaction of RAD52 with p300 and CBP will be reduced by ATM inhibition, thereby decreasing the acetylation of RAD52. Taken together, our results indicate that the DSB-induced acetylation of RAD52 occurs in the vicinity of DSB sites in an ATM-associated manner.
We next examined whether RAD52 acetylation influences the accumulation of the RAD52-associated proteins, RPA and RAD51, at DSB sites. Expression of the RAD52 (10xR) mutant protein disrupted the IR-induced colocalization of RAD51 with γH2AX at 6 h in MSCs (Fig 14A and 14B) and at 4 h in HEK293 cells (Fig 14C and 14D). Furthermore, RAD52 (10xR) mutant expression in MSCs did not affect RAD51 foci formation or colocalization with γH2AX from 0.5 to 2 h after IR (Fig 14E). A time course experiment revealed that RAD51 colocalized with γH2AX at 2 h after irradiation in both RAD52 (Wt) and RAD52 (10xR)-expressing HEK293 knock-in cells, with no significant difference (Fig 15A and 15B). Thereafter, at 4 and 6 h after irradiation, the colocalization decreased only in the RAD52 (10xR)-expressing cells (Fig 15A and 15B). These findings suggest that RAD52 acetylation is dispensable for the initial loading of RAD51 at DSB sites, but is required for the sustained retention of RAD51 at DSB sites.
If the DNA resection is affected by RAD52 acetylation, then RPA foci formation should also be affected. However, the expression of the RAD52 (10xR) mutant protein did not affect the colocalization of RPA with γH2AX (Fig 16A). Therefore, our result suggests that DNA resection is not affected by RAD52 acetylation. Previously, BRCA1 was demonstrated to function in the loading of RAD51 at DSB sites via the PALB2-mediated interaction with BRCA2 [42]. IR-induced phosphorylated BRCA1 foci were observed at DSB sites in RAD52 (10xR) and NLS-RAD52 (13xR)-expressing cells (Fig 16B). These findings suggest that the non-acetylated RAD52 protein disturbs the colocalization of RAD51 at DSB sites, but does not influence BRCA1 foci formation. These results are consistent with the aforementioned finding that the initial loading of RAD51 at DSB sites was not affected by RAD52 acetylation.
RAD52 depletion is reportedly synthetically lethal with the BRCA2 deficiency, and inhibits cell growth in BRCA2-deficient cells [11]. Therefore, we examined whether BRCA2 depletion also inhibits cell growth in RAD52 (10xR)-expressing cells. BRCA2 depletion inhibited cell growth in RAD52 (10xR)-expressing cells, but not in RAD52 (Wt)-expressing cells (Fig 17A).
Expression of the RAD52 (10xR) mutant did not affect the γH2AX foci formation after irradiation (Fig 9B and 9E). This result might be due to the existence of a backup DSB repair system by the NHEJ pathway, and is consistent with the previous report showing that the inactivation of mouse RAD52 reduces HR, but does not affect the resistance to ionizing radiation [7]. However, the NHEJ pathway is dispensable for the repair of cross-linking DNA damage, but the HR pathway is required for its repair with the Fanconi anemia DNA repair pathway [44]. In the survival assay of cells treated with the DNA cross-linker cisplatin, the RAD52 (10xR)-expressing cells were more sensitive to cisplatin than the RAD52 (Wt)-expressing cells (Fig 17B). These results suggest that the acetylation of RAD52 is involved in HR repair. Therefore, we next examined whether the expression of the RAD52 (10xR) mutant affects the HR efficacy. We quantified IR-induced sister chromatid exchanges in HEK293 cells expressing an empty vector or a vector encoding RAD52 (Wt) or RAD52 (10xR). Expression of RAD52 (Wt) did not influence the IR-induced sister chromatid exchanges, whereas the RAD52 (10xR) mutant expression decreased the frequency of sister chromatid exchanges (Fig 17C).
In order to confirm the requirement of RAD52 acetylation for HR repair, we used a reporter assay with a cell line bearing a direct repeat green fluorescent protein (DR-GFP) reporter cassette [43,45,46,47]. The DR-GFP reporter cassette comprises two inactive GFP genes in a direct repeat orientation. One of the genes, SceGFP, contains an I-SceI cleavage site that is absent in the human genome. The other gene, iGFP, comprises the internal GFP fragment. HR repair between SceGFP and iGFP is induced when a specific DSB at the I-SceI site is introduced by the expression of I-SceI endonuclease. Since HR repair generates an intact GFP gene, the HR repair efficiency can be monitored as the frequency of GFP-positive cells (Fig 18A).
Using this HR assay system, we first examined the impact of the depletion of RAD52, RAD51, and BRCA2 proteins. These proteins were efficiently depleted by siRNA treatment (S7C–S7G Fig). The depletion of RAD51 almost completely inhibited I-SceI-induced HR repair, as expected (Fig 18B). The HR repair efficiency was reduced more by the BRCA2 depletion than by the RAD52 depletion (S11 Fig), which is consistent with the notion that BRCA2 and RAD52 function in different pathways of RAD51-dependent HR repair [11]. Then, we examined the impact of the RAD52 acetylation-deficient mutation on HR repair (Fig 18C and 18D). We used two types of HeLa pDR-GFP cells, expressing either the HA-tagged RAD52 (Wt) or RAD52 (10xR) protein. The RAD52 (Wt) and RAD52 (10xR) proteins were expressed by the native and CMV promoters, respectively, and the expression levels of the proteins were almost the same in both types of cells (Fig 18E and S9F Fig). The endogenous untagged RAD52 protein was depleted by a treatment with an siRNA targeted to the 3'UTR region of RAD52 (S7C Fig). In both cell lines, the expression of the RAD52 (10xR) protein inhibited HR repair (Fig 18C and 18D). Collectively, these findings demonstrate that the acetylation of RAD52 is required for HR repair.
The interaction of human RAD52 with human RAD51 was previously detected in a yeast two-hybrid analysis [15]. The interactions of yeast Rad52 with yeast Rpa1, Rpa2, Rpa3 and Rad51 have also been detected [14]. Therefore, we analyzed the interactions between the human acetylation mimic RAD52 (10xQ) and its target-proteins, using a yeast two-hybrid analysis. Glutamine (Q) is widely used to mimic acetylated lysine (K), because the effect of the lysine-to-glutamine substitution is similar to the effect of the acetylation of lysine [48]. The yeast cells lacked p300 and CBP. Therefore, we constructed the plasmids expressing either GAL4-DBD or the NLS-GA4-AD-fused RAD52 (10xQ) mutant for a yeast two-hybrid analysis (S3 Fig). When the NLS-GAL4-activation domain (AD)-fused protein interacts with the GAL4-DNA binding domain (DBD)-fused protein, the reporter gene, His3, is expressed, and the yeast cells show growth on an SC-Leu-Trp-His agar plate containing 25 mM 3-Amino-1,2,4-Triazole (3AT). Interactions of the GAL4-DBD-fused human RAD52 (Wt) with the NLS-GA4-AD-fused human RAD52 (Wt), RAD51, RPA1, RPA2 or RPA3 were observed in the yeast two-hybrid analysis (S12 Fig). Another reporter gene, lacZ, can also be used in this yeast two-hybrid analysis system. Therefore, we quantitatively examined the protein-protein interactions of RAD52, by using a liquid β-galactosidase assay (Fig 19A–19E). Both RAD52 (Wt) and RAD52 (10xQ) were expressed almost equally in yeast cells (Fig 19F). The self-interaction of RAD52 was increased 1.7-fold by the 10xQ mutation (Fig 19A). The interactions of RAD52 with RAD51, RPA1, RPA2 and RPA3 were remarkably increased by the 10xQ mutation (Fig 19B–19E). These results suggest that the interactions of RAD52 with these proteins are enhanced by its acetylation.
Based on our findings, we propose the following working model (Fig 20). Following DSB formation, human RAD52 and p300/CBP are recruited to DSB sites, and interact with each other near the DSB sites, thereby inducing RAD52 acetylation. Although 13 lysine residues can be acetylated, the acetylation efficiency of each site is probably different. SIRT2 and SIRT3 are also recruited to DSB sites shortly after DSB induction, and deacetylate RAD52. During HR, RAD52 interacts with RPA or DNA. These interactions may prevent the ongoing acetylation of RAD52 by p300/CBP. As a result, the acetylation level of RAD52 is diminished by SIRT2 and SIRT3. The acetylation sites of RAD52 are located in the DNA binding region (K133), and also within regions involved in protein-protein interactions. Indeed, the acetylation-mimicking RAD52 showed increased interactions with RPA and RAD51 in yeast cells (Fig 19B–19E), and acetylated RAD52 binds ssDNA more robustly than non-acetylated RAD52 in vitro (S13 Fig). These findings suggested that acetylated RAD52 plays a critical role in the maintenance of RAD51 recruited to DSB sites. Non-acetylated RAD52 dissociates prematurely from the DSB sites, and thus impairs the retention of RAD51 at the DSB site and prevents the completion of HR. The change in the levels and sites of acetylation during HR might control several activities of the multifunctional RAD52 protein.
HR is a multistep process mediated by the concerted actions of many proteins. Regulation of each protein in both positive and negative manners is probably required to achieve the multistep and multi-enzyme processes. At present, the precise molecular mechanisms by which RAD52 exerts its functions in HR have not been elucidated. It is possible, however, that the strength of the RAD52 interactions with DNA and proteins could change during HR, which would enable RAD52 to change interaction partners and act as a multifunctional protein. Acetylation may provide important contributions to this regulation.
Feng et al. demonstrated that the depletion of human RAD52 has small effects on decreasing HR and IR-induced RAD51 foci formation, as compared to the depletion of human BRCA2 [11]. They also demonstrated, however, that the depletion of human RAD52 in a BRCA2-depleted background further impairs both HR and RAD51 foci formation [11]. Based on these observations, they proposed that two pathways lead to RAD51-dependent HR: One is a dominant pathway in which BRCA2 mediates the recruitment of RAD51 to DSB sites, and the other is a pathway in which RAD52 alternatively mediates RAD51 recruitment [11]. The latter pathway becomes evident when the dominant BRCA2-mediated pathway is disrupted. Importantly, in the present study, non-acetylatable RAD52 impaired the localization of RAD51 foci at DSB sites, even in the presence of the dominant BRCA2 pathway, suggesting that the non-acetylatable RAD52 may competitively interfere with the BRCA2-mediated pathway. With regard to the reduced co-localization of RAD51 with γH2AX, the acetylation-defective RAD52 10xR mutant may have a dominant-negative effect by binding to RAD51 and preventing the localization of RAD51 at DSB sites. In the presently determined kinetics of RAD51 foci, the RAD52 10xR mutant does not inhibit the initial step of RAD51 recruitment to DSB sites in the presence of BRCA2. We suppose that RAD51 transiently accumulates and interacts with RAD52 at DSB sites; however, RAD51 might subsequently dissociate from the DSB sites along with RAD52 in this mutant. Here we have described the novel post-translational acetylation of RAD52, and demonstrated that the failure to acetylate RAD52 critically impacts HR. Our results suggest that an inhibitor of RAD52 acetylation could be exploited for anticancer therapy.
Our results demonstrated that DSB-induced RAD52 acetylation requires ATM kinase (Fig 20). We also found that inhibiting the function of ATM caused a decrease in the accumulation of RAD52 and p300/CBP at DSB sites. From these results, we drew the following hypothesis: In the absence of DSBs, p300, CBP, and RAD52 are diffused in the nucleus. After DSB-induction, these proteins accumulate at DSB sites, which will promote the interaction between RAD52 and p300 or CBP, thereby inducing the acetylation of RAD52. Accordingly, in the absence of functional ATM, the interaction of RAD52 with p300 and CBP will not be promoted, leading to the loss of acetylated RAD52.
Our findings revealed that the accumulation of SIRT2 and SIRT3 at DSB sites is also triggered by ATM. Similar results were reported for the other HDAC, SIRT1, with ATM being required for its accumulation at DSB sites [49]. Human RAD52 is phosphorylated by c-ABL, and this phosphorylation is required for the IR-induced foci formation of RAD52 [29]. The activation of c-ABL is induced in response to various types of DNA damage, including DSB, and the activation depends on ATM. Therefore, ATM might be required for the IR-induced foci formation of RAD52 at DSB sites through the c-ABL-mediated phosphorylation of RAD52. ATM functions in the DNA damage-induced phosphorylation of p300 [41]. Therefore, we examined whether ATM-mediated phosphorylation is required for the accumulation of p300 at DSB sites, and concluded that p300 phosphorylation is dispensable for its accumulation at DSB sites. To our knowledge, there are no reports that these HATs and HDACs possess DNA binding activity. Therefore, it is unclear how they accumulate near DSB sites. One possibility is that they are recruited to DSB sites via interactions with a protein that exists at DSB sites. ATM phosphorylates several proteins at DSB sites. Taken together, our results suggest that such an ATM-phosphorylated protein at DSB sites interacts specifically with several HATs and HDACs, thereby recruiting them to DSB sites. In contrast to the requirement for ATM in the accumulation of several HATs and HDACs at DSB sites, SIRT1 is required for the accumulation of ATM at DSB sites [49]. Furthermore, ATM is also acetylated by another HAT, Tip60 [50]. ATM acetylation enhances its activation as a protein kinase in response to DNA damage. Thus, there is interplay between acetylation and phosphorylation in response to DNA damage.
The roles of HATs and HDACs are important topics in many fields of biology and medicine. In this paper, we found the novel roles of HATs (p300/CBP) and HDACs (SIRT2 and SIRT3) in the HR process through the acetylation of human RAD52, indicating that human RAD52 is required for HR, depending on its acetylation status. We further demonstrated that ATM protein kinase is required for DSB-induced RAD52 acetylation, as well as for the accumulation of RAD52, p300/CBP, SIRT2, and SIRT3 at DSB sites. These findings indicate the presence of crosstalk between acetylation and phosphorylation. Therefore, the ATM kinase activation/RAD52 acetylation axis may be important for HR repair.
At DSB sites, several HATs and HDACs regulate histone acetylation, which is required for DNA repair. In addition to histone acetylation, we have demonstrated that HATs (p300/CBP) and HDACs (SIRT2 and SIRT3) directly regulate the acetylation of the non-histone protein, RAD52. We speculate that HATs and HDACs target more DSB repair proteins at DSB sites. These findings provide important information for future studies using in vitro reconstitution systems in the context of chromatin, to clarify the molecular mechanisms of DSB repair.
Confluent cells were exposed to X-rays at a dose rate of 1 Gy/min at room temperature. The cells were immediately subcultured in media with bromodeoxyuridine (3 μg/ml). After adding colcemid for 6 h, the cells were harvested and treated with a hypotonic KCl solution (75 mM) at 37°C for 20 min, followed by methanol:acetic acid (3:1) fixation. After three rounds of fixation, the cells were dropped onto slides. The slides were treated with Hoechst 33258 (10 μg/ml) for 20 min and exposed to a black light for 30 min at 55°C. Finally, the slides were treated with 2XSSC (saline sodium citrate) solution for 20 min at 65°C. The slides were stained using 5% filtered Giemsa solution mixed in Gurr. Images of metaphase cells were obtained using a Zeiss Axioplan microscope (Carl Zeiss) equipped with a QImaging Exi Aqua Cooled CCD camera (QImaging, Surrey, Canada). Sister chromatid exchanges were counted per chromosome.
|
10.1371/journal.pgen.1006561 | Rapid turnover of DnaA at replication origin regions contributes to initiation control of DNA replication | DnaA is a conserved key regulator of replication initiation in bacteria, and is homologous to ORC proteins in archaea and in eukaryotic cells. The ATPase binds to several high affinity binding sites at the origin region and upon an unknown molecular trigger, spreads to several adjacent sites, inducing the formation of a helical super structure leading to initiation of replication. Using FRAP analysis of a functional YFP-DnaA allele in Bacillus subtilis, we show that DnaA is bound to oriC with a half-time of 2.5 seconds. DnaA shows similarly high turnover at the replication machinery, where DnaA is bound to DNA polymerase via YabA. The absence of YabA increases the half time binding of DnaA at oriC, showing that YabA plays a dual role in the regulation of DnaA, as a tether at the replication forks, and as a chaser at origin regions. Likewise, a deletion of soj (encoding a ParA protein) leads to an increase in residence time and to overinitiation, while a mutation in DnaA that leads to lowered initiation frequency, due to a reduced ATPase activity, shows a decreased residence time on binding sites. Finally, our single molecule tracking experiments show that DnaA rapidly moves between chromosomal binding sites, and does not arrest for more than few hundreds of milliseconds. In Escherichia coli, DnaA also shows low residence times in the range of 200 ms and oscillates between spatially opposite chromosome regions in a time frame of one to two seconds, independently of ongoing transcription. Thus, DnaA shows extremely rapid binding turnover on the chromosome including oriC regions in two bacterial species, which is influenced by Soj and YabA proteins in B. subtilis, and is crucial for balanced initiation control, likely preventing fatal premature multimerization and strand opening of DnaA at oriC.
| Initiation of replication is a key event in the cell cycle of all living cells, and is mediated by the ATPase DnaA in bacteria, and by ORC proteins in eukaryotic cells. DnaA binds to several high affinity binding sites at the origin region of replication (oriC) on the bacterial chromosome, triggers the unwinding of the DNA duplex nearby, and additionally supports loading of the DNA helicase, which in turn leads to the establishment of the DNA replication machinery. How the binding of DnaA to oriC and the triggering of duplex opening are regulated is under extensive investigation. Using two different fluorescence microscopy techniques, we show that DnaA binding and unbinding to oriC is very rapid in two bacterial species and occurs in the range of few seconds. Moreover, DnaA binds to several additional sites on the chromosome, but with an even shorter binding half-time than at oriC: average residence time throughout the chromosome is about 200 ms, as determined by single molecule microscopy. In the absence of two negative regulators, YabA and Soj, DnaA in Bacillus subtilis binds longer to oriC and to other sites on the chromosome, accompanied by a higher frequency of initiation per cell cycle, whereas the expression of a DnaA mutant protein that shows even faster exchange rates results in decreased initiation frequency. Our data reveal that DnaA exchanges rapidly at oriC, and that tight regulation of turnover is important for proper initiation control. We also show that YabA has a dual role, a) in tethering DnaA to the replication machinery and restricting its mobility within the cell and b) in increasing DnaA turnover at oriC, both of which activities reduce the risk of reinitiation during later stages in the cell cycle.
| All cells must be able to integrate environmental and internal physiological cues into the decision when to commence the duplication of the genome in order to initiate the proliferation cycle. Nature appears to have invented the process of replication initiation only once, because the key players, called ORC in eukaryotic and in archaeal cells, and DnaA in bacteria, are conserved AAA (ATPases Associated with diverse cellular Activities) proteins, whose ATPase activity leads to conformational changes that are transduced into mechanical force or into switch-like processes. For DnaA, ATP hydrolysis confers a crucial role in initiation control [1, 2]. DnaA has been shown to associate with several high affinity binding sites at the origin region on the circular chromosome and upon an unknown molecular trigger, to extend binding to several adjacent low-affinity binding sites, which induces the formation of a helical, right handed polymeric structure [2–4]. Superhelical torsional stress of this superstructure leads to strand opening at an adjacent AT rich region, which is stabilized by an indispensable repeating trinucleotide motif term the “DnaA-trio”, to which DnaA binds with its ssDNA binding region [37]. Subsequently, DnaA interacts with helicase loader proteins to achieve the establishment of replication forks via loading of helicase and further replication proteins [5, 6]. The time point of initiation of replication starts the cell cycle, and must be in tune with extracellular stimuli as well as with the energetic state of the cell. A failure to limit initiation of replication to once per cell cycle leads to growth defects and in severe cases ultimately to lethality.
The activity and regulation of DnaA has been described in most detail in Escherichia coli, where several negative regulatory systems cooperate to limit the generation of an initiator complex to once per cycle [5, 6]. Key aspects are the sequestration of oriC regions based on hemimethylation, the activation of ATPase activity of DnaA via active replication forks and sequestration of free DnaA to a sink of DnaA-binding sites that are duplicated soon after replication initiation. Several of the regulatory proteins described in E. coli are restricted to enteric bacteria, and are not found in most other species. In Bacillus subtilis, a model organism for Gram positive bacteria, two negative regulators of DnaA activity are known, YabA and Soj (a ParA type protein) [7]. YabA is a tetrameric protein that serves as an adaptor to recruit DnaA to DnaN, the sliding clamp of DNA polymerase [8]. It has been speculated that DnaA is thereby sequestered from origin regions after initiation of replication, consistent with the finding that DnaA is more dispersed in yabA deleted cells, and is visible at oriC regions in mutant cells to a larger extent than in wild type cells [9]. YabA has also been shown to a) be associated with oriC DNA in vivo, and b) to interfere with cooperative DNA binding of DnaA in vitro [10, 11]. During the cell cycle, YabA is mostly present at the replication forks [8, 12]. Therefore, it is not clear how and where YabA exerts its function in restricting the activity of DnaA in vivo. Soj protein is another regulator of DnaA, which binds directly to the initiator protein [13]. Dependent on whether Soj has bound ATP, and is in a dimeric state, or ADP, in a monomeric state, it either activates DnaA or inhibits its activity by interfering with the multimerization of DnaA [14], a key step in initiation of replication at oriC [15]. Overall, Soj acts as a negative regulator, because a soj deletion leads to increased initiation events during the cell cycle [16].
We wished to understand the interaction of DnaA with oriC and the replication machinery in vivo in more detail, and to investigate the connection between DnaA and its two regulators. We therefore employed fluorescence recovery after photobleaching (FRAP) and single molecule fluorescence microscopy to find that DnaA association at oriC is short-lived, and upon only moderate elongation of its dwell time, i.e. in the absence of one of its negative regulators, overinitiation takes place, revealing that initiation control occurs every few seconds in bacteria, and is thus extremely dynamic.
DnaA has been shown to be present at the origin of replication region(s) early in the cell cycle, but predominantly at the replication machinery during most of the cell cycle, after duplicated origin regions have been separated towards opposite cell poles [9]. DnaA can be expressed as a fully functional N-terminal YFP fusion protein in B. subtilis [9], and as a sandwich fluorescent protein fusion in E. coli [17, 18], as the sole source of DnaA in the cell. The expression level of YFP-DnaA driven by the xylose promoter was adjusted to be similar than that of DnaA driven by its original promoter in B. subtilis (S1A Fig), allowing us to follow its localization in live cells throughout the cell cycle. Previously, we have performed statistical localization studies of YFP-DnaA, showing that DnaA mostly co-localizes with the clamp loader complex of the replication machinery (visualized through DnaX-CFP), after the origin regions have separated towards opposite cell poles [9]. Fig 1A shows an example of a time lapse experiment of YFP-DnaA (original locus, CDS7, Table 1) in a large cell that has initiated replication before the next cell cycle (this is frequently the case under the growth conditions used in this study, tD = 92 min, [9]), where DnaA co-localizes with oriC regions for 20 minutes. When duplicated origins have separated, a single YFP-DnaA focus remains between origins in the left cell half, likely co-localizing with the central replication machinery, while in the right cell half, DnaA remains at one origin region for some time. From minute 140 to minute 170, DnaA is no longer seen at origin regions, while at minute 200, two origins show YFP-DnaA foci, with a central DnaA focus remaining. This example shows that DnaA can be seen to accumulate at replication forks as well as at origin regions after replication has been initiated (as judged from the separation of duplicated origin regions). We were only able to obtain ten complete time lapse experiments with YFP-DnaA foci being visible at all time intervals, not permitting us to make any statistical statements, but the data show that DnaA has the capacity to visibly move between replication machinery, cytosol and origins during the cell cycle. Thus, although a fraction of DnaA is tethered to the replication machinery throughout most of the cell cycle, DnaA also accumulates at oriC after replication initiation, which could potentially cause unwanted reinitiation events. Therefore, additional control mechanisms must exist that prevent reinitiation at oriC, which include the two negative DnaA regulators, Soj (ParA) and YabA.
YabA has been reported to largely co-localize with the replication machinery [8, 12]. On the other hand, in vivo and in vitro data show that YabA can influence binding of DnaA to oriC sequences [5, 6, 10], throughout the cell cycle, and to affect multimerization of DnaA [19]. We wished to address the question how YabA can affect oriC binding of DnaA when based on epifluorescence microscopy YabA is visually absent from oriC during most of the cell cycle [8, 12]. We therefore visualized a functional [8, 12] YFP-YabA fusion throughout the cell cycle, examining its co-localization with both, oriC regions, and with DnaX as a marker for the replication machinery (yfp-yabA, lacI-cfp/lacO array at oriC, dnaX-mCherry). Fig 1B shows that in 54% of cells, YabA co-localizes with the central DnaX focus, and is not visible at bipolar oriC regions. In 10% of cells, all three markers co-localize (initiation) and in 9% of cells, YabA co-localizes with DnaX and with one of the two oriC regions. In the remaining cells (27%), a YabA focus is present, but does not co-localize with either oriC or DnaX. Therefore, YabA can also visibly accumulate at oriC during ongoing replication. We analysed the localization pattern of YFP-YabA more precisely, grouping cells according to cell length, which correlates with the age of the cell. S2 Fig shows that in small, medium sized and large cells, YabA is present at the replication forks and not at oriC regions in about 50% of cells (panels B, D and G), and co-localizes with oriC in 10% of young cells, in 6% of medium-age cells, and in 13% of old cells (panels C, F and H). Interestingly, YabA does not co-localize with either oriC or DnaX in about 25% of cells, possibly interacting with DnaA that is bound at promoter regions on the chromosome. YabA was seen to co-localize with only one of two origins in 9% of middle aged cells (S2 Fig, panel E), showing that oriC localization can be asymmetric. Therefore, YabA largely follows the localization pattern of DnaA [9] (Fig 1A), and a considerable fraction of YabA can be asymmetrically positioned, both at the forks and at oriC. The patterns of localization of DnaA and of YabA could be explained if they were to alternate between binding at oriC and at replication forks in a frame of few minutes or even less, which is shown below.
To shed light on the connection between YabA, DnaA and binding to origin regions, we determined the turnover rate of YFP-DnaA binding to oriC, using FRAP analysis. Large cells containing two YFP-DnaA foci usually contain YFP-DnaA co-localizing with oriC regions, because replication has terminated [9], and were therefore used for the analyses. From 32 experiments, we found a half time recovery for YFP-DnaA of 2.5 ± 0.3 seconds (YFP-DnaA original locus, strain PG1037, Table 2); for a typical FRAP experiment see Fig 2A. Therefore, turnover of DnaA binding is extremely rapid. In contrast, LacI binds to the lacO site for about 4 minutes on average [20], and TetR has a half-time recovery of 2 to 2.5 minutes on a tetO array in eukaryotes [21, 22]. When we repeated experiments using LacI-GFP binding to a lacO array (the “origin tag”) in B. subtilis, we determined a half-time recovery of 2.4 ± 0.58 minutes for LacI-GFP binding (S3A and S3B Fig), which is quite similar to the recovery of the tetO array/TetR-GFP system. By normalizing the FRAP data by the whole cell intensity (see Material and methods), we determined if a fraction of DnaA molecules remain bound to oriC on a timescale much longer than that of the experiment (i.e. if a fraction of DnaA does not exchange during the course of the experiment). From Table 2, it can be deduced that the stationary fraction of YFP-DnaA at oriC is about 10% (strain PG1037) and is therefore very low, indicating that most DnaA molecules constantly turn over at this chromosomal site. In some cells, there was no measurable stationary fraction, while in others it was about 20%, indicating considerable heterogeneity in this aspect during the cell cycle, although it must be kept in mind that FRAP analyses are inherently noisy. To rule out that FRAP recovery is affected by even a possible slight overproduction of DnaA, we grew cells in the presence of 0.1% xylose, leading to reduced levels of YFP-DnaA (S1A Fig). Half-time recovery in these cells was 2.39 ± 0.85 s (S3C Fig), showing that our experiments are not affected by unphysiological levels of YFP-DnaA.
We wished to determine if ectopically expressed YFP-DnaA behaves in a similar manner as the fusion expressed from the original gene locus. Ectopically induced YFP-DnaA reduces the expression of wild type DnaA through auto-repression of DnaA at the dnaA promoter [23], which we verified by Western blotting (S1B Fig). We expressed YFP-DnaA for 30 min with 0.1% xylose (w/v) prior to FRAP experiments. Under this condition, somewhat more YFP-DnaA than non-tagged DnaA was present within the cells, but not exceeding the amount of DnaA in cells lacking ectopically expressed YFP-DnaA (S1B Fig, lane 4). Importantly, half time FRAP recovery was similar for ectopically expressed YFP-DnaA (2.9 ± 0.2 s, n = 61) (strain AHV2, Tables 1 and 2), compared with YFP-DnaA expressed from the original gene locus, showing that YFP-DnaA turns over similarly in the presence of wild type DnaA in a merodiploid strain, compared to being the sole source of DnaA in the cell.
To ensure that FRAP recovery reflects YFP-DnaA bound to oriC, we analysed recovery in strain ME15, which ectopically expresses YFP-DnaA and carries an origin-CFP tag (Table 1). From 10 experiments, we found a recovery time for YFP-DnaA co-localizing with oriC of 2.7 s ± 0.5 s (S4A and S4C Fig, Table 2), showing that the FRAP data adequately describe the turnover of YFP-DnaA at oriC.
To find out if turnover of DnaA at oriC differs from that at the replication machinery, we performed FRAP analysis of YFP-DnaA foci not co-localizing with oriC (S4B and S4D Fig). This fraction of foci largely represents replication fork-bound DnaA, and has a half-time turnover that was very similar to that of oriC-localized YFP-DnaA (3.08 ± 0.5 s, n = 10). To further substantiate this point, and to investigate the localization pattern of a DnaA version that is unable to bind to dsDNA, we generated a mutant allele of dnaA that carries a mutation in the DNA-binding domain, dnaAR387C, which is expected to disrupt DNA binding based on studies on E. coli DnaA [24]. DnaA could be efficiently purified (Fig 3A) and showed little degradation (Fig 3B). To test for DNA binding in vitro, we used a 624 bp fragment containing oriC, to which purified wild type DnaA bound with high affinity in gel shift experiments (Fig 3C), in contrast to a 532 bp fragment lacking dnaA boxes (Fig 3D). Contrarily, DnaAR387C showed only non-specific binding to the oriC fragment (Fig 3E). To further quantify this point, we immobilized a 632 bp PCR fragment containing oriC on an SPR (surface plasmon resonance) chip. Wild type DnaA showed higher affinity to oriC DNA in the presence of ATP (Fig 3G) than in the absence (Fig 3F), whereas DnaAR387C did not show any DNA binding activity in SPR experiments irrespective of the absence or presence of ATP (Fig 3F and 3G). Interestingly, the DNA-binding mutant DnaA protein still formed foci at the replication machinery, but not at oriC regions (co-localization in only 0.5% of the cells, n = 200, see overlay in Fig 4A), verifying the findings that DnaA is bound to the replication forks via protein/protein interactions to YabA and to DnaN (the sliding clamp of the polymerases) [8], and to oriC via direct DNA binding. The accumulation of DNA-binding deficient DnaA at the replication machinery in addition to that of wild type DnaA present in the cells reveals that there are sufficient binding sites to allow for the visual accumulation of mutant DnaA. Half-time FRAP recovery of DnaAR387C was 2.5 s ± 0.4 s (Fig 4A and 4B, ME21, Table 2), and not significantly different from that of wild type DnaA (Fig 4C), confirming that DnaN-bound DnaA also has a very high turnover rate. Thus, while DnaA is bound to DnaN via YabA, rather than to DNA at oriC, the turnover rate is similar for replication machinery-bound DnaA compared with oriC-bound protein.
We also determined FRAP recovery for YabA (Fig 2C and 2D). YFP-YabA expressed additionally to YabA had a half-time recovery of 2.3 ± 0.2 s (strain JDS190, Table 2), and 2.3 ± 0.2 s when expressed as sole source of YabA in the cell (strain KS173, Table 2), and thus very close to that seen for DnaA (Fig 2E). However, the immobile fraction of YabA was 20%, somewhat higher than that of DnaA (Table 2). We cannot determine if this fraction is present at oriC or at the replication machinery, because oriC-co-localized YabA is more difficult to visualize than oriC-bound DnaA, making localization analysis much noisier and less reliable. These data suggest that individual molecules of YabA and DnaA do not remain bound to oriC or to the replication forks for more than few seconds and that neither DnaA nor YabA have a significant stationary subpopulation of more than 10 or 20%, respectively, at oriC.
To use a second method for analyzing DnaA dynamics in live cells, we visualized and tracked single YFP-DnaA molecules using single molecule fluorescence microscopy and single molecule tracking (SMT). To facilitate tracking of YFP-DnaA by generating fewer signals per cell, we took advantage of the fact that ectopically expressed YFP-DnaA (strain ME15) behaved similarly in FRAP experiments as YFP-DnaA expressed from the native locus (Fig 4A). We induced YFP-DnaA from the amylase locus with a very low concentration of xylose (0.01% instead of 0.5%) to generate only few molecules per cell that can be easily tracked, and continuously illuminated the cells at 514 nm with an exposure and interval time of 41 ms per frame. After bleaching most of the few chromophores in the cells, we observed movement of single YFP-DnaA molecules, which can be recognized by a characteristic single bleaching step to background noise (S5A and S5B Fig). We linked single molecule localizations in consecutive frames either manually or automatically and quantitatively analyzed the dynamics and the kinetics of movement. Fig 5A shows an example of a dynamic YFP-DnaA track, in which a single YFP-DnaA molecule moves from the center of the cell towards one pole and then back, to bleach in a single step between two frames (also see S1 and S2 Movies for exemplary tracks). The corresponding heat map (Fig 5B) shows that the molecule did not arrest at one site for more than one interval, and Fig 5C shows that the step size of movement has mostly been in increments of more than 230 nm. Molecules that move in steps of less than 230 nm are considered static (230 nm corresponds to 2.3 pixels in our setup where movement between pixels is just about detectable), e.g. shown in S5C, S5D and S5E Fig for YFP-DnaA (see also S3 Movie), or in Fig 5G and 5H for YFP-YabA. Importantly, we were able to observe molecules undergoing transitions between mobile and static events in addition to examples of purely mobile or immobile movements (Fig 5F).
For all tracks observed from the different fluorescent protein fusions (Table 3) we first calculated apparent diffusion constants (D*) from the mean square displacement (MSD). Taking into account all molecules (185 trajectories of an average length of 5.7 time points), YFP-DnaA shows an apparent diffusion rate of 0.68 μm2/s (Table 3), which is about an order of magnitude less than that of free GFP (7 to 14 μm2/s) [25, 26]. Even accounting for the fact that YFP-DnaA has a size of 78 kDa and might migrate as a dimer, the diffusion rate is low, which indicates that movement of DnaA is restrained. This is most likely generated through its non-specific interaction with DNA since it also acts as a transcription factor at several sites on the chromosome [27]. An overlay of all tracks monitored in a single cell shows that movement of DnaA occurs mostly in the middle of the cell, and rarely at the cell poles (Fig 5D), consistent with the idea that most dynamic DnaA molecules move through the nucleoid and are constrained in their diffusion through DNA-interactions. However, we cannot distinguish between constrained movement and freely diffusing DnaA molecules, which should also exist in the cells.
We also analyzed the step size distribution of all the fluorescent protein fusions to identify subpopulations of movement (Fig 6). Particles undergoing Brownian motion move in increments that follow a normal (Gaussian) distribution with mean zero (the longer the displacement, the further away is the value from “0”) and a variance σ2, which scales with the diffusion coefficient. Thus, the higher the variance of the distribution, the higher is the diffusion coefficient of the particle. In case of YFP-DnaA, the shape of the histogram could not be described well by a single Gaussian distribution (S6A Fig). So we applied a multivariate Gaussian distribution fit assuming two modes of motion (Fig 6A), which also reflects our previous observations (Fig 5A and 5B and S5C and S5D Fig). Thereby, we determined a static/slow moving fraction (dotted line, Fig 6A), with an apparent diffusion constant of 0.027 μm2/s (Table 4), and a dynamic fraction (dashed line, Fig 6A) with D* = 0.51 μm2/s. From the area underneath the two curves, we can determine that 20% of DnaA molecules are static and 80% are dynamic molecules revealing that most DnaA molecules are moving in search for binding sites on the chromosome. This more discriminative analysis of DnaA movement reveals two distinct populations, and together with the heat maps shows that DnaA can change between static and dynamic movement within short periods of time (few hundred milliseconds).
DnaA was surprisingly dynamic so as a control we wished to analyze a protein presumably diffusing much slower. Therefore, we tracked Spo0J-GFP (1104 molecules, average track length 8.5 time points), which binds to at least 10 specific sites surrounding the origin regions on the chromosome [28]. In agreement with slow recovery in FRAP experiments [29], we found that Spo0J molecules are largely static (Fig 6B, Table 4), likely representing molecules sequestering parS sites. Similarly, heat maps of Spo0J-YFP show entirely static molecules (S9B Fig).
Additionally, we performed two control experiments to rule out that we are not missing a dynamic fraction of DnaA under our conditions. First, we tracked YFP-DnaA expressed from the original gene locus as sole source of the protein. We induced with 0.1% xylose, which leads to lower cellular levels of DnaA (S1A Fig), and tracked the movement using faster (10 ms) stream acquisition settings. YFP-DnaA showed a very similar step size distribution under this condition compared to the low-induced ectopically expressed YFP-DnaA tracked in time intervals of 41 ms (S6B Fig). We found that 19% of the steps showed static movement (D* = 0.2 μm2/s), and 81% of the steps corresponded to dynamic movement. Further analysis of the square displacement distribution using a three-component analysis indicated that we are unlikely to miss a highly mobile fraction of YFP-DnaA (S6C Fig). Second, we used our setup to track mNeongreen (a variant of LanYFP from Branchiostoma lanceolatum, distantly related to GFP [30]) in E. coli cells, using fast stream acquisition with an exposure time and interval time of 4 ms (S4 Movie). An overlay of mNeongreen tracks obtained in two cells, reveals that it diffuses throughout the cells (S7A Fig), in contrast to the more centrally restrained movement of YFP-DnaA (Fig 5D). The motion can be described by a single population fit referring to dynamic molecules, which diffuse with D* = 3.3 μm2/s, much faster than the dynamic fraction of YFP-DnaA with D* = 0.51 μm2/s (S7B Fig). These experiments show that our experimental setup is able to track slowly and quickly diffusing molecules.
We wished to determine the average residence time of YFP-DnaA on all sites on the chromosome. Average residence time was computed by considering the intervals, both forward and backward in time, where a track stays within a radius of 230 nm. Hence, each track is decomposed into a set of non-overlapping time intervals. To remove the contribution of slowly diffusing mobile molecules, we consider only static intervals of length greater than or equal to three frames. For YFP-DnaA, and given our track length distribution, we estimated that the probability of a free molecule being erroneously counted as static is 9%. S8 Fig shows a histogram of the percentage of molecules stopping for 3, 4, 5, 6 and 15 time intervals. From all stopping events, the cumulated periods of residence for YFP-DnaA are shown in Fig 7A, from which we calculated an average residence time of 197 ms (Table 3). This number from SMT is much lower than the turnover at oriC determined by FRAP experiments because this approach measures transient binding events at all places on the chromosome besides oriC (e.g. promoters having dnaA boxes plus non-specific DNA binding), or at the replication machinery. The small relative number of YFP-DnaA molecules bound at oriC or at the replication machinery compared to the dynamic molecules also contributes to this disparity. Nonetheless, we can conclude that single DnaA molecules stay transiently at individual sites for about 200 ms on average, revealing extremely rapid turnover on its chromosomal binding sites.
In contrast to YFP-DnaA, few mobile YFP-YabA molecules were observed, and mostly static spots appeared and disappeared. The fact that YabA was mostly static can be seen from the overlay of tracks in a single cell (Fig 5E), which is strikingly different from that of DnaA (Fig 5D), and in agreement with the idea that YabA binds mostly to the centrally located replication machinery, or to origin regions positioned around the quarter sites in the cell, with the measured recovery half-time of 2.5 s (Fig 2C). Consequently, the apparent diffusion coefficient of YFP-YabA is much lower than that of DnaA with D* = 0.16 μm2/s as calculated by MSD analysis (Table 3, 171 molecules, average track length 8.5 time points) and the step size distribution (Fig 6C) shows a much higher static fraction of YabA compared with that of DnaA (Fig 6A). 52% of YabA molcules were static, and 48% were mobile (Fig 6C, Table 4). YFP-YabA exhibited similar residence times as Spo0J-YFP (259 and 275 ms, Table 3) and resided for longer times in a radius of 230 nm compared to YFP-DnaA molecules. When we restricted the dwell time analysis to tracks that were at least 10 frames long, we found that YFP-YabA and Spo0J-YFP stayed about 2.5 and 5 frames longer respectively in the same radius than YFP-DnaA (S1 Table and S10A and S10B Fig). These data show that YabA does not diffuse through the nucleoid, neither in a complex with DnaA, nor alone in a constrained motion like DnaA, because a rapidly diffusing fraction similar to that of DnaA is not observed (while a YabA-YFP tetramer has about the same size as a YFP-DnaA dimer). Rather, YabA arrests at sites where likely DnaA and/or DnaN are present, and dissociates off to move freely through the cell. YabA could therefore act as a “chaser” for DNA-bound DnaA, leading to the dissociation of DnaA assemblies, possibly together with Soj. Note that a similar experimental setup showed that different regions on the chromosome do not move at this time scale [31], such that the mobile molecules detected for DnaA and for YabA do indeed move between different sites on the chromosome rather than the chromosome moving with DnaA, or through the entire cell for YabA.
ADP-bound Soj has been shown to negatively affect the multimerization of DnaA [14], which is assumed to be an important step in replication initiation. Also, YabA has been shown to decrease the co-operativity of DnaA binding to DNA in vitro [10]. The deletion of soj leads to a very modest over-initiation phenotype (1.3 times more origins than wild type cells [16]), while a yabA deletion leads to a slower growth rate and a 1.7 fold increase in oriC number [8]. We wondered if a loss of any of the negative regulators of DnaA would have an effect on the dwell time of DnaA at oriC. Indeed, half-time recovery was extended to 3.8 ± 0.3 s in the absence of YabA (Fig 2F and 2G, strain KS171 Table 2), and to 4.3 ± 0.4 s in cells devoid of Soj/Spo0J (Fig 2I and 2J, strain KS175, Table 2). Note that yabA mutant cells have more background signal for YFP-DnaA (Fig 2F), i.e. contain more diffusive YFP-DnaA molecules, because DnaA is no longer tethered to the replication forks [9]. Statistical analysis shows that the recovery of YFP-DnaA in the absence of YabA is significantly lower than in wild type cells (Fig 2H), with a p-value of 0.002. Absence of Soj/Spo0J is also statistically highly relevant (Fig 2K), with p = 0.001.
These data strongly suggest that both YabA and Soj act as chasers of DnaA at oriC. Consistent with this idea, when we deleted soj/spo0J, YFP-DnaA molecules became more stationary, which can be seen in the more narrow step size distribution (Fig 6D; 36% static and 64% dynamic, compared to 20/80% for wild type cells, Table 4), and showed increased residence times (Fig 7C and S8B Fig) in SMT experiments. From the resting times of 3 or more frames, we calculated average residence time of 197 ms for YFP-DnaA in wild type cells, and 257 ms in soj/spo0J mutant cells (KS92, 125 molecules, average track length 6.6 time points, Table 3). For tracks longer than 10 frames, the difference was no longer detectable showing that the loss of soj/spo0J affects only short residence times (S1 Table and S10C Fig). Taken together it is apparent that in the absence of Soj/Spo0J, individual DnaA molecules transiently arrest for longer periods of time at their chromosomal binding sites, and not only at oriC. It was difficult to track YFP-DnaA in the absence of YabA, for a reason that is unclear to us, so we could not determine the effect of the absence of YabA on single molecule movement of DnaA.
The above data suggest that an increase in the residence time of DnaA at oriC leads to over-initiation of replication and that both YabA and Soj act to decrease the residence time of DnaA, both at longer residency-time targets, i.e. the initiation complex at oriC and the replication forks, and at shorter-bound targets on the chromosome. If this were true, then a reduced average residence time of DnaA should lead to under-initiation. To test this notion, we generated a mutant DnaA protein that has strongly reduced ATPase activity (Fig 8A), DnaAE183Q. We reasoned that this mutant protein might be compromised for initiation activity. Purified DnaAE183Q bound to oriC DNA with similar affinity as wt DnaA (Fig 8B), but did not show an increase in binding activity after addition of ATP as is true in the case for wild type DnaA (Fig 8C). Expression of DnaAE183Q from an ectopic site on the chromosome and in the presence of wild-type dnaA resulted in the elongation of cells, and the number of origins per cell decreased considerably compared to that in wild type cells, which was especially evident in rich medium, where wild type cells have between 4 to 8 origins, whereas 60 minutes after induction of mutant DnaA, cells contained between 2 and 4 origins (Fig 8D and 8E). When we investigated the expression of the mutant allele in minimal medium, cells also became considerably longer that wild type cells (Fig 8F), had a decreased number of nucleoids (Fig 8G) and of origins (Fig 8H), and also showed fewer YFP-DnaA foci per cell (Fig 8I). These data show that the mutant protein acts in a dominant negative way and leads to under-replication. Interestingly, tracking of YFP-DnaAE183Q revealed that the protein arrests less frequently than the wild type protein (average residence time 163 ms, strain ME20, Table 3, Fig 7D, S10D Fig). The difference in residence time is even more apparent when only tracks longer than 10 time points are used for the calculation of average residence time: here, wt DnaA has a residence time of 342 ms, while underinitiating mutant DnaA has 205 ms (S1 Table). Consequently, DnaAE183Q (144 tracks with an average length of 5.7 time points) showed a higher apparent diffusion coefficient 0.98 μm2/s than wild type YFP-DnaA (0.68 μm2/s) (Table 3) as calculated using MSD analysis, and which is also apparent from the increase in the fraction of the mobile molecules (91% versus 80% for wtDnaA) and a decrease in the static fraction (9 versus 20%) (Fig 6E and Table 4). These findings suggest that a decrease in average residence time of DnaA on the chromosome leads to a reduction in initiation frequency.
A functional DnaA-YFP sandwich (sw) fusion has been shown to localize as foci in E. coli, either at oriC regions, or also at the DAT locus, which contains many dnaA boxes [17]. In another study, DnaA was reported to form foci along the cell membrane, in a helical pattern [18]. Of note, epifluorescence laser illumination (stream, exposure time = 200 ms) was used, such that only many fluorescent proteins being present at a certain position will be seen as a spot, rather than single FP molecules as in SMT. Intriguingly, DnaA-YFPsw was highly mobile, fluorescent signals were clearly seen to move along the chromosome between the cell halves (S5 Movie and Fig 9A). We quantified the intensity of fluorescent patches in the cell halves, relative to each other and found that DnaA oscillates between opposite cell halves in a time frame of few seconds. The plot shows that an increase in intensity in one cell half is accompanied by a correlated decrease in the other half. When the normalized intensity in one half is plotted against that in the other half (Fig 9A, right panel), a clearly inverse correlation is apparent. When the signal of a tetO array bound by TetR-YFP is analysed under identical conditions (Fig 9C), no correlation between signals in the two cell halves is observed (Fig 9C, right panel). We wondered whether the removal of DnaA from one part of the chromosome (including oriC) might be driven by RNA polymerase engaged in transcription. We therefore performed experiments in cells that were treated with rifampicin for 60 minutes. Fig 9B shows that DnaA-YFPsw still oscillated between chromosome regions, without any noticeably altered timing, and still showed a highly inverse correlation between normalized intensities (Fig 9B, right panel). These data suggest that coordinated binding and unbinding of DnaA also occurs on a time scale of a few seconds in E. coli, independent of transcription. Interestingly, DnaA oscillation is much faster than that of MinD, a membrane-associated regulator of cell division, which oscillates between the cell poles in a frame of about 20 seconds, driven by the MinE “chaser” [32, 33]. It will be highly interesting to investigate the involvement of ATP in this process and to identify the chaser of E. coli DnaA. In SMT experiments, E. coli DnaA showed rapid movement (S6 Movie), however, diffusion rates were slower (0.34 μm2/s) and residence time longer (258 ms, Fig 7E) than those of B. subtilis DnaA (Table 3, 279 molecules, average track length 9.6 frames). Like BsDnaA, EsDnaA showed patterns of static and solely dynamic movement, as well as changes between movement and resting periods (S9A Fig). Additionally, 70% of EcDnaA were dynamic (Table 3, Fig 6F), compared with 80% for BsDnaA, revealing that the protein is highly mobile, but somewhat less dynamic than its Gram positive counterpart. Therefore, DnaA clearly has very fast binding and unbinding dynamics in two model bacteria, and possibly in many other bacteria.
It has been elegantly shown that DnaA binds to several high affinity binding sites within oriC, and can form a helical superstructure upon spreading to further sites at oriC, leading to strand opening and to the recruitment of DNA helicase loaders [15, 34–36]. Our data show that indeed, DnaA is visibly present at oriC regions throughout the Bacillus cell cycle, however, its average residence time is surprisingly short, around 2.5 s. In E. coli, the observed oscillatory dynamics also suggest a residence time at oriC in the lower seconds range. Evidently, the cell ensures at a time frame of seconds that an oriC-bound DnaA supercomplex can not build up prematurely. A fast binding turnover is clearly helpful if not crucial for avoiding the formation of DnaA multimers at oriC, whose stable generation would lead to initiation. Additionally, while over-initiation is detrimental or even fatal for cells, high DnaA turnover may be advantageous when cells have to adapt to environmental changes, and may allow rapid stabilization of DnaA at oriC and thereby rapidly trigger initiation or allow for an increase in initiation frequency when needed. It should be noted that we have studied the dynamics of DnaA in bulk culture, so residence times may vary considerable within the cell cycle, and it is likely that prior to initiation of replication, DnaA exchange at oriC is strongly decreased, leading to spreading into the DnaA-trio sequences adjacent to the dnaA boxes [37].
We studied the localization of two mutant alleles of DnaA that were generated based on studies performed on E. coli DnaA. Mutation E204Q (B. subtilis: E183Q) leads to reduced ATPase activity [38, 39], while mutation R407C (B. subtilis: R387C) abolished dsDNA binding [40]. We show that the corresponding mutations in B. subtilis have the same effect, and ectopic induction of DnaAE183Q leads to a dominant negative effect in initiation activity, showing that reduced ATPase activity interferes with the action of DnaA even if wild type protein is present. Purified DnaAE183Q still bound to oriC DNA like wild type DnaA, but affinity could not be stimulated by the addition of ATP, as is the case for wild type protein, showing that ATPase activity affects high affinity binding of DnaA. YFP-DnaAE183Q formed fewer fluorescent foci in the cells and led to a strong underinitiation phenotype. Conversely, non-DNA binding YFP-DnaAR387C did not affect initiation frequency, and still visibly assembled at replication forks but not at origin regions. These data reinforce the idea that DnaA binds to the replication machinery via the YabA adaptor protein and DnaN, and to origin regions via direct DNA binding. Mutant DnaA accumulated at the replication forks in the presence of wild type DnaA, showing that there are ample binding sites, and that neither DnaN nor YabA are limiting factors to attach additional DnaA molecules, in support of the idea that DnaN acts as a sink for DnaA to titrate a considerable amount of the inducer away from duplicated and separated origin regions.
While in B. subtilis, based on our data, molecules at oriC exchange between the replication forks, origins and other sites on the chromosome in a seemingly stochastic manner, in E. coli, an oscillation between two cell halves (and thus most likely between origins) can be observed. The mechanism of this coordinated association and dissociation is unknown, but may involve a “chaser” molecule, in analogy to MinE leading to MinD disassembly at a cell pole [41, 42]. It will be highly interesting to elucidate how this oscillation is driven at a molecular level. In any event, DnaA shows rapid turnover of molecules on the chromosome in two model bacteria compared with a residence time of 4 min for lac repressor being bound to lacO [20], 2 to 2.5 minutes for tetO-bound TetR [21, 22], 2.5 minutes for LacI-GFP binding to a lacO array (this work), or 2 minutes for the SMC protein remaining within condensation centres located near the origin regions on the chromosome [29].
Interestingly, the ParB partner of Soj (ParA), the Spo0J protein, shows very static behavior. Compared with DnaA, we could not observe any exchange between bound molecules surrounding origin regions and possible free molecules. The number of free Spo0J-YFP found in this study is likely an underestimate of the actual number, because Spo0J-YFP was tracked at 41 ms for comparison to DnaA dynamics, so in this case fast diffusing molecules may have been lost, but clearly, binding of Spo0J is highly static, such that it is able to generate long range connections within the origin region, as has been seen in Hi-C analyses [43], over long periods of time.
Importantly, an increase in residence time through the deletion of DnaA regulators YabA and Soj (ParA) in B. subtilis, leads to over-initiation, showing that rapid turnover at oriC is modulated by two interactor proteins, both of which increase the rate of dissociation of DnaA, in agreement with findings that YabA and Soj interfere with cooperative binding of DnaA or with multimerization, respectively, in vitro [10, 44]. Therefore, YabA confers a dual role, acting as a tether of DnaA to the replication machinery, and also as a chaser of DnaA binding at oriC [10, 11]. Mutant DnaA protein that has lost DNA-binding activity is no longer able to bind to oriC, but is still sequestered to the replication machinery, verifying the idea that a considerable number of DnaA molecules are bound at the forks, and—although rapidly exchanged—are not available for oriC binding, until replication has terminated. Additionally, a decrease in DnaA ATPase activity (E183Q mutation) leads to decreased binding to oriC and other sites on the chromosome, showing that this aspect of DnaA activities also serves as an additional layer of replication control in B. subtilis.
Our data suggest that binding of many DnaA molecules to oriC visibly occurs at many times in the cell cycle, and that one major function of YabA and of Soj is to counteract this event in order to prevent over-initiation. However, YabA clearly has a second function in recruiting DnaA to the active replication machinery. The fact that generally, YabA can be seen to accumulate at the replication machinery, rather than at oriC, yet interacts with origin regions throughout the cell cycle, suggests that there are more binding sites at the forks than at oriC, which is also true for DnaA. This is corroborated with recent experiments showing that DnaN dimers remain bound to the lagging strand for an extended time, and are thus deposited behind the replication forks [45, 46]. Thus, DnaN provides many binding sites for YabA and DnaA, for which DnaA evidently has similar association/dissociation kinetics. Therefore, DnaA moves between oriC, different sites on the chromosome, and the replication machinery, with the latter providing many binding sites to act as a datA-like sink for DnaA binding. Together, DnaN binding via YabA, inhibition of cooperative DNA binding of DnaA at oriC by YabA, and inhibition of multimerization, which would lead to strand opening at oriC, by Soj provide efficient means to restrict initiation activity to once per cell cycle.
SMT experiments show that DnaA has very short residence times in B. subtilis as well as in E. coli, in the range of few hundred milliseconds. It should be noted that the values of 200 or 260 ms, respectively, determined in our study are underestimates of the true dwell time at binding sites due to interference by photobleaching, i.e. bleaching of molecules that are bound on the DNA. However, the actual values in the cell will not be strikingly different. SMT is therefore a powerful technique to determine in vivo protein turnover rates at loci that appear to consist of static foci using epifluorescence microscopy, which has also been shown for e.g. proteins involved in DNA mismatch repair [47] or in chromosome segregation [48]. Our work reveals that two bacterial species from different phyla avoid long residence times of the initiator protein and employ tight turnover rates to ensure tight initiation control throughout the cell cycle.
All cells imaged in this study were growing exponentially, and in a non-synchronized manner. Average dwell times (exchange rates) may be different during certain periods of the cell cycle, e.g. for YFP-DnaA during initiation of replication. For most of the cell cycle, initiation activity of DnaA is repressed to prevent reinitiation events, and therefore, the data determined by FRAP and SMT analysis predominantly reflect the dynamics of DnaA during its initiation-repressed state of activity.
In order to study the localization behavior of YFP-YabA with respect to the origin of replication and the replication machinery, a strain with a CFP-tagged origin and an mCherry-tagged replication machinery was generated. To combine the CFP tagged origin with YFP-YabA, competent cells of PG208 (spo0J::lacO cassette, thrC::lacI-cfp) were transformed with chromosomal DNA of JDS190 (amyE::Pxylyfp-yabA, ΔyabA), which gave rise to strain KS119. The chloramphenicol resistance in PG1159 (dnaX-mcherry) was changed to tetracylcline resistance, resulting in strain KS114. Cells of KS119 were transformed with chromosomal DNA of KS114 (dnaX-mcherry) to obtain KS140 (origin-cfp, amyE::yfp-yabA, ΔyabA, dnaX-mcherry). To analyze YFP-DnaA in a ΔyabA background, KS171 was generated transforming competent PG1037 cells (Pxyl yfp-dnaA) with chromosomal DNA of KS173 (ΔyabA::phleo amyE::yfp-yabA). To test for successful yabA deletion a test-PCR was performed. To study YFP-DnaA in a Δsoj-spo0J background, PG1037 cells were transformed with chromosomal DNA of AG1505 (Δsoj-spo0J::spec), giving rise to KS175 (Δsoj-spo0J::spec Pxyl yfp-dnaA).
For single molecule microscopy a strain with an ectopic copy of yfp-dnaA in a Δsoj-spo0J background was generated. To this end AHV2 (amyE::Pxylyfp-dnaA) cells were transformed with chromosomal DNA of AG1505 (Δsoj-spo0J::spec) resulting in KS192 (Δsoj-spo0J::spec amyE::yfp-dnaA). To construct a strain with YFP-YabA under an IPTG (isopropyl β- D -1-thiogalactopyranoside) inducible promoter, yfp-yabA was amplified from chromosomal DNA of JDS190 using primer 1036 5’-ctg gct agc aga aag gag att cct agg atg and 4737 5’-cat gca tgc cta ttt ttt att taa gaa tga cag and cloned into pDR111 via NheI and SphI restriction sites (KS142). Competent wild type cells were transformed using pDR111-yfp-yabA to generate KS149. Competent KS183 cells (yabA::phleo) were transformed with the plasmid pKS142 (Pspac-yfp-yabA) or chromosomal DNA of JDS190 (amyE::yfp-yabA) to give rise to KS173 and KS167, respectively.
To visualize yfp-spo0J, we generated a strain that expresses the fluorescent protein fusion protein from the amyE locus under control of the inducible xylose promoter, We amplified the gene using primers spo0J_apa_up (5`-CAT GGG CCC ATG GCT AAA GGC CTT GGA-3`), spo0J_eco_dn (5`-CAT GAA TTC TGA TTC TCG TTC AGA CAA AAG -3`) and chromosomal DNA of B. subtilis strain PY79. We ligated the PCR product into similarly digested plasmid pSG1193 and transformed it into competent PY79 cells. For tracking of single mNeonGreen molecules, we constructed a strain carrying mNeonGreen under control of an IPTG-inducible promoter. The gene was amplified from a plasmid carrying an E. coli adapted mNeonGreen allele (kind gift of the group of Kristina Jonas) using primers mNGreen_EC_FW (5`-CACCGGTGGCGGCGGTTCTATG-3`) and mNGreen_RV (5-TAAGCATGCTTACTTGTACAGCTCGTCCATGC-3`). The gene was first inserted into the pENTR-D-Topo vector and then transferred to pHGWA using the Gateway cloning system. All strains are listed in Table 1.
Protein purification was performed in two consecutive steps. The purification of (His)6-wild type/mutant DnaA initially began with affinity chromatography using an ÄKTA Prime apparatus (GE Healthcare) and Nickel-Sepharose columns (HisTrap HP 1 ml, GE Healthcare) and was continued by size-exclusion chromatography using an ÄKTA FPLC apparatus (GE Healthcare) and a gel filtration column (Superdex 200 10/300 GL, GE Healthcare). Prior to purification, the respective proteins were overexpressed in E. coli Rosetta (DE3) pLysS cells carrying a pETDuet-1 vector (Novagen) with an (indirectly) IPTG-inducible T7 promoter, six encoded histidines and the full gene sequence of the dnaA variants. Transformants were grown under vigorous shaking in LB-medium at 37°C to exponential phase (OD600 0.8) and induced for 60 minutes with 1 mM IPTG. Subsequently, the cells were centrifuged (20 minutes, 4°C, 5000 rpm) and the pellet was resuspended in HEPES A (50 mM HEPES, 300 mM NaCl, pH 7.5). To prevent protein degradation a protease inhibitor was added (Complete, Roche). Afterwards, the cells were French-pressed (AMINCO French Press, Laurier Research Instrumentation) in two consecutive cycles at approximately 20000 psi and the lysate was centrifuged (30 minutes, 4°C, 16000 rpm). The clear supernatant was passed through a filter (pore-size 0.45 μm, Filtropur S, Sarstedt) before injection into the loop of the ÄKTA Prime apparatus (preequilibrated with HEPES A and HEPES B [50 mM HEPES, 300 mM NaCl, 500 mM imidazole, pH 7.5]). The proteins were loaded onto the Nickel-Sepharose column, the column was washed with 20% HEPES B and the protein eluted with 100% HEPES B in fractions of 1 ml and checked by SDS-PAGE. Fractions containing significant amounts of the desired protein were assembled and loaded onto size exclusion chromatography columns (preequilibrated with HEPES A). The peak fractions were analyzed by SDS-PAGE and only pure protein fractions were assembled and stored at 4°C.
ATPase activity was measured by using a coupled spectrophotometric pyruvate kinase/lactate dehydrogenase assay as previously described [49]. The reaction was monitored at 340 nm for 30 minutes at room temperature.
EMSA were performed with increasing amounts (0–80 pmol) of (His)6-wild type/mutant DnaA and either DnaA-box containing oriC-DNA (0.9 pmol, linear DNA-fragment, 624 nt) or control DNA without DnaA-boxes (0.9 pmol, linear DNA-fragment, 532 nt,) under ATP-containing (2.5 mM) and ATP-free conditions. The reaction mixture with a final volume of 20 μl (27.5 mM HEPES (pH 7.6), 0.25 mM EDTA, 1.25 mM magnesium acetate, 2.5% glycerol (v/v), 0.025 mg/ml BSA, 135 mM NaCl) was incubated for 30 minutes at room temperature. Subsequently, the protein-DNA samples were mixed with 6xDNA loading buffer (30% glycerol (v/v), 300 mM boric acid, 300 mM Tris, 0.5 mg/ml bromphenol blue) and run on native polyacrylamide gradient gels (4–12%, Anamed) in 50 mM boric acid and 50 mM Tris at a constant voltage (constant 200 V, 2 hours, Power source, VWR). Afterwards, the gel was placed in a beaker containing running buffer and DNA-stain (dilution 1:60000, GelRed nucleic acid gel stain, Biotium) and rotated for 20 minutes at room temperature prior to DNA-visualization by ultraviolet light (UV Transilluminator, UVP). The values for the apparent binding constant Kapp, referred to as the protein concentration at which half of the total amount of linear DNA in the reaction is bound (room temperature, pH 7.6), were estimated from the respective gel-shift experiments.
The BIAcore 3000 apparatus (GE Healthcare) used for surface plasmon resonance (SPR) was applied to investigate protein-DNA interactions in real-time. Linear oriC-DNA (0.25 pmol, 624 nt) was biotinylated both at its 5´ and 3´ ends (standard PCR with biotinylated primers) and non-covalently bound to a streptavidin coated sensor chip (~1700 RU, Sensor chip SA, GE Healthcare). The system was preequilibrated and permanently flushed (flow rate 20 μl/min) at room temperature with SPR-binding buffer (50 mM HEPES, 300 mM NaCl, 2.5 mM MgCl2, pH 7.6) containing or lacking ATP (2.5 mM). (His)6-wild type/mutant DnaA (2.5 μM) preincubated for one minute with or without 2.5 mM ATP in SPR-binding buffer was subsequently applied to the sensor chip (volume 75 μl) at a flow rate of 20 μl/min, i.e. for 225 seconds, followed by protein dissociation from the DNA. Protein-DNA interactions were measured in real-time over a period of 700 seconds. The response of the interaction of the protein to the oriC-DNA containing chamber was subtracted from unspecific binding to the chip surface monitored in a second DNA-free chamber. SPR-wash buffer (100 mM NaOH, 500 mM NaCl) was injected to remove bound proteins from the chip.
Cells were grown in S750 (B. subtilis) or M9 (E. coli) minimal medium until they reached mid-exponential phase. To ensure continuous nutrition supply and to immobilize the cells they were covered with a pad consisting of S750 minimal medium and 1% (w/v) agarose. For microscopy, an Olympus BX51 microscope with a digital CCD camera (Cool Snap EZ, Photometrics) controlled by the Metamorph 6.3-Software (Meta Imaging Software) was used.
For FRAP experiments a Zeiss Axio Observer A1 with a TIRF objective (100x, immersion oil, NA: 1.45, Zeiss) was used. Images were acquired using a digital EMCCD camera (Evolve, Photometrics). A 515 nm laser (Visitron Systems, Munich) was used to excite and bleach the sample. Acquisition was controlled using the VisiView 2.1.4 software (Visitron Systems, Munich). A custom made macro was applied for acquiring the FRAP sequences. Analysis of FRAP data was performed using Fiji ImageJ [50] and as described by Kleine Borgmann [31]. Acquired streams were aligned using the StackReg plugin. Fluorescence intensities of the region of interest were measured and background fluorescence subtracted. To correct for acquisition bleaching the fluorescence of a control cell was measured and used to normalize the data (single normalization) [51]. Data were additionally normalized to their pre-bleach levels. In order to facilitate fitting, clusters of three individual cells with similar intensities were grouped together, full-scale normalized to correct for different bleach depths and averaged. Fits to the post-bleach data points of each cluster were obtained using the nonlinear regression algorithm from Wolfram Mathematica 8.0. As a model, the function f(t)=A(1−e−krt), with recovery rate kr and recovery level A, was used. The recovery half time t1/2=Ln 2kr was calculated for each cluster. Then the mean of the recovery half times and its standard error were computed. In the plots, one single experiment is visualized together with the corresponding fit. The Student’s t-test was used to compare the means of the recovery halftimes. In order to determine if an immobile fraction was present, we performed double normalization to whole cell intensity [51]. Data from individual cells were normalized by the whole intensity. This corrects for both acquisition bleaching and the initial laser pulse. Proceeding as above, the recovery level A, can then be interpreted as the fraction of molecules which are mobile on the timescale of the experiment.
For single molecule microscopy an Olympus IX71 with an Olympus TIRF objective (100x, ApoN, NA: 1.7) was used. Image acquisition was accomplished using a back-illuminated EMCCD camera (Andor, iXon Ultra 897). The centre of a 20 fold expanded beam from a 100 mW multiline argon laser (JDS Uniphase, laser head: 2219-G5MLS) was focused on the back-focal plane and operated during image acquisition with 150 to 200 W/cm2. An appropriate YFP-laser filter-set was used. For image acquisition the program Andor Solis 4.21 was applied. Streams of 1500 frames of 41 ms or 24.5 Hz were acquired. For faster image acquisition only a subset of 128 × 128 pixel of the chip was read out. High refractive index glass cover slips (n = 1.78) for the Olympus objective were used. Of note, cells continued to grow after imaging, showing that there is little to no photodamage during imaging.
Acquired streams were loaded into Fiji ImageJ [50] and pixel sizes (100 nm) and time increments were calibrated. Tracking of single molecules was achieved using the ImageJ plugin MtrackJ, or u-track 2.0 [52]. Trajectory x/y-coordinates were imported in SMMtrack and parameters like the apparent diffusion constant and the mean square displacement were calculated. SMMtrack (https://github.com/SMMTrack) is a Delphi executable that reads trajectory-data and analyzes the increase of the mean square displacement (MSD). It performs a mean weighted fit on MSD curves of either a whole data set, or just individual trajectories. Also it visualizes individual tracks in superposition with its “heat map”, where each time point in the trajectory emits a constant amount of “heat” into its neighborhood, which will ultimately accumulate to higher degrees of temperatures (= darker shades of red) over time. In order to distinguish real trapping events from trajectories crossing over an additional cooling effect is employed by adding a bell-shaped heat mask for the current trajectory point to a damped version of the previous temperature distribution. The final heat map shows the point wise maximal temperatures of the whole trajectory. SMMtrack also analyzes trapping events within individual trajectories. Trapping events were defined as the time interval in which the trajectory stays within a threshold radius around some trajectory point. This threshold radius is determined by the resolution limit of 230 nm of our microscopic setup. The trapping time distributions in Fig 7 and S10 Fig were derived from disjoint trapping intervals of observed trajectories. Also histogram data was generated of the x/y-proportions of the single step lengths. To estimate the probability that a freely diffusing molecule will be erroneously counted as static we proceed as follows. Firstly, to make the calculation analytically tractable, we allow the bounding circle defining the region in which jumps are considered static to move with the track i.e. we consider individual displacements rather than the net displacement starting from an initial time point as used above. The following false positive rate is therefore likely a slight overestimate. The probability that a molecule with diffusion constant D diffuses less than rmax within one frame interval Δt is given by
p=1-e-rmax24DΔt.
Then the probability that a track of length n contains at least 3 consecutive static frames is then given by
P(n)=∑r=3n(n-2n-r)pr(1-p)n-r,
where the binomial coefficient is the number of ways to arrange n-r non-static jumps in a track of length n-2 (3 consecutive static jumps having being grouped together). Given our distribution of tracks lengths, we can use this result to estimate the false positive rate. For YFP-DnaA, we find a false positive rate of 9%.
|
10.1371/journal.pmed.1002492 | Progression of the first stage of spontaneous labour: A prospective cohort study in two sub-Saharan African countries | Escalation in the global rates of labour interventions, particularly cesarean section and oxytocin augmentation, has renewed interest in a better understanding of natural labour progression. Methodological advancements in statistical and computational techniques addressing the limitations of pioneer studies have led to novel findings and triggered a re-evaluation of current labour practices. As part of the World Health Organization's Better Outcomes in Labour Difficulty (BOLD) project, which aimed to develop a new labour monitoring-to-action tool, we examined the patterns of labour progression as depicted by cervical dilatation over time in a cohort of women in Nigeria and Uganda who gave birth vaginally following a spontaneous labour onset.
This was a prospective, multicentre, cohort study of 5,606 women with singleton, vertex, term gestation who presented at ≤ 6 cm of cervical dilatation following a spontaneous labour onset that resulted in a vaginal birth with no adverse birth outcomes in 13 hospitals across Nigeria and Uganda. We independently applied survival analysis and multistate Markov models to estimate the duration of labour centimetre by centimetre until 10 cm and the cumulative duration of labour from the cervical dilatation at admission through 10 cm. Multistate Markov and nonlinear mixed models were separately used to construct average labour curves. All analyses were conducted according to three parity groups: parity = 0 (n = 2,166), parity = 1 (n = 1,488), and parity = 2+ (n = 1,952). We performed sensitivity analyses to assess the impact of oxytocin augmentation on labour progression by re-examining the progression patterns after excluding women with augmented labours. Labour was augmented with oxytocin in 40% of nulliparous and 28% of multiparous women. The median time to advance by 1 cm exceeded 1 hour until 5 cm was reached in both nulliparous and multiparous women. Based on a 95th percentile threshold, nulliparous women may take up to 7 hours to progress from 4 to 5 cm and over 3 hours to progress from 5 to 6 cm. Median cumulative duration of labour indicates that nulliparous women admitted at 4 cm, 5 cm, and 6 cm reached 10 cm within an expected time frame if the dilatation rate was ≥ 1 cm/hour, but their corresponding 95th percentiles show that labour could last up to 14, 11, and 9 hours, respectively. Substantial differences exist between actual plots of labour progression of individual women and the ‘average labour curves’ derived from study population-level data. Exclusion of women with augmented labours from the study population resulted in slightly faster labour progression patterns.
Cervical dilatation during labour in the slowest-yet-normal women can progress more slowly than the widely accepted benchmark of 1 cm/hour, irrespective of parity. Interventions to expedite labour to conform to a cervical dilatation threshold of 1 cm/hour may be inappropriate, especially when applied before 5 cm in nulliparous and multiparous women. Averaged labour curves may not truly reflect the variability associated with labour progression, and their use for decision-making in labour management should be de-emphasized.
| Dr Emmanuel Friedman’s studies on normal and abnormal labour progression have defined how labour should be managed since the mid-1950s until today. Although Friedman’s studies were conducted among pregnant women in the United States, the general belief that labour progression is the same in humans led to universal application of their findings, and the expectation that the cervix dilates by at least 1 cm/hour in all women during established labour.
Since the early 2000s, however, researchers using new statistical methods to study labour found evidence to suggest that the patterns of labour progression as described by Friedman may not be accurate for the current generation of women giving birth. While these newer findings have informed changes in recommended labour practices in some settings, they have also generated a lot of controversy.
As a result of persistent questions as to whether racial characteristics influence labour progression patterns, recent studies have been conducted among different populations, but not yet in any African population.
We conducted an analysis of prospectively collected observational data of 5,606 women who presented in early labour (at or before 6 cm of cervical dilatation) following spontaneous labour onset and gave birth vaginally in 13 maternity hospitals in Nigeria and Uganda. None of these women experienced serious adverse outcomes for themselves or their babies.
We applied advanced statistical and computational methods (survival analysis and Markov techniques) to determine how long it took the cervix to dilate by 1 cm from one level of dilatation to the next until full dilatation (10 cm) and how long it took the cervix to reach full dilatation based on the dilatation at the time of labour admission. We also used two separate methods to plot population average cervical dilatation time curves (labour curves) for the women in our sample.
Contrary to the generally held view, we found that labour progressed more slowly in our study population than previously reported. On average, the rate of cervical dilatation was less than 1 cm/hour for some women until 5 cm of cervical dilatation was reached among those undergoing their first, second, or subsequent labours.
Labour was very slow in some women throughout the first stage, including the early part of the period that is traditionally known as the ‘active phase’, when the ‘normal’ cervical dilatation rate is expected to be at least 1 cm/hour or faster. While on average the labour progression in first-time mothers was generally similar to their counterparts in the US, China, and Japan, there are also important differences in the slowest-yet-normal (95th percentile) group of women in our study population.
The average labour curves derived from our study population are substantially different from those published from the pioneer work of Friedman. They also do not truly reflect the variations shown in the labour progression of individual women in our study.
The application of population average labour curves could potentially misclassify women who are slowly but normally progressing as abnormal and therefore increase their chances of being subjected to unnecessary labour interventions. We propose that averaged labour progression lines or curves are not used for decision-making in the management of labour for individual women.
As labour may not naturally accelerate in some women until a cervical dilatation of 5 cm is reached, labour practices to address perceived slow labour progression should not be routinely applied by clinicians until this threshold is achieved, provided the vital signs and other observations of the mother and baby are normal.
In the absence of any problems other than a slower than expected cervical dilatation rate (i.e., 1 cm/hour) during labour, it is in the interest of the woman that expectant, supportive, and woman-centred labour care is continued.
| From the mid-1950s until the 1980s, Dr Emmanuel Friedman published a series of landmark studies describing the patterns of labour progression in nulliparous and multiparous women [1–9]. The classic sigmoidal labour curve derived from his work has defined the fundamental basis of labour management for more than six decades. Although Friedman’s studies were limited to obstetric populations in the US, the general notion that the labour progression pattern is largely consistent in humans has led to universal application of their findings and the expectation that the cervix dilates by at least 1 cm/hour in all women. This long-held assumption was the basis for the introduction of ‘Active Management of Labour’ protocols by O’Driscoll and colleagues in the 1970s [10], to ‘normalize’ women’s labour patterns in accordance with the ‘1 cm/hour rule’. However, the escalating rates of unnecessary labour interventions over the last two decades, particularly oxytocin augmentation and cesarean section [11], have renewed interest in what constitutes normal labour progression.
Since the late 1990s and early 2000s, there is increasing evidence to suggest that the descriptions of the relationship between the duration of first stage of labour and cervical dilatation patterns and the definitions of labour dystocia as earlier described may not be appropriate [12–16]. Labour interventions such as induction, oxytocin augmentation, and epidural anaesthesia are now more common, while instrumental and breech vaginal births are becoming rare. The generation of women giving birth in contemporary practice is older, and with increasing body mass index and fetal weight.
In addition, newer research has taken advantage of methodological advancements in computational techniques to address the limitations of studying labour progression and constructing labour curves in the 1950s and 1960s [17]. While these advancements have led to novel findings and new guidance on labour care [18], they are also a subject of intense debate [19–21]. Suggestions that there may be racial and ethnic differences in labour progression patterns as a result of differences in pelvic configurations and sociocultural aspects have promoted research in different obstetric populations [22]. While contemporary labour curves have been published for white, Hispanic, and Asian obstetric populations [14–16], no modern labour curves exist for sub-Saharan African women.
As part of the WHO’s Better Outcomes in Labour Difficulty (BOLD) project, which aimed to develop an innovative and effective labour monitoring-to-action tool [23], we examined the patterns of labour progression in a prospective cohort of women in Nigeria and Uganda who gave birth vaginally without adverse birth outcomes following a spontaneous labour onset.
Scientific and technical approval for this study was obtained from the Review Panel on Research Projects (RP2) of the UNDP/UNFPA/UNICEF/WHO/World Bank Special Program of Research, Development and Research Training in Human Reproduction (HRP), Department of Reproductive Health and Research, WHO. Ethical approval was obtained from the WHO Ethical Review Committee (protocol A65879), the Makerere University School of Health Sciences Research and Ethics Committee, Uganda (protocol #SHSREC REF 2014–058), University of Ibadan/University College Hospital Ethics Committee (UI/EC/14/0223), Federal Capital Territory Health Research Ethics Committee, Nigeria (protocol FHREC/2014/01/42/27-08-14), and Ondo State Government Ministry of Health Research Ethics Review Committee, Nigeria (AD 4693/160). The study was conducted according to the Declaration of Helsinki of the World Medical Association.
The WHO BOLD research project was primarily designed to identify the essential elements of labour monitoring that trigger the decision to use interventions aimed at preventing poor labour outcomes, with the aim of developing a new labour monitoring-to-action tool. The study protocol and detailed methodological considerations have been published elsewhere [23]. In brief, this was a prospective, multicentre, cohort study of women admitted for vaginal birth with single live fetuses during early first stage of labour across 13 hospitals in Nigeria and Uganda. This included women undergoing induction of labour and those with spontaneous labour onset who presented at cervical dilatation of ≤ 6 cm. Women with multiple pregnancies, gestational age less than 34 weeks, elective cesarean section, and those who were unwilling to participate or incapable of giving consent due to obstetric emergencies were excluded. 9,995 women (56.1%) out of 17,810 women who were screened in all hospitals during the study period met these inclusion criteria and participated in the study.
Participating hospitals had a minimum of 1,000 deliveries per year with stable access to cesarean section, augmentation of labour, and instrumental vaginal birth. Estimation of gestational age at birth was in accordance with individual institutional practices, which relied upon the woman’s first date of the last menstrual period in the majority of cases. Labour was managed by midwives or obstetric residents and/or obstetricians. Doppler fetal monitor was used to assess fetal vital status at hospital admission and for intermittent monitoring throughout labour. Labour management protocol, as well as the number and timing of pelvic examinations, were not standardized across participating institutions. None of the institutions subscribed to the ‘Active Management of Labour’ protocol during the study period. Although the partograph was a standard element in all labour protocols, adherence to its application for labour management during the study period varied widely across hospitals.
Eligible women were recruited into the study between December 2014 and November 2015. From the medical record, trained research nurses prospectively extracted detailed information on sociodemographic, anthropometric, obstetric, and medical characteristics of study participants at hospital admission, multiple assessments for labour monitoring and interventions performed throughout the first and second stages of labour, and maternal and neonatal outcomes following labour. Attending staff were approached to complement medical records data when needed. Data collection was limited to the hospital stay of the mother and baby, and there was no follow-up after hospital discharge.
The current study used information on maternal baseline and admission characteristics, repeated assessments of cervical dilatation over time, maternal and neonatal characteristics throughout labour, and perinatal outcome data. This analysis was focused on describing the labour patterns of women without adverse birth outcomes and not on determining correlation to clinical outcomes (See S1 STROBE Checklist). From a total of 8,957 singleton births with consistent time records in the database, we restricted our analysis to examine labour progression to 5,606 women on the basis of the following inclusion criteria (Fig 1): term births (between 37 weeks and 0 days and 41 weeks and 6 days) with vertex presentation and spontaneous labour onset. We excluded women who had labour induction, previous uterine scar, or intrapartum cesarean section. To examine the labour patterns in women with normal perinatal outcomes, we excluded women whose labour resulted in severe adverse outcomes, which was defined as occurrence of any of the following: stillbirth, early neonatal death, neonatal use of anticonvulsant, neonatal cardiopulmonary resuscitation, 5-minute Apgar score < 6, maternal death or organ dysfunction associated with labour dystocia, or uterine rupture. Furthermore, we excluded women who gave birth to neonates with severe congenital malformation and those with fewer than two cervical dilatation assessments during the first stage of labour (since a single data point cannot be used to generate a labour pattern for the individual woman).
We grouped women in the selected sample into three parity groups (0, 1, and 2+) to explore any differences in labour patterns according to parity. We used two independent approaches to analyse labour progression patterns and construct average labour curves for the selected sample. In the first approach, we performed survival analyses to estimate the time it took to progress from one level of cervical dilatation to the next (called ‘sojourn time’) (i.e., from 3 to 4 cm, 4 to 5 cm, 5 to 6 cm, until full dilatation [10 cm]). We used both complete (where available) and interval-censored times to estimate the distribution of times for progression from one integer centimetre of dilatation to the next, with an assumption that the labour data are log-normally distributed. Based on this model, the median, 5th, and 95th percentiles were calculated. We used the same approach to derive the cumulative duration of labour for women presenting at different cervical dilatations (3 cm, 4 cm, 5 cm, and 6 cm) to evaluate any potential differences in the patterns of labour progression. To illustrate the ‘slowest-yet-normal’ labour patterns, we plotted the 95th percentiles for the cumulative duration of labour based on the cervical dilatation at admission. To construct average labour curves, we applied a nonlinear mixed model that best fit our data instead of polynomial models used by previous authors [12–16, 24]. We expressed cervical dilatation for subject i in time j (yij) as a function of time (tij) according to the following three-parameter logistic growth model:
yij=β0+β11+exp(−(tij−(β2+bi)))
in which β0 is dilatation value when tij → −∞, β1 is the asymptotic curve height, and β2 is the inflection point and at this time value when the dilatation reaches half of its height. For simplicity, we estimated β0, β1 as fixed effects and included the random term bi in the inflection point and assumed that this term follows a normal distribution, i.e., bi∼N(0,σb2). Given that women in this analysis entered the cervical dilatation time curve at different dilatations but all ended at full dilatation (10 cm), the starting point (time = 0) on the x-axis was set at full dilatation (10 cm), which was reached by all women in the sample and then calculated backwards (e.g., 1 hour before 10 cm becomes −1 hour and so on). This x-axis (time) was then reverted to a positive value. For example, instead of −12 → 0 hours, it became 0 → 12 hours. We used R-Cran version 3.2 for these statistical analyses [25].
In the second approach, we applied a multistate Markov modelling technique to examine the labour progression patterns in the same sample. This mathematical modelling technique from matrix algebra describes the transitions that a cohort of individuals make among a number of mutually exclusive and exhaustive health states during a series of short time intervals [26]. As cervical dilatation progression is a state- and time-related phenomenon during a period ranging from labour onset through to full cervical dilatation and birth of the baby (i.e., there is a finite set of states), the labour process can be considered a mathematical model that is suitable for the application of multistate Markov modelling. We therefore represented the sequence of labour progress as states based on every observed centimetre from 2 to 10 cm until birth of the baby—the ‘absorbing state’, as illustrated in S1 Fig. At a time t, the woman is in state S(t). The model was designed as a progressive unidirectional model, which only allows a choice of a way out of a particular state, but once a woman has left a state she cannot return. The next state to which a woman moves and the time of the change are governed by a set of transition intensities for each pair of states r and s. The transition intensity represents the instantaneous likelihood of moving from state r to state s. The full set of intensities for the system form the matrix Q.
A Markov process is based on the transition matrix with a probability structure P(u, t + u). The (r, s) entry (the elements of entire matrix) of P(u, t + u), is the probability of being in state s at a time t + u, given the state at time u is r. P(u, t + u) is calculated in terms of Q. Assuming that the transition intensity matrix Q is constant over the interval (u, t + u), as in a time-homogeneous process, P(u, t + u) = P(t) and the equations are solved by the matrix exponential of Q scaled by the time interval, P(t) = Exp(tQ) (S1 Fig). We used msm package for R Project programming environment to fit the multistate Markov model [26]. We generated random observations of cervical dilatation based on the transition matrix P(t) for the entire duration of labour (S2 Fig) to derive average labour curves according to parity and calculated the median, 5th, and 95th percentiles of sojourn times and cumulative duration of labour according to cervical dilatation at admission.
In order to assess the influence of oxytocin augmentation on the described labour patterns, we applied the survival analyses and multistate Markov models to perform sensitivity analyses comparing labour progression patterns of all women with that of a population excluding women with oxytocin augmentation (i.e., our entire study population versus study population excluding women with augmented labours).
The plan for the above survival analyses was first presented at an expert meeting convened by the WHO in November 2016, following which the analyses were started. In February 2017, after a review of the preliminary results of these analyses, the WHO study-coordinating unit requested an independent application of multistate Markov models to the same sample of women in order to determine whether the findings are consistent between the two analytical approaches. From June to July 2017, sensitivity analyses were conducted using the two analytical approaches to assess the influence of oxytocin augmentation on the described labour patterns for the study population, following the suggestions of the BOLD project technical advisory group and study co-authors.
A total of 5,606 women were included in these analyses. Table 1 presents the characteristics of these women by parity. In the selected sample, 54.7% of the women were from Uganda and 45.3% were from Nigeria. Nulliparous women were younger than the multiparous women, constituted over a third of the study sample, and were evenly balanced between the two countries. There was a slight increase in maternal body mass index at birth as parity increased. At labour admission, spontaneous rupture of the membranes had occurred in a quarter of nulliparous women and in about one-fifth of multiparous women. The cervix was well effaced (thin or very thin) in half of the nulliparous and in slightly higher proportions in the multiparous groups. Median cervical dilatation was 4 cm, and the fetal head was not engaged in over 90% of women in all parity groups. There was no caput succedaneum or moulding in over 99% of the women at the time of admission.
In terms of labour interventions, 40% of nulliparous women received oxytocin infusion for labour augmentation, compared with 28% of multiparous women. The median number of vaginal examinations per woman throughout first stage was 3. Presence of a labour companion was observed at least on one occasion in more than half of the women and on two or more occasions in at least a third. While over two-thirds of the women were observed to have taken oral fluids at least once during labour, less than half of them were observed to have done so two or more times. In comparison, oral feedings were observed less frequently, although the observed pattern was similar across parity groups. Severe caput succedaneum and third-degree moulding of the fetal head were rarely seen in any of the parity groups. Labour analgesia and operative vaginal birth were used in less than 2% in the study population; a reflection of the current clinical practices in the study hospitals. While the gestational age at birth was similar across the parity groups, there was an average of a 100-g increase in birth weight with increasing parity.
Table 2 presents the detailed analyses of labour progression based on the two analytical approaches and compares these with the findings of Zhang et al. [14]. The table shows that, based on survival analyses, the median time for the cervix to dilate by 1 cm was longer than the generally accepted limit of 1 hour until a cervical dilatation of 5 cm was achieved in nulliparous women and until 5 cm was achieved in multiparous women. In all parity groups, the median rate of progression doubles as the cervix reaches 6 cm with a median time shorter than 1 hour. Labour progression afterwards escalated more rapidly as it advanced towards 10 cm in all parity groups. Likewise, multistate Markov modelling shows that the median time needed to advance by 1 cm was more than 1 hour until 5 cm was achieved in both nulliparous and multiparous women, and labour progression became more rapid from 7 cm. The distribution of data from both analysis methods show a wide variability around the median for each level of advancement, though this was more pronounced in the survival analyses data. The 95th percentiles of the distribution of sojourn times indicate that labour could progress much more slowly for some women and still result in vaginal birth without adverse birth outcomes. The data show that it was not unusual for nulliparous women to spend more than 7 hours to advance from 4 to 5 cm and over 3 hours to advance from 5 to 6 cm. For some women, the 95th percentile data suggest that throughout the first stage of labour, it took more than 1 hour for cervical dilatation to advance by 1 cm irrespective of the parity groups. The table also shows that the pattern of median times to advance from early to advanced first stage of labour is largely consistent with the findings of Zhang et al. [14], although our 95th percentiles show even wider variability.
Fig 2 shows that the ‘average labour curves’ derived from multistate Markov models for both nulliparous and multiparous women progressed gradually from 4 cm with fairly linear trajectories as they advanced towards 10 cm. The slopes of the curves for multiparous women were steeper than that of the nulliparous women.
The nonlinear mixed models, however, produced smooth labour curves for both nulliparous and multiparous women, which proceeded gradually with a slight upward inclination from around 5 cm and no clear inflection points through 10 cm (S3 Fig). Inflection points appear outside the normal range of observations. Within the range of observed data for cervical dilatation, the curves appear to accelerate from 5 cm, with steeper slopes as they advanced towards 10 cm in multiparous compared to nulliparous women.
S1 Video, S2 Video, S3 Video, and S4 Video are video displays comparing actual plots of cervical dilatation pattern of individual women (starting from 4 cm) with (1) the average labour curves constructed from our study population and (2) the 1 cm/hour alert line of the partograph. The videos show that a substantial proportion of nulliparous and multiparous women crossed the 1 cm/hour alert line as they progressed during labour. The videos also show that substantial differences exist between actual plots of labour progression for individual women and the population average curves.
Table 3 shows the cumulative duration of labour from the cervical dilatation observed at admission (e.g., at 3 cm, 4 cm, 5 cm, or 6 cm) to the next centimetre until 10 cm. The table shows that the median times estimated by the two analysis methods are mostly consistent but also have wide variability in data distribution expressed by their corresponding 5th and 95th percentiles. The rapid progression of cervical dilatation in advanced labour as shown by the sojourn times (in Table 2) is also expressed by the progressively shorter cumulative duration of labour as cervical dilatation on admission increased from 4 to 6 cm. The median rates of ‘linear dilatation’ increased from 1 cm/hour for nulliparous women admitted at 4 cm to 1.3 cm/hour for those admitted at 6 cm. While the median times for nulliparous women admitted at 4, 5, and 6 cm to achieve full dilatation were within the same time frame for dilatation progressing at ≥1 cm/hour, their 95th percentiles show that it was not uncommon to have labours lasting up to 14, 11, and 9 hours in the same categories of women, respectively. The observed cumulative duration of labour in women arriving in labour before 4 cm shows that some of these women did not deliver vaginally until almost 24 hours after admission. The overall patterns are similar for multiparous women, although the medians and their corresponding 95th percentiles were generally shorter than for nulliparous women.
Fig 3, Fig 4, and Fig 5 illustrate the 95th percentiles (in Table 3) plotted as connected staircase lines with specified dilatation at admission having its own corresponding line. Based on the dilatation at admission, women falling to the right of these lines (or thresholds) can be regarded as having protracted or unusually slow labour. From the survival analysis data, for example, if a nulliparous woman who was admitted at 4 cm takes longer than 10 hours to reach 6 cm. Likewise, a nulliparous woman admitted at 6 cm can be considered to be experiencing a protracted labour if she takes longer than 7 hours to reach 8 cm or longer than 9 hours to reach 10 cm. The patterns of cumulative labour duration are similar for all parity groups until 6 cm, when the staircase lines become steeper for multiparous compared to nulliparous women.
Table 4, Table 5, and Table 6 show the results of the sensitivity analyses of labour progression based on our two analytical approaches. As shown in Table 4, the median, 5th, and 95th percentile times to advance by 1 cm were generally shorter when women who had oxytocin were excluded from the study population. The differences between the median times were generally small, less than half an hour in nearly all cases, and mostly confined to the early part of labour (i.e., between 3 and 5 cm). For nulliparous women, the differences in median times ranged from 5 to 22 minutes, while for parity = 1 and parity = 2+ women, it ranged from 1 to 33 minutes and from less than 1 minute to 27 minutes, respectively. The differences in median times centimetre by centimetre became insignificant as labour advanced.
Table 5 and Table 6 show the cumulative duration of labour from the cervical dilatation observed at admission to the next centimetre until 10 cm, excluding women who had oxytocin augmentation. The slightly faster progression of cervical dilatation in the absence of oxytocin augmentation as shown by the sojourn times (in Table 4) is also expressed by the shorter median cumulative duration of labour in all scenarios. For example, considering the cumulative duration of labour for 3 to 10 cm, 4 to 10 cm, 5 to 10 cm, and 6 to 10 cm, the differences in median times were all less than 1 hour regardless of the analysis method used, and the faster progressions were more obvious in women arriving early in labour (i.e., at 3 and 4 cm cervical dilatation).
Fig 6 shows the average labour curves by parity groups after excluding women with oxytocin augmentation. Excluding women who received oxytocin augmentation did not lead to any major change in the pattern or the trajectories of the curves for any parity group. However, the small difference in the labour curves of multiparous groups (as shown in Fig 2) disappeared when women who received oxytocin augmentation were excluded from the analysis. Fig 7, Fig 8, and Fig 9 illustrate the changes in the 95th percentiles (in Table 5 and Table 6) plotted as connected staircase lines for women who received oxytocin augmentation compared to all women. The shorter cumulative labour duration is also reflected in the 95th percentiles for all parity groups regardless of the dilatation at admission, except for nulliparous women admitted at 3 cm, which showed more variability.
Understanding the natural progression of labour presents unique challenges in current obstetric practice. Nevertheless, a gradual shift towards approaches to reduce labour interventions deserves evidence-based information on the upper limits of normal labour to guide practice, especially now that modern analytical methods are available. Contrary to the generally held view, our study shows that in this obstetric population, labour appears to progress more slowly than previously reported [1–3, 27, 28]. The median time needed for the cervix to dilate by 1 cm exceeded 1 hour until dilatation was at least 5 cm in both nulliparous and multiparous women. Labour tended to progress more slowly in the early part of traditional active phase and more rapidly after 6 cm. Considerable variability exists in the distribution of times needed to advance by 1 cm and the duration of labour among women who gave birth vaginally without adverse birth outcomes. For instance, based on 95th percentile thresholds, some nulliparous women took more than 7 hours to advance from 4 to 5 cm, and more than 3 hours to advance from 5 to 6 cm. This pattern of progression was observed irrespective of the analysis method we applied.
While the cumulative duration of labour indicates that a substantial proportion of nulliparous women admitted in labour at 4, 5, and 6 cm achieved full dilatation within an expected time frame if the dilatation rate was ≥ 1 cm/hour, their 95th percentiles show that labour in these women could last up to 14, 11, and 9 hours, respectively, and still lead to a vaginal birth without untoward effects on the mother and baby. Labour could be considerably slow to advance from 3 to 4 cm, and women admitted before 4 cm could have long labours that ultimately end in uncomplicated vaginal birth. Substantial differences exist between actual plots of cervical dilatation over time for individual women and the ‘average labour curves’ derived from our population-level data.
To our knowledge, this is the first attempt to employ modern statistical and computational mathematical methods to assess the patterns of labour in any African population. We used two analytical approaches to determine labour progression and construct labour curves from the same sample in an attempt to explore whether the resulting patterns are independent of analysis methods. We applied these methods to a relatively large and prospectively collected data set from two sub-Saharan African countries comprising multiethnic groups. However, two main limitations need to be highlighted.
First, our study is prone to selection bias that is inherent in the designs of studies of labour patterns in current obstetric practice [17]. Women excluded from our analysis due to cesarean section during the first or second stage of labour may have a different pattern of labour progression compared with women who had vaginal births. Our perception is that this will not impact our study findings, not only because such women constituted 12% of women in whom vertex delivery was anticipated, but also because the inclusion of women who had cesarean sections as a result of labour dystocia during the first stage or failed operative vaginal birth during the second stage could have biased our results towards even longer labours. Additionally, construction of our labour curves was dependent on using 10 cm as the starting point through a reverse approach, and therefore, it was essential that all women in our study sample reached full dilatation. Nevertheless, the exclusion of women whose labours were induced and those with nonvertex presentation implies that our findings may not be applicable to these women. Our findings also need to be interpreted within the context of non- or low use of epidural anaesthesia and instrumental vaginal birth. As these interventions tend to be associated with slower labours, it is reasonable to assume that their low rates in this population would have biased the current findings towards shorter rather than longer labour duration.
Second is the measurement bias that could have been introduced due to inherent subjectivity in cervical dilatation assessments and a lack of standardization of frequency of pelvic examinations across participating hospitals. Additionally, clinical assessments of cervical dilatation can only be estimates that are rounded up to the nearest centimetre. Given the total number of women analysed for each parity group, any bias from intra- and inter-observer variations is likely to be random with potential impact on the data spread but with minimal effects on the point estimates. However, it is possible that the accuracy of our estimations could have been affected by smaller sample sizes in the subgroups that were used to explore various obstetric characteristics. For example, fewer women in our analysed sample presented to the labour ward at 3 cm or less compared to 4 cm and above in all parity groups. While this reflects the prevailing practices in the study hospitals and most maternity units around the world, it is possible that smaller numbers of women did not permit an equally robust analysis of the passive phase of labour and could have contributed to even wider variability in cervical dilatation profiles during this stage.
Our findings provide new data from the perspective of a sub-Saharan African population to support the observations reported in similar studies by Zhang [12–14], Suzuki [16], Shi [15], and their colleagues, which suggest that labour progresses more slowly than previously thought. Similar to these studies, our study reveals that the variability of labour progress in a cohort of nulliparous and multiparous women with vaginal birth is greater than generally appreciated. This variability is apparent even in an obstetric population as selected as ours and is independent of our analysis methods, centimetre of cervical dilatation, or cervical dilatation of the woman at admission.
Despite the general similarities in the nulliparous labour progression pattern between our study and those by Zhang [14], Suzuki [16], and Shi [15] et al., there are important differences in the 95th percentiles reported for sojourn times and cumulative durations of labour. Our 95th percentile times indicate that labour can even be slower than what was reported by Zhang [14] and Shi et al. [15], in their American and Chinese populations, respectively, but not as long as Suzuki et al. [16] reported for Japanese women. While this may be due to the differences in the methods for analysing labour progression, a more logical explanation is the heterogeneity in these study populations in terms of labour interventions and demographic characteristics. For instance, oxytocin augmentation among nulliparous women was more common in the US population (47%) studied by Zhang et al. [14] and our study population (40%), but infrequent (6.5%) in the Japanese population studied by Suzuki et al. [16].
The described patterns of labour progress from our study deviate substantially from what Friedman’s curve indicates [1–3]. The classic sigmoidal pattern was not observed in our average labour curves. This may be due to the fact that the majority of the women in our study were not admitted early enough in labour to substantially reflect the pattern of the passive phase of labour and because of the lack of documented assessment of 9-cm dilatation in our cohort, which precluded exploration of any deceleration between 9 and 10 cm. In his series of 500 nulliparous women [2], Friedman used the mean values of the four separate phases of individually plotted sigmoid curves to derive the mean labour curve and reported 1.2 cm/hour as the minimum value of ‘phase of maximum slope’ based on the 95th percentile point on the distribution curve. The nulliparous average curves from our cohort are less steep, and the 95th percentile values from one level of dilatation to the next during the traditional active phase yielded median rates between 0.1 and 0.5 cm/hour between 4 and 10 cm. It remains unclear to what extent an average labour curve depicts the variability associated with individual women’s labour progress, and its value in clinical practice is becoming increasingly questioned. The differences illustrated by the video displays of individual labour profiles, compared to the average labour curves for this cohort, indicate how unreliable a population average curve is in representing an individual woman’s labour progression profile.
In an attempt to overcome the shortcomings of Friedman’s labour curves, Zhang et al. [12] proposed the use of repeated measures analysis with polynomial modelling as a superior method for constructing labour curves, given its flexibility to fit labour data. Other investigators using the same statistical method have confirmed a similar pattern of labour curves published by Zhang et al [12–14]. However, we found that the polynomial model was not appropriate for our data, as it presents a behaviour that is incompatible with labour curve modelling. Rather, we applied multistate Markov modelling to overcome the unpredictable nature of cervical dilatation [29], since its models can accommodate the inherent randomness in cervical dilatation over time [30] and it has the advantage of providing a better representation of real life scenarios from more angles by including empirical observations. We also applied a nonlinear mixed model because of its advantages in terms of interpretability, parsimony, and validity [31]. Although the curves obtained from our nonlinear mixed models are similar to those constructed through polynomial models by previous authors [12, 14, 15], they should be interpreted with caution, as the model appears dependent on extrapolation beyond the normal range of observations for women in the sample.
An interesting finding in our study is the median cumulative duration of labour (e.g., from 4 to 10, 5 to 10, and 6 to 10 cm), which, when considered linearly, suggests that the cervix was dilating at ≥ 1 cm/hour. However, such interpretation hides the nonlinearity of labour progression patterns for most women and does not account for slower progress at the beginning of the traditional active phase and faster progress when active phase is advanced. This implies that some women within the 95th percentile boundary as shown in our study will be categorised as having protracted labour if current labour standards were applied. For instance, a woman with reassuring maternal and fetal conditions who remains at 4 cm for 4 hours may be subjected to oxytocin augmentation when she could still be within her normal limits before advancing to 5 cm. Application of interventions too soon when a woman is still within the boundaries of her normality probably accounts for escalating rates of interventions to expedite labour globally.
One subject of debate in the analysis of labour progression patterns in contemporary practice is the potential impact of oxytocin augmentation on observed labour patterns. A widely held view is that the inclusion of women with augmented labour is likely to produce faster labour progression profiles, and the restriction of analysis to women without labour augmentation will generate labour profiles that reflect natural labour progression. However, we found the contrary, as the exclusion of women with augmented labours from our study population resulted in generally faster labour progression patterns. Although unexpected, this finding was not surprising, as it reflects the impact of Friedman’s original curves and their derivative tools on labour management even today. Women with augmented labours were those assessed by labour attendants as having slower than normal progression based on a preconceived expectation of 1 cm/hour cervical dilatation. Therefore, their exclusion from the analysed study population leaves a highly selected population of women whose labour progression, by the assessment of the labour attendants, conformed to this preconceived expectation and did not require labour augmentation. While the overall clinical implications of the altered progression in terms of labour duration are minimal, our findings support the inclusion of women with augmented labours in the analysis of labour progression in the context where use of oxytocin is the norm so as to facilitate applicability of their findings.
We acknowledge that the described labour patterns from this cohort may be related to the demographic characteristics and prevailing clinical practices in our study settings. Nevertheless, a number of clear messages emerged from our study. First, population average labour curves are at best estimates that may not truly reflect the variability associated with labour progress and could potentially misclassify individual women. It appears that average labour curves are dependent on the underlying assumptions and principles governing the statistical methods from which they are derived. We conclude that population average labour curves are merely useful for illustrative purposes.
Secondly, our labour progression data clearly demonstrate that a minimum cervical dilatation rate of 1 cm/hour throughout the period traditionally described as active phase may be unrealistically fast for some women and should therefore not be universally applied as a threshold for identifying abnormally progressing labour. Likewise, for most nulliparous and multiparous women, labour may not accelerate until a threshold of at least 5 cm is reached. The implication is that a cervical dilatation rate slower than 1 cm/hour throughout the first stage of labour, especially before 5 cm, should not be an indication for interventions to expedite labour provided maternal and fetal vital signs and other observations are normal. It would be useful for labour care providers to consider the upper boundaries reported in this cohort when reviewing whether an intervention is justified. It is important to note, however, that the presented percentile values are insufficient to define abnormal labour that requires interventions to avert adverse outcomes. As this is a selected sample of women without adverse birth outcomes, we cannot conclude from the current analysis whether women with cervical dilatation progressing beyond our percentile values (or other specific boundaries) have comparatively higher risk of adverse birth outcomes. As cervical dilatation is a reflection of a complex interaction of biological, physical, and psychological factors during the course of labour, it is imperative that women with a suspicion of protracted labour be carefully evaluated to exclude developing complications (e.g., cephalopelvic disproportion) and to ensure that the woman’s physical and emotional needs are being met. In the absence of any problems other than a slower than expected cervical dilatation (i.e., 1 cm/hour), it is in the interest of the woman that expectant, supportive, and woman-centred labour care is continued.
We propose that averaged lines or curves are not used for decision-making in the management of labour for individual women. Efforts should focus on developing individualised (or personalised) labour management algorithms that optimize woman-centred health outcomes. Decision-analysis models and machine learning technologies that are available today can assist in achieving this objective.
|
10.1371/journal.pcbi.1002749 | The Role of Flexibility and Conformational Selection in the Binding Promiscuity of PDZ Domains | In molecular recognition, it is often the case that ligand binding is coupled to conformational change in one or both of the binding partners. Two hypotheses describe the limiting cases involved; the first is the induced fit and the second is the conformational selection model. The conformational selection model requires that the protein adopts conformations that are similar to the ligand-bound conformation in the absence of ligand, whilst the induced-fit model predicts that the ligand-bound conformation of the protein is only accessible when the ligand is actually bound. The flexibility of the apo protein clearly plays a major role in these interpretations. For many proteins involved in signaling pathways there is the added complication that they are often promiscuous in that they are capable of binding to different ligand partners. The relationship between protein flexibility and promiscuity is an area of active research and is perhaps best exemplified by the PDZ domain family of proteins. In this study we use molecular dynamics simulations to examine the relationship between flexibility and promiscuity in five PDZ domains: the human Dvl2 (Dishevelled-2) PDZ domain, the human Erbin PDZ domain, the PDZ1 domain of InaD (inactivation no after-potential D protein) from fruit fly, the PDZ7 domain of GRIP1 (glutamate receptor interacting protein 1) from rat and the PDZ2 domain of PTP-BL (protein tyrosine phosphatase) from mouse. We show that despite their high structural similarity, the PDZ binding sites have significantly different dynamics. Importantly, the degree of binding pocket flexibility was found to be closely related to the various characteristics of peptide binding specificity and promiscuity of the five PDZ domains. Our findings suggest that the intrinsic motions of the apo structures play a key role in distinguishing functional properties of different PDZ domains and allow us to make predictions that can be experimentally tested.
| Proteins that are capable of binding to many different ligands are said to have broad specificity. This is sometimes also referred to as promiscuity. Whether a protein is promiscuous or not can sometimes be readily explained by the structure of the protein and the ligand in terms of electrostatic and steric effects. Sometimes however, this simple interpretation can struggle to explain the experimentally observed data. A prominent case in point is the PDZ domains. These small protein domains bind to unstructured regions of other proteins and are involved in many signaling pathways. Some PDZ domains appear to be more promiscuous than others, but this has been difficult to explain purely on the basis of the composition of residues in the binding groove. In this work we examine the dynamics and conformational flexibility of five key PDZ domains and demonstrate that despite similar folds, these proteins can exhibit quite different dynamics. Furthermore the difference in the dynamic behavior appears to correlate with the observed promiscuity. Our findings suggest that knowledge of the dynamic behavior of the PDZs can be used to rationalize the extent of expected promiscuity. Such knowledge will be critical for drug design against PDZ domains.
| A number of structural studies comparing holo and apo forms of proteins have demonstrated that ligand binding is often coupled to conformational changes of the interacting partners [1]–[3]. The real challenge is, however, to uncover the exact sequence of events resulting in the observed structural changes. Two main models, the induced fit (Koshland) and the conformational selection (or population shift) hypothesis (see [4] for a review), have been introduced to describe the limiting cases of the complex process of molecular recognition [5]–[8].
According to the induced fit model, ligand binding happens first and the formation of a ‘weak complex’ is followed by the conformational rearrangement of the protein that results in stronger binding [9]. By contrast, in the conformational selection model, the intrinsic dynamics of the protein lead it to spontaneously transition between a stable unbound and a less stable ‘bound conformation’. As the apo protein actually visits the bound state with significant probability, the ligand can bind directly to this conformation shifting the distribution of conformers towards the bound population. As recently reviewed [4], it seems likely that the induced fit and conformational selection mechanisms often act together in the ligand recognition process.
Furthermore, in terms of protein-protein interactions, it is increasingly clear that many proteins display functional promiscuity which requires them to be able to interact with multiple partners [10]. If the conformational selection mechanism is involved in promiscuous ligand binding, this assumes that the protein needs to visit multiple (often dissimilar) binding conformers capable of binding the different ligands. An example of structural evidence of such multi-specificity can be found in the X-ray crystallography study of the SPE7 antibody (a monoclonal immunoglobulin E raised against a 2,4-dinitrophenyl hapten) that has been shown to adopt different binding conformers and is consequently able to bind to multiple unrelated antigens [11]. Another example is the NMR study of apo ubiquitin which has found that this protein exists in an ensemble of conformers that are almost identical to the conformations of ubiquitin in complex with 46 different binding partner proteins [12]. One of the best known examples of a promiscuous enzyme is cytochrome P450 which has been shown to adopt a great variety of active site conformations and is able to bind and transform diverse substrates [13].
As shown by these examples, the intrinsic dynamics of promiscuous proteins let them visit multiple unrelated binding conformers and the property of multispecificity seems to be related to conformational flexibility. Promiscuous proteins that are able to bind to multiple partners through conformational selection need to explore a larger conformational space than those that bind to only a single partner. More rigid binding sites therefore may have restricted specificity with the benefit of higher binding affinity.
Indeed, a study of human cytochrome P450 enzymes has found that while a relatively rigid member of the family (CYP2A6) has narrow substrate specificity, the most flexible member (CYP3A4) is also the most promiscuous one [14]. Functionally promiscuous proteins could be of key importance for the emergence of new functions in protein evolution. Recent research about the relationship between binding promiscuity, conformational flexibility and evolvability of proteins has been reviewed by Tokuriki et al. [15], [16]. As discussed in these reviews, these studies suggest that for proteins that exist in equilibrium between a highly populated native state (interacting with a native ligand) and less populated conformers (binding to alternative partners), mutations can gradually shift the equilibrium towards a promiscuous conformer. This can eventually lead to a new dominant primary function. While mutations may be neutral with regards to the original function (i.e. hardly change the relative occupancy of the native conformer), they may cause significant increase in the occupancy of the alternative conformer. On the other hand, point mutations that reduce the occupancy of promiscuous conformers may result in a decreased flexibility (rigidification) but increased specificity (and higher affinity) for the native ligand as for example observed in the process of antibody maturation [17]. Promiscuity may therefore be a common feature of highly evolvable proteins.
Despite their highly conserved overall fold and binding sites, PDZ (PSD-95, Dlg, ZO-1) domains have been found to have surprisingly diverse binding specificities [18]. PDZ domains bind peptidic ligands, usually located at the C-terminus of partner proteins. A recent study at the genome level confirmed that this location is dominant [19], but other modes of interaction have also been reported [20]–[22].
Although a series of different classification systems have been proposed aiming to organize PDZ domains based on their preference towards peptide ligands there is no consensus on the best way of classification [23], [24] although some progress has been made towards mapping determinants of specificity [25]. PDZ specificity turned out to be unexpectedly complex as many PDZ domains are able to bind to multiple ligands that belong to different classes of peptide motifs. This property is often referred to as degenerate specificity, multivalent specificity or most commonly, binding promiscuity [10]. In addition, single peptides have been shown to bind to multiple PDZ domains. The complex picture of PDZ-peptide interactions therefore makes it rather difficult to develop a simple specificity-based classification scheme.
In addition, very little is known about what determines the specificity and promiscuity of PDZ domains. To address this question, Stiffler et al. [26] have used protein microarrays and quantitative fluorescence polarization to study the binding specificity of 157 mouse PDZ domains and found only a weak correlation between the pairwise sequence divergence of PDZ domains and their divergence in selectivity space. The fact that overall sequence similarity proved to be a poor predictor of PDZ domain function indicates that the majority of sequence variation in the PDZ family is neutral with regards to peptide-binding selectivity. This also suggests that binding specificity is mostly determined by only a subset of residues that are likely to be located in the binding pocket of the domain [26].
In order to study the sequence determinants of specific ligand recognition, Tonikian et al. [25] performed mutagenesis at ten binding site positions in the Erbin PDZ domain. As a result, they identified several mutations that altered binding specificity. Since not all of these critical residues were in direct contact with the ligand, Tonikian et al. concluded that both direct interactions and cooperative, long-range effects may play important roles in determining the specificity of PDZ domains [25].
In a recent study, using a combinatorial peptide library and site-directed mutagenesis, Shepherd et al. [27] have found that only four point mutations were enough to switch between the distinct binding specificities of the Tiam1 (T-cell lymphoma invasion and metastasis 1) PDZ and Tiam2 PDZ domains. Gee et al. [28] have come to similar conclusions after performing in-vitro mutagenesis studying the PDZ domains of PSD-95 (postsynaptic density protein 95) and α1-syntrophin. By identifying a few critical sequence positions, they have found that single-amino acid substitutions can alter specificity and affinity of PDZ domains for their ligands. The fact that ligand specificity relies on minor sequence modifications, while the chemistry of the binding pocket and the overall fold are well conserved, suggests a very favorable flexibility property of the PDZ domain fold [29]. PDZ domains are both versatile and robust because mutations frequently change their specificities without a loss of function [25]. Similar robustness under high mutational pressure has also been observed for other peptide-binding domains, for example the WW [30] and SH3 domains [31].
On the other hand, a number of experimental and computational studies (outlined below) have shown that the conformational dynamics of PDZ domains may also play a crucial role in determining binding specificity. These results suggest that the intrinsic fluctuations of PDZ structures are also likely to be related to the selectivity for peptide ligands. Recently, Gerek et al. [32] used a modified coarse-grained elastic network model to find characteristic residue fluctuation patterns for PDZ domains belonging to different specificity classes. By clustering these residue fluctuation profiles, they have identified common motion characteristics of Class I and Class II type PDZ domain interactions [32].
Basdevant et al. performed 20–25 ns molecular dynamics simulations of 12 PDZ domain complexes and used the MM/PBSA (Molecular Mechanics/Poisson-Boltzmann Surface Area) method to analyze electrostatic, nonpolar and configurational entropy contributions to the binding free energies [33]. Their results show that the degree to which the dynamics of the peptide ligands are coupled to those of the PDZ domains varies highly. They concluded that complex-specific dynamical or entropic responses may form the basis of the selective recognition of peptides. It is important to note that different flexible docking strategies have already been proposed to be able to incorporate the effect of binding site flexibility in structure-based drug design studies targeting PDZ domains [34], [35].
Another aspect that has been investigated is the role of temperature on binding behaviour. Staneva and Wallin [36] applied an all-atom Monte Carlo based approach to analyze various aspects of the process of peptide binding to PDZ domains. They found that the probability that peptide ligands can occupy the correct bound state in the simulations increased sharply with the decrease of temperature. In another study, Cecconi et al. [37] have analyzed the temperature-dependence of the unbinding of peptide ligands from PDZ domains. They have found that the free-energy landscape determining the kinetics of ligand escape is sensitively dependent on the temperature. However, PDZ-peptide complexes are stabilized within a physiologically relevant temperature interval.
Given all of the above, we were interested in the role of conformational dynamics in determining the ligand binding specificity of PDZ domains. In particular, given the possible relationship between flexibility and promiscuity, we wanted to examine how well the property of multi-specificity of these domains is correlated with the flexibility of their binding pockets. We were also interested to examine to what extent PDZ domains obey the conformational selection versus induced fit mechanism. We thus selected five, well-characterized, PDZ domains: Dvl2 PDZ capable of binding both C-terminal and internal (i.e. not at the terminus of a protein) peptides and shows large conformational changes between binding modes, Erbin (ERBB2 interacting protein) PDZ which binds both class I and class II ligands, but comparison with the apo structure reveals very little conformational change, InaD PDZ1 for which it is known that peptides bind in different modes, but structural information is thus far only available for one mode, PTP-BL PDZ2 for which induced fit has been predicted to be important in the binding process and GRIP1 PDZ7 for which structural studies suggest that the binding cleft is not capable of binding peptides in the expected manner for PDZ domains.
All five of the aforementioned PDZ domains are of clinical interest due to their central role in disease pathways. Four of these PDZ domains (Dvl2 PDZ, Erbin PDZ, InaD PDZ1 and PTP-BL PDZ2) are promiscuous in the sense that they are able to interact with multiple partners. However, while for example, Dvl2 PDZ is capable of interacting with peptides using different binding modes (binding both classical C-terminal and non-classical internal peptides), Erbin PDZ is able to interact only with very similar peptide binding modes.
On the basis of this, one can formulate a definition of strong promiscuity, which is the ability to interact with multiple ligands that require the binding pocket to adopt significantly different conformations. In this sense, Dvl2 PDZ is promiscuous and Erbin PDZ is not. If conformational selection plays a role in the recognition of peptides, the above-defined property of promiscuity must correlate with intrinsic conformational flexibility since the binding pocket needs to visit all different conformations required for binding multiple ligands. In this paper we explore the relationship between the dynamics, promiscuity and flexibility of PDZ domains. The results have implications for many protein-protein interaction pathways.
To explore the role of conformational selection and flexibility and its relationship to promiscuity of binding we examined five well-documented PDZ domains (Table 1, Figure 1) with 200 ns molecular dynamics simulations (See Methods). Although the sequence identity is between 19 and 30%, the structural similarity, as measured by the root-mean squared deviation of the Cα carbons, of these domains is high, especially in the binding site (Table 2).
To compare the inherent flexibility of the five PDZ binding pockets, we used a measure of the overall fluctuation, Θ, which reflects the mean pairwise distance variance of binding pocket residues (See Methods for details). This approach has the added advantage that it is not superposition dependent as it only depends on distances rather than coordinates. The overall fluctuation was calculated for the five conformational ensembles of the 200 ns MD simulation trajectories (40000 snapshots for each PDZ domain). We assessed the convergence of the trajectories via calculation of the root mean square inner product (RMSIP) and obtained values between 0.59 and 0.69 for the binding pocket residues (and high overlaps for the full proteins as well) from the simulations which according to Lagerge and Yonetani [38] suggests adequate convergence (see Supporting Information, Text S1, for more details). The Θ fluctuation values of the five binding pockets (i.e. the five sets of binding site residues defined by the multiple sequence alignment) are summarized in Table 3. As discussed in Methods, the Θ measure shows the size of conformational space the binding pocket explores in the simulation. Table 3 shows that despite the high structural similarity of the five binding sites (Table 2), one can see large differences in the extent of their intrinsic fluctuation. The InaD PDZ1 and Dvl2 PDZ binding sites have the most flexible binding pockets, while the binding site of Erbin PDZ is the most rigid of these five PDZ domains. The Θ value of Dvl2 PDZ is almost twice as large as that of Erbin PDZ.
These results are in good agreements with the conclusions of experimental studies [22], [39]–[42] that have found that Erbin PDZ binding site shows little structural variability while the Dvl2 PDZ binding site is flexible showing large structural variation. The results suggest that the rigidity/flexibility of these binding sites demonstrated in other studies by comparison of apo and holo crystal structures can be explained by the intrinsic dynamics of the apo proteins.
The flexibility of the binding pocket of the Dvl2 PDZ domain has been discussed in the literature before [22]. Therefore we decided to compare the dynamics between Dvl2 PDZ and Erbin PDZ domains. The difference in the overall fluctuation of the two binding pockets can also be seen in their fluctuation matrices (Figure 2A,B), defined as the matrix of variance of pairwise residue distances. We also define “flexibility” as a measure of the maximum range any pairwise residue distance can exhibit (see Methods). The flexibility matrices, which essentially capture extreme movements, reveal that, as expected, there are regions of high flexibility for Dvl2 PDZ. They also reveal, unexpectedly that although the fluctuation matrices suggested that Erbin PDZ is quite rigid, they also highlight that there is flexibility in terms of the distance between K396 (located at the C-terminal of the α1 helix) and the β2 strand and in particular S335 (see Supporting Information Figure S1). Taking the result of the fluctuation and flexibility matrices together suggests that a section of the binding site can open up considerably, but that these extremes in conformation are infrequent and essentially the Erbin PDZ binding site behaves as a rigid structure.
To better understand the role that intrinsic dynamics might play in ligand binding to the Dvl2 PDZ domain, we performed the fluctuation and flexibility analysis on an experimentally derived ensemble. We took the structure of the apo Dvl2 PDZ domain (PDB code: 2rey) and four crystal structures of different ligand-bound conformations (PDB codes: 3cbx, 3cby, 3cbz and 3cc0 which are also referred to in the literature as the pep-C1, pep-N1, pep-N2 and pep-N3 complexes [22]). The pep-C1 structure exemplifies C-terminal ligand binding, whereas the other three illustrate internal ligand binding. The flexibility matrix was computed for this ensemble and is shown in Figure 3. The matrix shows us which binding pocket residue pairs have the largest relative displacement between the apo and ligand-bound structures. The experimentally derived flexibility matrix has remarkable similarity to the simulation-based fluctuation and flexibility pattern (Figure 2A and C) with a correlation of 0.74 and 0.68 respectively. The largest displacements seen experimentally are for residues I14 and S15 with respect to the α2 helix, which is the same as that observed in the simulations. This suggests that the Dvl2 PDZ domain is capable of visiting conformations that are consistent with the ligand-bound conformations even in the absence of ligands.
We were interested to know how the snapshots of the MD simulations were distributed in conformational space. To that end we performed multi-dimensional scaling on snapshots taken every 50 ps (the first ns of the trajectories were excluded). The input was thus a 3981×3981 matrix containing the pairwise dRMSD dissimilarity values of 3981 conformers. Groups of similar conformers were identified with the k-means cluster analysis and clustering was validated with the silhouette index measure (see Methods). The optimal number of clusters corresponding to the maximal overall average silhouette index (SOVER = 0.411) was found to be 2. Figure 4A shows the results of multi-dimensional scaling (MDS) where conformations are represented by dots on a 2D-map and similar conformers are adjacent. The map suggests that the conformational space can be split into two distinct contiguous clusters.
When the ligand-bound structures were included in this MDS analysis (Figure 4B) it was found that one (pep-N2) resides within cluster two, whilst the other three sit within cluster one. Examination of ligand-bound conformations with, superposed on, the medoid conformations (ie. those conformations that are most representative of the ensemble) from the two clusters (Figure 4C, and D) shows that the key difference lies in the motion of I14 and S15 (at the N-terminal of the β2 strand) relative to the α2 helix. Thus the intrinsic fluctuations observed for the Dvl2 PDZ domain allow access to two distinct conformational states that have been captured in ligand-bound structures.
In contrast, MDS performed on the data for the Erbin PDZ domain shows just one cluster and the two experimental structures lie within this cluster (Figure 4E). The complex with the class I peptide is located close to the cluster center, while the complex with the class II peptide is placed closer to the edge of the cluster. The Erbin PDZ domain is promiscuous in the sense that it binds multiple peptides, but the experimental structures and this analysis shows that these peptides are essentially binding to the same conformational state and thus it does not satisfy the “strong” definition of promiscuity defined here.
The binding pocket of the InaD PDZ1 domain has the largest overall fluctuation of the five PDZ binding pockets examined here (Table 3). The fluctuation matrix (Figure 5A) shows that the part of the InaD PDZ1 domain that fluctuates the most is the three residues at the C-terminal end of the α2-helix (I91, K92, and E93) with regards to the entire β2-strand. However, the flexibility matrix (Figure 5B) shows that this PDZ domain can undergo even larger distortions within the binding cleft. MDS analysis of the conformational ensemble identified two main clusters (Figure 5C) with the known experimental structure of the InaD PDZ1 domain in complex with the NorpA peptide (PDB code: 1ihj) belonging to cluster one. The overall average silhouette index, SOVER was 0.43 and as can be seen the division between clusters is not as distinct as for the Dvl2 PDZ.
The presence of two distinct clusters for InaD is intriguing and raises the question of whether the second cluster has biological relevance. Besides the NorpA peptide, the InaD PDZ1 domain has been shown to bind to the unconventional myosin NinaC. Intriguingly the experimental results [43] suggest that InaD PDZ1 may interact with NinaC in a different mode than it does with NorpA [44]. If conformational selection plays a role in this interaction, then one would expect the InaD PDZ1 domain to be relatively flexible in the apo state and able to visit distinct regions of conformational space, which is exactly what is observed here. Conformations in Cluster 1 are likely to be relevant for NorpA binding whilst excursions into Cluster 2 may be essential for NinaC peptide binding.
Taken together these data suggest that the InaD PDZ1 domain is likely to satisfy the ‘strong’ definition of promiscuity as it probably binds to different partners using considerably different binding modes. Thus we would predict from this that any structure of the InaD PDZ1-NinaC complex would be placed into Cluster 2 of the MDS analysis. These results are in support of the earlier anticipation of Kimple et al [44] and Wes et al [43] that InaD PDZ1 binds to NorpA and NinaC using different binding modes. Based on our results we would predict that the difference is expected to be in the shift of the C-terminal end of the α2-helix (I91, K92 and E93) with respect to the β2-strand.
Although InaD PDZ1 (and also Dvl2 PDZ) has distinct conformation clusters defined by k-means clustering, it is perhaps also useful to define states in terms of kinetics. We performed a temporal analysis to ascertain whether our geometrically defined states are supported by a kinetic definition simply defined by asking is the intra-cluster relaxation time faster than the inter-cluster transition time (see Supporting Information, Text S1, for details). For InaD PDZ1, the average inter-cluster transition time was 14.1 ns whilst the intra-cluster relaxation time was 100 ps. Similarly for Dvl2 PDZ, the average inter-cluster transition time was 7.03 ns whereas the intra-cluster relaxation time was again 100 ps. Thus, this analysis suggests that the conformational clusters defined by the dRMSD similarity measure correspond to kinetically separated, metastable states of the protein.
The fluctuation pattern of PTP-BL PDZ2 (Figure 6A) shows that this domain [45] has a considerably rigid binding site, similar to the Erbin PDZ domain. However, the flexibility pattern (Figure 6B) reveals that the N-terminal end of the α2 helix is flexible with regards to the β2 strand. MDS analysis (Figure 6C) shows that the majority of conformations appear to be distributed within a single compact cluster, which also has a large number of outliers.
The results here place the experimental ligand-bound conformation in the main conformational cluster, but that does not rule out the possibility that induced-fit plays an essential role in the binding process. Indeed for PTP-BL PDZ2, the binding to the Adenomatous Polyposis Coli-protein (APC) peptide has been proposed to occur through induced fit [46]. In order to investigate this, the structural differences between the APC-bound conformation and the most similar (neighboring) conformations sampled in the apo MD simulation were characterized using Q values (see Methods) which is introduced as a quantitative measure of similarity. Table 4 summarizes the results of this analysis for the PDZ domains where there is a complex reported. It can be seen that the complex of PTP-BL PDZ2 domain with the APC peptide is the least similar to the apo MD simulation ensemble. It has the highest Q(1) and Q(10) values (0.37 Å and 0.39 Å) which represent the average dRMSD dissimilarity between the ligand-bound conformer and the most similar and ten most similar stimulation snapshots, respectively. By contrast, the complex of InaD PDZ1 with the NorpA peptide has significantly lower Q(1) and Q(10) values (0.15 Å and 0.18 Å) indicating that this ligand-bound binding site conformation is more closely approached in these simulations. We also performed an additional simulation of the PTP-BL PDZ2 domain in complex with the APC peptide to examine whether the presence of the peptide kept conformational space closer to the ligand-bound crystal structure. As expected, the Q(1) and Q(10) values are lower (see Supporting Information, Text S1) for the simulation with the peptide bound compared to apo (Table 4), lending further support to the induced fit mechanism.
Although due to sampling limitations, we are unable to tell if the apo structures get any closer to the peptide-bound conformations in reality, the data presented here suggest that, out of the five PDZ domains studied here, PTP-BL PDZ2 is the most likely to involve an induced fit mechanism when binding to the APC peptide. Figure 6D shows the mean absolute difference distance matrix (Δ) pattern calculated between the peptide-bound structure and the 100 most similar snapshots. We can see that the largest deviations are found in the distances between S28 and L85 and between V29 and R86. The Δ pattern suggests that these two inter-residue distances are altered the largest extent upon binding to the APC peptide.
Visual inspection of the PTP-BL PDZ2 trajectory shows some subtle rearrangements of the protein from the starting crystal structure. The movement between residues S28 and L85 along with V29 and R86 appears to be facilitated by re-arrangement of the “pre-β2” loop and the “post-α2” loop and the movement of K10, the side-chain of which appears to act as a helix cap for the α2 helix most of the time. As these movements occur, water molecules penetrate deeper into the cleft but are then expelled as the cleft returns to conformations more similar to the starting structure. However, the whole structure appears to be further stabilized by the formation of salt-bridge between residues D22 and K50 which is not initially present in the crystal structure (see Supporting Information, Figure S2). The overall change in shape of the pocket in this extreme is similar in nature to opening of the Erbin PDZ domain (which occurs infrequently – see Supporting Information, Figure S1).
The solution structure of the GRIP1 PDZ7 domain [47] suggests that the α2/β2 binding pocket adopts a “closed conformation” and has a significantly smaller carboxyl peptide-binding site than other PDZ domains which would restrict its ability to interact with peptides. However, it is the case that other PDZ domains with similar closed pockets appear to be able to open up in order to incorporate a peptide ligand such as LARG PDZ domain [40]. Thus we examined the conformational dynamics of the GRIP1 PDZ7 domain to see if it would open up to a conformation capable of peptide binding.
The fluctuation and flexibility patterns of the GRIP1 PDZ7 binding pocket (Figure 7A and B) show that the N-terminal end of the β2-strand has notable fluctuation with regards to the C-terminal end of the α2-helix. On the other hand, the patterns also show that the C- terminal end of the β2-strand has little mobility with regards to the N-terminal end of the α2-helix. Since the bottom of the binding pocket is located between the C-terminal end of the β2-strand and the N-terminal end of the α2-helix, their low relative fluctuation suggests that the base of the binding site does not open significantly.
In order to examine this in more detail, and in a comparative way to the other PDZ domains studied here, the distance between the C-terminal residue of the β2-strand and the N-terminal residue of the α2-helix was used to characterize to what extent the base part of the binding pockets in all the PDZ domains is open. Figure 7C shows these distance distributions for each PDZ domain. The distributions for Erbin PDZ and PTP-BL PDZ2 are almost identical (both are approximately Gaussian functions with a mean of 6.28 Å and 6.4 Å, and standard deviation of 0.29 Å and 0.49 Å, respectively) indicating that the base parts of the binding groove of these two PDZ domains behave in a very similar fashion. The distribution of the InaD PDZ1 domain, however, has larger spread (a standard deviation of 0.58 Å), but the mean distance is about the same (6.35 Å) as for Erbin PDZ and PTP-BL PDZ2. Interestingly, the distance distribution of Dvl2 PDZ is a superposition of two Gaussian distributions (with a mean of 6.0 Å and a standard deviation of 0.58 Å). However, the location of one of the two superposed Gaussian curves agrees well with the distributions observed for Erbin PDZ and PTP-BL PDZ2.
Most importantly, the distance distribution of GRIP1 PDZ7 (which can be approximated well as a single Gaussian distribution) is significantly shifted relative to the other four distributions. It has a mean of only 5.68 Å, and a standard deviation of 0.39 Å. The probability that the base part of the binding pocket is open with an extent larger than 0.6 Å is considerably lower in the case of the GRIP1 PDZ7 domain but is high in the four other PDZ domains. These results show that the bottom of the binding groove of the GRIP1 PDZ7 domain is closed and it remains closed in the course of the 200 ns MD simulation unlike in other PDZ binding sites. This unique property of GRIP1 PDZ7 is probably the reason why this PDZ domain has been found to be unable to bind to carboxyl peptides.
The intrinsic dynamics of the binding sites of five PDZ domains have been compared in this paper, based on 200 ns all-atom molecular dynamics simulations of the apo structures. Despite the remarkable structural similarity of the five PDZ folds and binding sites, their fluctuation and flexibility properties have been found to be surprisingly different. Furthermore, the differences of their mobility correlate well with differences of their functional properties suggesting that intrinsic dynamics is an important determinant of function.
The binding sites of InaD PDZ1 and Dvl2 PDZ are the most flexible of those of the five PDZ domains and this high degree of flexibility is likely to be necessary for them to be able to interact with multiple partners using significantly different binding modes, a property referred to as “strong promiscuity”. The Erbin PDZ domain, by contrast, has a rigid binding site and while it is also promiscuous, it interacts with very similar peptides using very similar binding modes. We do not count interactions with proteins at distal sites such as that reported for the Erbin-Smad3 MH2 interaction [48] which appears to be well away from the classical PDZ interaction groove. The results presented here are consistent with the proposed link between binding site flexibility and promiscuity discussed in other studies [14], [16].
Currently there is no experimental structure available of the complex of InaD PDZ1 with the NinaC peptide. Based on the results presented in this study, we predict that InaD PDZ1 interacts with NinaC in a significantly different binding mode than it does with NorpA, a conclusion also made by Kimple et al [44] and Wes et al [43]. This hypothesis should be readily testable via structural characterization experiments.
The results for PTP-BL PDZ2 have revealed that the conformational space explored by the apo protein is the most different from the APC peptide-bound conformation compared to the other PDZ-peptide complexes. These results, in accordance with experimental data, suggest that the induced fit mechanism may be crucially involved in the binding of PTP-BL PDZ2 to the APC peptide and play a larger role in the recognition mechanism compared to other PDZ domains. Overall it seems likely that conformational selection and induced fit both appear to play roles in binding of PDZ domains to their peptides. One can formulate the two mechanisms into distinct roles; Firstly, conformational selection seems to be an essential mechanism for PDZ domains to visit regions of the conformational space that are close to different ligand-bound states. Visiting these regions is probably necessary for the formation of weak (initial) complexes. Once a weak complex is formed, the induced fit mechanism, as a fine-tuning step, could lead to minor changes in the shape of the binding pocket stabilizing the PDZ-peptide complex. The extent to which these mechanisms are required is likely variable across the PDZ domain family.
The MD simulations confirm that GRIP1 PDZ7 has a closed canonical binding site which is consequently unable to accommodate carboxyl peptides. The binding pocket does not appear to undergo a transition from its closed state to an open state in the course of the 200 ns trajectory. These results agree with the experimental observations that GRIP1 PDZ7 cannot interact with carboxyl ligands.
The results highlight how one fold can exhibit quite different dynamics. For PDZ domains this issue should be borne in mind when considering structure-based drug-design [49]. Considering conformational selection in the docking strategies of virtual screening is a promising new paradigm recently reviewed by Amaro and Li [50]. Furthermore, describing binding site flexibility was suggested to be crucial for designing compounds of high selectivity for a given drug target [51]. As the dynamics of the PDZ binding pocket seems to be a key factor determining the ability to interact with different peptides, the flexibility of the binding site should also be taken into account alongside steric and electrostatic effects [52] in rational drug design. From this work we would anticipate that intrinsic dynamics would play a role in other systems ranging from influencing large domain movements through to allosteric transitions. As simulation times approach experimental timescale, particularly for NMR, it will become possible to assess how well these observations fit into solvable models for conformational selection and induced fit such as the one proposed by Zhou [53].
All-atom 200 ns MD simulations were performed for the five apo PDZ domains summarized in Table 1 with the GROMACS software package [54], [55] using the OPLS force field [56] in an NPT ensemble. Pressure coupling was performed using the Berendsen barostat with a time constant, tau, of 1.0 ps. The systems were coupled using a Berendsen heat bath [57] with a tau value of 0.1 ps. Electrostatics were treated with a Particle Mesh Ewald scheme with a real-space cut-off of 10 Å. The neighbour list cut-off was also set to 10 Å and was updated every 10 steps. The proteins were solvated in explicit SPC water [58] and Na+ and Cl− ions added to make up a neutral solution of 150 mM. A short steepest descents minimization of 225000 steps was performed, followed by a short restrained run of 200 ps whereby the Cα atoms of the protein were restrained by a harmonic potential with a force constant of 1000 kJmol−1 mn−2. Snapshots from the trajectories were saved every 5 ps for analysis. Convergence was assessed via root mean square inner product (RMSIP) between sections of the trajectories (see Supporting Information, Text S1, for more details).
Let A and B denote two proteins that consist of NA and NB residues, respectively. In this study, residues are represented by their α-carbon atoms. An alignment between the two structures defines a mapping between the two sets of residues. Let N denote the number of aligned residue pairs (after removing positions aligned to gaps). The two sets of aligned residues are described by the NxN distance matrices of their α-carbon atoms denoted by dA and dB: i.e. the matrix entry is the distance of α-carbon atoms of aligned residues i and j in structure A.
The difference distance matrix δ between structure A and B is defined as:(1)
Positive entries in this matrix indicate pairs of atoms of larger distance in structure A than in structure B. This matrix can be used to characterize the location and extent of structural differences between two different proteins or two conformations of the same protein.
The dRMSD (distance root mean square deviation) measure of dissimilarity between the two structures is defined as:(2)
We use this measure instead of the standard RMSD dissimilarity because dRMSD is not dependent on structural superposition.
Let S = {S1, S2, …, SK} denote an ensemble of conformations of a protein represented by its α-carbon atoms. Let the number of its residues be N.
We define an NxN matrix as the F fluctuation matrix, which describes the extent of the pairwise fluctuation of α-carbon atoms. Matrix F contains the variances of the distance of each α-carbon pair, where the variance is calculated over the whole ensemble. It is precisely defined as: (3)where is the mean distance of α-carbon atoms i and j in the ensemble. We have previously described the use of a similar matrix where standard deviation rather than variance of the distances was used [59].
Although variance describes the spread of a distance distribution characterizing the relative fluctuation of two atoms, it is not always informative about how much the distance between two atoms can change. Even if the distance of two atoms significantly deviates from their mean distance in some conformations, the variance may still be low provided that most of the variation is around the mean. To measure the pairwise flexibility of two atoms (i.e. the maximal difference of their distance in the ensemble), the flexibility matrix denoted as X is introduced. Matrix X describes the range of distance distribution for each pair of atoms:(4)
Note that the above definitions of F and X matrices allow that two pairs of atoms that have equal pairwise fluctuation can have considerably different pairwise flexibility.
While the F matrix contains pairwise atomic fluctuation values, a measure of the overall fluctuation of the whole structure (or a subset of residues) was also introduced. This overall fluctuation measure denoted by Θ was defined as the root mean square of dRMSD dissimilarity of each structure with regards the mean distance matrix calculated for the whole S ensemble. In other words, Θ is a measure for the size of conformational space the protein explores in the ensemble. It is easy to see that the above definition is equivalent to the root mean of the entries of F fluctuation matrix calculated for the same conformational ensemble. The precise definition of overall fluctuation is therefore(5)where is the mean distance matrix of the ensemble.
Equivalent binding site residues were defined on the basis of a multiple sequence alignment (MSA) (Figure 1B). The binding groove of PDZ domains is located between the β2 strand and the α2 helix. Two sequence regions were therefore selected in the MSA that correspond to the conserved structural elements of the β2 strand and α2 helix (or α1 helix, in Erbin PDZ). The binding sites were characterized by 5×10 submatrices of the δ, F and X matrices describing the relative structural difference, fluctuation and flexibility of the α-helix and the β-strand.
MD simulation trajectory snapshots were clustered with k-mean cluster analysis, a simple unsupervised learning algorithm [60], [61]. The method can be used for partitioning N data points (here, protein conformations) into k disjoint subsets (or clusters) denoted by C1 , C2 , …, Ck . The parameter k is fixed a priori. The goal of the algorithm is to find the optimal partitioning of conformations to minimize the within-cluster sum of squares (WCSS):(6)where the dRMSD measure is used to capture the similarity of conformations and Ci is the mean distance matrix of cluster i. Since k is an arbitrary parameter, the goodness of clustering results was estimated using the Silhouette Index cluster validity measure (see below) [62]. The optimal k-value that provided the highest overall average Silhouette Index was selected.
Once the conformational ensemble is clustered, the following Silhouette Index measure is calculated for each conformation:(7)where a(i) is the average dRMSD dissimilarity of conformation i to all other conformations in the same cluster and b(i) is the minimum of average dRMSD dissimilarities of conformation i to all other clusters. The silhouette index is between −1 and 1: if S(i) it is close to 1, it means, the conformation is well-clustered; if S(i) is close to 0, it means the conformation could be assigned to another cluster as well; if S(i) is close to −1, it means the conformation is misclassified. The goodness of clustering result was then measured by the overall average silhouette index SOVER which is simply the average of S(i) for all conformations in the ensemble:(8)
Multidimensional scaling (MDS) (also known as Principal Coordinates Analysis) is a dimensionality reduction method often used to visualize high-dimensional data on a two-dimensional map [63]. The input of the method is a dissimilarity matrix that contains distances (dissimilarities) between pairs of objects calculated in a high-dimensional space. The output is a configuration of points embedded into lower (ideally, two or three)-dimensions. In Classical MDS (CMDS) (also referred to as Torgerson-Gower scaling) [64] used in this study, the goal is that the Euclidean distances between the outputted points should approximately reproduce the original dissimilarity matrix.
In order to study the difference between induced fit and conformational selection binding, a simple definition is introduced to measure how similar conformations are sampled in an apo simulation to a given experimental ligand-bound structure. Let S(k) denote the set of k most similar conformations (neighboring conformers) with regards to a reference experimental structure E (ranked based on the dRMSD dissimilarity measure). The following Q(k) value is defined as the average dRMSD dissimilarity of conformations in S(k) with regards to structure E:(9)
In this study the quantities Q(1), Q(10), Q(100) and Q(200) were used to characterize the similarity of the most similar, 10 most similar, 100 most similar and 200 most similar conformations to an experimental ligand-bound structure of interest.
|
10.1371/journal.pntd.0001874 | Cytokine Responses to Novel Antigens in an Indian Population Living in an Area Endemic for Visceral Leishmaniasis | There are no effective vaccines for visceral leishmaniasis (VL), a neglected parasitic disease second only to malaria in global mortality. We previously identified 14 protective candidates in a screen of 100 Leishmania antigens as DNA vaccines in mice. Here we employ whole blood assays to evaluate human cytokine responses to 11 of these antigens, in comparison to known defined and crude antigen preparations.
Whole blood assays were employed to measure IFN-γ, TNF-α and IL-10 responses to peptide pools of the novel antigens R71, Q51, L37, N52, L302.06, J89, M18, J41, M22, M63, M57, as well as to recombinant proteins of tryparedoxin peroxidase (TRYP), Leishmania homolog of the receptor for activated C kinase (LACK) and to crude soluble Leishmania antigen (SLA), in Indian patients with active (n = 8) or cured (n = 16) VL, and in modified Quantiferon positive (EHC+ve, n = 20) or modified Quantiferon negative (EHC−ve, n = 9) endemic healthy controls (EHC).
Active VL, cured VL and EHC+ve groups showed elevated SLA-specific IFN-γ, but only active VL patients produced IL-10 and EHC+ve did not make TNF-α. IFN-γ to IL-10 and TNF-α to IL-10 ratios in response to TRYP and LACK antigens were higher in cured VL and EHC+ve exposed individuals compared to active VL. Five of the eleven novel candidates (R71, L37, N52, J41, and M22) elicited IFN-γ and TNF-α, but not IL-10, responses in cured VL (55–87.5% responders) and EHC+ve (40–65% responders) subjects.
Our results are consistent with an important balance between pro-inflammatory IFNγ and TNFγ cytokine responses and anti-inflammatory IL-10 in determining outcome of VL in India, as highlighted by response to both crude and defined protein antigens. Importantly, cured VL patients and endemic Quantiferon positive individuals recognise 5 novel vaccine candidate antigens, confirming our recent data for L. chagasi in Brazil, and their potential as cross-species vaccine candidates.
| Visceral leishmaniasis is a parasitic infection that results in death in susceptible people unless they are treated. Current drugs are expensive and toxic, and there are no vaccines in use in humans. We know that it is possible to become immune to infection with this parasite because people who have been cured using drug treatment are resistant to further infection. In addition, a large percentage of people infected with the parasite remain asymptomatic and develop a specific immune response that can be measured using crude leishmanial antigens. We hypothesized that these resistant people might hold the key to understanding the kind of immune response required for protection. In this paper we compared the immune response to a series of novel vaccine candidates in people with active disease, in those drug-cured from the disease, and in the naturally resistant individuals. We show that immune individuals make strong cytokine responses to five of eleven novel vaccine candidates that were tested, making them ideal candidates to take forward in the development of a defined vaccine against leishmaniasis.
| Visceral leishmaniasis (VL), also known as kala-azar, is a potentially fatal disease caused by obligate intracellular parasites of the Leishmania donovani species complex. VL is a serious public health problem in indigenous and rural populations in India, accounting for enormous morbidity and mortality, as well as major costs to both local and national health budgets. The estimated annual global incidence of VL is 200,000 to 400,000, and >90% of these cases occur in India, Bangladesh, Sudan, South Sudan, Ethiopia and Brazil [1]. Interestingly, 80 to 90% of human infections are subclinical or asymptomatic, and this asymptomatic infection is associated with strong cell-mediated immunity [2], [3], [4], [5], [6]. Only a small percentage of infected individuals develop severe disease [7], [8], and patients who recover from VL display resistance to reinfection [5]. This suggests the development of protective immunity and provides a rational basis for the development of vaccines that impart potent cell-mediated immune responses. Furthermore, the factors that skew the immune response toward T helper 1 (Th1) or Th2/T regulatory (Treg) cell dominance are partially understood, and it is believed that direct interaction between parasite antigens and host immune cells participate to shape the subsequent pathogenic or protective immune responses [9], [10].
A key mechanism by which T cells mediate their effector functions is through the production of cytokines. However, heterogeneity of CD4+ T-cell cytokine responses has made it difficult to define immune correlates of protection after vaccination in leishmaniasis. In murine cutaneous leishmaniasis, the degree of protection in vaccinated mice was predicted by the frequency of CD4+ T cells simultaneously producing interferon-γ (IFN-γ), interleukin (IL)-2 and tumour necrosis factor (TNF, formerly TNF-α) [11]. These multi-functional effector CD4+ T cells elicited by all vaccines tested were unique in producing high amounts of IFN-γ [11]. In our own studies comparing vaccines with different efficacies in mice, we found that the balance between antigen-specific CD4 T cell-derived pro-inflammatory IFN-γ and regulatory IL-10 (and to a lesser extent IL-4 and IL-5), rather than magnitude of IFN-γ per se, provided the best correlate of a protective immune response [12]. A strong tumour necrosis factor-α (TNF-α) response concurrent with IFN-γ has also been shown to be important in models of VL [13]. A crucial step in vaccine development against human disease requires improved understanding of the functional heterogeneity of T-cell cytokine responses generated by candidate vaccine antigens. For example, one study in malaria reported that peptide-specific IFN-γ to a conserved epitope of the circumsporozoite surface protein was strongly associated with protection of humans again infection and disease [14], providing a precise target for vaccine design. Without a convincing single marker of protective immunity against leishmaniasis, vaccine development has to rely on screening a range of cytokines to gauge the balance between Th1 and Th2/T regulatory (Treg) responses.
Advances in our understanding of Leishmania pathogenesis and of the generation of host protective immunity, together with completed Leishmania genome sequences, have opened new avenues for vaccine research. Although significant progress has been made to understand mechanisms of VL immunity in humans [9], [10], there is no effective vaccine available for humans against any form of leishmaniasis. Drugs used in leishmaniasis therapy are significantly toxic, expensive and faced with increasing resistance. Limitations in pharmacotherapy argue for the development of a vaccine for VL. Vaccination with live virulent parasites, termed leishmanization, was practiced from ancient times until recently in many endemic areas [15]. Vaccine trials involving whole, killed parasites were conducted in the 1970s and 1980s [16], [17]. Although no overall statistically significant protection has been associated with any trial of these killed vaccines [18], [19], [20], [21], [22], a common theme has been protection in persons who showed conversion of Leishmania-specific delayed type hypersensitivity (DTH) skin test responses during the trial, whether or not they received the vaccine. The latter points to the importance of understanding the immune response in exposed individuals who become infected but do not progress to clinical disease, which in the Indian endemic area has been equated to a positive modified Quantiferon response to leishmanial antigens in whole blood assays [23].
The genome sequence and proteome data (∼33.6 Mb genome and ∼8300 protein coding genes) of Leishmania major [24] provides a rich source of potential vaccine candidates. We recently described the identification of novel Leishmania antigens delivered as DNA vaccines to susceptible BALB/c mice, and identified 14 protective candidate antigens in a screen of 100 amastigote-expressed genes [25]. To determine their potential as vaccine candidates for humans, we here evaluate the ability of 11 of these novel Leishmania vaccine candidates, along with soluble Leishmania antigen (SLA), recombinant Leishmania homolog of the receptor for activated C kinase (LACK), and tryparedoxin peroxidase (TRYP) proteins, to stimulate cytokine responses in whole blood from active and cured VL patients, and from modified Quantiferon positive and negative endemic health controls (EHC), in India.
The study was approved by the Ethics Committee of the Banaras Hindu University, Varanasi, India. Written informed consent was obtained from all adult subjects included in the study, or from the parents or guardians of individuals less than 18 years of age. Subjects belonged to 4 clinically well characterized groups: (i) active VL: cases of parasitologically confirmed, active VL (n = 8); (ii) cured VL: subjects who were definitively cured of VL and shown to have no parasites in splenic aspirates at least 6 months after treatment (n = 16); (iii) EHC with a positive antigen-specific IFN-γ response measured by modified Quantiferon (Cellestis, Chadstone, Australia) assay (cf. below) (EHC+ve, n = 20); and (iv) EHC testing negative by modified Quantiferon assay (EHC−ve, n = 9). Subjects having fever within the past month, and children less than five years of age, were excluded. Follow up visits were made to the homes of the EHC+ve and cured subjects 6 and 12 months after enrolment to monitor for the development of active VL. Demographic and clinical characteristics of participants enrolled in the vaccine study are summarized in Table 1. None of the cured VL or EHC subjects developed clinical VL during the 1 year follow-up.
SLA from an Ethiopian strain of L. donovani (LV9) or L. major (LV39) were prepared at the Cambridge Institute for Medical Research, University of Cambridge School of Clinical Medicine, UK as described previously [12]. SLA from an Indian strain of L. donovani was prepared at the Infectious Disease Research Laboratory, Banaras Hindu University, according to the published protocol of Scott and co-workers [26]. The protein concentration was estimated using the BCA method [27]. SLA was stored at −80°C until use.
Recombinant LACK and TRYP proteins were prepared as described [11], [12], with large-scale preparation, endotoxin removal and protein estimation out-sourced to Novexin Ltd. (Cambridge, UK). As previously described [28], overlapping 13–20-mer peptides (minimal overlap of 12 amino acids to ensure complete coverage of epitopes, 7–31 peptides/antigen depending on amino acid length of the protein) representing 11 (R71, Q51, L37, N52, L302.06, J89, M18, J41, M22, M63, M57) of 14 novel antigens identified [25] were synthesized commercially (Peptide2.0, Chantilly, VA), initially solubilized in dimethylsulfoxide (final concentration in the well <0.1% DMSO), and pooled (for peptides within each antigen) in endotoxin-free phosphate-buffered saline at a final concentration of 50 µg/mL per individual peptide. Peptides pools were stored at −80°C. Due to the size of peptides in the pools, some natural processing is required for epitope selection prior to epitope binding to class II and presentation to CD4+ T cells.
The Quantiferon (Cellestis, Chadstone, Australia) whole blood assay was conducted on 147 endemic healthy individuals according to the manufacturer's instructions or to our published modifications [6], [23]. From these data, 20 highly positive (above the cut-off value generated by the ROC curve) individuals were selected as the EHC+ve study group, and 9 individuals selected at random from below the cut-off value as the EHC−ve study group. Blood (5 mL) was collected into heparinised tubes, and samples diluted 1 in 8 in serum-free complete medium comprising RPMI supplemented with 2 mM L-glutamine, 100 µg/mL streptomycin, 100 IU/mL penicillin (Gibco, USA). Diluted blood (180 µL/well) was plated into 96-well U-bottomed plates (Nunc, Rochester, USA) and antigen added in triplicate wells at a final concentration of 10 µg/mL for TRYP, LACK and SLA, 5 µg/mL for PPD, PHA and the 11 novel antigen peptide pools, and made up to a volume of 200 µL. Plates were incubated at 37°C in 5% CO2 for 24 hours, 72 hours or 6 days. Supernatants from replicate wells were harvested, pooled and stored at −80°C until analysed by ELISA.
Cytokine release (IFN-γ, TNF-α and IL-10) by antigen stimulated whole blood cells was measured at 24 hours, 72 hours, or 6 days. IFN-γ, TNF-α and IL-10 were measured using matched antibody pairs (BD Pharmingen, Franklin Lakes, NJ, USA) by ELISA. The limit of detection for these ELISAs was 31 pg/mL. Background levels in non-stimulated control wells were deducted from antigen-stimulated values to determine antigen specific cytokine responses (with negative values recorded as zero). To control for inter-plate and intra-plate variation, a positive-control supernatant (1∶4 and 1∶8 dilution of PHA stimulated Non Endemic Healthy Controls (NEHC) whole blood pooled supernatant) was used in duplicate on each ELISA plate. The mean variability of these duplicate measurements was 2.53% (intra-plate variation). The coefficient of variation between plates (inter-plate variation) was 20.74% for IFN-γ, 10.78% for TNF-α and 18.04% for IL-10.
Because data were generally not normally distributed (as determined using the Kolgomorov-Smirnov test), data are plotted using box and whiskers (Tukey) plots, and statistical differences (P<0.05) between pairs of groups were determined using nonparametric 2-tailed Mann-Whitney tests. Nominal P-values are presented throughout (i.e. without correction for multiple testing). Plots were generated using GraphPad Prism 5 (San Diago, USA), and statistical analyses were performed using GraphPad Prism 5 or SPSS software v18.0.
To investigate antigen specific production of IFN-γ, TNF-α and IL-10 cytokines, diluted whole blood from different patient groups was initially stimulated with SLA from an Indian L. donovani strain. Comparison of responses over time post stimulation in active VL cases showed that all 3 cytokines were highest, and less variable, at the 24 hour time point (Fig. 1A–C). Active cases made variable responses to PPD reflecting prior exposure to Mycobacterium and indicating that ability to make a response to mycobacterial antigens is not compromised in active VL patients. Cured cases made similarly variable responses to PPD (Fig. S1). Between group comparisons at 24 hours post stimulation showed that active VL, cured VL, and EHC+ve study groups all made higher IFN-γ responses relative to EHC−ve subjects (Fig. 1D). The observation that active VL cases make a significant amount of IFN-γ is in line with our recent observations for whole blood assays using undiluted blood in a modified Quantiferon assay [6], [23]. Of interest, while cured VL and active VL groups generated TNF-α concomitant with IFN-γ, the EHC+ve group did not (Fig. 1E), suggesting that production of this cytokine might relate to the pathogenic role of TNF-α in VL disease [29]. Note, however, that this was not true for responses to putative vaccine candidates outlined below, which all elicited TNF-α concomitant with IFN-γ in the EHC+ve group. Importantly, only the active VL group made IL-10 in response to Indian L. donovani SLA (Fig. 1F), supporting previous data indicating that IL-10 is a key regulatory cytokine in VL patients [10], [30].
Cytokine responses to 3 different Leishmania strains, an Indian L. donovani strain (designated SLA), an Ethiopian L. donovani strain (LV9) and L. major strain (LV39) were compared (Fig. 2). For IFN-γ responses the two L. donovani preparations stimulated equivalent responses (Fig. 2A). Interestingly, SLA prepared from the local Indian L. donovani strain elicited significantly stronger 24 h TNF-α (Fig. 2B) and IL-10 responses (Fig. 2C) compared to Ethiopian L. donovani (p = 0.0003; p = 0.004) or the L. major strain (p = 0.028; p = 0.007). The L. major antigen was more variable in eliciting responses across all cytokines and time points.
The ability of diluted whole blood assay samples to respond to mitogenic stimulation with PHA (Tables S1 and S2) confirmed the viability of the cells from all donors.
We previously demonstrated that high CD4-derived IFN-γ to low IL-10 ratios predicted vaccine success in mice when comparing TRYP and LACK as DNA with/without Modified Vaccinia Ankara vaccines [12], [31]. Here, we examined immune responses to these potential vaccine antigens in clinically well characterized groups of human subjects. The full set of results for IFN-γ, TNF-α and IL-10 responses to TRYP (Fig. S2) and LACK (Fig. S3) at 24 hours, 72 hours and 6 days post stimulation in active VL, cured VL, EHC+ve and EHC−ve study groups is provided in the figures S2 and S3. Of note, although the EHC−ve group comprised negative responders to Indian SLA by the modified Quantiferon assay, their cytokine responses to TRYP and LACK was rarely significantly different as a group from cured VL and EHC+ve groups. Hence, we conclude that there are exposed individuals amongst this EHC group, although we did not test non-endemic healthy control responses to these antigens. As this exposure status is variable and equivocal, we exclude them from further analysis of between group responses.
Results for active VL, cured VL and EHC+ve groups are summarised in figures 3 (TRYP) and 4 (LACK). For both antigens, the pattern of IFN-γ and TNF-α responses across the 3 groups is generally established at 24 hours, and clear cut by 72 hours and 6 days, post stimulation. For these two cytokines, responses were significantly lower in active VL compared to cured VL and EHC+ve groups at 72 hours and 6 days post stimulation. The pattern of responses for IL-10 was similar (i.e. higher in cured VL and EHC+ve compared to active VL), but more clearly apparent at 24 hours post-stimulation. This led to interesting between group differences in the ratios of IFN-γ to IL-10 and TNF-α to IL-10 at 24 hours, when ratios were significantly higher in the active VL group compared to cured VL and EHC+ve groups (particularly for TRYP), compared to 6 days of stimulation where the reverse was true for both antigens. For both antigens, the ratios of IFN-γ to IL-10 and TNF-α to IL-10 were highest in the EHC+ve group, suggesting that a potent pro-inflammatory response relative to modest levels of IL-10 may correlate with protection from disease in this confirmed Quantiferon positive exposed EHC group.
We previously identified 14 protective Leishmania antigens in a screen of 100 candidates delivered as DNA vaccines to susceptible BALB/c mice [25]. We measured IFN-γ and TNF-α as effector pro-inflammatory cytokine responses to peptide pools for each of 11 of these antigens in diluted whole blood assays, and IL-10 as a measure of their ability to elicit a regulatory cytokine response. A full summary of responder status on a categorical scale (− = <20 pg/ml; + = 20–99 pg/ml; ++ = 100–249; +++ = 250–499 pg/ml; ++++ = 500–10000 pg/ml; ++++ = >10000 pg/ml) to each of the 11 antigens, and to control SLA, PPD and PHA stimulations, is provided for all individuals in Tables S1 and S2. As these antigens were based on L. major sequence data, the antigens are presented throughout in order of their percent identity to L. infantum, as reported by us previously [28]. As would be predicted on the basis of genetic heterogeneity in HLA-restricted T cell responses and other background genetic and environmental factors, not all individuals make a response to individual candidate vaccine antigens. Using cured VL patients as an initial evaluation of percent responders (≥20 ng/mL above background) with time post stimulation, we observed maximal IFN-γ responders at 24 hours and 72 hours post stimulation (Fig. 5A), with 55–87.5% responders to 5 of the novel antigens (R71, L37, N52, J41 and M22; of these L37 exceptional in eliciting the highest sustained IFN-γ responses at day 6 post stimulation, see also Table S2). Comparing across groups for the 72 hour time point (Fig. 5B), we observed 40–65% responders to these 5 novel antigens in the EHC+ve group, with ≥25% of active VL cases also making IFN-γ responses to these antigens. Looking across cytokine responses for these 5 antigens (Fig. 6), we observe a similar profile of TNF-α responses in cured VL and EHC+ve groups as we observed for IFN-γ, but no IL-10. Even amongst active VL cases, N52 was the only antigen to elicit IL-10 responses (Fig. 6L). As for TRYP and LACK, a small number (22–33%) of responders was observed in the EHC−ve group, consistent with evidence of exposure in these individuals despite their negative response in the modified Quantiferon assay. Alternatively, these might represent non-specific responses to these antigens as we did not include non-endemic controls in our study. In summary, we have identified five Leishmania antigens from 11 putative vaccine candidates tested that stimulate potent pro-inflammatory recall responses in exposed but protected individuals (cured VL patients and EHC+ve) in the absence of regulatory IL-10, providing potential immunotherapeutic or vaccine targets for future investigation.
A variety of defined antigens have been investigated as vaccine antigen candidates for VL in animal models [32], [33], [34], but few have advanced to human clinical trials [35], [36]. One limitation in the search for an effective vaccine for leishmaniasis is the lack of information on immunological correlates of natural and vaccine-mediated protection in humans. In recent studies we have highlighted the use of a modified Quantiferon assay to screen for naturally exposed resistant individuals in the Indian study area [6]. That assay relies on 3 mL of undiluted whole blood. Here we show that individuals positive by the modified Quantiferon assay are also positive in our 96-well plate assays using diluted whole blood, providing the means to more efficient screening in large-scale epidemiological studies as has been used previously in studies of mycobacterial diseases [37], [38], [39]. Importantly too, our 96-well plate assay also showed that active VL patients were positive for IFN-γ in these diluted whole blood 96-well plate assays. Our initial demonstration [23] that active VL patients are positive for IFN-γ in the modified Quantiferon assay was remarkable given the numerous previous studies that had failed to observe cellular proliferation or IFN-γ release after stimulation of peripheral blood mononuclear cells from active VL patients with crude Leishmania antigen [10], [40], [41], [42]. Ability to measure this IFN-γ response in the diluted whole blood assay described here will also facilitate more efficient screening of active VL cases using smaller blood volumes in a 96-well plate format.
In human and murine cells infected in vitro, and in mice in vivo, clearance of Leishmania parasites requires IFN-γ. However, IFN-γ alone does not predict vaccine-mediated protection in mice [12], [31], [43], [44]. Rather, the simultaneous production of IFN-γ, IL-2 and TNF-α by a particular subset of CD4 T cells [11], and/or the balance between pro-inflammatory IFN-γ/TNF-α and regulatory IL-10 [12], [31], [44], [45], have been variously shown to be predictive of vaccine outcome. Epidemiological studies indicate that patients drug-cured from L. donovani infection are protected against subsequent clinical disease [46], and it is thought that exposed individuals who test as positive to crude leishmanial antigens in the modified Quantiferon assay employed in our study area in India are infected asymptomatic individuals who are resistant to developing active VL disease [6]. Therefore, in the analysis of human immune responses to known and novel antigens presented here, we hypothesized that ability to stimulate IFN-γ, TNF-α and IL-10 in cured VL and EHC+ve individuals, compared to active VL cases, would provide some insight into their potential as vaccine candidates.
Our investigations focused initially on the known vaccine candidates TRYP and LACK. Although others have found LACK protective in murine models of cutaneous leishmaniasis [47], in the virulent model of visceralising L. major LV39 infection in mice we found that TRYP was protective but LACK was not [12]. Although the vaccine-induced IFN-γ responses were similar between the two antigens in mice, lower IL-10 was elicited by TRYP than LACK, resulting in higher IFN-γ to IL-10 ratios as correlates of protective immunity. In the human studies described here, we found that TRYP and LACK were equivalent to each other in the magnitudes of IFN-γ, TNF-α and IL-10 responses elicited, and in generating higher IFN-γ to IL-10 and TNF-α to IL-10 ratios in putatively protected cured VL and EHC+ve individuals than in active VL cases. It was of interest that in India, the asymptomatic EHC+ve group had equivalent responses to the cured VL group, whereas in our recent study [28] of the same antigens (and antigen preparations) in Brazil, we found that asymptomatic DTH+ve individuals had lower ratios of IFN-γ to IL-10 and TNF-α to IL-10 compared to cured VL patients. This was due to higher IL-10 responses in the DTH+ve group compared to the cured VL group, leading us to suggest that a measure of modulation of the pro-inflammatory response by IL-10 in the DTH+ve group might contribute to the protective response. In active VL disease, high levels of TNF-α contribute to fever and cachexia, and are detrimental [29], and it is not yet known what role in pathogenesis is played by the strong 24 hour IFN-γ responses observed in whole blood assays in active VL [6], [23].
In our analysis of novel vaccine candidates, we found that 5 antigens (R71, L37, N52, J41 and M22) elicited IFN-γ and TNF-α responses in a high percentage of cured VL (55–87.5%) and EHC+ve (40–65%) subjects. This represents remarkable replication of recent findings from an area endemic for L. infantum chagasi in northern Brazil, where 4 of these antigens (R71, L37, N52 and M22; same preparations of peptide pools) also elicited strong IFN-γ and TNF-α responses in both cured VL and exposed asympotmatic DTH+ individuals [28]. In Brazil, responses to J41 were only observed in the cured VL group, but the sample size for DTH+ individuals was small (n = 4). Strong responses were also observed in Brazil to two additional antigens, L302.06 and M18, for which a lower percentage (<30%) of responders were observed in India. This may reflect small samples sizes, differences in amino acid sequences of the parasites, and/or differences in HLA alleles between the two populations. On balance, all of these antigens remain strong candidates in the context of a multivalent cross-species vaccine against leishmaniasis. R71 and L37 are ribosomal proteins with high (100 and 99%, respectively) percentage identity at the amino acid level between L. major and L. infantum [28]. N52 is a V-ATPase subunit F which also has high (94%) identity across the two species. J41 and M22 are hypothetical proteins of unknown function which, despite lower percent identities (73% and 61%, respectively) between L. major and L. infantum, appear to provide cross-reactive epitopes that are recognised in both Brazil [28] and India. An important contrast between the two endemic regions was the almost complete lack of IL-10 responses to these novel antigens (same preparations of peptide pools) in the Indian study in both cured VL and EHC+ve groups, whereas in the Brazilian study cured VL subjects who were positive for IFN-γ and TNF-α responses also produced IL-10. N52 was also unique in being the only antigen to stimulate IL-10 responses in active VL patients, suggesting that responses to this antigen might provide an important early diagnostic biomarker for disease-associated IL-10 in VL. Further studies are needed to evaluate more carefully the differences in cytokine responses to individual antigens in active compared to cured VL groups, as well as between cured VL and exposed asymptomatic DTH+ or modified Quantiferon positive groups. Unlike Brazil [4], [5], DTH responses have not provided a sensitive means of evaluating cell mediated immune response in cured VL or exposed individuals in India [48], pointing to potential differences in cell-mediated responses between DTH+ compared to Quantiferon positive exposed asymptomatic individuals that might hold the key to uncovering the true correlates of vaccine-induced immunity in leishmaniasis.
Results of our study demonstrate that only a percentage of individuals respond to vaccine antigens that have individually been shown to be protective in mice. This suggests that defined vaccine for use in humans will need to be complex multi-epitope/antigens vaccines. To date, only one multicomponent vaccine, Leish-111f, has been assessed in a large clinical trial [49]. Our recent small-scale clinical trial in a L. donovani endemic area showed Leish-F1-MPL-SE was safe and well tolerated in people with and without prior VL exposure and induced strong antigen-specific T cell responses [36]. The data presented here, and in our earlier study from Brazil [28], provide evidence to support a number of novel candidates that could be taken forward as vaccines against human leishmaniasis.
|
10.1371/journal.pcbi.1003315 | Prediction of Drug-Target Interactions for Drug Repositioning Only Based on Genomic Expression Similarity | Small drug molecules usually bind to multiple protein targets or even unintended off-targets. Such drug promiscuity has often led to unwanted or unexplained drug reactions, resulting in side effects or drug repositioning opportunities. So it is always an important issue in pharmacology to identify potential drug-target interactions (DTI). However, DTI discovery by experiment remains a challenging task, due to high expense of time and resources. Many computational methods are therefore developed to predict DTI with high throughput biological and clinical data. Here, we initiatively demonstrate that the on-target and off-target effects could be characterized by drug-induced in vitro genomic expression changes, e.g. the data in Connectivity Map (CMap). Thus, unknown ligands of a certain target can be found from the compounds showing high gene-expression similarity to the known ligands. Then to clarify the detailed practice of CMap based DTI prediction, we objectively evaluate how well each target is characterized by CMap. The results suggest that (1) some targets are better characterized than others, so the prediction models specific to these well characterized targets would be more accurate and reliable; (2) in some cases, a family of ligands for the same target tend to interact with common off-targets, which may help increase the efficiency of DTI discovery and explain the mechanisms of complicated drug actions. In the present study, CMap expression similarity is proposed as a novel indicator of drug-target interactions. The detailed strategies of improving data quality by decreasing the batch effect and building prediction models are also effectively established. We believe the success in CMap can be further translated into other public and commercial data of genomic expression, thus increasing research productivity towards valid drug repositioning and minimal side effects.
| Small drug molecules usually bind to unintended off-targets, leading to unexpected drug responses such as side effects or drug repositioning opportunities. Thus, identifying unintended drug-target interactions (DTI) is particularly required for understanding complicated drug actions. It remains expensive nowadays to experimentally determine DTI, so various computational methods are developed. In this study, we initiatively demonstrated that target binding is directly correlated with drug induced genomic expression profiles in Connectivity Map (CMap). By improving data quality of CMap, we illustrated three important facts: (1) Drugs binding to common targets show higher gene-expression similarity than random compounds, indicating that upstream ligand binding could be characterized by downstream gene-expression change. (2) It is found that some targets are better characterized by CMap than others. To guarantee efficiency of DTI discovery, prediction models should be specifically built for those well characterized targets. (3) It is broadly observed in the predicted DTI that ligands for the same target may collectively interact with common off-target. This observation is consistent with published experimental evidence and can help illustrate the mechanisms of unexplained drug reactions. Based on CMap, our work established an efficient pipeline of identifying potential DTI. By extending the success in CMap to other genomic data sources, we believe more DTI would be discovered.
| Drug promiscuity refers to the phenomenon that small molecule drug binds to multiple protein targets. In recent years, drug promiscuity has gained broad attention [1]–[3], because unintended drugs-target interactions (DTI) are often associated with drug repositioning [4] and side effects [5]–[8]. Although biotechnology evolves and new biochemical assays arise [9], [10], it remains time-consuming and expensive nowadays to experimentally discover unknown DTI, especially when multiple compounds and proteins are simultaneously involved. This situation therefore provides a strong incentive to develop new computational methods, which could screen potential DTI with high throughput and low cost.
By binding to targets with complementary structures, drug molecules profoundly modify the behavior of downstream genes and lead to specific reactions. Along this route of drug action, various biological informations could be correlated to target binding and be analyzed with computational models. For example, methods have been established to predict DTI by ligand/protein structures [11]–[16] and clinical side effects [17]. On the other hand, although there are researches addressing drug-induced target expression [18], it has been rarely studied that drug-induced downstream gene-expression changes may directly indicate target promiscuity, thus missing a possible technique of DTI discovery. Here we suppose that drugs binding to specific target are generally prone to influencing the target-related downstream genes [19], [20], so the pattern of gene-expression change could reflect the characteristics of target binding (Figure 1A). One of the most reliable and comprehensive sources of drug-induced genomic expression data is the Connectivity Map (CMap), which includes 6100 human cell cultures (i.e. 6100 CMap ‘instances’) treated by 1309 bioactive compounds [21]. We initiatively found that drugs interacting with the same target generally lead to similar gene-expression profiles in CMap. This observation enlightened us to apply CMap expression similarity as a guilt-by-association metric, that high similarity between different drugs may imply interactions to the same target (Figure 1B)
However, one of the major impediments of CMap data analysis is so-called ‘batch effect’ [22], i.e. cells under the same culture condition lead to highly similar expression patterns, even if they are treated by totally different compounds. In order to overcome the batch effect and make CMap data reflect more signal than noise, a variety of new protocols are successively developed [18], [22]–[24], suggesting the importance of this issue. In order to adjust batch effect as well as keep the integrity of CMap data, we implemented here a novel method to bridge the gap between different batches upon homogeneous drug treatments. Comparing adjusted data with original CMap, we saw that our adjustment procedures lead to improved efficiency of connecting drugs with common protein targets, which solidly facilitated the discovery of potential DTI.
In order to accurately predict DTI with gene-expression profiles, we primarily improved the reliability of CMap data. Ideally, the gene-expression profile of each CMap instance should be solely determined by the bioactivity of treating compound. But the signal is confounded by batch variation, which makes the gene-expression profiles of different batches much less comparable. Iskar et al. [18] used a ‘mean-centering’ method to remedy the batch effect, but at the cost of abandoning many instances in small batches. To present a complete evaluation of CMap based DTI discovery, we therefore developed a novel method that not only overcomes batch variation but also retains all instances (Text S1, Figure S1 and Table S1). We hypothesize that if two instances belonging to different batches are treated by the same drug, the drug action should be homogeneously reflected in two gene-expression profiles, so their difference should be mainly attributed to batch variation. Based on this hypothesis, we select the instances treated by the same drug as ‘bridges’ between two batches, so batch variation is estimated by the difference between bridge instances (see Methods). If the estimated quantity of batch variation is added to the original gene-expression profiles, two different batches could be, in a sense, regarded as derived from the same cell culture and merged into one (Figure 2A).
To bridge the batch variation across all CMap instances, we primarily selected 10 big batches (with not less than 30 instances each) that share a variety of bridge instances ( and Table S3). These 10 big batches are merged together, then other batches are further merged via bridges and so on (Figure 2B and Text S2). Finally, all 6100 instances of 302 batches are unified into an adjusted dataset (freely available upon request).
Across different cell lines and treatment dosages, the instances treated by the same compound are collectively considered, that the fold changes of gene-expression are averaged to obtain a single ‘synthetic expression profile’ for each compound. To measure the gene-expression similarity between two different compounds, we calculated the Bridge Adjusted Expression Similarity (BAES, see Methods) by using a protocol similar as the Gene Set Enrichment Analysis (GSEA) algorithm described in the original CMap publication [21]. In a total, 856,086 BAES scores were calculated across all 1309 CMap compounds. In the same way, we also calculated the gene-expression similarity for original (unadjusted) CMap data.
We assess the efficacy of CMap adjustment by evaluating the correlation between BAES and well-known drug-target interactions. DrugBank database, so far, is one of the most acknowledged sources of drug target information [25]. We therefore mapped the drugs enrolled in DrugBank to the CMap compounds, obtaining 2084 interactions between CMap compounds and 731 DrugBank targets.
We expect that drugs binding to common target result in higher pairwise similarity in gene-expression profiles than random compounds. And it is observed that the BAES significantly outperforms the unadjusted expression similarity [26], in terms of scoring compound pairs that share at least one target in DrugBank (Figure 3). This test corroborates that after batch effect adjustment, CMap expression profiles would better characterize the genomic reactions of ligand binding. Thus, BAES could be used as a guilt-by-association metric to detect potential drug-target interactions, that drugs show high BAES may interact to the same target.
To demonstrate the genuine power of CMap based DTI prediction, we adopted a type of naïve model without any fitting process. For a given target, its designated ligands recorded in DrugBank are defined as ‘benchmarks’. Given the correlation between DTI and gene-expression similarity, we expect the true ligands to show higher BAES to benchmarks than random compounds do. Thus, the likelihood of DTI can be measured by the average BAES between a candidate compound and a series of benchmark ligands (Figure 4A), that higher BAES should indicate higher ‘likelihood of interaction’ (LOI, see Methods).
This model is applied to each human protein target, and the performance is evaluated with leave-one-out cross validation (LOOCV) (see Methods). Taking peroxisome proliferator-activated receptors gamma (PPAR-γ, encoded by PPARG gene) as an example, the 9 PPAR-γ ligands enrolled in CMap are set as benchmarks. All 1309 CMap compounds, including the benchmark ligands (positive set) and other compounds (negative set), are ranked by LOI in LOOCV. Two criteria are used to determine whether DTI is effectively characterized by BAES. Primarily, the area under receiver operating characteristic (ROC) curve should be high and robust. Additionally, the benchmark ligands should be particularly enriched in the drugs with high LOI, thus ensuring the practicability of detecting hidden ligands from the top-ranked drugs. We therefore calculated the 95% confidence interval of area under curve (AUC) [26] and the odds ratio of positive set enrichment. In the above example of PPAR-γ, we can see that most benchmark ligands of PPAR-γ show relatively high LOI, leading to robust ROC curve and significant enrichment of positive set (Figure 4B).
For most tested targets (72 out of 78, accounting for 92%), the benchmark ligands are distinguished from other CMap compounds (i.e. AUC>0.50), suggesting the general efficiency of BAES model. On the other hand, examining the robustness of ROC curve and the benchmark enrichment in top-ranked drugs, we find that individual targets are differentially characterized by CMap (Figure 5 and Data S1). The well characterized targets with robust ROC curve (i.e. the lower bound of AUC confidence interval is over 0.50) and significant benchmark enrichment (i.e. the p-value is less than 0.05) are more likely to be found among neurotransmitter receptors, ion channels, nuclear receptors and cyclooxygenases. Such distinction of performance indicates that the ligands binding of some targets, but not others, can particularly result in extensive and intensive changes at mRNA level, which is exactly detectable in CMap. So instead of building a universal model to predict interactions across all drugs and targets, we suggest that specified models should be established for individual targets (especially the targets well characterized by CMap), in order to increase the chance of detecting true DTI.
Besides the well characterized targets, we found that a variety of targets (such as some neurotransmitter receptors, several calcium ion channels and monoamine oxidases etc.) also exhibit high odds ratio of benchmark enrichment, but not high significance level (Data S1). These observations could be attributed to the limited number of enrolled benchmark ligands (i.e. a small positive set). As a result, the power of Fisher's exact test is impaired, due to too few true positive and false negative samples. For example, serotonin receptor HTR1B showed even better ROC curve than the well characterized target HTR1A, but could not pass the significance test (Figure S2).
This suggests that the sufficiency of benchmark ligands information is critical to the robustness and reliability of CMap based DTI prediction. We therefore look forward to translating the success in CMap into other large-scale genomic expression data resources (such as Gene Expression Omnibus [27] built by NCBI and classified data submitted to FDA by drug developers [28]) or high-throughput data derived from individual studies [29]. Since the data variation brought by the difference of experiment conditions can be effectively adjusted with appropriate computational methods (e.g. BAES), we believe that many external data could turn to be comparable to CMap profiles [30]. Then expression profiles of additional ligands (not enrolled in CMap) can be further used as benchmark ligands, in order to improve the DTI prediction models.
By using the CMap based model, the compounds showing high LOI to particular target are identified, thus providing drug-target pairs with potential interactions. We notice that in some cases, the designated ligands of one certain target tend to collectively interact with another specific off-target (Figure 6A). For instance, the designated ligands of opioid receptor OPRD1 generally exhibit high LOI to calcium channel CACNA1C (Figure 6B), even if OPRD1 and CACNA1C have no DrugBank ligands in common (i.e. they are ‘distant targets’). This phenomenon indicates that the ligands of some targets, as a whole, are likely to share common off-targets. Upon the term of ‘drug-target interactions’, we define such interactions between one target and a family of ligands for another target as ‘target-target interactions’.
As a unique output of CMap based model, target-target interactions are broadly observed across the DrugBank targets (Figure 7A & Data S2), some of which are consistent with previous experimental and clinical evidences. So unlike sporadic drug-target interactions, the target-target interactions are not just proposing individual cases of drug promiscuity, but providing explanations as to complicated actions that prevail in a family of drugs.
For instance, the designated ligands of neurotransmitter receptors generally showed high LOI to the cardiac ion channels (Figure 7B). Cardiac ion channels, present in the membranes of cardiac cells, control the movement of ions across membranes and determine the rate of heartbeat. Modification of these ion channels by drugs can bring about fatal arrhythmias [31]. On the other hand, the major ligands of neurotransmitter receptors are antipsychotic drugs, which are intended to selectively act on central nervous systems. However, as a whole, antipsychotics showed profound association with risk of arrhythmias [32]. Although previous studies have found a few direct interactions between individual antipsychotics (e.g. pimozide, haloperidol and sertindole etc.) and several ion channels [33]–[35], the mechanisms for antipsychotics induced cardiotoxicity remain unclear. Such target-target interactions in CMap suggest that the promiscuous interactions may not be limited to only a handful of antipsychotics and ion channels, but prevalent across many of them. In a systematic view [36], even moderate disturbance to multiple ion channels can add up to fatal impact, while it can be hardly explained by any single ion channel. Therefore, to understand the detailed mechanisms of drug induced cardiotoxicity, the binding affinities towards a variety of cardiac ion channels are recommended to be addressed.
Another example is the target-target interactions concerning with cyclooxygenases (PTGS1 and PTGS2, also known as COX-1 and COX-2). The ligands of cyclooxygenases are largely nonsteroidal anti-inflammatory drugs (NSAIDs), which are expected to relieve inflammation and pain. On the other hand, the NSAIDs are surprisingly reported to reduce cancer risk, by indirectly influencing carcinogenesis pathways [37]. However, the NSAIDs exhibit high LOI to estrogen receptors (targets for breast cancer drugs) in CMap based models (Figure 7C and Figure S3), suggesting that the anti-cancer activity of NSAIDs may also be attributed to direct interactions with anti-cancer drug targets. Consistent with our discovery, a recent study has initiatively identified an NSAID (i.e. diclofenac) targeting estrogen receptors [15]. Thus, we expect more hidden ligands for estrogen receptors to be found from NSAIDs, leading to a new prospective of anti-cancer drug development.
In the present study, several important facts are initiatively discovered. First of all, we demonstrate that drug-induced gene-expression changes are directly correlated with ligand binding, and can be used solely to predict drug target. By adjusting the batch variation, CMap expression similarity can be used as the only indicator of DTI, which provides another cost-effective way of off-target identification. We therefore developed a prediction model, based only on gene-expression profiles. This model is suitable for those studies based only on limited information, such as the studies without large-scale gene network or expensive animal model.
Secondly, we find that not all targets are equally characterized in CMap, i.e. ligands binding to different targets would disturb the expression of different genes. Thus, unlike many one-size-fit-all methods interested in predicting all kinds of DTI, we prefer the target-specific models based on benchmark ligands. Especially for a series of well characterized targets, the ligands are proved to be highly predictable.
Finally, besides proposing sporadic hidden ligands or off-targets, researchers are paying more attention to integration of groups of drugs. For example, Iorio et al. [24] have integrated drugs into communities with similar mode of action, so as to find drugs acting on unexpected pathways. Similarly, we used CMap based model to specifically identify collective interactions between multiple drugs and targets (i.e. target-target interactions). This can help explain the reactions of not individual drugs but drug families, and increase the productivity of studies on drug repositioning and side effects [38].
Meanwhile, we are acutely aware that more effort should be made by learning from other CMap based studies related to the DTI issue. Primarily, DTI discovery is a very complicated problem that requires analyses of various types of information. In a recent study, Iskar et al. [39] have successfully identified a series of transcriptional modules by combining CMap with microarray data of rat models. These transcriptional modules then contribute to a better understanding of drug repositioning and identification of therapeutic targets. Following this example, we plan to further combine our model with other drug-related information (e.g. chemical-protein interactome [8], [40]), thus improving the power of CMap and the efficiency of DTI prediction. In addition, although our current work is focused on drug-target binding, it is well known that the impact of DTI has to be carried out through downstream biological pathways. As an example, Iorio et al. [24] have used CMap expression profiles to identify drugs acting on unexpected pathways. Enlightened by this study, we would extend our target prediction model to the level of downstream pathways, so as to better understand the biological implications of off-targets.
Taken together, we expect our model, along with other related works, can provide a full range of solutions to transcriptomic data analysis for researchers with different interests. By activating CMap and other transcriptomic data sources, gene-expression information would be readily integrated into DTI discovery pipelines in subsequent studies.
The raw data of expression change fold in CMap is downloaded from CMap website (http://www.broadinstitute.org/cmap/). Suppose two different batches (say batch A and B) have n (n>0) pairs of instances treated by the same drug (i.e. bridges). For one certain gene, the expression change fold in the i-th bridge is designated as E(A,i) and E(B,i) in two batches, respectively. Taking all n pairs into consideration, we calculated the average variation of gene-expression profile between two batches in the logarithm form as
According to this quantified variation value, the expression of all instances (not limited to bridge instances) in batch B are transferred intowhich approximates the change fold as if the instances are derived from cell cultures in batch A. And for all the 22,283 genes quantified by CMap microarray platform, the batch variation is bridged one gene after another, following the above procedures. As two different batches are merged into one, the merged new batch is again bridged with other batches and so on, until all CMap batches are adjusted. The data after adjustment as well as the R code can be downloaded at http://cpi.bio-x.cn/cmap/adjusted.zip
By merging batches and combining instances treated by the same compound, we obtained a synthetic expression profile for each of the CMap compounds. The 22,283 genes are then ranked by fold change, that the most up-regulated genes are ranked at top and down-regulated at bottom. Every compound is in turn selected as reference, whose top and bottom ranked 250 genes are used as signature to query all compounds by GSEA algorithm. A pair of drugs, say drug A and B, could have two similarity scores, one score by querying B with A's signature and the other score in opposite. The BAES is defined as the average value of these two scores. Following the same procedure, we also calculated the similarity score with the original unadjusted CMap data to make a comparison.
Given the direct correlation between drug-target binding and BAES score, we assume that the likelihood of a candidate compound (symbolized as C) binding to a specific target (symbolized as T) can be reflected by the overall expression similarity between the compound C and designated ligands of the target T. Suppose target T has N ligands enrolled in CMap, the likelihood of C interacting with T is estimated as followswhere BAESi represents the BAES score between the compound C and the i-th designated ligand of target T.
We perform leave-one-out cross validation to evaluate how well each target is characterized by CMap. Each CMap compound, including benchmark ligands and background drugs, serves as test set in turn, and all other compounds as training set. The LOI of the compound in test set is determined by its average BAES score to the benchmark ligands in training set. Then the benchmark ligands (positive compounds) and other CMap compounds (negative compounds) are classified by LOI, whose performance is illustrated with ROC curve. The 95% confidence interval of area under ROC curve is computed by the pROC package [26] for R environment (http://www.r-project.org/), with 2000 replicates of bootstrap test. The LOI corresponding to 90 percent specificity is set as the threshold to discriminate positive and negative compounds. The enrichment for benchmark ligands above threshold is calculated as an odds ratio (OR):in which TP, TN, FP and FN represent true positive, true negative, false positive and false negative samples, respectively. To assess the statistical significance of enrichment, we performed Fisher's exact test based on the 2 by 2 contingency table corresponding to the four factors of odds ratio. To ensure the efficiency of bootstrapping and statistical test, the evaluation is confined to a total of 78 DrugBank human protein targets with at least 5 designated ligands enrolled in CMap.
|
10.1371/journal.pcbi.1004663 | Discovery of Influenza A Virus Sequence Pairs and Their Combinations for Simultaneous Heterosubtypic Targeting that Hedge against Antiviral Resistance | The multiple circulating human influenza A virus subtypes coupled with the perpetual genomic mutations and segment reassortment events challenge the development of effective therapeutics. The capacity to drug most RNAs motivates the investigation on viral RNA targets. 123,060 segment sequences from 35,938 strains of the most prevalent subtypes also infecting humans–H1N1, 2009 pandemic H1N1, H3N2, H5N1 and H7N9, were used to identify 1,183 conserved RNA target sequences (≥15-mer) in the internal segments. 100% theoretical coverage in simultaneous heterosubtypic targeting is achieved by pairing specific sequences from the same segment (“Duals”) or from two segments (“Doubles”); 1,662 Duals and 28,463 Doubles identified. By combining specific Duals and/or Doubles to form a target graph wherein an edge connecting two vertices (target sequences) represents a Dual or Double, it is possible to hedge against antiviral resistance besides maintaining 100% heterosubtypic coverage. To evaluate the hedging potential, we define the hedge-factor as the minimum number of resistant target sequences that will render the graph to become resistant i.e. eliminate all the edges therein; a target sequence or a graph is considered resistant when it cannot achieve 100% heterosubtypic coverage. In an n-vertices graph (n ≥ 3), the hedge-factor is maximal (= n– 1) when it is a complete graph i.e. every distinct pair in a graph is either a Dual or Double. Computational analyses uncover an extensive number of complete graphs of different sizes. Monte Carlo simulations show that the mutation counts and time elapsed for a target graph to become resistant increase with the hedge-factor. Incidentally, target sequences which were reported to reduce virus titre in experiments are included in our target graphs. The identity of target sequence pairs for heterosubtypic targeting and their combinations for hedging antiviral resistance are useful toolkits to construct target graphs for different therapeutic objectives.
| An average of three influenza pandemics occurred in each century over the last 300 years. As occurrence of the next influenza pandemic is definite, developing new antivirals is imperative since resistance to the remaining class of antivirals has been reported occasionally, and vaccines are ineffective in the initial wave of a pandemic. The typical evolutionary traits of viruses, which manifest as multiple virus subtypes in circulation and perpetual viral genomic mutations, require the development of subtype-specific antivirals that ultimately acquire resistance. Being a rapidly evolving and highly contagious virus that manifest the most subtypes, this is particularly acute for influenza A. Our approach to overcome these challenges is to identify and characterize influenza A virus sequences for RNA targeting that can theoretically address all strains from the most prevalent human-infecting subtypes (i.e. simultaneous multi-subtype targeting) that can hedge against antiviral resistance. We uncover an extensive list of target sequence pairs and their specific combinations for which they can be selected for novel therapeutics development that will be effective on multiple circulating seasonal strains and future pandemic strains. As our approach is applicable to other viruses, the methods are general for use in the selection of antiviral therapeutic targets.
| An average of three influenza pandemics occurred in each century over the last 300 years [1]. The time interval between consecutive pandemics and their respective mortality are however irregular; while the 1918 H1N1 Spanish flu was estimated to kill 50 million people, the 2009 H1N1 Swine flu pandemic was probably responsible for 100,000 to 200,000 deaths [2]. As substitution of a few specific amino acids can be sufficient to alter host tropism [3,4], it is relatively easy for a novel influenza A viral subtype previously circulating in animals against which the general human population lacks antibody-mediated immunity to cause future pandemics.
Vaccines and anti-viral drugs respectively are the main biologics and pharmaceuticals tools to reduce the morbidity and mortality of a pandemic. Depending on the circulating viral strains and the recipients’ age, vaccine effectiveness can be lower than 40% [5]. Moreover, vaccines are unlikely to be available in the initial wave of a pandemic as current vaccine approaches are lineage and subtype-specific and such vaccines are typically developed after the new antigenically distinct pandemic virus has emerged. Anti-viral drugs are classified by their target viral proteins, typically–M2 ion-channel inhibitors (amantadine and rimantadine) and neuraminidase inhibitors (e.g. oseltamivir and zanamivir) [6]. The former are now ineffective against current circulating H3N2 and H1N1 (2009) subtypes [7–9]. Neuraminidase inhibitors are the sole antiviral option in a pandemic while incidences of resistance have been reported occasionally [10–15]. When coupled with so-called permissive mutations, the classical Tamiflu-resistance mutation H274Y (H275Y) can become more prevalent and in the case of the previous seasonally circulating H1N1 virus, became fixed in circulating viruses rapidly in 2008 [16]. Developing new antivirals to anticipate resistance in seasonal as well as potential future pandemic viruses is thus imperative. Unfortunately, the multiple circulating influenza A virus subtypes coupled with the perpetual genomic mutations and segment reassortment events challenge the development of effective therapeutics against multiple circulating and future strains.
An arsenal of protein inhibitors that each binds to distinct sites of all expressed viral proteins is one strategy for heterosubtypic targeting and to hedge against inevitable antiviral resistance. While chemical protein inhibitors constitute most pharmaceuticals, their targets are limited [17–18]. Alternatively, prior studies have demonstrated viral RNA targeting via siRNA or antisense oligonucleotides (AONs) as viable antiviral strategies [19–42]. They can potentially target any sequences within a viral RNA segment leading to RNA degradation that is elicited by either RNAi or RNase-H, or inhibition of RNA splicing or translation by steric hindrance effects [43]. Of particular clinical relevance is the demonstration that intranasal AON inhalation is an efficient delivery vehicle to the respiratory tract and lungs in animal studies [21,24,27–28,35]. Additionally, rational computational methods [44] to identify optimum RNA target sites can facilitate rapid development of an AON library for hedging against antiviral resistance and for targeting novel viral strains in a pandemic.
To examine heterosubtypic RNA targeting, we identify and characterize conserved RNA target sequences in the eight influenza A virus segments from subtypes infecting humans and animals. Analyses on 168,986 segment sequences derived from 51,661 human and animal strains reveal thousands of specific pairs of target sequences that can address all prevalent circulating human strains simultaneously. Novel strategies for target sequence selection to hedge against antiviral resistance illuminate countless sets of target sequence combinations with distinct hedging capacities. Together, the target sequences and their specific combinations discovered in this pan-virus subtype study is a useful resource for the development of effective RNA therapeutics, which targets viral RNA, mRNA or cRNA, against multiple circulating and future strains.
Five subtypes representing the most prevalent human infecting Influenza A viruses in the past four decades were studied–H1N1 (before 2009; hitherto refer to as H1N1), 2009 pandemic H1N1pdm09 (hitherto refer to as PD09), H3N2, H5N1 and H7N9. Although both H5N1 and H7N9 subtypes are primarily avian influenza viruses, they have periodically caused human infections with occasional reports of human adaptive mutations and therefore pose a significant risk of pandemic potential. For each subtype, all available sequences of each of the eight viral segments were downloaded from curated databases (refer to S1 Text). 123,060 segment sequences from 35,938 strains of which 70,723 are unique were analysed (S1 Table breakdowns the sequence counts by subtype and segment).
Two sets of RNA target sequences were obtained. Sequences in the “5-S” set were selected to optimally target the five subtypes simultaneously whereas sequences in the “3-S” set were selected to target H1N1, PD09 and H3N2 simultaneously. The 5-S set was obtained as depicted in Fig 1. First, the consensus sequence of the entire coding segment was determined for every segment of each subtype from the respective unique sequences. Next, for each segment, sequence alignment was performed on all the respective consensus sequences from the five subtypes simultaneously. Consensus motifs defined as sections of the consensus sequence with perfect alignments were identified. Finally, target sequences of at least 15 nucleotides were selected from the respective collection of consensus motifs in each segment; this minimum target length is chosen for RNA binding specificity and thermodynamic stability. The 3-S set was obtained similarly by omitting H5N1 and H7N9 segment sequences. Section A in Table 1 summarizes both the 5-S and 3-S sets; note that 5-S is a subset of 3-S and the much smaller number of conserved target sequences in 5-S illustrates, not surprisingly, that target sequence conservation strongly depends on the number and selection of strains. Both segments 4 and 6, which code for the more variable hemagglutinin and neuraminidase surface proteins respectively, cannot be targeted as they do not share any 15-mer sequences between the subtypes consensus. When the target sequences counts were normalized with the respective segment coding lengths, conserved target sequences appear enriched in segment 7 (coding for M1 and M2 proteins in alternative frames) in 5-S and 3-S, and in segment 1 (coding for PB2 protein) of 3-S.
To evaluate coverage of intra-subtype variation for every conserved target sequence in a segment, it was matched against every unique sequence of the respective segment from five subtypes in 5-S or from three subtypes (H1N1, PD09 and H3N2) in 3-S. No target sequence was found in all relevant unique segment sequences although there are always some in each target segment that are found in more than 95% of the respective unique sequences (S1 and S2 Figs). The coverage against human corresponding animal subtypes (aH1N1 aH3N2 aH5N1 and aH7N9) and three groups of collective subtypes labelled as “H00N00”, “zoonotic” and “exotic” (Materials and Methods) were also determined. The H00N00 group consists of eight subtypes that have infected humans but are not or no longer in large-scale human circulation whereas the zoonotic and exotic groups respectively consist of 78 and 19 animal subtypes with zoonotic potential. 45,926 sequences from the six internal segments of which 32,961 are unique, were consolidated from 15,728 strains and analysed (S2 Table). Notably, there are human target sequences that are found in more than 90% of the unique sequences in each target segment of each animal subtype and in each group of subtypes (S3 and S4 Figs present the coverage of every target sequence against each human and corresponding animal subtypes, and against each group of subtypes). Hence, both 5-S and 3-S sets are relatively conserved in a total of 109 human and animal subtypes; for more coverage analyses, see S5 and S6 Figs. Coverage against Influenza B virus was 0%. In order to achieve 100% coverage in human subtypes, we next considered target sequence pairs.
Target sequences within a segment were paired (each pair is termed a Dual). In the 5-S set, effective Duals in four segments (1, 2, 3 and 7) and in the 3-S set, effective Duals in six internal segments can cover all unique sequences of respective target segments (Section B in Table 1). That is, one or both of the target sequences constituting an effective Dual are found in all unique segment sequences. A significant fraction of single target sequences in segments 1, 3 and 5 forms effective Duals (Section B in Table 1). The distribution of the target sequence positions from the effective Duals in each segment are depicted in S7 Fig.
Next, overlapping single target sequences in a segment (tallied in Section B in Table 1) were grouped as clusters. Two clusters are paired when target sequences between the two clusters form one or more effective Duals. The cluster pairings in both 5-S and 3-S are depicted as graphs in which a pairing is denoted by an undirected edge connecting two clusters depicted as vertices (Fig 2). The utilization of cycle graphs in Fig 2 (i.e. vertices connected in a closed chain) in selecting effective Duals for hedging against antiviral resistance will be discussed.
Alternatively, target sequences from two segments were paired (each pair is termed a Double). A Double can target a virus strain when one or both of its target sequences is found in either one or both of the strain’s respective segment sequences. The coverage of a Double is defined as the fraction of total virus strains in five subtypes in 5-S or in three subtypes in 3-S that it can target. The coverage of every Double was determined for every segment pairing (each with at least 10,000 strains, S3 Table). As summarized in Table 2, for each of the 15 target segment pairings, there are effective Doubles with 100% strain coverage. In contrast to Duals (Section B in Table 1), most of the single target sequences are included in effective Doubles (Table 3). The collective distribution of the target sequence positions from effective Doubles in each segment are given in S8 Fig.
For every single target sequence tallied in Table 3, the number of segment partners (NSP) was determined by counting the number of segments with which it can form an effective Double with their respective single target sequences; NSP thus ranges from one to five, as illustrated in Fig 3A. Interestingly, the frequency distributions of NSP in each segment show that the number of target sequences with a high NSP is common and in some segments, the minimum NSP of all single target sequences is greater than unity (Figs 3B and S9A). This indicates a high reusability of a target sequence to form effective Doubles with different target segments.
As there are single target sequences in every segment whose NSP is five (Fig 3B), we investigate the size distribution of all 6-vertices segment partner graph formed by one (NSP = 5) target sequence from each of the six internal segments. The graph size is the sum of edges with each edge connecting two vertices (representing target sequences) denoting an effective Double; hence, the size indicates the number of effective Doubles in a particular segment partner graph. As shown in Figs 3C and S9B, the modal number of effective Doubles per graph in 5-S and 3-S is 13 and 7 respectively, and every graph in 5-S has at least 10 effective Doubles. In addition, 808,704 and 3,944,376 graphs in 5-S and 3-S respectively have 15 effective Doubles, which are termed as complete graphs i.e., every pair of distinct vertices is connected by a unique edge (S9B Fig). Identical analyses of 6-vertices segment partner graphs constructed by single target sequences with NSP ≥ 1 (5-S) and with NSP ≥ 4 (3-S) also reveal significant number of graphs with a big size (≤ 14) (S10 Fig). The existence of big-sized and complete graphs creates the possibility to hedge against antiviral resistance, as described next.
To mitigate antiviral resistance, a minimum of three target sequences is needed so that there is a target sequence to replace one that has become resistant. In this context, a resistant target sequence is unable to achieve 100% coverage of all subtypes strains when paired with another non-resistant target sequence. We define the hedge-factor to evaluate the extent a set of target sequences can potentially mitigate antiviral resistance. When depicted in a graph, wherein vertices denote selected target sequences and an edge represents an effective Dual or Double, the hedge-factor is the minimum number of resistant target sequences that will eliminate all the edges therein i.e. abolish the set’s therapeutic effectiveness to achieve 100% coverage. The hedge-factor in a set is maximal when the target sequences form a complete graph (demarcated with a border in Fig 4A) i.e. every pair of distinct vertices is connected by a unique edge; in an n-vertices complete graph, maximum hedge-factor is n– 1 for n ≥ 3. A set of target sequences that forms a complete graph has the advantage of maintaining the 100% coverage of all strains by any pair of target sequences, which is valuable when target sequences need to be interchanged upon developing resistance.
For effective Duals, Fig 4B gives the maximum hedge-factor (number of grey nodes) in each target segment. Complete graphs, which are mostly 3-vertices graphs, are formed only in segments 1, 3 or 7 –S1 (5-S: two 3-vertices; 3-S: three 3-vertices), S3 (3-S: two 3-vertices) and S7 (S-5: two 3-vertices; 3-S: one 4-vertices and seven 3-vertices). Generally, complete graphs in 5-S have higher hedge-factors than in 3-S, and the highest hedge-factor of five is attained in segment 7. In fact, although the number of effective Duals in segment 7 is lowest among the three target segments in 3-S (Section B in Table 1), they form the biggest and most number of complete graphs. For effective Doubles, complete graphs involving 3, 4, 5 or 6 segments resulting in hedge-factors from two to five exist (Fig 4C). Myriad combinations of effective Doubles and effective Duals can be used to construct target sequence graphs that surpass the hedge-factor limit of five from using Doubles or Duals alone, two of which are depicted in Fig 4D.
Monte Carlo simulations were used to study the effect of hedge-factor on the mutation counts and time elapsed for a set of n target sequences to become resistant i.e. lose its 100% coverage. In the simulation model (Materials and Methods and S2 Text), it is considered resistant when n– 1 target sequences are resistant; a target sequence becomes resistant when its target site acquires a mutation and thereby cannot achieve 100% coverage. Thus, the mutation count is the minimum number of mutations in a virus to become resistant to a set of target sequences, with the corresponding time elapsed obtained by dividing the mutation counts with the average total annual mutation events in a virus [45]. Mutation events in 100,000 viruses were simulated for each set of target sequences to determine the median mutation counts and median time to resistance. Fig 5A plots the medians for sets of effective Duals forming complete graphs for a range of hedge-factors in segments 1, 3 and 7 depicted in Fig 4B. Expectedly, as the hedge-factor of a set is increased, substantially more mutation events and longer time is required to attain resistance. Likewise, complete graphs of effective Doubles in Fig 4C with higher hedge-factors possess considerably larger capacity to hedge against resistance (Fig 5B). Lastly, the target sequence graphs that combine effective Doubles and effective Duals in Fig 4D to augment the hedge-factor can further increase their hedging capacities (HF = 6(1) in Fig 5B).
The target sequences were aligned with up to one mismatch to the genomes and transcriptomes from human, pig and chicken hosts (Materials and Methods). Up to 4.7% and 3% of them in 5-S and 3-S sets respectively were found in one or more of the hosts’ transcriptomes whereas up to 22% and 12.7% of target sequences in 5-S and 3-S sets respectively were found in one or more of the hosts’ genomes (Section A of S4 Table). The 56 and 165 respective target sequences in 5-S and 3-S sets that hit the human transcriptome were mapped to 36 and 133 human genes (Section B of S4 Table). Among the 27 and 89 respective genes whose expression data were known, which correspond to 45 and 122 hit target sequences in 5-S and 3-S sets respectively, not all are expressed in tissues of the respiratory system. Nevertheless, when all the hit target sequences from each set independent of their tissue expression status were removed, complete heterosubtypic coverages by either effective Duals or effective Doubles (Tables 4 and 5), as well as maximal hedging against resistance by complete graphs (S11A Fig) are retained; all the data are made available at S1 Text. Similarly, when the respective target sequences in 5-S and 3-S sets that hit either the human genome or transcriptome were excluded, complete heterosubtypic coverages by effective Duals or Doubles, and hedging against resistance by complete graphs are conserved (Tables 4 and 5, S11B Fig and S1 Text).
The capacity to drug most RNAs motivates the investigation on viral RNA targeting to address multiple circulating human subtypes and to mitigate antiviral resistance. 123,060 segment sequences and 35,938 virus strains from H1N1 (prior 2009), PD09 (2009 pandemic H1N1pdm09), H3N2, H5N1 and H7N9 representing the most prevalent human infecting subtypes over the past four decades were used to identify and characterize two sets of target sequences, each with a minimum length of 15 bases, for their coverage in targeting the multiple subtypes either singly or in pairs. A total of 1,183 conserved target sequences in the 5-S set and 5,523 conserved target sequences in the 3-S set (only H1N1, PD09 and H3N2 subtypes were analysed) were identified in all but segments 4 and 6 (Section A in Table 1). Notably, simultaneous heterosubtypic targeting of all the subtypes is achieved when specific pairs of same-segment (effective “Duals”) or two-segment (effective “Doubles”) target sequences are used. In 5-S and 3-S respectively, large numbers of effective Duals (1,662 and 29,124) and effective Doubles (28,463 and 280,351) exist (Section B in Table 1 and Table 2). The target selection space of Doubles is larger (Section B in Table 1 vs. Table 2)–(a) there are about 10 times more effective Doubles than effective Duals; (b) almost all single target sites can be paired to form effective Doubles but not effective Duals.
The specific pairings of multiple effective Duals, effective Doubles or both can generate distinct sets of target sequences each with different potential to hedge against antiviral resistance, as indicated by its hedge-factor (Fig 4). As the hedge-factor is maximal when target sequences in a set form a complete graph, they would be top choices for target selections. The number of possible complete graphs is enormous particularly those formed among effective Doubles (Figs 3C and S9B). Importantly, because effective Doubles in the six internal segments can form complete graphs (Fig 4C vs. Fig 4B) unlike effective Duals where complete graphs exist only in three segments (1, 3 and 7), they can hedge against resistance arising from segment reassortment events. That is, when a target segment undergoes reassortment and thereby becomes resistant to a target sequence, there are options to target another segment. Multi-segment targeting could be essential to address the observation that a third of avian flu A virus samples harbours at least one reassorted segment [46–47], although the reassortment frequency is likely lower in human subtypes. The Monte Carlo simulation results corroborate the use of combinatorial targets to significantly prolong antiviral resistance in HIV infections to a timescale typical for a chronic disease [48]. Incidentally, several target sequences in 5-S and 3-S sets that have been experimentally validated to reduce virus titre can be paired to form effective Duals, effective Doubles, and target graphs with hedge-factor of 2 (S3 Text). Thus, they can be readily be used for animal studies prior human clinical trials.
Although the time to resistance of a set of target sequence is primarily dependent on the hedge-factor, it is affected by target sequence length and target segment mutation rate (Fig 5). At a given hedge-factor, the time to resistance correlates inversely with target sequence length since a long target sequence has more nucleotides to acquire a mutation. Target sequences in a relatively slower mutating segment have longer time to resistance–for instance, sets of effective Duals in segment 7 has the longest time to resistance for the same hedge-factor and target sequence length (Fig 5A). In summary, Fig 5 provides a reference for specifying key parameters pertaining to target segment(s), hedge-factors, target sequence length, and hedging capacity. Together with Fig 4, they offer the toolkits for assembling sets of target sequence for specific therapeutic objectives.
Besides efficacy and efficiency, off-target side effects are major obstacles to a successful drug in human studies. In the context of viral RNA suppression as a therapeutic strategy, off-target effects occur when a viral target site is also found in the host cell transcriptome whose expression is essential. Notably, when target sequences with potential off-target effects were removed, the reduced numbers of effective Duals, effective Doubles and complete graphs still remain enormous as drug targeting space; for example, the 3-S set has 22,380 effective Duals, 242,803 effective Doubles, and 905,850 complete graphs of size 15 (Tables 4 and 5 and S11 Fig). Other off-target effects result from polypharmacological properties of a drug through binding to unspecific target sites, high-dose non-specificity effects, immunological response [49] or other non-sequence dependent effects [43,50]. Specific nucleic acid chemistry and modifications such as morpholino and 2’-O-methyl with phosphorothioate or phosphorodiamidate backbones have improved binding specificity [43,51–53] and could help to overcome some of these problems.
Due to the relative small number of available H5N1 and H7N9 strains (S1 Table), they were excluded in the determination of conserved target sequences in the 3-S set. Consequently, the number of target sequences, effective Duals and effective Doubles in 3-S is about an order of magnitude larger than in 5-S (Table 1); an expanded target space is typically useful for therapeutic development. Notably, consideration of H5N1 and H7N9 subtypes does not affect the coverage of target sequences against human subtypes, as the coverage distributions against H1N1, PD09, H3N2 and H00N00 in both sets are similar (S12 Fig). In contrast, coverage of target sequences in the 5-S set was greatly improved over in the 3-S set against all animal subtypes as well as both zoonotic and exotic groups of animal subtypes (S12 Fig), except for segments 2 and 8. Thus, one can select target sequences in segments 2 and 8 from the 3-S set for their larger target space and target sequences in segments 1, 3 5 and 7 from the 5-S set for their better extensibility against cross-species subtypes to pre-empt future strains that cross from animal to human hosts.
Curated strains with incomplete genomes and segment sequences with non-full-length were both included in the analyses for considering as many sequence variations as possible. When only complete-genome strains were analysed for both 5-S and 3-S sets, the number of target sequence pairs and their combinations do not change significantly. This is expected because the analyses of both effective Duals and effective Doubles do not require genome completeness. Moreover, all the effective Duals, effective Doubles and complete graphs of size 15 from the all-genome analysis are complete subsets of those in the complete-genome analysis (S5 Table). On the other hand, when only full-length segment sequences were analysed, significantly more target sequence pairs and their combinations were obtained in both 5-S and 3-S sets. Similarly, all the effective Duals, effective Doubles and complete graphs of size 15 from the all-length analysis are complete subsets of those in the full-length analysis, except for five effective Duals (out of 29,124) and 54 effective Doubles (out of 280,351) in the 3-S set (S6 Table). Therefore, not all results from the complete-genome and full-length analyses respectively are applicable to complete- and incomplete-genome strains, and to full-length and non-full-length segments. In short, considering only complete genomes and full-length segments overestimates conservation of candidate sites while additional variations observed in incomplete genomes/segments lead to un-selection of more sites as not being conserved. Therefore, an analysis considering all available influenza sequences will provide the most robust selection hedging against further genomic changes and natural virus evolution.
This study inevitably poses the following questions–which segment to target when Duals are used, which segment combinations to target when Doubles are used, and how to prioritise a set of RNA drugs targeting multiple sequences for clinical trials? Reduction of virus titre upon either knockdown or knockout of specific segment have been reported [19–42], however, there is no comparative study to determine the relative suppression by each targeted segment on virus viability and replicability. Moreover, only four out of 15 combinations of double segments targeting have been reported; segment 5 paired with segment 1, 2, 3 or 7 [26,29,31–33]. The feasibility to target a segment is also dependent on the accessibility of its target sequence’s secondary structures for efficient drug binding. The secondary structures and thereby binding accessibility of a target sequence between subtypes can vary due to nucleotide variations among segment sequences (S13 Fig). To facilitate the selection of a set of target sequences that lead to efficient RNA therapeutics targeting viral RNA, mRNA or cRNA, the following resources for respective 5-S and 3-S sets are made available for download–coverages of all target sequence against all analysed strain sequences from human and animal subtypes, pairings of all effective Duals and Doubles, and binding accessibilities [54] of every target sequence (S1 Text); two versions (inclusion or exclusion of hit target sequences to the human transcriptome or genome) are provided per resource.
It is possible that RNA therapeutics could develop antiviral resistance easier than protein-targeting drugs via silent mutations. However, at least three simultaneous mutations at a target sequence are required to abrogate the AON efficiency [30,34], which further substantially increase the mutation counts and prolongs the elapsed time required for a set of target sequences to become resistant (S14 Fig). The availability of drugs for simultaneous heterosubtypic targeting is likely to be more effective and therefore may slow down and reduce the severity of pandemic and seasonal flu infections, which limits the reservoir of hosts for the virus to evolve. In addition, clinical administration of a drug cocktail with more than two RNA drugs to further delay antiviral resistance is worth exploring albeit two targets are theoretically sufficient to address all prevalent subtypes. Nonetheless, since few genomic modifications suffice to create a novel virus strain with strongly altered transmission phenotype [3–4], the strategy of selecting a resistance-hedging set of multiple target sequences is particularly relevant as some of the target sequences are likely to remain effective against a new strain. This is corroborated from the results that there are human target sequences that are found in more than 90% of the unique sequences from a total of 109 human and animal subtypes of differing zoonotic potential. (S3, S4, S5, and S6 Figs). Finally, the concept of Duals, Doubles and hedge-factor can potentially be applied to other viruses that manifest multiple subtypes to develop RNA therapeutics addressing the subtypes simultaneously and for mitigating antiviral resistance.
Influenza A virus nucleotide sequences from both human and animal hosts were downloaded from GenBank for all eight segments for H1N1 (before 2009), PD09 (2009 pandemic H1N1), H3N2 and H5N1 subtypes. For the H7N9 subtype, nucleotide sequences for both human and animal hosts were downloaded from the Global Initiative on Sharing All Influenza Data (GISAID) Epiflu database; we acknowledge the authors, originating and submitting laboratories of the sequences analysed from the Database, listed in gisaid_acknowledge_table_processed.txt. To further assess the degree of conservation and extensibility of target sequences, three other groups of influenza viruses were downloaded from GenBank. The first group consists of viruses from any other subtypes with history of infecting humans: H1N2, H2N2, H6N1, H7N2, H7N3, H7N7, H9N2 and H10N8, collectively named as “H00N00”. The second group, named as “zoonotic”, is considered as having a higher zoonotic potential. To obtain this set of viruses, every protein from all viruses from the H00N00 group was used as query to the tachyon server [55] and all animal strains that are within the top 50 hits with a tachyon score of 0.8 or more would be considered as having zoonotic potential. The third group of viruses, named as “exotic”, are viruses from less common host sources like Equine, Canine, Ferret, Cat, Seal, Tiger, Pika, Mink, Bat, Penguin, Bovine, Wild boar, Raccoon dog, Camel, Leopard, Muskrat, Cheetah, Feline, Stone marten, Panda, Civet, Whale, Giant anteater, Blow fly or Beetle origin. Sequences used in the analyses were downloaded on 29th April 2014. To ensure that only unique sequences were analysed for each subtype, the redundant identical sequences were removed with Cd-hit [56] by allowing a maximal sequence identity of 100%. Although UTRs were not used in identifying target sequences, they were not removed from the sequences as they were used for secondary structure predictions when designing AONs.
A computational model is developed to simulate the random single nucleotide substitution events in each of the eight viral segments; the model assumptions are discussed in S2 Text. Monte Carlo simulations were applied on the model as follows:
The segment mutation rates (S2 Text) are used to compute the probability of each segment, Pr(Segment), where the next substitution mutation will occur in the Monto Carlo simulations, and the time to resistance. Pr(Segment) is calculated by normalizing the segment’s mutation rate with the total number of substitution events from all the segments in a year. Lastly, the time to resistance is computed by dividing the mutation counts with the total number of substitution events from all the segments in a year. The simulation was implemented in JavaTm programming language; source codes can be downloaded at http://mendel.bii.a-star.edu.sg/SEQUENCES/HEDGING_DRUG_RESISTANCE/source-codes/index.html.
The human (GCF_000001405.28_GRCh38.p2), pig (GCF_000003025.5_Sscrofa10.2) and chicken (GCF_000002315.3_Gallus_gallus-4.0) genomic and transcriptomic sequences were downloaded from the NCBI genome resource (ftp://ftp.ncbi.nlm.nih.gov/genomes/all/). The blastn program [57] was used with both the default parameters, and also with parameters adjusted for short sequence searches (e-value of 1000, word size 7, no complexity masking) to search the target consensus sequences against the human transcriptome and genome for possible cross-reactivity. In order to reduce cross-reactivity, target sequences that have a hit in the human transcriptome were removed. The hits had maximally one mismatch and no target sequence was found to have more than one mismatch from the above search criteria. In any case, sequences with two- and three-mismatches are inefficient and ineffective to target respectively [30,34]. The genes in the human transcriptome that match the target sequences for up to one mismatch were examined for their gene expression profile in 84 tissue types using data obtained from Gene Atlas (Human U133A/GNF1H, GSE1133, http://biogps.org/downloads, click the link to gnf1h-gcrma.zip) [58]. Perl scripts were used to map the accessions from the human ptome to their respective genes using the NCBI gene2accession data file (ftp://ftp.ncbi.nlm.nih.gov/gene/DATA/gene2accession.gz); the scripts can be downloaded at http://mendel.bii.a-star.edu.sg/SEQUENCES/HEDGING_DRUG_RESISTANCE/source-codes/index.html.
|
10.1371/journal.pcbi.1006075 | Spatial modeling of the membrane-cytosolic interface in protein kinase signal transduction | The spatial architecture of signaling pathways and the interaction with cell size and morphology are complex, but little understood. With the advances of single cell imaging and single cell biology, it becomes crucial to understand intracellular processes in time and space. Activation of cell surface receptors often triggers a signaling cascade including the activation of membrane-attached and cytosolic signaling components, which eventually transmit the signal to the cell nucleus. Signaling proteins can form steep gradients in the cytosol, which cause strong cell size dependence. We show that the kinetics at the membrane-cytosolic interface and the ratio of cell membrane area to the enclosed cytosolic volume change the behavior of signaling cascades significantly. We suggest an estimate of average concentration for arbitrary cell shapes depending on the cell volume and cell surface area. The normalized variance, known from image analysis, is suggested as an alternative measure to quantify the deviation from the average concentration. A mathematical analysis of signal transduction in time and space is presented, providing analytical solutions for different spatial arrangements of linear signaling cascades. Quantification of signaling time scales reveals that signal propagation is faster at the membrane than at the nucleus, while this time difference decreases with the number of signaling components in the cytosol. Our investigations are complemented by numerical simulations of non-linear cascades with feedback and asymmetric cell shapes. We conclude that intracellular signal propagation is highly dependent on cell geometry and, thereby, conveys information on cell size and shape to the nucleus.
| Frequently, cells detect signals at their surface, which are transmitted to the nucleus. The influence of cell shape and size is often neglected and cells are regarded as well-mixed compartments. However, the advance of modern microscopy has unraveled heterogeneous distribution of signaling molecules in the cell and variations depending on cell shape, size and organelle arrangement. Understanding spatial signaling usually involves solving mathematical equations in space and time including approximations or sophisticated numerical methods. We provide exact analytical solutions for the steady state of two different spatial arrangements of a generic linear signaling cascade model. Furthermore, the dynamic process is investigated using advanced computational techniques. Implications are drawn on single-cell variation in signal transduction and on spatial regulation by cell size and shape.
| Cells need to respond to a large variety of external stimuli such as environmental changes or extracellular communication signals. Signals transmitted from cell surface receptors to target genes in the nucleus are frequently transduced by cascades of covalent protein modifications. These modifications consist of inter-convertible protein forms, for instance, a phosphorylated and an unphosphorylated protein. Signaling cascades occur in many different variations including mitogen-activated protein-kinase (MAPK) cascades and small GTPase cascades.
Signal transduction mechanisms carried out by networks of protein-protein interactions are highly modular and regulatory behavior arises from relatively simple modifications [1]. The spatial arrangement of signaling cascades varies in different biological systems. We focus on the localization of signaling components, which can be tethered to the cell-membrane or freely diffuse in the cytosol. Tethering of signaling molecules to the cell-membrane can be mediated by lipidation modifications [2–6], co-localization by membrane-bound scaffolds [7] or membrane anchoring proteins [8]. Frequently, the first steps of signal transduction occur at the membrane and are then continued into the cytosol. We investigate linear signaling cascades with different realizations of spatial arrangements of signaling components as shown in Fig 1. Here, we focus on the membrane-cytosolic interface, which is included in the signaling motif shown in Fig 1(B) and 1(C).
In many experimental and theoretical studies on signaling cascades, the cell is regarded as a number of well-mixed compartments with no variation in size, shape or organelle location. Attempts of a quantitative description of signaling cascades with a focus on temporal aspects have been made in [9–12]. However, the spatial description of signaling processes has received less attention despite its relevance in understanding cell morphology and growth regulation in time and space [13]. Examples of spatial effects on the length scale of single cells range from the yeast mating process [14, 15] to the propagation of spatial information in hippocampal neurons which is controlled by cell shape and vice versa [16, 17].
Since the cytosol scales with cell volume and the cell membrane with the cell surface area, reactions on the membrane and in the cytosol scale with the cell-surface to cell-volume ratio. For instance, we obtain an area/volume ratio of ∝ 3/Rcell for a spherical cell geometry, where Rcell is the cell radius. We will show that this scaling affects the global phosphorylation rate of signaling proteins that diffuse in the cytoplasmic volume, which depends on cell size. While cytosolic gradients naturally occur from the membrane to the nucleus, membrane-bound components can only form gradients along the membrane, which changes the response to heterogeneous signals. Furthermore, the diffusion on the membrane is much slower for membrane-bound components than for cytosolic components [18]. Both of these factors are expected to largely change signal transduction properties of the pathway.
An analysis and comparison of spatial signal transduction motifs in response to spatially homogeneous and heterogeneous signals is presented in this study. The natural extension of widespread used ordinary differential equations are bulk-surface partial differential equations [19, 20]. Here, bulk refers to the cellular compartments that are represented as a volume such as the cytoplasm or the nucleus, while surface refers to all cellular structures that are represented as an area such as the cellular or nuclear membrane. Since their introduction to cell signaling systems [21], bulk-surface partial differential equations have been successfully employed in several models for cell polarization [18, 22–24]. However, membrane-cytosolic interfaces at different stages of a signaling cascade have not yet been investigated.
We start with an analysis of two different motifs with simplified linear kinetics, which allows to develop exact analytical solutions of the steady state. Both motifs differ in their cell size dependence and we show further that their behavior can be drastically different from the assumption of well-mixed compartments. The time-scaling of signal transduction is investigated using the method of local accumulation times [25]. We continue by investigating the response and sensitivity to spatially heterogeneous signals such as signaling gradients for symmetrical and asymmetrical cell shapes. In the last section, we proceed with numerical investigations of systems with negative feedbacks which lead to cell-size dependent oscillations. A Fourier analysis in time is used to provide insight into the dependency of oscillation frequency and amplitude on cell size. Depending on the spatial motif, cell size limits for the extinction of oscillatory behavior are obtained.
We start with a linear signaling cascade with different localizations of the membrane-cytosolic interface as shown in Fig 1. We employ a simple cascade model from [9], in which stimulation of a receptor leads to the consecutive activation of several down-stream protein kinases. This model is extended into space in the following. We assume a linear cascade with N components, where the first M < N components are localized at the membrane while the remaining N − M components are assumed to freely diffuse in the cytosol. The equations for the membrane-bound components read
∂ P n ∂ t = D mem Δ Γ P n + v n a - v n d on the cell membrane , for n = 1 , … , M . (1)
Here, P 1 ( x → , t ) , … , P M ( x → , t ) are the local concentrations of signaling molecules on the cell membrane. The activation rate of the first signaling component v 1 a is assumed to be dependent on the input signal, which is denoted by P 0 ( x → , t ). The input signal on the cell surface can be a trigger on the cell membrane or arise from an extracellular signal. All of these species are functions of space and time, where x → is a point on the membrane and t is the time. Diffusion along the cell membrane, which is assumed to be a two-dimensional curved surface in three-dimensional space, is described by the Laplace-Beltrami operator ΔΓ and the diffusion coefficient Dmem. Since the membrane is a surface in three-dimensional space with negligible thickness, the natural unit for concentrations of the cell membrane-bound species Pn (n = 1, …, M), is molecules per area. Molecular concentrations of signaling molecules are frequently provided in nanomolar or micromolar (nM or μM). For convenience, we therefore use the units nanomolar or micromolar times micrometer (nMμm or μMμm) for the membrane-bound signaling molecules. Note that 1 μMμm ≈ 602 molec/μm2.
The phosphorylation rates v n a as well as the dephosphorylation rates v n d have units molecules per area and time. If the input signal is homogeneous in space, meaning P 0 ( x → , t ) = P 0 ( t ), all spatial fluxes Dmem∇ΓPn are zero and the equation system for the membrane-bound species can be described by an equivalent system of ordinary differential equations (S1 Appendix). In contrast to the membrane-bound signaling components P1, …, PM, the signaling component PM+1 can freely diffuse in the cytosol. For the modeling of the membrane-cytosolic interface, we need to include diffusion in the cytosol and reactions on its boundaries, which are the membranes. These processes are modeled by a reaction-diffusion equation
∂ P M + 1 ∂ t = D cyt Δ P M + 1 - v M + 1 d in the cytosol, (2)
with the boundary condition
- D cyt ∇ P M + 1 · n → = v M + 1 a - v i on the cell membrane. (3)
Since PM+1 is activated by the upstream component PM, which is tethered to the membrane, there is a phosphorylation reaction only at the cell membrane but not in the interior of the cytosol. This reaction is, therefore, modeled as a boundary condition. The reactions at the membrane-cytosolic interface are described by the phosphorylation rate v M + 1 a and the inactivation rate vi, both with units molecules per area and time. The species PM+1 diffuses freely in the cytosolic volume with the diffusion rate Dcyt and therefore its local concentration is described in units molecules per volume. The dephosphorylation rate v M + 1 d in the cytosol is given in molecules per volume and time. Note that the inactivation rate and vi can be invoked by membrane-bound phosphatases or saturation of phosphorylation at the membrane. Both, va and vi, comprise the kinetics at the membrane-cytosolic interface. For the flux on all other membrane enclosed organelles we assume a zero-flux condition
- D cyt ∇ P M + 1 · n → = 0 . (4)
The equations for the components of the downstream cytosolic cascade read
∂ P n ∂ t = D cyt Δ P n + v n a - v n d , in the cytosol , for n = M + 2 , … , N . (5)
The concentrations of the cytosolic components at position x → in the cytosolic volume at time t are described by functions P n ( x → , t ) with n = M + 1, …, N. For the cytosolic components we assume zero-flux conditions:
- D cyt ∇ P n · n → = 0 , on the cell membrane, (6) - D cyt ∇ P n · n → = 0 , on the nuclear membrane, for n = M + 2 , … , N . (7)
In classical MAPK cascades the last component of the cascade, which is the phosphorylated MAPK, is imported into the nucleus. Examples range from Hog1 nuclear import in yeast [26, 27] to the import of ERK in mammals [28]. In this case, the boundary condition Eq (7) on the nucleus for the last cytosolic component PN needs to be modified to
- D cyt ∇ P N · n → = - ϵ P N , (8)
where ϵ represents a nuclear-import reaction rate on the nuclear membrane. Unless otherwise stated, a zero-flux boundary condition is assumed on the nucleus throughout this paper.
We will test and compare systems with three components N = 3 as shown in Fig 1, where the spatial arrangement of the components is varied. Here, M = 2 describes the case of two membrane-bound and one cytosolic element (motif Fig 1B) and M = 0 the case of only cytosolic components (motif Fig 1C). In the following the case M = 2 is referred to as mixed membrane-cytosolic (MMC) and M = 0 as pure cytosolic (PC) cascade.
We start this section with an analysis of a spherical cell and then generalize the analysis to arbitrary cell shapes. A spherical cell of radius Rcell with a spherical nucleus of radius Rnuc placed in the center of the cell is assumed in the following.
The input signal is denoted by P0(t) and is assumed to be homogeneous on the cell surface. The concentrations of protein kinases are described by functions Pi(r, t) depending on space and time. Note, since the cellular geometry is radially symmetric and the input signal P0 acts homogeneously on the cell membrane, these functions depend only on the radial distance from the cell center, denoted by r, and time t. In the following analysis, the kinetic rates are linearized, meaning that we assume v n a = α n P n - 1 and v n d = β n P n for the phosphorylation and dephosphorylation, respectively. The inactivation rate vi at the membrane-cytosolic interface is as well linearized by vi = γPM+1. Note that the interface kinetics can be reformulated as v M + 1 a - v i = γ ( α M + 1 γ P M - P M + 1 ), from which it can easily be seen that the activation at the membrane saturates at P M + 1 = α M + 1 γ P M. The model equations for the mixed membrane-cytosolic cascade (MMC) with linearized kinetics read
∂ P 1 ∂ t = D mem Δ Γ P 1 + α 1 P 0 - β 1 P 1 on the membrane , (9) ∂ P 2 ∂ t = D mem Δ Γ P 2 + α 2 P 1 - β 2 P 2 on the membrane , (10) ∂ P 3 ∂ t = D cyt Δ P 3 - β 3 P 3 in the cytosol , (11)
and boundary conditions for the cytosolic species P3:
- D cyt ∇ P 3 · n → = α 3 P 2 - γ P 3 on the membrane , (12) - D cyt ∇ P 3 · n → = - ϵ P 3 at the nucleus . (13)
There are several estimates of phosphatase activity and diffusion coefficients for MAPK signaling components. The diffusion coefficient, Dcyt, of globular cytosolic proteins has been shown to be in the range 1 - 10 μm2s−1, while the diffusion coefficient of membrane-bound components, Dmem, is much lower with a value in the range of 10−3 - 0.1 μm2s−1 [29–31]. The phosphatase rates βn range over three orders of magnitude 0.1 - 100 s−1 [32, 33]. In the case of Fus3, which is the MAPK in the mating pathway of the yeast S. cerevisiae, the diffusion coefficient and cytosolic dephosphorylation rate were estimated to be 4.2 μm2s−1 and 1 s−1, respectively [14]. See Table 1 for an overview on parameter values and units.
We begin with a steady state analysis of this system in the parameter regimes of interest and assume that the signal P0 is constant over time. Here and in the following we indicate the steady distribution with a bar, meaning that P ¯ n denotes the steady state of Pn. The steady state of the first two elements is given by P ¯ 1 = α 1 β 1 P ¯ 0 and P¯2=α1α2β1β2P¯0. For the steady state of P3, the solution is given by
P ¯ 3 ( r ) = A i 0 ( r β 3 D cyt ) + B k 0 ( r β 3 D cyt ) , (14)
where i0 and k0 are modified spherical Bessel functions of the first and second kind, respectively [37]. Note that i0 is increasing with r (distance from the cell center), while k0 is a decreasing function of r. The coefficients A and B are derived in S1 Appendix. If we neglect the nucleus or there is no nucleus in the cytosol, meaning Rnuc = 0, the coefficient B becomes zero. The steady state solution for different cell sizes is shown in Fig 2.
The concentration is maximal at the cell membrane and decays towards the nucleus. An estimate of the decay length Lgradient of the intracellular gradient (with highest concentration at the membrane) is given by L gradient = D cyt / β 3 [32]. This decay length can be compared with the actual cell size. Their ratio is called the Thiele modulus, a dimensionless measure defined as Φ = R cell / L gradient = β 3 R cell 2 / D cyt [33]. For Φ ≫ 1 strong intracellular gradients and concentration heterogeneities of signaling molecules are to be expected, while for Φ ≪ 1 the concentration is almost homogeneous. Since the Thiele modulus relates the diffusion coefficient and degradation rate to cell size, it is an important parameter to investigate gradient formation [33] and signal propagation for several cascade levels [38]. However, in a three-dimensional space Lgradient can not be interpreted as the actual gradient anymore, since its derivation is based on the assumption of a one-dimensional geometry. In addition, if an excluding volume such as the nucleus is assumed, the one-dimensional Lgradient overestimates the concentration gradient. For example, if we assume Dcyt = 4.0 μm2/s and β = 1.0 s−1, we obtain Lgradient = 2.0 μm. In the case of the classical one-dimensional simplification a decay proportional to ∝ exp(−x/Lgradient) is assumed, which suggests a concentration decrease in a distance of x = 2 μm by a factor of exp(−x/Lgradient) ≈ 0.37. However, in a spherical cell with radius Rcell = 3 μm with excluding volume Rnuc = 1 μm, the concentration decreases only by a factor of 0.77 in a distance of 2 μm from the cell membrane. The effect of cell size on intracellular concentration gradients is shown in Fig 2.
The cell size dependence in cell signaling systems does not only arise by the characteristic length scale for intracellular gradient formation, but by the change of average intracellular concentration levels with cell size. We start with the simplifying assumption that there is no nucleus or excluding volume in the cytosol, meaning Rnuc = 0. In this case the steady state solution reads
P ¯ 3 ( r ) = α 3 P ¯ 2 D cyt β 3 i 1 ( Φ ) + γ i 0 ( Φ ) i 0 ( r β 3 D cyt ) . (15)
The modified spherical Bessel functions i0 and i1 are monotonically increasing functions with lim Φ → 0 i 0 ( Φ ) = 1 and lim Φ → 0 i 1 ( Φ ) = 0. We obtain P ¯ 3 ( R cell ) ≈ α 3 P ¯ 3 / γ for cells with small Φ and, therefore, the phosphorylation reaction at the membrane is at saturation in this case. For large Φ, we obtain from lim Φ → ∞ i 1 ( Φ ) / i 0 ( Φ ) = 1 the lower bound P ¯ 3 ( R cell ) ≈ α 3 P ¯ 3 / ( D cyt β 3 + γ ). These estimates also hold in the case of an excluding volume which is the nucleus and we obtain the estimate for the concentration P ¯ 3 ( R cell ) at the cell membrane:
α 3 P ¯ 2 D cyt β 3 + γ ≤ P ¯ 3 ( R cell ) ≤ α 3 P ¯ 2 γ . (16)
The dependence of absolute concentration levels on the membrane-cytosolic interface is shown in Figs 2 and 3 for a set of different parameters. For a large inactivation rate at the membrane-cytosolic interface γ > D cyt β 3, the cell size dependence decreases. Therefore, cell size dependence is mainly determined by γ and D cyt β 3 but is independent of the phosphorylation rate α.
We can further investigate the evolution of the average concentration levels, which depends on the concentration at the cell membrane and the strength of the intracellular gradient. In case of arbitrary cell shapes with cell volume Vcell and cell membrane area Mcell, the average concentration is obtained from
P m avg = 1 | M cell | ∫ M cell P m d A for 1 ≤ m ≤ M , (membrane-bound components), P n avg = 1 | V cell | ∫ V cell P n d V for M + 1 ≤ n ≤ N , (cytosolic components). (17)
In [33], analytical solutions for the average concentration in a spherical cell and a slab have been derived as functions of the Thiele modulus. However, since the derivations are restricted to cell geometries, where an explicit analytical solution of the reaction diffusion equation is available, we introduce an alternative approach to estimate average concentration levels depending on the cytosolic volume and cell membrane area. Since the signal propagates from the cell membrane to the cytosol, the cell membrane can be regarded as a source, while the cytosolic volume, where phosphorylated signaling molecules are dephosphorylated, can be regarded as a sink. This idea can be derived mathematically by integration of Eq (2) and application of Green’s theorem, which results in
α M + 1 | M cell | P ¯ M avg - γ ∫ M cell P ¯ M + 1 d A = β M + 1 | V cell | P ¯ M + 1 avg . (18)
Here, the production on the left hand side of the equation depends on the cell membrane area, which is balanced by the degradation in the cytosol on the right hand side of the equation. On the basis of the equation of mass conservation (see S1 Appendix) in reaction diffusion systems, we introduce the following measure:
Λ M + 1 = α M + 1 | M cell | P ¯ M avg γ | M cell | + β M + 1 | V cell | . (19)
This measure has the property Λ M + 1 = P ¯ M + 1 avg for γ = 0 or β = 0, which holds for arbitrary cell shapes. Furthermore, for a spherical cell the estimate P ¯ M + 1 avg ≤ Λ M + 1 ≤ P ¯ M + 1 max holds for β > 0 and γ > 0 (see S1 Appendix and Fig 3). Therefore, we use ΛM+1 as a proxy for the average concentration for arbitrary cell shapes, which can be easily calculated.
A comparison of the estimate Λ3 to the average concentration is shown in Fig 3 for different cell shapes, which are a spherical cell, a rod shaped cell, a cell with one protrusion and a cell with two protrusions. These cell shapes occur for example in S. cerevisiae, S. pombe and haploid S. cerevisiae stimulated with mating pheromone [39, 40]. The MMC cascade was simulated for these shapes with varying cell size. The measure Λ3 is an exact predictor for the average concentration in the case γ = 0 for all cell shapes and slightly overestimates the average concentration for γ = 1 μm s−1.
Furthermore, we investigated the concentration differences of P ¯ 3 between cell membrane and nucleus and compared them to the average concentration in the cytosol. For the spherical cell the average concentration levels of P ¯ 3 in the cytosol as well as on the membrane were the lowest, which is expected since the surface to volume ratio is the lowest among all shapes. The concentration differences between membrane and nucleus were the highest for the spherical cell and the cell with two protrusions and the average concentration at the nucleus decreased almost to zero for large cells. For the rod shape cell the concentration differences were the smallest, since the distance along the short axis is small and the concentration does not drop as sharply as for the other cell shapes.
We furthermore established a correspondence to the evolution of the average concentration levels in time. In the case γ = 0 and for arbitrary cell shapes, the average concentration levels follow the system of ordinary differential equations
d P 1 avg ( t ) d t = α 1 P 0 avg ( t ) - β 1 P 1 avg ( t ) , (20) d P 2 avg ( t ) d t = α 2 P 1 avg ( t ) - β 2 P 2 avg ( t ) , (21) d P 3 avg ( t ) d t = | M cell | | V cell | α 3 P 2 avg ( t ) - β 3 P 3 avg ( t ) , (22)
where P 1 avg and P 2 avg are the average concentration levels in molecules per cell membrane area. This system of ordinary equations can be obtained by integrating Eqs (9)–(13) over their respective spatial domains. See S1 Appendix for details of the derivation. The steady state for the average concentration of P ¯ 3 is given by
P ¯ 3 avg = | M cell | | V cell | α 1 α 2 α 3 β 1 β 2 β 3 P ¯ 0 avg . (23)
Therefore, the average concentration level scales with the ratio of membrane area to cytosolic volume which is given by | M cell | | V cell |. The effective global phosphorylation rate for the average concentration of active signaling molecules in the cytosol is therefore determined by α ˜ 3 = | M cell | | V cell | α 3. These relations give us a correspondence between widespread used ordinary differential equations and the bulk-surface partial differential equations employed in this paper. In summary, we have strong cell size dependence, with decreasing concentrations for larger cells.
In the following we consider a pure cytosolic (PC) cascade with three elements, in which all elements diffuse freely through the cytosol. The reaction-diffusion system is given by
∂ P 1 ∂ t = D cyt Δ P 1 - β 1 P 1 in the cytosol , (24) ∂ P 2 ∂ t = D cyt Δ P 2 + α 2 P 1 - β 2 P 2 in the cytosol , (25) ∂ P 2 ∂ t = D cyt Δ P 3 + α 3 P 2 - β 3 P 3 in the cytosol (26)
with boundary conditions on the membrane
- D cyt ∇ P 1 · n → = α 1 P 0 - γ P 1 , (27) - D cyt ∇ P 2 · n → = - D cyt ∇ P 3 · n → = 0 , (28)
and at the nucleus
- D cyt ∇ P 1 · n → = - D cyt ∇ P 2 · n → = 0 , (29) - D cyt ∇ P 3 · n → = - ϵ P 3 . (30)
Note that the membrane-cytosolic interface occurs at the first cascade level, meaning that only P1 is activated at the membrane with rate α1 P0 − γP1. In the special case of β1 = β2 = β3 = β analytical approximations to cytosolic cascades in a one-dimensional system have been derived in [41, 42]. While a one-dimensional cellular geometry can be used to study gradient formation qualitatively, spatial effects such as the cell surface to volume ratio are neglected. Therefore, we derived exact analytical solutions to the linear system in three dimensions. The steady state solutions for P ¯ n ( r ) are expanded as follows
P ¯ n ( r ) = ∑ k = 1 n A n , k r k - 2 exp ( β D cyt r ) + ∑ k = 1 n B n , k r k - 2 exp ( - β D cyt r ) . (31)
The algebraic expressions of the coefficients An, k and Bn, k and their derivation are given in the S1 Appendix. In comparison to the MMC cascade, which was discussed in the previous section, the third cascade element P3 is more evenly distributed in the cell and concentration gradients are much more shallow (see Fig 2).
In order to quantify the concentration differences in a cell of arbitrary shape, we measure the concentration variance in the cell. Therefore, we introduce the variance as
Σ n 2 = ∫ V cell ( P ¯ n - P ¯ n avg ) 2 d V (Variance). (32)
This measure has a close correspondence to the variance in image analysis [43, 44]. In contrast to image analysis, where the square of the deviation of the fluorescence intensity from the average fluorescence intensity of a marker is integrated pixel-wise, the integration here is continuous. As for the analog in image analysis, the normalized variance is calculated as a measure for the deviation from the average [44, 45]. While this measure is frequently used in auto-focus algorithms in image analysis, we suggest the normalized variance as a measure for the degree of localization of signaling molecules within a cell. An estimate the propagation of the normalized variance in the cytosolic cascade is given by
Σ n 2 P ¯ n avg | V cell | ≤ C n Σ n - 1 2 P ¯ n - 1 avg | V cell | , with C n = α n β n ( D cyt C s 2 d 2 + β n ) 2 . (33)
Here, Cs is a constant depending on cell shape and d is the cell diameter. Note that C = Csd is the Poincaré constant from the well known Poincaré inequality [46]. For convex cell shapes the estimate holds for C s = 1 π. In the case of a convex cell shape of a small cell as yeast (without protrusion), we therefore have the estimate Cn ≈ 0.3 ≤ 1 for Dcyt = 3 μm2/s, αn = βn = 1 s−1 and a cell diameter of d = 6 μm. For this parameter set, the normalized variance decreases at least by 70% at the second cytosolic cascade level and by 90% at the third cytosolic cascade level (compared to the first cascade element). In general, for Cn < 1, the normalized variance of the intracellular concentrations decreases with increasing cascade level and concentration differences in the cell are balanced out (see Fig 3).
Similar to the previous section, we derived an estimate for the average concentration. In this case, we employed the estimate Λ1 to P ¯ 1 avg, since P1 is the cascade element that is activated at the membrane-cytosolic interface for the PC cascade. The average concentration P ¯ 3 avg is related to P ¯ 1 avg by P ¯ 3 avg = α 2 α 3 β 2 β 3 P ¯ 1 avg and, therefore, we used the approximation α 2 α 3 β 2 β 3 Λ 1 for P ¯ 3 avg. As in the case of the MMC, this approximation is exact for γ = 0 and overestimates the average concentration sligthly for γ = 1 μm s−1 (see Fig 3).
Exact expressions for the steady states of the average concentration of signaling components in the case γ = 0 and for arbitrary cell shapes are given by
P ¯ 1 avg = | M cell | | V cell | α 1 β 1 P ¯ 0 avg , P ¯ 2 avg = | M cell | | V cell | α 1 α 2 β 1 β 2 P ¯ 0 avg , P ¯ 3 avg = | M cell | | V cell | α 1 α 2 α 3 β 1 β 2 β 3 P ¯ 0 avg . (34)
Therefore, the average concentration of the third cascade element P ¯ 3 avg takes the same values in the MMC and PC cascade. The major distinction of both spatial motifs is given by the fact that the concentration differences obtained at the cell membrane and nucleus are larger in the MMC cascade than in the PC cascade. Similarly as in the previous section, we can formulate a system of ordinary differential equations for the time evolution of average concentrations
d P 1 avg ( t ) d t = | M cell | | V cell | α 1 P 0 avg ( t ) - β 1 P 1 avg ( t ) , (35) d P 2 avg ( t ) d t = α 2 P 1 avg ( t ) - β 2 P 2 avg ( t ) , (36) d P 3 avg ( t ) d t = α 3 P 2 avg ( t ) - β 3 P 3 avg ( t ) . (37)
The dependence of absolute concentration levels on the membrane-cytosolic interface is shown in Fig 2 and figure in S1 Appendix for a set of different parameters.
Time-resolved image-based analysis has shown that MAPK signaling pathways respond with a measurable signal in the nucleus in time scales of seconds to a few minutes. The Hog1 pathway response (phosphorylated MAPK) in budding yeast is at about 80% of its maximal activity within a minute [26]. Another example is the Src activation/deactivation cycle, where oscillations and pulses take place in the regime of seconds [47]. The timing of signal transduction in linear signaling cascades for well-stirred homogeneous systems has been analyzed in [9]. They concluded for weakly activated signaling cascades that phosphatases have a more pronounced effect than kinases on the rate and duration of signaling, whereas signal amplitude is controlled primarily by kinases. A thorough analysis of linear models assuming a homogeneous distribution of signaling molecules for different kinds of external stimuli has been recently worked out in [12]. We extended and compared these findings to spatial signal transduction omitting the simplification of homogeneous concentrations. How long does it take to establish an intracellular concentration gradient? How does diffusion change the timing of signal transmission from the cell membrane to nucleus? Which concentration differences are expected until a steady state is established? The time-scale analysis for spatial models is more difficult than for models based on a will-mixed assumptions due to high computational costs. Therefore, we used the recently introduced measure of accumulation times [25, 48]. The approach of Pn(r, t) to its steady state P ¯ n ( r ) at radial distance r from the cell center and time t can be characterized using the local relaxation function
ρ n ( r , t ) = ( P ¯ n ( r ) - P n ( r , t ) ) / P ¯ n ( r ) . (38)
The difference ρn(r, t1) − ρn(r, t2) can be interpreted as the fraction of the steady state level P ¯ n ( r ) that accumulated in the time interval [t1, t2]. In an infinitesimal time interval [t, t + dt] the fraction of accumulated activated signaling molecules at steady state is given by - ∂ ρ n ( r , t ) ∂ t d t. The local accumulation time is defined as [25]
τ n ( r ) = - ∫ 0 ∞ t ∂ ρ n ( r , t ) ∂ t d t .
The accumulation time can be derived from the steady state solution even if no closed form of the time-dependent solution is known [25].
The timing of the average concentrations given in the system of ordinary differential equations for the MMC cascade (20)–(22) and the PC cascade (35)–(37) are the same and can be analytically expressed as
τ 3 = 1 β 1 + 1 β 2 + 1 β 3 . (39)
This expression also coincides with signaling times calculated by Heinrich et al. [9]. However, for the spatial model the local accumulation times at the membrane and nucleus differ. The accumulation is generally faster at the membrane and slower at the nucleus, where the degree of the difference increases with cell size (see Fig 4). Furthermore, the two spatial motifs show significant differences. For the MMC cascade the accumulation time for the second element P2 is exactly 1 β 1 + 1 β 2 on the membrane, while it is shorter for the cytosolic species (compare Fig 4).
The accumulation time of P3 at the nucleus is, as expected, much longer. For small cells the intracellular concentration is spatially homogeneous and the approximation 1 β 1 + 1 β 2 + 1 β 3 holds, while the time for signal propagation to the nucleus increases with cell size. An analytical solution of the accumulation times for P3 for the MMC cascade and the special case of Rnuc = 0 can be derived [49], which is given in the S1 Appendix. However, for larger cells, the time for signal propagation to the nucleus increases with cell size. For the PC cascade, the increase in accumulation time at the nucleus with cell size is less pronounced than for the mixed-membrane cytosolic cascade.
While a constant stimulus was applied to calculate the accumulation times, we also tested a decaying signal P 0 ( t ) = P 0 max exp ( - λ t ), with P 0 max = 100 n M μ m and solved the time-dependent system numerically. A comparison of the MMC and PC is shown in Fig 4. Interestingly, the concentration level at the membrane for the PC cascade decreases from the first cascade species P1 to the second cascade species P2 and than increases again from the second cascade species P2 to the third cascade species P3, while there is an increase from the preceding cascade species to the next cascade species at the nucleus. This phenomenon is caused by the concentration differences from cell membrane to nucleus, which is larger for P1 than for P2 in the PC cascade. Note that the parameters were chosen to be α n β n = 2, which means a twofold increase for the average concentration levels from one signaling cascade element to the next. Therefore, the spatial system can behave entirely different than the homogeneous system. The accumulation time at the membrane was much faster for γ = 1 μm/s than for γ = 0 and changed only slightly with cell size. However, for larger cells the accumulation time of the signal at the nucleus was almost independent of γ. Therefore, the difference of accumulation times at the membrane and nucleus increased with γ (also compare figure in S1 Appendix). In case of the MMC cascade the accumulation time at the nucleus for a cell with Rcell = 12 μm almost doubled compared to a small cell with Rcell = 2 μm, while for the PC cascade the increase of accumulation time with cell size was less pronounced.
For calculation of higher moments of the time scaling and the special case of a cell without nucleus we refer to [49]. An analysis for time scaling of a linear cascade in one spatial dimension with four elements including higher moments has been carried out in [50].
In the following we analyze signal transduction of heterogeneous external signals. For example, in cultures of mixed haploid yeast cell populations [40] as well as in microfluidic devices [51], the external pheromone signal, which triggers a MAPK cascade, is not homogeneously distributed but forms gradients in the extracellular medium. The activated signaling cascade is spatially localized and triggers subsequent directed growth in S. cerevisiae [14] as well as S. pombe [15]. Furthermore, properties of protein-protein interactions and morphological changes can be tightly connected [52].
Therefore, we investigate the signal transduction in response to an external heterogeneous signal for same cell shapes as in Fig 3, which were a spherical cell, a rod shaped cell, a cell with one protrusion and a cell with two protrusions. These cell shapes occur for example in S. cerevisiae, S. pombe and during their response to stimulation with mating pheromone [53].
We tested the linear signaling cascade with a graded stimulus of the form
P 0 ( x → ) = P 0 sig [ 1 + P 0 slope ( x 1 - x 1 mid ) ] , x → = ( x 1 , x 2 , x 3 ) , (40)
where P 0 sig and P 0 slope are constants describing the basal signal strength and the slope of the signal, respectively. Here, we chose the origin of coordinates to be in the center of the cell and, therefore, x → mid = ( 0 , 0 , 0 ). In this way, we obtain an input signal gradient which increases linearly in x1-direction for P 0 slope > 0 and decreases linearly for P 0 slope < 0. The concentration at x 1 mid is given by P 0 sig.
We tested the influence of asymmetries in cell shape in response to the graded stimulus Eq (40) and investigated the spatial distribution of the last signaling component of the MMC and PC cascade, which is P3. In Fig 5(A) and 5(C), the concentration profile of P ¯ 3 on the cell membrane as well along a slice through the cell in response to a homogeneous signal is shown as control. Since the spherical cell is radially symmetric, no gradient was induced on the membrane. For the rod shaped cell, we observed a shallow gradient on the cell surface with higher concentration at the poles, the intracellular concentration profile exhibited two areas of low concentration, which were separated by the nucleus. This effect was more pronounced for the MMC cascade. For the PC cascade, the concentration was almost homogeneously distributed. For the asymmetric cell shapes with one and two protrusions, a gradient from the distal end (front) to the spherical part (back) was established in response to a homogeneous input signal. The protrusion, therefore, can be compared to a pocket in which higher concentrations of cytosolic signaling molecules are established. Mathematically this effect can be explained by the geometry dependence of the eigenfunctions of the Laplace operator [54], which can be employed to characterize the solution of the reaction-diffusion equations for a certain cell geometry. Therefore, these asymmetric cell shapes can already induce a gradient of signaling molecules from front to back.
In Fig 5(B) and 5(D), the responses to a signal with P 0 slope = 0 . 03 μ m - 1, which is increasing in x1-direction, and a signal with P 0 slope = - 0 . 03 μ m - 1, which is decreasing in x1-direction, were simulated and opposed to the response to a spatially homogeneous signal with P 0 slope = 0. To measure the response, we define the gradient of the n-th cascade element naturally as the difference of concentrations at two points over the euclidean distance of these two points. In the case of the kinase concentrations, the gradient was computed from ( P n ( x → front ) - P n ( x → back ) ) / | x → front - x → back |. Here, x → front and x → back are the extreme points in x1-direction on the cell membrane or nucleus. Both motifs, the MMC and PC cascade, behave differently in the transduction of signal gradients. The gradient of the third cascade level P3 along the cell membrane and the nucleus decreased for the MMC cascade with cell size for all shapes. For the PC, the gradient increased with cell size to a maximum value and then decreased for larger cell sizes, which suggests an optimal cell size for gradient detection and transmission. This effect was expected, since for small cells the concentration was almost homogeneous in the cytosol and concentration differences were balanced by diffusion. However, with increasing cell size the average concentration level decreased in the cell and at the nucleus. As a consequence, also the absolute gradient decreased.
The rod shaped cell showed a stronger response than the spherical cell shape, since concentrations were higher at the poles and the cell was aligned along the gradient. Furthermore, the compartmentalization induced by the nucleus in the thin rod shaped cell had a pronounced effect on the P3 gradient, since diffusion in the cytosol from front to back was hindered. For the cells with one and two protrusions the gradient of P3 was strongly biased with an increase in the direction of the protrusions. Note that both motifs behave differently for the transmission of the gradient to the nucleus. While for the MMC cascade, the shape dependence was more pronounced and the gradient in the cell interior was almost decoupled from the gradient on the membrane for the asymmetric cell shapes, the PC cascade transmitted the gradient more reliably into the cell interior and the nucleus.
In summary, we observed a strong influence of cell size on localization and establishment of gradients by signaling cascade elements. For the cell with a protrusion the concentration of P3 was higher in the protrusion than in the opposite distal end, which is the spherical part of the cell. This effect emerged due to a higher local surface to volume ratio in the protrusion region. Therefore, a larger portion of cytosolic signaling molecules, which diffuse freely in the cytosol, is phosphorylated in the protrusion part leading to a gradient from the protrusion tip to the opposite distal end of the cell. The influence of cellular asymmetries has also been investigated in [23] for gradients of the small Rho-GTPase Cdc42 during cell polarization. However, this system reacts in the opposite way, since the flux of molecules during the establishment of a polarity site is directed from the cytosol onto the membrane and, therefore, a gradient from the distal end to the protrusion is established.
These effects occur due to the different architectures of both systems. In the PC and MMC signaling cascades, we have signal transduction from the membrane to the nucleus and, therefore, a diffusive flux of activated signaling molecules from the membrane into the cytosol, while in the polarization system the flux of signaling molecules during the establishment of a polarity site is directed from the cytosol onto the membrane, which is the opposite direction. Therefore, both system respond differently to cellular asymmetries with respect to gradient formation. This interplay of both systems is especially interesting, since in many organisms a polarization system is interacting with a MAPK cascade [55, 56] and might, therefore, precisely control cell shape and size.
For spherical cell shapes we furthermore investigated more complex external signal gradients, meaning heterogeneities with multiple maxima and minima (see S1 Appendix). As in [18, 57], a heterogeneous signal on a sphere can be decomposed using spherical harmonics
P 0 ( θ , ϕ , t ) = ∑ l = 0 ∞ ∑ m = - l l A 0 , l m ( t ) Y l m ( θ , ϕ ) , (41) A 0 , l m ( t ) = ∫ 0 2 π ∫ 0 π P 0 ( θ , ϕ , t ) Y l m * ( θ , ϕ ) sin ( θ ) d θ d ϕ . (42)
In this decomposition the amplitudes of higher order, where the order is denoted by l, are generally more strongly damped than gradients or spatial heterogeneities of lower order. In this manner, the results shown here can be extended to complex spatial signals on the cell surface. We provide full analytical solutions for the MMC and PC for a sphere with excluding nucleus (see S1 Appendix).
In this section, we analyze the influence of cell size on signal transduction for an oscillating cascade consisting of two membrane-bound and one cytosolic member (MMC) and a cascade of three cytosolic elements (PC), meaning for M = 2 and M = 0, respectively. The case of a negative feedback and a constant homogeneous signal is investigated in the following.
Negative feedbacks are a frequent regulation element in signaling cascades and can be induced by the dephosphorylation of upstream components by the MAPK or phosphatases [39, 58–61]. Examples are given by Tyr phosphatases, which can induce a negative feedback [47] and dual specificity phosphatases (DUSPs) [59]. Some negative feedbacks, as for instance induced in the Src-Tyr cycle, lead to oscillations on the time scale of seconds [47], while others act on much longer time scales. For instance, during the yeast pheromone response the MAPK Fus3 undergoes sustained oscillations in the range of 2-3 hours, which control the periodic formation of mating projections. In this process Sst2 acts as a negative regulator of the G-Protein signaling at the membrane, while deactivation in the cytosol is mediated by the MAPK phosphatase Msg5 [39]. A classical and most simple example of an oscillator with negative feedback and non-linear reaction terms is the Goodwin oscillatory system [62, 63].
We adapt the mentioned, modified system and formulate the problem using partial differential equations by adding a diffusion term and formulating the boundary conditions accordingly to the models mentioned before. The phosphorylation and dephosphorylation rates for both models read as
v 1 a = P 0 1 + ( P 3 / K m ) p , v 1 d = β 1 P 1 , (43) v 2 a = β 2 P 1 , v 2 d = β 2 P 2 , (44) v 3 a = β 3 P 2 , v 3 d = β 3 P 3 , (45)
according to Eqs (1)–(5), respectively. The activation rate v 1 a contains the negative feedback, since a high concentration of P3 leads to a lower activation of P1. We assume a constant external signal P 0 ( x → , t ) = 100 n M μ m.
For the first model (MMC), the deactivation with rate v 3 d takes place in the cytosol, whereas the activation occurs on the membrane and is therefore modeled as a boundary condition with vi = 0 according to Eqs (2) and (3), as P3 is a solely cytosolic species. We assume zero-flux conditions for P3 on the nucleus, meaning that we set the nuclear-import reaction rate ϵ = 0 (compare Eq (8)).
For the second model (PC), all species are solely cytosolic, hence the activation rate v 1 a for P1 is a boundary condition describing the activation of P1 on the membrane. We assume a zero-flux condition for P3 on the nucleus (ϵ = 0) and for P1 and P2 on both nucleus and membrane, meaning the whole boundary.
Both models contain non-linear kinetics as well as negative feedbacks, resulting in oscillations. Furthermore, in both models the activation rate for P1 depends on a parameter p > 0. It is shown, e.g. in [64], that for p > 8 the ODE system consisting of three species destabilizes, and that for longer cascades, i.e. for larger N, the system becomes instable for even lower values of p > 1.
Since an analytical solution is unknown for both models, numerical methods have to be employed to solve the systems. For simplicity reasons and due to the high computational overhead we solved the systems in two dimensions, using a disc to model the cell. We used a fixed-point scheme to solve the non-linear equations.
We chose the parameters β1 = β2 = β3 = 0.125 s−1, Dcyt = 1 μm2s−1, Dmem = 0.01 μm2s−1, Rnuc = 1 μm and Rcell = 2 μm. The initial conditions were P1 = P2 = 10 nMμm and P3 = 10 nM for the MMC cascade and P1 = P2 = P3 = 10 nM for the PC cascade. For Km = 100 nM and the feedback strength p = 10, both models oscillated as expected, Fig 6A.
Therefore, in the case of a relatively small cell size of Rcell = 2 μm, both spatial models behave similarly to the original model, which was formulated as a system of ordinary differential equations. An analysis of the oscillation frequencies and the mean concentration can be seen in Fig 6B. Based on previous experiments and plots, the frequency analysis was conducted after t = 200 s, when the frequency and corresponding amplitudes of the mean concentrations for both models can be assumed to be constant. Both models show a very similar behavior, whereas the frequency and mean concentration are higher for the pure cytosolic model, as can be explained by the fact that the reactions do not only occur at the membrane, but everywhere in the cytosol.
In a next step we varied the cell size 1.5 μm ≤ Rcell ≤ 4.0 μm and again conducted an analysis of the frequencies and corresponding amplitudes for both models. The results of the analysis for the third component of both models are plotted in Fig 6C. As pointed out before, the frequency for both models is very similar, but the amplitude of the signal is higher for the cytosolic model.
Oscillations in the first, mixed cascade model only occur for a cell size up to Rcell ≤ 2.5 μm, and in the second, pure cytosolic model for a cell size up to Rcell ≤ 3.0 μm. The inital oscillations die down fast and both models converge against a steady state if the cell size is chosen bigger.
Stimulated by the progress in cell imaging and the increasing need to understand intracellular dynamics, we investigated and discussed a general approach of modeling cellular signal transduction in time and space. Signaling cascades of covalent protein modifications, such as mitogen-activated protein-kinase (MAPK) cascades and small GTPase cascades, occur in a plethora of variations [1, 13, 65]. The first signal component can be activated at the cell membrane by a membrane-bound enzyme such as a kinase or a guanine nucleotide exchange factor in GTPase signaling, while deactivation can occur at the membrane or in the cytosol, for instance, mediated by a phosphatase or GTPase activating protein [66]. Therefore, activation and deactivation can be spatially separated, which creates a number of different spatial arrangements and combinations in signal transduction.
We investigated signaling cascades with different spatial arrangements of signaling components. We showed that modeling of the membrane-cytosolic interface is crucial as well as the ratio of membrane area and cytosolic volume, which are both spatial properties. The results imply strong cell size and shape dependence of signal transduction within cells, which are likely to contribute to single cell variation in response to extracellular stimuli. We suggest that cells measure the cell membrane to cell volume ratio to coordinate growth and differentiation. For asymmetric cell shapes also local changes in cell volume to cell membrane ratio becomes important for intracellular signaling. Widely used time-dependent models of ordinary differential equations can naturally be extended into space by using bulk-surface differential equations. Applying this extension to a class of linear signal transduction models, we compared the assumption of a well mixed cell with two different spatial signal transduction motifs. We derived and discussed criteria that can be used to test the well-mixed assumption and showed that kinetics that connect membrane-bound species with cytosolic species naturally cause size dependence. The results are, therefore, of general importance for kinetic models of signal transduction.
Our findings have relevant biological implications. Since the signals transduced by linear signaling cascades from the cell membrane to the nucleus decrease exponentially on a length scale of a few microns, our theoretical findings suggest a strong cell size dependence in response to extracellular stimuli. Furthermore, the global cell volume to cell membrane area is important for average concentration levels. Mating projections as they occur in yeast act as pockets for signaling molecules, which can support biochemical feedbacks. Adaptations as lamellipodia in keratocytes or invaginations such as T-tubuli in myocytes can locally increase the accumulation of signaling molecules. These cellular structures are able to directly provide a feedback on signaling.
We suggest the normalized variance as a measure to quantify concentration differences and localization of signaling molecules, which can be obtained from spatially resolved microscopy data additionally to mean intensity levels of a fluorescence marker. For example, it would be enlightening to measure average concentration and normalized variance together with cell size and morphology. Interesting studies of the response in cell populations often lack the response behavior attributed to cell size and morphology. Examples range from the switch-like behavior in populations of oocytes [67] to the pheromone response in yeast cells [68, 69]. Therefore, single cell data where the cell size is assigned to these measurements is needed for a faithful quantitative investigation of the pathway, to disentangle biochemical properties of protein-protein interactions and morphological properties such as size and shape of whole cells. Targeting signaling proteins by lipidation modifications such as palmitoylation, prenylation or myristoylation [2, 3, 5] could change the sequestration of a signaling cascade from a pure cytosolic (PC) cascade to a mixed-membrane bound (MMC) cascade. In the case of the mixed-membrane MMC the geometry and size dependence is more pronounced, since the first signaling elements are tethered to the membrane. In contrast, for the investigated PC cascade, localization and strong intracellular gradients are reduced, but depending on the kinetic parameters, the geometric information can also be better transmitted through the whole cell.
In non-linear signaling systems, the differences that we observed in the linear signaling cascade models are likely to be amplified. Non-linear kinetics can amplify gradient formation, which leads to even stronger intracellular concentration differences [70]. This also holds for absolute concentration levels that can behave in a switch-like manner depending on the kinetics [67, 71]. Furthermore, higher order kinetics can amplify the accumulation time differences in different cellular locations [72], which can lead to spatial oscillations and phosphoprotein waves.
The analysis of the signaling cascade model can be extended to more complex spatial heterogeneities for example by using the Laplace series as suggested in [18, 57]. With this approach localized signals arising from membrane structures like lipid rafts, septins or co-localization due to protein-protein interactions can be represented. Since these are often precursors for cell shape and organelle structures, the interplay with cell shape and morphology needs to be addressed by future research. The intrinsic geometry dependence of signaling systems has recently been shown for ellipsoidal cell shapes in the MinE-MinD system [24, 73, 74], but also in the yeast system [23, 75–77]. Recent developments of mathematical methods such as the finite element method for bulk-surface equations [19, 20] as well as stability analysis techniques of these systems [23, 78–82] are expected to provide further insight in the behavior of cellular signal transduction.
We used the finite-element software FEniCS [83, 84] to solve the arising partial differential equations in the Python programming language. The meshes were generated using the computational geometry algorithms library (CGAL) [85]. The non-linear equations were solved using a fixed-point scheme.
|
10.1371/journal.ppat.1007209 | A fully-virulent retargeted oncolytic HSV armed with IL-12 elicits local immunity and vaccine therapy towards distant tumors | Oncolytic herpes simplex viruses (oHSVs) showed efficacy in clinical trials and practice. Most of them gain cancer-specificity from deletions/mutations in genes that counteract the host response, and grow selectively in cancer cells defective in anti-viral response. Because of the deletions/mutations, they are frequently attenuated or over-attenuated. We developed next-generation oHSVs, which carry no deletion/mutation, gain cancer-specificity from specific retargeting to tumor cell receptors—e.g. HER2 (human epidermal growth factor receptor 2)—hence are fully-virulent in the targeted cancer cells. The type of immunotherapy they elicit was not predictable, since non-attenuated HSVs induce and then dampen the innate response, whereas deleted/attenuated viruses fail to contrast it, and since the retargeted oHSVs replicate efficiently in tumor cells, but spare other cells in the tumor. We report on the first efficacy study of HER2-retargeted, fully-virulent oHSVs in immunocompetent mice. Their safety profile was very high. Both the unarmed R-LM113 and the IL-12-armed R-115 inhibited the growth of the primary HER2-Lewis lung carcinoma-1 (HER2-LLC1) tumor, R-115 being constantly more efficacious. All the mice that did not die because of the primary treated tumors, were protected from the growth of contralateral untreated tumors. The long-term survivors were protected from a second contralateral tumor, providing additional evidence for an abscopal immunotherapeutic effect. Analysis of the local response highlighted that particularly R-115 unleashed the immunosuppressive tumor microenvironment, i.e. induced immunomodulatory cytokines, including IFNγ, T-bet which promoted Th1 polarization. Some of the tumor infiltrating cells, e.g. CD4+, CD335+ cells were increased in the tumors of all responders mice, irrespective of which virus was employed, whereas CD8+, Foxp3+, CD141+ were increased and CD11b+ cells were decreased preferentially in R-115-treated mice. The durable response included a breakage of tolerance towards both HER2 and the wt tumor cells, and underscored a systemic immunotherapeutic vaccine response.
| There is increasing interest in oncolytic viruses (OVs), following the approval of OncovexGM-CSF, and the success of a number of them in clinical trials. Most OVs, particularly the oHSVs, are attenuated to varying degree. In contrast, the tropism-retargeted oHSVs are fully-virulent, highly effective oncolytic agents, and appear to be highly safe in mice. Up to now, it was unknown how efficacious the retargeted oHSVs are as immunotherapeutic agents. Here, the demonstration that they elicit local immune response and systemic therapy vaccine effects opens the possibility that the fully-virulent retargeted oHSVs may be highly efficacious oncolytic-immunotherapeutic agents.
| Oncolytic viruses (OVs) meet the need for novel anticancer agents characterized by low toxicity and low negative impact on the quality of life of patients [1–4]. Oncolytic herpes simplex viruses (oHSVs) stand for their efficacy in a number of clinical applications [5,6]. The most successful oHSV, OncovexGM-CSF, was approved against metastatic melanoma [7,8]. The Clinical trials.gov website lists 22 open or recently completed trials with oHSVs [9–13]. Much of the current interest in OVs stems from their immunotherapeutic properties. Thus, oHSVs, and OVs in general, boost the immune response to the tumor, exert a therapeutic vaccine effect with no requirement for the identification of the tumor-specific or patient-specific neoantigens [14–17]. In combination with checkpoint inhibitors (CPIs), they enhance the efficacy of the blockade therapy [18–21]. They can be engineered to express anti-checkpoint antibodies [22].
The oHSVs in clinical practice or trials are attenuated to varying degrees, and gain their cancer specificity from the attenuation [3,5,23,24]. In essence, safety was achieved at the expense of virulence. The attenuated oHSVs infect preferentially, but not exclusively, the cancer cells. Attenuation was attained through genetic engineering, as in the Δγ134.5 viruses, including OncovexGM-CSF, or through natural mutations [23,25,26]. In some examples, multiple deletions resulted in high attenuation, to the point that oHSVs replicated to limited extent even in the tumor cells [27] and were scarcely efficacious as single agents. In OncovexGM-CSF, the attenuation is reversed by the immediate early expression of US11, a viral protein which counteracts protein kinase R [28]. To circumvent the attenuation effects, OVs are employed as vectors for the transgenic expression of cytokines or CPIs. Indeed, the first cytokine-expressing oHSV was designed by Martuza and Rabkin some 20 years ago [29]. OncovexGM-CSF is armed with GM-CSF, which activates APCs, boosts the immune response to the tumor, and enables a distant effect [23]. A key modulator of the cancer immune response is IL-12. This cytokine targets a variety of immune cells, activates effector cells, induces IFNγ secretion which boosts and sustains the immune response [30–32]. In humans, the systemic administration of IL-12 was marred by toxicity. The expression of IL-12 from OVs, in particular oHSVs, raised the hopes to benefit from local administration, without the toll of systemic toxicity. The IL-12-armed Δγ134.5 oHSVs showed efficacy in preclinical models [19,29,33–37], and one of them is in clinical trial against glioblastoma multiforme [38].
An alternative approach to safety-through-attenuation centres on specific tropism for the cancer cells, achieved by retargeting the virus tropism to cancer-specific receptors of choice, and detargeting from the natural receptors [39–41]. The retargeted oHSVs carry no deletion/attenuation. In their target cancer cells they are fully virulent. Because they infect no other cell than the specifically-targeted cancer cells, they promise to be highly safe. When injected intraperitoneally, they caused no harm to tumor-free mice up to the maximum tested dosage (108 PFU) [42]. The tumor cell receptor selected in our laboratory was HER2 (human epidermal growth factor receptor 2) [43–48] overexpressed in a number of cancers [49]. The HER2-retargeted oHSVs named R-LM249 and R-LM113 exerted a strong therapeutic efficacy in immunodeficient mice [42,50–52]. A single virus administration practically ablated tumor growth [42]. When administered intraperitoneally (i.p.) in a model of peritoneal carcinomatosis they rendered 60% of mice tumor-free [51]. The immunodeficient mouse model underscores the therapeutic effect against primary tumors, accounts for the oncolytic effect of the virus, but is inadequate to evaluate the immunotherapeutic effects. The central question that prompted this study was to what extent a fully virulent HER2-retargeted oHSV, armed with IL-12, exemplified here by R-115, was able to elicit a local immune response, lymphocytes migration to the tumor and activation, and ultimately local and distant immunotherapeutic efficacy. The question stemmed from intrinsic differences between retargeted oHSVs and the deleted/mutated oHSVs in clinical use. A major difference relates to the innate response, a phenomenon that also impacts on adaptive immunity. As mentioned, most oHSVs are defective in the synthesis of the γ134.5 product, a protein that contrasts the host innate response to the virus [26]. Hence, these viruses are defective in counteracting the innate response. In contrast, fully virulent non-attenuated HSVs first elicit an innate response (e.g. secretion of IFN type I, TNF, etc.), but then dampen it through a number of molecular mechanisms (e.g. secretion of IL-10, IL-6) that limit the hostile microenvironment and ultimately favour viral replication [53–57]. It was thus unclear to what extent the regulation of the innate response put in place by a virulent oHSV would affect its ability to modify the immunosuppressive tumor microenvironment, and to elicit a strong adaptive durable response. Additional differences include the efficient replication and the lack of off-target infection. Specifically, the retargeted oHSVs replicate to near wt-virus yields in human target cancer cells [48], and fail to infect cells other than the targeted cancer cells [44]. In contrast, the currently employed oHSVs infect various cell types in the tumor bed. We report that R-LM113, and its IL-12 encoding derivative R-115 inhibited the growth of the primary treated tumor, completely prevented the growth of distant untreated tumors, elicited local and systemic immune response and thus induced a vaccine-like response. In all assays the IL-12-armed R-115 was more effective than the unarmed R-LM113.
A major difficulty encountered when switching from immunodeficient to immunocompetent mice is that murine cells are scarcely permissive to HSV infections, and the viral replication may be 2–3, or more, logs lower in murine cancer cells than in human cancer cells [58,59]. This is an obvious obstacle in the preclinical studies, and strongly underestimates the efficacy of oHSVs. Further limitations in our experimental model were that HER2-retargeted oHSV only infected HER2-expressing cells, and that the appropriate host for these cells are HER2-transgenic/tolerant mice. Here, we made use of the C57BL/6 HER2-transgenic/tolerant mice, although the parental strain is among the least sensitive to HSV. The murine B16 melanoma cells and the Lewis lung carcinoma (LLC1) cells were made HER2-transgenic by lentiviral transduction, selection with puromycin and single cell cloning. The HER2-LLC1 cells expressed HER2 at higher level than the HER2-B16 cells (Fig 1, compare A to C) and at similar level as the SK-OV-3 cells (Fig 1E), a HER2-expressing human ovary cancer cell line. The HER2-LLC1 and HER2-B16 cells were homogeneous clones (Fig 1B and 1D). SK-OV-3 are shown for comparison (Fig 1F). In both the HER2-LLC1 and HER2-B16 cells, the HER2 expression was stable for more than 30 consecutive passages.
HER2-LLC1 and HER2-B16 cells were compared for ability to support the replication of R-LM113 and R-115 [44,60]. The latter is a R-LM113 derivative, which expresses the murine interleukin 12 (mIL-12) (see Fig 1G for a schematic representation of R-LM113 and R-115 genomes), in the amount of 200–400 pg/105 SK-OV-3 cells [60]. Fig 1H reports the plating efficiency of both viruses and shows that, on average, the amounts of viruses required to infect a single SK-OV-3, HER2-LLC1 and HER2-B16 cell were 1, 20, and 2500 PFUs, respectively. Based on these figures, we next carried out a virus growth experiment in the three cell lines. The cells were infected at 0.1 PFU/cell, according to the virus titre determined in the respective cell line. Under these conditions, R-LM113 and R-115 grew in HER2-LLC1 cells at higher yields than in HER2-B16 cells, and at one order of magnitude lower yields than in the susceptible human SK-OV-3 cells (Fig 1I).
As mentioned in the preceding paragraph, the selected mice were the C57BL/6 HER2-transgenic/tolerant mice [61]. To provide formal evidence of HER2 tolerance, we evaluated the engraftment efficiency of HER2-LLC1 or wt LLC1 cells in the HER2-transgenic/tolerant and in the wt C57BL/6 mice. Fig 2A and 2B shows that the HER2-LLC1 cells exhibited a reduced tumor growth and an about 50% reduction in the engraftment ability in wt mice relative to HER2-transgenic/tolerant mice, quantified as reduced tumor growth at d 25 (Fig 2E), and in the Kaplan-Meier survival curves (Fig 2F). In contrast, when the wt-LLC1 cells were implanted in the two types of mice, there was no substantial difference (Fig 2C–2F). Thus, the wt mice, but not the HER2-transgenic/tolerant mice, exhibited a resistance to the HER2-LLC1; the resistance was not put in place against the wt-LLC1 cells. We conclude that the HER2-transgenic mice were indeed tolerant to HER2.
The family of HER2-retargeted oHSVs exhibits a high safety profile in nude mice, by virtue of the tropism detargeting from the natural receptors, and retargeting to HER2 [42]. We asked whether the high safety profile was maintained in the immunocompetent (wt-C57BL/6) mice, and in the HER2-transgenic/tolerant counterparts. We injected wt-HSV-1(F), R-LM113 and R-115 i.p. in wt- and HER2 transgenic/tolerant mice. Since there was no difference between the HER2-transgenic/tolerant and the wt-mice, the results for the two types of mice are presented cumulatively. HSV-1(F) killed 5/6, and 2/6 mice injected with 2x109 or 1x108, respectively (Fig 2G). Thus, the LD50 was about at least 2 orders of magnitude higher than expected. A LD50 higher than 1x106 PFU was reported for wt HSV-1 in wt-C57BL/6 upon i.p. administration [62]. Of note, none of the mice injected with R-LM113 or R-115 died, irrespective of whether they were wt or HER2 transgenic/tolerant. The results extend to immunocompetent mice the high safety profile of the HER2-retargeted oHSVs [42]. Furthermore, we evaluated R-115 biodistribution to blood and a number of organs, upon 4 consecutive intratumoral injections at 3–4 days distance. Fig 2H shows that no viral genome was detected in any of the organs.
The HER2-LLC1 cells were implanted in the right flanks of HER2-transgenic/tolerant C57BL/6 mice. Mice received 4 locoregional injections of R-LM113 or R-115, at three-four days distance, starting at d 3 after tumor implantation (early treatment) (see schedule in Fig 3A). The tumor growth was delayed or halted in some of the mice receiving R-LM113 relative to vehicle-treated mice (compare Fig 3C to 3B). 7/20 mice were tumor-free (Fig 3C). In the R-115-treated mice the tumor growth regressed in 8 animals, or its engraftment was inhibited, resulting in 15/22 tumor-free animals (Fig 3D). Comparison of the tumor volume at d 21 after virus treatment showed a statistically significant reduction in R-LM113-treated mice, and a higher reduction in R-115-treated mice (Fig 3E). The difference between the two treatment arms was statistically significant, and highlighted the IL-12 contribution. The Kaplan-Meier survival curves showed highly significant differences among the three treatment groups (Fig 3F). At d 18 or 30, a subset of the virus-treated mice (R-LM113, n = 4; R-115, n = 8) were implanted with a 1° contralateral challenge tumor in the opposite flank; the tumor volumes at d 20 and d 27 after its implantation is reported in Fig 3G. The key finding was that all the mice which survived the primary tumor were protected from the challenge tumor (Fig 3G), irrespective of whether they were treated with R-LM113 or R-115. We conclude that the early viral treatment induced a reduction in tumor growth. R-115 was clearly and statistically more effective than R-LM113. All mice which survived the primary tumor were resistant to a 1° contralateral challenge tumor. Some of the mice which had survived the primary tumor and had received the 1° challenge were included in the long survivor group. The relatively high amounts of viruses administered at each dose were not surprising in view of the results in Fig 1H and 1I, which indicate that it takes almost 20 PFUs (as titrated in SK-OV-3) to infect a single HER2-LLC1 cell, and that both viruses replicated in HER2-LLC1 at 1–2 log lower yields than in the human SK-OV-3 cells.
Mice implanted with HER2-LLC1 tumor cells received 4 consecutive peri- intra-tumoral injections of R-LM113 or R-115, starting at 10 days after implantation (Fig 4A). The kinetics of tumor growth are shown in Fig 4B–4D. The R-115-treated mice exhibited a delay or reduction in tumor growth, and 3/18 were tumor-free. A significant reduction in tumor volume at d 24 is documented in Fig 4E, relative to vehicle-treated and to R-LM113-treated mice (Fig 4C–4E). In the latter group 1/12 was tumor-free (Fig 4C). The survival curves show a statistically significant difference between R-115- and R-LM113-treated mice, or control mice (Fig 4F). At d 18 or 30 mice were implanted with a 1° contralateral challenge tumor in the opposite flank, which was rejected in 100% of the animals. The volumes of the challenge tumors, and of the tumors in naïve mice is reported in Fig 4G. These experiments show that (i) R-115 treatment was more effective than R-LM113 treatment. This confirms and extends the early treatment data, and highlights a role of IL-12. (ii) All the mice which survived the primary tumor developed a durable resistance and were spared from the 1° challenge tumor. Some of the mice which had survived the primary tumor and had received the 1° challenge were included in the long survivor group analyzed in Fig 5. As mentioned, above, the moderate efficacy on primary tumor likely reflects the low susceptibility/permissivity of HER2-LLC1 cells to the viruses.
The long-term survivor (LS) group included mice which survived the primary tumor, and were unaffected by the 1° challenge (see, Figs 3 and 4) (n = 4 and n = 8 in the R-LM113 and in the R-115 arms, respectively). At d 80–90 after primary tumor implantation, they received the implantation of a 2° untreated contralateral HER2-LLC1 tumor (Fig 5A for a schematic diagram of treatments). 3/4 mice in the R-LM113 arm, and 8/8 in the R-115 arm were fully protected, whereas the naïve mice were not (Fig 5B–5D). The size of the second contralateral tumor at d 15 and 28 after implantation shows no or reduced tumor growth in mice of both arms, with highest difference in the R-115-treated mice (Fig 5E).
The splenocytes from the long-term survivors which survived the 2° challenge were incubated with the HER2-LLC1 tumor cells to detect tumor-specific reactivity; 5 samples (63%) in the R-115 arm, and all 3 samples in the R-LM113 survivors’ group exhibited a significant IFNγ response (Fig 5F), and a tendency towards higher reactivity in the R-115-treated mice. In the three long-term survivors which did not show a significant response at sacrifice, but resisted tumor engraftment, the sacrifice may have taken place too long after the second challenge, or our assays were not sufficiently sensitive.
At sacrifice, the sera of the long-term survivors that resisted the 2° challenge exhibited a statistically significant reactivity to HER2-LLC1 and to wt-LLC1 cells in both treatment groups, and higher reactivity in the R-115-treated mice (Fig 5G). No reactivity was observed towards the unrelated murine tumor B16 cells, highlighting the specificity of the response (Fig 5G). The results underscore a systemic tumor-specific cell-mediated and humoral immune response in the long-term survivors, argue that the protection from distant tumor growth was immune-mediated, and suggest that the viral treatments broke the tolerance to HER2 as well as to the tumor neoantigens.
Next, we evaluated whether mice developed a long-term antibody response to HSV-1. The long survivor sera were reacted with HSV-1-infected or uninfected rabbit skin cells. As a positive control, we included a monoclonal antibody to HSV-1 glycoprotein D, named HD1 [63]. The mice treated peri- intra-tumorally with R-115 or R-LM113 developed antibodies to the virus (Fig 5H).
To shed light on the immune basis of the acquired resistance and to identify the immune effectors associated with the therapeutic effects exerted by R-LM113 and R-115, we evaluated the modifications elicited by the viruses in tumor infiltrating cells and in immune markers. A new group of mice was treated as depicted in Fig 4A, i.e. with 4 consecutive virus administrations, starting at d 10 after tumor implantation. Tumors were explanted 6–7 days after completion of the virus treatment. Based on the virus effect on tumor growth, we subdivided the animals in responders and non-responders. The first exhibited a regression or slowdown in tumor growth. The latter exhibited a tumor growth similar to that in the untreated arm. The tumor growth curves are shown in Fig 6A–6E. A most striking result was the difference in the responder/non-responder ratio between the two arms, namely 3/12 and 18/13 in the R-LM113 and R-115 arms. This corresponds to 20% and 58% responders/treated animals in the two arms. The tumor volumes in the five groups at d 22 is reported in Fig 6F. HSV genome copies in the tumors ranged from 300 to 775000 copies/100ng DNA, did not differ between the two treatment arms, nor between responders and non-responders, and was in overall agreement with previous observations [36]. In the blood and in a number of tissues, there was practically no detectable virus genome, as shown in Fig 2H. Thus, virus presence was strictly limited to the tumor. The presence of non-responder mice may reflect both the stochastic nature of the immune response and individual variations in the repertoire of immune cells, and the fact that the amounts of administered viruses were insufficient to elicit a total response, as seen in the efficacy data.
Next, we characterized the immune cells present in tumors and spleens. Although the number of responder mice in the R-LM113 arm was very small, and data should therefore be interpreted cautiously, two patterns emerged. Thus, CD4+ and CD335+ cells—both total and expressing the CD69 activation marker—were increased in the tumors of all responder mice (Fig 6H–6J). CD8+, both total and expressing the CD69 activation marker, Foxp3+, and CD141+ cells were increased and CD11b+ cells were decreased preferentially in the responder R-115-treated tumors (Fig 6K–6O). The distribution of these cells in the spleens mirrored that in the tumors (Fig 6Q–6X). PD-L1 was increased in tumor from R-115-treated mice, and in the spleens of mice treated with either virus (Fig 6P and 6Y). In as much as PD-L1 expression is regulated downstream of the IFN R signaling, its increase in the responder group was likely induced at least in part by IFNγ. The results suggest that some changes to tumor infiltrating and spleen infiltrating lymphocytes/monocytes were likely elicited by either virus, and, conversely, that the CD8+ cell infiltration was preferentially induced by the IL-12 encoding R-115.
Next, we analyzed the tumors from Fig 6 by reverse transcription quantitative PCR (qRT-PCR). The specimens from both the R-LM113- and R-115-treated mice exhibited a highly significant difference in Ifng and Tbet mRNA levels, relative to the vehicle-treated mice (Fig 7A and 7B). Tbet is a transcription factor which contributes to drive a type 1 T helper (Th1) response and controls IFNγ expression. There was a clear tendency towards higher response in the R-115 arm, even though a statistical significance analysis was not carried out, given that a single sample was present in the responder R-LM113 group. The results provide a second line of evidence for activation of the intratumoral immune response elicited by the retargeted oHSVs, with a trend for higher activation in the R-115 responder mice.
An unbiased analysis of the immune markers at the protein level was carried out by Luminex Multiplex, in tumor specimens taken at 6–7 days after virus treatment. Even in this assay, the number of available specimens for the responder R-LM113-treated mice was too low; therefore the R-LM113-derived specimens were considered as a single group. In specimens from both virus treatments there was a clear increase in IFNγ and Granzyme B, especially in the R-115 responders (Fig 7C and 7D). There was an increase also in TNFα and IL-2 in the R-115 specimens (Fig 7E and 7F), highlighting a polarization towards a Th1 response and an activation of the effector cells. IL-10 and IL-6 were specifically increased in the R-115 specimens (Fig 7G and 7H). The IL-12 was detectable at d 5 and increased at d 7 after completion of R-115 treatment (Fig 7I). In R-LM113-treated mice no increase in IL-12 was observed at d 5 relative to vehicle-treated mice. The data extend the evidence for a de-repression of the immunosuppressive microenvironment induced especially by R-115. We note that TNFα, IL-6 and IL-10 are induced also by HSV, so their presence may in part be consequent to infection [54,55,57].
We report on the first efficacy studies of fully virulent, HER2-retargeted oHSVs, either unarmed (R-LM113) or armed with mIL-12 (R-115). Their peri- intra-tumoral administration led to a reduction in the growth of the primary tumor, particularly in the R-115-treated mice, to intratumoral and systemic immune responses which almost completely prevented the engraftment of distant untreated tumors. In essence, the intralesional vaccination promoted local and systemic immunity. Pertinent to the model system and the main results is the following.
Inasmuch R-LM113 and R-115 were retargeted to HER2, the murine cancer cells were made HER2-transgenic, and the mice were the HER2-transgenic/tolerant C57BL/6 [61]. The safety profile of both viruses was very high, as mice resisted i.p. doses as high as to 2x109 PFU, which were lethal for 83% of the mice injected with wt-HSV. Furthermore, upon peri- intra-tumoral delivery, there was no detectable R-115 infection in organs other than the injected tumors. The selected tumor cells were the HER2-LLC1, which are markedly less sensitive/permissive than the human cancer cells. This animal system is therefore adequate to investigate the immunotherapeutic efficacy of the HER2-retargeted oHSVs, but is hardly predictive of the therapeutic effects on primary tumors. The high resistance of murine tumor cells to HSV is shared with a large part of preclinical studies on the therapeutic effects of oHSVs [58,59], and clearly results in underestimation of the oncolytic and immunotherapeutic effects of oHSVs.
The key efficacy and immunotherapeutic data were as follows. The IL-12 armed R-115 reduced and delayed tumor growth more efficiently than the unarmed R-LM113. Practically all the mice which survived the primary tumor—from either the R-LM113 or R-115 arm—were protected from a distant challenge tumor, and, after several weeks, from a subsequent re-challenge. Hence, the highest effect was the abscopal one. A specific systemic immune response was detected at sacrifice both in splenocytes and in sera of the long-term survivors. The durable vaccine effect adds to the strong oncolytic effects exerted on the primary human tumors in immunodeficient mice [42,50,51].
The immune basis of the therapy was documented by two lines of evidence, i.e. the early modifications to the immune suppressive tumor microenvironment and the establishment of a systemic response. With respect to the early modifications, the tumors in responder mice contained significant amounts of immunostimulatory factors, at higher frequencies/amounts in the R-115 arm. Beside IL-12, these included IFNγ, IL-2, Granzyme B, Tbet and TNFα,—typical effectors of IL-12, Th1 polarization and natural killer (NK) cell activation, in addition to IL-10 and IL-6. The tumor microenvironment of the responder mice was further characterized by infiltration and activation of immune cells. Although the number of R-LM113 responders was very low and data should be considered cautiously, two patterns emerged. CD4+ and CD335+ NK cells appeared to be induced by either virus. These cells may well represent the first line of anti-tumor defence put in place by virulent oHSVs. CD8+ and CD141+ cells, PD-L1+ tumor cells, and Foxp3 T regulatory cells were preferentially induced by R-115, along with a decrease in the number of intratumoral CD11b+ cells. The role of each immune cell subpopulation of in virus-induced antitumor activity is the subject of intense studies. The impact of the CD8+ cell infiltration in unleashing the immunosuppressive phenotype of the tumor and in sensitizing otherwise refractory cancers to checkpoint inhibitors was documented recently also in clinical specimens from patients treated with OncovexGM-CSF + pembrolizumab [18,64]. More controversial appears to be the role of NK cells. These cells and their effectors IFNγ, TNFα, Granzyme B were selectively increased in the R-115-treated tumors. The impact of the NK cell recruitment in the Maraba virus-induced eradication of cancer was well established [34]. In contrast, in glioblastoma viro-immunotherapy induced by a transcriptionally retargeted oHSV the activation and recruitment of NK cells to the treated tumor diminished the anti-tumor efficacy [65]. Specific NK cell depletion experiments will shed light on the role played by NK cells on the antitumor activity of the HER2-retargeted o-HSVs. The increase in Foxp3+ T-regulatory lymphocytes was observed upon treatment with other OVs, e.g. Newcastle disease virus, and may reflect an autoregulatory mechanism put in place upon CD8+ cell-mediated response [66]. Also the monocytic lineage was affected by the viral treatment; in particular CD11b+ cells were decreased in R-115 responders. These cells include cells that contribute to the immunosuppressive properties of the tumor microenvironment, including myeloid-derived-suppressor cells [67]. Not surprisingly they were present in higher amounts in the tumors of non responder mice, and may contribute to resistance. Of the cytokines induced by R-115, TNFα, IL-6 and IL-10 are part also of the signature of the antiviral innate response to HSV, finalized to limit the virus-hostile microenvironment and to favour virus replication [53–55,57]. Hence, the immune-related molecules in the tumor microenvironment are likely to be induced in part by the virus itself, in part by the virus-encoded IL-12. The former are not necessarily detrimental to the control of tumor growth. It has been proposed that the OV-induced antiviral immune responses may exert intrinsic anticancer benefits and may be critical for establishing clinically desired antitumor immunity [68,69]. Studies with a recombinant o-poliovirus suggested activation of dendritic cells and immune adjuvancy as a result of canonical innate anti-pathogen response [70].
The systemic response was detected in splenocytes and as durable antibody response towards both the HER2-LLC1 and the wt-LLC1 cells. This finding implied that the virus treatment broke the tolerance not only to the HER2-LLC1 but also to the wt-LLC1 cells, and argued that, especially for R-115, a systemic route of delivery was not required to achieve a systemic vaccine protection. The high efficacy against the engraftment of distant untreated tumors, implanted long after virus treatment, and the Th1 polarization bare strong similarities with the therapeutic effect of a IL-12-armed oncolytic measles virus [33]. In that case, the viral backbone was that of a measles virus vaccine strain.
Differences are readily evident between current efficacy data and those exerted by attenuated Δγ134.5 oHSVs. The major effect of OncovexGM-CSF on B16 melanoma cancer cells in C57BL/6 mice was exerted on the primary treated tumor. The effects on the distant untreated tumors were limited [58] and a long-term protection was only achieved when the OncovexGM-CSF was combined with anti-CTLA-4 [58]. The different effects seen in that study and in current investigation likely reflect the different action of the cytokines expressed by the respective oHSV, i.e. the GM-GSF in OncovexGM-CSF and the IL-12 in R-115, the permissivity of the respective murine cancer cells to HSV infection, as well as the genomic properties of the viral backbones, namely a deleted/attenuated viral genome in OncovexGM-CSF, and a full non-attenuated genome in R-115. The importance of the genomic backbone is clearly apparent when the effects elicited by R-115 are compared to those elicited by the IL-12-armed G47 [19]. The two recombinants express the same cytokine, and differ strikingly in their genome. The IL-12-G47 carries multiple deletions, whereas R-115 carries no deletion. As a monotherapy against a model glioblastoma tumor, G47 alone or the IL-12-G47 were barely effective [19]. A strong efficacy required the combination of IL-12-G47 with both CPIs, the anti-CTLA-4 and anti-PD-1 Abs. While the glioblastoma model likely exemplifies a refractory and immunosuppressive tumor, and, possibly, a resistant tumor to HSV infection, the current finding that the IL-12-armed R-115 reduced the growth of the primary tumor and fully protected the mice from the growth of distal tumors in the absence of combination with CPIs argue that the fully virulent retargeted oHSVs have the potential to be highly active oncolytic-immunotherapeutic agents.
Human ovary SK-OV-3 cancer cells (Roswell Park Memorial Institute) were cultured in RPMI- Glutamax (Life Technologies #61870–010) containing 10% fetal bovine serum (FBS). LLC1 and B16 cells were purchased from ATCC and cultured in Dulbecco modified Eagle medium (Life Technologies, #31966–021) containing Glutamax, High Glucose and 10% FBS.
R-LM113 was described [44]. R-115 is a derivative of R-LM113; like the parental R-LM113, it is retargeted to HER2. It expresses the murine IL-12 (mIL-12) under the CMV promoter [60]. Viruses were cultivated and titrated by plaque assay in SK-OV-3 cells. The mIL-12 production was quantified in the supernatant of R-115-infected cells by means of mIL-12 ELISA kit (EMIL12, Thermo Fisher Scientific) according to manufacturer instruction.
The HER2 receptor expressed in murine cancer cells was a chimera in which the C-tail signaling portion was replaced with the C-tail of the non-signaling nectin1 receptor. The construct was named as HER2-nectin, and herein referred to as HER2. For the engineering, the ectodomain of human HER2 receptor was amplified from pcDNA-HER2 plasmid [71] with primers HER2_NheI_f GCGGCCGCGCTAGCATGGAGCTGGCGGCCTTGTGCCGC and HER2_HpaI_r AGAGATGATGGAGTTAACAGGGCTGGCTCTCTGCTCGGCGGG. The HER2 ectodomain amplicon spans from ATG start codon up to the amino acid 650 (GenBank NM_004448). pCF18HNK was described [72]. The plasmid was cut with NheI-HF and HpaI (New England Laboratories) to delete the ectodomain of human nectin1 from amino acid 1 to 330 (GenBank NM_203285.1 aa 1–330). The HER2 ectodomain amplicon was cut with NheI and HpaI and then ligated into NheI/HpaI digested pCF18HNK to generate pHER2-Nectin plasmid. The entire HER2-Nectin chimeric ORF was sequenced. Then, a NheI/XbaI fragment from pHER2-nectin plasmid was subcloned in the NheI/XbaI digested lentiviral expression vector pLV-EF1-MCS-SV40-Puro. The resulting expression vector was called pLV-HER2-nectin-puro.
The B16 melanoma and the LLC1 cells were made transgenic for HER2-nectin expression by lentiviral transduction, as detailed [73]. Transduced cells were selected by means of puromycin, enriched for HER2 expressing cells with microbeads (Miltenyi Biotech) following incubation with an anti-HER2 mouse IgG antibody (SantaCruz Biotecnology, clone 9G6). Single-cell clones were obtained by limiting dilution. Clones were checked for stable HER2 expression for up to 30 passages in cell culture by flow cytometry (BD Accuri) with anti-HER2 MGR2 antibody (Vinci-Biochem, #ALX-804-573-C100).
To measure the virus growth, HER2-LLC1, HER2-B16 and SK-OV-3 cells were infected at an input multiplicity of 0.1 PFU/cell (as titrated in the respective cell line) for 90 min at 37°C. Unabsorbed virus was inactivated by means of acidic wash (40 mM citric acid, 10 mM KCl, 135 mM NaCl, pH 3). Replicate cultures were frozen at the indicated times (24 and 48 h) after infection and the progeny was titrated in SK-OV-3.
To determine the relative plating efficiency, replicate aliquots of R-115 or R-LM113 were plated onto HER2-LLC1, HER2-B16 and SK-OV-3 cells. The infected monolayers were overlaid with medium containing agar. The plaques were scored 3 days later. The results represent the average of triplicates ± SD.
C57BL/6 mice transgenic for and tolerant to HER2 (B6.Cg-Pds5bTg(Wap-ERBB2)229Wzw/J) [61] were obtained from Wayne State University through The Jackson Laboratories, and bred in the facility of the Department of Veterinary Medical Sciences, University of Bologna. The animals for tumor implantation were HER2-transgenic (HER2-TG). HER2-LLC1 cells were implanted subcutaneously in the left flank of six-to-eight weeks old HER2-TG C57BL/6 mice in 250 μL of PBS, 0.2x106 cells/mouse. The start of the virus treatment is detailed in the Results section. Mice received 4 loco-regional or peri- intra-tumoral (p.i.t.) injections of the respective virus suspension, diluted in PBS, 1x108 PFU/mouse, at 3–4 days distance. Mice in the control group received PBS (vehicle). Each treatment group consisted of 5, 10 or more mice, as detailed in the Figure legend. Tumor volumes were scored twice weekly by measuring the largest and the smallest diameter by means of a calliper. Tumor volume was calculated using the formula: largest diameter x (smallest diameter)2 x 0.5. Mice were killed when tumor volumes exceed 1000–2000 mm3, ulceration occurred, or animals exhibited distress or pain. Where indicated, mice received a contralateral tumor, made of HER2-LLC1 cells implanted subcutaneously in the right flank, 0.2x106 cells/mouse. The contralateral tumors were not treated.
Freshly explanted spleens were smashed through a 70 μm cell strainer in PBS with a sterile 5ml syringe plunger to isolate splenocytes. Red blood cells in spleen and tumor samples were lysed with ACK buffer (NH4Cl 150 mM, NaHCO3 10 mM, 1mM EDTA), resuspended in medium (RPMI 1640 containing 10% heat inactivated FBS, 1% penicillin/streptomycin, 0.05 mM β-Mercaptoethanol), counted and seeded in 24 well plate. Splenocytes (1x106 cell/well) were incubated with 1x105 HER2-LLC1 cells in 0.5 ml medium, and cocultured for 48 h. Media were collected and the amount of secreted IFNγ was quantified by ELISA (IFN-gamma Mouse ELISA Kit, Thermo Fisher).
Tumors were minced, resuspended in lysis buffer (Tris-HCl pH 7.4 50 mM, NaCl 250 mM, EDTA 5 mM, Na3VO4 1 mM, NaF 50 mM, NaN3 0.02%, Sodium deoxycholate 0.5%, NP40 1%, Nα-p-tosyl-L-lysin chloromethyl ketone hydrochloride 0.3 mM, Nα-p-tosyl-L-phenylalanine chloromethyl ketone 0.3 mM, PMFS 1 mM), in a proportion of 500μL of lysis buffer for 100 mg of tumor. Samples were sonicated with Bioruptor (Diagenode), using program HIGH for 20 minutes (30sec ON 30sec OFF) and centrifuged for 30 minutes at 11000 x g. The protein content of supernatants was determined by means of the Bio-Rad protein assay (Bio-Rad); the supernatants were analyzed by means of a Magnetic Luminex Assay (R&D) and a mouse premixed Multi-Analyte kit. The custom-made kit included: TNFα (BR14), IL-12 p70 (BR15), IL-2 (BR22), IL-4 (BR25), IL-6 (BR27), IL-10 (BR28), IL-17A (BR30), IFNγ (BR33), CXCL10/IP-10 (BR37), Granzyme B (BR63). Supernatants were 1:1 or 1:5 diluted with the Calibrator Diluent RD6-52 and analyzed according to manufacturer instructions. Standard curve with 1:1 diluted lysis buffer was employed to quantify TNFα, IL-2, IL-4, IL-6, IL-10, IL-17A, IP-10 and Granzyme B in 1:1 diluted samples, while the quantification of IFNγ was performed in 1:5 diluted samples, using the corresponding standard curve in 1:5 diluted lysis buffer. Data were analyzed according to the manufacturer instructions. Results were expressed as pg of each analyte per mg of total proteins. Quantification of IL-12 p70 was under the LOD for Luminex analysis, hence the analyte was quantified by means of the Mouse IL-12 ELISA kit.
Tumors were homogenized. A few mgs of the homogenates were employed for total RNA purification with the Nucleospin RNA kit (Macherey-Nagel) according to the manufacturer’s protocol (including the on-column DNaseI treatment). 1.2 μg of RNA was employed for the cDNA synthesis using the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems), following the manufacturer’s instruction. qRT-PCR reactions were prepared as follows: 2 μL of the diluted (1:4) cDNA samples were mixed with 5 μL of TaqMan Fast Advanced Master Mix (Applied Biosystems) and 0.5 μL TaqMan probes in a final volume of 10 μL. The TaqMan probes employed for the assay were: mm01168134_m1 (Ifng), mm00450960_m1 (Tbx21/Tbet), mm01612987_g1 (Rpl13a). qRT-PCR reactions were performed in a StepOnePlus System (Applied Biosystems), following the protocol indicated in the Master Mix. The levels of expression were determined using the ΔΔCt method, normalized on the Rpl13a housekeeping gene.
Tumors and organs were homogenized; a few mgs of the homogenates or 50 μL of blood were employed for DNA purification with the Nucleospin Tissue kit (Macherey-Nagel) according to the manufacturer’s protocol. HSV genome copies (gc) were quantified by qRT-PCR. 2 μL of the dilutions or 30 ng of the tissue DNAs were mixed with 5 μL of TaqMan Fast Advanced Master Mix (Applied Biosystems) and 0.5 μL HSV probe in a final volume of 10 μL. The HSV probe contained the following oligonucleotides: DnapolFw (CATCACCGACCCGGAGAGGGAC), DnapolRev (GGGCCAGGCGCTTGTTGGTGTA) and DNA_Pol_PROBE (FAM-CCGCCGAACTGAGCAGACACCCGCGC-TAMRA), 2.50μM each. qRT-PCR reactions were performed in a StepOnePlus System (Applied Biosystems), following the protocol indicated in the Master Mix. The amount of gc was determined by comparison with a standard curve, prepared using purified genomic HSV DNA, and expressed as gc/100ng of DNA.
LLC1 and HER2-LLC1 cells were trypsinized, rinsed and resuspended in flow cytometry buffer (PBS + 2% FBS). For each sample, 0.25x106 cells were reacted with mouse serum, diluted 1:60, in 96 well plate in ice for 1 hour, rinsed with flow cytometry buffer and finally incubated with anti mouse PE (1:400, Beckton Dickinson). Data were acquired on BD C6 Accuri. Cell enzyme-linked immunosorbent assay (CELISA) was performed as described [74]. Briefly, RS cells were infected with HSV-1 (F) at 3 PFU/cell, in 96 well plate. Twenty-four h later they were fixed with paraformaldehyde, reacted with mouse serum diluted 1:60, or with MAb HD1 diluted 1:400, followed by anti-mouse peroxidase. Finally, peroxidase substrate o-phenylenediamine dihydrochloride (OPD; Sigma-Aldrich) was added and plates were read at 490 nm with GloMax Discover System (Promega Corporation).
Single cell suspensions were prepared from freshly isolated HER2-LLC1 tumors and spleens at sacrifice. Tumors were minced in small pieces and digested with collagenase (1 mg/ml) for 1.5 h at 37°C. The resulting cell suspensions were passed through 70 μm cell strainer and rinsed with flow cytometry buffer. Spleens were processed as described above, and then treated as the tumor samples. Red blood cells in spleen and tumor specimens were lysed by means of ACK buffer, samples were pelleted and resuspended in flow cytometry buffer. Subsequently for each sample 2x106 cells were blocked with α-CD16/32 Ab (clone 93, eBioscience), and then reacted with the antibodies CD4-FITC (clone GK1.5, eBioscience), CD8a-PE (clone 53–6.7, eBioscience), CD45-FITC (clone 30-F11, eBioscience), CD45-Percp-Cy7 (clone 30-F11, eBioscience), CD335-PE (clone 29A1.4, eBioscience), FoxP3-PE (clone 150d/e4, eBioscience), CD11b-FITC (clone M1/70, eBioscience), PD-L1-APC (clone MIH5, BD), CD141-PE (clone LS17-9, eBioscience) and CD69-PercP (clone H1-2F3, eBioscience). Data were acquired on BD C6 Accuri. Only samples which provided at least 100000 events were included in subsequent analysis.
Animal experiments were performed according to European directive 2010/63/UE, Italian laws 116/92 and 26/2014. The experimental protocols were reviewed and approved by the University of Bologna Animal Care and Use Committee (‘‘Comitato per il Benessere degli Animali”, COBA), and approved by the Italian Ministry of Health, Authorization # 86/2017-PR to Prof. Anna Zaghini.
Statistical analyses are reported in figure legends, as it applies: *, p < 0.05; **, p < 0.01; ***, p < 0.001; ****, p < 0.0001; ns, non-significant.
|
10.1371/journal.pntd.0001839 | Molecular Diagnostics for Lassa Fever at Irrua Specialist Teaching Hospital, Nigeria: Lessons Learnt from Two Years of Laboratory Operation | Lassa fever is a viral hemorrhagic fever endemic in West Africa. However, none of the hospitals in the endemic areas of Nigeria has the capacity to perform Lassa virus diagnostics. Case identification and management solely relies on non-specific clinical criteria. The Irrua Specialist Teaching Hospital (ISTH) in the central senatorial district of Edo State struggled with this challenge for many years.
A laboratory for molecular diagnosis of Lassa fever, complying with basic standards of diagnostic PCR facilities, was established at ISTH in 2008. During 2009 through 2010, samples of 1,650 suspected cases were processed, of which 198 (12%) tested positive by Lassa virus RT-PCR. No remarkable demographic differences were observed between PCR-positive and negative patients. The case fatality rate for Lassa fever was 31%. Nearly two thirds of confirmed cases attended the emergency departments of ISTH. The time window for therapeutic intervention was extremely short, as 50% of the fatal cases died within 2 days of hospitalization—often before ribavirin treatment could be commenced. Fatal Lassa fever cases were older (p = 0.005), had lower body temperature (p<0.0001), and had higher creatinine (p<0.0001) and blood urea levels (p<0.0001) than survivors. Lassa fever incidence in the hospital followed a seasonal pattern with a peak between November and March. Lassa virus sequences obtained from the patients originating from Edo State formed—within lineage II—a separate clade that could be further subdivided into three clusters.
Lassa fever case management was improved at a tertiary health institution in Nigeria through establishment of a laboratory for routine diagnostics of Lassa virus. Data collected in two years of operation demonstrate that Lassa fever is a serious public health problem in Edo State and reveal new insights into the disease in hospitalized patients.
| In the past, diagnostic testing for Lassa fever patients in Nigeria has been performed nearly exclusively outside of the country. Patients thus were managed on-site based on clinical suspicion alone, posing risks to patients and health care workers and exhausting resources. To tackle this problem, we established a diagnostic PCR laboratory directly at a referral hospital serving a Lassa fever endemic area in Nigeria. Long-term collaboration between partners in the North and the South was crucial to implement this project. Training of laboratory staff in the partner institutions and on-site, mobilization of local human and financial resources, good management of the laboratory, a basic quality management and control system, and a stable supply chain for consumables and reagents were among the key factors for success. The laboratory reliably delivered results in a short turnaround time, despite some problems due to PCR contamination. The service has improved patient and contact management including treatment with ribavirin and led to better protection of health care workers against hospital-acquired infections. The data provide new insights into disease progression and a basis for further optimization of case management including supportive treatment.
| Lassa fever is a viral hemorrhagic fever that was first described in 1969 in the town of Lassa in the North-East of Nigeria [1]. It is endemic in the West African countries of Sierra Leone, Guinea, Liberia, and Nigeria ([2], [3] and references therein). Cases imported to Europe indicate that Lassa fever also occurs in Côte d'Ivoire and Mali [4], [5]. The causative agent is Lassa virus, an RNA virus of the family Arenaviridae. Its natural host is the rodent Mastomys natalensis [6], [7], which lives in close contact to humans. Mastomys shed the virus in urine [8] and contamination of human food is a likely mode of transmission. The virus may be further transmitted from human to human, giving rise to mainly nosocomial epidemics with case fatality rates (CFR) of up to 65% [9]–[12]. However, most of the Lassa virus infections in the communities are probably mild [13].
Clinically, Lassa fever is extremely difficult to distinguish from other febrile illnesses seen in West African hospitals, at least in the initial phase [14], [15]. Gastrointestinal symptoms, pharyngitis, and cough are frequent signs. Late complications include pleural and pericardial effusions, facial edema, bleeding, convulsion, and coma. In the terminal stage patients often go into shock, although bleeding itself is usually not of a magnitude to produce shock [10], [14]–[22]. The only drug with a proven therapeutic effect in humans is the nucleoside analogue ribavirin. Drug efficacy decreases if treatment is commenced at day 7 or later [23], making early diagnostics critical for survival.
Lassa virus can be detected in blood at an early stage of illness. Death occurs about two weeks after onset of illness with fatal cases showing higher levels of viremia than those who survive. In survivors, virus is cleared from circulation about three weeks after onset of symptoms [24]–[26]. IgM and IgG antibodies are detectable only in a fraction of patients during the first days of illness, and patients with fatal Lassa fever may not develop antibodies at all [24], [26], [27]. Therefore, RT-PCR is a valuable tool for rapid and early diagnosis of Lassa fever [24], [25], [28], [29].
So far, diagnostic testing of samples from Lassa fever patients has been performed almost exclusively outside of Africa. Only the laboratory at the hospital in Kenema, Sierra Leone, which has become operational since 2004 (after civil war forced its closure in 1993), is able to perform Lassa fever testing for patients [30]. In Nigeria, the situation improved with the implementation of Lassa virus PCR testing at a research laboratory of the University of Lagos, which facilitated retrospective laboratory confirmation of Lassa fever cases in various parts of the country [31], [32]. However, none of the hospitals in the endemic areas of Nigeria has the capacity to perform Lassa virus tests. Case management is thus mainly based on non-specific clinical criteria [14], [15] and in the worst cases, health care workers became infected while they treated patients without knowing they had Lassa fever [32].
The Irrua Specialist Teaching Hospital (ISTH) has faced these challenges for many years. ISTH serves as a referral hospital in Edo State, one of the many Nigerian States with evidence of Lassa fever [1], [9], [12], [17], [32]–[37]. In 2001, ISTH was designated as a Centre of Excellence in the management of Lassa fever, along with two other federal tertiary health institutions. It set up awareness campaigns to sensitize hospital staff and the public to the severity of Lassa fever infection and need for treatment and prevention. Ribavirin was periodically supplied to the hospital by the Federal Ministry of Health and given to suspected cases. Prevalence and case fatality figures based on clinical suspicion and pilot laboratory investigations in 2003 and 2004 suggested a high incidence of Lassa fever in Edo State [31], but the true magnitude of the problem remained obscure. In 2007, the management of ISTH was dissatisfied with the level of response and attention given to Lassa fever and took bold steps to address the situation. Amongst these was the establishment of the Institute of Lassa Fever Research and Control (ILFRC). The rationale for the institute was based on the need to build capacity to adequately respond to the epidemics observed in the region in terms of manpower development and training, laboratory diagnosis, and adequate case management as well as the dire need for focused research and advocacy. A collaborative effort was made to establish a laboratory for molecular diagnostics of Lassa fever, which was considered crucial for appropriate case and contact management, including early treatment and postexposure prophylaxis with ribavirin [23], [38], [39]. The diagnostic and research laboratory was built in 2008 and started operation in September 2008. We describe here the establishment of a diagnostic service for Lassa fever and analyze the data recorded during two years of operation.
The study was classified as a service evaluation and granted exemption from ethical review by the Research and Ethics Committee of ISTH. Lassa fever PCR diagnostics, patient management, and public health measures are part of routine clinical practice at ISTH. The choice of treatment or diagnostics was that of the clinician and patient according to professional standards or patient preference. Neither PCR diagnostics nor treatment with ribavirin was experimental in nature. All data described in the manuscript stem from existing records that have been generated as part of the regular clinical practice. The service was evaluated 2 years after implementation. Service evaluation is exempt from ethical review according to the National Code of Health Research Ethics, National Health Research Ethics Committee, Federal Ministry of Health, Nigeria.
The following case definition, taking into account published signs and symptoms of Lassa fever, was used as a guideline for identifying suspected cases and requesting molecular testing for Lassa virus (from the diagnostic laboratory request form):
However, as the laboratory provides routine diagnostic service for patients, there was flexibility in applying this case definition. Clinical experience and suspicion were taken into account as well. In addition, in the interest of time, Lassa virus testing was often performed before typhoid fever or malaria had been excluded. Samples were also sent from other parts of Nigeria for Lassa virus RT-PCR testing.
Ribavirin treatment was usually commenced on clinical grounds before laboratory testing. If RT-PCR was negative, it was terminated. However, in cases with a strong clinical suspicion for Lassa fever, ribavirin treatment was continued even if the RT-PCR was negative. If the RT-PCR was positive, treatment was continued or commenced.
If a patient was suspected of having Lassa fever in any of the clinical or outpatient departments of ISTH, staff of ILFRC collected an EDTA blood sample. The blood was centrifuged and virus RNA was purified from plasma by using the diatomaceous earth method as described [40], [41]. In brief, 140 µl and 14 µl, respectively, of each plasma sample were mixed with 560 µl chaotropic lysis buffer AVL (Qiagen) containing 5.6 µg carrier RNA (Qiagen, no. 19073). AVL has been shown to inactivate enveloped RNA viruses [42]. The lysate was incubated at room temperature for 10 min. About 100 mg of diatomaceous earth (Sigma, no. D3877) and subsequently 560 µl of ethanol was added to the lysate and the slurry was incubated with vigorous agitation for 10 min at room temperature in a shaker. The diatomaceous earth was pelleted by centrifugation, and the pellet was washed three times, first with 500 µl of buffer AW1 (Qiagen, no. 19081), second with 500 µl of buffer AW2 (Qiagen, no. 19072), and finally with 400 µl of acetone (each washing step included vortexing with wash fluid, centrifugation for 2 minutes at maximum speed in a table top centrifuge, and removal of supernatant). The pellet was dried at 56°C for 20 minutes until the acetone was completely evaporated. To elute the RNA from the diatomaceous earth, the pellet was resuspended in 100 µl of water (Aqua ad injectabilia), the slurry was vortexed, incubated 1 minute at room temperature, centrifuged at maximum speed, and the supernatant was transferred to a new tube. The RNA was immediately used for PCR. The Lassa virus RT-PCR targeting the GPC gene was performed using QIAGEN OneStep RT-PCR Kit reagents (Qiagen, no. 210210 or 210212) as described [28]. The 25-µl assay contained 5 µl RNA, 0.6 µM primer 36E2 (ACC GGG GAT CCT AGG CAT TT), 0.6 µM primer LVS-339-rev (GTT CTT TGT GCA GGA MAG GGG CAT KGT CAT), 0.4 mM dNTP, 1× RT-PCR buffer, 1× Q-solution, and 1 µl enzyme mix. The reaction was performed in a Primus25advanced thermocycler (PeqLab, Erlangen, Germany) using the following temperature profile: 50°C for 30 min, 95°C for 15 min, followed by 45 cycles of 95°C for 30 s, 52°C for 30 s and 72°C for 30 s. All pre-PCR pipetting was performed with filter tips. PCR products were separated in a 1.5% agarose gel containing ethidium bromide and visualized by UV light. Gel images were recorded with a digital camera. As a positive control, inactivated culture supernatant of cells infected with Lassa virus strain CSF was used. All consumables, reagents, chemicals, kits, and plastic materials were purchased in Germany or the US and transferred to ISTH.
PCR products generated in the diagnostics from September 2008 through February 2011 were stored at −20°C and sequenced retrospectively using primer 36E2. The automated base calling was proof-read by visual inspection of the electropherograms. A representative set of 35 sequences has been sent to GenBank and assigned the accession nos. JN651366-JN651400. Phylogenetic analysis included all novel GPC sequences (n = 204) as well as Lassa virus sequences available from GenBank by May 2011. The program jModelTest 0.1.1 [43] identified the general time-reversible model of sequence evolution with a gamma distribution of among-site nucleotide substitution rate variation (GTR+gamma) as the substitution model that best describes the data in the nucleotide sequence alignment of the partial GPC genes (284 taxa, 237 sites). The gamma+invariant sites model was not considered because it was not favored by jModelTest, and because the two parameters estimated under this model (the gamma distribution shape parameter and the proportion of invariant sites) are highly correlated and may be poorly estimated depending on the number of taxa [44]. Phylogenies were inferred by the Bayesian Markov Chain Monte Carlo method implemented in BEAST software [45] using the following parameters: GTR+gamma; 107 steps with sampling every 105th step; and two independent runs combined (effective sampling size >100 for all parameters). To avoid over-parameterization – considering that the sequences were short – simple molecular clock and demographic models were chosen, that is strict clock with mean substitution rate fixed at 1 and constant population size.
Demographic data as well as major symptoms at presentation were recorded on the request form that accompanied the blood sample. The department responsible for the patient provided clinical chemistry data generated in the Clinical Pathology Department of ISTH and data on treatment and outcome. Data were entered into a database (Excel, Microsoft) maintained at ILFRC. All patients from whom samples were processed in the Lassa fever diagnostics laboratory from January 2009 through December 2010 were included in the analysis. Each case was included only once; further testings on the same case were not considered. The data set was checked using plausibility criteria. A PCR result was corrected before analysis if the sequence of the PCR product indicated a false positive result, e.g. if the sequence corresponded to that of the positive control. Statistical comparison of unpaired groups was performed for continuous parameters with the Mann-Whitney test and for frequencies with two-tailed Fisher's Exact test. A critical p value of 0.01 was considered appropriate, given the large number of tests performed on the data set. The p value was further lowered to 0.001–0.0005 according to Bonferroni correction if multiple tests were conducted within one category (e.g. within the category profession).
The laboratory was built in 2008 on the campus of ISTH. Equipment was provided by Bernhard-Nocht-Institute for Tropical Medicine (BNI), Harvard University, and University of Ibadan after initiating collaborations with the hospital in 2007. These partners also performed on-site trainings before and regularly during operation of the laboratory. In addition, staff of ILFRC was trained for 3 months in PCR technology at BNI in Hamburg and at Harvard University in Cambridge, Massachusetts. The laboratory started operation in September 2008. It is divided into three zones to minimize PCR contamination: a “clean” area for all pre-PCR manipulations, a “grey” area for amplification, and a “dirty” area for post-PCR manipulations (Figure 1A). The clean area features separate rooms for i) sample inactivation, ii) RNA extraction and PCR setup, and iii) mastermix preparation. The workflow starts with sampling blood using a closed syringe system. The sample is processed the same day or, if it arrives late, it is stored at 4°C and processed the next day. The syringe container is opened within a plexiglas box in the inactivation room and the plasma is mixed with chaotropic buffer to inactivate the virus and prepare RNA, or aliquoted for storage at −20°C (Figure 1B). From each plasma sample, 140 µl and 14 µl are inactivated and processed separately. The reason for testing two different volumes per sample is to avoid false negative results due to PCR inhibition, which appears to be a particular problem with samples from patients with severe hemorrhagic fever [46]. If the undiluted sample (140 µl) would be false negative due to inhibition, the 1/10-volume sample (14 µl) was expected to be positive due to the dilution of the inhibitor. In addition, running two PCRs in parallel on a patient sample enhances the reliability of the diagnostic process, as in most Lassa fever cases both samples were positive due to the high virus load (see below). RNA is extracted from samples and a negative control by the diatomaceous earth method [40] in the RNA extraction room. This method is inexpensive, and has been demonstrated to recover RNA over a broad concentration range with high efficacy [47]. The RT-PCR mastermix is prepared in the mastermix room and transferred to the extraction room for setting up the reaction. The closed PCR vials are transferred to the PCR thermocycler in the “grey” area. After completion of the PCR run, the closed vials are transferred to the detection room in the “dirty” area for agarose gel electrophoresis and gel documentation. PCR results are technically evaluated and transmitted to the clinics on a report form (Figure 1C). If either the undiluted or the 1/10-volume sample was positive, additional tests were performed to exclude PCR contamination. Gel pictures were transmitted to BNI staff via the internet for process monitoring. Standard laboratory procedures were defined in a set of quality management documents.
During 2009 through 2010, the laboratory has processed blood samples of 1650 patients. Testing a second sample was requested for 57/1650 patients; all other patients were tested once. Retrospective verification of positive test results by sequencing the PCR products revealed that in 13/1650 cases (0.8%), the result was probably false positive result due to PCR contamination. The sequences of the corresponding PCR fragments matched exactly that of the positive control or the sequence of a highly positive sample processed before. For data analysis, these samples were re-classified as negative, as well as samples which have initially been reported as indeterminate (n = 12; e.g. faint PCR signals that were not confirmed in a second blood sample). Applying these criteria, 1452 cases (88%) were Lassa RT-PCR negative, and 198 cases (12%) were positive. Undiluted and 1/10-volume RNA extract were positive in 138 (70%) cases and one of both was positive in 60 (30%) of the Lassa fever cases. The outcome was known for 170 cases with confirmed Lassa fever: 61 died and 109 survived. Thus, the CFR is 31% if calculated based on all Lassa fever patients or 36% if calculated based on cases with known outcome.
The vast majority of the patients came from Edo State, in particular from the Local Governmental Areas (LGA) surrounding ISTH, namely Esan West, Esan Central, Esan North East, Etsako West, and Owan West (Figure 2). Several patients also came from the neighboring state of Ondo, and a few samples were sent from other parts of Nigeria. The median age of Lassa fever patients was 32 years. Those who died from the disease were older than those who survived (p = 0.005) (Figure 3). The proportion of males and females was equal, and no association with outcome was observed. Children and students made up about one third of the patients. Adult patients had various professions and no specific profession was associated with the outcome of Lassa fever. When comparing the demographic data of Lassa fever negative versus positive patients, no statistically significant differences were observed with the exception of age, which tended to be higher among the positive patients (p = 0.006).
Patients presented at the hospital 5 days (median) after onset of symptoms. Lassa fever positive patients presented slightly later than those who tested negative (median difference 1.5 days, p = 0.009). A blood sample for Lassa fever diagnostics was taken from 75% of the patients at the day of presentation or the following day without differences among the groups. Lassa fever patients stayed longer in hospital and had a longer duration of illness than those who tested negative (p = 0.002 and p = 0.01, respectively). Clear differences in these two categories were observed between fatal cases of Lassa fever and survivors. Survivors stayed 10 days in hospital and had duration of illness of 16 days, while patients died from Lassa fever 2 days after admission and 10 days after onset of symptoms (figures are median; p<0.0001 and p<0.0001, respectively) (Figure 3). The time window for therapeutic intervention was extremely short: 25% of the fatal cases died one day after admission and the same day the blood sample was taken for Lassa virus RT-PCR; 75% died within 4 days of hospitalization and within 3 days after sampling for PCR.
Patients with signs of Lassa fever were seen in virtually all clinical and outpatient departments of ISTH. The vast majority of requests for Lassa fever testing came from the emergency departments, the medical wards, and the outpatient departments. Lassa fever patients were significantly more frequent among patients attending the adult emergency department (detection rate 19%; p<0.0001) while they were underrepresented among patients attending the outpatient departments (detection rate 5%; p<0.0001). Thus, nearly two thirds of laboratory-confirmed Lassa fever cases were seen in the emergency departments of ISTH.
Median axilliary body temperature recorded at the time of presentation was 37.5°C for all patients tested. Thus, most of them had a body temperature lower than indicated in the case definition (≥38°C). Body temperature was slightly higher for those who tested positive (p = 0.003). However, patients who had a fatal outcome had 0.8°C lower temperature (median difference) than those who survived (p<0.0001) (Figure 3). Indeed, nearly half of the fatal cases had normal axilliary temperature (35.5°C–37°C) [48] and 75% had ≤38°C. A few had hypothermia (≤35.5°C).
The CFR among Lassa fever patients was three times higher than the fatality rate among patients who tested negative for Lassa fever (p<0.0001), and negative patients were 10-times less frequently admitted to hospital than positive patients (p<0.0001). Ribavirin treatment was given to nearly all laboratory-confirmed Lassa fever patients, although several patients who tested negative received the drug as well, at least initially. While all survivors received the drug, 23% of Lassa fever patients with fatal outcome did not receive ribavirin because they died the day of presentation or the next day.
The 1/10-volume sample (defined as 2+ score) was more frequently positive in fatal cases than in survivors (p = 0.001), indicating higher virus load in fatal cases. Clinical symptoms reported at presentation largely matched the known symptoms of Lassa fever [10], [14]–[20]. The only symptom that was significantly more frequent among fatal cases was bleeding (p = 0.0001). Urea and creatinine blood levels were available for a subset of patients. Both values were clearly elevated in fatalities compared to survivors (p<0.0001 for urea and creatinine) (Figure 3).
Two levels of seasonality were observed (Figure 4). First, the total number of tests followed a seasonal pattern. From April through October, about 25% fewer patients were tested than in the remaining months. Second, the percentage of Lassa fever positive samples dropped by about 50% in the same period. Both effects led to a seasonal pattern with high Lassa fever incidence in the hospital from November through March (dry season) and low incidence from April through October (rainy season).
The short PCR fragments obtained in routine diagnostics were sequenced and subjected to phylogenetic analysis. Sequences from Edo and Ondo State cluster within lineage II (Figure 5). New sequences obtained from samples sent from Adamawa State (Nig11–205 and Nig11–208 from Yola) and Ebonyi State (Nig11–186 from Abakaliki and Nig10–148 from Izzi) also cluster with lineage II, while sequences of samples from Nasawara State (Nig09–072 from Akwanga) and the Federal Capital Territory (Nig09–121 and Nig09–193 from Abuja) cluster with lineage III. This is in agreement with the known geographical distribution of Lassa virus lineages in Nigeria [3], [49]. In addition, one sequence from Edo State (Nig09–045) was found to cluster with the new putative lineage that was recently described in Edo State and is defined by sequence Nig05-A08 (marked with “?” in Figure 5) [49]. All other sequences from Edo and Ondo State form a separate clade within lineage II that can be further subdivided into three clusters (A, B, and C), although the posterior probability support for cluster C is weak (Figure 6). The analysis of larger sequences might substantiate this tree topology. Strains within these three clusters do not show a strict geographical clustering, presumably because the sequences are too short to further resolve the relationships. However, there are a few visible associations between the origin of the strains and their phylogeny. Cluster A contains strains from Esan West and Uhunmwode, cluster B contains mainly strains from Esan Northeast and Central, while cluster C contains strains from all parts of Edo State as well as the strains from Ondo State. Even though the sequences are too short for an association study, it is worth mentioning that there is no obvious clustering of strains from patients with fatal outcome; they are randomly distributed over the whole phylogenetic tree.
Pilot investigations initiated in 2003 by the University of Lagos, ISTH, and BNI suggested that Edo State is a hot-spot for Lassa fever [31]. These data led to the decision to establish at ISTH a routine diagnostic service for Lassa fever to facilitate appropriate case management. It was further decided to use PCR as it offers case detection at an early stage [24], [25]. The training of staff members of ISTH in theory and practice of diagnostic PCR in partner institutions and on-site has been of paramount importance for the implementation of this technology at ISTH. Featuring separate areas for virus inactivation, RNA extraction, and mastermix preparation, the laboratory complies with basic standards of diagnostic PCR facilities. To guarantee smooth operation of the laboratory, technologies and workflow of the diagnostic facility at BNI in Hamburg were “mirrored” at ISTH, with only minor modification. This strategy facilitated training, interchangeability of protocols, and troubleshooting.
The most recent version of the GPC gene-specific RT-PCR assay [28] was chosen as a PCR assay and samples were tested using the regular extraction volume and 1/10-volume to minimize the problem of PCR inhibition [46]. Indeed, several samples tested positive with the 1/10-volume RNA preparation only and contamination was excluded in most of them by sequencing. It is likely that in these cases amplification of the undiluted sample was inhibited. A modification to the BNI procedures has been the use of diatomaceous earth for RNA preparation instead of a commercial kit to reduce the costs [40].
The main problem that arose during operation of the laboratory was PCR contamination, which was confirmed by sequencing of the PCR products. It was probably facilitated by the high analytical sensitivity of the assay (15 virus genome copies per reaction are detected with a likelihood of 95%) [28] and the need to open the reaction tubes for detection in agarose gel. The staff of the laboratory was aware of this inherent problem of PCR and protocols were developed to minimize and cope with it. We found retrospective evidence for reporting a false positive result in about 1% of the cases tested, which corresponds to an analytical specificity for the whole diagnostic process of 99% and an overall positive predictive value of 90%. However, with one exception, all contaminations manifested in only one of the two reactions (undiluted and 1/10-volume) performed on each sample. Thus, the positive predictive value for the majority of positive samples, namely those positive in both reactions is 99%, while the positive predictive value for samples positive in only one of the two reactions is 80%. Overall, we consider these good performance characteristics for a PCR diagnostics performed in a resource-limited setting.
Complementing the RT-PCR by antibody testing would further improve the reliability of the diagnostics. On the one hand, serology may be used to confirm the PCR diagnosis in patients who have already developed antibodies during the acute phase. On the other hand, antibody testing facilitates detection of patients in the convalescent stage when the virus load has dropped below the detection limit of the PCR [25], [26], [50].
The demographic data of the patients did not provide clues as to risk factors associated with Lassa fever; profession, geographic origin, and gender did not differ significantly between patients who tested positive and those who tested negative. However, a seasonal pattern of Lassa fever incidence was observed with the lowest number of cases during April through October, which corresponds to the rainy season. Similar, though not identical seasonal fluctuations have been described in Sierra Leone and Guinea [14], [15], [51]. The reason for a decrease in incidence during rainy season is not clear. Behavioral changes may play a role, as the number of all patients tested also decreased during this time, which may suggest that patients attend the hospital less frequently in rainy than in dry season. In addition, rodent dynamics and climate factors influencing the efficacy of virus transmission from the reservoir to humans may be involved [2], [52].
The CFR of 31% is high, though in the range of previous reports. In hospitalized patients with endemic Lassa fever, the CFR ranged from 9.3% to 18% [14]–[16], [51], [53], [54]. During nosocomial outbreaks, the CFR appears to be higher, ranging from 36% to 65% [9]–[12]. However, these figures are not fully comparable, due to differences in case definitions and diagnostic methods used in the various studies. Half of the patients with febrile illness attended ISTH at day 5 after onset of symptoms or later, and 50% of those with Lassa fever even at day 6 or later. The efficacy of ribavirin treatment decreases with progression of the disease and is hardly effective after day 6 [23]. Thus, a large number of Lassa fever patients attending ISTH did not benefit as greatly as they could have from the administration of ribavirin early in their disease course. This may explain the CFR of one third. Indeed, the time for therapeutic intervention is extremely short, as 50% of the fatal cases die before day 10 of illness and within 2 days in hospital—often before ribavirin treatment could be commenced. The severity of Lassa fever also explains why most of the patients attend the emergency department rather than the general outpatient departments.
A few parameters were identified which differ significantly between fatal cases of Lassa fever and survivors and are not yet documented in the literature. An important finding was lower body temperature in fatal cases. Often the temperature was not elevated at all or even below the normal range (<35.5°C). This resembles the sepsis-associated hypothermia which is a predictor of poor outcome [55], [56]. Although hypothermia is common in end-stage shock and organ failure of any etiology and not specific for Lassa fever, this sign has implications for the case definition of Lassa “fever”, which apparently needs to be revised to facilitate sensitive detection of cases in the terminal stage. Another factor associated with fatal outcome was higher age. Elderly people are also at higher risk of dying from sepsis, which is thought to be related to a reduced immune status or immune dysfunction [57], [58]. Bleeding was identified as the only clinical sign at presentation associated with poor outcome. Creatinine and blood urea levels were strongly elevated in the fatal cases suggesting renal failure. The semiquantitative PCR data also indicated higher virus load in patients with poor prognosis, which is consistent with published data [26].
The burden of Lassa fever in most regions of Nigeria is not known, as hospitals are not able to detect Lassa fever patients by laboratory testing. In future, surveillance systems including laboratory confirmation in reference centers need to be implemented in the country. Taken together, the data from ISTH indicate that fatal cases of Lassa fever are characterized by the following criteria:
We propose to use the above set of criteria as a surveillance tool to identify hospitals that are attended by Lassa fever patients. While this case definition is less sensitive, as it targets fatal cases only (“the tip of the iceberg”), it may be more specific for Lassa fever than existing ones.
The sequences generated from the short PCR products confirmed previous studies showing that Lassa virus strains from Edo State cluster phylogenetically with lineage II [3], [49]. Although the sequences originate from the same geographical area, they are quite diverse, which is in agreement with previous reports [3], [6]. A first attempt to correlate Lassa virus sequences with outcome did not reveal associations at least on the level of the clades that were resolved by the phylogenetic program. More sophisticated studies are warranted to look into possible links between virus genetics and clinical presentation.
In conclusion, routine diagnostics for Lassa fever has been established at ISTH. There are two major advantages for case management. First, early detection of a Lassa fever case improves protection of staff from nosocomial Lassa virus transmission. In the pre-diagnostic era, an unrecognized Lassa fever patient may have been cared for on a regular ward for several days, before clinical signs raised the suspicion of Lassa fever and appropriate measures were taken. Now, Lassa fever cases are transferred immediately to a specific ward where they are appropriately managed. In addition, close contacts to the Lassa fever patients, including hospital staff, can be monitored or offered ribavirin post-exposure prophylaxis as early as possible if the contact was very close [39]. Second, ribavirin treatment can be commenced early in all Lassa fever cases or may be terminated in non-cases if it had been provisionally commenced on clinical suspicion. Treatment is no longer based on clinical criteria, which is neither sensitive nor specific. However, rapid on-site diagnosis alone probably does not reduce the case fatality rate as long as most Lassa fever patients present too late for ribavirin treatment to be efficacious.
The open PCR platform may be used also for molecular testing for other pathogens. In addition, it provides the basis for research involving the Lassa fever patient and the optimization of the supportive treatment, including renal dialysis and intensive care. Steps to upgrade the laboratory with equipment for viral load determination, serology, blood chemistry, and hematology have been undertaken.
|
10.1371/journal.pgen.1003589 | H-NS Can Facilitate Specific DNA-binding by RNA Polymerase in AT-rich Gene Regulatory Regions | Extremely AT-rich DNA sequences present a challenging template for specific recognition by RNA polymerase. In bacteria, this is because the promoter −10 hexamer, the major DNA element recognised by RNA polymerase, is itself AT-rich. We show that Histone-like Nucleoid Structuring (H-NS) protein can facilitate correct recognition of a promoter by RNA polymerase in AT-rich gene regulatory regions. Thus, at the Escherichia coli ehxCABD operon, RNA polymerase is unable to distinguish between the promoter −10 element and similar overlapping sequences. This problem is resolved in native nucleoprotein because the overlapping sequences are masked by H-NS. Our work provides mechanistic insight into nucleoprotein structure and its effect on protein-DNA interactions in prokaryotic cells.
| The information required to build and maintain a cell is written into an organism's DNA in the form of genes. When individual genes are “read,” the DNA code is transcribed into an mRNA molecule by RNA polymerase. Hence, the DNA sequence adjacent to the start of a gene must contain a signal to recruit RNA polymerase. In certain instances this signal is difficult to differentiate from the background DNA sequence. For example, many bacterial chromosomes contain discrete sections of DNA with a high percentage of A and T nucleotides. Because RNA polymerase recognises an AT-rich signal sequence, these chromosomal regions can be ambiguous. In this paper we address the long-standing question of how RNA polymerase specifically recognises such DNA target sites. We show that a crucial factor is local nucleoprotein organisation. Hence, the manner in which DNA is folded, in conjunction with primary DNA sequence, facilitates specific RNA polymerase interactions with DNA.
| Transcription is initiated by binding of RNA polymerase to specific DNA sequences known as promoters [1]. Following promoter recognition the resulting complex undergoes a process of isomerisation. Hence, ∼14 base pairs (bp) of DNA, close to the transcription start site, are unwound [2]. RNA polymerase then engages in abortive cycles of initiation before escaping the promoter to form an elongation complex [3]. It has long been known that promoter unwinding is facilitated by the weak base stacking interactions associated with AT-rich DNA. Thus, the eukaryotic TATA box (5′-TATAAA-3′) is unwound during transcription initiation [4]. Similarly, the prokaryotic −10 hexamer (5′-TATAAT-3′), recognised by Domain 2 of the RNA polymerase σ70 subunit, participates in DNA opening [5]. Because DNA elements recognised by RNA polymerase are AT-rich, chromosomal regions, where DNA AT-content is unusually high, prove particularly challenging templates for recognition. For example, the horizontally acquired sections of some bacterial chromosomes have an elevated AT-content. As a result, RNA polymerase may bind cryptic promoters [6] or initiate transcription promiscuously [7].
In Escherichia coli, gene regulatory regions are targeted by chromosome folding proteins [8]. Hence, in addition to their architectural role, these proteins can influence RNA polymerase-DNA interactions [9]. The Histone-like Nucleoid Structuring (H-NS) protein recognises AT-rich DNA and is associated with horizontally acquired genes [10]–[13]. The prevailing view is that, when bound at such regions, H-NS silences transcription [14]. However, the precise mechanism remains elusive; models proposing exclusion of RNA polymerase from, and trapping of RNA polymerase at, H-NS bound regions have both been proposed [15]. Since these models are not mutually exclusive a third possibility is that a myriad of different configurations exist. Interestingly, two recent studies have reported close association between RNA polymerase and H-NS [16], [17]. In one case, H-NS stimulated rather than repressed gene expression [17].
In this work we describe an undocumented role for H-NS; facilitating the correct recognition of promoters by RNA polymerase. The ehxCABD operon from Shiga toxin-producing E. coli (STEC) has an unusually high AT-content. Consequently, the operon regulatory region contains multiple sequences that resemble −10 hexamers. We show that, despite the apparent ambiguity of this DNA template, RNA polymerase initiates transcription specifically from a single promoter in vivo. However, in vitro, RNA polymerase is unable to differentiate between this promoter and adjacent binding sites. We show that H-NS plays a critical role by blocking access of RNA polymerase to the adjacent binding sites. Thus, H-NS ensures correct positioning of RNA polymerase.
The ehxCABD operon is located on the pO157 plasmid and its derivatives. The operon encodes an enterohemolysin and proteins for its post-translational modification and export [18]. The 248 bp of regulatory DNA immediately upstream of the operon has an AT-content of 71%. H-NS has been implicated in regulating expression of the operon but a comprehensive molecular analysis is lacking [19]–[21]. As a first step we determined which section of the regulatory DNA contained promoter activity. Note that the ehxCABD regulatory DNA has an almost identical sequence in multiple E. coli serotypes and we arbitrarily used the ehxCABD regulatory sequence described by Rogers et al. [20]. We began by generating DNA fragments carrying discrete sections of the ehxCABD regulatory region (illustrated in Figure 1Ai). The fragments encompass 248 bp of DNA adjacent to the first gene in the operon (fragment F1), the downstream part of this region (fragment F2) or the upstream section of the locus (fragment F3). We assayed each fragment for promoter activity using two plasmid based systems (illustrated in Figure 1Aii). Hence, pRW50 and pLux encode the reporter proteins β-galactosidase and Luciferase respectively. Note that pRW50 was used to report promoter activity in E. coli K-12 whilst pLux was used with E. coli O157:H7 as a control for effects of STEC encoded transcriptional regulators. The raw activity data, for each DNA fragment, in each plasmid, is summarised in Figure 1B. Our results show that the F1 and F3 fragments stimulate transcription, to similar levels, in all of the assays. No detectable transcription was driven by the F2 fragment. Therefore, the ehxCABD promoter must be located in the upstream portion of the regulatory region common in both F1 and F3.
Our next aim was to identify transcription start sites in the F3 fragment. To do this we conducted mRNA primer extension experiments. We used RNA extracted from E. coli JCB387 cells, carrying the F3 fragment cloned in plasmid pRW50. Our analysis yielded two extension products of 155 and 154 nucleotides (nt) in length (Figure 1C). The transcript start, corresponding to the more abundant 154 nt extension product, is labelled +1 in Figure 1D. A consensus extended promoter −10 element (5′-TGnTATAAT-3′) was found 8 bp upstream of the transcription start site. A four out of six match to a promoter −35 element (5′-TTGACA-3′) was observed further upstream. Throughout this work we refer to this promoter, highlighted green in Figure 1D, as PehxCABD. The two primer extension products, differing in length by a single nt, both likely originate from this promoter. Importantly, we confirmed that PehxCABD was the only promoter present in the F1 fragment. Thus, using RNA extracted from E. coli JCB387 cells carrying the F1 fragment cloned in plasmid pRW50, we observed only primer extension products corresponding to PehxCABD (Figure S1).
Our primer extension analysis shows that, in vivo, RNA polymerase initiates ehxCABD transcription with precision (Figure 1). This is remarkable given the abundance of potential −10 hexamer sequences in this regulatory region (two such sequences are highlighted red in Figure 1D). To better understand how specificity is achieved we examined recognition of the naked F3 fragment by RNA polymerase. We utilised two in vitro DNA footprinting techniques. First, we exploited the properties of Fe2+ chelated Bromoacetamidobenzyl-EDTA (FeBABE). FeBABE is a DNA cleavage reagent that can be attached to specific cysteine side chains in proteins. Once attached, FeBABE cleaves nucleic acids within a 12 Å radius of the attachment site. Thus, FeBABE conjugated with the RC461 derivative of E. coli σ70 cleaves promoter −10 elements [22]. Figure 2Ai shows the pattern of FeBABE cleavage observed with the F3 fragment. As expected, the PehxCABD −10 element was cleaved (highlighted by green box in Figure 2Ai). However, we also observed DNA cleavage at additional sites overlapping PexhCABD (highlighted by red stars in Figure 2Ai). In complementary experiments KMnO4 footprinting was used to detect DNA unwinding by RNA polymerase. We observed DNA melting at the PehxCABD −10 element (highlighted by a green box in Figure 2Aii) and at additional sites (highlighted by yellow stars in Figure 2Aii). It did not escape our attention that the additional sites of FeBABE and KMnO4 reactivity align with each other and with sequences that resemble −10 hexamers highlighted in Figure 1D. Nevertheless, we were concerned that the additional FeBABE and KMnO4 reactivity signals might originate from RNA polymerase bound at PehxCABD. To exclude this possibility we ran identical reactions with unrelated cbpA P6 promoter DNA. In these experiments no DNA cleavage products were observed other than those at the cbpA P6 −10 hexamer. We conclude that the naked PehxCABD F3 fragment must contain multiple overlapping RNA polymerase binding sites.
Factors present in vivo must influence RNA polymerase interactions with PehxCABD. Such factors may explain why the additional RNA polymerase binding sites observed in vitro do not generate transcripts in vivo. Our attention turned to H-NS, which is known to recognise AT-rich regulatory regions and influences ehxCABD expression [19]–[21]. Thus, we used chromatin immunoprecipitation (ChIP) to measure binding of RNA polymerase and H-NS to PehxCABD in vivo. Recall that, in ChIP experiments, a cell's nucleoprotein is cross-linked with formaldehyde, extracted, and then fragmented by sonication. Antibodies directed against the protein of interest are then used to select DNA fragments with which the protein is cross-linked. Finally, PCR is used to identify recovered DNA fragments. Figure 3A shows PCR analysis of DNA immunoprecipitated with anti-RNA polymerase (β subunit) or anti-H-NS. Control experiments, in which we analysed total cellular DNA, or DNA recovered from a mock immunoprecipitation, are also shown. The PehxCABD DNA is detected in the total DNA sample, the anti-β, and anti-H-NS immunoprecipitates. Importantly, the PehxCABD DNA was not detected in the mock immunoprecipitate. In a set of control PCR reactions we probed the lacZ and yabN loci. Note that these loci are not transcribed in the conditions used here and are not bound by H-NS. As expected, lacZ and yabN were not detected in the immunoprecipitates.
We next reconstituted co-association of RNA polymerase, H-NS and PexhCABD in vitro. Electrophoretic Mobility Shift Assays (EMSA) were used to probe the complexes formed. The result is shown in Figure 3B. The data show that RNA polymerase (lane 2) and H-NS (at two different concentrations, lanes 3 and 5) form distinguishable complexes with the DNA. When H-NS and RNA polymerase are added in unison an additional complex can be detected (boxed in lanes 4 and 6). To confirm that this additional complex contained both H-NS and RNA polymerase the band was extracted, submitted to tryptic digest, and the resulting peptides analysed by mass spectrometry. Both RNA polymerase and H-NS were present in the excised band.
To more precisely understand the ternary H-NS-RNA polymerase-DNA complex we repeated our σ70RC461-FeBABE analysis. The data show that, in the presence of H-NS, the signal for RNA polymerase binding at the PehxCABD −10 element is retained. Conversely, binding of RNA polymerase at adjacent sites is lost (Figure 4A). In a complementary experiment we used DNAse I footprinting to locate H-NS binding in the absence of RNA polymerase. The data show that H-NS recognises the same AT-rich region, extending from +10 to −30, as the transcriptional apparatus (Figure 4B). Thus, the binding sites for H-NS and RNA polymerase overlap. To assess how H-NS effects RNA polymerase interactions with PexhCABD in vivo we repeated our primer extension analysis. We used RNA extracted from wild type E. coli K-12 and cells lacking hns. As described above, RNA from wild type cells yielded two extension products of 155 and 154 nt in length (Figure 4C lane 5). These extension products were also observed when we analysed RNA from Δhns cells (Figure 4C lane 6). Strikingly, RNA from Δhns cells yielded a further 9 extension products of between 138 and 194 nt in length. These additional primer extension products align with the additional sites of RNA polymerase binding observed in Figure 4A. Finally, it is noteworthy that, in order to observe the primer extension products in Lane 6 of Figure 4C, we had to “overload” the sample onto the gel. This suggests that the net result of reduced RNA polymerase binding specificity is a reduction in transcription. Consistent with this, we observed reduced expression from the F3 fragment, in cells lacking H-NS, using our LacZ reporter assay (Figure S2).
Our data suggest that PehxCABD is flanked by at least two overlapping elements that can bind RNA polymerase. If this model is correct there should be competition between RNA polymerase molecules for binding the various targets. A logical consequence of this competition would be reduced transcription from PehxCABD. To test this model we disrupted either the PehxCABD −10 hexamer or the overlapping RNA polymerase binding elements. The mutations utilised are illustrated in Figure 1D. Figure 5A shows LacZ activity data from wild type E. coli cells carrying the various promoter::lacZ fusions. The -41G mutation increases LacZ expression that is further increased by the -7T-5T-4T mutations. Conversely, the -13G mutation, in the canonical PehxCABD −10 element, reduces LacZ expression.
We next sought to confirm the stimulatory effect of H-NS on specific recognition of PehxCABD by RNA polymerase. Thus, we compared the effects of H-NS and the -41G mutation using in vitro transcription assays. The F3 DNA fragment was cloned upstream of the λoop terminator in plasmid pSR. In the context of this construct PehxCABD produces transcripts, of 178/179 nt in length, that can be quantified after electrophoresis. Additional transcripts, corresponding to the Δhns primer extension products in Figure 4C, should also be generated. On this basis, we expected to detect an abundant 162 nt transcript (corresponding to the 138 nt extension product in Figure 4C) and scarce transcripts sized between 183 nt and 218 nt (equivalent to the primer extension products in the 159–194 nt range). The results of the analysis with and without H-NS are shown in Figure 5Bi alongside a set of “marker” transcripts (Lane 1). Lane 2 shows the result in the absence of H-NS. As expected we observed two intense bands corresponding to the 178/179 and 162 nt products. Note that because the bands in the 183–218 nt range are less abundant and poorly resolved in this assay they were not clearly visible. The 108 nt “RNAI” transcript is from the pSR replication origin and acts as an internal control. Addition of H-NS to the reactions specifically stimulated transcription from PehxCABD (Lanes 2–5). Figure 5Bi shows the effect of the -41G mutation, it is indistinguishable from the effect of H-NS. Note that both the addition of H-NS, and the introduction of the -41G mutation, resulted in a decrease in the relative abundance of the 162 nt transcript compared to the RNAI control transcript (Figure 5B).
The data presented here demonstrate that nucleoprotein organisation, as well as primary DNA sequence, controls the specificity of regulatory DNA for RNA polymerase. In our model, RNA polymerase competes with itself for binding to AT-rich sequences overlapping PehxCABD (Figure 6). In the context of native nucleoprotein this self-competition is negated. This is because RNA polymerase has instead to compete with H-NS (Figure 6). Hence, evolution of RNA polymerase binding targets likely involves a trade-off between attaining the optimal DNA sequence for correct chromosome folding and precise transcription initiation. We note the PehxCABD has a consensus extended −10 element. Such sequences are incredibly rare, being found in only 3 of the 554 documented promoters in E. coli [23]. We speculate that, in very AT-rich gene regulatory regions, closer matches to the consensus RNA polymerase recognition elements are highly beneficial. Thus, in the presence of H-NS, RNA polymerase is able to recognise PehxCABD because of its close similarity to a consensus promoter. Conversely, adjacent AT-rich sequences are ignored. Interestingly, the net effect of H-NS on transcription from PehxCABD is positive and this results from correct positioning of RNA polymerase by H-NS (Figures 4 and 5). Park and co-workers [17] recently documented a mechanism for positive regulation of malT by H-NS. Although H-NS exerts its effect on malT by binding the malT mRNA there are some clear parallels with the mechanism described here. Hence, the incoming ribosome is unable to correctly recognise the 5′ end of the malT mRNA because the Shine Dalgarno sequence is ambiguous. H-NS corrects mispositioning of the ribosome by binding to an adjacent AU-rich element. We note that the effect of H-NS on binding of RNA polymerase to PehxCABD is similar to the effect of CRP on binding of RNA polymerase to the acsP2 promoter [24]. However, the molecular mechanisms underlying the effects are different. Hence, at acsP2, CRP makes direct contacts with RNA polymerase that ensure it engages the promoter precisely.
Rogers et al. [20] previously studied a 1338 bp DNA fragment carrying 126 bp of the ehxCABD gene regulatory region, the entire 516 bp ehxC gene, and 695 bp of ehxA. The fragment was fused to lacZ and, on detection of LacZ expression, it was concluded that a promoter must be located within the 126 bp regulatory section of the 1338 bp fragment. We show that, when examined in isolation, the 126 bp of DNA immediately upstream of ehxC is not able to promote transcription (see the F2 fragment in Figure 1). Similarly, no mRNA species were found to originate in this section of the regulatory region (highlighted blue in Figure S1). Thus, the only plausible explanation for the observations of Rogers et al. is that they unwittingly measured transcription from spurious promoters located within the AT-rich ehxCABD coding sequence. More recently, Iyoda and co-workers [21] examined the full ehxCABD regulatory region (similar to our F1 fragment). The authors found that deleting the upstream part of the regulatory region greatly reduced transcription. Building on the assumptions of Rogers et al. (2009) the authors presumed that they had removed the binding site for a transcriptional activator. A speculative binding site for the activator was identified; this sequence aligns with the PehxCABD consensus extended −10 hexamer. Clearly, a more likely explanation is that Iyoda and co-workers had simply removed PehxCABD. Taken together, these data suggest that control of ehxCABD expression is more complex than previously thought. In particular, the possibility that additional promoters exist within the ehxCABD coding sequence is intriguing [20]. Should any such promoters be repressed by H-NS, as suggested by Rogers et al. [20], this would further ensure specific transcription initiation from PehxCABD. We also speculate that small differences in the DNA sequence of the ehxCABD regulatory region, in different E. coli isolates, may provide information about how H-NS regulated promoter regions evolve. Further biochemical and genetic dissection of the ehxCABD locus should provide the necessary insight.
Wild type E. coli strains JCB387 and M182 have been described previously [25], [26]. The Δhns M182 derivative (JRG4864) is described by Wyborn et al. [27]. Plasmids pRW50 and pLux are described by Lodge et al. [28] and Burton et al. [29] respectively. Plasmid pSR is described by Kolb et al. [30]. More detailed descriptions of strains and plasmids are provided in Table S1.
H-NS and RNA polymerase were prepared as described previously [22], [25]. DNA fragments for footprinting and EMSA experiments were derived from Qiagen maxi-preparations of plasmid pSR. Thus, the ehxCABD F3 fragment was excised from pSR by sequential digestion with HindIII and then AatII. After digestion fragments were labelled at the HindIII end using [γ-32P]-ATP and polynucleotide kinase. DNAse I and KMnO4 footprints were then performed as described by Grainger et al. [25]. FeBABE footprinting reactions were completed according to the methodology of Bown et al. [22]. Radio-labelled DNA fragments were used at a final concentration of ∼10 nM. Note that, apart from the KMnO4 reactivity assays, all in vitro DNA binding reactions contained a vast excess (12.5 µg ml−1) of Herring sperm DNA as a non-specific competitor. We checked that our reaction conditions were meaningful by comparing the affinity of H-NS for PehxCABD and the well-characterised H-NS target proU. We found that the affinity of H-NS for the two DNA fragments was similar in our conditions (Figure S3). Footprints were analysed on a 6% DNA sequencing gel (molecular dynamics). The results of all footprints and EMSA experiments were visualized using a Fuji phosphor screen and Bio-Rad Molecular Imager FX.
The in vitro transcription experiments were performed as described previously Savery et al. [31] using the system of Kolb et al. [30]. A Qiagen maxiprep kit was used to purify supercoiled pSR plasmid carrying the different promoter inserts. This template (∼16 µg ml−1) was pre-incubated with purified H-NS in buffer containing 20 mM Tris pH 7.9, 5 mM MgCl2, 500 µM DTT, 50 mM KCl, 100 µg ml−1 BSA, 200 µM ATP, 200 µM GTP, 200 µM CTP, 10 µM UTP with 5 µCi [α-32P]-UTP. The reaction was started by adding purified E. coli RNA polymerase. Labelled RNA products were analysed on a denaturing polyacrylamide gel.
Luciferase assays were done as described by Burton et al. [29] using E. coli O157:H7. β-galactosidase assays were completed using the protocols of Miller [32] with E. coli JCB387, M182 or the Δhns derivative. All assay values are the average of three independent experiments and, in all cases, cells were grown aerobically, at 37°C, in LB media. The ehxCABD F1 fragment was synthesised by DNA2.0 (USA). The F3 fragment was generated using overlapping oligonucleotides (5′-ggctgcgaattctatcttacaaatcaatcatctgagtgttataatataacttagctgtgatatgtgtaagaatgtttaggcaat-3′ and 5′-cgcccgaagcttcatctctcccaaccaaaacaacattagcgataataatatattgcctaaacattcttacacatatca-3′). Similarly, F2 was generated using 5′-ggctgcgaattctgtttttagatgcttcttgcttaaaagaatataattcctgttcttttatatagagttctttaca-3′ and 5′-cgcccgaagcttcataatgtttaaacaaataagaaaattcagtaaatgtaaagaactctatataaaagaac-3′. Mutations were introduced using derivatives of these oligonucleotides. All ehxCABD regulatory region sequences are numbered with respect to the transcription start point (+1) and with upstream and downstream locations denoted by ‘−’ and ‘+’ prefixes respectively.
Transcript start sites were mapped by primer extension, as described in Lloyd et al. [33], using RNA purified from strains carrying the F3 DNA fragment cloned in pRW50. The 5′ end-labelled primer D49724, which anneals downstream of the HindIII site in pRW50 was used in all experiments. Primer extension products were analysed on denaturing 6% polyacrylamide gels, calibrated with size standards, and visualized using a Fuji phosphor screen and Bio-Rad Molecular Imager FX.
Chromatin Immunoprecipitation was done exactly as described previously [8], [34]. Briefly, formaldehyde crosslinked nucleoprotein, obtained from growing JCB387 cells carrying the F3 fragment in plasmid pRW50, was fragmented by sonication. Some of this sample was retained as the “total DNA” fraction. DNA cross-linked with RNA polymerase or H-NS was then precipitated using a rabbit polyclonal antibody against H-NS or an antibody against the RNA polymerase β-subunit (Neoclone). A control mock immunoprecipitation (with no antibody) was done in parallel. After immunoprecipitation the protein-DNA complexes were de-cross-linked and the DNA was recovered using a Qiagen PCR purification kit. Recovered DNA was resuspended in 50 µl of elution buffer and 1 µl of this solution was used as a template in a 50 µl PCR. The reactions were run for 28 cycles of amplification before 5 µl was loaded onto a 7.5% polyacrylamide gel. After electrophoresis PCR products were visualised with ethidium bromide. The oligonucleotide primers for amplification of the yabN [34] and lacZ [8] open reading frames, in their chromosomal context, have been described previously. To amplify PehxCABD we used 5′-ggctgcctcgagtatcttacaaatcaatcatctgagtgttataatataacttagctgtga-3′ and 5′-cgcccgggatcccatctctcccaaccaaaacacattagcg-3′.
|
10.1371/journal.pgen.1006025 | Mapping Topoisomerase IV Binding and Activity Sites on the E. coli Genome | Catenation links between sister chromatids are formed progressively during DNA replication and are involved in the establishment of sister chromatid cohesion. Topo IV is a bacterial type II topoisomerase involved in the removal of catenation links both behind replication forks and after replication during the final separation of sister chromosomes. We have investigated the global DNA-binding and catalytic activity of Topo IV in E. coli using genomic and molecular biology approaches. ChIP-seq revealed that Topo IV interaction with the E. coli chromosome is controlled by DNA replication. During replication, Topo IV has access to most of the genome but only selects a few hundred specific sites for its activity. Local chromatin and gene expression context influence site selection. Moreover strong DNA-binding and catalytic activities are found at the chromosome dimer resolution site, dif, located opposite the origin of replication. We reveal a physical and functional interaction between Topo IV and the XerCD recombinases acting at the dif site. This interaction is modulated by MatP, a protein involved in the organization of the Ter macrodomain. These results show that Topo IV, XerCD/dif and MatP are part of a network dedicated to the final step of chromosome management during the cell cycle.
| DNA topoisomerases are ubiquitous enzymes that solve the topological problems associated with replication, transcription and recombination. Type II Topoisomerases play a major role in the management of newly replicated DNA. They contribute to the condensation and segregation of chromosomes to the future daughter cells and are essential for the optimal transmission of genetic information. In most bacteria, including the model organism Escherichia coli, these tasks are performed by two enzymes, DNA gyrase and DNA Topoisomerase IV (Topo IV). The distribution of the roles between these enzymes during the cell cycle is not yet completely understood. In the present study we use genomic and molecular biology methods to decipher the regulation of Topo IV during the cell cycle. Here we present data that strongly suggest the interaction of Topo IV with the chromosome is controlled by DNA replication and chromatin factors responsible for its loading to specific regions of the chromosome. In addition, our observations reveal, that by sharing several key factors, the DNA management processes ensuring accuracy of the late steps of chromosome segregation are all interconnected.
| DNA replication of a circular bacterial chromosome involves strong DNA topology constraints that are modulated by the activity of DNA topoisomerases [1]. Our current understanding of these topological modifications comes from extensive studies on replicating plasmids [2, 3] These studies suggest that positive supercoils are formed ahead of the replication fork, while precatenanes are formed on newly replicated sister strands. At the end of a replication round, unresolved precatenanes accumulate in the region of replication termination and are converted to catenanes between the replicated sister chromosomes. Neither precatenanes or catenanes have been directly observed on chromosomes but their presence is generally accepted and failure to resolve them leads to chromosome segregation defects and cell death [4].
Topo IV is a type II topoisomerase formed by two dimers of the ParC and ParE subunits and is the main decatenase in Esherichia. coli [5]. in vitro, its activity is 100 fold stronger on catenated circles than that of DNA gyrase [6]. Topo IV activity is dependent on the topology of the DNA substrate; Topo IV activity is strongest on positively supercoiled DNA and has a marked preference for L-braids, which it relaxes completely and processively. Topo IV can also unlink R-braids but only when they supercoil to form L-plectonemes [7–9]. In vivo, DNA gyrase appears to have multiple targets on the E. coli chromosome [10–12], whereas Topo IV cleavage sites seem to occur less frequently [11]. Interestingly, Topoisomerase IV activity is not essential for replication itself [13] but is critical for chromosome segregation [14]. The pattern of sister chromatid separation has been shown to vary upon Topo IV alteration, leading to the view that precatenanes mediate sister chromatid cohesion by accumulating for several hundred kilobases behind the replication forks keeping the newly replicated DNA together [13, 15]. The regulation of Topo IV and perhaps the accessibility of the protein to chromosome dimers was proposed to be an important factor controlling chromosome segregation [15, 16]. Topo IV activity can be modulated by a number of proteins including MukB and SeqA. MukB, is an SMC-related protein in E. coli and is reported to bind to the C-terminus of Topo IV [17] to enhance Topo IV unlinking activities [18, 19]. MukB also appears to be important in favoring the formation of Topo IV foci (clusters) near the origin of replication [20]. SeqA, a protein involved in the control of replication initiation, and Topo IV also interact [21]. These interactions may play a role in sister chromatid segregation at the late segregating SNAP regions near the origin of replication of the chromosome [16].
Beside its role in the resolution of precatenanes, Topo IV is mostly required in the post-replicative (G2) phase of the cell cycle for the resolution of catenation links. Indeed, Espeli et al. showed that Topo IV activity is mostly observed during the G2 phase, suggesting that a number of catenation links persist after replication [22]. Recent cell biology experiments revealed that in G2, the terminal region (ter) opposite oriC segregates following a specific pattern [23–25]. Sister ter regions remain associated from the moment of their replication to the onset of cell division. This sister-chromosome association is mediated by the Ter macrodomain organizing protein, MatP [26]. At the onset of cell division, the FtsK DNA-translocase processes this region, releasing the MatP-mediated association. This process ends at the dif site, when the dimeric forms of the sister chromosomes are resolved by the XerC and XerD recombinases. A functional interaction between the MatP/FtsK/XerCD-dif system and Topo IV has long been suspected. FtsK interacts with Topo IV, enhancing its decatenation activity in vitro [27, 28] and the dif region has been reported as a preferential site of Topo IV cleavage [29]. This functional interaction has been poorly documented to date and is therefore remains elusive.
In this study we have used genomic and molecular biology methods to characterize Topo IV regulation during the Escherichia coli cell cycle on a genome-wide scale. The present work revealed that Topo IV requires DNA replication to load on the chromosome. In addition, we have identified two binding patterns: i) regions where Topo IV binds DNA but is not engaged in a cleavage reaction; ii) numerous sites where Topo IV cleavage is frequent. We show that Topo IV-mediated removal of precatenanes is influenced by both local chromatin structure and gene expression. We also demonstrate that at the dif site, Topo IV cleavage and binding are enhanced by the presence of the XerCD recombinase and the MatP chromosome-structuring factor. The enhancement of Topo IV activity at dif promotes decatenation of fully replicated chromosomes and through interaction with other DNA management processes, this decatenation ensures accurate separation of the sister chromosomes.
To identify Topo IV binding, we performed ChIP-seq experiments in ParE and ParC Flag tagged strains. The C-terminus fusions of ParE and ParC replaced the wild-type (WT) alleles without any observable phenotypes (S1 Fig). We performed three independent experiments, two ParE-flag IPs and one ParC-flag IP, with reproducible patterns identified in all three experiments. A Pearson correlation of 0.8, 0.9 and 0.7 was observed for ParC-ParE1, ParE1-ParE2 and ParC-ParE2 respectively. A map of enriched regions observed in each experiment is represented on Fig 1A (red circles). Four of the highly-enriched sites are illustrated at a higher magnification in Fig 1A—right panels. Interestingly one of these sites corresponds to the dif site (position 1.58Mb), which has previously been identified as a strong Topoisomerase IV cleavage site in the presence of norfloxacin [29]. We also observed strong enrichment over rRNA operons, tRNA and IS sequences. To address the significance of the enrichment at rRNA, tRNA and IS, we monitored these sites in ChIP-seq experiments performed in the same conditions with a MatP-flag strain and mock IP performed with strain that did not contain any flag tagged protein. Both MatP and Mock IP presented significant signals on rRNA, tRNA and IS loci (S2 Fig). This observation suggested that Topo IV enrichment at rRNA, tRNAs and IS was an artifact of the ChIP-Seq technique. By contrast no enrichment was observed at the dif site in the MatP and mock-IP experiments (S2 Fig), we therefore considered dif to be a genuine Topo IV binding site and compared every enriched region (>2 fold) with the dif IP. We filtered the raw data for regions presenting the highest Pearson correlation with the dif signal (>0.7). This procedure discarded many highly enriched regions (Fig 1A orange circles). We identified 19 sites throughout the chromosome where Topo IV IP/input signal suggested a specific binding for at least two of the experiments (Fig 1A, outer circle histogram, S1 Table). Most Topo IV binding sites span a 200 bp region. These sites frequently overlapped intergenic regions, with their mid-points located inside the intergenic region, and did not correlate with any identifiable consensus sequence. In addition to dif, which exhibited a 10-fold enrichment, three other sites were strongly enriched. These sites corresponded to positions 1.25Mb (9.4x), 1.85Mb (31x) and 2.56Mb (19x) on the chromosome (Fig 1A, right panels). Beside these specific sites, Topo IV IP showed non-specific enrichment in the oriC proximal half of the chromosome. This bias was not a consequence of locus copy number, as the enrichment remained after copy number normalization (Fig 1B). We used MatP-Flag IP [30] and a control IP in a strain that does not contain a Flag tagged gene to differentiate non-specific Topo IV binding from experimental noise (S3A Fig). In addition, Topo IV enrichment was also observed in GC rich regions of the chromosomes (S3B Fig). Importantly, the ori/ter bias was not a result of the GC% bias along the chromosome since it was still explicit after GC% normalization (S3C Fig). More precisely, the Topo IV binding pattern closely followed gene dosage for a ~3Mb region centered on oriC (S3D and S3E Fig and S1 Text). In the complementary ter-proximal region, gene dosage (input reads) was higher than the ChIP-seq profile, suggesting that the nonspecific Topo IV binding was lower or lasts for a shorter time in the cell cycle (since these data are population-averaged). The Terminus region that is depleted in Topo IV binding (1.6Mb) surpassed, by far, the size of the Ter macrodomain (800kb).
The influence of Topo IV on sister chromatid interactions [15] prompted the question of how Topo IV would follow replication forks and bind to the newly replicated sister chromatids throughout the cell cycle. We performed ChIP-seq experiments in E. coli dnaC2 strains under conditions suitable for cell cycle synchronization of the entire population. Synchronization was achieved through a double temperature shift, as described previously [15]. Using these conditions, in each cell, S phase is initiated on one chromosome, lasts for 40–45 min and is followed by a G2 phase (20 min) (S4 Fig). We analyzed ParE binding before the initiation of replication, in S phase 20 min (S20) and 40 min (S40) after the initiation of replication and in G2 phase. The synchronization of replication in the population was monitored by marker frequency analysis of the Input DNA (Fig 1C). The profile observed for bacteria that did not replicate at non-permissive temperature was strictly flat, but the S20 replication profile presented two sharp changes of the marker frequency slope around positions 500kb and 2700kb. This suggested that each replication fork had crossed approximately 1000 to 1300 kb in 20 min. The S40 replication profile demonstrated that most cells had finished replication, with the unreplicated region being limited to 300 kb around dif in no more than 20% of the bacteria. In G2 phase the marker frequency was flat. We used flow cytometry to demonstrate that at G2, the amount of DNA in each bacterium was double compared to that of the G1 bacteria, indicating that cytokinesis has not yet occurred (S4 Fig). We analyzed Topo IV binding at specific binding sites (Fig 1D). Binding at these sites was strongly impaired in the absence of replication. Binding at every site started in the S20 sample and was maximal in the S40 or G2 samples, without showing any marked decrease, even in the oriC-proximal region. These observations suggest that Topo IV binds to specific sites during S phase. However, since enrichment was observed for non-replicated loci and was maintained for a long time after replication, it was not compatible with a model of Topo IV migration with the replication forks. Synchronization experiments with a higher temporal resolution are required to clarify this observation.
To measure Topo IV cleavage at the binding sites, we took advantage of the fact that norfloxacin covalently links Topoisomerase II to the gate segment of DNA and prevent its relegation [31]. We first monitored Topo IV activity on the Topo IV enriched regions (1.2, 1.8, 2.5, 3.2 Mb and dif) by incubating bacteria with norfloxacin for 10 minutes before genomic extraction and performing Southern blot analysis to detect the cleaved DNA products [10, 29]. This revealed cleavage fragments induced by both DNA Gyrase and Topo IV poisoning in the WT strain, but only Topo IV cleavage in a nalR strain where DNA Gyrase is resistant to norfloxacin. Among the 5 tested sites, only two displayed clear Topo IV cleavage at the expected position (Fig 2A). As expected, the dif site exhibited strong cleavage. Moreover cleavage was also observed at position 2.56 Mb. However the 1.2, 1.8 and 3.2 Mb sites did not show any Topo IV mediated cleavage in the presence of norfloxacin.
The above result prompted us to investigate Topo IV cleavage at the genome-wide scale. We performed IPs in the presence of norfloxacin as a crosslinking agent instead of formaldehyde. Following this step, all downstream steps of the protocol were identical to that of the ChIP-Seq assay. We referred to this method as NorflIP. The NorflIP profile differed from the ChIP-seq profile (Fig 2B). Regions immunoprecipitated with Topo IV-norfloxacin cross-links were frequently observed (Fig 2C orange circle). Similarly to the ChIP-seq experiments, the NorflIP profile revealed strong enrichment over the rRNA operons and IS sequences but not at the tRNA genes (S5A Fig). We used a Southern blot cleavage assay to demonstrate that these signal did not correspond to Topo IV cleavages (S5B Fig). The NorflIP peaks correspond to a ~170 bp forward and reverse enrichment signal separated by a 130 bp segment, which is not enriched. This pattern is the consequence of the covalent binding of Topo IV to the 5’ bases at the cleavage site. After Proteinase K treatment the cleaving tyrosine residue bound to the 5’ extremity resulted in poor ligation efficiency and infrequent sequencing of the cleaved extremities. (S6A and S6B Fig) This observation confirmed that we were observing genuine Topoisomerase cleavage sites. We used this pattern to define an automatic peak calling procedure (S6C Fig) that identified between 134 and 458 peaks in the three NorflIP experiments, two experiments performed with ParC-Flag and one with ParE-Flag (Fig 2C purple circles and Fig 2D). We observed a total of 571 possible sites in the three experiments with about half of the sites common to at least two experiments and approximately 88 sites common to all three experiments (S1 Table). We analyzed sequencing reads for the three experiments around the dif, 0.2 Mb and 1.92Mb positions. It revealed abrupt depletions of forward and reverse reads in a 100bp center region suggesting that it corresponds to the site of cleavage. We extrapolated this result for every peak to estimate the cleavage positioning of Topo IV (~150bp downstream of the center of the forward peak, S6D Fig) We manually validated 172 sites that were common to ParC-1 and ParE-1 experiments (S1 Table) for further analysis.
The Topo IV cleavage at the dif site was the most enriched of the chromosome (~ 30 fold), fourteen sites were enriched from 5 to 10 fold and other positions were enriched from 2 to 5 fold (Fig 2E). Most NorflIP sites did not correspond to significant peaks in the ChIP-seq experiment (Fig 2E). We also did not observe any cleavage for the majority of the strong binding sites observed by ChIP-seq. This is illustrated for the binding site at 1.85 Mb (Fig 2E). We verified several Topo IV cleavage sites by Southern blot, a significant cleaved DNA fragment was observed at the expected size for each of them (Fig 2F). Southern blotting experiments following DNA cleavage in the presence of norfloxacin on synchronized cultures revealed that, like its binding, Topo IV cleavage is coordinated with DNA replication. In good agreement with ChIP-seq experiments, increased cleavage was observed as soon as 20 minutes after initiation of replication for the dif and 2.56 Mb sites (Fig 2G).
The general genomic distribution of Topo IV cleavage sites was not homogeneous; a few regions had a large number of sites clustered together, while the 1.2Mb– 2.5 Mb region contained a low density of sites (Fig 2H). We further analyzed the distribution of cleavage sites in the terminus and the oriC regions. In the terminus region, the average distance of consecutive cleavage sites was long (around 30 kb in the 1.5–2.5 Mb region) compared to 8 kb in the 0.8–1.5 Mb or the 2.5–3.1 Mb regions (S7A Fig). The oriC region displays a mixed distribution (S7B Fig), a high density of sites near oriC flanked by two depleted regions, including the SNAP2 region [16]. At the gene scale, the mid-point of Topo IV cleavage signal can be localized inside genes (82%) or intergenic regions (16%) but it presents a bias toward the 5’ or 3’ gene extremities (S7C Fig). Since the cleavage signal spans approximately 200bp, nearly 50% of the sites overlapped, at least partly, with intergenic regions that account for only 11% of the genome. Finally, we did not identify any robust consensus between sets of Topo IV cleavage sites. The only sequence traits that we identified are a bias for GC dinucleotides near the center of the sites (S7D Fig) and an increased spacing of GATC motifs around cleavage sites (S7E Fig).
The bias in the distribution of cleavage sites (Fig 2H) was very similar to the Topo IV binding bias revealed by ChIP-seq (Fig 1C). NorflIP and ChIP-seq data were compared on Fig 3A. Despite the lack of corresponding ChIP-seq enrichment at the position of most highly enriched NorflIP sites, a number of consistencies were observed between these two data sets. Overall the NorflIP and ChIP-seq datasets had a Pearson correlation of 0.3 and the averaged data (1 kb bin) revealed a Pearson correlation of 0.5. First a small amount of local enrichment in the ChIP-seq experiments was frequently observed in the regions containing many cleavage sites (Fig 3A and 3C). This led us to consider that trapped Topo IV engaged in the cleavage reaction could contribute to a small amount of local enrichment in the ChIP-seq experiments. Second, both Topo IV cleavages and binding sites were rare in highly expressed regions (Fig 3A), only one of the 172 manually validated Topo IV cleavage site overlapped a highly expressed region. However cleavages sites were more frequently, than expected for a random distribution, observed in their vicinity (Fig 3C and S8 Fig). Thirty percent (50/172) of the Topo IV sites are less than 2 kb away from the next highly expressed transcription unit (Fig 3).
We explored correlations between the localization of Topo IV cleavages and binding sites of various NAPs thanks to the Nust database and tools [32]. A significant correlation was only observed for Fis binding sites (Fig 3B). Sixty eight genes present both Fis binding [33] and Topo IV cleavage (P value 2x10-03). Thirty-three of the 172 manually validated cleavage sites overlapped at least partially with a Fis binding site, 80 of them are located less than 400 bp away from a Fis binding site. At the genome scale this correlation is difficult to observe (Fig 3A), but close examination clearly revealed overlapping Topo IV cleavages and Fis binding sites (Fig 3C). Fis binding sites are more numerous than Topo IV cleavage sites, therefore a large number of them do not present enrichment for Topo IV (Fig 3C). By contrast, Topo IV peaks are excluded from H-NS rich regions (Fig 3A, 3B and 3C). Only one of the 172 manually validated Topo IV cleavage site overlapped with an H-NS binding site. As observed for highly expressed regions TopoIV cleavage sites were frequently observed at the border of H-NS rich regions (Fig 3C). Moreover H-NS rich regions contain less Topo IV than the rest of the chromosome (Fig 3A–3D and S9A Fig). H-NS rich regions correspond to an AT rich segment of the chromosome (Fig 3C and 3D). Indeed background level of Topo IV binding and cleavage were significantly reduced in AT rich regions (S9B Fig). In rare occasions binding of H-NS has been observed in regions with a regular AT content (Fig 3C), notably Topo IV binding and cleavage were also reduced in these regions. This observation suggested that H-NS itself rather than AT content limits the accessibility of Topo IV to DNA. This observation was confirmed by the identification of Topo IV cleavage in regions with an AT content ranging from 20 to 80% (S9C and S9D Fig).
We performed Southern blot analysis of Topo IV cleavage on representative sites to test whether gene expression and chromatin factors influenced Topo IV site selection. First, we observed that the exact deletion of cleavage sites at position 1.92 Mb and 2.56 Mb did not abolish Topo IV cleavage activity (Fig 3D and 3E). Second, since these loci also contain a Fis binding site overlapping Topo IV cleavage signal, we deleted the fis gene. However, deletion of the fis gene did not modify Topo IV cleavage (Fig 3D and 3E). Finally we performed cleavage assays in the presence of rifampicin to inhibit transcription. To limit the pleiotropic effects of rifampicin addition we performed the experiment with a 20 min pulse of rifampicin. Rifampicin treatment abolished Topo IV cleavage (Fig 3E). These results suggest that gene expression rather than chromatin factors influences Topo IV targeting.
Our analysis confirms that the dif region is a hot spot for Topo IV activity [29]. Indeed, ChIPseq and NorflIP show that Topo IV binds to and cleaves frequently in the immediate proximity of dif. We measured DNA cleavage by Topo IV in the presence of norfloxacin in various mutants affecting the structure of dif or genes implicated in chromosome dimer resolution. Southern blot was used to measure Topo IV cleavage (Fig 4A). We observed that exact deletion of dif totally abolished Topo IV cleavage. Interestingly, the deletion of the XerC-binding sequence (XerC box) of dif was also sufficient to abolish cleavage, while the deletion of the XerD box only had a weak effect. Deletion of the xerC and xerD genes abolished Topo IV cleavage at dif. However, cleavage was restored when the catalytically inactive mutants XerC K172A or XerC K172Q were substituted for XerC (Fig 4B). This suggests that the role of XerCD/dif in the control of Topo IV activity is structural and independent of XerCD catalysis. Deletion of dif or xerC did not significantly alter cleavage at any of the other tested Topo IV cleavage sites (Fig 4C). This suggests that influence of XerC on Topo IV is specific to dif.
To evaluate the role of XerCD-mediated Topo IV cleavage at dif, we attempted to construct parEts xerC, parEts xerD and parCts xerC double mutants. We could not obtain parCts xerC mutants by P1 transduction at any tested temperature. We obtained parEts xerC and parEts xerD mutants at 30°C. The parEts xerC double mutant presented a growth defect phenotype at 30°C and did not grow at temperature above 35°C (Fig 4D). The parEts xerD mutant presented a slight growth defect at 37°C compared to parEts or xerD mutants. None of the parEts mutant grew above 42°C. Next, we used quinolone sensitivity as a reporter of Topo IV activity. To this aim, we introduced mutants of the FtsK/Xer system into a gyrAnalR (nalR) strain; Topo IV is the primary target of quinolones in such strains. The absence of XerC, XerD, the C-terminal activating domain of FtsK or dif exacerbated the sensitivity of the nalR strain to ciprofloxacin (Fig 4D). We therefore concluded that the impairment of Topo IV was more detrimental to the cell when the FtsK/Xer system was inactivated. Among partners of the FtsK/Xer system the absence of XerC was significantly the most detrimental, suggesting a specific role for XerC in this process.
The above results suggest an interaction between Topo IV and the XerCD/dif complex. We therefore attempted to detect this interaction directly in vitro (Fig 4E and 4F). We performed EMSA with two fluorescently labeled linear probes, one containing dif and the other containing a control DNA not targeted by Topo IV in our genomic assays. Topo IV alone bound poorly to both probes (Kd > 100nM). Binding was strongly enhanced when XerC or both XerC and XerD were added to the reaction mix. In contrast, Topo IV binding to dif was slightly inhibited in the presence of XerD alone. These results were consistent with the observation that deletion of the XerC box but not of the XerD box inhibited Topo IV cleavage at dif and pointed to a specific role for XerC in Topo IV targeting. The control fragment showed that these effects are specific to dif. Topo IV-XerC/dif complexes were stable and resisted a challenge by increasing amount of XerD (S10A Fig). The positive influence of XerCD on TopoIV binding was also observed on a negatively supercoiled plasmid containing dif. In the presence of XerCD (50nM), a delay in the plasmid migration was observed with 40nM of TopoIV. By contrast, 200 nM was required in the absence of XerCD (S10B Fig). The Southern blot cleavage assay showed that overexpression of the ParC C-terminal domain (pET28parC-CTD) strongly reduced cleavage at dif but enhanced cleavage at the Topo IV site located at 2.56Mb. This suggested that, as observed for MukB [17], Topo IV might interact with XerC through its C-terminal domain (Fig 4G).
We assayed the effects of the reported Topo IV modulators and proteins involved in chromosome segregation the activity of Topo IV at dif. MukB has previously been shown to influence the activity of Topo IV [17, 18]. We measured Topo IV cleavage in a mukB mutant at dif and at position 2.56 Mb, cleavage was reduced at dif but no significant effect was observed at position 2.56Mb (Fig 5A). We did not detect any effect of a seqA deletion on Topo IV cleavage at either position (Fig 5B). We next assayed the effect of MatP, which is required for compaction and intracellular positioning of the ter region as well as for the its progressive segregation pattern ending at dif [25, 26]. The Topo IV cleavage at dif was significantly impaired in the matP mutant (Fig 5C). The Topo IV cleavage site at position 1.9Mb is included in the Ter macrodomain, but cleavage at this site was almost unchanged in the absence of MatP (Fig 5C). Introduction of a matP deletion into the nalR strain yielded an increase in ciprofloxacin sensitivity (Fig 5D). We also constructed a parEts matP double mutant. Growth of this strain was significantly altered compared to the parEts parental strain at an intermediate temperature (Fig 5E). Such a synergistic effect was not found when combining the matP deletion with a gyrBts mutation. Taken together, these results led us to consider that MatP itself or the folding of the Ter macrodomain might be important for Topo IV targeting at dif.
Since the FtsK/Xer/dif system is dedicated to post-replicative events that are specific to a circular chromosome, it was tempting to postulate that the activity of Topo IV at dif is also dedicated to post-replicative decatenation events and is strictly required for circular chromosomes. To address this question, we used E. coli strains harboring linear chromosomes [34]. In this strain, expression of TelN from the N15 phage promotes linearization of the chromosome at the tos site inserted a 6kb away from dif. Indeed, chromosome linearization suppresses the phenotypes associated with dif deletion [34]. We analyzed cleavage at the dif site by Topo IV in the context of a linearized chromosome. Cleavage was completely abolished; showing that Topo IV activity at dif is not required on linear chromosomes. This effect was specific to the dif site, since cleavage at the 1.9Mb site remained unchanged after chromosome linearization (Fig 5F). We next assayed if the phenotypes associated with matP deletion, i.e., formation of elongated cells with non-partitioned nucleoids [26], depend on chromosome circularity. Strikingly, most of the phenotypes observed in the matP mutant were suppressed by linearization of the chromosome (Fig 5G). Interestingly, the frequency of cleavage at dif sites inserted far (300 kb) from the normal position of dif or in a plasmid were significantly reduced compared to the WT situation (S11 Fig) confirming that Topo IV cleavage at dif is specific to circular chromosomes.
Whole genome analysis of Topo IV binding by ChIP-seq revealed approximately 10 Topo IV binding sites across the E. coli genome. Among them, only 5 sites were strongly enriched in every experiment and these were mapped to positions 1.25, 1.58 (dif), 1.85, 2.56 and 3.24 Mb. We did not identify any consensus sequence that could explain specific binding to these sites. Band shift experiments at the dif site and the 1.25 Mb site revealed that Topo IV binding is not sequence-dependent.
This led us to favor models involving exogenous local determinants for Topo IV binding as it is the case for the dif site in the presence of XerC. Because XerC is only known to bind to dif, we could speculate that other chromatin factors might be involved in Topo IV targeting. Topo IV and Fis binding sites [33] overlap more frequently than expected (Nust P value 10e-03 [32]. Topo IV and Fis binding sites overlap at the positions 1.25 and 2.56 Mb; it is therefore possible that Fis plays a role in defining some Topo IV binding sites. However our EMSA, cleavage and ChIP experiments did not show any cooperative binding of Topo IV with Fis. In spite of its co-localization with Topo IV, Fis does not contribute in defining Topo IV binding or cleavage sites. Nevertheless, the role of the chromatin in Topo IV localization was also illustrated by the strong negative correlation observed for the Topo IV and H-NS bound regions. H-NS rich regions were significantly less enriched for nonspecific Topo IV binding than the rest of the chromosome.
We postulated that loci where Topo IV is catalytically-active could be identified by DNA cleavage mediated by the quinolone drug norfloxacin. We designed a new ChIP-seq strategy that consisted of capturing DNA-norfloxacin-Topo IV complexes. We called it NorflIP. Three independent experiments show that Topo IV was trapped to a large number of loci (300 to 600) with most of these loci observed in two out of three experiments. A hundred of these loci were identified in all three experiments. Dif presented a strong signal in the NorflIP as in the ChIP-seq but this is not the case for most of the other ChIP-seq peaks. NorflIP peaks presented a characteristic pattern suggesting that they are genuine DNA-norfloxacin-Topo IV complexes. Considering that norfloxacin does not alter Topo IV specificity, our results suggest that for Topo IV the genome is divided into five categories: i) Loci where Topo IV binds strongly but remains inactive for most of the cell cycle; ii) Loci where Topo IV is highly active but does not reside for very long time; iii) Loci where we observed both binding and activity (dif and 2.56 Mb); iv) regions where Topo IV interacts non-specifically with the DNA and where topological activity is not stimulated; v) regions where non-specific interactions are restricted (the Ter domain, chromatin rich regions (tsEPODs [35], H-NS rich regions). Detection of norfloxacin-mediated genomic cleavage by pulse field electrophoresis has previously revealed that when Topo IV is the only target of norfloxacin the average fragment size is 300–400 kb while it drops to 20 kb when Gyrase is the target [11]. This suggests that, for each cell, no more than 10 to 20 Topo IV cleavages are formed in 10 min of norfloxacin treatment. To fit this observation with our data, only a small fraction (10–20 out of 600) of the detected Topo IV cleavage sites would actually be used in each cell. This might explain why Topo IV cleavage sites were hardly distinguishable from background in the ChIP-seq assay (Fig 3). This is in good agreement with the estimation that the catalytic cycle only provokes a short pause (1.8 sec) in Topo IV dynamics [36]. The mechanism responsible for the choice of specific Topo IV cleavage sites is yet to be determined. As indicated by our findings that deletion of the cleavage site resulted in the formation of a new site or sites in the vicinity, cleavage is not directly sequence-related. We observed several biases that might be involved in determination of cleavage sites (GC di-nucleotide skew, GATC spacing, positioning near gene ends or intergenic regions, proximity with highly expressed genes and Fis binding regions). Interestingly inhibition of transcription with rifampicin inhibits Topo IV cleavage (Fig 3). This raises the possibility that transcription, that can be stochastic, may influence stochastic determination of Topo IV activity sites. The influence of transcription could be direct, if RNA polymerase pushes Topo IV to a suitable place, or indirect if the diffusion of topological constraints results in their accumulation near barriers imposed by gene expression [37, 38]. This accumulation could then, in turn, signal for the recruitment of Topo IV.
Synchronization experiments revealed that, like Topo IV binding at specific sites, Topo IV cleavage activity is enhanced by chromosome replication. Enrichment was the highest in late S phase or G2 phase; it seems to persist after the passage of the replication fork at a defined locus. Enrichment in asynchronous cultures was significantly reduced compared to S40 or G2 synchronized cultures suggesting that Topo IV is not bound to the chromosome for the entire cell cycle. Unfortunately our experiments did not have the time resolution to determine at what point of the cell cycle Topo IV leaves the chromosome and if it would leave the chromosome during a regular cell cycle. The role of DNA replication of Topo IV dynamics has recently been observed by a very different approach [36]. The authors propose that Topo IV accumulates in the oriC proximal part of the chromosome in a MukB and DNA replication dependent process. These observations are in good agreement with our data and suggest that Topo IV is loaded on DNA at the time of replication, accumulate towards the origin of replication and remains bound to the DNA until a yet unidentified event triggers its release. Formation of positive supercoils and precatenanes ahead and behind of the replication forks respectively, could be the reason for Topo IV recruitment. One could hypothesize that MukB is used as a DNA topology sensor that is responsible for redistribution of Topo IV. However we only detected a modest effect of mukB deletion on Topo IV cleavage at dif (Fig 5). Putative events responsible for Topo IV release could be, among others, complete decatenation of the chromosome, SNAPs release, or stripping by other proteins such as FtsK.
Non-specific Topo IV binding presents a very peculiar pattern; it is significantly higher in the oriC proximal 3Mb than in the 1.6Mb surrounding dif. This pattern is not simply explained by the influence of replication (S3 Fig). Interestingly, ChIP-seq and ChIP-on-Chip experiments have already revealed a similar bias for DNA gyrase [12] and SeqA [39]. The CbpA protein has been shown to present an inverse binding bias [40], with enrichment in the terminal region and a reduction in the oriC proximal domain. The HU regulon has also presented a similar bias [41]. The terminus domain defined by these biases always comprises the Ter macrodomain but it extends frequently beyond the extreme matS sites. The role of MatP in the definition of these biases has not yet been tested. The group of G. Mushelishvili proposed a topological model to interpret the DNA gyrase and HU regulon biases, suggesting that HU coordinates the global genomic supercoiling by regulating the spatial distribution of RNA polymerase in the nucleoid [41]. Topo IV could benefit from such a supercoiling gradient to load on the chromosome. Interestingly, the strongest Topo IV binding and cleavage sites are localized inside the Terminus depleted domain. One possibility could be that these sites minimize Topo IV binding to adjacent nonspecific sequences. Alternatively one can propose that a regional reduction of non-specific binding creates a selective advantage for optimal loading on to specific sites.
Dif was the strongest Topo IV cleavage site detected by NorflIP, it was also detected in the ChIP-seq assays. We have used Southern blot to analyze the determinants involved in this activity. The binding of XerC on the xerC box of dif and the region downstream of the xerC box are essential. In vitro, XerC also strongly favors binding of Topo IV at dif. Interestingly XerD and the xerD box did not improve Topo IV binding or cleavage. We propose that XerC works as a scaffold for Topo IV, simultaneously stimulating its binding and its activity. Topo IV activity at dif is also dependent on the circularity of the chromosome, suggesting that when topological constraints can be evacuated through chromosome ends, Topoisomerase IV does not catalyze strand passage at dif. This suggests that topological complexity is directly responsible for Topo IV activity. Topo IV cleavage activity at dif is not influenced by SeqA or FtsK, which are two known Topo IV partners. Interestingly, mukB and matP deletion mutants slightly reduced this activity. The synergistic effect observed when a matP deletion is combined with a parEts mutation suggests that MatP indeed influences Topo IV activity. The phenotypes of the matP mutant are rescued by the linearization of the chromosome. A similar rescue has been observed for the dif mutant [34]. Therefore it is likely that a significant part of the problems that cells encounter in the absence of matP corresponds to failure in chromosome topology management, either decatenation or chromosome dimer resolution [25]. In conclusion, we propose that genomic regulation of Topo IV consists of: (1) Topo IV loading during replication, (2) Topo IV binding to specific sites that may serve as reservoirs, (3) Topo IV activation to remove precatenanes or positive supercoils in a dozen of stochastically chosen loci (4) XerC and MatP ensuring the loading of Topo IV at the dif site for faithful decatenation of fully replicated chromosomes.
ParE-flag and ParC-flag C-terminus fusions were constructed by lambda red recombination [42]. Cultures were grown in LB or Minimal medium A supplemented with succinate (0.2%) and casamino acids (0.2%). Cells were fixed with fresh Formaldehyde (final concentration 1%) at an OD600nm 0.2–0.4. Sonication was performed with a Bioruptor Pro (Diagenode). Immunoprecipitations were performed as previously described 26. Libraries were prepared according to Illumina's instructions accompanying the DNA Sample Kit (FC-104-5001). Briefly, DNA was end-repaired using a combination of T4 DNA polymerase, E. coli DNA Pol I large fragment (Klenow polymerase) and T4 polynucleotide kinase. The blunt, phosphorylated ends were treated with Klenow fragment (3’ to 5’ exo minus) and dATP to yield a protruding 3- 'A' base for ligation of Illumina's adapters which have a single 'T' base overhang at the 3’ end. After adapter ligation DNA was PCR amplified with Illumina primers for 15 cycles and library fragments of ~250 bp (insert plus adaptor and PCR primer sequences) were band isolated from an agarose gel. The purified DNA was captured on an Illumina flow cell for cluster generation. Libraries were sequenced on the Genome Analyzer following the manufacturer's protocols.
Norfloxacin (final concentration 2μM) was added to the cultures at OD600nm 0.2 LB for 10 min before harvesting. Sonication and immunoprecipitation were performed as described for the ChIP-seq assay.
Sequencing results were processed by the IMAGIF facility. Base calls were performed using CASAVA version 1.8.2. ChIP-seq and NorflIP reads were aligned to the E. coli NC_000913 genome using BWA 0.6.2. A custom made pipeline for the analysis of sequencing data was developed with Matlab (available on request). Briefly, the number of reads for the input and IP data was smoothed over a 200bp window. Forward and reverse signals were added, reads were normalized to the total number of reads in each experiment, strong non-specific signals observed in unrelated experiments were removed, data were exported to the UCSC genome browser for visualization and comparisons. The strongest peaks observed with NorflIP experiments (dif and 1.9 Mb) present a characteristic shape (S6 Fig) that allows the automatic detection of lower amplitude peaks but preserves the characteristic shape. We measured Pearson correlation coefficient with the dif and the 1.9 Mb site for 600bp sliding windows over the entire genome. Peaks with a Pearson correlation above 0.72 were considered as putative Topo IV cleavage sites. Sequencing data are available on the GEO Repository (http://www.ncbi.nlm.nih.gov/geo/)with the accession number GSE75641. Data were plotted with the Circos tool [43] and UCSC Archaeal Genome Browser [44].
Cleavage of DNA by Topo IV in the presence of Norfloxacin was monitored by Southern blot as previously described [10]. DNA was extracted from E. coli culture grown in minimal medium supplemented with glucose 0.2% and casaminoacids 0.2%. Norfloxacin (final concentration 10μM) was added to the cultures at OD 0.2 for 10 min before harvesting. DNA was transferred by neutral blotting on nitrocellulose membranes. For synchronization experiments a flash freeze step in liquid nitrogen is included before harvesting. Quantification was performed with Image J software.
Experiments were conducted using Cy3-coupled probes harboring the dif site and a Cy5-coupled dye as control. Reactions were carried out in EMSA reaction buffer (1mM spermidine, 30mM potassium glutamate, 10mM DTT, 6mM magnesium chloride, 10% glycerol, pH 7.4). Reactions were incubated for 15 min at RT, loaded on 4% native PAGE gel at 25 volts and then run at 125 volts for 2 hours. Gels were then visualized using a Typhoon FLA 5000 scanner (GE healthcare Life Science). EMSA of plasmids were performed with unlabeled supercoiled plasmid in the same reaction buffer. Electrophoresis was performed in a 0.8% agarose gel in 0.5x TAE buffer at 4°C for 80 min at 150V. DNA labeling was performed with SYBR green.
|
10.1371/journal.ppat.1001211 | Glycosylation Focuses Sequence Variation in the Influenza A Virus H1 Hemagglutinin Globular Domain | Antigenic drift in the influenza A virus hemagglutinin (HA) is responsible for seasonal reformulation of influenza vaccines. Here, we address an important and largely overlooked issue in antigenic drift: how does the number and location of glycosylation sites affect HA evolution in man? We analyzed the glycosylation status of all full-length H1 subtype HA sequences available in the NCBI influenza database. We devised the “flow index” (FI), a simple algorithm that calculates the tendency for viruses to gain or lose consensus glycosylation sites. The FI predicts the predominance of glycosylation states among existing strains. Our analyses show that while the number of glycosylation sites in the HA globular domain does not influence the overall magnitude of variation in defined antigenic regions, variation focuses on those regions unshielded by glycosylation. This supports the conclusion that glycosylation generally shields HA from antibody-mediated neutralization, and implies that fitness costs in accommodating oligosaccharides limit virus escape via HA hyperglycosylation.
| Influenza A virus is highly susceptible to neutralizing antibodies specific for the viral hemagglutinin glycoprotein (HA), and is easily controlled by standard vaccines. Influenza A virus remains an important human pathogen, however, due to its ability to rapidly evade antibody responses. This process, termed antigenic drift, is due to the accumulation of amino acid substitutions that modify HA antigenic sites recognized by neutralizing antibodies. In this study, we perform bioinformatic analysis on thousands of influenza A virus isolates to better understand the influence of N-linked glycosylation on antigenic drift. HA from human IAV isolates can accommodate up to 6 oligosaccharides in its globular domain. We show that for H1, H2, and to a somewhat less extent H3, HAs, the number of glycosylation sites in the globular domain does not greatly modify the total degree of variation in antigenic sites, but rather focuses variation on sites whose access to antibodies is unaffected by glycosylation. Our findings imply that glycosylation protects HA from antibody neutralization, but functional impairment limits the number of oligosaccharides that HA can accommodate.
| The influenza A virus (IAV) hemagglutinin (HA) is a homotrimeric glycoprotein that initiates infection by attaching virus to host cell sialic acids and mediating fusion of viral and endosomal membranes [1]. HA consists of a fibrous stem inserted into the viral membrane supporting a globular domain containing three sialic acid binding sites (one per monomer). Trimerization of nascent HA is necessary for HA folding and export from the early secretory pathway [2], [3], [4]. Nearly all antibodies (Abs) that neutralize viral infectivity (“neutralizing antibodies”) recognize epitopes in the globular domain. Most Abs neutralize infection by sterically blocking access of sialic acid receptors to the HA [5], [6].
Neutralizing Abs are the principal selective force driving HA evolution in man. The rapid emergence of mutants that escape Ab neutralization is termed “antigenic drift”, and has prevented effective long-term vaccination against IAV. Based on locating single amino acid substitutions that enable escape from neutralization with monoclonal Abs (mAbs), physically distinct regions have been defined on the globular domains of H1 (Sa, Sb, Ca, Cb) and H3 (A, B, C, D, E) subtype HAs [7], [8], [9]. We term the region of HA containing these sites, consisting of residues 58–272, the globular domain. Differences in the location of the antigenic sites in the globular domain correlate with the differential location of consensus N-linked oligosaccharide attachment sites in the H1 (PR8) vs. H3 (HK) HAs used for antigenic analysis [9], [10].
This raises the important question of the influence of HA glycosylation on antigenic drift. Other viral glycoproteins (e.g. HIV gp160) mask potential antigenic sites by hyperglycosylation [11], [12]. Addition of glycans to the globular domain has been directly shown to block neutralization of HA by monoclonal and polyclonal Abs [13]. Why doesn't IAV employ this strategy to a greater extent? A potential clue comes from the distinct evolution of H3 vs. H1 HAs in humans. Despite circulating for far less time in humans (41 years), H3 viruses have accumulated approximately twice as many glycosylation sites in the globular domain than H1 subtype viruses (circulating for ∼70 years- 1918–1957, 1977-present) [14], [15], [16], [17]. This is consistent with the idea that there are distinct fitness costs to glycosylation that vary among HA subtypes [13], [18], [19], [20].
Despite the potential importance of HA glycosylation in IAV evolution, there is a paucity of bioinformatics analysis of the large number of sequences accumulating in data banks. Here, we provide bioinformatics evidence that supports a critical role for glycosylation in focusing antigenic variation on non-glycosylated regions of the HA globular domain.
We analyzed 1907 full-length H1 HA sequences from human, swine or avian viruses downloaded from the NCBI influenza virus resource. NetNGlyc prediction of glycosylation sites (Asn-Xaa-Ser/Thr, where Xaa is any amino acid except Pro) in the globular domain reveals the non-random distribution of probable glycosylation sites at nine locations (Figure 1a). With few exceptions, glycosylation sites are located within 5 residues on either side of a consensus site. Consequently, for further analysis we defined conserved glycosylation sites within an 11-residue sequence centered on the consensus site.
Consistent with previous findings that efficient HA folding and assembly requires glycosylation at conserved sites, glycosylation sites at or near residues 15, 26, 289, 483, and 542 occur in virtually all HAs [4], [20], [26], [27], [28], [29], [30] (note that throughout the manuscript we use the H3 HA numbering system). These sites are located in the stem region of the HA (rendered in green in Figure 2) and may be conserved due to proper association with glycan binding-ER chaperones that facilitate HA folding and assembly [30]. The distribution of glycosylation sites in the H1 globular domain is variable, and is distributed among three regions centered on residues 91, 129 and 162 are rendered in red (Figure 2). For further analysis, we chose ±5 amino acids on each side of the conserved glycosylation sites to define glycosylation regions. It is well documented for H2 HA that addition of consensus glycosylation sites at these regions results in the predicted glycosylation, as determined by mobility shifts in SDS-PAGE [13].
Due to their potential influence on antigenic drift, we focused our attention on the glycosylation sites in the H1 HA globular domain, which center on residues 91, 129, and 162. We temporally analyzed the presence of glycosylation sites in viruses isolated from 1918 to present. Though this analysis is hindered by the limited number of sequences available until 1995, two trends are apparent: an increase in glycosylation sites from zero/one as HA evolved from the 1918 strain to three sites typical for contemporary H1N1 viruses, and an abrupt reintroduction of a single glycosylation site with the appearance of SOIV in 2009 (Figure 1b, c). With three glycosylation sites, there are eight permutations of glycosylation status, all three (1), 2 of 3 (3), one of three (3), and none (1) (Figure 1d). Our analysis revealed the complete absence of HAs with glycosylation sites at positions 129 and 162. Interestingly, these sites are essentially adjacent in the 3-dimensional structure. Thus, it is not surprising that simultaneous glycosylation would have deleterious effects on HA folding, providing strong negative selection; what is more surprising that selection against the two sites is alleviated by a glycosylation site at residue 91, which is located further down the HA trimer (Figure 2). Since glycosylation occurs co-translationally, glycosylation at 91 would precede glycosylation at 129/162, and could limit the extent to which 129 and 162 are simultaneously glycosylated, accounting for its ability to modulate negative selection against adjacent sites. Alternatively, the absence of 129 162 dual glycosylation isolates may relate to historical evolution factors.
Does glycosylation focus drift on selected antigenic sites? We correlated the location of glycosylation sites with the variability at individual residues in the globular domain (Figure 3). This revealed that glycosylation alters the focus of sequence variation. In HAs lacking glycosylation sites in the antigenic domain, variability peaks near residue 135. Acquisition of a glycosylation site in the same region (129) results in reduction of variability in that region, and increase in variability at residues 78, 159, and 228, which represent the Cb, Sb, and Ca antigenic sites. Acquisition of two glycosylation sites at 91 and 129 now focuses variation at residues 165 and 190. With all three glycosylation sites utilized, variation is now focused around residues 190 and 191. Interestingly, positions 190 and 228 greatly influence HA receptor specificity for α-2,3 vs. α-2,6 of the sialic residue [31], [32].
Each combination of glycosylation sites generates a similar pattern: glycosylation minimizes variation around its own site while focusing variation onto non-glycosylated sites. The statistical significance of this conclusion is shown in Figure 4. A simple interpretation for this finding is that oligosaccharides shield antigenic regions from Ab neutralization, shifting variation to unshielded sites. This is consistent with the observation that viruses cluster in the PCA plot based on a common number of glycosylation sites in the globular domain and not year of collection (Figure S2), which demonstrates that they are highly homologous in the hypervariable regions of the globular domain.
If glycosylation can affect the pattern of antigenic variation, is there a correlation between number of antigenic domain glycosylation sites and antigenic variation? There is no clear relationship between the number of sites and the overall variability of amino acids between the residues that comprise the globular domain (58–272) (Figure 1d), indicating that the overall extent of glycosylation does not globally limit variation in antigenic regions.
We extended this approach to H2N2 and H3N2 HA sequences. H2N2 viruses possess a single glycosylation site in the globular domain, located at position 166 (Figure 5a). Consistent with the H1 data, analysis of H2N2 viruses showed limited variation near the sole glycosylation site at position 166 (Figure 5b).
H3N2 viruses have up to six glycosylation sites on the globular domain of HA (Figure 6). When we sorted H3 sequences based on number of glycosylation sites, we found a distinct trend compared to H1 HA in acquiring glycosylation sites. Position 168 is the most conserved position glycosylation in H3 sequences. When there is one glycosylation, it's nearly invariably at position 168 (with a few isolates with lone glycosylation at position 84). Double glycosylation is dominated by the 84, 168 pairing. Remarkably, triple gycosylation is dominated by 168 with two novel sites: 66 and 129. Adding glycosylation at residue 259 uniformly attains four-site glycosylation. Adding sites between residues 129 and 168 achieve higher order glycosylation.
We next examined the correlation between the location glycosylation site and regions of variability for H3 viruses with 2 to 4 sites in the globular domain (Figure 7). Analysis of other glycoforms was compromised by either paucity of isolates in a group or by the complexity of glycosylation pattern. Although there was a reasonable correlation between the presence of a glycosylation site and absence of variation in the residues surrounding the site, this relationship was less robust than for the H1 and H2 HAs (as indicated by the arrows pointing to exceptions).
Is it possible to predict the tendency of acquisition of glycosylation sites (losses as well as gains) as a function of likelihood of mutation of codons present in the glycosylation regions? We devised a simple algorithm, termed the flow index (FI) to model glycan site evolution based on the sequences present in the glycosylation regions of H1N1 viruses with a given oligosaccharide status. The tendency of mutating to (green arrow) or from (red arrow) a given glycosylation state is assigned is shown in Figure 8. Summing the probabilities to and from a given state provides a measure of the probability of remaining in that state (Table S1).
Despite its simplicity, this algorithm reasonably accurately reflects the prevalence of glycoforms among the H1 isolates in the database. The most notable exception is the 129+162 glycoform, which is not represented by any isolate despite having a facile mutational path. As noted above, this may be due to the proximity of these residues in the folded structure, which may interfere with folding of the globular domain (Figure 2). This exception points to the contribution of functional selection in the prevalence of glycoforms.
Unfortunately, we could not calculate a FI for H2 or H3 HA evolution due to either low number of sequences of viruses in a given group or the complexity of the glycosylation patterns.
The recent introduction of SOIV into the human population offers a unique opportunity to study IAV evolution in humans at high resolution in real time [33], [34], [35]. Nearly all of the 212 unique SOIV isolates downloaded on October 12th 2009 possess oligosaccharide site at position 91. How does the pattern of variation of SOIVs compare to human H1 isolates or classic Swine isolates that also possess a single oligosaccharide site at position 91?
As seen in Figure 9, despite their limited time in humans, SOIVs demonstrate a remarkable amount of variation, peaking around positions 225 and 264, with other hot spots at residues 77 and 135 (Figure 9b). This pattern differs from human H1N1 isolated from 1918 to present (Figure 9c), which show far less variation at residues 225 and 264 regions while focusing variation near 77, 135 and 190 regions. Classic swine viruses (Figure 9d) show a different pattern of variation, focused at residues 147 and 200 (note that the data shown in Figure 3b include all isolates with a single glycosylation site at position 91).
We again downloaded available unique full-length SOIV sequences on March 31st to examine variability over the course of six months (Figure 9a). Apart from variability at position 225 and 264, there is a higher variability near position 142 (Sb-site) that was absent in October.
Can we predict the evolution of SOIV glycoforms using the FI? FI of the October sequences for loss of oligosaccharide at 91 or gain at 129 or 162 is each calculated to be zero, i.e. two amino substitutions are required to insert a glycan at these regions. Indeed, the sequences downloaded on March 2010 showed a strong tendency to maintain glycosylation at position 91, though a few isolates lose glycosylation at 91 (10 isolates) or gain glycosylation at 162 (5 isolates) [36]. We therefore predict that a single glycosylation site at position 91 at the antigenic domain will predominate in SOIV isolates for a prolonged period.
It has been known for more than 70 years that IAV, unlike many viruses, demonstrates significant antigenic variation. The epidemiological significance of antigenic variation in IAV was unmistakable from the failure of the 1947 vaccination campaign [37], [38]. Despite considerable effort and significant gains in understanding HA antigenic structure, much remains to be learned about how drift occurs in humans. The revolution in nucleic acid sequencing technology provides enormous opportunities to better understand drift. Here, we utilize the NCBI influenza resource to examine the relationship between glycosylation in the HA globular domain and antigenic variation.
The ability of oligosaccharides to sterically block antibody binding to HA antigenic sites was clearly established with the original definition of antigenic sites on the HA structure using mAbs [39], [40]. Surprisingly, however, the more global effects of oligosaccharides on HA evolution have been examined in only a few publications [17], [41], [42]. We detect a clear influence of oligosaccharides in directing the focus of variation to the established neutralizing antibody binding sites on the H1 and H2 HAs. We also find a similar pattern among H3 HAs with 2–4 globular domain glycosylation sites, but note exceptions to the relationship (arrows in Figure 6), that might contribute to the finding of a prior bioinformatics analysis of H3 isolates that failed to detect a relationship between glycosylation and the locus of variability [41]. This potential difference in glycosylation in shaping HA evolution might be related to a major difference in H1 vs. H3 HA evolution: while H3 remained in human populations constantly from its introduction in 1968 until the present time, after the complete replacement of H1 in 1957 by H2 viruses, it re-appeared in 1977 in the form of the 1950 virus, almost certainly as a result of a re-introduction from a laboratory sample.
While there is a tendency towards adding oligosaccharides to the H1 HA with time, the process is slower than might be expected. H1 viruses have circulated in humans for at least 80 years in the period between 1918 and present time, yet only possess 3 globular domain glycosylation sites while H3 HAs have up to six glycosylation sites in the globular domain.
It is important to note that we have not experimentally established that antibody pressure is responsible for the influence of oligosaccharides on variation in the globular domain. Although it seems less likely, it is possible that oligosaccharides influence HA evolution by modulating the mutation space of globular domain residues.
We show that the sequence space in the regions of the globular domain where oligosaccharides can be accommodated appears to play a surprisingly robust role (since at most, only two amino acid changes are needed to create a glycosylation site) in influencing the evolutionary acquisition of additional glycosylation sites. Thus, although the FI is hampered by historical biases in the number of isolates collected during the course of IAV evolution in man (and by alterations in glycosylation that accompany adaptation to growth in eggs or cultured cells [43]), it nonetheless is able to predict the prevalence of HA glycoforms in H1N1 isolates. That the FI is a less than perfect prognosticator is expected, since sequence space does not completely account for oligosaccharide evolution. A critical missing factor is the fitness of the various glycoforms, both in terms of viral replication in the human host, and also the ability of virus to evade neutralizing antibodies. Oligosaccharides are well known to influence HA function, particularly binding to host cell receptors and of course, in shielding HA from Ab mediated neutralization.
Indeed, the major point of this work is that oligosaccharides influence HA evolution in antigenic regions. Notably, while the number of oligosaccharides in the globular domain has little gross effect on the overall variation (Figure 1d), it focuses variation on uncovered antigenic epitopes. This supports the idea that glycosylation is an effective strategy for deflecting neutralizing Abs. Why then, doesn't HA simply cover itself with oligosaccharides?
The likely answer is that HA simply can't block all neutralization sites with oligosaccharides and maintain its function. This may be a more difficult evolutionary task than it appears at first glance, since HIV gp160 is the exception among viral receptor proteins rather than the rule. Perhaps there are yet to be defined host molecules that recognize hyperglycosylated proteins to limit this strategy.
A first set of 4781 full length HA sequences (full-length sequences from all hosts and geographic origins) were downloaded on June 26th, 2009 from the influenza virus resource at the NCBI (http://www.ncbi.nlm.nih.gov/genomes/FLU). These include 1907 H1N1, 83 H2N2 and 2791 H3N2 sequences. A second set of 212 swine origin influenza virus (SOIV) H1N1 HA sequences) was downloaded on 12th October 2009, followed by a third set of 1339 full-length SOIV sequence was downloaded on March 31st 2010.
The NetNGlyc 1.0 web-server (http://www.cbs.dtu.dk/services/NetNGlyc) was used to predict N-Glycosylation sites (Asn-Xaa-Ser/Thr, where Xaa is any amino acid except Pro) of all HA sequences; a positive was scored when the jury returned a “+” score. According to NetNGlcy, 76% of positive scored sequons are modified by N-Glycans, with a bias towards Thr containing sequons [21], [22], [23], [24].
All HA sequences of H1N1 were aligned in a single common alignment using the program Muscle [25] with default parameters.
The amino acids composition of the sequences was used to perform a multivariate analysis called Principal Components Analysis (PCA). The PCA analysis of the amino acids composition was performed using the prcomp function of the R package (http://www.r-project.org). This analysis performs a decomposition of the variables, e.g. the abundance of each amino acid (20 variables), into each principal component. The first two components of the PCA, showed in the plots 1 and 2, preserve 59% of the total variability (Figure S2).
Amino acid variability was quantitated from position 58 through 272 (globular domain). Figure 3 shows the amount of variability in H1 HA at each position. Variability was quantitated by counting the number of different amino acids found at each position, i.e. a position where all sequences have the same amino acid, the value of variability is 1, while for example a variability value of 7 corresponds to a position that have 7 different possible amino acids. Likewise, variability of H2 HA and some H3 HAs at each position were calculated (Figure 4 and 6). Regions of the H1 HA globular domain 91, 129 and 162+/−5 amino acids were selected to calculate the Flow Index.
H1 HA sequences were sorted based on their glycosylation status (i.e., Ø; 91; 129; 162; 91,129; 91,162; 91,129,162). Sequences with the glycosylation sites at positions 129 and 162 were not found. The amino acids frequencies in each aligned amino acid position of these regions for each starting group were calculated. Then, using the amino acids frequencies at each position, a set of 10,000 “random” sequences of each group was generated. These “random” sequences, which maintain the amino acids frequencies of the actual sequences, correspond to the initial “pre-state” to run the simulations.
We performed two independent rounds of simulation (flow-charted in Figure S1). Since the tendency of the virus is to maintain its glycosylation status, a change in status rarely occurs in a single round of simulation. The first round (left side Figure S1) uses the amino acids frequencies of each pre-state. Then, choosing a position at random in a glycosylation region, an amino acid substitution based on the amino acid frequencies at the same position is made (i.e. random substitution guided by the amino acids frequencies of the pre-state sequence). Using this data set, we enumerated the number of times that single changes in glycosylation site number occurred (gain or loss) per 10,000 iterations, and calculated the Pdi, the probability of changing glycosylation status. In the second simulation round (right side, Figure S1), repeated rounds of simulation are performed until a change occurs, resulting in Pdi→j the probability of changing from a pre-state to a post-state that differ by a single glycosylation site (gain or loss). The Flow Index (FI) is defined as the product of the two rounds and provides a measure of the tendency of changing from a pre-state i to a post-state j.Since the FI is based on the frequency of amino acid of all sequences in the starting group, it is free of constraints imposed by a consensus sequence. In addition, the FI also takes selection into account, since only sequences of viable viruses are used in the simulated mutagenesis.
|
10.1371/journal.pcbi.1002267 | Dynamical and Structural Analysis of a T Cell Survival Network Identifies Novel Candidate Therapeutic Targets for Large Granular Lymphocyte Leukemia | The blood cancer T cell large granular lymphocyte (T-LGL) leukemia is a chronic disease characterized by a clonal proliferation of cytotoxic T cells. As no curative therapy is yet known for this disease, identification of potential therapeutic targets is of immense importance. In this paper, we perform a comprehensive dynamical and structural analysis of a network model of this disease. By employing a network reduction technique, we identify the stationary states (fixed points) of the system, representing normal and diseased (T-LGL) behavior, and analyze their precursor states (basins of attraction) using an asynchronous Boolean dynamic framework. This analysis identifies the T-LGL states of 54 components of the network, out of which 36 (67%) are corroborated by previous experimental evidence and the rest are novel predictions. We further test and validate one of these newly identified states experimentally. Specifically, we verify the prediction that the node SMAD is over-active in leukemic T-LGL by demonstrating the predominant phosphorylation of the SMAD family members Smad2 and Smad3. Our systematic perturbation analysis using dynamical and structural methods leads to the identification of 19 potential therapeutic targets, 68% of which are corroborated by experimental evidence. The novel therapeutic targets provide valuable guidance for wet-bench experiments. In addition, we successfully identify two new candidates for engineering long-lived T cells necessary for the delivery of virus and cancer vaccines. Overall, this study provides a bird's-eye-view of the avenues available for identification of therapeutic targets for similar diseases through perturbation of the underlying signal transduction network.
| T-LGL leukemia is a blood cancer characterized by an abnormal increase in the abundance of a type of white blood cell called T cell. Since there is no known curative therapy for this disease, identification of potential therapeutic targets is of utmost importance. Experimental identification of manipulations capable of reversing the disease condition is usually a long, arduous process. Mathematical modeling can aid this process by identifying potential therapeutic interventions. In this work, we carry out a systematic analysis of a network model of T cell survival in T-LGL leukemia to get a deeper insight into the unknown facets of the disease. We identify the T-LGL status of 54 components of the system, out of which 36 (67%) are corroborated by previous experimental evidence and the rest are novel predictions, one of which we validate by follow-up experiments. By deciphering the structure and dynamics of the underlying network, we identify component perturbations that lead to programmed cell death, thereby suggesting several novel candidate therapeutic targets for future experiments.
| Living cells perceive and respond to environmental perturbations in order to maintain their functional capabilities, such as growth, survival, and apoptosis. This process is carried out through a cascade of interactions forming complex signaling networks. Dysregulation (abnormal expression or activity) of some components in these signaling networks affects the efficacy of signal transduction and may eventually trigger a transition from the normal physiological state to a dysfunctional system [1] manifested as diseases such as diabetes [2], [3], developmental disorders [4], autoimmunity [5] and cancer [4], [6]. For example, the blood cancer T-cell large granular lymphocyte (T-LGL) leukemia exhibits an abnormal proliferation of mature cytotoxic T lymphocytes (CTLs). Normal CTLs are generated to eliminate cells infected by a virus, but unlike normal CTLs which undergo activation-induced cell death after they successfully fight the virus, leukemic T-LGL cells remain long-term competent [7]. The cause of this abnormal behavior has been identified as dysregulation of a few components of the signal transduction network responsible for activation-induced cell death in T cells [8].
Network representation, wherein the system's components are denoted as nodes and their interactions as edges, provides a powerful tool for analyzing many complex systems [9], [10], [11]. In particular, network modeling has recently found ever-increasing applications in understanding the dynamic behavior of intracellular biological systems in response to environmental stimuli and internal perturbations [12], [13], [14]. The paucity of knowledge on the biochemical kinetic parameters required for continuous models has called for alternative dynamic approaches. Among the most successful approaches are discrete dynamic models in which each component is assumed to have a finite number of qualitative states, and the regulatory interactions are described by logical functions [15]. The simplest discrete dynamic models are the so-called Boolean models that assume only two states (ON or OFF) for each component. These models were originally introduced by S. Kauffman and R. Thomas to provide a coarse-grained description of gene regulatory networks [16], [17].
A Boolean network model of T cell survival signaling in the context of T-LGL leukemia was previously constructed by Zhang et al [18] through performing an extensive literature search. This network consists of 60 components, including proteins, mRNAs, and small molecules (see Figure 1). The main input to the network is “Stimuli”, which represents virus or antigen stimulation, and the main output node is “Apoptosis”, which denotes programmed cell death. Based on a random order asynchronous Boolean dynamic model of the assembled network, Zhang et al identified a minimal number of dysregulations that can cause the T-LGL survival state, namely overabundance or overactivity of the proteins platelet-derived growth factor (PDGF) and interleukin 15 (IL15). Zhang et al carried out a preliminary analysis of the network's dynamics by performing numerical simulations starting from one specific initial condition (corresponding to resting T cells receiving antigen stimulation and over-abundance of the two proteins PDGF and IL15). Once the known deregulations in T-LGL leukemia were reproduced, each of these deregulations was interrupted individually, by setting the node's status to the opposite state, to predict key mediators of the disease. Yet, a complete dynamic analysis of the system, including identification of the attractors (e.g. steady states) of the system and their corresponding basin of attraction (precursor states), as well as a thorough perturbation analysis of the system considering all possible initial states, is lacking. Performing this analysis can provide deeper insights into unknown aspects of T-LGL leukemia.
Stuck-at-ON/OFF fault is a very common dysregulation of biomolecules in various cancer diseases [19]. For example, stuck-at-ON (constitutive activation) of the RAS protein in the mitogen-activated protein kinase pathways leads to aberrant cell proliferation and cancer [19], [20]. Thus identifying components whose stuck-at values result in the clearance, or alternatively, the persistence of a disease is extremely beneficial for the design of intervention strategies. As there is no known curative therapy for T-LGL leukemia, identification of potential therapeutic targets is of utmost importance [21].
In this paper, we carry out a detailed analysis of the T-LGL signaling network by considering all possible initial states to probe the long-term behavior of the underlying disease. We employ an asynchronous Boolean dynamic framework and a network reduction method, which we previously proposed [22], to identify the attractors of the system and analyze their basins of attraction. This analysis allows us to confirm or predict the T-LGL states of 54 components of the network. The predicted state of one of the components (SMAD) is validated by new wet-bench experiments. We then perform node perturbation analysis using the dynamic approach and a structural method proposed in [23] to study to what extent does each component contribute to T-LGL leukemia. Both methods give consistent results and together identify 19 key components whose disruption can reverse the abnormal state of the signaling network, thereby uncovering potential therapeutic targets for this disease, some of which are also corroborated by experimental evidence.
Any biological regulatory network can be represented by a directed graph G = (V, E) where V = {v1, v2,…, vn} is the set of vertices (nodes) describing different components of the system, and E is the set of edges denoting the regulatory interactions among the components. The orientation of each edge in the network follows the direction of mass transfer or information propagation from the upstream to the downstream node. Each edge can be also characterized with a sign where a positive sign denotes activation and a negative sign signifies inhibition. The source nodes (i.e. nodes with no incoming edges) of this graph, if they exist, represent external inputs (signals), and one or more nodes, usually sink nodes (i.e. nodes with no outgoing edges), are customarily designated as outputs of the network.
Boolean models belong to the class of discrete dynamic models in which each node of the network is characterized by an ON (1) or OFF (0) state and usually the time variable t is also considered to be discrete, i.e. it takes nonnegative integer values [24], [25]. The future state of each node vi is determined by the current states of the nodes regulating it according to a Boolean transfer function , where ki is the number of regulators of vi. Each Boolean function (rule) represents the regulatory relationships between the components and is usually expressed via the logical operators AND, OR and NOT. The state of the system at each time step is denoted by a vector whose ith component represents the state of node vi at that time step. The discrete state space of a system can be represented by a state transition graph whose nodes are states of the system and edges are allowed transitions among the states. By updating the nodes' states at each time step, the state of the system evolves over time and following a trajectory of states it eventually settles down into an attractor. An attractor can be in the form of either a fixed point, in which the state of the system does not change, or a complex attractor, where the system oscillates (regularly or irregularly) among a set of states. The set of states leading to a specific attractor is called the basin of attraction of that attractor.
In order to evaluate the state of each node at a given time instant, synchronous as well as asynchronous updating strategies have been proposed [24], [25]. In the synchronous method all nodes of the network are updated simultaneously at multiples of a common time step. The underlying assumption of this update method is that the timescales of all the processes occurring in a system are similar. This is a quite strong and potentially unrealistic assumption, which in particular may not be suited for intracellular biological processes due to the variety of timescales associated with transcription, translation and post-translational mechanisms [26]. To overcome this limitation, various asynchronous methods have been proposed wherein the nodes are updated based on individual timescales [25], [27], [28], [29], [30], including deterministic methods with fixed node timescales and stochastic methods such as random order asynchronous method [27] wherein the nodes are updated in random permutations. In a previous work [22], we carried out a comparative study of three different asynchronous methods applied to the same biological system. That study suggested that the general asynchronous (GA) method, wherein a randomly selected node is updated at each time step, is the most efficient and informative asynchronous updating strategy. This is because deterministic asynchronous [22] or autonomous [30] Boolean models require kinetic or timing knowledge, which is usually missing, and random order asynchronous models [27] are not computationally efficient compared to the GA models. In addition, the superiority of the GA approach has been corroborated by other researchers [29] and the method has been used in other studies as well [31], [32]. We thus chose to employ the GA method in this work, and we implemented it using the open-source software library BooleanNet [33]. It is important to note that the stochasticity inherent to this method may cause each state to have multiple successors, and thus the basins of attraction of different attractors may overlap. For systems with multiple fixed-point attractors, the absorption probabilities to each fixed point can be computed through the analysis of the Markov chain and transition matrix associated with the state transition graph of the system [34]. Given a fixed point, node perturbations can be performed by reversing the state of the nodes i.e. by knocking out the nodes that stabilize in an ON state in the fixed point or over-expressing the ones that stabilize in an OFF state.
A Boolean network with n nodes has a total of 2n states. This exponential dependence makes it computationally intractable to map the state transition graphs of even relatively small networks. This calls for developing efficient network reduction approaches. Recent efforts towards addressing this challenge consists of iteratively removing single nodes that do not regulate their own function and simplifying the redundant transfer functions using Boolean algebra [35], [36]. Naldi et al [35] proved that this approach preserves the fixed points of the system and that for each (irregular) complex attractor in the original asynchronous model there is at least one complex attractor in the reduced model (i.e. network reduction may create spurious oscillations). Boolean networks often contain nodes whose states stabilize in an attracting state after a transient period, regardless of updating strategy or initial conditions. The attracting states of these nodes can be readily identified by inspection of their Boolean functions. In a previous work [22] we proposed a method of network simplification by (i) pinpointing and eliminating these stabilized nodes and (ii) iteratively removing a simple mediator node (e.g. a node that has one incoming edge and one outgoing edge) and connecting its input(s) to its target(s). Our simplification method shares similarities with the method proposed in [35], [36], with the difference that we only remove stabilized nodes (which have the same state on every attractor) and simple mediator nodes rather than eliminating each node without a self loop. Thus their proof regarding the preservation of the steady states by the reduction method holds true in our case. We employed this simplification method for the analysis of a signal transduction network in plants and verified by using numerical simulations that it preserves the attractors of that system. In this work, we employ this reduction method to simplify the T-LGL leukemia signal transduction network synthesized by Zhang et al [18], thereby facilitating its dynamical analysis. We also note that the first step of our simplification method is similar to the logical steady state analysis implemented in the software tool CellNetAnalyzer [37], [38]. We thus refer to this step as logical steady state analysis throughout the paper.
It should be noted that the fixed points of a Boolean network are the same for both synchronous and asynchronous methods. In order to obtain the fixed points of a system one can solve the set of Boolean equations independent of time. To this end, we first fix the state of the source nodes. We then determine the nodes whose rules depend on the source nodes and will either stabilize in an attracting state after a time delay or otherwise their rules can be simplified significantly by plugging in the state of the source nodes. Iteratively inserting the states of stabilized nodes in the rules (i.e. employing logical steady state analysis) will result in either the fixed point(s) of the system, or the partial fixed point(s) and a remaining set of equations to be solved. In the latter case, if the remaining set of equations is too large to obtain its fixed point(s) analytically, we take advantage of the second step of our reduction method [22] to simplify the resulting network and to determine a simpler set of Boolean rules. By solving this simpler set of equations (or performing numerical simulations, if necessary) and plugging the solutions into the original rules, we can then find the states of the removed nodes and determine the attractors of the whole system accordingly. For the analysis of basins of attraction of the attractors, we perform numerical simulations using the GA update method.
The topology (structure) and the function of biological networks are closely related. Therefore, structural analysis of biological networks provides an alternative way to understand their function [39], [40]. We have recently proposed an integrative method to identify the essential components of any given signal transduction network [23]. The starting point of the method is to represent the combinatorial relationship of multiple regulatory interactions converging on a node v by a Boolean rule:where uij's are regulators of node v. The method consists of two main steps. The first step is the expansion of a signaling network to a new representation by incorporating the sign of the interactions as well as the combinatorial nature of multiple converging interactions. This is achieved by introducing a complementary node for each component that plays a role in negative regulations (NOT operation) as well as introducing a composite node to denote conditionality among two or more edges (AND operation). This step eliminates the distinction of the edge signs; that is, all directed edges in the expanded network denote activation. In addition, the AND and OR operators can be readily distinguished in the expanded network, i.e., multiple edges ending at composite nodes are added by the AND operator, while multiple edges ending at original or complementary nodes are cumulated by the OR operator. The second step is to model the cascading effects following the loss of a node by an iterative process that identifies and removes nodes that have lost their indispensable regulators. These two steps allow ranking of the nodes by the effects of their loss on the connectivity between the network's input(s) and output(s). We proposed two connectivity measures in [23], namely the simple path (SP) measure, which counts the number of all simple paths from inputs to outputs, and a graph measure based on elementary signaling modes (ESMs), defined as a minimal set of components that can perform signal transduction from initial signals to cellular responses. We found that the combinatorial aspects of ESMs pose a substantial obstacle to counting them in large networks and that the SP measure has a similar performance as the ESM measure since both measures incorporate the cascading effects of a node's removal arising from the synergistic relations between multiple interactions. Therefore, we employ the SP measure and define the importance value of a component v as:where NSP(Gexp) and NSP(GΔv) denote the total number of simple paths from the input(s) to the output(s) in the original expanded network Gexp and the damaged network GΔv upon disruption of node v, respectively. This essentiality measure takes values in the interval [0,1], with 1 indicating a node whose loss causes the disruption of all paths between the input and output node(s). In this paper, we also make use of this structural method to identify essential components of the T-LGL leukemia signaling network. We then relate the importance value of nodes to the effects of their knockout (sustained OFF state) in the dynamic model and the importance value of complementary nodes to the effects of their original nodes' constitutive activation (sustained ON state) in the dynamic model.
The T-LGL signaling network reconstructed by Zhang et al [18] contains 60 nodes and 142 regulatory edges. Zhang et al used a two-step process: they first synthesized a network containing 128 nodes and 287 edges by extensive literature search, then simplified it with the software NET-SYNTHESIS [42], which constructs the sparsest network that maintains all of the causal (upstream-downstream) effects incorporated in a redundant starting network. In this study, we work with the 60-node T-LGL signaling network reported in [18], which is redrawn in Figure 1. The Boolean rules for the components of the network were constructed in [18] by synthesizing experimental observations and for convenience are given in Table S1 as well. The description of the node names and abbreviations are provided in Table S2.
To reduce the computational burden associated with the large state space (more than 1018 states for 60 nodes), we simplified the T-LGL network using the reduction method proposed in [22] (see Materials and Methods). We fixed the six source nodes in the states given in [18], i.e. Stimuli, IL15, and PDGF were fixed at ON and Stimuli2, CD45, and TAX were fixed at OFF. We used the Boolean rules constructed in [18], with one notable difference. The Boolean rules for all the nodes in [18], except Apoptosis, contain the expression “AND NOT Apoptosis”, meaning that if Apoptosis is ON, the cell dies and correspondingly all other nodes are turned OFF. To focus on the trajectory leading to the initial turning on of the Apoptosis node, we removed the “AND NOT Apoptosis” from all the logical rules. This allows us to determine the stationary states of the nodes in a live cell. We determined which nodes' states stabilize using the first step of our simplification method, i.e. logical steady state analysis (see Materials and Methods). Our analysis revealed that 36 nodes of the network stabilize in either an ON or OFF state. In particular, Proliferation and Cytoskeleton signaling, two output nodes of the network, stabilize in the OFF and ON state, respectively. Low proliferation in leukemic LGL has been observed experimentally [43], which supports our finding of a long-term OFF state for this output node. The ON state of Cytoskeleton signaling may not be biologically relevant as this node represents the ability of T cells to attach and move which is expected to be reduced in leukemic T-LGL compared to normal T cells. The nodes whose stabilized states cannot be readily obtained by inspection of their Boolean rules form the sub-network represented in Figure 2A. The Boolean rules of these nodes are listed in Table S3 wherein we put back the “AND NOT Apoptosis” expression into the rules.
Next, we identified the attractors (long-term behavior) of the sub-network represented in Figure 2A (see Materials and Methods). We found that upon activation of Apoptosis all other nodes stabilize at OFF, forming the normal fixed point of the system, which represents the normal behavior of programmed cell death. When Apoptosis is stabilized at OFF, the two nodes in the top sub-graph oscillate while all the nodes in the bottom sub-graph are stabilized at either ON or OFF. As shown in Figure 3, the state space of the two oscillatory nodes, TCR and CTLA4, forms a complex attractor in which the average fraction of ON states for either node is 0.5. Given that these two nodes have no effect on any other node under the conditions studied here (i.e. stable states of the source nodes), their behavior can be separated from the rest of the network.
The bottom sub-graph exhibits the normal fixed point, as well as two T-LGL (disease) fixed points in which Apoptosis is OFF. The only difference between the two T-LGL fixed points is that the node P2 is ON in one fixed point and OFF in the other, which was expected due to the presence of a self-loop on P2 in Figure 2A. P2 is a virtual node introduced to mediate the inhibition of interferon-γ translation in the case of sustained activity of the interferon-γ protein (IFNG in Figure 2A). The node IFNG is also inhibited by the node SMAD which stabilizes in the ON state in both T-LGL fixed points. Therefore IFNG stabilizes at OFF, irrespective of the state of P2, as supported by experimental evidence [44]. Thus the biological difference between the two fixed points is essentially a memory effect, i.e. the ON state of P2 indicates that IFNG was transiently ON before stabilizing in the OFF state. In the two T-LGL fixed points for the bottom sub-graph of Figure 2A, the nodes sFas, GPCR, S1P, SMAD, MCL1, FLIP, and IAP are ON and the other nodes are OFF. We found by numerical simulations using the GA method (see Materials and Methods) that out of 65,536 total states in the state transition graph, 53% are in the exclusive basin of attraction of the normal fixed point, 0.24% are in the exclusive basin of attraction of the T-LGL fixed point wherein P2 is ON and 0.03% are in the exclusive basin of attraction of the T-LGL fixed point wherein P2 is OFF. Interestingly, there is a significant overlap among the basins of attraction of all the three fixed points. The large basin of attraction of the normal fixed point is partly due to the fact that all the states having Apoptosis in the ON state (that is, half of the total number of states) belong to the exclusive basin of the normal fixed point. These states are not biologically relevant initial conditions but they represent potential intermediary states toward programmed cell death and as such they need to be included in the state transition graph.
Since the state transition graph of the bottom sub-graph given in Figure 2A is too large to represent and to further analyze (e.g. to obtain the probabilities of reaching each of the fixed points), we applied the second step of the network reduction method proposed in [22]. This step preserves the fixed points of the system (see Materials and Methods), and since the only attractors of this sub-graph are fixed points, the state space of the reduced network is expected to reflect the properties of the full state space. Correspondingly, the nodes having in-degree and out-degree of one (or less) in the sub-graph on Figure 2A, such as sFas, MCL1, IAP, GPCR, SMAD, and CREB, can be safely removed without losing any significant information as such nodes at most introduce a delay in the signal propagation. In addition, we note that although the node P2 has a self-loop and generates a new T-LGL fixed point as described before, it can also be removed from the network since the two fixed points differ only in the state of P2 and thus correspond to biologically equivalent disease states. We revisit this node when enumerating the attractors of the original network. In the resulting simplified network, the nodes BID, Caspase, and IFNG would also have in-degree and out-degree of one (or less) and thus can be safely removed as well. This reduction procedure results in a simple sub-network represented in Figure 2B with the Boolean rules given in Table 1.
Our attractor analysis revealed that this sub-network has two fixed points, namely 000001 and 110000 (the digits from left to right represent the state of the nodes in the order as listed from top to bottom in Table 1). The first fixed point represents the normal state, that is, the apoptosis of CTL cells. Note that the OFF state of other nodes in this fixed point was expected because of the presence of “AND NOT Apoptosis” in all the Boolean rules. The second fixed point is the T-LGL (disease) one as Apoptosis is stabilized in the OFF state. We note that the sub-network depicted in Figure 2B contains a backbone of activations from Fas to Apoptosis and two nodes (S1P and FLIP) which both have a mutual inhibitory relationship with the backbone. If activation reaches Apoptosis, the system converges to the normal fixed point. In the T-LGL fixed point, on the other hand, the backbone is inactive while S1P and FLIP are active.
We found by simulations that for the simplified network of Figure 2B, 56% of the states of the state transition graph (represented in Figure 4) are in the exclusive basin of attraction of the normal fixed point while 5% of the states form the exclusive basin of attraction of the T-LGL fixed point. Again, the half of state space that has the ON state of Apoptosis belongs to the exclusive basin of attraction of the normal fixed point. Notably, there is a significant overlap between the basins of attraction of the two fixed points, which is illustrated by a gray color in Figure 4. The probabilities of reaching each of the two fixed points starting from these gray-colored states, found by analysis of the corresponding Markov chain (see Materials and Methods), are given in Figure 5. As this figure represents, for the majority of cases the probability of reaching the normal fixed point is higher than that of the T-LGL fixed point. The three states whose probabilities to reach the T-LGL fixed point are greater than or equal to 0.7 are one step away either from the T-LGL fixed point or from the states in its exclusive basin of attraction. In two of them, the backbone of the network in Figure 2B is inactive, and in the third one the backbone is partially inactive and most likely will remain inactive due to the ON state of S1P (one of the two nodes having mutual inhibition with the backbone).
Based on the sub-network analysis and considering the states of the nodes that stabilized at the beginning based on the logical steady state analysis, we conclude that the whole T-LGL network has three attractors, namely the normal fixed point wherein Apoptosis is ON and all other nodes are OFF, representing the normal physiological state, and two T-LGL attractors in which all nodes except two, i.e. TCR and CTLA4, are in a steady state, representing the disease state. These T-LGL attractors are given in the second column of Table 2, which presents the predicted T-LGL states of 54 components of the network (all but the six source nodes whose state is indicated at the beginning of the Results section). We note that the two T-LGL attractors essentially represent the same disease state since they only differ in the state of the virtual node P2. Moreover, this disease state can be considered as a fixed point since only two nodes oscillate in the T-LGL attractors. For this reason we will refer to this state as the T-LGL fixed point. It is expected that the basins of attraction of the fixed points have similar features as those of the simplified networks.
Experimental evidence exists for the deregulated states of 36 (67%) components out of the 54 predicted T-LGL states as summarized in the third column of Table 2. For example, the stable ON state of MEK, ERK, JAK, and STAT3 indicates that the MAPK and JAK-STAT pathways are activated. The OFF state of BID is corroborated by recent evidence that it is down-regulated both in natural killer (NK) and in T cell LGL leukemia [45]. In addition, the node RAS was found to be constitutively active in NK-LGL leukemia [41], which indirectly supports our result on the predicted ON state of this node. For three other components, namely, GPCR, DISC, and IFNG, which were classified as being deregulated without clear evidence of either up-regulation or down-regulation in [18], we found that they eventually stabilize at ON, OFF, and OFF, respectively. The OFF state of IFNG and DISC is indeed supported by experimental evidence [44], [46]. In the second column of Table 2, we indicated with an asterisk the stabilized state of 17 components that were experimentally undocumented before and thus are predictions of our steady state analysis (P2 was not included as it is a virtual node). We note that ten of these cases were also predicted in [18] by simulations.
The predicted T-LGL states of these 17 components can guide targeted experimental follow-up studies. As an example of this approach, we tested the predicted over-activity of the node SMAD (see Materials and Methods). As described in [18] the SMAD node represents a merger of SMAD family members Smad 2, 3, and 4. Smad 2 and 3 are receptor-regulated signaling proteins which are phosphorylated and activated by type I receptor kinases while Smad4 is an unregulated co-mediator [47]. Phosphorylated Smad2 and/or Smad3 form heterotrimeric complexes with Smad4 and these complexes translocate to the nucleus and regulate gene expression. Thus an ON state of SMAD in the model is a representation of the predominance of phosphorylated Smad2 and/or phosphorylated Smad3 in T-LGL cells. In relative terms as compared to normal (resting or activated) T cells, the predicted ON state implies a higher level of phosphorylated Smad2/3 in T-LGL cells as compared to normal T cells. Indeed, as shown in Figure 6, T cells of T-LGL patients tend to have high levels of phosphorylated Smad2/3, while normal activated T cells have essentially no phosphorylated Smad2/3. Thus our experiments validate the theoretical prediction.
A question of immense biological importance is which manipulations of the T-LGL network can result in consistent activation-induced cell death and the elimination of the dysregulated (diseased) behavior. We can rephrase and specify this question as which node perturbations (knockouts or constitutive activations) lead to a system that has only the normal fixed point. These perturbations can serve as candidates for potential therapeutic interventions. To this end, we performed node perturbation analysis using both structural and dynamic methods.
In this paper we presented a comprehensive analysis of the T-LGL survival signaling network to unravel the unknown facets of this disease. By using a reduction technique, we first identified the fixed points of the system, namely the normal and T-LGL fixed points, which represent the healthy and disease states, respectively. This analysis identified the T-LGL states of 54 components of the network, out of which 36 (67%) are corroborated by previous experimental evidence and the rest are novel predictions. These new predictions include RAS, PLCG1, IAP, TNF, NFAT, GRB2, FYN, SMAD, P27, and Cytoskeleton signaling, which are predicted to stabilize at ON in T-LGL leukemia and GAP, SOCS, TRADD, ZAP70, and CREB which are predicted to stabilize at OFF. In addition, we found that the node P2 can stabilize in either the ON or OFF state, whereas two nodes, TCR and CTLA4, oscillate. We have experimentally validated the prediction that the node SMAD is over-active in leukemic T-LGL by demonstrating the predominant phosphorylation of the SMAD family members Smad2 and Smad3. The predicted T-LGL states of other nodes provide valuable guidance for targeted experimental follow-up studies of T-LGL leukemia.
Among the predicted states, the ON state of Cytoskeleton signaling may not be biologically relevant as this node represents the ability of T cells to attach and move which is expected to be reduced in leukemic T-LGL compared to normal T cells. This discrepancy may be due to the fact that the network contains insufficient detail regarding the regulation of the cytoskeleton, as there is only one node, FYN, upstream of Cytoskeleton signaling in the network. While the network is able to successfully capture survival signaling without necessarily capturing the cytoskeleton signaling, this discrepancy suggests that follow-up experimental studies should be conducted to determine the relationship between cytoskeleton signaling and survival signaling in the T-LGL network. We note that in the case of perturbation of TBET, PI3K, NFκB, JAK, or SOCS, the node Cytoskeleton signaling exhibits oscillatory behavior induced by oscillations in TCR. At present it is not known whether this predicted behavior is relevant.
Using the general asynchronous (GA) Boolean dynamic approach, we analyzed the basins of attraction of the fixed points. We found that the basin of attraction of the normal fixed point is larger than that of the T-LGL fixed point. The trajectories starting from each initial state toward the T-LGL fixed point (Figure 4) may be indicative of the accumulating deregulations that lead to the disease-associated stable survival state. Although the fixed points, being time independent, are the same for all update methods or implementations of time, the update method may affect the structure of the state transition graph of the system and the basins of attraction of the fixed points. We note that the GA method assumes that each node has an equal chance of being updated. If quantitative or kinetic information becomes available in this system, unequal probabilities may be implemented by grouping the nodes into several “priority classes” and assigning a weight to each class where higher weights indicate more probable transitions [51]. Incorporating such information into the state space may prune the allowed trajectories and give further insights into the accumulation of deregulations.
We took one step further by performing a perturbation analysis using dynamical and structural methods to identify the interventions leading to the disappearance of the disease fixed point. We note that our study has a dramatically larger scope than the previous key mediator analysis of Zhang et al [18]. For the dynamical analysis, we employed the GA approach instead of the random order asynchronous method and considered all possible initial conditions as opposed to performing numerical simulations using a specific initial condition. Zhang et al only focused on the node Apoptosis, and identified as “key mediators” the nodes whose altered state increases the frequency of ON state of Apoptosis. An increase in Apoptosis' ON state does not necessarily imply that apoptosis is the only possible final outcome of the system. In this work, after finding the fixed points, which completely describe the state of the whole system, we performed dynamic perturbation analysis by fixing the state of each node to its opposite state in the T-LGL fixed point and determining which fixed points were obtained and what their basins of attraction were. This way we were able to identify and distinguish the key mediators whose altered state completely eliminates the leukemic outcome, and those whose altered state reduces the basin of attraction of the leukemic outcome. Moreover, numerical simulations, as done in [18], may not be able to thoroughly sample different timing. In this study, using a reduction technique, we found the cases when timing does not matter with certainty (where there is only one fixed point), and also the cases in which timing and initial conditions may matter (where there are two reachable fixed points). For the perturbation analysis using the structural method, we used the simple path (SP) measure to identify important mediators of the disease outcome and observed consistent results with the dynamic analysis. Our dynamical and structural analysis led to the identification of 19 therapeutic targets (the first 19 nodes in the first column of Table 2), 53% of which are supported by direct experimental evidence and 15% of which are supported by indirect evidence.
Multi-stability (having multiple steady states) is an intrinsic dynamic property of many disease networks [52], [53], which is related to the presence of feedback loops in the network. In a graph-theoretical sense, a feedback loop is a directed cycle whose sign depends upon the parity of the number of negative interactions in the cycle. A positive/negative feedback loop has an even/odd number of negative interactions. It was conjectured that the presence of positive feedback loops in the network is necessary for multi-stability whereas the existence of negative feedback loops is required for having sustained oscillations [54]. From a biological point of view, the former dynamical property is associated with multiple cell types after differentiation while the latter is related to stable periodic behaviors such as circadian rhythms [55]. We note that the T-LGL signaling network consists of both positive and negative feedbacks and thus has a potential for both multi-stability and oscillations. Indeed, the negative feedback in the top sub-graph of Figure 2A causes the complex attractor shown in Figure 3. In contrast, the negative feedback on the node P2 of the bottom sub-graph is counteracted by the positive self-loop on the same node, thus no complex attractor is possible for the bottom sub-graph of Figure 2A. The two mutual inhibition-type positive feedback loops present in the bottom sub-graph and the self-loop on P2 generate the three fixed points, while the positive self-loop on Apoptosis maintains the normal fixed point once Apoptosis is turned ON.
Negative feedback loops can be a source of oscillations [56], homeostasis [56], or excitation-adaptation behavior [57]. Especially, when the activation is slower than the inhibitory interaction in the negative feedback, it can lead to sustained oscillations [56]. In the T-LGL network, the negative feedback loop between the T cell receptor TCR and CTLA4 modulates stimulus-induced activation of the receptor in such a way that CTLA4 is indirectly activated after prolonged TCR activation, whereas the inhibition of TCR by CTLA4 is a direct interaction [58]. That is, activation is slower than inhibition in the negative feedback and thus an oscillatory behavior reminiscent of that obtained by our asynchronous Boolean model would also be observed in continuous modeling frameworks as well. Although no time-measurements of the T cell receptor activity in T-LGL exist, it has been reported that there is variability for TCR activation in different patients ([43] and unpublished observation by T.P. Loughran), supporting the absence of a steady state behavior.
Our study revealed that both structural and dynamic analysis methods can be employed to identify therapeutic targets of a disease, however, they differ in implementation efficiency as well as the scope and applicability of the results. The structural analysis does not require mapping of the state space and thus is less computationally intensive and is more feasible for large network analysis, but it may not capture all the initial states and thus may miss or inaccurately identify some important features. The dynamic analysis method, while computationally intensive, yields a comprehensive picture of the state transition graph, including all possible fixed points of the system, their corresponding basins of attraction, as well as the relative frequency of trajectories leading to each fixed point. We demonstrated that the limitations related to the vast state space of large networks can be overcome by judicious use of the network reduction technique that we developed in our previous study [22]. We conclude that the structural method incorporating the cascading effects of node disruptions is best employed for quick exploratory analysis, and dynamic analysis should be performed to get a thorough and detailed insight into the behavior of a system. Overall, the combined analysis presented in this study opens a promising avenue to predict dysregulated components and identify potential therapeutic targets, and it is versatile enough to be successfully applied to a large variety of signal transduction and regulatory networks related to diseases.
|
10.1371/journal.ppat.1005703 | Chitosan Mediates Germling Adhesion in Magnaporthe oryzae and Is Required for Surface Sensing and Germling Morphogenesis | The fungal cell wall not only plays a critical role in maintaining cellular integrity, but also forms the interface between fungi and their environment. The composition of the cell wall can therefore influence the interactions of fungi with their physical and biological environments. Chitin, one of the main polysaccharide components of the wall, can be chemically modified by deacetylation. This reaction is catalyzed by a family of enzymes known as chitin deacetylases (CDAs), and results in the formation of chitosan, a polymer of β1,4-glucosamine. Chitosan has previously been shown to accumulate in the cell wall of infection structures in phytopathogenic fungi. Here, it has long been hypothesized to act as a 'stealth' molecule, necessary for full pathogenesis. In this study, we used the crop pathogen and model organism Magnaporthe oryzae to test this hypothesis. We first confirmed that chitosan localizes to the germ tube and appressorium, then deleted CDA genes on the basis of their elevated transcript levels during appressorium differentiation. Germlings of the deletion strains showed loss of chitin deacetylation, and were compromised in their ability to adhere and form appressoria on artificial hydrophobic surfaces. Surprisingly, the addition of exogenous chitosan fully restored germling adhesion and appressorium development. Despite the lack of appressorium development on artificial surfaces, pathogenicity was unaffected in the mutant strains. Further analyses demonstrated that cuticular waxes are sufficient to over-ride the requirement for chitosan during appressorium development on the plant surface. Thus, chitosan does not have a role as a 'stealth' molecule, but instead mediates the adhesion of germlings to surfaces, thereby allowing the perception of the physical stimuli necessary to promote appressorium development. This study thus reveals a novel role for chitosan in phytopathogenic fungi, and gives further insight into the mechanisms governing appressorium development in M.oryzae.
| Magnaporthe oryzae is a filamentous fungal pathogen which causes devastating crop losses in rice. Successful invasion of the host is dependent upon the ability of the fungus to remain undetected by the innate immune system of the plant, which recognizes conserved components of the fungal cell wall, such as chitin. Previous studies have demonstrated that infection-related changes in cell wall composition are necessary to allow the fungus to remain undetected during infection. One such change that has long been hypothesized to have a role as a 'stealth mechanism' is the deacetylation of the polysaccharide chitin by enzymes known as chitin deacetylases. The deacetylation of chitin produces a polysaccharide known as chitosan, which has previously been shown to accumulate specifically on infection structures in plant pathogenic fungi. However, in this study, we show that germling-localized chitosan is not required for pathogenicity, arguing against a role as a 'stealth mechanism' at this stage. Instead, chitosan is required for the development of the appressorium, a critical fungal infection structure required for the penetration of plant cells. This requirement can be attributed to chitosan mediating the adhesion of germlings to surfaces, which is required for the perception of physical stimuli.
| All fungal cells are encased within a cell wall. This complex and dynamic structural barrier is composed of interwoven polysaccharides and proteins. Indeed, the polysaccharide moiety makes up the majority of the fungal wall, being comprised of chitin (a polymer of β1,4-N-acetylglucosamine), β/α1,3-glucans and mannans [1]. There is, however, considerable variation in the proportion of these wall components between different cell-types and between fungal species [1]. The cell wall plays an essential role in maintaining cellular integrity in the face of the challenging and varied environmental conditions to which fungi are exposed. Indeed, the cell wall plays a far greater role than simply being the extracellular “coat of armour” [2]. Firstly, morphogenesis is heavily influenced by cell wall composition, and, conversely, localized changes in composition allow fungal cells to undergo morphogenesis, whilst maintaining cellular integrity [3]. This is illustrated by many studies, where chemical or genetic perturbation of cell wall composition has led to gross defects in fungal growth and morphogenesis (reviewed in [1, 4]). Secondly, the cell wall forms the interface between the fungal cell and its immediate environment. As a result, the structure and composition of the cell wall influences both physical and biological interactions which occur over the life-cycle of the fungus, as, for example, during host infection [2].
The hemibiotrophic fungus Magnaporthe oryzae causes significant losses of rice [5]. It is thus a notable pathogen but is also considered as a model organism to study appressorium formation in pathogenic fungi [6]. Disease occurs when three–celled asexual conidia land on the host, differentiate a short germ tube and thence an infection structure (the appressorium). This developmental progression occurs in response to hard, hydrophobic surface and following the perception of host-derived surface chemistries, such as cutin [7]. The maturing appressorium generates a considerable turgor pressure, a penetration peg emerges and penetrates the leaf cuticle [8]. Subsequently, invasive hyphae ramify though the host. Throughout this process the fungal wall undergoes extensive remodelling during rapid growth [9].
For successful infection M. oryzae must remain undetected by its host: but how does the fungus do this? Plants readily detect key molecular signatures of fungal cells, that is, Pathogen Associated Molecular Patterns (PAMPs). Fungal cell wall polysaccharides in particular represent a major source of PAMPs [10]. The best characterized of these are chitin oligomers, which are released from the fungal wall either as a result of endogenous cell wall remodelling, or due to the action of plant chitinases. Chitin oligomers are recognized in rice plants by the Pattern Recognition Receptor (PRR) CEBiP (Chitin Elicitor Binding Protein) [11]. Binding of chitin oligomers to CEBiP induces its dimerization and, in association with CERK1 (Chitin Elicitor Receptor Kinase-1), results in phosphorylation and activation of CEBiP [12]. This triggers PAMP Triggered Immunity (PTI) [13]. This response includes callose deposition, the production of degradative enzymes (chitinases and glucanases), and a burst of reactive oxygen species [14, 15]—all of which serve to restrict or kill the invading fungus. Phytopathogenic fungi have evolved a number of strategies to avoid triggering PTI. The first is by secretion of chitin-binding effectors, which compete with the chitin receptors to bind the oligomers, and thus enable full pathogenesis, as in Cladosporium fulvum and M.oryzae [16, 17]. The second strategy is to change cell wall composition to either alter identity, or to mask the presence of the PAMPs. For example, during infection-related development, M.oryzae synthesises α1,3-glucan as a component of the cell walls of germ tubes, appressoria and invasive hyphae [9]. Deletion of the sole gene encoding α1,3-glucan synthase (AGS1) results in enhanced susceptibility to enzymatic degradation, the triggering of PTI and thus loss of pathogenicity [18]. This suggests that α1,3-glucan acts as a 'stealth' mechanism, cloaking the fungus from recognition by the plant. We postulated that chitin deacetylation may play a similar role—as it constitutes another infection-related cell wall alteration, observed to occur in a number of fungal plant pathogens, including Uromyces fabae, Colletotrichum graminicola, Puccinia graminis and M.oryzae [9, 19].
Chitin deacetylation is catalyzed by a family of conserved carbohydrate esterase enzymes (CE-4 family) known as chitin deacetylases (CDAs) (E.C. 3.5.1.41). The removal of acetyl groups from N-acetylglucosamine residues of chitin results in the formation of chitosan, typically a heterogeneous polymer of β1,4-glucosamine and β1,4-N-acetylglucosamine [20]. The deacetylation of chitin confers two hypothetical benefits upon the invading fungus. Firstly, unlike chitin, chitosan cannot cause activation of CEBiP [12]. In addition, although various defence responses have been observed in plants treated with chitosan (reviewed in [21], it is unclear whether chitosan is as effective as chitin in this respect. Secondly, chitin deacetylation likely provides protection from plant chitinases, since chitosan is a poor substrate for these enzymes [22]. However, such roles for chitosan, acting as either a protectant or as a 'stealth' mechanism are not yet supported by direct experimental evidence. Moreover, chitin deacetylases genes are present in the genomes of all fungi [23]. Thus, chitosan may be a broadly-occurring cell wall component, with roles beyond those we hypothesize for plant pathogens. Indeed, such additional roles may relate to the different physical and chemical properties of chitin and chitosan. Chitin occurs in fungal cell walls as antiparallel chains, forming microfibrils with immense strength and rigidity [24]. However, chitin deacetylation creates primary amine groups, which are largely protonated and charged at physiological pH. Consequently, chitosan is polycationic, more hydrophilic and an amorphous polysaccharide, in stark contrast to its parent polymer chitin. Thus, deacetylation could have profound consequences on the structure and physical properties of the fungal cell wall, which, in turn, will impact on fungal growth and morphogenesis.
There is, however, little published data defining the role(s) of chitosan in fungal cell walls per se. The role of chitin deacetylation was first described in S.cerevisiae [25] Here, two functionally redundant CDAs deacetylate chitin specifically in the ascospore cell wall—the double knockout mutant Δcda1Δcda2 strain results in complete loss of chitosan from the ascospore wall. Whilst spore viability was unaffected, spores of the Δcda1Δcda2 strain were more susceptible to treatment with lytic enzymes, ether and heat shock, suggesting possible disorganization and increased permeability of the wall [25, 26]. Further experimentation showed that the outer dityrosine layer was absent in CDA deletion strains [27]. It is hypothesized that the dityrosine is cross-linked to the amine groups of the chitosan, thereby creating a rigid and impermeable ascospore cell wall [28]. Similarly, chitosan was shown to be a component of spore wall in Ashbya gossypii—deletion of the single CDA resulted in a complete loss of sporulation, suggesting that chitosan is required for spore development [29]. In the Basidiomycete fungus Cryptococcus neoformans, chitosan is a major component of the vegetative cell wall [30]. Deletion of 3 of the 4 CDA genes (CDA1-3) abolished chitosan synthesis [31]—these triple deletion strains were hypersusceptible to various cell wall perturbants, showed attenuated virulence and “leaked” melanin from their cell walls [32] -Chitosan is thus required for cell wall integrity in this fungus..
Deacetylation of chitin may play critical roles in fungal development, integrity and pathogenesis. Thus the two central hypotheses regarding its role are: i) chitin deacetylation is important for cellular development and morphogenesis and ii) chitosan acts mechanisitically as a 'stealth molecule' during plant infection in pathogenic fungi. Appressorium development in M.oryzae represents an amenable and relevant system to test these hypotheses. Firstly, chitosan is known to be a component of the germ tube and appressorium cell wall [9]. Secondly, protective cell wall components (α1,3-glucan) in appressoria are required for pathogenicity [18]; this may hold true for chitosan. Lastly, appressorium development likely requires increased cell wall flexibility—chitin deacetylation may enable such plasticity. Indeed, this may be the most likely scenario as deletion of a putative chitin deacetylase, CBP1 resulted in defective appressorium formation on artificial surfaces [33]. Yet, deletion of CBP1 also prevented hyphopodium and pre-invasive hypha formation on artificial surfaces (but not root surfaces), suggesting that chitosan plays a role in multiple infection-related cellular differentiation events [34]. However, chitosan synthesis was not consistently reduced in the cbp1 mutant [34] and neither was the chitin deacetylase activity of Cbp1 proven [33] and so the role of chitin deacetylation remains inconclusive. A comprehensive characterization of chitin deacetylases operating during appressorium development should provide a valuable insight into the roles of chitin deacetylation during infection in plant pathogenic fungi.
The localization of chitosan during appressorium development in M.oryzae has previously been investigated using Eosin Y (9). Whilst this dye binds to chitosan, via an electrostatic interaction, the specificity of this interaction has not been directly proven. We sought an alternative, more specific method, that is by immune-detection, with mAbG7, a monoclonal anti-chitosan antibody [35], and a polyclonal anti-chitosan antibody [19]. These antibodies revealed that chitosan localizes to the cell wall of the germ tube and the appressorium (Fig 1A & 1B), and is consistent with previous observations made with Eosin Y [9]. However, we wished to determine the precise timing of chitin deacetylation during appressorium development in order to assess whether it is, indeed, required for this process. To do this, we invoked use of the probe OGA488. This recently developed and highly specific probe [36] stains chitosan rapidly (within 15 minutes), and is thus ideally-suited to studying germlings in vivo at different stages of morphogenesis. Using this probe, germling development was tracked on an artificial hydrophobic surface inductive to appressorium formation [37]. After 1–2 hpi (hours post inoculation) the germ tube was weakly labelled, labelling strengthened upon onset of germ tube hooking (2–3 hpi), and intensified further during appressorium development (4 hpi). At this stage, the entire germ tube and appressorium wall were labelled—this being consistent with the immune-localisation data, (Fig 1C).
Next, we asked which of the 10 putative chitin deacetylase genes residing in the M.oryzae genome (http://www.broadinstitute.org/annotation/genome/magnaporthecomparative/MultiHome.html) are responsible for chitin deacetylation occuring prior to and during appressorium development. The 10 M.oryzae CDAs carry a polysaccharide/chitin deacetylase domain (Pfam 01522). All, except Cda9, have a predicted N-terminal signal peptide. Putative chitin binding domains (CBD; Pfam 00187) are present in Cbp1, Cda1, Cda2 and Cda7. Cbp1 also has a serine/threonine-rich repeat region at its C-terminus [33]. Cda4, Cda5 and Cda8 have putative single transmembrane domains, suggesting that these proteins localise to the membrane (Fig 2A).
Chitin deacetylation in Colletotrichum lindemuthianum is dependant upon a zinc-binding Asp-His-His triad, and four key active site residues: a catalytic base residue (Asp) and a catalytic acid residue (His), which interact with Arg and Asp residues respectively [38, 39]. Sequence alignment using Clustal Omega [40], revealed conservation of the zinc-binding triad and active-site resides in all M.oryzae CDA sequences, with the exception of a non-catalytic Asp in Cbp1 (MGG_12939) at position 153 of the alignment (S1 Fig).
To determine which of the 10 CDAs are expressed during appressorium development, we performed qRT-PCR analysis. Total RNA was extracted from rice leaves inoculated with WT conidia, at 5 hpi. Appressorium development was confirmed at this timepoint by microscopic observation (S2 Fig). Relative quantification of transcript abundances for all 10 CDAs revealed that CDA3 is the most highly expressed, followed by CDA2 and CBP1 (Fig 2B). Low levels of expression were observed for the remaining genes. In addition to this, we also analyzed the expression of the CDAs during later stages of leaf infection, at 36 hpi. In this case, CDA1 and CDA6 showed the highest transcript abundance (S2 Fig). These data align with published transcriptomics datasets [41, 42].
The combined transcriptional profiling data reveals that three particular CDA genes are likely involved in chitin deacetylation during appressorium development: CBP1, CDA2, and CDA3. CBP1 has previously been partially characterized [33], but we wished to further this work, in combination with CDA2, and CDA3 and thus generated single, double and triple deletion strains by a targeted gene replacement approach. All such mutants generated were confirmed by both PCR and by Southern Blot analysis (S3 and S4 Figs). Deletion strains are listed in S1 Table.
To compare appressorium development, conidia of the mutant and WT strains were inoculated onto an artificial hydrophobic glass surface. In the WT strain, over 80% of conidia developed appressoria after an 8 hr incubation (Fig 3A). However, very few appressoria formed in cbp1 at 8 hpi (3% ± 3% (SD n = 3))—germ tubes were abnormally elongated and failed to hook at the end distal from the spore (Fig 3B). This is consistent with the previous report [33]. Germination of cbp1 conidia was significantly higher than the WT strain at 1 hpi (Welch's ANOVA, p < 0.05, Fig 3A) and thus failure to form appressoria is not due to delayed or reduced germination. By 24 hpi, 83% ± 4% of cbp1 germ tubes formed appressoria (Fig 3A) that were misshapen in appearance (S6 Fig). Appressorium development is therefore both delayed and defective in cbp1. In the cda2, cda3 and cda2/cda3 strains, appressorium development and conidial germination progressed similarly to WT. However, deletion of CDA2 together with CBP1 resulted in much more severe defects in appressorium development than observed in the single deletion strain cbp1. Germlings of strain cda2/cbp1 failed to form appressoria by 8 hpi, as in cbp1, but extending the incubation period to 24 hpi resulted in only rare occurrences of appressorium development (27% ± 20 of germ tubes), in contrast to cbp1 (Fig 3A). The germ tubes of cda2/cbp1 appeared even more elongated (487 ± 42 μm in cda2/cbp1 compared with 264 ± 37 μm in cbp1 (S3 Table)), yet remained undifferentiated. This phenotype was also observed when conidia were germinated on other hydrophobic surfaces, including plastic and Parafilm (S5 Fig). CDA2 and CBP1 therefore exhibit partial redundancy. Conversely, deletion of CDA3 had no additive effects on appressorium development under the conditions tested: cbp1/cda3 demonstrated an identical phenotype to cbp1, and cda2/cbp1/cda3 a similar phenotype to cda2/cbp1.
To determine whether chitin deacetylation is reduced in the cda mutants, germlings of the deletion strains were stained with OGA488 (Fig 3B). WT germlings developed on an artificial hydrophobic surface showed strong staining of the cell wall in germ tubes and appressoria at 16 hpi, as described previously. Similar staining was observed in cda2/cda3. In cbp1 and cbp1/cda3, appressoria labelled with OGA488, but little fluorescence was observed on their elongate germ tubes. However, no chitosan staining was observed in strains cda2/cbp1 and cda2/cbp1/cda3, suggesting complete loss of chitin deacetylation activity, and providing further evidence of functional redundancy between cbp1 and cda2. By 24 hpi a small proportion of germ tubes in strains cda2/cbp1 and cda2/cbp1/cda3 formed appressoria. However, no chitosan was detected (S6A Fig), suggesting that it may not be an absolute requirement for appressorium morphogenesis. Yet, the presence of low concentrations of chitosan, or chitosan in cell wall regions inaccessible to the OGA488 probe cannot be completely discounted.
The highly elongate germ tubes observed in strains cbp1, cda2/cbp1 and cda2/cbp1/cda3 are a curious feature. Calcofluor White staining revealed that the mutant germ tubes are septated, with cross wall distributed along the elongate germ tubes of cbp1, cda2/cbp1 and cda2/cbp1/cda3 at regular intervals (S6B Fig). Taken together, these data suggest a clear link between the loss of chitin deacetylation, and loss of appressorium development.
We created fluorescent protein fusions, to better understand how chitin deacetylation by Cbp1 and Cda2 promotes appressorium development. C-terminal mCherry fusions of CBP1 and CDA2 were made, under the control of their respective native promoters, and transformed into their respective deletion background strains. Several independent transformant lines were characterized for each fusion and which exhibited identical patterns of fluorescence (S1 Table).
Strains expressing Cbp1:mCherry show fluorescence at all stages of germling morphogenesis (Fig 4A–4C), and demonstrate fully restored appressorium development (Fig 4D), indicating that the Cbp1:mCherry fusion protein is fully functional. Unexpectedly, fluorescence was observed in the cell wall at conidial apices, even in conidia which had not yet been harvested from the mycelium (Fig 4A). Substantial intracellular fluorescence was also observed at this stage, most likely localized to vacuoles. During initial stages of germ tube growth (1–2 hpi), weak fluorescence was observed at the lateral walls of the germ tube, but not the tip (Fig 4B). In addition, small punctae of intracellular fluorescence were sometimes observed along the length of the germ tube. At later stages of appressorium development (3–5 hpi), wall-localized fluorescence became stronger and localized to the entire germ tube and appressorium, although intracellular fluorescence was still apparent in some appressoria (Fig 4C). The appressorial wall remained fluorescent even at later stages (8–16 hpi, S7 Fig), although intensity decreased noticeably.
Similarly to Cbp1:mCherry, strains expressing Cda2:mCherry exhibited fluorescence during appressorium development, but with fluorescence typically appearing after hooking of the germ tube (Fig 4E). Strong fluorescence was also observed during subsequent appressorium formation (Fig 4F), and remained present at later stages (8 hpi), as observed with Cbp1:mCherry (S7 Fig). Unlike Cbp1:mCherry however, fluorescence was only very rarely observed intracellularly. Since the cda2 strain was apparently identical to the WT strain, functionality of the fusion protein could not be determined in the complemented strain. Attempts to create a Cda3:eGFP fusion protein did not yield fluorescent transformants, suggesting that the fusion protein is unstable.
The defects in appressorium development observed upon deletion of CDA genes was similar to that resulting from deletion of the hydrophobin gene MPG1 [43]. It was reported that appressorium development could be rescued in the mpg1 mutant by co-inoculation with WT conidia, suggesting that Mpg1 could act in trans [44]. To test whether chitosan could rescue the cda mutants, exogenous chitosan was added to conidia during germination on an artificial hydrophobic surface. Addition of 0.01% (w/v) or 0.001% (w/v) chitosan, surprisingly, restored appressorium development in all cda deletion strains (Fig 5A & 5B). Between 70–85% of cbp1, cda2/cbp1 and cbp1/cda3 conidia formed appressoria after an 8hr incubation in the presence of 0.01% or 0.001% chitosan, compared with 0–10% in the control treatment (water). In the triple deletion strain (cda2/cbp1/cda3) rescue was slightly lower (55% ± 17 (SD, n = 3) and 70% ± 16 (SD, n = 3) appressoria at 8 hpi for 0.01% and 0.001% chitosan, respectively. In the WT strain, 70–85% of conidia formed appressoria in all treatments. To further investigate this, different derivatives and formulations of chitosan were tested for their ability to restore appressorium development in the deletion strains (structures are shown in S8 Fig). Glycol chitosan, a soluble derivative of chitosan restored appressorium development in cda2/cbp1/cda3 at a concentration of 0.01% but not 0.001%. Carboxymethylchitosan (in which chitosan has been modified by the addition of an anionic carboxymethyl group) did not restore appressorium formation at either concentration. Lastly, the addition of chitosan in oligomeric form (oligochitosan, with a degree of polymerization of 22–33 residues) partially restored appressorium formation at 8 hpi, whereas glucosamine showed no restorative effects at all. Taken together, these data suggest that the cationic nature of exogenous chitosan is the most important factor determining rescue of the mutant phenotype, with steric factors also having some influence. Polymer length appears to be of little importance, given the broad ranges capable of restoring appressorium development.
To investigate how exogenous chitosan restores appressorium development in the cda mutants, fluorescently-labelled chitosan was used. FITC-chitosan, synthesized according to Qaqish et al [45], was added to conidia of the WT and cda2/cbp1/cda3 strains on an artificial hydrophobic surface. After 16 hr incubation, the FITC-chitosan was removed and the germlings imaged by confocal microscopy. The FITC-chitosan restored appressorium development in cda2/cbp1/cda3, and clear localization to the cell wall of germ tubes and appressoria (Fig 6A). In the WT strain, weak fluorescence was occasionally observed in the germ tubes. Exogenous chitosan is therefore associated with the cell wall of the deletion strains.
Chitosan localizes to the germ tube prior to appressorium development (Fig 1C). To determine whether this germ tube-localized chitosan is sufficient to induce appressorium development, FITC-chitosan was added to germinating conidia of WT or cda2/cbp1/cda3 strains for 2 hr only, before being washed off, and the germlings left to develop for a further 16 hr. In this way, chitosan was only present during the initial germ tube stage of germling morphogenesis, and was removed before appressorium development occurred. Appressorium development was restored in the deletion strain by this short incubation with FITC-chitosan. Here, fluorescence was only observed in the germ tube, and not the appressoria of cda2/cbp1/cda3 germlings, with similar localization in the WT (Fig 6B). Thus germ tube-localized chitosan seems sufficient to induce appressorium development.
Lastly, to investigate whether chitosan restores appressorium development at later stages of germling morphogenesis, conidia were incubated for 16 hr, FITC-chitosan added, and the conidia incubated for a further 8 hr. Again, appressorium development was restored in the cda2/cbp1/cda3 strain, although here the germ tubes were highly-elongate (Fig 6C). Fluorescence was observed along the entire length of these elongate germ tubes, but appeared more intense at regions proximal to its point of emergence from the conidium. In this case, WT germlings also exhibited weak fluorescence in both germ tubes and appressoria.
Appressorium development is highly defective in cda mutants on artificial hydrophobic surfaces, inductive to appressorium formation in the WT strain. To determine whether this results in a reduction in pathogenicity, equal concentrations of conidia of the deletion strains were inoculated onto detached rice leaves. As reported previously for cbp1 [33] and cda3 [46], and as expected for cda2 and cda2/cda3, pathogenicity was unaffected in these strains (S9 Fig). Surprisingly however, the double and triple deletion strains were also able to successfully infect detached rice leaves or whole plants, causing similar lesion numbers to the WT strain (Fig 7A and S9 Fig). Previously, appressorium development was found to be restored in the cbp1 strain on rice leaves [33]. To see if this was also the case in the other deletion strains, rice leaf sheaths were inoculated with conidia of the cda2/cbp1/cda3 strain, and incubated for 24 hr. This revealed that appressorium development was, indeed, restored upon germination on a leaf surface (Fig 7B). Not only this, but the appressoria of cda2/cbp1/cda3 were able to penetrate rice cells with similar efficiency to the WT: 67% (±7, (SD) n = 3) of WT appressoria had invasive hyphae at 24 hpi, compared with 47% (±10, (SD) n = 3) of cda2/cbp1/cda3 appressoria. To see if this effect was specific to the rice leaf surface, the experiment was repeated on onion epidermis. Again, appressorium development was restored in the cda2/cbp1/cda3 strain and invasive hyphae were observed at 24 hpi (Fig 7B).
The artificial hydrophobic surface lacks the laminated layer of cutin sandwiched between wax on the rice leaf surface cuticular rice leaf [47]. To test whether wax is sufficient to restore appressorium development in the cda2/cbp1/cda3 strain, hydrophobic glass coverslips were coated with the wax molecule 1-octacosanol, which has previously been used to induce appressorium development in M.oryzae [48]. Conidia of strain cda2/cbp1/cda3 germinated on this surface for 24 hr demonstrated partially restored appressorium development (65.6 ± 6.4% on 1-octacosanol compared with 4.6 ± 3.6% on the control surface) (Fig 8). However, germ tubes of the triple deletion strain remained extremely elongated (S10 Fig).
Cutin monomers are known to induce appressorium development in M.oryzae [7], To determine whether cutin restores appressorium development in the cda mutants, conidia of cda2/cbp1/cda3 were germinated on an artificial hydrophobic glass surface, in the presence of 1,16-hexadecanediol (HDD; cutin monomer) either on its own, or in combination with 1-octacosanol. HDD did not induce appressorium development in germlings of the cda2/cbp1/cda3 strain, and no synergistic effect was observed when HDD was used in combination with 1-octacosanol (Student's T-test, no significant difference at p < 0.05, compared with 1-octasosanol treatment alone) (Fig 8 and S10 Fig).
The rescue of appressorium development in the cda2/cbp1/cda3 strain on plant surfaces is perplexing. We hypothesized that chemical factors present on the leaf surface, such as wax, trigger upregulation of additional, functionally redundant chitin deacetylases, and restore chitin deacetylation. To test this, conidia of the cda2/cbp1/cda3 and WT strains were germinated on rice leaf sheaths or onion epidermis', and, subsequently stained with OGA488. Chitosan remained completely undetectable in the triple deletion strain, whilst strong labelling of germ tubes and appressoria was observed in the WT (Fig 9A). This suggests that restoration of appressorium development in cda2/cbp1/cda3 on plant surfaces is not due to the upregulation of redundant chitin deacetylases. Chitosan is thus not an absolute requirement for appressorium development; instead, this requirement appears to be surface dependant.
The appressorial cell wall is impregnated with an impermeable layer of melanin. This may mask detection of chitosan in the appressoria of cda2/cbp1/cda3. Mutant strain germlings were also stained with OGA488 at 4 hpi, prior to melanization of the appressoria. However, no OGA488 staining was detected in unmelanized cda2/cbp1/cda3 germlings, suggesting the melanin does not prevent the detection of chitosan (Fig 9B).
Appressorium development in M.oryzae is controlled by a number of regulatory pathways, which have been shown to respond to the physical and chemical cues present on the leaf surface [49]. In order to determine if there is interplay between chitosan and these particular relays, appressorium development assays were performed in the presence of chemical inducers of these pathways. Previously, appressorium development in cbp1 has been shown to be restored by the addition of cAMP, 1,16 hexadecanediol or diacylglycerol [33]. We tested the cda mutant strains by germinating conidia on a hydrophobic glass surface in the presence of 2.5 mM IBMX (a phosphodiesterase inhibitor which increases intracellular levels of cAMP), 200 mM 1,16 hexadecanediol (HDD, a cutin monomer) or 58 mM diacylglycerol (DAG, an activator of protein kinase C). Consistent with the previous report, all treatments restored appressorium development in cbp1 (Fig 10). Although the proportion of germ tubes forming appressoria was unchanged in this case (since observations were made at 16 hpi), germling morphology of cbp1 was now similar to WT, as evidenced by a significant reduction in germ tube lengths (Mann-Whitney U-test, p < 0.001 S3 Table). cbp1/cda3 showed a similar profile of sensitivity to the chemical inducers as cbp1. Intriguingly however, cda2/cbp1 was considerably less affected by all three chemical inducers: neither HDD or DAG caused a significant increase in appressorium development in this strain (Student's T-test, p < 0.05), whilst IBMX effected only a partial rescue towards the WT phenotype. In cda2/cbp1/cda3, IBMX treatment led to a small proportion of germ tubes forming aberrant appressoria by 16 hpi, although this increase was non-significant (Student's T-test, p < 0.05). Rescue of cbp1 by HDD and DAG is therefore dependent upon CDA2, and rescue by IBMX is dependent upon both CDA2 and CDA3.
Hydrophobicity is one of the key physical cues inducing appressorium development, and is perceived in a cAMP-dependant manner [50]. In order to further examine the relationship between hydrophobic surface sensing, chitosan and the chemical inducers, germling differentiation assays, with the WT and mutant strain cda2/cbp1/cda3, were performed on both hydrophobic and a hydrophilic glass surfaces. At 24 hpi on the hydrophobic surface, the response of cda2/cbp1/cda3 germlings to the inducers was much the same as at 16 hpi, although the proportion of appressoria formed with IBMX was higher (Fig 11A). As with the appressoria formed on the plant leaf surface, it was important to determine whether this rescue could be explained by restored chitin deacetylation. IBMX-treated germlings of the WT and cda2/cbp1/cda3 strains were stained with OGA488 at 24 hpi. This revealed that chitosan was indeed present on the germ tubes and appressoria of the triple deletion strain, albeit to a much lesser extent than the WT germlings (Fig 11B). This suggests that rescue by IBMX may simply be due to upregulation of redundant chitin deacetylases, as observed in cbp1, rather than the activation of an otherwise inactive signalling pathway.
Germinations were repeated on a hydrophilic glass surface (Fig 11C). No appressorium development was observed in the WT strain after a 24 hr incubation (consistent with previous observations [37]), but it was restored by the addition of IBMX or HDD. Both such inducers remained ineffective on cda2/cbp1/cda3. As exogenous chitosan fully restores appressorium development in the CDA deletion strains on a hydrophobic surface we extended this work to evaluate its effect on a hydrophilic surface. WT and cda2/cbp1/cda3 conidia, germinated in the presence of 0.01% (w/v) chitosan for 24 hr did not form appressoria. Next, combined treatments of chitosan and IBMX or with HDD were applied. Interestingly, these combinatorial treatments induced appressorium development in the cda2/cbp1/cda3 mutant. IBMX and HDD added in combination did not, however, induce appressorium development in the triple deletion strain. These experiments reveal two important facts: i) Chitosan acts independently of surface hydrophobicity, and ii) Chitosan relays through a separate pathway from IBMX or HDD.
During the course of these assays it was observed that CDA deletion strain germlings washed off artificial surfaces more readily than the WT strain. The percentage of WT and cda2/cbp1/cda3 germlings adhering to hydrophilic and hydrophobic glass surfaces was quantified at 2 hpi, at which point appressorium development had not yet occurred. Germlings of the triple deletion strain demonstrated significantly lower adhesion than the WT, on both surfaces (Student's T-test, p < 0.05) (Fig 12A & 12B). This loss of adhesion could not be explained by lower conidial germination in the cda2/cbp1/cda3 strain; 95% (±2 (SD), n = 3) of WT conidia had germinated by 2 hpi, compared with 98% (±1 (SD), n = 3) in cda2/cbp1/cda3. Remarkably, however, adhesion was fully restored on both surfaces by germinating the conidia in the presence of 0.01% chitosan (w/v), (Fig 12A & 12B). The adhesion of fungal cells to surfaces such as glass or plastic is mediated by cell wall glycoproteins [51]. These can be detected by staining with the fluorescently-labelled mannose-binding lectin Concanavalin A (FITC-ConA). Staining of WT or cda2/cbp1/cda3 germlings at 2 hpi with FITC-ConA revealed that staining intensity was often reduced in cda2/cbp1/cda3 strain (S11 Fig). Quantification of fluorescence intensity revealed significant (p < 0.01) differences between the WT and cda2/cbp1/cda3 germlings in 3 of the 4 experiments (2-way ANOVA with post-hoc Tukey test) (S11 Fig).
Chitin deacetylases are a conserved family of enzymes in fungi. They catalyze the removal of acetyl groups from chitin, forming chitosan. It has been hypothesized that the deacetylation of chitin may either act as a 'stealth mechanism' to prevent the activation of a PAMP-triggered immune response in the host plant, or result in crucial alterations to the physical and chemical properties of the cell wall necessary for cellular development and morphogenesis. In this study, we present evidence to refute these hypotheses. Instead, we show that chitin deacetylation during germling morphogenesis is not an absolute requirement for either appressorium development or for pathogenicity. Instead, chitin deacetylation plays a role in perception of physical stimuli during the early stages of germling morphogenesis.
Appressorium development is induced by the perception of physical and chemical stimuli present on the plant surface, with signals then relayed via two key regulatory pathways in M. oryzae. The Pmk1 pathway operates at the earliest stage of appressorium development, and is required for sensing of wax, and possibly for surface attachment. The cAMP pathway, is hypothesized to operate at the commitment phase of appressorial development, and is required for sensing hydrophobicity and cutin monomers [49]. Considerable cross-talk exists between these two pathways, but activation of both is required for appressorium development. Many previous studies have been devoted to characterizing the proteins operating in these pathways (reviewed in [49]). Significantly, deletion of the genes encoding such proteins results in phenotypes that are similar to those reported for the cda mutants. The data from our study, set in the context of previous data on signal perception in M. oryzae [49] allows the role played by chitosan to be elucidated.
When germinated on an artificial hydrophobic surface, conidia of the CDA deletion strains cbp1, cda2/cbp1 and cda2/cbp1/cda3 failed to undergo early differentiation events, and appressorium development was either severely delayed (in cbp1) or rarely occurred at all (in cda2/cbp1 and cda2/cbp1/cda3). Failure to differentiate appressoria on artificial surfaces is a phenotype shared by a number of other deletion strains with putative roles in surface sensing. For example, deletion strains of the signalling mucin Msb2, produce highly elongated, undifferentiated germ tubes on an artificial hydrophobic surface [48], but appressorium development is fully restored when inoculated onto a plant surface, just as in the cda strains described herein. In addition, cutin monomers were also unable to rescue msb2, although the effect of cAMP was not determined. Is there a functional link between Msb2 and chitosan? Msb2 has a single transmembrane domain, which anchors the protein to the plasma membrane, whilst the extracellular portion of the protein resides within the cell wall. This extracellular domain alone can partially suppress the defects associated with MSB2 deletion [52]. Chitosan may be required for cell wall-localization of Msb2; the absence of chitosan may result in loss of Msb2 from the cell wall, resulting in the observed phenotype. However, the precise mechanism by which Msb2 mediates surface sensing remains elusive and this limits speculation on the putative functional link between this protein and chitosan. However, in a recent study, combined deletion of CBP1 and MSB2 had additive effects—virulence and Pmk1 phosphorylation (in vegetative hyphae) were shown to be more reduced in msb2/cbp1 strains than in msb2 strains [52]. Loss of chitosan may therefore affect multiple surface-sensing related processes at the earliest stage of germling morphogenesis, which converge on the Pmk1 pathway. In support of this hypothesis, cbp1 germlings have also been shown to be unable to form hyphopodia on artificial hydrophilic surfaces, a phenotype shared by the pmk1 mutant, but not the cpka mutant [34].
Other proteins with putative roles in surface sensing are the hydrophobin Mpg1 [43, 53], and the putative G-protein coupled receptor Pth11 [54]. The mpg1 and pth11 mutants demonstrate similar defects in appressorium development. That is, germlings produce elongate germ tubes on hydrophobic surfaces, and fail to develop appressoria, as in the cda mutants. There are, however, several important differences in the phenotypes: Firstly, appressorium development is also defective on plant surfaces in mpg1 and pth11. Secondly, germ tubes of mpg1 and pth11 do undergo early differentiation events (germ tube hooking). Thirdly, cAMP can restore appressorium development in both mpg1 and pth11 [53, 54], but not in the cda mutants, although this is complicated by the presence of redundant CDAs that appear to be upregulated by cAMP [55, 56]. This suggests that Mpg1 and Pth11 act upstream of the cAMP pathway, whereas chitosan may not. This hypothesis is supported by the fact that chitosan and IBMX/cutin had synergistic effects on appressorium development on a hydrophilic surface. Yet, the fact that germination in the presence of 1-octacosanol, which acts upstream of the Pmk1 pathway (perceived by the ShoI receptor protein) [48], could only partially restore appressorium development in the cda2/cbp1/cda3 strain, may suggest that activation of the cAMP pathway is also defective in this mutant. On the other hand, cutin monomers, which act through the cAMP pathway [54], did not have a synergistic effect with wax, which may suggest the cAMP pathway is already active in the cda2/cbp1/cda3 strain. Partial rescue with 1-octacosanol could instead be explained by the composition of the wax, which may not effectively mimic the mixture present on leaf surfaces [57].
An alternative explanation for the observed phenotypes in the cda mutants, although one that is not mutually exclusive with those proposed above, is that the lack of germling adhesion is responsible. Germlings of cda2/cbp1/cda3 were much more easily detached than those of the WT, even under relatively gentle washing conditions. This suggests that the adhesive properties of the germ tube are defective in the absence of chitosan. In this scenario, germ-tube localized chitosan is required for the adhesion of the germ tube to the surface itself, and is therefore required for surface sensing, since such sensing mechanisms presumably require close contact between the germ tube and the surface in question. Thus, chitosan may act directly as a polysaccharide adhesin, analagous to the role of galactosaminogalactan (GAG) in the cell wall of Aspergillus fumigatus, which mediates adhesion to hydrophobic surfaces and host cells [58, 59]. Alternatively, chitosan may be required for the attachment of various adhesin-like proteins to the cell wall (or a combination of both). Lack of chitosan in cda2/cbp1/cda3 did result in a reduction of Concanavalin A staining, suggesting a reduction of mannan in the cell wall, but there are several possible explanations for this: i) Chitosan is directly required for attachment of mannoproteins to the cell wall, ii) Loss of chitosan results in large-scale changes in cell wall composition, affecting attachment of mannoproteins or resulting in a reduction in non-protein associated mannan, iii) Mannoproteins are produced, but secretion to the outer wall is affected by loss of chitosan or iv) Lack of surface sensing and activation of Pmk1 and/or cAMP pathways means that production of the mannoproteins does not occur in the first place, i.e. reduced germling adhesion is a consequence, and not the cause of defective surface sensing. Further investigations are required to distinguish between these possibilities.
The rescue of appressorium development in the cda mutants by exogenous chitosan and its association with the cell wall was an unexpected result, and it is unclear exactly how this occurs. Because the charge of the exogenous chitosan was of great importance, this may indicate that an electrostatic interaction is responsible for its association with the cell wall. This, in turn, suggests the existence of an anionic cell wall component, which may form a complex with the cationic chitosan. At present, little is known of anionic cell wall components in fungi. Uronic acids are known to be present in the cell walls of some fungi [60–63], as is phosphomannan [64–66], but their presence in M.oryzae has not been investigated. In addition to an electrostatic association, the possibility of enzymatic incorporation of exogenous chitosan must also be considered. CRH transglycosylases have previously been shown to be able to incorporate exogenous fluorescently-labelled oligosaccharides in S.cerevisiae [67]. It is therefore possible that other enzymes with similar activities may be responsible for the hypothesized incorporation of exogenous chitosan.
The role of chitosan in appressorium development is an intriguing one, and there remain several unresolved issues, as discussed above. Despite this, the model presented in Fig 13 is consistent with all of the data presented herein, and with previous studies characterizing surface sensing proteins in M.oryzae. In this model, germ-tube localized chitosan is required for the activation of the Pmk1 MAP kinase pathway in response to a surface, but independently of surface hydrophobicity. Optimal activation of the cAMP pathway may also require chitosan, but it is not possible to conclusively determine whether just one or both of the signalling pathways are defective in cda mutants from the current data.
The deacetylation of chitin was hypothesized to cause profound changes to the chemical and physical properties of the cell wall, which could be crucial for cellular morphogenesis. However, no evidence for this has been found in the present study. Germlings of the cda2/cbp1/cda3 strain developed appressoria that were morphologically indistinguishable from the WT strain on plant surfaces, despite the absence of detectable chitosan. Additionally, appressorium development in cda2/cbp1/cda3 was restored by a short incubation with FITC-chitosan that was only incorporated into the germ tube. Not only was chitosan not required for morphogenesis of appressoria, but appressorium function also seemed to be unaffected by the absence of chitosan, since plant penetration continued to be achieved successfully. Chitosan is therefore also unlikely to be a key structural component of the cell wall, in contrast to Cryptococcus neoformans where loss of chitosan impaired cellular integrity [31]. On the other hand, it is also possible that compensatory changes in cell wall composition occurred in the absence of chitosan, which were not investigated. The upregulation of other cell wall components may be sufficient to maintain cell wall integrity in the cda mutants. In a similar vein, it is not inconceivable that compensatory changes were also responsible for allowing appressorium morphogenesis to occur in the absence of chitosan i.e. there are multiple, redundant morphogenic mechanisms operating during appressorium development.
A second hypothesized role of chitosan is in the protection from plant chitinases. However, the findings presented here, together with evidence from previous studies do not support this role for chitosan, at least in germlings. Germlings of cda2/cbp1/cda3 in which chitosan was absent remained intact on leaf surfaces, in contrast to those of the ags mutant which lack α1,3-glucan and were destroyed, presumably by degradative enzymes on the leaf surface [18]. This suggests that α1,3-glucan may be the cell wall component with the protective role in M.oryzae germlings. It would also be valuable to investigate the hypothesized protective effects of chitin deacetylation during in planta growth. Chitosan staining was not performed on invasive hyphae in the cda mutants, although qRT-PCR analysis suggested that the chitin deacetylases operating during appressorium development are not highly expressed during invasive growth (S2 Fig). Instead, CDA1 and CDA6 appear to be involved in chitin deacetylation at this stage of infection. Further investigations should therefore focus on the characterization of these genes to determine the role of chitosan in invasive hyphae.
The deacetylation of chitin is hypothesized to require a degree of collaboration between the membrane-localized chitin synthases, and cell wall or membrane-associated chitin deacetylases [31, 68]. Evidence from this study may lend further support to this hypothesis. Deletion of CHS7 results in an almost identical phenotype to multiple CDA deletion [69], suggesting that Cbp1, Cda2 and Cda3 may deacetylate the chitin synthesized by Chs7. Further investigation is required to determine the relationship between chitin synthesis and deacetylation during appressorium development in M.oryzae.
In summary, the investigation of chitin deacetylation during appressorium development in M.oryzae has yielded unexpected and intriguing data. The deacetylation of chitin by at least three chitin deacetylases, with overlapping roles, is required for surface sensing in germlings. Yet, this requirement is surface dependant, due to the multiple, redundant mechanisms by which appressorium development can be induced in M.oryzae. Nevertheless, this study provides a novel insight into the mechanisms behind the perception of physical stimuli in M.oryzae, and also demonstrates a novel way in which the cell wall is crucial in acting as an interface between fungal cells and their environment. Importantly, evidence in support of the long-standing hypothesis regarding the role of chitin deacetylation was not found; chitosan is not required as a 'stealth' mechanism in germlings of M.oryzae, and so the role of this cell wall component in the interactions between phytopathogenic fungi and their hosts may need to be reconsidered. However, further work is required to determine whether or not chitosan acts as a 'stealth mechanism' during in planta growth of M.oryzae, as this was not determined in this study.
The wild-type (WT) rice pathogenic Magnaporthe oryzae strain Guy11 and mutant strains were cultured at 24°C with a 14 h light 10 h dark cycle. Strain maintenance and composition of media were essentially as described by Talbot et al [43].
RNA was extracted from rice leaves inoculated with conidia of the Guy11 strain using the Qiagen RNeasy RNA extraction kit, according to manufacturer's instructions. RNA concentration was determine using a ThermoScientific ND-1000 NanoDrop spectrophotometer, integrity was evaluated by gel electrophoresis. Genomic DNA was removed by using an RNase free DNase set (Qiagen), according to manufacturers’ instructions. Reverse transcription of 1 μg of RNA into cDNA was performed by using the Maxima First Strand cDNA synthesis kit (ThermoFisher Scientific), according to manufacturers instructions, using random hexamer primers. qRT-PCR was performed with Power SYBR Green PCR master mix (ThermoFisher Scientific), on an Applied Biosystems 7300 Real-time PCR system. Primers are listed in S2 Table (37–58). Primers were designed to span an intron where possible (not all CDA genes have introns), and efficiency of the primers was 85–104% (average efficiency 93.6%). No amplification was observed in samples that were not reverse transcribed (-RT control), in samples without RNA (NTC control), or in RNA extracted from mock-inoculated leaves. Relative transcript abundance was calculated by the efficiency correction method [70] as follows: Abundance = Etarget(Ct(reference)-Ct(target)).
Single CDA deletion strains were generated by replacing the coding sequences of CBP1 (MGG_12939), CDA2 (MGG_09159) and CDA3 (MGG_04172) with a hygromycin resistance cassette [71]. Briefly, sequences flanking the target genes (~1.5kb upstream, and ~1.1kb downstream) were amplified by PCR, using primers 7–10 (for CBP1), 13–16 (for CDA2) and 17–20 (For CDA3) (primers are listed in S2 Table). These fragments were joined to the hygromycin resistance gene by overlapping PCR, using primer pairs 7/4 and 3/10 (for the CBP1 deletion construct), 13/4 and 3/16 (for the CDA2 deletion construct), 17/4 and 3/20 (for the CDA3 deletion construct). The final, complete construct was made by overlapping PCR, amplifying the products from the previous reactions using primer pairs 7/10 (for CBP1), 13/16 (for CDA2) and 17/20 (for CDA3). The final PCR product was used directly for DNA-mediated protoplast transformation of WT Guy11 strain following protocols described by Talbot et al (39). Putative transformants were selected on minimal medium (MM) supplemented with 300 μg ml-1 hygromycin B (Calbiochem, Merck, Darmstadt, Germany). Deletion of the target gene was confirmed by both PCR and Southern Blot analysis, as described in Samalova et al [72].
Double CDA deletion strains were generated in the cda2 or cbp1 background strains. The coding sequences of CBP1 and CDA3 were replaced with a bialaphos resistance cassette. The deletion construct was made as described above, except primers 8 & 9 were substituted for 11 & 12 (for CBP1), and primers 18 & 19 for 21 & 22 (for CDA3). The CBP1 deletion cassette was transformed directly into protoplasts of the cda2 strain, to generate the cda2/cbp1 mutant. The CDA3 deletion cassette was transformed into both the cda2 and cbp1 background strains, to generate the cda2/cda3 and cbp1/cda3 mutants. Putative transformants were selected on defined complex medium (DCM) supplemented with 60 μg ml-1 Bialophos (Goldbio, St Louis, MO, USA). Deletion strains were confirmed as above.
To generate the triple cda2/cbp1/cda3 mutant, CDA3 was deleted in the cda2/cbp1 mutant. In this case, the coding sequence of CDA3 was replaced by a sulphonylurea resistant allele of the M.oryzae ILV gene (MGG_06868). Since the sulphonylurea transformation constructs were too large to be generated by over-lapping PCR, GAP-repair S. cerevisiae cloning [73] was used to assemble the constructs in pNEB1284 vector. Primers 5,6 and 23–26 were used to amplify the required DNA fragments. Putative transformants were selected on BDCM medium supplemented with 100 μg ml-1 chlorimuron ethyl (Sigma Aldrich, UK) and confirmed as specified above.
Standard molecular techniques [74] were used to prepare the complementation constructs with fluorescently tagged CDAs. A set of transformation vectors based on pUCAP was generated as described in Samalova et al. [75]. The vectors contain polyadenylation signal pATrpC and either bialophos or hygromycin resistance marker that was cloned into re-created SalI sites using primer pairs 1/2 or 3/4 respectively (see S2 Table and S12 Fig). For PCR amplification of CBP1 and CDA2, primer pairs 27/28 and 29/30 were used, respectively. Genomic DNA from the WT strain Guy11 was used as a template, and amplified using Herculase DNA polymerase (Agilent). This resulted in amplification of the coding sequence of the genes (without stop codons), together with 2 kb of native promoter sequence for CBP1 and 1.3 kb for CDA2. The PCR products were cloned into the AscI sites of the vector described above (S12 Fig), creating C-terminal mCherry fusions.
Conidia (2.5 x 105 ml -1) of Guy11 and complemented strains were collected from 10 day old plates and inoculated in 50 μl droplets onto hydrophobic glass cover-slip,; onion peels, or rice leaf sheaths; as described in Samalova et. al., [75] and incubated for specified times in the growth chamber. For viewing mCherry fluorescence, the samples were viewed using the C-Apochromat 40x/1.2 water corrected objective lens of a Zeiss LSM 510 Meta confocal microscope at 543 nm excitation from the HeNe laser and emission collected with BP565-615 filter for mCherry.
Calcofluor White staining was performed as follows: Conidia (2.5 x 105 ml -1) of the Guy11 and mutant strains were inoculated in 50 μl droplets onto hydrophobic glass cover-slips, and incubated for specified times in the growth chamber. After the incubation, the water droplet was removed from the cover-slip, and replaced with 100 μl of 0.05% Calcofluor White solution (Sigma-Aldrich, UK), and incubated for 20 min. Samples were then washed briefly with dH2O, and viewed using the Zeiss LSM510 microscope as above, with 405 nm excitation and emission collected with an LP420 filter.
OGA488 staining was performed essentially as described in [36]. Briefly, Conidia (2.5 x 105 ml-1) of the Guy11 and mutant strains were inoculated in 50 μl droplets onto hydrophobic glass cover-slips, onion epidermis, or rice leaf sheaths and incubated for specified times in the growth chamber. Samples were washed briefly with 25 mM MES (pH 5.6), and incubated with OGA488 (a generous gift from William Willats [36]) (diluted 1/1000 in 25 mM MES) for 15 min on ice. This was followed by 2–3 brief washes with 25 mM MES, after which the samples were viewed using the C-Apochromat 40x/1.2 water corrected objective lens of a Zeiss LSM 510 Meta confocal microscope, at 488 nm excitation and emission collected with an LP505 filter.
Staining of germlings with FITC-labelled Concanavalin A (ConA-FITC) (Sigma-Aldrich, UK) was performed as follows: Conidia (2.5 x 105 ml-1) of the Guy11 and mutant strains were inoculated in 50 μl droplets onto hydrophobic glass cover-slips, and incubated in the growth chamber for 2 hr. Cover-slips were washed briefly with PBS, then incubated with 40 μg/ml ConA-FITC for 20 min on ice, then washed briefly with PBS 2–3 times. Samples were viewed using the C-Apochromat 40x/1.2 water corrected objective lens of a Zeiss LSM 510 Meta confocal microscope, at 488 nm excitation and emission collected with an LP505 filter. Fluorescence intensity was quantified using ImageJ.
Staining with the monoclonal anti-chitosan antibody mAbG7 (a generous gift from Stefan Schillberg [35]) was performed as follows. Conidia (2.5 x 105 ml-1) of the Guy11 strain were inoculated in 50 μl droplets onto hydrophobic glass cover-slips, and incubated in the growth chamber for the specified time. Samples were first blocked by incubation with 2% BSA (w/v) in PBS for 1 hr at room temperature. Samples were washed 3 times with PBS/T (PBS +0.05% Tween 20), for 5 min each on an orbital shaker (~70 rpm) and then incubated with the primary antibody (mAbG7, at 10 μg/ml in PBS) for 1.5 hr at room temperature. This incubation was followed by 3 more washing steps as described above, and incubation with the secondary antibody (FITC-labelled anti Mouse IgM, 5 μg/ml in PBS) at room temperature for a further 1.5 hr. Finally, the secondary antibody was removed by 3 more washing steps with PBS/T as described previously and viewed under using the Zeiss LSM510 microscope, as described above for ConA-FITC staining. A negative control was included in all experiments, in which samples were only incubated with the secondary antibody.
Staining with the polyclonal anti-chitosan antibody (a generous gift from Holger Deising [19]) was performed as for the mAbG7, except that the antibody was used at a dilution of 1/100 from the original antiserum, and the secondary antibody (a FITC-labelled anti-rabbit IgG) was used at 10 μg/ml (in PBS).
Conidial germination and appressorium development were assessed at 1, 8, 16 or 24 hpi by following germling differentiation on hydrophobic glass cover-slips (Gerhard Menzel, Glasbearbeitungswerk GmbH & Co., Braunschweig, Germany). Conidia (2.5 x 105 ml-1) of the Guy11 and mutant strains were inoculated in 50 μl droplets onto hydrophobic glass cover-slips, and incubated in the growth chamber for the specified time. Samples were viewed under an Olympus BX50 microscope, and ~500 germlings in 3 independent experiments counted per strain/timepoint. For germinations in the presence of chemical inducers of appressorium development, IBMX (Sigma-Aldrich, UK) was used at 2.5 mM (from a 250 mM stock in DMSO), 1,16 hexadecanediol (Sigma-Aldrich, UK) at 200 μM (from a 50 mM stock in ethanol) and diacylglycerol (1,2-dioctanoyl-sn-glycerol) (Enzo Life Sciences) at 58 μM (from a 7.25 mM stock in DMSO). Chitosan and its derivatives were used at a final concentration of 0.01% or 0.001%, from a 1% stock (except for FITC-chitosan which was from a 0.08% stock). For germinations in the presence of wax (1-octacosanol) (Sigma-Aldrich, UK), the wax was first dissolved in chloroform to a concentration of 4 mg/ml. In a laboratory fume hood, 100 μl of this stock was pipetted onto a hydrophobic glass coverslip and the chloroform allowed to evaporate, leaving a layer of wax on the coverslip. Conidia were then inoculated onto this surface, as described above. For germinations on hydrophilic glass coverslips (Heathrow Scientific) the protocol was identical except that only 20 μl of conidial suspension was used. Coverslips were placed in square Petri dishes with damp filter paper and sealed with Parafilm to prevent evaporation.
Cuticle penetration was assessed by scoring the frequency with which appressoria formed penetration pegs and intracellular infection hyphae on rice leaf sheaths, after incubation at 24°C for 24 h.
Leaf infection assays were performed on blast-susceptible, 21-day-old seedlings of rice (Oryza sativa L.) cultivar CO39. Assays on detached leaves were performed as described in [72]. For assays on whole plants, 21-day-old seedlings of rice (Oryza sativa L.) cultivar CO39 were spray inoculated with 4 ml of conidial suspension at three different concentrations (1.25 x 105, 6.25 x 104 or 3.13 x 104 conidia/ml, in 0.2% (w/v) gelatine water). A mock inoculation of 0.2% (w/v) gelatine water was included as a negative control. Infection was assessed 4 days later.
Adhesion assays were performed on both hydrophobic and hydrophilic glass coverslips. 20 μl droplets of conidial suspension (either with or without 0.01% chitosan) of each strain were pipetted onto the coverslips, which were placed into a humidity chamber and incubated at 24°C for 2 hr to allow germination. At 2 hpi, half of the coverslips were removed and placed into a 50 mm Petri dish containing 5 ml dH2O, and shaken on an orbital shaker at 100 rpm for 5 min. Coverslips were then removed and mounted on slides for viewing under the 10X objective lens of an Olympus BX50 microscope. Conidia were counted from one field of view (for hydrophobic coverslips) or three fields of view (for hydrophilic coverslips), with a conscious effort made to locate the area of the coverslip with the highest conidial density. For each strain, the number of conidia counted on the washed coverslips was compared with those counted on the unwashed coverslips, and percentage conidial adhesion calculated as: 100—((conidia on unwashed coverslip—conidia on washed coverslip) /conidia on unwashed coverslip) x 100.
FITC-labelled chitosan was synthesized exactly as described previously [45].
|
10.1371/journal.pntd.0002815 | Trends in the Epidemiology of Pandemic and Non-pandemic Strains of Vibrio parahaemolyticus Isolated from Diarrheal Patients in Kolkata, India | A total of 178 strains of V. parahaemolyticus isolated from 13,607 acute diarrheal patients admitted in the Infectious Diseases Hospital, Kolkata has been examined for serovar prevalence, antimicrobial susceptibility and genetic traits with reference to virulence, and clonal lineages. Clinical symptoms and stool characteristics of V. parahaemolyticus infected patients were analyzed for their specific traits. The frequency of pandemic strains was 68%, as confirmed by group-specific PCR (GS-PCR). However, the prevalence of non-pandemic strains was comparatively low (32%). Serovars O3:K6 (19.7%), O1:K25 (18.5%), O1:KUT (11.2%) were more commonly found and other serovars such as O3:KUT (6.7%), O4:K8 (6.7%), and O2:K3 (4.5%) were newly detected in this region. The virulence gene tdh was most frequently detected in GS-PCR positive strains. There was no association between strain features and stool characteristics or clinical outcomes with reference to serovar, pandemic/non-pandemic or virulence profiles. Ampicillin and streptomycin resistance was constant throughout the study period and the MIC of ampicillin among selected strains ranged from 24 to >256 µg/ml. Susceptibility of these strains to ampicillin increased several fold in the presence of carbonyl cyanide-m-chlorophenyldrazone. The newly reported ESBL encoding gene from VPA0477 was found in all the strains, including the susceptible ones for ampicillin. However, none of the strains exhibited the β-lactamase as a phenotypic marker. In the analysis of pulsed-field gel electrophoresis (PFGE), the pandemic strains formed two different clades, with one containing the newly emerged pandemic strains in this region.
| Vibrio parahaemolyticus has been associated with several epidemics of foodborne diarrheal infection. Recent observations in several counties have shown the emergence of pandemic strains of V. parahaemolyticus with unique genetic features and their role in diarrheal outbreaks. Unlike other enteric pathogens, the appearance of pandemic strains of V. parahaemolyticus has not been associated with the economic/hygiene status of the population. The pandemic strains of V. parahaemolyticus continue to prevail in Kolkata, India since its appearance during 1996. The present communication describes not only the prevalence of pandemic serovars of V. parahaemolyticus, but also the appearance of novel serovars under the pandemic strain category. In addition, the trh gene was detected in some of the pandemic strains for the first time. In the newly emerged serovars genetic changes have occurred, as evidenced from the PFGE analysis. Overall, the antimicrobial susceptibility of pandemic strains remains unchanged for the past 20 years. The observations made in this study re-emphasize the importance of this pathogen and shows the recent genetic and serovar changes in the epidemiology of V. parahaemolyticus-mediated diarrhea.
| Vibrio parahaemolyticus is a Gram-negative bacterium, which is normally found in several niches of the coastal environments. In humans, this pathogen causes three major clinical syndromes: gastroenteritis, wound infections and septicemia [1]. Intestinal infections caused by this pathogen are mainly associated with the consumption of raw or undercooked seafood with clinical symptoms such as moderate to severe diarrhea, abdominal cramps, nausea, vomiting, with or without fever and tenesmus [1]. In infected individuals, the frequency of diarrhea may vary from 3 to 10 times per day and in the case of persistent diarrhea; the duration may last for 4–7 days. V. parahaemolyticus infection has been reported all over the world, either as sporadic diarrhea or contaminated food-related outbreaks [2], [3]. Generally, the isolation rate of this pathogen from diarrheal cases has been high in Asian countries [4]–[6]. A recent surveillance conducted during 1996–2010 in the US revealed an increase in the infection rate of V. parahaemolyticus [7].
To confirm their role in the diarrheal epidemiology, V. parahaemolyticus isolated from clinical, food and environmental sources are further tested for virulence and other genetic characteristics. The virulence of this pathogen has been attributed to the production two major factors: thermo-stable direct hemolysin (TDH) encoded by the tdh, and TDH-related hemolysin encoded by trh. Either or both of these genes have been commonly detected in clinical strains, but not always from food/environmental strains [8]. The emergence of the first pandemic strain of V. parahaemolyticus belonging to serovar O3:K6 has been reported from Kolkata during 1996 [9]. Since then, this pathogen has been associated with several large outbreaks of diarrhea in many countries [10].
In addition to virulence characteristics, V. parahaemolyticus strains have been tested for the prevalence of different serovars and pandemic marker genes encoded in the ToxRS region by using a group specific PCR (GS-PCR) [11]. This GS-PCR was developed based on the nucleotide sequence variations in the toxRS operon, which encode transmembrane proteins involved in the regulation of virulence-associated genes. This specific variation was found only in the pandemic strains of V. parahaemolyticus and hence used as a genetic marker for its detection. The toxRS gene sequence in the new pandemic strains has difference at 7 base positions compared with non-pandemic strains, of which 2 bases have been used to design primers in the GS-PCR. In an active surveillance of diarrheal infection, we monitor several enteric pathogens among acute diarrheal patients admitted at the Infectious Diseases Hospital (IDH), Kolkata, India. Since multiple antimicrobial resistances have been reported in other enteric pathogens [12]–[15], we examine the susceptibility patterns of V. parahaemolyticus strains. In this study, V. parahaemolyticus strains isolated during 2001–2012 from the hospitalized acute diarrheal patients were examined for serovar prevalence, virulence traits, antimicrobial resistance and genetic lineage of strains, along with the association of clinical symptoms of the cases.
Ethical approval has been obtained from the National Institute of Cholera and Enteric Diseases Ethics Committee (Ref.C-4/2012-T&E), and the enrolled patients/parent in the case of children in this study provided written informed consent.
Between January 2001 and December in 2012, every fifth diarrheal patient admitted at the IDH was enrolled in the active surveillance. During enrollment, patients or primary caretakers of children undertook a standardized questionnaire to solicit demographic, epidemiologic, and clinical information. Stool specimens were collected before the administration of antibiotics using sterile catheters and transported to the laboratory with 2 hrs. In the event of any anticipated delay, soaked swabs in stool specimens were stored in Carry Blair transportation medium (Difco, BD, Sparks, MD) at ambient temperature for 6–8 hrs.
FLC and RBC have been examined microscopically (Olympus CX41, Olympus Corporation, Tokyo, Japan) by smearing a thin layer of fresh stool on a glass slide and counts were made the under high power in five or more fields. Microscopic presence of RBC was further confirmed by Hemaoccult 11 (Smith Kline Diagnostics, San Jose, CA). The stool pH was determined using a portable pH meter (Jenway, Staffordshire, UK).
Stool specimens/swabs were processed for the detection of V. parahaemolyticus after enrichment in alkaline peptone water (Difco) with 1% NaCl and pH 8.5. After 4–6 hrs of incubation at 37°C, a loop full of culture was plated onto thiosulphate citrate bile salts sucrose agar (TCBS, Eiken, Tokyo, Japan), followed by incubation at 37°C overnight. Typical green colonies grown on the TCBS agar have been tested in triple-sugar iron agar, production of cytochrome oxidase, and tolerance to NaCl at various concentrations [16].
Somatic (O) and capsular antigen (K) of V. parahaemolyticus were detected using commercially available kits (Denka Seiken, Tokyo, Japan) that contained 9 pooled polyvalent K group antisera (KI to KIX), 65 monovalent K type antisera (K1 to K71; K2, K14, K16, K27, K35, K62 are not included), and 11 O group antisera (O1 to O11). Freshly grown cultures on nutrient agar (Difco) supplemented with 1% NaCl and heat-killed cells suspended in normal saline were used for K and O serotyping, respectively.
V. parahaemolyticus strains were tested for virulence traits such as tdh, trh genes and pandemic group specific (GS) toxRS gene using PCR assays as described previously [11], [17], [18].
Antimicrobial susceptibility test was performed by disc diffusion method in accordance with Clinical and Laboratory Standards Institute guidelines [19] using commercially available ampicillin (AM) (10 µg), azithromycin (AZM) (15 µg), ceftriaxone (CRO) (30 µg), chloramphenicol (C) (30 µg), ciprofloxacin (CIP) (5 µg), nalidixic acid (NA) (30 µg), norfloxacin (NOR) (10 µg), ofloxacin (OFX) (5 µg), streptomycin (S) (10 µg), tetracycline (TE) (30 µg), trimethoprim/sulfamethoxazole (SXT) (25 µg), discs (BD, Sparks, MD) in Mueller Hinton agar (MHA) (Difco). These antimicrobials are generally used in the empirical treatment of acute diarrheal cases and hence included in the susceptibility testing. MICs of ampicillin streptomycin and nalidixic acid have been determined by using an E-test (AB bioMèrieux, Solna, Sweden), following the manufacturer's instructions. Escherichia coli strain ATCC 25922 was used as the quality control strain for each batch of the assay. Since there is no published interpretive breakpoint to categorize susceptible/resistant V. parahaemolyticus strains, we have followed the interpretive breakpoint of E. coli strain ATCC 25922 in this study.
Simplex PCR assays were used to detect antibiotic resistance genes such as strA, aadA1 (encoding aminoglycoside [3′] adenylyltransferases), blaSHV, blaOXA and blaTEM (encoding β-lactamases) as described before [14], [20]. New primers (VP-bla F-CCTGTTGGTTGGGCTGATGGTT and VP-bla R-GAAGCGAAAGGTCTGTGT CTGTGA) were designed to detect chromosomally encoded V. parahaemolyticus beta-lactamase gene (VPA0477) and a qnr homologue VPA0095 (QnrVPF- CGAATATCCAGCCCGTCCAGTT and QnrVPR- AATCCAAAGCGCTAGAAGGGTGTA) using a DNA gene sequence of V. parahaemolyticus RIMD 2210633 (accession No. BA000032) with the DNAStar software (Madison, WI). Template DNA was prepared by boiling the cultures grown in Luria Bertani (LB, Miller) broth (Difco) for 10 min, rapidly cooled on ice followed by brief centrifugation at 10,000 rpm and the supernatant was used in the PCR.
Synergy testing was performed using MHA supplemented with or without the efflux pump inhibitor carbonyl cyanide-m-chlorophenyldrazone (CCCP, 1.5 µM) and ampicillin E-test strips [21].
General log-linear model (GLM) has been used to analyze the association of clinical parameters and stool characteristics with V. parahaemolyticus infection. In this analysis, all the variables were treated equally as “response” variables whose mutual association was explored. Using Newton-Raphson with Poisson method, the maximum likelihood parameter estimation model was obtained using SPSS version 19 software [SPSS, Inc., Chicago, IL]. In this analysis, age was grouped in four categories: 1 = up to 10 years, 2 = >10–20 years, 3 = >20–40 years and 4 = >40–≥60 years. The nature of diarrhea was categorized in three groups: 1 = watery, 2 = loose stool and 3 = bloody and mucoid stool. The duration of diarrhea was classified in two groups: 1 = up to 24 hrs and 2 = >24 hrs. Frequency of stool per day was considered in three groups: 1 = up to 5 times, 2 = >5–10 times and 3 = >10 times. Abdominal pain and vomiting were categorized in two groups, each with 1 = present and 2 = absent. Stool characteristics such as the stool consistency, pH, number of RBC, and number of pus cells were made in three categories, each with: 1 = liquid, 2 = mushy and 3 = formed; 1 = <7, 2 = ≥7–8 and 3 = >8; 1 = 1–10, 2 = 11–20 and 3 = absent; 1 = 1–10, 2 = 11–20 and 3 = absent, respectively. The categorical data can highlight the interrelationship in a log linear analysis.
PFGE has been made following the PulseNet International protocol [22]. About 40 V. parahaemolyticus pandemic strains belonging to diverse serovars have been selected in the PFGE, which includes all the newly identified pandemic serovars (n = 11), representative pandemic serovars (n = 26), along with 3 pandemic O3:K6 strains isolated before 2001 in Kolkata. Briefly, the chromosomal DNA of each strain was digested with NotI enzyme (Fermentas, Germany) at 37°C overnight. The XbaI (Fermentas) digested DNA of Salmonella Braenderup strain H9812 was used as a molecular weight marker. The restriction fragments were resolved in a CHEF Mapper system (Bio-Rad, Hercules, CA). The PFGE patterns were analyzed using the BioNumerics version 4.0 software (Applied Maths, Sint Martens Latem, Belgium) after normalization of the TIFF images with the size standard of strain H9812. Clustering was performed using the unweighted pair group method (UPGMA) and the Dice correlation coefficient with a position tolerance of 1.0%. The PFGE profiles of three O3:K6 pandemic strains isolated before 2001 (VP101, VP174 and VP232 isolated during 1996, 1997 and 1998, respectively) were included in the clonal comparison.
In a span of 12 years from 2001 to 2012, 178 (1.3%) V. parahaemolyticus strains were isolated from 13,607 diarrheal patients. The prevalence of V. parahaemolyticus was maximum in 2009 (Fig. 1). Although the isolation rate was low, diverse serovars were identified in this study (Table 1). Overall, the serovars O3:K6 (19.6%), O1:K25 (18.5%), O1: KUT (K-untypable, 11.2%), O3:KUT (6.7%), O4:K8 (6.7%), and O2:K3 (4.5%) were comparatively higher than the others.
In the GS-PCR, pandemic strains of V. parahaemolyticus were detected (68%) more than non-pandemic counterparts (32%). Among the pandemic strain category, serovars O3:K6 (91.4%; 32/35), O3:KUT (100%; 12/12), O1:KUT (80%; 16/20), O1:K25 (100%; 33/33) and O1:K36 (100%; 11/11) were predominantly detected. Though less in numbers, the other new serovars such as O2:K4, O4:KUT, O4:K4, O4:K13, O8:K21, and O10:K60 were identified as pandemic strains in the GS-PCR (Table 1). Based on the virulence gene PCR assay results, V. parahaemolyticus strains were categorized in four groups: tdh+trh+, tdh+trh−, tdh−trh+, and tdh−trh−. The most predominant virulence gene profile was tdh+trh− (94.9%, 169/178). V. parahaemolyticus strains with other gene profiles remained were: tdh−trh− (2.2%, 4/178), tdh−trh+ (1.7%, 3/178) and tdh+trh+ (1.1%, 2/178). When correlating virulence gene profiles with GS-PCR results, 97.5% (118/121) of the strains harbored only the tdh gene. However, 3 trh positive strains (2.5%, 3/121) were identified as pandemic strains in the GS-PCR. Of these, two trh positive pandemic strains belonged to O1:KUT and the other was identified as O3:KUT. Among the non-pandemic serovars, the tdh+trh− (89.5%, 51/57) profile was predominantly detected. However, 4 (7%) non-pandemic strains did not harbor any of these virulence markers, and 2 (3.5%) had the tdh+trh+ profile.
Ninety-eight percent (174/178) of the strains were resistant to ampicillin, 86% to streptomycin, 3.4% to nalidixic acid, and 1.7% to chloramphenicol. One non-pandemic strain with an unknown serovar (OUT:KUT) was resistant to trimethoprim-sulfamethoxazole, tetracycline, chloramphenicol, nalidixic acid ampicillin and streptomycin. Three strains were found to be susceptible to all the antimicrobials. Ampicillin resistance was common among pandemic and non-pandemic strains. The MIC of ampicillin against 10 randomly selected strains ranged from 24 to >256 µl/ml and 6 to 12 µl/ml for streptomycin. All the strains remained negative for β-lactamase-production.
All the strains were screened for strA, aadA1 and blaTEM genes that encode resistance to aminoglycosides and extended-spectrum β-lactamase (ESBL), respectively. Only two strains harbored strA, and one harbored with aadA1. All the strains were negative for blaTEM, blaSHV and blaOXA genes. However, the newly reported ESBL encoding open reading frame (ORF) VPA0477 was found in all the strains, including the strains susceptible to ampicillin. Except for two, the chromosomally encoded qnr homologue was detected in all the strains, irrespective of the quinolone resistant/susceptible phenotype. The qnr homologue negative nalidixic acid susceptible strains had 1–3 folds lower MIC values compared to the strains harboring this gene.
Synergy test results showed that the MIC of ampicillin was 1.5 to 16-folds less in the selected V. parahaemolyticus strains with CCCP as compared to the growth in the inhibitor-free medium (Table 2).
The GLM showed a significant association between V. parahaemolyticus infection and some of the stool characteristics and clinical symptoms. Liquid and mushy stool consistency, presence of mucus, alkaline stool (pH 8.0), presence of RBC up to 10 and ≥20 FCL counts were significantly associated with the V. parahaemolyticus infection (p<0.001) (Table 3). In the older than 30 years age group, short duration of diarrhea (≤24 hrs), frequency of stool more than 5 times/day, the presence of abdominal pain, and high frequency of vomiting were significantly associated with the V. parahaemolyticus infection (p<0.001) (Table 4). It is worth to mentioning that in the majority (78.1%; 139/178) of V. parahaemolyticus positive cases, this organism was detected as a sole pathogen and in the rest (21.9%; 39/178) as a mixed infection (data not shown). The other pathogens identified in 39 mixed infection cases included V. cholerae, V. fluvialis, Salmonella spp., Shigella spp., diarrhegenic E. coli, (ETEC, EPEC, EAEC), Campylobacter spp., Aeromonas spp., Rota virus, Adeno virus, Naro virus, Sappo virus, Giardia spp., Entamoeba histolytica, and Cryptosporidium spp.
Cluster analysis based on the NotI-PFGE profiles revealed two distinct clades (A and B) in the dendrogram (Fig. 2). Clade A had 26 V. parahaemolyticus pandemic strains, of which 46% (12/26) of the strains belonged to O3:K6, 27% (7/260) to O1:K25, 11% to O4:K68 (3/26) and 8% to O1:KUT (2/26). All these serovars have been previously reported and had an overall similarity of more than 75%, which includes three O3:K6 strains isolated during 1996–1998. In clade B, the serovar O10:K60 isolated between 2011 and 2012 was more frequent compared to others (57%, 4/7). One unusual O3:K6 serovar was also identified in this clade. From the dendrogram, it appears that the newly emerged pandemic servoars of V. parahaemolyticus are heterogeneous with about 50% genetic similarity with serovars placed in clade A (Fig. 2).
Previous studies conducted in Kolkata showed an abrupt appearance of pandemic O3:K6 serovar in 1996 with additional pandemic serovars such as O1:K25, O1:KUT and O4:K68 in subsequent years [9], [23]. Almost during the same period, a similar trend was reported from Thailand and Japan [24], [25]. Spread of pandemic strains of V. parahaemolyticus has been reported in several countries, either as a sporadic occurrence or associated with large foodborne outbreaks [10]. In this study, the isolation rate of V. parahaemolyticus during 2001–2012 ranged from 0.5% to 4%. The overall isolation rate was 1.3%, which closely matches a report from Bangladesh [26]. In 2009, an increased isolation rate (4.2%) of V. parahaemolyticus was detected compared to other years. The rise in the prevalence of V. parahaemolyticus during this period was not associated with any local outbreak. In 2009, O1:K36, O1:K25 and O3:K6 serovars were predominantly identified.
Overall, pandemic O3:K6 was isolated throughout the study period. Conversely, only three strains of O4:K68 serovar were identified, which was the second most dominant serovar during 1997–2000 in Kolkata. During 2001–2012, O4:K68 was replaced by serovars O1:K25 (18.5%) and O1:KUT (11.8%). A similar serovar succession has been reported in Thailand [27].
The major V. parahaemolyticus pandemic serovars identified in this study were O3:K6, O1:K25, O1:KUT, O3:KUT, O1:K36. Of these, O3:KUT and O1:K36 serovars were newly identified. In addition, O1:K30, O1:K38, O1:K56, O2:K4, O4:KUT, O4:K4, O4:K13, O4:K25, O4:K55 and O8:K21 and O10:K60 serovars were also positive in the GS-PCR assay and hence considered pandemic strains. Studies conducted in Peru, Norway and Chile have also shown emergence of new GS-PCR positive serovars such as O3:KUT, O3:K58, O3:K68 [28]–[30]. Universally, all the pandemic strains have 7 base variations in the nucleotide sequence of toxRS operon, which encodes transmembrane proteins involved in regulation of virulence-associated genes. These distinctive gene mutations were found in the non-pandemic strains of V. parahaemolyticus. Based on our results and other reports, it appears that several new serovars have emerged recently with pandemic strain attributes. However, in southern Thailand, the major pandemic serovars remained consistent for more than 6 years [24]. The other noteworthy aspect of this study was the emergence of trh-harboring pandemic strains. Generally, pandemic strains of V. parahaemolyticus harbor only the tdh gene. The trh gene association has not been reported previously. Serovars O1:KUT and O3:KUT harbored the trh gene, and the other two tdh and trh positive strains belonging to O1:KUT and O1:K30 were negative in the GS-PCR.
Several investigations have shown that clinical strains of V. parahaemolyticus are susceptible to many antimicrobial agents as compared to environmental strains [27], [31], [32]. Recently, ESBL-production and fluoroquinolone resistance was reported in V. parahaemolyticus isolated from food samples [32], [33]. V. parahaemolyticus remained highly susceptible to many antimicrobial agents, despite the fact that other enteric pathogens have developed multiple antimicrobial resistances in this region [12]–[15]. In other countries, ampicillin/trimethoprim-sulfamethoxazole resistance has been reported in V. parahaemolyticus [24], [27], [34].
It is known that ampicillin resistance is very common in V. parahaemolyticus [9]. Following this trend, 98% of the V. parahaemolyticus strains isolated in the present study showed resistance to ampicillin. However, in the MIC assay, ampicillin resistance varied from moderate to high level, with selected strains belonging to different serovars. When examined for the mechanism of ampicillin resistance, we found that the resistance was not related to the tested bla gene alleles, as all the strains were negative in the PCR assays. Ampicillin resistance was also not related to a chromosomally encoded β-lactamase ORF (VPA0477; accession no. BA000032) as this encoding gene was detected in both susceptible and resistant strains. In V. parahaemolyticus, the beta-lactamase ORF (VPA0477; accession no. BA000032) has not been annotated consistently in pandemic and pre-pandemic strains of genomes sequenced (accession nos. BA000032 and CP003973) and hence there is no experimental proof for the functional aspect of this encoding gene. However, we found that the observed ampicillin resistance was mediated by an efflux system. This mechanism was demonstrated by synergistic testing with the efflux pump inhibitor CCCP. The MIC of ampicillin for V. parahaemolyticus decreased considerably when tested with CCCP at the highest concentration 1.5 µM. When the concentration of CCCP increased to 2 µM and above, the growth of V. parahaemolyticus was inhibited.
Streptomycin was the other antimicrobial agent for which most of the V. parahaemolyticus strains were resistant. The MIC of streptomycin revealed that resistance was close to that of the susceptibility cutoff value (>8 µg/ml) in E. coli [35]. The mechanism of resistance for this antibiotic in V. parahaemolyticus was not related to the presence of strA or aadA1, as these genes were found in only three strains.
The pandemic and non-pandemic strains were susceptible to trimethoprim-sulfamethoxazole, ceftriaxone, fluoroquinolones, and very few pandemic strains were resistant to chloramphenicol and nalidixic acid. The chromosomally encoded qnr homologue VPA0095 (accession no. BA000032) have more than 50% similarity with the plasmid-mediated qnrA and qnrS [36]. This qnr homologue was detected in 176 of 178 strains screened in this study. Although these two strains displayed susceptibility for fluoroquinolones, the MIC value for nalidixic acid was 1–3 fold less compared to strains that harbored the qnr homologue VPA0095.
V. parahaemolyticus infection has been significantly associated with older age group with clinical symptoms of abdominal pain, nausea, vomiting and bloody stool [5]. We found that stool specimens of V. parahaemolyticus infected cases were significantly detected with alkaline pH with high RBC and FLC counts. A high RBC and FLC count in the stool is an indication of an inflammatory diarrhea. In healthy individuals, the pH progressively rises in the small intestine from pH 6 to 7.4 in the terminal ileum. The pH falls to 5.7 in the caecum and steadily increases to pH 6.7 in the rectum [37]. Due to large secretion of small intestinal fluid, the pH of diarrheal stool remains alkaline when excreted. The alkaline pH favors many enteric vibrios and considerably reduces the normal gut flora [38]. We are not ruling out the possibility of involvement of other pathogens as mixed infections among diarrheal patients. It is worth mentioning that in the majority of the V. parahaemolyticus positive cases, this organism was detected as a sole pathogen indicating the importance of V. parahaemolyticus as one of the major etiological agents of diarrhea in this region.
Previous reports revealed clustering of V. parahaemolyticus O3:K6 and O4:K68 serovars from India and Thailand with 78–91% similarity in the PFGE profiling [39]. In the subsequent years, several other serovars were genetically grouped with O3:K6 [27], [40]. In this study, we found that pandemic serovars such as O3:K6, O1:K25, O4:K68 and O1:KUT were clustered in one clade and several new serovars remained in the other. The overall similarity between the old pandemic serovars with new serovars remained only about 50%. Recently, similar genetic event has not been reported among pandemic strains of V. parahaemolyticus.
In this surveillance study, we found variation in the isolation rates of V. parahaemolyticus from hospitalized acute diarrheal patients. Combined genetic and molecular typing analysis verified emergence of newer pandemic serovars in this region. The tested V. parahaemolyticus strains reveled susceptibility towards a wide range of antimicrobials used in the treatment of diarrheal infection.
|
10.1371/journal.pcbi.1005215 | Combined Changes in Chloride Regulation and Neuronal Excitability Enable Primary Afferent Depolarization to Elicit Spiking without Compromising its Inhibitory Effects | The central terminals of primary afferent fibers experience depolarization upon activation of GABAA receptors (GABAAR) because their intracellular chloride concentration is maintained above electrochemical equilibrium. Primary afferent depolarization (PAD) normally mediates inhibition via sodium channel inactivation and shunting but can evoke spikes under certain conditions. Antidromic (centrifugal) conduction of these spikes may contribute to neurogenic inflammation while orthodromic (centripetal) conduction could contribute to pain in the case of nociceptive fibers. PAD-induced spiking is assumed to override presynaptic inhibition. Using computer simulations and dynamic clamp experiments, we sought to identify which biophysical changes are required to enable PAD-induced spiking and whether those changes necessarily compromise PAD-mediated inhibition. According to computational modeling, a depolarizing shift in GABA reversal potential (EGABA) and increased intrinsic excitability (manifest as altered spike initiation properties) were necessary for PAD-induced spiking, whereas increased GABAAR conductance density (ḡGABA) had mixed effects. We tested our predictions experimentally by using dynamic clamp to insert virtual GABAAR conductances with different EGABA and kinetics into acutely dissociated dorsal root ganglion (DRG) neuron somata. Comparable experiments in central axon terminals are prohibitively difficult but the biophysical requirements for PAD-induced spiking are arguably similar in soma and axon. Neurons from naïve (i.e. uninjured) rats were compared before and after pharmacological manipulation of intrinsic excitability, and against neurons from nerve-injured rats. Experimental data confirmed that, in most neurons, both predicted changes were necessary to yield PAD-induced spiking. Importantly, such changes did not prevent PAD from inhibiting other spiking or from blocking spike propagation. In fact, since the high value of ḡGABA required for PAD-induced spiking still mediates strong inhibition, we conclude that PAD-induced spiking does not represent failure of presynaptic inhibition. Instead, diminished PAD caused by reduction of ḡGABA poses a greater risk to presynaptic inhibition and the sensory processing that relies upon it.
| Postsynaptic GABAAR mediate inhibition by causing hyperpolarization or by preventing (shunting) the depolarization caused by concurrent excitatory input. Presynaptic GABAAR work differently, in the spinal cord at least. Because of their higher-than-equilibrium intracellular chloride concentration, the central terminals of primary afferent fibers are depolarized by activation of GABAAR. This so-called primary afferent depolarization, or PAD, nonetheless reduces spike propagation and synaptic release from those fibers because of shunting effects and sodium channel inactivation. But those inhibitory effects can be diminished under certain pathological conditions; in fact, the emergence of dorsal root reflexes suggests that PAD can become paradoxically excitatory. The biophysical basis for this paradoxical excitation has been hinted at by experiments, but here, for the first time, we use computational modeling and dynamic clamp experiments to decipher how distinct contributing factors interact to enable PAD-induced spiking. Our results suggest that PAD-induced spiking requires a shift in GABA reversal potential plus changes in intrinsic excitability that allow for repetitive spiking during sustained depolarization. Inhibitory effects of PAD are retained under conditions in which GABAAR activation causes transient spiking and are only lost if GABAAR activation can evoke repetitive spiking.
| Synaptic inhibition regulates transmission of sensory signals through the spinal cord. Importantly, numerous chronic pain conditions are associated with diminished inhibition [1–5] and pharmacological blockade of inhibition at the spinal level has been shown to reproduce many features of those chronic pain conditions [6–9]. Decreased transmitter release, reduced GABAA/glycine receptor function, and altered chloride regulation are all potential disinhibitory mechanisms, but pre- and postsynaptic inhibition are not equally susceptible to certain pathological changes; for instance, the potassium-chloride co-transporter KCC2 is not expressed in primary afferent neurons, meaning disinhibitory effects of KCC2 downregulation [10] are attributable entirely to reduced postsynaptic inhibition, in cells that express KCC2. KCC3 is expressed in some primary afferents and can extrude chloride under isosmotic conditions [11,12] but it remains unknown whether KCC3 is altered under pathological conditions. Yet selective disruption of presynaptic inhibition can cause mechanical and thermal hypersensitivity [13] and presynaptic expression of the α2 GABA receptor subunit is necessary for the antihyperalgesic effect of diazepam [14]. These observations affirm that presynaptic GABAAR-mediated inhibition also plays a key role in nociception.
Pre- and postsynaptic inhibition in spinal cord are mechanistically distinct. Postsynaptically, in mature spinal neurons, the reversal potential associated with GABAAR (EGABA) is normally around -70 mV [10], meaning GABAAR activation reduces depolarization caused by concurrent excitatory input. Presynaptically, in the central terminals of primary afferents, EGABA is normally around -35 mV because chloride is actively loaded into primary afferents by the sodium-potassium-chloride co-transporter NKCC1 [13,15–17], thus GABAAR activation causes depolarization. Contrary to the presumed excitatory effect of depolarization, primary afferent depolarization (PAD) mediates inhibitory effects via sodium channel inactivation and shunting [18–21]. However, PAD can sometimes trigger spikes that conduct antidromically, thus producing what is referred to as a dorsal root reflex (DRR) [22]. One theory holds that antidromically conducted spikes mediate an inhibitory effect by colliding with and blocking othrodromically conducted spikes originating in the periphery [23,24]; however, collisions are unlikely since the latency to travel the full length of the nerve is short relative to the interspike interval at realistic spiking rates. PAD-induced spikes are unlikely to trigger synaptic release from the PAD-affected branch because spike amplitude is attenuated, but PAD-induced spikes that manage to propagate to adjacent, PAD-free branches may trigger synaptic release [25]. The experiments required to test these model predictions are prohibitively difficult. The above theory was formulated for large myelinated proprioceptive afferents involved in locomotion; in contrast, within smaller afferents responsible for nociception, the prevailing view is that PAD-induced spikes occur only under pathological conditions and that DRRs contribute to neurogenic inflammation and hypersensitivity [22,26]. Within this context, PAD-induced spiking is thought to represent conversion of PAD from an inhibitory process to an excitatory one [22].
With respect to biophysical mechanisms, PAD-induced spiking requires GABAAR activation [27] and NKCC1-mediated chloride loading [28]. Enhanced chloride loading and the consequent depolarizing shift in EGABA has been hypothesized to facilitate PAD-induced spiking [29,30]. Nerve injury increases NKCC1 protein levels and PAD [13,31], and although total NKCC1 expression is not altered by inflammation [32,33], NKCC1 membrane trafficking and phosphorylation are affected by painful stimuli [34]. Notably, inflammation causes a depolarizing shift in EGABA [35] and promotes DRRs in C and Aδ fibers [36]. Increased GABAAR density and reduced low-threshold potassium channel density have also been hypothesized to promote DRRs [35,37] but the full set of requirements for PAD-induced spiking remains unclear. We sought to identify which biophysical changes, alone or together, enable PAD-induced spiking and how such changes impact PAD-mediated inhibition.
Changes in GABA conductance density ḡGABA, its associated reversal potential EGABA, and intrinsic excitability have all been implicated in PAD-induced spiking, as outlined above. To account for whether a neuron spikes transiently or repetitively, and whether spike threshold is sensitive to the rate of depolarization, we discuss excitability in terms of spike initiation dynamics [38]. Rather than characterize further how excitability and GABAergic signalling are pathologically altered, we sought to decipher how known pathological alterations contribute to PAD-induced spiking. To this end, we took an approach distinct from previous studies to determine how isolated and combined changes in each factor–ḡGABA, EGABA, and excitability–affect PAD-induced spiking. We began with a minimalist conductance-based computer model to generate predictions that we then tested experimentally in acutely dissociated dorsal root ganglion (DRG) neurons using dynamic clamp. Intracellular recording/stimulation in most axons is prohibitively difficult but sustained depolarization of the soma or axon by optogenetic-based photostimulation evokes transient spiking, although photostimulation of peripheral axon terminals can evoke repetitive spiking in some DRG neurons [39]. It remains unclear how central axon terminals respond to sustained depolarization. We assume here that somatic and axonal spike initiation properties are qualitatively similar, but if axons were more excitable (i.e. more prone to repetitive spiking) than somata, they would operate farther to the right along the “excitability” axis described below. We applied virtual GABA conductances rather than assuming the soma and axon have equivalent GABAAR densities. As a final step, we confirmed our results in a multicompartment axon model.
Starting with computer simulations, we co-varied EGABA and intrinsic excitability (controlled by βw; see Methods) while keeping ḡGABA fixed at 2 nS/pF. The light grey and dark grey regions of the resulting 2-D bifurcation diagram (Fig 1A) show the EGABA and βw combinations that produce transient and repetitive spiking, respectively. Spiking pattern was determined by the response to GABA conductance “steps”. To more accurately simulate different forms of synaptic transmission, other conductance waveforms were tested: phasic inhibition via intrasynaptic GABAAR was modeled by a “fast” synaptic waveform (τrise = 2 ms; τdecay = 20 ms; see Eq 6); tonic inhibition corresponds to the sustained component of the conductance step, but we also tested a “slow” synaptic waveform with intermediate kinetics to simulate spilled-over GABA asynchronously activating extrasynaptic GABAAR (τrise = 20 ms; τdecay = 200 ms). Fig 1B shows responses to each stimulus waveform for parameter combinations labeled a-f on Fig 1A. Under control conditions used in the experiments described in this study (EGABA = -35 mV and βw = -20 mV; point b), GABA conductance caused depolarization but no spiking. PAD-induced repetitive spiking required a combined depolarizing shift in EGABA and βw (point e) whereas transient spiking required a smaller increase in βw (point c) and could result solely from a large change in EGABA. By comparison, an isolated change in βw could not enable PAD-induced spiking. As illustrated in panel c of Fig 1B, slow-onset GABAAR input required stronger input to elicit spiking because transient spiking involves a spike initiation mechanism that is sensitive to the rate of depolarization [40].
Next, we repeated the 2-D bifurcation analysis for different ḡGABA values to produce a family of curves (Fig 1C). The dashed curve demarcating the minimum requirements for transient spiking shifted downward as ḡGABA was increased. The solid curve demarcating the minimum requirements for repetitive spiking also shifted downward for an initial increase in ḡGABA but shifted rightward as ḡGABA was increased further, indicating that GABAAR activation is maximally excitatory at intermediate densities. Somatic recordings have demonstrated somatic ḡGABA between 0.2 and 0.5 nS/pF [35] and the absolute ḡGABA values reported by Chen et al. [13] correspond to approximately 0.1 nS/pF after conversion to densities based on estimated surface areas. Axonal ḡGABA may differ from somatic ḡGABA (given precedents for differential ion channel distribution [41]) but measuring ḡGABA in central axon terminals is prohibitively difficult. Our experimental approach does not rely on measuring axonal ḡGABA but, instead, was designed to determine the minimum ḡGABA required (for different EGABA and intrinsic neuronal excitability) to enable PAD-induced spiking; comparing this value against measured ḡGABA (in the soma) reveals whether the density of native GABA receptors is sufficient to evoke spiking under different conditions. It remains unclear what ḡGABA would be necessary to evoke spiking in central axon terminals.
To test the simulation predictions described above, we conducted experiments in acutely dissociated DRG somata using an approach distinct from previous studies. Rather than activating native GABAARs by puffing GABA (which would produce a current whose conductance, reversal potential and kinetics are not easily measured or independently manipulated), we used dynamic clamp to apply a virtual conductance whose parameters are precisely and independently controllable. In this way, we quantified the minimum virtual ḡGABA required to elicit spiking under different conditions. Importantly, because virtual ḡGABA can be higher than native ḡGABA, the density of native GABAAR does not limit our studies; indeed, failure of GABA puffs to evoke spikes in previous studies [13,35,42] suggests that somatic ḡGABA is normally too low to produce spikes, but ḡGABA may be higher in central axon terminals. In dynamic clamp, the voltage recorded from a neuron is passed to a computer, which, in real time, uses voltage to calculate current that is injected back into the patched neuron, thereby introducing a virtual conductance [43]. This approach allows manipulations to be applied like in computer simulations but to real neurons, such that we can avoid modeling the neuron (and making any assumptions about intrinsic excitability) and test directly how virtual GABAAR input affects native voltage-gated channels controlling spike initiation. Notably, photostimulation-based testing of axonal excitability has revealed transient spiking comparable to that observed in somata [39] but the excitability of central axon terminals remains uncertain. If central axon terminal and somatic excitability are similar, then the requirements for PAD-induced spiking ascertained for the soma can be extrapolated to those terminals; on the other hand, if those terminals are more excitable, they would operate farther to the right on the excitability axis depicted in the inset of Fig 2A.
To begin, we tested virtual GABA conductances in neurons from naïve animals before and after reproducing the hyperexcitability associated with nerve injury by blocking Kv1-type potassium channels with 4-AP [44,45]; this corresponds in the model to setting βw to less negative values. Testing with different EGABA and stimulus waveforms, we systematically increased ḡGABA to try to elicit spiking. As illustrated for a typical cell in Fig 2A, PAD was most likely to produce spiking after application of 4-AP and a depolarizing shift in EGABA to -20 mV. Fig 2B summarizes the proportion of cells in which PAD-induced spiking was observed under each test condition. For cells from naïve animals tested with EGABA = -35 mV, 4-AP increased the proportion exhibiting PAD-induced spiking but not significantly (p = 0.079; Fisher’s exact test) whereas the 4-AP effect was highly significant for EGABA = -20 mV (p = 0.004). Shifting EGABA from -35 mV to -20 mV significantly increased the proportion of cells exhibiting PAD-induced spiking both before and after 4-AP (p < 10−3 and 10−4, respectively), consistent with the NKCC1 hypothesis of DRR generation [29,30]. But as predicted by our simulations, the proportion of cells with PAD-induced spiking was most significantly increased by the combination of 4-AP and a depolarizing shift in EGABA (p < 10−9). Within this data set, two cells were subsequently identified as outliers based on analysis of the minimum ḡGABA needed for PAD-induced spiking (see below); removing those outliers did not substantively alter the statistical results reported above.
Based on cells that exhibited PAD-induced spiking before and after 4-AP for EGABA = -20 mV, the minimum ḡGABA needed to elicit spiking was significantly reduced from 0.49 ± 0.07 nS/pF (mean±SEM) to 0.16 ± 0.03 nS/pF by 4-AP (p = 0.005, Tukey test following ANOVA described below) (Fig 2C left). Plotting the same data against soma diameter revealed a trend towards higher minimum ḡGABA for smaller cells, but soma diameter did not have a significant effect (p = 0.61) and nor did it interact significantly with the 4-AP effect (p = 0.29; two-way repeated measures ANOVA) (Fig 2C right). Notably, we report all conductances as densities to correct for the direct effect of membrane surface area on our measurements; however, soma diameter is known to correlate with fiber type [46], and so the insignificant effect of cell size (after normalization by surface area) argues that minimum ḡGABA does not differ significantly between myelinated (A) and unmyelinated (C) neurons. Of the cells that exhibited PAD-induced spiking for both EGABA values after 4-AP, the minimum ḡGABA needed to elicit spiking was significantly reduced from 0.30 ± 0.07 nS/pF to 0.11 ± 0.02 nS/pF by shifting EGABA from -35 mV to -20 mV (p = 0.022, paired t-test) (Fig 2D). Of the 10 cells tested with both fast and slow gGABA waveforms at EGABA = -20 mV after 4-AP, 7 responded to both stimuli with transient spiking and 2 responded with repetitive spiking. Among transient spiking cells, the slow waveform required higher ḡGABA than the fast waveform to elicit transient spiking (0.46 ± 0.09 nS/pF vs 0.27 ± 0.09 nS/pF) which, although not a statistically significant difference (p = 0.25; paired t-test), is consistent with a spike initiation mechanism sensitive to the rate of depolarization. By comparison, the two repetitive spiking cells required exactly the same minimum ḡGABA for the fast and slow waveforms, consistent with a spike initiation mechanism sensitive only to the amplitude of depolarization [40]. Comparing the responses to gGABA steps and ramps illustrates that the latter are far less effective in eliciting transient spiking (Fig 2E). All of these experimental data are consistent with simulation results in Fig 1 and S1 Fig.
Like 4-AP, nerve injury increased the proportion of cells exhibiting PAD-induced spiking (bars on right side of Fig 2B). Compared against naïve cells without 4-AP, nerve injury caused no change in the proportion of cells exhibiting PAD-induced spiking for EGABA = -35 mV (p = 1) whereas it did significantly increase that proportion for EGABA = -20 mV (p = 0.028; Fisher’s exact tests). Nerve injury and treatment of naïve cells with 4-AP resulted in a similar proportion of cells with PAD-induced spiking when tested with EGABA = -35 mV and -20 mV (p = 1 and 0.40, respectively). Among nerve-injured cells, shifting EGABA from -35 mV to -20 mV significantly increased the proportion with PAD-induced spiking (p = 0.001). Consistent with the combined effects of 4-AP and altered EGABA, the proportion of cells with PAD-induced spiking was most significantly increased by the combination of nerve injury and a depolarizing shift in EGABA (p < 10−5).
Testing with current injection (Istim) confirmed that 4-AP had the intended effect of increasing excitability yet, despite responding to Istim steps with repetitive spiking, most neurons responded to gGABA steps with transient spiking, as illustrated in Fig 3A. Specifically, PAD-induced repetitive spiking was not observed in any nerve-injured neurons and was seen in only two neurons after 4-AP application. All neurons were tested with a broad range of ḡGABA to confirm that repetitive spiking could not eventually be achieved by applying a stronger conductance. Increasing ḡGABA above the minimum required to elicit transient spiking consistently caused attenuation of the spike height and clamped the subsequent voltage near EGABA (Fig 3B). Based on our simulation results (see Fig 1A), we reasoned that the lack of repetitive spiking was due to 4-AP or nerve injury not causing a sufficient increase in excitability. To test this hypothesis, we further increased excitability by using dynamic clamp to introduce a virtual sodium conductance like that upregulated after nerve injury [45]. As predicted, PAD-induced repetitive spiking was made possible by this additional manipulation (Fig 3C). Although we managed to reproduce PAD-induced repetitive spiking, the extent of the required manipulations suggests that naturally occurring pathological changes cause few neurons to become sufficiently hyperexcitable that PAD will induce repetitive spiking. That said, if the central terminals of axons are more excitable (i.e. more prone to repetitive spiking) than the soma, PAD would be more likely to elicit repetitive spiking than suggested by our data.
The above results demonstrate that depolarizing GABA current can induce transient spiking under conditions associated with nerve injury. This does not, however, exclude PAD from retaining its inhibitory effects, especially given that inhibition stems from sodium channel inactivation and shunting. In fact, although PAD may induce a single spike at its onset, shunting effects persist as long as GABAAR are activated. This raises the important question of whether more spikes (arising in the periphery or ectopically in the soma or a neuroma) are blocked by PAD than are induced by PAD in the central axon terminals.
Our initial model did not include sodium channel inactivation for the sake of simplicity; therefore, our next step was to modify the model so that a certain proportion of sodium channels, controlled by parameter p, experience inactivation (Eqn. 7). Using this new model, we set βw to 0 mV to facilitate repetitive spiking and conducted 2-D bifurcation analysis to determine the p and EGABA combinations associated with different effects of PAD (Fig 4A). The grey region shows parameter combinations for which a gGABA step (2 nS/pF) applied alone elicits spiking (sample traces a and d in Fig 4B). The green region shows parameter combinations for which the same gGABA step inhibits spiking induced by Istim steps (sample traces c-e in Fig 4B). Importantly, the green and grey regions overlap, thus demonstrating that PAD can induce spikes yet nonetheless block spikes originating by other means. Fig 4C shows the 2-D bifurcation analysis repeated for different ḡGABA values. The region of PAD-induced spiking remained unchanged (not illustrated) but the region of PAD-mediated inhibition expanded as ḡGABA was increased, suggesting that stronger GABAAR activation manages to terminate spiking despite a smaller proportion of inactivatable sodium channels.
To measure PAD-mediated inhibition in real DRG neurons, we combined gGABA and Istim steps as done for simulations in Fig 4B. Fig 5A shows a typical neuron in which Istim elicited repetitive spiking. Interposing a gGABA step during the Istim step caused reduction or complete cessation of repetitive spiking depending on ḡGABA and EGABA. Note that spikes occurring during the gGABA step were shorter than those occurring outside the gGABA step, consistent with the shunting effect of the virtual GABA conductance. Applying the gGABA step before the onset of Istim confirmed that the former could elicit transient spiking yet still inhibit the repetitive spiking otherwise driven by Istim (Fig 5B). Using the same stimulus sequence, we measured rheobase (i.e. the minimum Istim required to elicit spiking) for each level of ḡGABA (Fig 5C). Rheobase was significantly increased by increments in ḡGABA (p = 0.013, two-way repeated measures ANOVA) but was not significantly affected by EGABA (p = 0.52) (Fig 5D). These data confirm that PAD elicited in the cell body of DRG neurons mediates shunting inhibition even under conditions in which it can induce spiking.
Activation of the calcium-activated chloride channel ANO-1 in primary afferent neurons can evoke or exacerbate pain, especially under inflammatory or neuropathic conditions [47–50]. Notably, intracellular chloride tends to be elevated under those conditions (see Introduction), which may explain why ANO-1 activation is excitatory rather than inhibitory. Consistent with this, ANO-1 modulation of spiking evoked by current injection is sensitive to intracellular chloride level [51] but demonstration that ANO-1 itself evokes spiking was based on a chloride reversal potential of -18 mV [49]. Given its activation requirements [52], we predicted that ANO-1 channels would not be activated by the GABAergic input underlying PAD; recall that GABAAR activation is necessary for PAD [22]. Nonetheless, to rule out a contribution by ANO-1, we repeated virtual PAD experiments (like in Fig 2) before and after blockade of ANO-1 channels by bath-applied 10 μM T16Ainh-A01 (A01) (Fig 6). Based on the pipette solution, the chloride reversal potential for ANO-1 was -20 mV but EGABA for virtual gGABA was set to -35 or -20 mV in dynamic clamp. As predicted, ANO-1 blockade had no significant effect on the minimum ḡGABA needed to evoke spiking for EGABA = -20 mV (p = 1.0, paired t-test; Fig 6A) and nor did it significantly affect the depolarization evoked by different ḡGABA for EGABA = -35 mV (p = 0.77, two-way repeated measures ANOVA; Fig 6B) or have any effect on rheobase, input resistance, or resting membrane potential (p > 0.05, paired t-tests). The data above are based exclusively on capsaicin-responsive cells (see Fig 6C) since ANO-1 channels are expressed primarily in cells that express TRPV1 [47]. Notably, the response to capsaicin was reduced by ANO-1 blockade (Fig 6D), consistent with Takayama et al. [49] and thus verifying the efficacy of our A01. Based on these results, we conclude that ANO-1 channels are not activated and, therefore, do not contribute to PAD under our experimental conditions.
All simulations described thus far were conducted in a single compartment model. This adequately simulates spike initiation occurring in proximity to the recording electrode, as occurs when recording from an isolated DRG soma. Although spontaneous or PAD-induced spiking may arise at the site of PAD, an important inhibitory effect of PAD in the intact fiber is to block the orthodromic propagation of spikes originating in the periphery. To test for conduction block, we converted our single-compartment model into a 3-compartment model (Fig 7A). Although still very simple compared with past models used to study this topic [e.g. 19,25,53], this model suffices to qualitatively illustrate key points relevant for the present study. Each compartment was further subdivided into equipotential segments. Based on its small diameter and the absence of nodes, this model simulates continuous propagation in an unmyelinated fiber. By applying GABA conductance to the middle compartment, we tested if that conductance can induce spikes (originating within that compartment) and/or block the propagation of other spikes (evoked at the far end of adjacent compartment).
For an EGABA value of -35 mV, gGABA never evoked spiking (consistent with the single-compartment model) but it did block spike propagation (Fig 7B, left column). Interestingly, blocked propagation could occur even in the absence of sodium channel inactivation, therein supporting claims that shunting mediated by gGABA mediates an inhibitory effect. When EGABA was shifted to -20 mV, gGABA evoked a single spike that propagated in both directions away from the center compartment (Fig 7B, right column). Yet despite this excitatory effect, propagation of other spikes was blocked in two of the three conditions illustrated. Sample traces were chosen to illustrate that large gGABA could block propagation in the absence of sodium channel inactivation but a smaller gGABA could achieve the same effect when combined with sodium channel inactivation. Fig 7C demonstrates that sodium channel inactivation can accumulate over time, thus eventually blocking spikes traveling as part of a train. These results confirm that PAD does not abruptly lose its inhibitory effects once able to induce its own spikes.
Using computer simulations and an experimental approach distinct from previous studies, we have identified which pathological changes are necessary and sufficient to enable PAD-induced spiking. We determined that a depolarizing shift in EGABA is necessary yet insufficient to enable PAD-induced spiking in most DRG neurons. An increase in intrinsic excitability (i.e. altered spike initiation properties) is also necessary, especially to enable PAD-induced repetitive spiking. Neurons may experience both changes after nerve injury or inflammation, meaning PAD-induced spiking could occur in certain pathological conditions [22,26,29]; however, other factors such as the requirement for fast depolarization suggest that PAD-induced spiking is probably rare (see below), but this depends on the excitability of central axon terminals, which still remains uncertain. Intriguingly, our data also suggest that PAD continues to mediate presynaptic inhibition under conditions in which it can induce transient spiking. Although seemingly paradoxical, the co-existence of excitatory and inhibitory effects has been observed previously in studies of presynaptic inhibition in crayfish [54] and is consistent with the biophysical mechanisms responsible for each effect. This is unlike what happens postsynaptically in spinal neurons, where paradoxical excitation develops only after inhibition fails [10,55]. Our data argue that increased PAD has a net inhibitory effect, meaning paradoxical excitation via enhanced PAD poses less risk to somatosensory processing than disinhibition caused by reduced PAD.
The GABA conductance density required for PAD-induced spiking under normal conditions is evidently quite high, so much so that we were able to elicit spiking in only 2 of 29 neurons despite testing with virtual ḡGABA several times greater than the typical density measured in somata [13,35]. This is consistent with previous failures to elicit spikes by puffing GABA on the soma [13,35,42]. Puffed GABA also failed to elicit calcium signals when applied to the central terminals of GCaMP-expressing primary afferents [13], and Verdier et al. [56] observed GABA-induced spiking in only 4 of 77 neurons tested in the trigeminal nucleus. The value of ḡGABA in central axon terminals remains an open question but evidence points to reduced expression of presynaptic GABAARs following nerve injury [13,57,58], which suggests that presynaptic inhibition is weakened by reduction of ḡGABA rather than ḡGABA becoming strong enough that PAD induces spiking. That said, the minimum ḡGABA needed for PAD-induced spiking is reduced by increased neuronal excitability (Fig 2C) and by a depolarizing shift in EGABA (Fig 2D). Unlike an increase in ḡGABA, which increases inhibitory effects due to shunting, increased neuronal excitability and depolarized EGABA can encourage PAD-induced spiking without enhancing PAD-mediated shunting. Studying transient spiking cells in the chick cochlear nucleus, Monsivais and Rubel [59] found that depolarizing GABAAR input could elicit spiking after blockade of the low-threshold potassium current known to be responsible for transient spiking [60]. The same GABAAR input normally inhibited stimulus-evoked spiking by activating the low-threshold potassium current and thereby elevating spike threshold [59]. Those data are entirely consistent with results presented here. Notably, PAD-induced spiking would be more likely in central axon terminals if those terminals are more excitable that we have assumed based on extrapolation from somatic data.
Intracellular chloride could be depleted during PAD if chloride uptake via NKCC1 became saturated (at least transiently) and thus failed to keep pace with chloride efflux via activated GABAA channels. The potential for altered chloride concentration is exacerbated by the small caliber of central axon terminals, especially C fibers, since intracellular volume is small compared to surface area [61]. Chloride depletion, if it occurred, would cause an activity-dependent hyperpolarizing shift in EGABA, the implication being that EGABA may be near -20 mV only at the onset of GABAAR activation. Given that PAD-induced spiking depends on a depolarized EGABA value, a hyperpolarizing shift would discourage PAD-induced repetitive spiking. That said, the transient spiking observed in our dynamic clamp experiments was not due to chloride depletion since the virtual GABA current is mediated by current injection through the patch pipette rather than by chloride efflux across the cell membrane. In effect, PAD-induced repetitive spiking may be more difficult to evoke under natural conditions, and transient spiking may rely even more heavily on abrupt depolarization than our experiments suggest.
Following on the above points, both simulations and experiments demonstrated that smaller pathological changes in EGABA and/or excitability are required to enable PAD-induced transient spiking than are required for PAD-induced repetitive spiking. This has important implications. Even if sustained, PAD is likely to produce only one spike at its onset (if it produces any spikes at all) and will likely not produce any spikes unless its onset is abrupt. This is because transient spiking involves a spike initiation mechanism that is sensitive to the rate of depolarization [40]. Sensitivity to gGABA onset kinetics would be inconsequential if presynaptic inhibition was phasic, which is to say that the GABAARs are clustered within the synaptic cleft and therefore receive an abrupt pulse of GABA upon its vesicular release [62], but evidence points toward a more tonic mode of action (unlike the phasic inhibition studied in the crayfish neuromuscular junction [63]) as outlined below. Recording from mammalian primary afferent terminals to measure the activation kinetics (and density) of the GABAAR current is prohibitively difficult, but immunocytochemical evidence argues that C fiber terminals are devoid of gephyrin [64]. Since gephyrin is usually necessary for GABAAR clustering [65], its absence suggests that GABAARs are distributed more diffusely. Electrophysiological evidence for high-affinity GABAARs in primary afferent neurons [37] supports this view since such receptors have a δ subunit [66] in place of the γ subunit that is necessary for clustering [62,67]. If primary afferent GABAARs are indeed distributed extrasynaptically, and are thus activated asynchronously as GABA diffuses beyond the synaptic cleft, then gGABA will have slow onset kinetics and is unlikely to elicit transient spiking. Only the most hyperexcitable fibers (i.e. those capable of PAD-induced repetitive spiking) are likely to exhibit any PAD-induced spiking. And whereas PAD-induced transient spiking relies on abrupt GABAAR activation, PAD-mediated inhibition does not; instead, PAD-mediated inhibitory effects will last throughout the duration of the PAD. In other words, slow activation of extrasynaptic GABAARs–arguably the most likely scenario at least for C fiber terminals (see above)–will not cause PAD-induced spiking but will cause PAD-mediated inhibition.
Notably, dorsal root reflexes (DRRs) have typically been studied using electrical stimulation of a nerve or dorsal root to synchronously activate a large number of afferent fibers [e.g. 68]. Notwithstanding differential conduction latencies, such input will evoke a large burst of GABA release, causing GABAAR activation that is ideally suited for PAD-induced transient spiking. It is not obvious that those same fibers would exhibit PAD-induced spiking under more natural conditions (i.e. less synchronous inputs). However, Dubuc et al. [69] observed antidromic spiking in 19% of cat dorsal root fibers during fictive locomotion. It has long been recognized that dorsal root reflexes are more common in certain afferents (e.g. stretch receptors) with direct evidence for DRRs being weakest in C fibres [22]. However, Lin et al. [36] reported spontaneous and von Frey-evoked antidromic spiking in all fiber types and, moreover, found that intradermal capsaicin selectively increased antidromic spiking in C and Aδ fibers. Based on more recent observations, including data presented here, one may suspect that chloride regulation, GABA receptor clustering and/or intrinsic excitability differ between afferent types. Somatic recordings suggest that important differences do indeed exist [70] but definitively resolving this requires comparison of axon terminals (rather than somata) and is therefore technically difficult. Notably, Dubuc et al. [69] observed repetitive antidromic spiking, as have others [e.g. 71], which argues that the excitability of certain afferent terminals is quite high. The role of axonal excitability warrants closer attention in future studies. Observation that cooling increases DRRs [72] likely holds important clues. Please see [5] for a recent review of other factors.
As already explained, PAD-induced spiking does not equate with failure of presynaptic inhibition. The resilience of presynaptic inhibition is best appreciated by comparing how pre- and postsynaptic inhibition fail. As KCC2 is downregulated postsynaptically, EGABA undergoes a depolarizing shift that directly compromises the inhibitory effect of GABAergic input [61]. The same shift in EGABA that reduces postsynaptic inhibition is what eventually results in paradoxical excitation. This shift from inhibition to paradoxical excitation is evidently not what happens presynaptically. In primary afferent terminals, the changes required for paradoxical excitation–a shift in EGABA and increased excitability–do not undermine the inhibitory effect; in fact, the relatively high ḡGABA required for PAD-induced spiking also encourages PAD-mediated inhibition. This conclusion contradicts past assumptions on this matter. Furthermore, whereas the risk of paradoxical excitation increases postsynaptically during sustained GABAergic input (because of chloride accumulation), presynaptically, the balance shifts towards inhibitory effects over time as sodium channel inactivation accumulates and if intracellular chloride is depleted. The greatest risk to presynaptic inhibition is reduced PAD rather than enhanced PAD.
To conclude, we have demonstrated that combined changes in EGABA and intrinsic excitability enable PAD-induced transient spiking. However, unless neurons become so hyperexcitable that PAD can induce repetitive spiking, slow (asynchronous) activation of extrasynaptic GABAARs is unlikely to elicit any spiking. On the other hand, PAD will continue to mediate presynaptic inhibition. In practical terms, our results suggest that presynaptic inhibition is a viable therapeutic target whose enhancement carries little risk of causing paradoxical excitation.
All experiments were approved by the University of Pittsburgh IACUC and by The Hospital for Sick Children Animal Care Committee.
Starting from a previously published model [45,73], our single compartment, conductance-based model is described as follows:
CdVdt=Istim−g¯Nam∞(V)(V−ENa)−g¯Kw(V−EK)−gleak(V−Eleak)−gGABA(t)(V−EGABA)
(1)
where activation variable m changes instantaneously with voltage V according to
m∞(V)=0.5[1+tanh(V−βmγm)],
(2)
whereas w changes more slowly according to
dwdt=ϕww∞(V)−wτw(V),
(3)
w∞(V)=0.5[1+tanh(V−βwγw)],
(4)
τw(V)=1cosh(V−βw2γw).
(5)
Neuronal excitability was varied by changing parameter βw [38]. Injury-induced hyperexcitability can be reproduced by shifting βw from its normal value of around -20 mV to less negative values [73]. Setting βw to less negative values reflects a multitude of potential injury-induced molecular changes including reduced KV1-type potassium current, which we model experimentally using 4-AP application, and increased sodium current, which we model experimentally using dynamic clamp (see below); the effect of such changes, occurring alone or together, is to alter spike initiation [45]. All other neuronal parameters were fixed as reported previously [38] at the following values: C = 2 μF/cm2; sodium conductance ḡNa = 20 mS/cm2, ENa = 50 mV, βm = -1.2 mV, γm = 18 mV; potassium conductance ḡK = 20 mS/cm2, EK = -100 mV, ϕw = 0.15, γw = 10 mV; leak conductance gleak = 2 mS/cm2, Eleak = -70 mV.
Stimulating current Istim was not applied unless indicated. Maximal GABA conductance density ḡGABA and reversal potential EGABA were varied. Units for ḡGABA were converted to nS/pF for comparison with experimental measurements. The normal EGABA value in primary afferent is around -35 mV based on measurements using different techniques [12,13,35,42]. GABA conductance was activated as a step or as a synaptic waveform described by
gGABA(t)=g¯GABAx[−e−tτrise+e−tτdecay],
(6)
which comprises an exponential rise to maximum (with time constant τrise) followed by an exponential decay back to baseline (with τdecay). The peak is normalized to 1 by factor x before being scaled by ḡGABA. Kinetics are reported in the Results section.
For simulations reported in Figs 4 and 7, sodium channel inactivation h was applied to a proportion of sodium channels defined by p, thus giving the following current balance equation
CdVdt=Istim−pg¯Nam∞(V)h(V−ENa)−(1−p)g¯Nam∞(V)(V−ENa)−g¯Kw(V−EK)−gleak(V−Eleak)−gGABA(t)(V−EGABA).
(7)
Changes in h are described by the same equations used to describe w (Eq 3–5) where βh = -28 mV, γh = -14 mV, and ϕh = 0.005.
All simulations in single compartment models were conducted in XPP. Bifurcation analysis was conducted using AUTO via the XPP interface. The multicompartment model was built in NEURON. Ion channels were modeled as above except that both ḡNa and ḡK were increased to 30 mS/cm2. Additional parameters were as follows: axial resistivity Ra = 150 Ω·cm, diameter = 1 μm, compartment length = 1 mm, d_lambda = 0.01. GABA conductance ḡGABA was modeled as a uniform density throughout the middle compartment.
All experiments were carried out on adult (200–340 g) male Sprague-Dawley rats (Harlan, Indianapolis, IN and Charles River, Montreal, Quebec). A subset of animals received spinal nerve ligation (SNL) 2–5 days before terminal experiments [74]. Under isoflurane anesthesia, the paraspinal muscles were separated to access the L6 process, which was carefully removed. The L5 spinal nerve was tightly ligated with 6–0 silk suture. All nerve-injured animals maintained motor function but developed neuropathic pain as inferred by guarding of the affected paw.
To collect DRG neurons, rats were deeply anesthetized by subcutaneous injection of anesthetic cocktail (1 ml/kg of 55 mg/ml ketamine, 5.5 mg/ml xylazine, and 1.1 mg/ml acepromazine) or by isoflurane (4% for induction; 2.5% for maintenance). DRG (L4 and L5 in naïve animals; L5 in nerve-injured animals) were surgically removed to chilled MEM-FBS culture media and desheathed. DRG were then enzymatically treated for 45 minutes in culture media composed of 89% MEM, 370 units/ml penicillin and 370 μg/ml streptomycin, 1% MEM vitamin solution (all from Life Technologies), and 1.2 mg/ml collagenase Type 4 (Worthington Biochemical Corp). DRG were mechanically dissociated by trituration with a fire-polished Pasteur pipette, and further enzymatically treated for 5 minutes in Ca2+- and Mg2+-free Hanks’ balanced salt solution (HBSS; Life Technologies Inc), containing 2.5 mg/ml trypsin (Worthington Biochemical Corp) and 0.02% sterile ethylenediaminetetraacetic acid (EDTA; Sigma-Aldrich Canada Ltd). Trypsin activity was subsequently inhibited by the addition of MEM-FBS supplemented with 0.625 mg/ml MgSO4 (Caledon Labs). Dissociated cells in MEM-FBS were plated on glass coverslips previously coated by a solution of 0.1 mg/ml poly-D-lysine, and incubated in MEM-FBS at 37°C, 5% CO2, and 90% humidity for 2 h. Coverslips were then transferred to a HEPES-buffered Leibovitz’s L-15 media containing glutamine (Life Technologies Ltd), 10% FBS, 100 units/ml of penicillin and 100 μg/ml streptomycin, and 5 mM D-glucose (Caledon Labs) and stored at room temperature until used for experiments for 2–28 hours later. Spiking properties do not change appreciably over this period and nor do neurites develop based on storage at room temperature, omission of laminin from coverslips, and the growth factor-free culture medium used.
Coverslips with cultured cells were transferred to a recording chamber constantly perfused with room temperature, oxygenated (95% O2 and 5% CO2) artificial cerebral spinal fluid containing (in mM) 126 NaCl, 2.5 KCl, 2 CaCl2, 2 MgCl2, 10 D-glucose, 26 NaHCO3, 1.25 NaH2PO4. Cells were recorded in the whole-cell configuration with >70% series resistance compensation using an Axopatch 200B amplifier (Molecular Devices; Palo Alto, CA). Electrodes (2–5 MΩ) were filled with a recording solution containing (in mM) 125 KMeSO4, 5 KCl, 10 HEPES, 2 MgCl2, 4 ATP, 0.4 GTP as well as 0.1% Lucifer Yellow; pH was adjusted to 7.2 with KOH and osmolality was between 270 and 290 mosmol/L. For experiments on the contribution of ANO-1 channels, KMeSO4 was reduced to 67 mM and KCl was increased to 63 mM to give ECl = -20 mV. Data were low-pass filtered at 2 KHz, digitized at 20 KHz using a CED 1401 computer interface (Cambridge Electronic Design, Cambridge, UK), and analyzed offline. Virtual GABA conductance was applied via dynamic clamp using Signal 5 software (CED). The virtual conductance was modeled as a step or as a synaptic waveform described by Eqn. 6. To express the virtual conductance as a density and thus exclude direct effects of cell size, we normalized absolute conductance values by membrane capacitance C because C is proportional to the surface area of the cell. Capacitance was measured for each cell based on responses to small (50 pA) hyperpolarizing current steps, where C = τmembrane / Rin. To increase cellular excitability in neurons from naïve animals, potassium channels were blocked with bath applied 4-aminopyridine (4-AP). In a subset of experiments with 4-AP, a virtual voltage-dependent sodium conductance was also inserted via dynamic clamp using the equations and parameters reported by Ratté et al. [45]. Neurons from nerve-injured animals are already hyperexcitable and were not, therefore, subject to manipulations (i.e. 4-AP or virtual sodium conductance) intended to increase excitability.
All data and computer code are available from the corresponding author upon request.
|
10.1371/journal.pgen.1006278 | Evolution of New cis-Regulatory Motifs Required for Cell-Specific Gene Expression in Caenorhabditis | Patterning of C. elegans vulval cell fates relies on inductive signaling. In this induction event, a single cell, the gonadal anchor cell, secretes LIN-3/EGF and induces three out of six competent precursor cells to acquire a vulval fate. We previously showed that this developmental system is robust to a four-fold variation in lin-3/EGF genetic dose. Here using single-molecule FISH, we find that the mean level of expression of lin-3 in the anchor cell is remarkably conserved. No change in lin-3 expression level could be detected among C. elegans wild isolates and only a low level of change—less than 30%—in the Caenorhabditis genus and in Oscheius tipulae. In C. elegans, lin-3 expression in the anchor cell is known to require three transcription factor binding sites, specifically two E-boxes and a nuclear-hormone-receptor (NHR) binding site. Mutation of any of these three elements in C. elegans results in a dramatic decrease in lin-3 expression. Yet only a single E-box is found in the Drosophilae supergroup of Caenorhabditis species, including C. angaria, while the NHR-binding site likely only evolved at the base of the Elegans group. We find that a transgene from C. angaria bearing a single E-box is sufficient for normal expression in C. elegans. Even a short 58 bp cis-regulatory fragment from C. angaria with this single E-box is able to replace the three transcription factor binding sites at the endogenous C. elegans lin-3 locus, resulting in the wild-type expression level. Thus, regulatory evolution occurring in cis within a 58 bp lin-3 fragment, results in a strict requirement for the NHR binding site and a second E-box in C. elegans. This single-cell, single-molecule, quantitative and functional evo-devo study demonstrates that conserved expression levels can hide extensive change in cis-regulatory site requirements and highlights the evolution of new cis-regulatory elements required for cell-specific gene expression.
| Diversification of mechanisms regulating gene expression of key developmental factors is a major force in the evolution of development. However, in the past, comparisons of gene expression across different species have often been qualitative (i.e. ‘expression is on versus off’ in a certain cell) without precise quantification. New experimental methods now allow us to quantitatively compare the expression of gene homologs across species, with single cell resolution. Moreover, the development of genome editing tools enables the dissection of regulatory DNA sequences that drive gene expression. We use here a well-established “textbook” example of animal organogenesis in the microscopic nematode, Caenorhabditis elegans, focusing on the expression of lin-3, coding for the main inducer of the vulva, in a single cell called the anchor cell. We find that the lin-3 expression level is remarkably conserved, with 20–25 messenger RNAs per anchor cell, in species that are molecularly as distant as fish and mammals. This conservation occurs despite substantial changes and compensation in the regulatory elements required for cell-specific gene expression.
| Developmental systems operate in the presence of stochastic, environmental and genetic perturbations. While the output of a developmental system may be under stabilizing selection and remain mostly invariant, many internal variables such as the expression of a key gene or the activity of signalling pathways can be sensitive to perturbations. To reach a quantitative understanding of developmental systems, a key approach is to measure the sensitivity of the developmental system output to induced variation in an intermediate developmental phenotype. Whether and how this intermediate developmental phenotype varies within and among species then becomes a relevant evolutionary question [1]. The present work addresses the evolution of the expression level of the inducer of vulval development, lin-3, on which we previously performed a sensitivity analysis by manipulating its genetic dosage and addressing the phenotypic consequences for the developmental system [2].
The site and level of transcription of a gene can be modulated both in cis to the gene through cis-regulatory DNA sites directly influencing its transcription, or in trans due to evolution of trans-factors modifying the cellular context in which the gene is acting [3]. cis-regulatory sites containing binding sites for transcription factors often occur upstream of the coding region or in introns. These binding sites are often organized in modules, hence the designation as cis-regulatory modules (CRMs), acting in concert to enhance or repress gene expression in a given tissue at a given time. Changes in the number, relative order, orientation and spacing of transcription factor binding sites can affect transcription, often in a tissue-specific manner [4–6]. Tissue-specificity of CRMs is important for organismal evolution as it is thought to contribute to evolutionary novelty by minimizing pleiotropy [7–12]. Comparative studies in closely related species have revealed that transcriptional regulation can evolve through either extensive rewiring, or quantitative variation in the molecular components of a conserved network [11,13–17]. In particular, changes in cis-regulatory elements directly influencing the expression of critical developmental regulators have been shown to be a driving force for evolutionary innovation and phenotypic novelty in a variety of organisms. One example in Caenorhabditis concerns evolution between C. elegans and C. briggsae in the expression pattern of the transcription factor lin-48 in the excretory system, resulting in a morphological change in excretory cell position. In this case, lin-48 expression was gained in the excretory duct cell of C. elegans due to the acquisition of upstream binding sites for the transcription factor CES-2 [18,19].
Several features now make nematodes excellent experimental systems to understand gene expression evolution. First, rhabditid nematode species present a great advantage because homologous cells are easy to identify [20] so gene expression can be measured in a given cell. Second, the model organism Caenorhabditis elegans and other congeneric nematodes are amenable to functional genetics, transgenesis and now genome editing [21–26]. While transgenesis in C. elegans has long relied on formation of extra-chromosomal arrays containing many copies of the injected DNA that rearrange in an uncontrolled manner [27], the integration of a single copy at a defined locus is now possible, either at the endogenous locus using CRISPR/Cas9-mediated replacements [24–26,28] or at a controlled insertion locus using Mos1-mediated single-copy insertions (MosSCI) [29]. Third, Caenorhabditis species are highly divergent at the molecular level [30,31]. For example, C. elegans is as molecularly distant to C. briggsae as human is to mouse, and C. angaria as far as zebrafish to mouse [31], providing an opportunity to study the turnover of regulatory sequences at a large evolutionary scale where the nucleotide turnover is many times saturated yet the cellular context unchanged [32]. Many new Caenorhabditis species have recently been found and fully sequenced genomes are now available [33,34] (M. Blaxter, pers. comm.). Finally, the recent advent of quantitative methods, such as single-molecule fluorescent in situ hybridisation (smFISH) [35,36], allows to compare gene expression across species. The intensity of the conventional in situ hybridization signal cannot be meaningfully compared among species (regardless of whether the same probes or different probes targeting orthologs are used), while in the smFISH method the number of spots reflecting individual mRNA molecules can be counted, allowing a quantitative study of gene expression evolution.
Here, we take advantage of these recent developments to study the expression and requirement of lin-3, a model developmental gene involved in C. elegans vulval induction. The vulva is the egg-laying and copulatory organ of nematodes, and the C. elegans vulva is now a ‘textbook’ example of animal organogenesis [37]. C. elegans vulval development involves induction of three ventral epidermal cells (P5.p-P7.p) in response to the secretion of the LIN-3 signal from the anchor cell of the somatic gonad. LIN-3 activates the EGF receptor in the vulval precursor cells closest to the anchor cell and thereby acts as the upstream major inducer of vulval fates, in three precursor cells out of the six competent cells (Fig 1A). Induction of vulval fates involves interactions between EGF-Ras-MAPK, Notch and Wnt signalling, including some established pathway crosstalks [38]. We previously showed by modulating lin-3 expression via single-copy transgenesis that the genomic level of lin-3 expression is limited within a four-fold range for the vulva to develop normally in the C. elegans N2 background [2].
The C. elegans lin-3 gene has two alternative promoter regions, each including transcriptional and translational start sites. The lin-3 anchor cell isoform is driven by a specific cis-regulatory module lying immediately 5' of the second promoter, which is located in the first intron of the mRNA driven by the upstream promoter. Within this region, a 59 bp element was shown to be sufficient to drive expression in the anchor cell, acting as a transcriptional enhancer if placed upstream of a minimal promoter [39]. Anchor cell expression was shown to rely on two types of transcription factor binding sites in this 59 bp element, conserved in C. briggsae [39] (Fig 2): an NHR-binding site and two E-boxes. The lin-3(e1417) mutation substitutes a single nucleotide within the NHR-binding site and results in a strong reduction of lin-3 expression in the anchor cell [2,39]. This site can be bound in vitro by nuclear hormone receptors such as C. elegans NHR-25. The two E-boxes surround the NHR-binding site (E-boxL for left to the NHR and E-boxR for right), each consisting of the conserved sequence “CACCTG” but on opposite DNA strands to each other. When either of them is mutated in a lin-3::GFP transgene context, GFP expression in the anchor cell is strongly reduced [39]. We refer for simplicity to the ensemble of these three regulatory elements as the “regulatory triplet”.
We show here that a relative stability in lin-3 mRNA expression in the anchor cell and conservation of LIN-3 vulval induction activity contrasts with the turnover of cis-regulatory binding sites at the lin-3 locus. We show that the difference in requirement of regulatory elements for anchor cell expression is due to evolution in cis to the lin-3 locus without a need to infer evolution in trans. This evolution in cis occurs in a very short 58bp region upstream of the lin-3 vulval specific isoform. This study uncovers the evolution of new cis-regulatory motifs required for cell-specific gene expression.
To determine the level of intraspecific variation in lin-3 expression, we quantified lin-3 expression in different C. elegans wild isolates. In the reference strain N2, a mean level of 25.4 lin-3 mRNA spots was detected using smFISH [2,40] (Fig 1B; S1A Table). We found that the mean and range of lin-3 expression in the anchor cell at the time of vulval induction are comparable between the C. elegans reference strain N2 and the most genetically divergent C. elegans isolates such as DL238 and QX1211 (S1A Fig; S1A Table).
We further explored lin-3 expression in different rhabditid species. First, we searched for the lin-3 ortholog in other available genomes (S2 Fig). The LIN-3 proteins can be aligned along their whole length, with a conserved signal peptide, EGF and trans-membrane domains. Interestingly, the most conserved parts of the proteins are the N-terminal part following the signal peptide and the intracellular domain [41].
We designed smFISH probes for the lin-3 gene of C. briggsae, C. afra, C. angaria and Oscheius tipulae and found that lin-3 is expressed in a single cell within the somatic gonad, immediately dorsal to P6.p, which we identified by DAPI staining as the anchor cell (Fig 1C–1E; S1B Fig; S1B Table). Similar to C. elegans, we also detected lin-3 expression at a lower level in the gonad outside the anchor cell and in the pharynx. We quantified fluorescent spots in the anchor cell and found no significant difference between C. elegans and C. briggsae (mean of 26.5±1 standard error in C. elegans vs. 25±1 in C. briggsae) (Fig 1F). In C. angaria and O. tipulae, we only found a small decrease compared to C. elegans (Fig 1F). Although lin-3 was clearly detected in the anchor cell of C. afra (S1B Fig), the inferior quality of the hybridisation signal compared to the background did not allow us to quantify fluorescent spots in this species. We conclude that despite the great genetic distance between these nematodes [31], the mean number of lin-3 mRNAs is remarkably conserved at least in C. briggsae and may only vary within a narrow range in C. angaria and O. tipulae.
The vulval cell fate pattern is conserved throughout the Rhabditidae family, to which the Caenorhabditis and Oscheius genera belong [42], nevertheless molecular underpinnings of vulval induction in species other than C. elegans remain mostly unknown. lin-3 RNAi experiments in C. briggsae so far produced a weak effect [43]. In Pristionchus pacificus, an outgroup and the only nematode species for which we currently have substantial molecular information related to vulval induction, vulval formation relies on Wnt signalling and is thought to be independent of the EGF pathway [44,45].
To address whether the lin-3 homolog plays a functional role in vulval induction in different Caenorhabditis species, we used a combination of RNAi and pharmacological inhibition. First, we used recently established strains of C. remanei and C. briggsae that are rendered sensitve to RNAi administered by feeding due to the expression of the C. elegans intestinal transporter sid-2 [21,46]. lin-3 RNAi treatment in these C. briggsae and C. remanei strains resulted in substantial reduction in vulval induction (Fig 2A; S3A–S3D Fig). We observed vulval cell fate phenotypes upon lin-3 RNAi that are not found in C. elegans, but are in keeping with published results revealing cryptic variation in vulval fate patterning following anchor cell laser ablations. Specifically, we found that P(5–7).p adopted a 2°-3°-2° cell fate pattern in C. remanei and a 2°-2°-2° pattern in C. briggsae [17,43]. Second, we used the MAP kinase (MEK) inhibitor U0126 that inhibits the downstream signalling events following EGF receptor activation. Application of this inhibitor has been previously shown to decrease vulval induction in O. tipulae [47]. Consistent with this result, we also obtained evidence for loss of overall vulval induction both in C. angaria and C. afra (Fig 1B; S3D Fig). Thus, we conclude that lin-3 is expressed in the anchor cell and plays a conserved role in inducing vulval fates in the Caenorhabditis genus.
Three transcription-factor binding sites, an NHR-binding site and two E-boxes, are required for lin-3 expression in the anchor cell of C. elegans [39]. In light of the conserved expression pattern and level, we wondered whether these regulatory elements required for AC expression of lin-3 are also conserved. The regulatory triplet was found to be present in different species of the Elegans group of Caenorhabditis including C. briggsae (Figs 3, S4 and S5). However, in the sister clade, called the Japonica group, we were able to find the two E-boxes, but no putative NHR-binding site within a window of 2.5 kb upstream of the translational start site of the vulval isoform of lin-3. In further outgroup species, such as C. angaria, we only found a single E-box, and no NHR-binding site in this region. One E-box within the lin-3 CRM was also detected in the outgroup Oscheius tipulae (Fig 3). In C. sp. 1, we were able to detect a single ATG and the first E-box was only found 2 kb upstream. Overall, these observations suggest that the NHR-binding site was acquired in the branch leading to the Elegans group of the Caenorhabditis genus. The evolution of the second E-box at the base of the Caenorhabditis genus remains unclear: the second E-box may have been acquired in the branch leading to the Elegans supergroup or else be lost in the Drosophilae supergroup. No other sequence similarity could be found in the region upstream of the ATG of the vulva-expressed isoform of lin-3 (S4 Fig).
The above results raised an interesting conundrum. How is it possible that some elements that are required for lin-3 anchor cell expression in C. elegans are completely missing in related species, without any significant consequence for lin-3 spatial and quantitative expression?
We first aimed to confirm that one E-box is not sufficient for lin-3 expression in the anchor cell in C. elegans. We used CRISPR-mediated genome editing [48] to select deletions of cis-regulatory elements of the C. elegans lin-3 gene. We generated a variety of alleles, in which either all three elements are deleted (mf90), or NHR and E-boxR are deleted leaving E-boxL intact (mf72-mf74) or only E-boxR is left intact (mf75), the latter recapitulating the cis-regulatory context of the C. angaria lin-3 upstream module (Fig 4A). All these alleles result in fully penetrant vulvaless phenotypes with no cell induced to a vulval fate, thus a stronger phenotype than the lin-3(e1417) allele with one-nucleotide substitution in the NHR binding-site (Fig 4B). We used smFISH to detect lin-3 transcripts and found no lin-3 expression in the anchor cell, which was visualised by the unperturbed expression of lag-2. Interestingly, we still detected lin-3 expression in the gonad of these mutant animals (Fig 4C and 4D). We conclude that these new lin-3 alleles are anchor cell-specific null alleles.
These results confirmed that one E-box in the upstream cis-regulatory module of lin-3 is not sufficient for lin-3 expression in the anchor cell of C. elegans—whereas it appears sufficient in species of the Drosophilae group such as C. angaria.
The evolution in the requirement of transcription-factor binding sites for lin-3 expression in the anchor cell could be due to changes in cis or in trans to the lin-3 locus or both. We reasoned that if differences in trans were important, we would expect lin-3 genomic fragments derived from species missing one or two cis-regulatory elements from the regulatory triplet to be unable to be expressed in the anchor cell of C. elegans. We tested this hypothesis and obtained multiple lines of evidence suggesting no role for changes in trans to the lin-3 locus in explaining the differential binding site requirement.
First, we overexpressed in C. elegans a C. angaria lin-3 genomic fragment containing 200 bp of upstream sequence, the coding region and the 3’ UTR. This fragment drove anchor cell expression of Can-lin-3 and triggered vulval hyperinduction in C. elegans, further showing that the Can-LIN-3 protein could activate the C. elegans LET-23/EGF receptor (S6A and S6B Fig). Vulval hyperinduction was also observed when an equivalent genomic fragment from C. elegans was expressed in C. angaria or a fragment from C. afra was expressed in C. elegans (S6C and S6D Fig). These results indicate that the injected lin-3 fragments from different Caenorhabditis species contain the necessary information for anchor cell-specific expression, despite the fact that a superficially equivalent C. elegans fragment with only one E-box, as in the new lin-3 alleles described above, cannot be expressed in this cell.
Since the regulatory triplet for C. elegans anchor cell expression is missing in these transgenes, we tested whether sequences in the introns, exons or 3'UTR sequences were required for expression of the C. angaria transgene in the anchor cell. To this end, we fused the Can-lin-3 upstream sequences to a fragment containing the C. briggsae lin-3 coding sequence and 3’ UTR. We expressed this fragment in C. elegans N2 and again observed clear expression in the anchor cell. As expected, in control injections containing only the promoterless C. briggsae fragment, the transgene was not expressed anywhere in the body (S7 Fig). To further strengthen these results, we fused the lin-3 cis-regulatory modules amplified from C. elegans, C. briggsae, C. afra and C. angaria to sequences encoding an unrelated protein, the fluorescent protein Cherry, and the unrelated unc-54 3'UTR. In all cases, we observed clear expression in the anchor cell (Fig 5A), indicating again that these short cis-regulatory modules alone contain the necessary information for anchor cell-specific expression in C. elegans. We conclude that evolution within the 200 bp upstream cis-regulatory module of lin-3 is sufficient to explain the difference in requirement of regulatory elements for anchor cell expression within Caenorhabditis.
Above, we used multicopy transgenesis, which may cause sufficient expression and hyperinduction due to summing of weak transcriptional activity of many copies. We thus next asked whether the C. angaria lin-3 fragment had quantitatively a similar activity to that of its C. elegans counterpart when introduced in single copy at a targeted genomic location outside the lin-3 locus (using MosSCI transgenesis, see Methods). We found that a single-copy Can-lin-3 insertion in C. elegans N2 is expressed in the anchor cell (Fig 5B) and does not cause hyperinduction, like an equivalent Cel-lin-3 transgene copy [2]. Most interestingly, this single copy transgene could completely rescue the induction and brood size of lin-3(e1417) mutants, both in homozygous and hemizygous states (Figs 5C, S8). This quantitative behavior of the Can-lin-3 transgene (rescue in the hemizygous and homozygous state, no effect when added to the endogenous locus) recapitulates the activity of a C. elegans copy inserted at the same genomic location [2]. This experiment shows that the C. angaria lin-3 gene driven by its cis-regulatory element acts in a similar quantitative manner to the C. elegans fragment, even in the absence of the regulatory triplet.
To pin down the regulatory elements in the C. angaria transgene that are required for anchor cell expression, we mutated the E-box, which is the only distinguishable regulatory element in this short upstream region. We found that Can-lin-3 genomic fragments with a mutated E-box lose their ability to be expressed in the anchor cell of C. elegans and to trigger vulval hyperinduction when expressed as multi-copy transgenes (Fig 6B, 6D and 6E). This shows that the single E-boxR of C. angaria is necessary for lin-3 expression in the anchor cell of C. elegans.
Changes in the flanking sequences to core binding sites have been shown to contribute to binding efficiency of transcription factors, so we reasoned that perhaps the difference in requirement of regulatory elements for lin-3 expression in the anchor cell may rely on nucleotides adjacent to the single E-box. To this end, we synthesised a chimeric CRM, where a 58 bp central portion harbouring the regulatory triplet in C. elegans was replaced with 58 bp from C. angaria containing E-boxR (Fig 6E). We first showed that this chimeric fragment can be expressed in the C. elegans anchor cell when used in multiple-copy extra-chromosomal array transgenesis (Fig 6C). Furthermore, we used genome editing at the Cel-lin-3 locus to replace the endogenous lin-3 CRM with this chimeric CRM. We found that the genome-edited animals expressed lin-3 in the anchor cell at a normal level and produced a phenotypically wild-type vulva (Fig 6F; S2 Table).
These results demonstrate that the difference in requirement of cis-regulatory elements between C. elegans and C. angaria is explained by compensatory evolution within a very short cis-regulatory fragment (58 bp), rendering the presence of a second E-box and the NHR binding site unnecessary in C. angaria. Despite this loss of transcription factor binding sites, the activity of the cis-regulatory module in driving transcription in the anchor cell remains at the same quantitative level.
The compensation could be explained by the gain of new transcription factor binding sites in the C. angaria 58 bp regulatory region. To identify putative transcription factor binding sites, we performed a motif discovery approach in the anchor cell cis-regulatory lin-3 regions of Caenorhabditis species close to C. angaria and an exhaustive search of transcription factors that could bind the 58 bp sequence (see Methods). We found the GTTTATG sequence, a possible Forkhead-binding site, to be significantly over-represented. This sequence is only one bp to the right of the C. angaria E-box. We tested whether modifying this sequence in the 58 bp C. angaria replacement would change the lin-3 expression level. Indeed, when scrambling these 7 bp (see Methods; S2 Fig), lin-3 expression was reduced significantly to about 60% of the wild-type level (mf95 allele in Fig 6F; t-test, p-value < 6 10−8). However, as expected from a less than two-fold decrease [2], this new replacement, like the intact C. angaria CRM, produced phenotypically wild-type vulva cell fate induction (Fig 6F). Thus, we could affect the expression of the C. angaria CRM by modifying a motif adjacent to the E-box. This motif contributes to the compensation in cis in the 58 bp, but does not explain all of it, as lin-3 expression in the mf95 mutated replacement allele was still much higher than with a single C. elegans E-box.
This study addressed the level of expression of a critical developmental regulator in a single cell. We showed that both lin-3 expression level in the anchor cell and its requirement for the induction of vulval cell fates are conserved in Caenorhabditis and Oscheius nematode species. We found that the mean lin-3 mRNA level in the anchor cell only varies within 30%, despite the vast genetic divergence in this group—corresponding to that found among the most diverged vertebrates. We previously showed using quantitative perturbations that the mean level of lin-3 expression in C. elegans needs to stay within a four-fold range for a correct vulva pattern to arise and that the mean C. elegans N2 level is in the very middle (on a log scale) of this permissible zone. Therefore, it is likely that stabilizing selection acting on vulva formation [49] leads to stasis both in lin-3 expression level and in its effect on vulval induction.
By contrast with this evolutionary stasis in vulval pattern and in the lin-3 mRNA level, we showed that this cell-specific level of lin-3 expression involves substantial turnover of key cis-regulatory elements, namely the appearance of a novel binding site (NHR) and the turnover of a second copy of an existing binding site (E-box). Each of these elements is required for anchor cell expression in C. elegans yet is absent in some Caenorhabditis species. We further focused on the difference in requirement of cis-regulatory elements for lin-3 expression between C. elegans and C. angaria. A 58 bp fragment from C. angaria with a single E-box can replace the three C. elegans binding sites, demonstrating that compensatory evolution within this short cis-regulatory fragment at the lin-3 locus is sufficient to explain this difference in transcriptional regulation
Among evo-devo studies that center on comparisons of gene expression patterns and the evolution of cis-regulatory sequences, this is to our knowledge the first study taking advantage of the latest available capabilities to edit genomes and to quantify the level of mRNA expression at the single-cell level in a multicellular eukaryote.
Gene expression may evolve due to changes in cis or in trans to a given locus, two possibilities that are not mutually exclusive. Cis-regulation may occur from sites quite distant to the transcriptional unit due to long-range chromatin interactions. Our data provided strong support for compensatory cis-changes, and this in a DNA fragment directly upstream of the translational start site of the vulva specific isoform of lin-3. We cannot exclude that some further trans-changes facilitate the difference in requirement of regulatory elements between the two species. However, the cis-regulatory changes that we uncovered in this work are at least sufficient to explain the difference in requirement of regulatory elements for anchor-cell-specific gene expression in Caenorhabditis.
We have narrowed down the compensatory changes that allow the C. angaria lin-3 to be expressed in the anchor cell in a very short region of 58 bp. To explain the compensatory changes, we performed an exhaustive search of transcription factor binding sites and found a putative Forkhead binding site immediately adjacent to the E-box in C. angaria and absent from the replaced 58 bp region of C. elegans. Mutation of this site significantly lowered lin-3 expression, but insufficiently to affect the vulval induction level and it thus only partially explained the compensatory evolution in cis (Fig 6E). We further note that, because this putative Forkhead binding site is immediately adjacent to the E-box, we cannot distinguish between two scenarios: a role for another specific transcription factor binding site versus an alteration of the affinity of the E-box itself. An alternative model would indeed be that compensation occurs through a stronger affinity of the E-box in the C. angaria regulatory region, while the C. elegans E-box is insufficient to drive expression. Such differences in affinity may arise from changes in the sequences flanking the core binding sites as it has been shown for bHLH factors binding to E-boxes [50,51]. Variation in the flanking sequences next to core transcription factor binding sites has also recently been shown to influence both the levels and sites of gene expression for another developmentally important gene [52]. We conclude that the GTTTATG sequence contributes to the compensation, but does not explain it entirely.
Here we described some evolution in cis-regulatory elements that occurs without consequences at the level of gene expression, as observed in many other genes and various groups of organisms [53–56]. This cis-regulatory element turnover in the absence of phenotypic consequence can be viewed as an extension to the notion of developmental systems drift, which posits that distinct molecular mechanisms may underlie the emergence of similar developmental phenotypes [57]. In a similar way, the conservation of gene expression pattern and level may depend on distinct molecular mechanisms due to the loss and gain of binding sites. Indeed, if the invariant output phenotype that we consider is lin-3 expression level in the anchor cell, the molecular events leading to it, such as transcription factor binding, do vary in evolution.
The best-studied example for conservation of gene expression pattern despite turnover of cis-regulatory elements is the stripe 2 enhancer of the Drosophila pair-rule gene even-skipped. The minimal stripe 2 enhancer (eve2) in D. melanogaster is a DNA region of approximately 500 bp that consists of multiple binding sites for activators such as Bicoid and Hunchback and for repressors such as Giant and Krüppel: their combination allows a confined expression in the second stripe along the antero-posterior axis of the early Drosophila embryo [58]. Compared to the described lin-3 cis-regulatory module, the eve2 stripe element involves more transcription-factor binding sites and results in expression in a group of cells (nuclei) rather than in a single cell. Similar to the lin-3 CRM, the transcription-factor binding sites change in Drosophila species in a way that binding sites required for correct expression in D. melanogaster are absent in the stripe 2 element of other species, though without leading to alteration in the expression domain, due to compensatory cis-changes [53,59]. Here we went further in replacing the endogenous cis-regulatory sequences at the locus by those of a distant species, and show a quantitative rescue of gene expression and vulval induction.
One previous example in C. elegans of turnover of binding sites involves lin-48 expression in hindgut cells, which is conserved between C. elegans and C. briggsae despite turnover of EGL-38 upstream response elements [60]. This turnover shows both similarities and differences to the described evolution of lin-3 cis-regulatory elements. The similarity is that there is an increase in the number of EGL-38 response elements in C. elegans. However, in the lin-48 case, there is evolution towards redundancy because the gain in one EGL-38 response element decreases the reliance on the existing element for correct gene expression.
More recently, evolution of cis-regulatory elements between C. elegans and C. briggsae has been studied by placing exogenous cis-regulatory elements from C. briggsae into C. elegans. A main result over several genes whose expression is conserved between the two species is the appearance of ectopic gene expression domains in these transgenic experiments, implying evolution both in cis and in trans [61,62]. In one case, the ability of the unc-47 proximal promoter from C. briggsae to drive ectopic expression in some C. elegans neurons was mapped next to a conserved cis-regulatory motif [61].
We note that the C. angaria fragment conveys the same level of transcriptional activity yet that a few vulval cell fate patterning "errors" occur in the replacement lines (Fig 6F). We observed both hypoinduced and hyperinduced variants in each of the two replacement lines (S2 Table), but the very low frequency of these variants make them difficult to study quantitatively. In the case of the eve2 enhancer, the minimal stripe element is embedded within a larger region of approximately 800 bp, and these flanking sites contribute to robustness to some genetic and environmental perturbations [63]. In Caenorhabditis, the distal promoter of unc-47, although largely not conserved, is also important for robust gene expression, acting perhaps in a sequence-independent manner [64]. It remains unclear whether any regions within and/or outside the lin-3 CRM can play a similar role in stabilizing expression of lin-3 in Caenorhabditis to different perturbations.
The distribution of lin-3 cis-regulatory elements in different Caenorhabditis nematodes and the mapping of changes on the phylogeny suggests as the most likely evolutionary scenario a gain of regulatory sites: the likely acquisition of an E-box before the common ancestor between the Elegans and Japonica groups and a gain of an NHR-binding site before the origin of the Elegans group. In addition, these sites not only appeared, but also became indispensable for lin-3 anchor cell expression at least in C. elegans.
The acquisition of such new short regulatory motifs (6 bp) is easy and gains of regulatory motifs have been observed in other systems as well [65]. Given the high robustness of vulval development to several perturbations, the evolution towards a dependence on a higher number of sites for anchor cell expression is counter-intuitive and suggestive of evolution towards fragility. It is currently unclear what drove the evolution of these novel motifs with a conserved gene expression, whether selection or drift. Gains in interconnectedness between components of transcriptional networks may often occur non-adaptively, for example if they do not disrupt the underlying regulation [66]. Such gains can also be reshaped in equivalent network configurations and eventually become necessary depending on the evolution of the transcriptional network [67].
A complete list of strains used in this study is presented in the supplement (S3 Table). All strains were maintained at 20°C and handled according to standard procedures [68]. We used the Bristol N2 strain as a reference C. elegans strain on standard NGM plates with OP50 as a food source. The U0126 treatments were performed by supplying the DMSO-dissolved inhibitor to NGM plates at a concentration between 10–150 μM and letting synchronised L2 stage nematodes develop into L4 larvae. Control treatments in this case were performed by growing nematodes on plates supplemented with DMSO only.
For the Can-lin-3 rescue of the C. elegans lin-3(e1417) mutant, JU2495 hermaphrodites were crossed to JU2498 males and the F1 or F2 progeny were analysed for hemizygous or homozygous insertion phenotypic rescue, respectively.
The lin-3 genomic sequences of the different species were accessed in WormBase (www.wormbase.org; version WS252) or from the Caenorhabditis Genomes Project by Mark Blaxter's laboratory (http://bang.bio.ed.ac.uk:4567) or from Matt Rockman’s laboratory. The Oscheius tipulae genome was sequenced and assembled as a collaborative effort between M. Blaxter's and our lab (Besnard, Kotsouvolos et al., in preparation) and is available (http://oscheius.bio.ed.ac.uk/). We first used the TBLASTN algorithm conditioning only to the most identical hits, favouring those with high similarity in the N-terminal part and signal peptide, and lower e-value. Afterwards, we proceeded to predict gene bodies in these contigs using FGNESH (http://www.softberry.com) with a hidden Markov model specific to C. elegans. Finally, manual curation and annotation of the lin-3 sequences were performed using as a reference the amino-acid sequence of the closest available lin-3 ortholog.
To study the evolution of the regulatory triplet in the Caenorhabditis clade, we analysed the promoter regions upstream of the downstream ATG corresponding to the N-terminal exon homologous to that known to be expressed in the AC of C. elegans (S2 Fig). First, to address whether the cis-regulatory C. elegans NHR-binding sites and E-boxes were present in the other species, we performed a scan in the promoters with the position weight matrices of HLH-2 and NHR-proteins available in JASPAR [69] using matrix-scan [70] and a n = 2 Hidden Markov Model specific to C. elegans (Fig 3). Similarly, we looked in these regions for DNA patterns known to be binding sites of bHLH proteins [51] using the dna-pattern tool present in the RSAT suite [71]. Once we had the position of these sites across the promoter regions, we proceeded to plot their location using RSAT feature-map tool (S5 Fig).
Additionally, we looked for DNA motifs different from the cis-regulatory C. elegans NHR and E-boxes binding sites by performing a motif-discovery approach in Caenorhabditis lin-3 promoters using the RSAT tool oligo-analysis [71]. The top over-represented words of length 6, 7 and 8 base pairs were compared to known motifs available in JASPAR. We thus identify the GTTTATG to the right of the E-box. Finally, to identify possible transcription factors acting on the AC lin-3 expression in the 58 bp C. angaria fragment, we performed an exhaustive search of the full JASPAR motif repertoire in the 58 bp replaced sequence using RSAT matrix-scan. This search found the putative Forkhead-binding motif and a putative overlapping bZIP-binding motif (Fos/Jun repressors). The 7 bp modification in the mf95 replacement also affected this predicted binding site of bZip transcription factors.
All lin-3 CRMs reside directly upstream of the ATG of the vulval isoform of lin-3. To create the lin-3 CRM::Cherry::unc-54 constructs, we used a three-fragment Gateway approach merging the lin-3 CRMs cloned in pDONOR P4-P1R, the Cherry ORF cloned in pDONOR 221 and the unc-54 3’UTR cloned in pDONOR P2R-P3. All primer sequences containing attB4 forward and attB1 reverse recombination sites used to amplify the CRMs from gDNA from different species are shown in S4 Table.
unc-54 3’ UTR was amplified from N2 genomic DNA using primers unc-54attB2 and unc-54attB3. Worm-optimised Cherry was amplified from pAA64 using primers containing the attB1and attB2 sites. All constructs were injected at 10 ng/μl with myo-2::GFP as co-injection marker and pBluescript as carrier DNA.
To create the Can-lin-3 insertion by MosSCI, we amplified a 2.9 kb lin-3 fragment from C. angaria genomic DNA using primers Canlin-3AvrII and Canlin-3XhoI. The amplicon was cloned into pCFJ151 (chromosome II targeting vector) [29] as an AvrII/XhoI fragment. Injections and recovery of insertions were performed using the direct insertion protocol, as previously described.
To overexpress lin-3 fragments in C. elegans or C. angaria, we amplified genomic fragments amplified from C. elegans (5.2 kb), C. angaria (3.2 kb) and C. afra (5.1 kb) using primer pairs RH9for/RH9rev, Canlin-3F2/Canlin-3R1 and Caflin-3oxF2/Caflin-3oxR1, respectively. The PCR products were injected directly (30 ng/μl) together with pBluescript as carrier and myo-2::GFP as co-injection marker.
To mutagenize the E-box in the C. angaria lin-3 CRM, the above 3.2 kb fragment was cloned into pGEM-Teasy and the 5’-CAGGTG-3’ sequence was modified to 5’-CAGGAA-3’ using primers t211a_g212a/ t211a_g212a_anti and standard in vitro site directed mutagenesis.
The chimeric construct replacing a 58 bp region containing the C. elegans regulatory triplet (5’-cacctgtgtattttatgctggttttttcttgtgaccctgaaaactgtacacacaggtg-3’) with a similar in length sequence from C. angaria containing only one E-box (5’-attttttgtcaaagatttttcggcgccaggtgtgtttatgactcatgttagggccgag-3’) was synthesised by Genewiz. This construct was used as PCR template to permute 7 bp to the right of the C. angaria E-box (5’-CAGGTGtGTTTATG-3’ to 5’-CAGGTGtTTGGATT-3’).
The chimeric construct to drive Cbr-lin-3 under the C. angaria CRM was built using fusion PCR. Briefly, the Can-lin-3 CRM was amplified from C. angaria genomic DNA with primers Canlin-3 F2 and CaACFusion and the Cbr-lin-3 region coding region and 3’ UTR from C. briggsae genomic DNA with primers Cbrlin-3F1 and Cbrlin-3R1. The two amplicons were then fused together using a third PCR reaction with primers Canlin-3F2 and Cbrlin-3R1. The final product was injected as a PCR fragment at 20 ng/μl concentration.
smFISH was performed in synchronized populations of L3 stage animals using short fluorescently labelled oligos as probes, as previously described [2]. The animals were age-synchronized by bleaching, followed by hatching of embryos in M9 buffer. The L1 larvae were then placed onto culture plates with food until the L3 stage, as determined by Nomarski microscopy, and then fixed.
The C. elegans lin-3 and lag-2 probes have been previously described [2]. The low level of genetic divergence within C. elegans allowed us to detect fluorescent spots while using the same FISH probe as in the N2 strain.
For all other species we followed the same protocol as with C. elegans with the following two modifications to decrease the more pronounced background fluorescence. We used 20% formamide in the hybridisation and wash solutions and performed three washes post-hybridisation instead of two in C. elegans. Given that we are using different probes consisting of fewer oligos for the detection of lin-3 transcripts in these species together with slightly more stringent hybridisation conditions, the observed difference in the number of fluorescence spots may thus even be due to technical rather than biological reasons. The sequences of the new lin-3 probes can be found in S5 Table. The probes were labelled with Quasar 670 (Biosearch Technologies) and diluted to 100–200 nM for the overnight hybridisation.
RNAi was performed by feeding the animals with dsRNA-expressing bacteria, as previously described [2]. The C. elegans lin-3 RNAi feeding clone used in this study is from the Ahringer RNAi library (Source Bioscience). A Cre-lin-3 fragment was amplified using oligos Crelin-3RNAiF1 and Crelin-3RNAiR1 that contain an XhoI restriction site. The PCR product was cloned into L4440 as an XhoI fragment. To create the C. briggsae lin-3 RNAi clone, a fragment was amplified using primers Cbrlin-3RNAiF1 and Cbrlin-3RNAiR1 and then cloned into pDONR 221 (Invitrogen) using attB1F and attB2R universal oligonucleotides. The lin-3 fragment was sequence verified and transferred to a Gateway compatible L4440 plasmid. Both constructs were transformed into E. coli HT115 for use in C. elegans feeding.
To score the vulval cell fate pattern, nematodes were mounted with M9 on 3% agar pads containing 10 mM sodium azide and analysed under Nomarski optics. Standard criteria were used to infer cell fates based on the topology and number of cells at the L4 stage [43,72]. Half fates were assigned when two daughters of the Pn.p cells acquired distinct fates after the first cell division.
We followed the CRISPR/Cas9 target design and used reagents as previously described [48]. We targeted the following region at the C. elegans lin-3 CRM 5’-accctgaaaactgtacacacAGG-3’ with AGG representing the PAM motif. We replaced the unc-119 target site under the pU6 promoter [48] with the lin-3 target site using fusion PCR first with primers E-box2A gRNA-F/ U6prom HindIII and E-box2A gRNA-R/ oligos U6prom EcoRI F followed by amplification of the full sgRNA fragment with U6prom EcoRI F/ U6prom HindIII R. The only modification was that we did not clone the lin-3 sgRNA in a vector but injected it directly as a PCR product (40 ng/μl, together with 40 ng/μl eft-3::Cas-9 and myo-2::GFP as co-injection marker).
To replace the endogenous lin-3 cis-regulatory element of C. elegans by a 58 bp lin-3 element from C. angaria, we first obtained a chimeric double-stranded DNA as homologous recombination template, using Gibson assembly of C. elegans lin-3 promoter extremities with 58 bp of the C. angaria lin-3 upstream sequence. In a similar fashion, we obtained a homologous recombination template identical to the previous but with modified bases next to the C. angaria E-box. Oligonucleotide sequences are found in S4 Table. C. elegans N2 animals were injected with a DNA mix containing the Peft-3::Cas9 plasmid, the pU6::dpy-10 sgRNA plasmid (co-CRISPR marker), the Ebox-2A sgRNA containing plasmid and the double-stranded DNA repair templates (independently), with final concentrations of 50, 40, 100, and 30 ng/μl, respectively. On plates with a high number of animals displaying the Dpy phenotype, the F1 progeny were singled, and their progeny screened by PCR. Broods from independent P0 animals were found positive and rendered homozygous (two independent lines for each replacement). Both replacements were confirmed by Sanger sequencing. The resulting lines were given allele names mf91 and mf92 for the first replacement, and mf95 and mf112 for the second one.
|
10.1371/journal.pgen.1000120 | Mechanisms of Cell Cycle Control Revealed by a Systematic and Quantitative Overexpression Screen in S. cerevisiae | Regulation of cell cycle progression is fundamental to cell health and reproduction, and failures in this process are associated with many human diseases. Much of our knowledge of cell cycle regulators derives from loss-of-function studies. To reveal new cell cycle regulatory genes that are difficult to identify in loss-of-function studies, we performed a near-genome-wide flow cytometry assay of yeast gene overexpression-induced cell cycle delay phenotypes. We identified 108 genes whose overexpression significantly delayed the progression of the yeast cell cycle at a specific stage. Many of the genes are newly implicated in cell cycle progression, for example SKO1, RFA1, and YPR015C. The overexpression of RFA1 or YPR015C delayed the cell cycle at G2/M phases by disrupting spindle attachment to chromosomes and activating the DNA damage checkpoint, respectively. In contrast, overexpression of the transcription factor SKO1 arrests cells at G1 phase by activating the pheromone response pathway, revealing new cross-talk between osmotic sensing and mating. More generally, 92%–94% of the genes exhibit distinct phenotypes when overexpressed as compared to their corresponding deletion mutants, supporting the notion that many genes may gain functions upon overexpression. This work thus implicates new genes in cell cycle progression, complements previous screens, and lays the foundation for future experiments to define more precisely roles for these genes in cell cycle progression.
| All cells require proper cell cycle regulation; failure leads to numerous human diseases. Cell cycle mechanisms are broadly conserved across eukaryotes, with many key regulatory genes known. Nonetheless, our knowledge of regulators is incomplete. Many classic studies have analyzed yeast loss-of-function mutants to identify cell cycle genes. Studies have also implicated genes based upon their overexpression phenotypes, but the effects of gene overexpression on the cell cycle have not been quantified for all yeast genes. We individually quantified the effect of overexpression on cell cycle progression for nearly all (91%) of yeast genes, and we report the 108 genes causing the most significant and reproducible cell cycle defects, most of which have not been previously observed. We characterize three genes in more detail, implicating one in chromosomal segregation and mitotic spindle formation. A second affects mitotic stability and the DNA damage checkpoint. Curiously, overexpression of a third gene, SKO1, arrests the cell cycle by activating the pheromone response pathway, with cells mistakenly behaving as if mating pheromone is present. These results establish a basis for future experiments elucidating precise cell cycle roles for these genes. Similar assays in human cells could help further clarify the many connections between cell cycle control and cancers.
| The budding yeast Saccharomyces cerevisiae undergoes a cell cycle similar to other eukaryotic organisms except for the lack of nuclear envelope dissolution during mitosis and the production of daughter cells via budding, and thus budding yeast has become a model system for studying eukaryotic cell cycle progression [1] due to its rapid division, the availability of genetic tools, and homology to higher eukaryotic cell cycle processes. Numerous genes and proteins are involved in directing cells through the 4 major cell cycle phases, the growth gap phase G1, the DNA synthesis (S) phase, a second growth gap phase G2, and the mitotic (M) cell division phase [2],[3]. Extensive effort has been made to decipher the mechanisms of cell cycle control. However, given the extreme complexity of the cell cycle, with ∼300–800 genes regulated in a cell cycle-dependent manner [4]–[6], the complete set of cell cycle regulators, effectors, and helper proteins has yet to be determined.
Classically, conditional temperature-sensitive mutants have been very effective for studying yeast cell cycle division. Hartwell and colleagues identified more than 50 cell division cycle (CDC) genes required at specific stages in cell cycle division, by identifying conditional temperature-sensitive mutants with specific arrest points [7]–[10]. Gene dosage has been another powerful approach to study gene function. Either increasing (overexpression) or decreasing gene dosage (gene deletion or gene knockdown) can influence the activity of genes and lead to detectable phenotypes. Most large-scale cell cycle screens have focused on studying cell cycle progression by employing loss-of-function approaches such as gene deletion, RNAi, and promoter shutoff [11]–[13] and have successfully identified many cell cycle genes. However, loss-of-function mutations can often be masked, such as in the cases of genes acting as negative regulators or genes compensated for by redundant functions [14]–[16]. In contrast, overexpression of a gene product can potentially overcome such effects and often leads to a more detectable effect on cellular function [16]. Overexpression also offers the opportunity to identify and study gain-of-function mutations.
In order to identify additional cell cycle genes, especially those difficult to identify in loss-of-function studies, large-scale screens focusing on the effects of overexpression-induced gain-of-function of genes in cell-cycle progression are needed. Stevenson et al. performed the first such large-scale overexpression screen for cell cycle genes by expressing a moderated GAL promoter-driven cDNA library and sheared genomic DNA pool in ARS-CEN vectors [17]. Although 113 genes, including those causing only slight effects on the cell cycle, were identified from this screen, this screen was unsaturated due to the coverage of the cDNA library and incomplete gene annotation. Therefore, completion of the S. cerevisiae genome sequence and the systematic cloning of all genes into overexpression vectors now allow a more comprehensive analysis of the set of genes.
Analysis of overexpression phenotypes using cell sorting to assay the distribution of cells in different cell cycle stages has the advantage of being more quantitative and discerning than simple growth screens. However, flow cytometry has not been carried out comprehensively to cover all genes in the genome. In the present work, we performed a near-saturating screen for yeast genes having overexpression-induced defects in cell cycle progression, taking advantage of the availability of a yeast open reading frame (ORF) clone collection covering 91% of the yeast complete ORF set, including dubious ORFs [14]. After measuring the fraction of cells in different phases of the cell cycle via high-throughput flow cytometry for each of 5,556 individual ORFs and performing secondary validation assays, we identified 108 genes whose overexpression leads to significant changes in the timing of passage through the G1 or G2/M stages of the cell cycle. 82 of these genes are newly implicated in the cell cycle, with the majority likely to affect cell cycle progression via gain-of-function mechanisms.
The yeast ORF collection was obtained from Open Biosystems, in which each ORF was cloned into a 2μ plasmid under control of the GAL1 promoter in order to provide highly elevated expression when supplemented with galactose [14]. Control strains were constructed by transforming the empty precursor vector BG1766 to the ORF host strain Y258 (MATa pep4-3, his4-580, ura3-53, leu2-3,112) and plating on synthetic complete medium lacking uracil.
The plasmid PGAL1-SKO1 was also transformed into ste2Δ, ste4Δ, ste5Δ, ste20Δ, ste11Δ, fus3Δ, far1Δ, fus1Δ, kar4Δ, sst2Δ, dig2Δ deletion strains [18] (ResGen/Invitrogen) and a Fus1-GFP strain [19] ( Invitrogen), as well as their parent strain BY4741 (MATa his3Δ leu2Δ met15Δ ura3 and then plated on synthetic complete medium lacking uracil.
Yeast ORF strains were induced in parallel with the corresponding empty vector (BG1766) control strain. Cells were initially grown in 96-well plates (Corning 3595) with 170 µl SD-URA medium for 1–2 days at 30°C, and then 5 µl cells were inoculated into fresh 96-well plates with 170 µl SC-URA, 2% raffinose medium. After 12 hours growth in raffinose medium, cells were re-inoculated to fresh plates with 100 µl SC-URA, 2% raffinose medium at a final O.D.600nm of 0.15 and grown for 1 hour. 70 µl SC-URA medium with 5% galactose (final concentration 2%) was added, and cells were grown for 8–10 hours at 30°C.
Flow cytometry analyses were performed as in [20]. Briefly, ∼2×106 cells were harvested and fixed in 200 µl 70% ethanol, treated with 1mg/ml RNAse A (Sigma) for 4 hours at 37°C, then incubated with 1mg/ml Proteinase K (Sigma) for 1 hour at 50°C. ∼8×105 cells were then resuspended in 200 µl 50 mM sodium citrate with Sytox green (Invitrogen) at a final concentration of 1.5 µM, performing the above liquid transfers using a Biomek FX robot (Beckman Coulter). Samples were analyzed by flow cytometry, using a Becton Dickinson FACSCalibur with BD HTS auto sampler, controlled by Plate Manager and Cellquest pro software (BD Biosciences). Well-to-well contamination was minimized by flushing with ddH2O between each pair of samples. In order to maximize measured events while minimizing data collection time for 5,556 strains, we collected the shorter of either 20,000 events/strain or 30-seconds acquisition time/strain. Thus, for the extremely slow growing strains, the number of events collected in 30 seconds may drop below 20,000 events.
Analysis of DNA profiles was automated using ModFit 3.0 software (Verify Software house, Inc), fitting the histograms of 1C and 2C cells with Gaussian distributions (Figure 1C) and calculating the goodness-of-fit via the Reduced Chi Square (RCS) method. For quality control, DNA profiles with RCS>5 and event number<5000 were discarded. Empirically, we observed the resolution of the S phase cell distribution to not be of sufficiently high quality to merit systematic analysis; we thus focused instead on the well-resolved G1 and G2/M phase cells. The percentage of cells under each DNA peak (1C peak or 2C peak) was calculated by dividing the number of events under each peak by the total number of events under all peaks, and the ratio (1C/2C) of the percentage of cells under the 1C peak to that under the 2C peak was calculated for each strain. The base 2 logarithm of the 1C/2C ratio was calculated for each strain; the distribution of Log2 (1C/2C) values (abbreviated LR below) was fit well by a Gaussian distribution (R2 = 0.97) (Figure 2A), allowing each ORF strain i to be assigned a Z-score, calculated as (LRi−<LR>)/σLR.
Additionally, we manually categorized strains as diploid and 3C: 208 strains appeared diploid (e.g., had 2C and 4C peaks, rather than 1C and 2C) based upon the flow cytometry data and 56 strains showed notable 3C peaks and were assigned into the 3C category. Follow-up validation of these trends showed that the DNA content of these strains did not change upon galactose induction, suggesting these to be artifacts of these strains rather than an inducible effect of gene overexpression, and thus these strains were not studied further. These strains are listed in Table S6.
108 ORF strains showing reproducible cell cycle arrest were grown and induced as described above. After induction, cells were fixed in 70% ethanol and treated with 1mg/ml RNAse A (Sigma), and then stained with 1 µM Sytox green (Invitrogen). Cells were examined via phase contrast microscopy and fluorescence microscopy using a Nikon Eclipse 800 fluorescence microscope. From differential interference contrast (DIC) images, we used ImageJ software (National Institute of Mental Health) to measure the length of the bud and mother cell for an average of 100 cells for each of the 108 strains. Bud size was assigned by dividing the bud length by the length of mother cell. Cells with a ratio of 0 were classified as ‘no bud’; cells were categorized into ‘small bud’ when the ratio was between 0 and 0.4, and ‘large bud’ when the ratio was higher than 0.4 [7]. We further examined the large-budded cells and counted three types of nuclear morphology: an undivided nucleus in one cell body (class I), an undivided nucleus in the bud neck (class II), and divided nuclei in two cell bodies (class III) [21]–[23]. An average of 50 cells was counted for each of 87 G2/M strains.
The 77 of 82 genes not previously implicated in cell cycle defects (and 3 positive controls, TUB2, PAC2, and CST9) were assayed for growth defects in three conditions: SC-URA, 2% galactose; SC-URA, 2% galactose plus 15 µg/ml nocodazole, and SC-URA, 2% galactose plus 50 µM hydroxyurea [15],[24]. 4 dubious ORFs (YLL066W-B, YBR131C-A, YLR123C, YJL077W-A) were not included in the growth assays, as well as one gene (PPZ1) in the G2/M category. Cells were grown overnight in SD-URA medium, and then washed with SC-URA, 2% raffinose medium and grown in SC-URA, 2% raffinose medium for one hour at 30°C before being spotted onto agar plates. Six 10-fold serial dilutions were made for each strain, with the O.D.600nm of the first series at 0.2. 10 µl of each series was spotted onto SC-URA, 2% galactose plates and SC-URA 2% galactose plates containing the appropriate drugs, and grown at 30°C. Plates were photographed after 2–3 days growth in SC-URA, 2% galactose plates, or 5–8 days in the plates supplemented with drugs.
The plasmids PGAL1-YPR015C and pRS412::ADE2 [cir+] were transformed into the strain Cry1 (MATa ade2-1, ura3-1, leu2-3, 112, trp1, his3-11), plating transformants on synthetic complete medium lacking uracil and adenine. A single colony was picked and diluted in ddH2O. ∼104 cells were inoculated into SC-URA, 2% galactose medium and grown for 10 generations at 30°C, before plating ∼200 cells on a YPD plate. After growing 2–3 days at 30°C, plates were shifted to 4°C to maximize the color changes. Red and white colonies were counted, where red colonies have lost the centromere-containing plasmid and white colonies have retained it.
The SKO1 overexpression strain was induced in parallel with the corresponding empty vector (BG1766) control strain with 2% galactose in selective medium for 8 hours, as described above. Total RNA isolation and processing, microarray hybridization, and data analysis were performed as described previously [25], hybridizing RNA isolated from the SKO1 ORF strain against RNA from the empty vector control strain. For each strain, two biological replicates were analyzed, each by two technical (array) replicates. Differentially expressed genes were selected as having a minimum expression ratio (corresponding to the absolute value of Log(base2) of R/G normalized ratio (Median)) > = 1.5 for at least 2 arrays. The significance of differential expression was calculated using the error model of Hughes et al. [26].
Yeast cells were induced 8 hours, then fixed in growth medium with 1/10 volume 37% formaldehyde for 1 hour at 30°C. Fixed cell cultures were spheroplasted with 0.025 mg/ml zymolyase 20T (Seikagaku corporation) for 1 hour at 30°C. Cells were then spotted onto poly-L-lysine coated microscope slides. Cells on the slide were permeablized in −20°C methanol for 6 minutes, followed by −20°C acetone for 30 seconds. Cells were blocked with 3%BSA in PBS for 30 minutes at 30°C in a humid chamber, followed by incubation with 4 µg/ml mouse anti alpha-tubulin monoclonal primary antibodies (Invitrogen) for 1 hour and 4 µg/ml Texas Red conjugated goat anti-mouse secondary antibody (Invitrogen) for 2 hours at 30°C. After washing three times with PBS, cells were mounted with 60 µl VECTASHIELD hard set mounting medium with 1.5 µg/ml DAPI (Vector Laboratories, Inc), and imaged at 100x magnification with a Nikon Eclipse 800 microscope.
To analyze the effect of overexpression of yeast genes on cell cycle progression, we applied high-throughput flow cytometry to screen 5,556 strains of a yeast ORF collection [14] for genes that induce delay or arrest at particular cell cycle stages when overexpressed. Figure 1 outlines the overall approach. Excess accumulation of cells with either one copy (1C) or two copies (2C) of DNA content indicates a defect in progression through a particular cell cycle stage (G1 or G2/M, respectively). Thus, in order to search for such defects induced by overexpression of a particular yeast gene, we analyzed asynchronous cell cultures and determined the distributions of DNA content, assaying if cells from each given ORF overexpression strain exhibited a skewed distribution relative to control cells. In all, ∼5,700 DNA histograms were acquired and quantitatively analyzed, measuring the ratio of 1C/2C cells for each strain, i.e., the ratio of cells in the G1 phase to cells in the G2/M phase. We observed the Log2 (1C/2C) ratios of the 5,556 ORF strains and of 139 replicate analyses of control strains to be approximately normally distributed and well-fit by a Gaussian distribution (R2∼0.97) (Figure 2A). Therefore, for each strain, we calculated a Z-score for its distribution of DNA content across cells and could thus identify the ORF strains with significantly higher accumulations of cells in the G1 or G2/M growth phases. Based on this Z-score, 2 categories were assigned: ORF strains with Log2 (1C/2C) ratios in the left tail of the Gaussian distribution were considered to have significant G2/M delays, in which cells accumulated with two copies of DNA. Similarly, ORF strains with Log2 (1C/2C) ratios in the right tail of the distribution showed significantly higher proportions of cells with one copy of DNA, and were considered to exhibit G1 delays. Examples are shown in Figure 1C. We could assign genes to the G1 and G2/M categories using different confidence levels (Figure 2B). At the 95% confidence level, 198 genes were identified whose overexpression caused cell cycle detects; only 3 of 139 control strains exceeded this threshold. As the large-scale screen was based upon only a single culture per ORF strain, we further selected those strains with reproducible defects. Of the 198 strains, 108 were validated at least twice by manual flow cytometry analysis (DNA histograms are shown in Figure S1). Additionally, we tested that all 108 genes identified showed cell cycle delay phenotypes only upon induction in galactose, and that the phenotype for each hit therefore derived specifically from the GAL-promoter-driven gene. Of the 108 genes, 21 caused a significant accumulation of cells in the G1 phase, 87 genes in the G2/M phase. These genes are listed in full in Table S1.
The size of the bud relative to the size of the mother cell is the most notable morphological landmark of the cell cycle stages in budding yeast. Bud size was the basis of classical cell cycle screens [7]–[10],[27],[28], allowing the identification of mutants blocked at specific stages of the cell cycle: DNA replication occurs when bud size is small, nuclear division occurs when the bud is about three-fourths the size of the mother cell, and cell separation when the bud is approximately equal in size to the mother cell. In order to independently validate genes in the G1 and G2/M categories using bud size, we measured the ratio of bud size to mother cell size for the 108 ORF strains identified by flow cytometry as having cell cycle defects. Genes in the G1 category caused clearly elevated populations of unbudded cells when overexpressed, and the 20 of 21 genes in the G1 category tested for bud size all exhibited a higher percentage of unbudded cells than control strains (Figure 3A), with 12 being more than 2 standard deviations higher than controls, as shown in Figure 3A. For example, 92% of cells were unbudded and only 2% of cells were large-budded when TRM5 was overexpressed. In contrast, only 57% of wild type cells were unbudded, and 28% were large-budded (Figure 4B, Table S1). Of 87 strains in the G2/M category, 85 exhibited a higher percentage of large-budded cells than control strains (Figure 3B). For instance, at least 60% of cells had large buds when TUB2 and SPC97 were overexpressed (Table S1). Consistent with previous observations, TRM5, TUB2 and SPC97 are known to cause cell cycle delays when their normal function is perturbed [13],[21],[29],[30]. SPC97 is an example of the successful recovery of genes known to be important for the cell cycle; it encodes a structural constituent of the spindle pole body, and performs a key role in mitotic spindle formation. 47 strains in the G2/M category had proportions of large-budded cells more than two standard deviations higher than controls, as shown in Figure 3B. Bud size analysis thus provided a useful independent validation of the DNA content observations, with genes validated by both flow cytometry analysis and bud size distributions being the most likely to affect cell cycle progression.
One major expected cause of defective cell cycle progression is chromosome instability, especially chromosome loss and non-disjunction. Chromosome loss is characteristic of defects in DNA metabolism, while non-disjunction typically reflects defects in mitotic segregation [15]. To help address which chromosomal functions were primarily affected by the overproduction of the identified ORFs, we examined the strains' sensitivities to hydroxyurea and nocodazole. Hydroxyurea (HU) is an inhibitor of ribonucleotide reductase, an enzyme necessary for DNA synthesis. Nocodazole (NOC) is a microtubule depolymerizing drug that prevents formation of the mitotic spindle. Genes involved in DNA metabolism and the DNA replication checkpoint are often sensitive to HU, whereas genes sensitive to microtubule drugs are often involved with the mitotic checkpoint and mitotic spindle formation [15]. Due to the presence of the spindle checkpoint control, yeast mutants affecting spindle structure normally show cell-cycle arrest in mitosis [31]. We tested the 77 genes potentially newly implicated in the cell cycle for their sensitivity to HU and NOC separately. In the absence of the drugs, we observed all but 4 tested strains (all but IMG1, DHR2, GPT2, and YGR109W-A) to show strong growth defects indicative of toxicity of the overexpressed proteins. A semiquantitative score for growth defects, from 0 (no defect) to 3 (strong defect), shows the 77 strains have an average defect of 2.5. Beyond this intrinsic toxicity, we observed 22 strains to be specifically sensitive to NOC, 6 to be specifically sensitive to HU, and 13 strains to show sensitivity to both (Table S4 and Figure S2). As expected, TUB2 and PAC2 exhibited the non-disjunction-relevant phenotype, sensitivity to nocodazole but not hydroxyurea; TUB2 and PAC2 are required for normal microtubule function and mitotic sister chromatid segregation [21],[24]. We might expect that genes in the same category as TUB2 and PAC2 might be directly or indirectly involved in microtubule function or functions related to chromosome segregation, consistent with nearly all (21 of 22) genes having increased sensitivity specifically to NOC arresting at the G2/M phase when overexpressed.
We examined in more detail the functions for the 108 genes that caused cell cycle defects when overexpressed. Among these genes, 26 are known to be involved in different aspects of cell cycle progression, 21 are essential ORFs, 17 are transcription factors, 20 ORFs are uncharacterized, and 4 are dubious ORFs (Table S2 and Figure 2D). Importantly, of the 26 genes identified in the screen that were previously known for having cell cycle defects, 24 were consistent with the previously observed phenotypes. Of 8 Cdc28p cyclins included in the ORF collection, we recovered 5 (CLN1, CLB2, CLB3, CLB5, and CLB6). A number of known essential genes cause cell cycle defects when down-regulated [13]; we recovered 67% of these genes in this screen. These observations validate the general quality of the current screen by indicating that cell cycle defects caused by overexpression of these 108 genes do not generally result from random effects of overexpression, but rather the 108 genes are strongly enriched for known regulators of the cell cycle.
We tested to see if the 108 genes were cell cycle regulated or showed obvious expression level biases. They do not appear to be cell cycle regulated, as the set of 108 hits is not significantly enriched for cell-cycle regulated genes as measured by Spellman et al. [6] (p>0.05, hypergeometric probability). Analysis of the overexpression levels of the genes show typical induction by 5- to >15-fold over the native expression levels, for proteins of both low and high native levels (Figure S3). We analyzed the distribution of steady state native expression levels of proteins identified in this screen, and do not observe a significant bias in the native levels of the hits; the median expression level of the proteins we identified, measured in rich medium [32], is 2025 copies per cell, versus 2250 copies per cell expected (for all proteins).
We also compared the 108 genes with those previously identified by Sopko et al. [16] and Stevenson et al. [17] and observe a significant (p<0.05, hypergeometric probability) but small overlap, with 15 of the 108 genes observed previously and 93 new to this study (Figure S4). Genes observed in at least two of the three assays are strongly statistically enriched for direct regulators of the cell cycle (e.g., the cylins CLB3, CLB2, and CLB5, and components of the spindle pole body BIM1, TUB2, SPC42, SPC98, KAR1). Analysis of enriched functions (using Funspec [33]) among genes observed in ≥2 assays reveals the most strongly enriched functions also relate to the cell cycle, with the strongest enrichment observed for the MIPS annotations “cell cycle and DNA processing” (p<10−7), “cell cycle” (p<10−6), and “mitotic cell cycle and cell cycle control” (p<10−6).
In the next two sections, we describe the G1 and G2/M genes in more detail.
The 87 G2/M genes showed dramatic enrichment in cell cycle-related Gene Ontology (GO) biological process annotations, including regulation of CDK activity [GO:0000079] (p<9×10−7), microtubule-based process [GO:0007017] (p<2×10−6), cell cycle [GO:0007049] (p<4×10−6), cytoskeleton organization and biogenesis [GO:0007010] (p<8×10−6), microtubule cytoskeleton organization and biogenesis [GO:0000226] (p<8×10−6), G2/M transition of mitotic cell cycle [GO:0000086] (p<5×10−5), DNA replication and chromosome cycle [GO:0000067] (p<5×10−5), and related processes. These genes include CLB2, CLB3, CLB5, CDC31, KAR1, SPC97, PAC2, TUB2, NIP100, SLK19, ASK1, AME1, MAD2, and ACT1, which have direct roles in regulating the G2/M transition and related processes such as microtubule nucleation, chromosome segregation, and mitotic spindle checkpoint control. Additionally, 7 genes identified in previous large-scale studies [13],[16],[17] (SPO13, SEC17, MYO2, PRP31, ARF1, TFG2, and SHE1), although not directly involved in mitotic cell cycle control, were also observed in this study. Of 63 genes newly identified in this screen (3 were not tested for growth phenotype), 56 caused slow growth upon induction and the overexpression of 21 genes lead to specific sensitivity to nocodazole.
In order to better classify the genes by the nature of their overexpression defects, i.e., as to whether the cells exhibited M phase arrest or whether chromosome segregation defects led to G2/M arrest, 3 classes of nuclear morphology were assigned based on the patterns of DNA staining, as shown in Figure 4 D–F: an undivided nucleus in one cell body (class I, pre-M), an undivided nucleus in the bud neck (class II, early-M), and divided nuclei in two cell bodies (class III, late-M) [17]. In control strains, 60% of the cells exhibited class III nuclear morphology, with chromosomes in these cells successfully segregated, while only 11% of cells showed class I morphology, and 26% of cells class II morphology. We observed 20 ORF strains to have significantly elevated percentages (95% confidence level) of cells with class I morphology, 13 ORF strains with class II, and 17 ORF strains with class III (Figure 5). Among the 33 genes in the Class I and II, 9 have direct roles in regulating G2/M transition (CLB2, CLB3 and CLB5), or related important events in the mitotic cell division phase (ACT1, TUB2, NIP100, PAC2, CDC31, SPC97). For example, Spc97p is a component of the microtubule-nucleating Tub4p (gamma-tubulin) complex and overproduction of SPC97 causes microtubule defects, which in turn gives rise to a failure of chromosome segregation and a early M phase arrest (Figure 5B) [29]. We therefore reasoned that 24 newly implicated Class I and II genes causing a similar phenotype to that of SPC97 might play direct or indirect roles in chromosome segregation, especially for genes whose overexpression also leads to hyper sensitivity to nocodazole (GEA2, RFA1, HOS3, YPR015C, AVO2, CBF1, SHE1, and TEA1; Figure S2). We characterized two of these genes, RFA1 and YPR015C, in more detail.
YPR015C encodes an uncharacterized putative transcription factor known to exhibit synthetic lethality with and be functionally linked to CTF4 [34],[35]; both genes have zinc finger motifs. CTF4 encodes a chromatin-associated protein required for sister chromatid cohesion, which in turn regulates high-fidelity chromosome segregation (Hanna et al., 2001). Deletion of CTF4 increases chromosome instability and causes early mitotic delay [36]–[38]. We observe overexpression of YPR015C to give rise to a very similar phenotype to deletion of CTF4. YPR015C overexpression causes hyper sensitivity to nocodazole and slight sensitivity to hydroxyurea (Figure S2), and an elevated population of large-budded cells with the nucleus in the bud neck (Figure 6B). In order to test whether the overexpression of YPR015C also leads to chromosome instability, we overexpressed YPR015C in the strain Cry1 (MAT a ade2-1, ura3-1, leu2-3, 112, trp1, his3-11) carrying the low copy centromere-containing plasmid pRS412::ADE2 [cir+]. Overexpression of YPR015C doubled the rate of loss of centromere plasmids: 36% in the YPR015C overexpressing strain vs. 16% in the wild type control strain, indicating chromosome instability and mis-segregation.
Bud size and nuclear morphology indicated that cells arrested in early mitosis phase when YPR015C was overexpressed (Figure 6B). To test whether the early mitotic delay caused by the overexpression of YPR015C is due to activation of the DNA damage checkpoint or the spindle assembly checkpoint, we overexpressed YPR015C in the background of rad9Δ or mad2Δ mutants in which the DNA damage or spindle assembly checkpoints were removed, respectively. Cell cycle progression in these mutants was measured by DNA content analysis of galactose-induced cultures (Figure 6E). We observed that the YPR015C-induced early mitotic delay was dependent on the DNA damage checkpoint and not the spindle assembly checkpoint, in contrast to the early mitotic delay caused by deletion of CTF4, which is dependent on the spindle checkpoint [37]. Interestingly, three ribonucleotide reductases (RNR2, RNR3, RNR4) are the most significantly up-regulated genes following overexpression of YPR015C [39], and these three ribonucleotide reductases are regulated by the DNA replication and DNA damage checkpoint pathways [40]. Since transcriptional response, DNA replication, DNA repair, and chromosome condensation are the major chromatin restructuring events in cohesin operation [37], it appears that overexpression of YPR015C may interfere with chromosome cohesion, inducing defects in mitotic chromosome segregation via a different mechanism than CTF4.
RFA1 is another gene involved in DNA replication whose overexpression leads to G2/M delay. The Rfa1p protein is a subunit of the heterotrimeric replication protein A (RPA), which is involved in DNA replication, repair, and the DNA damage checkpoint [41],[42]. RFA1 is essential for yeast viability, an RFA1 null mutant is inviable [18]. However, several point mutations of RFA1 caused accumulation of large-budded [43] or dumb-bell shaped cells with a single nucleus in the bud neck [42] at the nonpermissive temperature and had defects in DNA replication and DNA repair [42]–[45]. We observe ∼73% of large-budded cells of the RFA1 overexpression strain showed a butterfly-shaped nucleus in their bud necks, similar to phenotype of SPC97 overexpression (i.e., asymmetric chromosome segregation) and fewer than 10% of large-budded cells had chromosomes segregated into two cell bodies (Figure 7B), suggestive of chromosome mis-segregation. In contrast, 63% of large-budded cells of the parental control strain had the chromosomes successfully segregated into two cell bodies.
Furthermore, we observed that the RFA1 overexpression strain had short mitotic spindles, with spindle pole bodies not clearly attached to the nucleus (Figure 7C, lower row). This defect is distinct from the spindle morphology caused by overexpression of SPC97 (Figure 7C, middle row); Spc97p is a component of the microtubule-nucleating Tub4p (gamma-tubulin) complex and is involved in spindle pole body separation and mitotic spindle formation. Cells either carrying point mutations [29] or overexpressing SPC97 (Figure 7C, middle row) had short spindles and elongated cytoplasmic microtubules, but the spindle pole appeared normally attached to the nucleus. Given that Rfa1p is a single-stranded DNA binding protein involved in DNA replication, it seems likely that overexpression of RFA1 disrupts DNA replication and leads to the observed spindle morphology defects, giving rise to the observed early mitotic delay. Such a role would also be consistent with the observation that DNA replication proteins can act as cohesion proteins and play important roles in regulating spindle integrity and maintaining the tension on chromosomes exerted by spindle microtubules [37],[46],[47].
While the strains arresting in G2/M phase were strongly enriched for cell cycle associated functions, diverse mechanisms are known to induce G1 arrests [16],[17]. This diversity was reflected in the enrichment of GO biological process annotations among the G1 arresting ORFs: no pathway was enriched at p<0.001 when calculated by the method of [33], consistent with previous overexpression studies [16],[17]. When calculated as in [25], the strongest enrichment consisted of negative regulators of transcription from RNA polymerase II promoters (GO:0000122; p<4×10−4).
Among the 21 genes inducing G1 delays, 6 (29%) are uncharacterized or dubious ORFs. The only functional information available for YOR131C and YDR493W is localization: YOR131C is localized in the nucleus and cytoplasm, and YDR493W is localized in mitochondria [18],[19],[48]. Our data further associate these two genes with cell cycle progression, either directly or indirectly. Tma64p is another protein of unknown function, previously identified in a mass spectrometry-based proteomic screen of yeast ribosomal complexes [49]. Tma64p associates with ribosomes, has a RNA binding domain and interacts with Rps4bp, a component of the small (40S) ribosomal subunit [50]. Moreover, it has been suggested that there might be a strong connection between ribosomal biogenesis and G1 transit [11],[13]. Therefore, the G1 delay caused by overexpression of TMA64 may suggest a role in ribosomal biogenesis.
The weak enrichment observed for transcriptional regulators derives from 4 transcription factors involved in responding to environmental stress that were observed in the G1 category. Three are transcriptional repressors (MIG3, NCB2, and SKO1), and the fourth (GAT4) is unclear as to mode of action. We observed unusual cellular morphology upon overexpression of SKO1, and examined this repressor in more detail.
We observed overproduction of SKO1 to strongly inhibit cell growth and arrest cells at the G1 phase (Figure 8A). Bud size analysis showed that 90% of cells had no bud when SKO1 was overexpressed (Table S1). SKO1 is a basic leucine zipper (bZIP) transcription factor of the ATF/CREB family, involved in osmotic and oxidative stress responses. The Sko1p protein forms a complex with Tup1p and Ssn6p to both activate and repress transcription [51]–[53]. Surprisingly, overproduction of SKO1 resulted in formation of shmoos, cell morphology changes that are normally seen in mating yeast in response to mating pheromone (Figure 8B). We reasoned that the elevated expression of SKO1 might activate the pheromone response pathway either directly or indirectly, causing shmoo formation and a mating-associated G1 arrest.
Since Fus1p is a marker protein induced during shmoo formation that localizes to the shmoo tip when the pheromone response pathway is activated [54], we tested SKO1 activation of the pheromone response pathway by examining the localization of Fus1p when SKO1 was overexpressed. We transformed PGAL1-SKO1 plasmids into a MATa strain in which FUS1 was C-terminally tagged with green fluorescent protein (GFP) [19]. Upon SKO1 overexpression, Fus1-GFP localized to the shmoo tip (Figure 8B), resembling its localization pattern upon alpha factor treatment, demonstrating that the morphological changes are accompanied by general activation of the mating pathway, thus explaining the G1 cell cycle arrest phenotype of the SKO1 ORF strains.
To further explore which genes involved in the pheromone MAP kinase pathway were activated by the overexpression of SKO1, we performed cDNA microarray profiling and found that the activated genes were highly enriched in pheromone response and mating genes. Significantly upregulated genes (p<0.01) included MFA1, STE2, BAR1, FAR1, FUS1, KAR4, FIG1, FIG2, GIC2, PRM4, PRM5, PRM8, AGA1, and AGA2, as listed in Table S5. To establish direct genetic interactions between SKO1 and pheromone response pathway, we overexpressed SKO1 strains in ste2Δ, ste4Δ, ste20Δ, ste11Δ, ste5Δ, kar4Δ, fus3Δ, far1Δ, fus1Δ, sst2Δ, and dig2Δ strains, and examined whether or not SKO1 overexpression induced shmoo formation in these deletion strains. We did not observe SKO1-induced shmoo formation in ste2Δ, ste4Δ, ste20Δ, ste11Δ, ste5Δ, kar4Δ, and far1Δ strains (Figure 8C), indicating that these genes are required for shmoo induction by SKO1 overexpression. FUS3 is functionally compensated by KSS1, FUS1 is downstream of the pheromone response signal transduction pathway, SST2 and DIG2 are inhibitors in the pathway; deletion of these genes affects neither pheromone nor SKO1-dependent shmoo induction. The observed effects of SKO1 overexpression on cell cycle progression thus appear to be indirect, activating the pheromone response pathway in a manner dependent upon the pheromone receptor (STE2) and MAP kinase signal transduction pathway, and this activation in turn results in G1 arrest through the normal mating pheromone-mediated pathway.
Overexpression of a normal gene product can result in gain-of-function, but may also mimic loss-of-function phenotypes [16], such as in cases where precise levels of a protein are required, with either too much or too little equally disruptive. In order to systematically assess the extent of these phenomena amongst the phenotypes of the overexpression strains, we took advantage of quantitative cell morphology data (bud count data) for deletion strains collected in the Saccharomyces cerevisiae Morphology Database (SCMD) [55] and compared them to our quantitative bud count data. Of 108 genes from this screen, 77 also appear in SCMD (21 essential genes and 10 additional genes are not included in SCMD) (Figure 9). We selected genes from our screen with significantly elevated populations (p<0.05) of unbudded cells or large-budded cells. In the G1 category, there were 12 strains from our screen whose percentages of cells without buds were significantly higher than that of wild type. Of these 12 G1 genes, only one also led to a significantly elevated population of unbudded cells when deleted, as measured by SCMD. Therefore, our rough estimate is that 11/12 (92%) of genes in the G1 category exhibit an overexpression phenotype distinct from the loss-of-function phenotype, at least as measured with regard to proportions of unbudded cells. Similarly, 44 (94%) genes in the G2/M category caused a significantly elevated proportion of large-budded cells when overexpressed but not when deleted, versus 3 that resembled the loss-of-function phenotype (Figure 9, Table S3). Thus, the majority of the overexpressed genes in this paper appear to exhibit a phenotype distinct from the loss-of-function case, supporting the previously hypothesized notion that gain-of-function may be common amongst the overexpression phenotypes [16].
SKO1 appears to represent such an example of a gain-of-function leading to differences between the overexpression phenotype and the corresponding deletion phenotype. When overexpressed, SKO1, which encodes a transcription repressor responsive to salt and osmotic stresses, activates the pheromone response pathway and leads to a strong G1 arrest, but the deletion of SKO1 has no detectable arrest or mating phenotype (Figure 8B). Moreover, transcriptional profiling of cells overexpressing SKO1 revealed that genes involved in the pheromone response pathway are significantly upregulated. However, genes involved in the pheromone response pathway do not appear to be regulated by SKO1 under normal culture conditions, at least as measured by chromatin-immunoprecipitation of SKO1 [56]. Therefore, our results suggest that SKO1 regulates genes in the pheromone response pathway through a gain-of-function mechanism, e.g., such as by enabling binding to a cryptic or lower affinity promoter when overexpressed.
In this paper, we describe a near-saturating screen for yeast genes whose overexpression causes cell cycle delays and which are thus likely to function in cell cycle progression. We individually examined the effects of overexpression on cell cycle progression for each of ∼5,556 yeast ORFs, and report the 108 genes with the most significant and reproducible cell cycle defects. 82 of these genes have not been reported in previous large-scale screens [13],[16],[17], probably due to different overexpression conditions and strain backgrounds, false positives in large-scale screens [11], or more likely, false negatives, e.g., such as might derive from variable 2 micron plasmid copy numbers [57] increasing phenotypic variability and thus allowing cell cycle defects to escape detection. Our analysis thus complements previous screens. These results lay the foundation for future experiments to elucidate the precise roles of these genes in cell cycle progression, such as the mechanisms of RFA1 and YPR015C. Overexpression screens such as we have described here provide complementary information to loss-of-function studies and therefore offer new opportunities for discovery of genetic interactions, such as by systematically testing the overexpression plasmids in deletion strains to screen for phenotype suppression or synthetic interactions. Finally, since overexpression is an efficient technique in human cell culture and since regulation of cell proliferation is an important aspect of studying human diseases, we anticipate that a similar effort to this work in human cell lines could accelerate our understanding of cell cycle control in mammalian systems and help to further clarify the many connections between cell cycle control and cancer. |
10.1371/journal.pcbi.1002378 | A Cell-based Computational Modeling Approach for Developing Site-Directed Molecular Probes | Modeling the local absorption and retention patterns of membrane-permeant small molecules in a cellular context could facilitate development of site-directed chemical agents for bioimaging or therapeutic applications. Here, we present an integrative approach to this problem, combining in silico computational models, in vitro cell based assays and in vivo biodistribution studies. To target small molecule probes to the epithelial cells of the upper airways, a multiscale computational model of the lung was first used as a screening tool, in silico. Following virtual screening, cell monolayers differentiated on microfabricated pore arrays and multilayer cultures of primary human bronchial epithelial cells differentiated in an air-liquid interface were used to test the local absorption and intracellular retention patterns of selected probes, in vitro. Lastly, experiments involving visualization of bioimaging probe distribution in the lungs after local and systemic administration were used to test the relevance of computational models and cell-based assays, in vivo. The results of in vivo experiments were consistent with the results of in silico simulations, indicating that mitochondrial accumulation of membrane permeant, hydrophilic cations can be used to maximize local exposure and retention, specifically in the upper airways after intratracheal administration.
| We have developed an integrative, cell-based modeling approach to facilitate the design and discovery of chemical agents directed to specific sites of action within a living organism. Here, a computational, multiscale transport model of the lung was adapted to enable virtual screening of small molecules targeting the epithelial cells of the upper airways. In turn, the transport behaviors of selected candidate probes were evaluated to establish their degree of retention at a site of absorption, using computational simulations as well as two in vitro cell-based assay systems. Lastly, bioimaging experiments were performed to examine candidate molecules' distribution in the lungs of mice after local and systemic administration. Based on computational simulations, the higher mitochondrial density per unit absorption surface area is the key parameter determining the higher retention of small molecule hydrophilic cations in the upper airways, relative to lipophilic weak bases, specifically after intratracheal administration.
| Local administration of therapeutic agents or bioimaging probes is commonly used to maximize concentrations at a desired site of action and to minimize side effects or background signals associated with distribution in off-target sites. However, in the specific case of inhaled, small molecule therapeutic agents or bioimaging probes, cell impermeant molecules may rapidly disappear from the sites of deposition via mucociliary clearance [1], [2]. Conversely, cell- permeant small molecules can rapidly diffuse away and disappear from the site absorption, down their concentration gradient [3]. Therefore, we decided to explore an integrative simulation approach (Figure 1) to study how the physicochemical properties of small molecule probes may be optimized to maximize local targeting and retention in the upper respiratory tract.
Previously, we constructed multiscale, cell-based computational models of airways and alveoli to predict the relative absorption, accumulation and retention of inhaled chemical agents [4]. In these models, the transport of small molecules from the airway surface lining to the blood or from the blood to the airway surface lining were modeled using ordinary differential equations (ODEs) [5], [6]. These ODEs described the transport of drug molecules across a series of cellular compartments bounded by lipid bilayers (Figure 1A,), which form the surface of each airway generation, modeled as a tube (Figure 1B). For a monoprotic base, the concentration of molecule in each subcellular compartment was divided into two components: neutral and ionized [7], [8]. Accordingly, two drug specific properties were used as input to simulate the transport process across each lipid bilayer: the logarithms of the octanol∶water partition coefficient of the neutral form of the molecule (i.e., logPn) and the pKa of the molecule. The logarithm of the octanol∶water partition coefficient of the ionized form of the molecule (i.e., logPd) can be derived from logPn or it can be incorporated as an independent input parameter that can be measured or calculated with cheminformatics software. For different compartments with different pHs and lipid fractions, the free fraction of the neutral and ionized forms of molecules was calculated according to the molecule's pKa, logPn, and logPd, using the Henderson-Hasselback equation and the laws of mass action.
Anatomically, the structure of the airways was modeled as a tree-like branching system of cylinders with progressively narrowing diameter [9] (Figure 1C). Starting with the trachea as the trunk of the tree and ending in the alveoli as the leaves, each branching segment corresponded to an airway ‘generation’ characterized by a particular surface area, blood flow, and cellular organization [4],[10]. Histologically, the walls of the airways or alveoli were modeled as multiple layers of epithelial, interstitial and endothelial cells separating the air from the blood. Several structural and functional differences between the airways and alveoli are noteworthy: 1) cartilage and smooth muscle are present only in the interstitium of the airways; 2) the surface area of the alveoli is two orders of magnitude larger than airways; and 3) while the blood flow to the alveoli corresponds to 100% of cardiac output from the right ventricle, the blood flow of the airways is approximately 1% of the cardiac output from the left ventricle [11], [12].
To predict a molecule's absorption and retention in each airway generation, the transport properties of small molecules across cellular membranes, as well as the local partitioning of molecules into lipid in different subcellular compartments can be calculated with the Fick and Nernst-Planck equations to describe the transport of the neutral and charged species of the molecule [4]. In simulations, combinations of logP and pKa spanning a range of values were used as input to simulate the changes in concentration of molecules of varying chemical structure, as they are absorbed from the airway surface lining liquid into the blood or vice versa.
Here, we applied this cell-based transport model as a virtual screening tool, to identify compounds with differential distribution profiles in airways and alveoli, after intratracheal (IT) or intravenous (IV) administration. In addition, two innovative in vitro cell based assays were developed to assess the absorption and retention of molecules across multiple layers of cells along the lateral (Figure 1 D–F) and transversal planes of a cell monolayer (Figure 1G)). Finally, in vivo microscopic bioimaging experiments were performed to visualize the distribution of fluorescent probes in the lung after either IT or IV administration (Figure 1H). The results revealed that the mitochondrial sequestration of hydrophilic, cell-permeant cations can provide an effective mechanism for maximizing their local exposure and retention at a site of absorption. Accordingly, mitochondriotropic cations may be useful as fiduciary markers of local, inhaled drug deposition patterns in the upper respiratory tract.
All of the equations and default parameter values were based on our published model [4]. The ODEs that describe this lung pharmacokinetic (PK) model were solved numerically in a Matlab® simulation environment (Version R2009b, The Mathworks Inc, Natick, MA). The ODE15S solver was used to address the issue of the stiffness in ODEs, and the relative and absolute error tolerance was set as 10−12 to minimize numerical errors. The Matlab scripts used for virtual screening and simulation purposes are provided, together with detailed instructions for running them, in the Supplementary Materials (Text S1, S2, S3, 4, S5, S6). The results of detailed parameter sensitivity analysis are also provided in the Supplementary Materials (Text S7).
For virtual screening, the airway and alveoli were linked to a systemic pharmacokinetic model through their respective blood compartments using a single compartment PK elimination model (eq. 1) [13]:(1)Where Vb is the volume of the blood compartment; Cb is the concentration in the blood; and CL is the clearance. The same initial dose (1 mg/kg) was used as an input parameter to simulate IT instillation experiments in the airways and alveoli, respectively. For virtual screening, clearance in the systemic circulation was set to zero. The logPn (−2 to 4 with interval of 0.1 units) and the pKa (5 to 14 with interval of 0.2 units) of monobasic compounds were independently varied and used as input parameters, in all possible combinations. For each set of physicochemical input parameters (logPn and pKa) two important pharmacokinetic indexes were calculated: 1) the percentage of mass deposited in the airways and alveoli (relative to the total mass in whole lung); and, 2) the concentration in the alveolar and airway regions, calculated as the sum of the masses in all the compartments in said regions of the lung divided by the sum of all the compartment volumes in that region. The area under the tissue concentration curve (AUC) for the airways and alveoli was calculated using the trapezoidal rule. The AUC ratio of airways to alveoli after inhalation was calculated by dividing the AUC of the airways by the AUC of the alveoli for every combination of logPn and pKa that were used as input.
For comparison, simulations were also run to simulate an intravenous (IV) bolus injection, with the initial concentration in venous blood as calculated with eq. 2:(2)The volume of venous and artery blood was set to 13.6 and 6.8 ml, respectively [13], [14]. The concentration in the blood was fixed (clearance set to 0) with the assumption of no significant plasma protein binding and a drug concentration blood to plasma ratio of 1.
Based on the results of virtual screening, two fluorescent probes were selected for further testing: Hoechst® 33342 (Hoe, Molecular Probes, CA, USA) to represent a highly hydrophobic, weakly basic molecule that can serve as a reference marker for a readily absorbed probe with limited intracellular retention; and, Mitotracker® Red (MTR, Molecular Probes, CA, USA) to represent a more hydrophilic cation that could serve as a candidate fiduciary marker for local inhaled drug deposition and absorption patterns. MTR was modeled with a single, fixed positive charge and a logPd = 0.16. Hoe was modeled as a lipophilic, monobasic molecule with a pKa = 7.8 and a logPn = 4.49 (calculated with ChemAxon, www.chemaxon.com). These physicochemical properties were used as input parameters to calculate the time dependent changes of the probe concentrations in the airways and alveoli, respectively. For simulations of IT instillation, the same initial concentration (1 mM) of MTR and Hoe was assumed as the initial condition for the airways and alveoli. The same initial dose used for IT instillation was also used for IV administration. Blood clearance was fixed to 0 for simulations, unless otherwise noted.
A customized transwell insert system was constructed using a polyester membrane with microfabricated pore arrays precisely machined using a focused ion beam (Hitachi FB-200A) [15] (Figure 1 E). These membranes support cell growth and the pores serve as a point source for compound administration to single cells on a cell monolayer (Figure 1F)). The pore arrays were comprised of 3 µm diameter cylindrical pores, arranged 20 µm apart in a 5-by-5 square array. Pores were also arranged 40, 80 and 160 µm apart in 3-by-3 symmetrical arrays. The pores were individually machined using a high brightness Ga liquid metal ion source coupled with a double lens focusing system. The perforated membranes were glued (Krazy Glue®) to the bottom of hollow Transwell® holder (Costar 3462 or 3460), creating a permeable support for cell growth (Figure 1D). The integrity of the insert system was tested by adding 5 mM Trypan Blue (dissolved in Hank's balanced salt solution; HBSS) to the insert wells [15]. The insert was considered intact if there was no evidence of Trypan Blue leakage from the edge of the insert membrane. For assessing lateral cell-cell transport, Madin-Darby canine kidney (MDCK) cells were purchased from ATCC (CCL-34™) and grown (37°C, 5% CO2) in Dulbecco's modified Eagle's medium (DMEM, Gibco 11995) containing 10% FBS (Gibco 10082), 1× non-essential amino acids (Gibco 11140) and 1% penicillin/streptomycin (Gibco 15140). MDCK cells were seeded on polyester membranes containing the pore arrays at a density between 1×105–2×105 cells/cm2 and were grown until a confluent cell monolayer formed (Figure 1F). To evaluate the effect of pore arrays on cell monolayer intactness, MDCK cells were washed and incubated in transport buffer (HBSS buffer supplemented with 25 mM D-glucose, pH 7.4) for 30 min followed by transepithelial electrical resistance (TEER) measurement using Millipore Millicell® ERS. Cell monolayers were used for experiments only if the background subtracted TEER values were higher than 100 Ω·cm2 and if the cells covering the pore arrays appeared as an intact monolayer.
To assess cell-to-cell transport along the plane of the monolayer (Figure 1D–F), fluorescent dyes were added into the basolateral compartment of the transwell system (at time 0). The dynamic staining pattern in the cells was imaged (Nikon TE2000S epifluorescence microscope equipped with a triple-pass DAPI/FITC/TRITC filter set (Chroma Technology Corp. 86013v2)). The 12-bit grayscale images were acquired using a CCD camera (Roper Scientific, Tucson, AZ). For measurements, individual cells or nuclei in these images were manually outlined using the region tool in MetaMorph® software (Molecular Devices Corporation, Sunnyvale, CA). The average and standard deviation of cellular or nucleus fluorescence intensity was measured using MetaMorph®, after subtracting the background fluorescence intensity estimated from the unstained regions of the monolayer distant to the pores. The rate of Hoe staining in the nucleus was measured as the slope of fluorescence increase normalized by the slope of increase in the first nucleus (closest to the pore).
Normal human bronchial epithelial (NHBE) cells (Clonetics™, passage 1; Lonza, Walkersville, MD) were cultured (37°C, 5% CO2) and seeded (passage 2) at 2.5×105 cells/cm2 on a Transwell® insert (Corning Inc., Lowell, MA; area: 0.33 cm2, pore size: 0.4 µm) in NHBE differentiation media (Lonza, Walkersville, MD) The apical media was aspirated after 24 h of cell seeding and the cells on the polyester membrane were maintained in media only in the basolateral compartment of the air-liquid interface culture (ALC) [16], [17]. On day 8 of ALC, the integrity of the cell layers on the membrane was assessed by light contrast microscope and by transepithelial electrical resistance (TEER) [18]. After equilibration of the cell layers on the insert with pre-warmed HBSS buffer (10 mM HEPES, 25 mM D-glucose, pH 7.4) for 30 min (37°C, 5% CO2), TEER values were obtained and cells with TEER values of ∼600 Ω•cm2 were used for the transport and retention assays [16], [17], [19], [20], [21].
NHBE cell multilayers grown on the inserts were examined with a Zeiss LSM 510-META laser scanning confocal microscope (Carl Zeiss Inc., Thornwood, NJ) with a 60× water immersion objective on day 8 of ALC culture. For the confocal analyses, three different cell-permeant dyes were prepared by dilution with HBSS buffer 10 µg/ml Hoe; 2.5 µM LysoTracker® Green (LTG, Molecular Probes, CA); and 1 µM MTR). After the cell multilayers were washed with HBSS, 240 µl of dye mixture (80 µl of each dye in HBSS) was added to the apical compartment and 600 µl of HBSS was added to the basolateral side. After 30 min, transport of the dyes across the cell layers was measured by placing the insert into a two-chambered slide (Lab-TeK®; Thermo Scientific Nunc co., Rochester, NY) and acquiring images along the Z-axis (interval, 1 µm) in three fluorescence channels (coherent enterprise laser (364 nm) for Hoe, Argon laser (488 nm) for LTG, and Helium neon 1 laser (543 nm) for MTR). The distribution of probes applied in the apical compartment of the NHBE cell multilayer cultures was assessed in 3D reconstructions of the acquired images of probe distribution, using MetaMorph® software (Figure 1G). The relative distributions of MTR, Hoe, and LTG dyes across the multilayers were assessed by imaging analyses through multiple Z-stacks. After background subtraction, the integrated intensity of each fluorescence channel per cell was summed in each cell layer and divided by the total integrated intensity in all the layers to calculate the percentage of relative distribution of the integrated fluorescence signal of each dye associated with inner layer or the exposed surface layer of the NHBE cell multilayer.
The distribution of MTR and Hoe in airways and alveoli after IT and IV injection in live mice were determined by microscopic imaging of cryopreserved lung tissue sections and confirmed by visual inspection followed by quantitative imaging of high resolution tiled mosaics assembled from fluorescence images of tissue sections (Figure 1H). For these experiments, male C57BL/6J mice (Jackson Laboratory, Bar Harbor, ME; 8 weeks, 20–30 g) were used and the protocol was approved by the University of Michigan's animal care and use committee in accordance with the National Institutes of Health Office of Laboratory Animal Welfare “Principles of Laboratory Animal Care.” MTR (50 ug in 10 ul DMSO) and Hoe (90 ul of 10 mg/ml in ddH20) were mixed so that the final concentration of MTR and Hoe was 0.94 and 14.61 mM, respectively. Mice received either 50 µl of dye mixture or 50 µl saline (control) via IV tail veil injection or IT instillation [22]. For IV administration, conscious mice were briefly restrained and for IT instillation mice were anesthetized with isoflurane gas, and the dose was delivered to the airway via the oral route as previously described.
In order to study the differential regional distribution of fluorescent dyes in the lung, mice were anesthetized with ketamine/xylazine 40 minutes after dosing. A thoracotomy was performed and a heparinized blood sample was acquired by cardiac puncture. The trachea was cannulated (20G luer stub) after which the lungs were inflated with ∼1 mL of a 30% sucrose-optimal cutting temperature (OCT; Tissue-Tek, Sakura Finetek USA, Torrance, CA USA) mixture and removed en bloc. The lungs were immersed in OCT and were immediately frozen (at −80°C) [23].
For microscopy, coronal lung sections (7 µm) were imaged using an epifluorescence Olympus BX-51 microscope equipped with the standard DAPI, FITC and TRITC filter sets. A series of low-magnification (×4) left and right lung section images were electronically captured with an Olympus DP-70 high-resolution digital camera using Image J software (ImageJ 1.44b, National Institutes of Health, USA; http://rsb.info.nih.gov/ij). In order to permit comparisons of image brightness and fluorescence, images for each lung section were acquired using the same illumination and image acquisition settings. Mosaics of the entire lung were tiled using Photoshop® (version 4; Adobe Systems Inc., San Jose, CA) and quantitative image analysis was carried out using the integrated morphometric analysis function of MetaMorph®. Background subtracted fluorescence intensity values over the airways and alveoli were measured, as the integrated value of all pixels per unit area of the manually selected airway and alveolar tissue regions, using the images acquired with the DAPI channel. In turn, the same airway and alveolar tissue regions were used to measure the MTR fluorescence signal using the images acquired with the TRITC channel.
For virtual screening experiments, molecules with maximal tissue exposure (AUC) in the airways after inhalation were identified by using combinations of logPn and pKa as input parameters in a multiscale, cell-based lung transport model (Figure 2). For weak bases, lower lipophilicity and higher pKa promoted intracellular retention and led to greater local exposure relative to the alveoli (Figure 2A, B). The calculated airway/alveoli exposure ratio (Figure 2C) ranged from 100 to 700 and increased with lowered logPn (increasing hydrophilicity) and higher pKa (greater positively charged fraction at physiological pH) Essentially, cell-permeant, hydrophilic molecules harboring a fixed positive charge showed the greatest accumulation and retention in the cells of the upper airway relative to the alveoli, following IT administration.
To probe the role of the route of administration, simulations were also performed by independently varying logPn and pKa to calculate the mass deposition pattern in the airways and alveoli under steady state conditions after IV administration (Figure 2D–F). In this manner we established the relationship between the physicochemical properties of small molecules and absolute and relative mass distribution in the airways (Figure 2D) and alveoli (Figure 2E). Following IV administration, the majority of the mass was deposited in the alveoli irrespective of the physicochemical properties of the molecules (Figure 2F); the airways held less than 20% of total drug mass in the lungs. Compounds with low logPn and high pKa tended to exhibit the largest airway to alveoli mass ratios, which paralleled the results obtained after IT administration.
In order to validate the results of these virtual screening experiments, two fluorescent bioimaging probes, MTR and Hoe, were selected for more detailed analysis. MTR is a cell-permeant, hydrophilic cation, and Hoe is a cell permeant, hydrophobic weak base. Based on the screening results (Figure 3) and more detailed simulations (Figure 3), the concentration profiles of these two fluorescent molecules in the airways and alveoli were markedly different after IT (Figure 3 A, B) and more similar after IV (Figure 3 C, D) administration. When given IT, the predicted MTR concentration, 40 to 60 min after administration, was nearly 10-fold higher in the airways than in the alveoli (Figure 3A). Conversely, the predicted concentration of Hoe in the airways was two-fold higher in alveoli than in airway (Figure 3B). When given IV, the predicted concentration of MTR in the airways was almost the same as that in alveoli (Figure 3C). However, the predicted concentration of Hoe in the airways was higher in alveoli (Figure 3D). Thus, MTR should be retained in the airways specifically after IT administration, whereas Hoe should not be retained in airways relative to alveoli regardless of the route of administration.
Next, cell based assays were used to establish the intracellular retention of MTR and Hoe at a site of absorption. For this purpose, a transwell insert system with micro-fabricated pores was constructed. After seeding MDCK epithelial cells on the patterned pore arrays and adding hydrophobic fluorescent compounds in the basolateral side of cell monolayer, the time course dye uptake in the cells sitting above the pores and the kinetics of lateral transport from the cells lying on top of the pore to the neighboring cells was visualized by fluorescence microscopy.
Three hours after the addition of Hoe to the basolateral compartment, only cells that were within close vicinity of pores were stained, indicating that the cells formed a tight seal with the pores such that each pore fed almost exclusively into cells that were in immediate contact with the pores (Figure 4). Monitoring of the cell-to-cell diffusion of Hoe over time, indicated that the pores served as point sources of sustained dye supply to the adjacent cells (Figure 4A–D) and for cells grown on membranes with pores spaced by 80 µm (Figure 4C) or 160 µm (Figure 4D), each pore could be considered as the single point source of dye molecules. Quantitative image analysis revealed that the rate of staining rapidly decreased as the distance of the cells from the pores increased (Figure 4E, F). Remarkably, only cells in the vicinity of each pore were labeled.
As controls, cells were stained with Hoe plus BCECF-AM from the basolateral compartment (Figure 4G–I). BCECF-AM is a nonfluorescent cell-permeant ester, which generates a cell-impermeant, fluorescent molecule upon intracellular hydrolysis. While the extent of Hoe diffusion was dependent on the distance from the pores (Figure 4G), the green fluorescence of the hydrophillic ester hydrolysis product (BCECF) was exclusively restricted to the first layer of cells that were in direct contact with pores (Figure 4H, I).
Similar to the Hoe staining pattern, MTR also exhibited a highly constrained diffusion pattern with most of the staining restricted to the vicinity of each pore (Figure 5). After two-hours of staining from the basolateral compartment with both Hoe (Figure 5A) and MTR (Figure 5B), only cells within 60 microns of the pore being stained with both probes (Figure 5C). The normalized fluorescence intensity of MTR and Hoe were similar in the first and second layers of cells, but MTR showed higher penetration into the third layer (Figure 5D).
In the transversal direction, the absorption and retention of MTR and Hoe across multiple layers of cells was also assessed in primary NHBE cells differentiated as multilayers in ALC (Figure 6). For the experiments, MTR and Hoe were simultaneously added in the apical side of the cells and intracellular accumulation was assessed using 3D reconstructions of the cell multilayers (Figure 6). As a positive control, LTG was also included in the apical HBSS buffer. Thirty minutes after the addition of probes to the apical compartment, both MTR and Hoe staining were constrained to the first, outer surface layer of cells (Figure 6, left). The cells beneath the surface layer of cells were stained with LTG (Figure 6, right), indicating that the limited penetration of both MTR and Hoe. Different transport patterns of MTR, Hoe and LTG across the cell multilayers were verified by image quantitation using MetaMorph® software in the multiple Z-stack images of NHBE cell multilayers. Approximately 96%±2.76% of MTR or 96%±2.48% Hoe of the dye was retained in the surface cell layer whereas 50%±15.62% of LTG fluorescence was associated with the surface cell layer. Tukey's multiple comparison test following ANOVA (one-way analysis of variance) test showed statistically significant difference between MTR and LTG (p-value<0.0001) and also between Hoe and LTG (p-value<0.0001), but not between MTR and Hoe with p-value larger than 0.05 (α = 0.05).
As an ultimate test of the results of in silico virtual screening experiments, mice were administered a mixture of MTR and Hoe by either IV tail vein or IT instillation and the distribution of the molecules in the lungs was assessed by fluorescent microscopy (Figure 7). Hoe distributed throughout the lungs regardless of route of administration (Figure 7A, B) with fluorescence in both alveoli and airways (Figure 7C, D)). Following IV administration, MTR also distributed throughout the lung in both airways and alveoli (Figure 7E). Conversely, IT administered MTR resulted in highly uneven fluorescence distribution (Figure 7F). Most importantly, the airway regions showed comparable MTR fluorescence in airway vs. alveoli after IV (Figure 7G) but higher MTR fluorescence intensity in airways compared with the alveoli following IT delivery (Figure 7H).
To confirm these observations quantitative image analysis was performed to compute background subtracted integrated intensity of alveolar and airway regions, to quantify the relative, differential fluorescence intensity distribution of Hoe and MTR in airway and alveoli. The fluorescence MTR/Hoe ratio ranged from 2.42 to 3.27 for IT administration. For MTR and Hoechst, the mean (± s.d.) percent airway delivery was 23.9%±5.8% and 8.8%±2.7%, respectively (based on 422 region measurements from a single lung). For IV administration, the fluorescence MTR/Hoe ratio ranged from 0.95 to 1.45. The mean (± s.d) percent airway delivery for MTR and Hoe were 7.5%±2.5% and 7.1%±1.8%, respectively (based on 383 region measurements from a single lung). The images and measurements were consistent with local intracellular retention of MTR in the airways compared with Hoe, following IT (but not IV) instillation. These in vivo results paralleled the in silico simulation results (Figure 3).
In order to identify the most important parameters that might explain the differences in local retention of MTR and Hoe, a parameter exchange analysis was performed using computational simulations. For this purpose, individual parameters of the airway were exchanged with those of the alveoli, one at a time, and the simulations were rerun to calculate the exposure (AUC) of MTR and Hoe. Based on the results of this simulation analysis (Table 1) the volume of interstitial smooth muscle cells together with the volume of mitochondria were the primary factors determining the retention of MTR in the upper airways relative to alveoli. Secondarily, the surface areas of epithelial and endothelial cell layers were important, affecting retention in opposite directions. Taken together, these results suggest that the mitochondrial density per unit absorption surface area is the key histological organization parameter responsible for the higher retention of MTR in upper airways after IT administration.
In traditional pharmacokinetic studies, drug distribution in the lungs is analyzed in a homogeneous and well-stirred compartment [13], [24]. Here, we have elaborated an integrated, cell-based approach to model local drug absorption and transport phenomena, aimed at identifying cell-permeant molecules that are retained in the cells of the upper airway upon local pulmonary administration via the inhaled route. This integrated approach can be exploited for bioimaging probe development or for optimizing the local concentration of pulmonary medications [25], [26].
Locally acting, inhaled medications are of considerable interest for treating various pulmonary ailments, including asthma, chronic obstructive pulmonary disease (COPD) and pulmonary hypertension [11], [27], [28]. The therapeutic benefits of inhaled medications include targeted drug delivery, rapid onset of action, low systemic exposure with a resultant reduction in systemic side effects [29], . Nevertheless, measuring local drug concentrations in the lungs is challenging. Previously, regional differences in local lung exposure have received little attention in the context of small molecule targeting and delivery. Inhaled drug development efforts ignore the possibility that local differences in drug exposure could influence regional differences in drug transport properties that are associated with structural and functional characteristics of the airways and alveoli [31], [32], [33]. Accordingly, the approach presented here is significant because it furthers our understanding of how inhaled drug molecules and bioimaging probes behave after local administration to the lungs. These findings have important implications in pulmonary drug development.
Our simulations and experiments indicate that route of administration, histological organization and circulatory parameters can affect the retention and distribution of different molecular agents in various regions of the lung based on the lipophilicity and ionization properties of molecules, and as such, may be of pivotal importance for the optimization of drug targeting [25], [26]. Specifically, we considered two major and clearly distinguishable regions of the lungs: the airways and the alveoli, which are histologically and physiologically distinct. Extensive studies have demonstrated that the regional lung deposition of drugs is largely dependent on the aerodynamic particle size generated by delivery devices [28], [31], [34], [35]. Here, we introduce the concept that other parameters (e.g., the chemical properties of molecules) may be as important for predicting the behavior of pulmonary delivered of drugs. This is evidenced from our simulations which indicated that, after absorption into the blood, the majority of drug mass (>80% of total mass in lungs) is predicted to accumulate in the alveoli because of its larger volume and higher lipid content and compounds with high lipophilicity and low pKa will accumulate to even a greater extent in the alveoli. Although inhaled drug targeting leads to most of the drug mass deposited in the upper airways, without significant intracellular retention, the molecules can be rapidly absorbed and circulate back to the lung to accumulate in the alveoli. In theory, only molecules that are retained in the cells of the upper airways at the local site of administration can be effectively targeted to the upper airways.
To study the transport properties of small molecules in airways and in alveoli, we conducted simulations concentrated on characterizing the behavior of two fluorescent compounds, MTR or Hoe, because they exhibited large differences in simulated transport behaviors. In addition, two in vitro cell based assays were developed to test the local cellular uptake and retention properties of small molecules: 1) primary NHBE cell cultures comprised of cell multilayers differentiated on transwell insets in the presence of an air-liquid interface; and 2) MDCK cell monolayer cultures on microfabricted pore arrays to establish the lateral cell-to-cell transport kinetics of small molecules, along the plane of the cell monolayer. In the case of Hoe and MTR, both in vitro assays confirmed that the probes were taken up and largely retained by cells in the immediate vicinity of site of absorption and that the extent of diffusion followed a dye concentration gradient from the pores. Our in vitro findings indicated that the lateral cell-to-cell diffusion of MTR and Hoe was highly constrained. These in vitro results confirmed that both Hoe and MTR were retained intracellularly at a significant level in the presence of a transcellular concentration gradient both in the apical-to-basolateral and lateral directions. These results were also informative in terms of the time scale of intracellular accumulation and the relative labeling intensity afforded by these two fluorescent probes in the presence of a transcellular gradient. However, the in vitro assays did not reveal a major difference in the local retention of MTR and Hoe. Based on this observation, the behavior of these probes in these in vitro assays appeared most consistent with the predicted behavior of the probes in the alveoli.
Nevertheless, the results of in vivo studies closely paralleled those obtained in silico, in that MTR was retained in airways upon local IT administration while Hoe distributed in both airways and alveoli irrespective of the route of administration. Although in vitro results were useful to confirm the high, local intracellular retention of the probes, the in silico model is a better representation of the three-dimensional organization and physiological parameters of the in vivo situation. Parameter sensitivity analysis indicates that mitochondrial uptake of hydrophilic cations, in relation to the surface area over which absorption occurs, is the critical histological component responsible for high exposure of MTR when given via IT instillation. This is because as MTR traverses from the lumen of the airway into the interstitium, it is rapidly taken up into the mitochondria, driven by the high negative membrane potential of the mitochondrial inner membrane. Conversely, release of MTR from the mitochondria out into the circulation is very slow because the membrane potential slows its release. In the case of alveoli, the alveolar epithelial cells have much higher apical and basolateral plasma membrane surface areas relative to the mitochondrial membrane surface area. The higher cell surface areas facilitate mass transport of MTR across the cells and into the circulation, which reduces MTR accumulation in mitochondria.
In contrast to MTR, Hoe is a lipophilic weakly basic compound with a pKa of 7.5. Therefore, at physiological pH, half of the Hoechst molecules exist in a highly membrane-permeant, neutral form. Transmembrane diffusion of the neutral form of Hoe is orders of magnitude faster than that of a cationic form. So there is no significant accumulation or retention of Hoe in either the airways or the alveoli. When administered by IV injection, the direction of distribution is from blood to the tissue. The distribution between blood and tissue is mostly a function of the partitioning or binding of molecules from the circulation to the tissue, which is dependent on the cell density of the tissue, the membrane content of the tissue, and the affinity of the probes for membranes and intracellular components in the tissue. Thus, after IV administration, both Hoe and MTR tended to partition more into alveoli than into the airways.
In conclusion, we have elaborated an integrated in silico-to-in vitro-to-in vivo modeling approach which has applicability toward the optimization of site-specific targeting of locally-administered molecules. In the process, we have found that MTR is a candidate fiduciary marker for local drug deposition and absorption patterns in the airways. Due to the compartmental nature of the lungs, computational simulations can be linked to upstream process, such as pulmonary particle deposition, dissolution and mucus clearance, as well as to downstream processes that can be captured by pharmacodynamic models [36], [37], [38]. With additional effort this approach can be expanded to include macromolecules, acidic, zwitterionic molecules as well as molecules possessing multiple ionization sites, to further development of probes of lung structure and function [39], [40].
|
10.1371/journal.ppat.1007609 | Human IFIT proteins inhibit lytic replication of KSHV: A new feed-forward loop in the innate immune system | Kaposi’s sarcoma-associated herpesvirus (KSHV) is causally associated with Kaposi’s sarcoma, primary effusion lymphoma (PEL) and multicentric Castleman’s disease. The IFIT family of proteins inhibits replication of some viruses, but their effects on KSHV lytic replication was unknown. Here we show that KSHV lytic replication induces IFIT expression in epithelial cells. Depletion of IFIT1, IFIT2 and IFIT3 (IFITs) increased infectious KSHV virion production 25-32-fold compared to that in control cells. KSHV lytic gene expression was upregulated broadly with preferential activation of several genes involved in lytic viral replication. Intracellular KSHV genome numbers were also increased by IFIT knockdown, consistent with inhibition of KSHV DNA replication by IFITs. RNA seq demonstrated that IFIT depletion also led to downregulation of IFN β and several interferon-stimulated genes (ISGs), especially OAS proteins. OAS down-regulation led to decreased RNase L activity and slightly increased total RNA yield. IFIT immunoprecipitation also showed that IFIT1 bound to viral mRNAs and cellular capped mRNAs but not to uncapped RNA or trimethylated RNAs, suggesting that IFIT1 may also inhibit viral mRNA expression through direct binding. In summary, IFIT inhibits KSHV lytic replication through positively regulating the IFN β and OAS RNase L pathway to degrade RNA in addition to possibly directly targeting viral mRNAs.
| The innate immune response to infections is triggered by recognition of pathogens as foreign or non-self. Recognition of invading pathogens is carried out by various sensors or pattern recognition receptors (PRRs) that detect conserved features of pathogens including lipids, nucleic acids and proteins. PRR activation triggers pathways that ultimately lead to pathogen destruction, including the interferon response. Interferons, in turn induce many interferon-stimulated genes, which inhibit or destroy a wide variety of pathogens, including viruses. IFITs are a family of interferon induced proteins that are thought to recognize RNAs and have antiviral effects primarily on RNA viruses. Kaposi’s sarcoma-associated herpesvirus (KSHV), a DNA virus, is associated with Kaposi’s sarcoma and lymphoid malignancies. In this study we show that IFITs restrict replication of KSHV and does so not only by inhibiting KSHV mRNA abundance but also by enhancing other effectors of the interferon response. This study reveals that the innate immune response can control not only invading viruses but ones that reactivate from latency, that IFITs can inhibit herpesvirus replication and that IFITs may amplify the innate immune response by a feed-forward mechanism.
| Kaposi’s sarcoma-associated herpesvirus (KSHV, HHV8) is causally associated with Kaposi’s sarcoma (KS), primary effusion lymphoma (PEL) and multicentric Castleman’s disease (for a review, see reference [1]). KSHV maintains a persistent latent infection in B lymphocytes, from which it occasionally reactivates, enters a lytic cycle of replication, and produces infectious virions. Transmission occurs by both sexual and nonsexual contact as well as blood and organ transfusion. Cell-mediated immunity is important for limiting KSHV reactivation and pathogenesis. KSHV infection activates several pattern recognition receptors (PRRs), including cGAS, IFI16, RIG-I, NLRP1, and several Toll-like receptors (TLRs) which play an important role in promoting the innate immune response [2–7]. KSHV pathogen associated molecular patterns (PAMPs) recognized by the innate immune system remain to be fully characterized but are primarily thought to reside on viral glycoproteins and nucleic acids [8].
Much of the work done on the innate immune response to KSHV has used systems in which the cellular response to incoming virus has been examined. These studies have shown that cytoplasmic and endosomal viral nucleic acids may be detected by one or more PRRs [9] and that viral glycoproteins may activate PRRs upon viral entry [9–12]. Several DNA sensors may be important in recognition of viral DNA, including cyclic GMP synthetase (cGAS) and IFI16. IFI16 has recently been shown to act in the nucleus to activate a nuclear inflammasome in response to KSHV and EBV infection [2, 13, 14]. cGAS mediated recruitment of STING and IRF3 activation requires association with a ribonucleoprotein complex which is remodeled by foreign DNA [15]. Several components of these cytoplasmic innate immune pathways are involved in the innate immune response to KSHV. TLR9 appears to act as a sensor for incoming KSHV DNA and partly contributes to the activation of IFN-α [10]. Both NLRP1, a protein component of the inflammasome, and IFI16 may restrict KSHV reactivation, since depletion of NLRP1 or IFI16 results in increased lytic replication [3, 16]. Similarly, RIG-I, a cytosolic RNA sensor, may be important in limiting KSHV reactivation from latency as KSHV lytic replication was enhanced in RIG-I-/- cells. The importance of these pathways is emphasized by the fact that KSHV counteracts the host response via vIRF-1, LANA and ORF52 [4, 6, 17].
The PAMPs displayed by herpesviruses are predicted to differ considerably depending on the stage and type of their replicative cycles. During latency, herpesviruses exist as chromatinized nuclear episomes. During stringent latency, few, if any, lytic proteins or RNAs are synthesized, and virion DNA is not produced [1]. In contrast, once reactivation occurs, and the virus enters the lytic replicative cycle, abundant amounts of viral mRNAs and non-coding RNAs are produced and newly replicated genomes are produced, encapsidated, and egress from the nucleus prior to tegumentation, final envelopment and transit through the plasma membrane. Thus, while virion DNA is not expected to be exposed to endosomes and cytoplasm in the same context as during primary infection of the cell, there is nevertheless ample of opportunity for virus components such as mRNAs, non-coding RNAs, and viral proteins to be detected by cytoplasmic PRRs and trigger innate immune responses. We therefore wished to extend our study of host cellular factors to cytoplasmic PRRs that could contribute to establishment of an antiviral state and restrict lytic KSHV replication and reactivation from latency. These include human IFIT proteins that have been recently demonstrated to play important roles in the inhibition of several viruses besides herpesviruses [18, 19].
The IFN induced tetratricopeptide repeat containing proteins (IFITs) are among the most highly interferon-induced proteins [20]. They constitute a family of related genes that have been identified in a wide variety of mammals from mouse to man [21]. The human genes, encoded on chromosome 10, are IFIT1 (ISG56), IFIT2 (ISG54), IFIT3, IFIT5 and IFIT1B [21]. Several of the IFITs have been implicated in an important antiviral response pathway dependent on recognition of foreign RNAs. In mice, Ifit1, a homolog of human IFIT1B, specifically recognizes uniquely modified viral RNAs that lack 2'O-methylation of their 5' mRNA caps (cap0-mRNAs) [22, 23]. Viruses that replicate in the cytoplasm that have either “snatched” a cap or encode their own 2”O-methyltransferase may thereby evade recognition as non-self [24]. In humans however, IFIT1 protein differs significantly from IFIT1B, and may play a broader role in antiviral function, with different RNA binding specificities [21]. IFIT1 forms a tripartite complex with IFIT2 and IFIT3 and binds to 5’ mRNA caps [25]. In addition to inhibiting replication of viruses that are predicted to have 2’O-methylated caps, IFIT1 inhibits papillomavirus replication by binding its E1A protein [26]. IFIT1 inhibits translation of viral mRNAs by preferentially binding their 5’ cap and preventing association with eukaryotic initiation factors [23]. In addition, IFIT1 has been shown to affect protein translation by interacting with eukaryotic initiation factor eIF3 and may thereby inhibit additional virus families by different mechanisms [18]. It was therefore of interest to determine whether IFIT proteins could inhibit KSHV, whose replication strategy differs considerably from other virus families in which IFITs have been shown to exert antiviral activity.
In this study, we examined the effect of IFIT proteins on KSHV replication by depleting IFITs under conditions whereby highly efficient KSHV lytic replication and infectious virion production was enabled in epithelial cells. KSHV lytic gene expression, DNA replication and virion production were enhanced by depletion of IFITs. Further, IFIT expression, which was undetectable during latent infection, was induced during the course of KSHV reactivation and lytic replication. Using deep sequencing of mRNA, we analyzed the effects of IFITs on KSHV and cellular transcript accumulation during lytic KSHV replication. In addition to IFIT effects on the viral transcriptome, we discovered an unexpected positive effect on the expression of other members of the interferon-induced response that is predicted to amplify the antiviral effect of IFIT proteins.
IFIT genes are strongly induced by several viral infections, including the betaherpesvirus hCMV [18, 19, 27]. Although several innate immune pathways are induced by KSHV infection or reactivation, it was not known whether IFITs were induced by KSHV lytic replication during reactivation from latency. We therefore examined the status of IFIT protein expression in iSLK/Bac16 cells stably transduced with a doxycycline-inducible viral transactivator, KSHV ORF50/Rta [59]. These Rta-inducible SLK cells (iSLK) are stably and latently infected with the Bac16 KSHV strain that expresses hygromycin resistance and GFP, and robust and synchronous reactivation of KSHV from latency is achieved by doxycycline treatment. [30]. Infected cells were 100% GFP positive when maintained under hygromycin selection (S1 Fig). iSLK/Bac16 cells were treated with doxycycline and cells were harvested at serial time points from 0–72 hrs at 12 hr intervals. We measured IFIT1 and IFIT3 protein levels in iSLK cells by Western blotting. Both IFIT1 and IFIT3 were not detectable in uninduced cells but were expressed after KSHV reactivation (Fig 1 and S2 Fig). IFIT1 protein was first detectable at 36hr post induction (p.i.) and continued to increase to 72 hr. IFIT3 was first detectable at 12 hr and reached peak expression by 48 hr. (Fig 1A and 1B). In order to confirm that exogenous Rta expression itself did not affect IFIT expression, we also assessed IFIT1 and IFIT3 expression in doxycycline treated or untreated uninfected iSLK cells. Neither IFIT1 and IFIT3 was detectable at 48 hr and 72 hr (S3 Fig). Induction of KSHV Bac16 lytic replication was confirmed by immunoblotting for ORF57 which was expressed by 12hr after induction (Fig 1C). qPCR also showed IFIT1, IFIT2 and IFIT3 expression peaking by 36hr which was consistent with the results of Western blotting (Fig 1D). These results clearly demonstrate that KSHV lytic replication induces IFIT expression.
In order to confirm induction of IFITs as a result of KSHV replication and to determine their cellular location, we examined lytically induced KSHV infected cells by immunofluorescence microscopy. iSLK cells were grown on glass coverslips and treated with 1 μg/ml doxycycline to induce virus lytic replication. Cells were fixed at 48hr, 72hr and 96hr post induction. Immunofluorescence staining for IFIT3 was performed and revealed cytoplasmic expression (Fig 2A). A small percentage of cells were IFIT3 positive before induction, possibly in cells which undergo spontaneous lytic gene expression. However, the percentage of cells expressing IFIT3 increased progressively after lytic induction, and was approximately 45-fold higher by 96 hrs (Fig 2B). These results confirmed the immunoblotting data and demonstrate that IFIT3 is expressed in the cytoplasm. IFIT1 exhibited similar cytoplasmic localization to IFIT3 (S4 Fig).
In order to investigate IFIT1, IFIT2 and IFIT3’s effect on KSHV lytic replication and reactivation from latency, we measured virion production in KSHV infected cells (iSLK/Bac16) after depletion of IFITs. IFIT depletion was carried out by lipid-mediated transfection of iSLK cells with siRNA specific for IFITs (Fig 3A–3C). Cells were transfected with siRNAs 6 hr prior to inducing lytic replication. We first measured IFIT RNA abundance at 48 hr post induction by qPCR (Fig 3A). Approximately 90% depletion of IFIT mRNAs was apparent in cells induced to permit KSHV replication. We next examined expression of IFIT1 and IFIT3 by immunoblotting from samples harvested at 48 hr post induction. Approximately 90% depletion of IFIT1 protein was achieved by 48 hr post-transfection as assessed by densitometry of the Western blot (Fig 3B). There was approximately 73% depletion of IFIT3 (Fig 3C). In order to assess the effect of IFIT depletion on KSHV reactivation and virion production, cells were transfected with either IFIT siRNAs or control siRNA, and 6 hours later KSHV reactivation was induced by addition of doxycycline. Virion-containing supernatant was harvested at 120 hours after induction of lytic replication. Infectious virus production was measured by infection of 293T cells with serial dilutions of virus supernatant followed by flow cytometry of infected cells. Virus titer in the supernatant can thus be accurately quantitated as GFP-transducing units [29]. As shown in Fig 3D and S5 Fig (IFIT KD and virion titration repeated in a separate experiment), IFIT depletion led to a marked increase in virion production (25–32 fold), compared to control siRNA-transfected cells induced in parallel. There was no microscopically detectable release of infectious virus in the absence of doxycycline from either IFIT depleted cells or in control cells, indicating that Rta is still absolutely required for lytic replication. These data indicate that the IFITs act as a restriction factor for KSHV virus production.
IFIT1, IFIT2 and IFIT3 may form a tripartite complex and cooperate in RNA binding [23, 25, 30]. Therefore, we performed virion titration as was done in the previous experiments to examine the effect on virion production of individual depletion of each IFIT. Individual depletion of the three IFITs had a similar effect on virion release, with each IFIT depletion leading to a marked increase in virion production (about 22–25 fold) (Fig 3E). The magnitude of this effect is similar to that observed upon depletion of all three IFITs together. These data suggest that each of the three IFITs is important for restriction of KSHV virus production.
We next wished to ask at which stage of KSHV lytic replication IFITs might be exerting an inhibitory effect on KSHV virion production. To determine whether the IFIT effect was due to inhibition of KSHV DNA replication, we measured KSHV genome abundance by qPCR on DNA samples from cells that were induced to replicate after depletion or mock depletion of IFITs. The results demonstrated that intracellular KSHV genome copy numbers increased at least 9-fold upon IFIT KD (Fig 3F). However, this increase was not as large as the increases observed in infectious virus titer (Fig 3D, S5 Fig), suggesting that IFITs may affect other steps in the lytic KSHV cycle in addition to DNA replication.
IFIT depletion enhanced KSHV virion production (25–32 fold) while KSHV DNA copy number increased only 9-fold, suggesting that IFITs may restrict expression of late genes that are needed for virion formation, egress or infectivity. In order to assess the global effect of IFITs on KSHV lytic gene expression, we performed high-throughput deep sequencing of mRNA from KSHV-infected cells in which IFITs were depleted prior to induction of lytic replication. KSHV-infected iSLK cells were transfected with either control siRNA or IFIT siRNAs as was done in the previous experiments to examine the effect on virion production. Six hours after siRNA transfection, cells were treated with doxycycline to induce KSHV lytic replication, and cells were harvested at 48 hours post induction, RNA was isolated, oligo-dT selected, and processed for deep sequencing. The effects of IFIT KD on lytic cycle transcription were compared to the transcriptional profile of induced cells transfected with control siRNA. A comparison of the transcriptional profiles is presented in Fig 4A. Consistent with its effect on virus production, IFIT KD was associated with broad enhancement of KSHV lytic gene expression. Approximately two thirds of genes demonstrated increased expression: 6 genes increased > 2-fold and 35 genes increased 1.2-2-fold. 30 genes did not exhibit increases (less than 20% change) while 15 genes decreased 20% -85% (Fig 4B). Of the six genes whose expression was increased more than two-fold, 5 are involved in lytic KSHV DNA replication: ORF56, the helicase primase; ORF54, the deoxyuridine triphosphatase; ORF6, the single-stranded DNA-binding protein; ORF70, thymidylate synthase; and ORF57, a post-transcriptional regulator that preferentially enhances mRNA accumulation of several genes involved in DNA replication [31]. The sixth gene ORF47, encodes glycoprotein L. The preferential enhancement of genes involved in KSHV DNA replication is consistent with the effect of IFIT KD on KSHV DNA replication shown above. However, the increased expression of the majority of KSHV lytic genes suggested that IFITs may have general effects beyond inhibition of specific KSHV genes.
In order to determine whether IFITs have a generally inhibitory effect on KSHV replication, we set out to examine the effect of IFIT KD in the BCBL1 KSHV-infected primary effusion lymphoma (PEL) cell line. The TREx BCBL1-Rta cell line employed carries a doxycycline-inducible Rta gene, allowing robust KSHV lytic replication upon doxycycline treatment [32], kind gift of Jae Jung.
TRExBCBL1-Rta cells were treated with doxycycline and cells were harvested at serial time points from 0–48 hrs at 12 hr intervals. IFIT1 and IFIT3 protein were not detectable (S6A Fig, S6B Fig) by Western blotting although KSHV ORF57 was strongly induced (S6C Fig). IFIT expression in iSLK Bac16 cells was easily detectable under the same conditions. These results indicated that the innate immune response in TRExBCBL1-Rta is different from that in iSLK/Bac16, consistent with prior reports that IFIT expression is nonfunctional in a large percentage of cancer cell lines and primary cancer cells [33, 34]. Regardless, we asked whether KD of IFIT mRNAs could affect KSHV gene expression even though IFIT protein expression was undetectable by immunoblotting. Lentiviruses containing shIFIT1 were constructed and tested by infecting induced iSLK/Bac16. Three clones of shIFIT1s (258, 316 and 581) achieved efficient knockdown of IFIT1 in iSLK/Bac16 without cellular toxicity (S6D Fig). These three lentiviral preparations were combined and used to infect TRExBCBL1-Rta at an MOI of 15. After infection by lentivirus, cells were either treated or mock-treated with doxycycline and harvested at 48hr post induction. Flow sorting was performed to confirm transduction by GFP-expressing lentivirus. Lysates from control or IFIT1 shRNA transduced TREX BCBL1-Rta cells showed no differences in ORF57 (early) or K8.1 (late) lytic protein expression (S6G Fig and S6H Fig). These results indicate that IFITs are not expressed and do not restrict KSHV lytic replication in BCBL1 cells.
IFIT depletion enhanced KSHV DNA replication and upregulated viral mRNA expression, contributing to increased virion production. We also wished to determine the effects of IFIT KD on cellular gene expression. There were 99 cellular genes whose transcript abundance decreased by 50% or more upon IFIT KD when KSHV lytic replication was induced. Analysis of this gene set by GO Enrichment Analysis (http://geneontology.org/page/go-enrichment-analysis) [35–37] showed high enrichment in genes assigned to the type 1 interferon pathway. 11 of the 99 genes whose transcript abundance decreased on IFIT KD were assigned to this pathway (Table 1). This represents a 35 fold enrichment over expected (p value 6.94 X 10−14). Among these ISGs, the OAS family (OAS1, OAS2, OAS3, OASL) was most significantly enriched. We performed qPCR for these OAS genes to validate and confirm the RNA Seq data (Fig 5A). All OAS genes were significantly down regulated (p<0.0002) to less than 17% after IFITs were depleted compared to mock depletion, suggesting that IFITs may affect KSHV replication through the OAS-RNase L pathway. Since several other ISGs belonging to the type 1 interferon pathway were also downregulated by IFIT KD, we performed qPCR to measure the expression of the most upstream regulator, IFN β. Lytic replication of KSHV induced IFN β expression in the control transfection (Fig 5B). IFN β expression was downregulated significantly (p<0.0001) to 7% in IFIT KD cells compared to the control (Fig 5B). Therefore, IFITs appear to enhance IFN β production and downstream ISG expression during KSHV lytic replication, and depletion of IFITs results in a blunted type 1 interferon response.
The 2’,5’-oligoadenylate (2-5A) synthetase (OAS)-RNase L system is an interferon-induced antiviral pathway. Induction of OAS proteins by IFN and viral replication leads to synthesis of 2’5’ oligoadenylates which activate RNase L. OASs (OAS1, OAS2, OAS3) synthesize 2’-5’ oligoadenylates and activate RNase L leading to degradation of viral and cellular RNAs, thereby restricting viral infections [38]. Activated RNase L preferentially cleaves target RNAs produced by viruses as well as several cellular RNAs at specific sites [39, 40]. To confirm that the OAS expression that was inhibited by IFIT KD was functionally relevant, we compared RNase L activity in IFIT depleted and mock-depleted cells. Donovan et al. have demonstrated that RNA cleavage by activated RNase L can be quantitatively measured by RtcB-ligase assisted qPCR [40]. In this assay, RtcB ligase, which is capable of ligating 2’,3’-cyclic phosphates (generated by RNase L cleavage) to 5’OH RNAs, is used to ligate all such ends in the total cellular RNA pool to an RNA-DNA adapter with a 5’OH group. The ligated RNA is then reverse-transcribed and the cDNA is analyzed by qPCR. By using forward primers complementary to specific individual cleavage sites in known RNase L targets, measurement of the cleavage at each such site is achieved, serving as a quantitation of RNase L activity (Fig 6A). In order to serve as an internal control, U6 RNA, which has a naturally occurring cyclic 2’3’ phosphate, was also analyzed and used for normalization. We prepared and purified recombinant RtcB (S7 Fig) and then performed a ligation-PCR assay to measure the effect of OAS downregulation on RNase L activity. The RtcB analysis demonstrated that upon IFIT KD, RNase L cleavage decreased significantly at site 36 in tRNA-His, site 27 in non-protein-coding RNA RNY4 and site 30 in non-protein-coding RNA RNY5. The differences between RNase L directed cleavage in the presence and absence of IFITs were statistically significant as shown in Fig 6B–6D. Consistent with lower cleavage activity of RNase L, total RNA yield increased significantly upon IFITs KD compared to the control (Fig 6E). Thus, IFITs may inhibit KSHV replication through the OAS-RNase L pathway.
IFIT1 has been shown to preferentially recognize certain types of capped RNA [12, 41]. Recently published data indicate that IFIT3 stabilizes IFIT1 and increases its affinity for cap0 mRNAs (See Fig 7 for a diagram of the various types of RNA 5’ caps) [41–43]. However, IFIT1 is also capable of binding cap1 mRNAs (Fig 7), albeit at lower affinities [41]. It was therefore possible that IFIT1 might recognize and inhibit translation or stability of KSHV mRNAs directly in addition to indirect effects mediated via other ISGs as shown with OASs. We performed immunoprecipitation experiments to determine if IFIT1 bound viral mRNA specifically or preferentially. iSLK cells were treated with doxycycline to induce KSHV replication, lysates were harvested at 48 hr post induction and immunoprecipitated with IFIT1 and IFIT3 antibodies. Immunoprecipitated RNAs were isolated and measured by qPCR. As shown in Fig 8A, viral RNAs were enriched 5~11-fold in the immunoprecipitation using IFIT1 and IFIT3 antibodies compared to control IPs. The cellular GAPDH mRNA was enriched 4.8-fold. It should be noted that the degree of binding of individual RNAs to IFITs was not related to their overall abundance. For example, ORF6, which was present at 330 FKPM was enriched similarly to K4, which was highly abundant at 50,000 FKPM (Fig 8B). Uncapped MT-ADP6 RNA, a cellular mitochondrial transcript [44], as well as uncapped snoRNAs U15 [45] and U16 [46, 47] were enriched only 1.7-fold (Fig 8A). U6 RNA, which has a gamma-monomethyl phosphate cap [48], was enriched in only 2.2-fold in the immunoprecipitates (Fig 8A). U1, U2 and U5 have a trimethylation cap [49], and they were similarly enriched less than 1.8-fold (Fig 8A). All these snoRNA are expressed at more than 10,000 copies per cell which is a much higher abundance compared to cellular genes [50, 51]. Therefore, IFIT1 and IFIT3, while preferentially recognizing cap0 structures, are also capable of binding to both viral and cellular capped mRNA (cap1 or cap2). However, their ability to bind uncapped, monomethyl capped, or trimethyl capped RNAs appears to be extremely limited.
In this study we examined the effects of cellular IFIT1, IFIT2 and IFIT3 on KSHV lytic replication. Heretofore, IFIT proteins have been primarily implicated in antiviral responses against RNA viruses [18, 19]. IFIT1, IFIT2 and IFIT3 form multimeric complexes that initially were shown to bind 5’ tri-phosphate RNAs [25]. Such RNAs are produced by several negative strand RNA viruses such as Rift Valley virus, vesicular stomatitis virus and influenza virus against which IFITs exhibit antiviral activity [19]. However, IFITs were subsequently shown to also preferentially bind cap0 mRNAs which lack 2’O-methylation at the first and second transcribed ribonucleotides (as seen in cap1 and cap2 mRNAs, Fig 7). Many viruses that replicate in the cytoplasm, including flaviviridae, poxviridae and coronaviridae, have evolved enzymes to independently perform 2’O-methylation of their mRNAs. Mutants of these viruses that have lost 2’O methyltransferase activity exhibit increased susceptibility to IFIT dependent immune responses, suggesting that IFITs allow discrimination between self and non-self RNAs [12, 52]. Although IFIT induction by RNA viruses is common, herpesviruses may also induce IFIT gene expression, by direct or indirect mechanisms. Indeed, IFIT2 and IFIT3 were identified as hCMV induced genes (cigs) over twenty years ago by the use of differential display [27]. HSV infection also leads to IFIT induction, albeit less strongly than CMV infection [27]. It was therefore of interest to determine whether a gammaherpesvirus such as KSHV could induce IFIT gene expression. KSHV, similar to other herpesviruses which undergo lytic replication in the nucleus, are presumed to have cap structures similar if not identical to host mRNAs [24]. Nevertheless, IFIT1 has also been shown to exert inhibitory effects on translation independent of mRNA sequestration by interacting directly with eIF3 [18], also raising the question of whether IFITs could establish an antiviral state that would inhibit KSHV virion production.
We first established that KSHV reactivation and lytic replication results in IFIT induction. While all three IFIT mRNAs were measurably induced upon KSHV lytic replication, we were only able to detect increased expression of IFT1 and IFIT3 proteins. While the IFIT2 antibodies we employed were able to detect exogenously overexpressed IFIT2, we did not detect IFIT2 protein expression by either immunoblotting or immunofluorescence microscopy. This may be due to minimal IFIT2 protein induction as a consequence of KSHV replication but our finding that IFIT2 depletion enhanced KSHV production suggests that functional IFIT2 is present and the failure to detect IFIT2 protein is likely due to inadequately sensitive IFIT2 antibodies. Nevertheless, it is clear that although KSHV lytic reactivation occurs from the nucleus, PAMP exposure sufficient to engage PRRs and induce ISG expression occurs. Future studies to examine the nature of the non-self signatures, whether DNA, RNA or protein, that evoke the innate immune response to reactivating herpesviruses, and whether the PRRs that recognize them are nuclear and/or cytoplasmic, will be very informative. Both IFIT1 and IFIT3 proteins that we detected by IF studies were localized to the cytoplasm in KSHV infected cells. Although virtually all studies have focused on the interaction of IFITs with cytoplasmic RNAs, it has been suggested that IFIT1 may also have transcriptional activating functions [53].
We examined the potential role of IFITs as KSHV inhibitory proteins by knocking down IFITs and then inducing KSHV lytic replication. As expected, IFIT expression was minimal in the absence of KSHV replication, and expression of IFITs was also severely curtailed after siRNA treatment. Blocking IFIT production resulted in a 25–30 fold increase in infectious KSHV virion production. Consistent with these findings, IFIT KD also led to increased lytic KSHV mRNA accumulation. Importantly, the increase in mRNA abundance, while not completely equal amongst all KSHV mRNA, was nevertheless broad, with over 65% of lytic mRNAs increasing in abundance. However, these findings pose a difficulty in interpretation due to the fact that herpesvirus late gene transcription, including that of KSHV, is dependent on DNA replication [54]. Since the most highly IFIT restricted KSHV mRNAs encode proteins that are either essential or important for KSHV lytic DNA replication, the broad inhibitory effect on KSHV mRNAs may be partly indirect, with late gene repression by IFITs due to the inhibitory effect of IFITs on viral DNA replication. We confirmed that IFITs do have an inhibitory effect on DNA replication by directly measuring KSHV DNA abundance in the presence and absence of IFITs.
KSHV is thought to most likely enter the human host by oral epithelial cell infection [55] and has been demonstrated to infect a variety of human epithelial cells including oral keratinocytes as well as epithelial cell lines [56, 57]. These experiments were carried out in the iSLK/Bac16 cell line, an epithelial cell line that supports efficient KSHV lytic replication and has served as a model for KSHV infection and reactivation from latency [58]. We also examined the potential role of IFITs in KSHV reactivation from latent infection in a PEL cell line, BCBL1 [59]. However, cells from BCBL1 did not express detectable IFITs and it was therefore not possible to determine whether IFITs may play a role in restricting KSHV replication in B lymphocytes. As loss of ISG expression is not uncommon in many human tumors, these findings do not rule out the possibility of IFITs playing a physiological role in regulating KSHV replication in B lymphocytes in vivo.
These data also raised the possibility that IFITs exerted at least some of their antiviral function by other indirect mechanisms that did not depend on specific targeting of KSHV mRNAs, especially since there are no known differences in cap structures between herpesvirus mRNAs and host cellular mRNAs [24]. Our analysis of the cellular transcriptome suggested an effect of IFITs on the type I interferon pathway, as several ISGs, including several known to be important for establishing an antiviral state, decreased upon IFIT KD in comparison to infected cells in which IFITs were not depleted. These findings do not differentiate between transcriptional or post-transcriptional effects of IFITs in enhancing ISG expression. However, the fact that interferon mRNA levels were decreased in the absence of IFITs suggests that the simplest model for a positive feedback loop maintained by IFITs might be an effect on type I IFN transcription or RNA stability. A recent report implicated IFIT1 in nuclear regulation of transcription, both acting to negatively regulate the inflammatory response as well as enhancing IFN β1 transcription [53]. The IFN β 1 response to pathogens was also blunted in IFIT1-depleted cells in this study. Our data support a model in which IFITs maintain an antiviral state by promoting enhanced IFN and ISG production.
OAS proteins are established components of the innate immune response to viruses. OASs synthesize 2’-5’ oligoadenylates and activate RNase L leading to degradation of viral and cellular RNAs and thereby block viral infections as well as amplification of IFN α/β by RNase L-generated small RNAs [38]. Because OAS mRNAs were the most highly downregulated upon IFIT KD, we examined whether this correlated with a functional decrease in potential antiviral activity. By using the RtcB-ligase assay, which measures the production of cyclic 3’ phosphate moieties at specific RNase L cleavage sites, we were able to determine that ISG KD does lead to a functional decrease in RNase L activity. The observed decrease in RNase L activity is consistent with the generalized decrease in KSHV lytic mRNA abundance and in total cellular RNA.
We also examined the ability of IFITs to bind several types of capped RNA. By direct immunoprecipitation of IFIT proteins, we found that cap1/2 host cell mRNAs and KSHV mRNAs were widely represented in IFIT immunoprecipitates. Despite the fact that IFIT complexes were originally isolated by using triphosphate uncapped RNAs as bait, we found very little representation of naturally uncapped RNAs such as certain mitochondrial or snoRNAs [44–47, 60] in IFIT immunoprecipitates. Whether interactions of IFITs with cap1/2 mRNAs exert negative effects on their stability or translation or if such interactions could even have positive effects on target mRNA remains to be determined. Although the enhanced affinity of IFIT1 complexes for cap0 mRNAs has been adduced as evidence of a PAMP recognition by IFITs that allows them to distinguish between self and non-self, the fact remains that the majority of viral mRNAs are 2’-O-methylated. In addition, some viruses, such as parainfluenza virus, whose mRNAs are 2’-O-methylated, are nevertheless inhibited by IFITs [61]. Our findings demonstrating that IFITs exert antiviral effects on KSHV, a herpesvirus, which has neither genomic RNAs nor atypically capped mRNAs, provide further evidence that IFITs may have antiviral effects beyond direct sequestration of mRNA. Although IFIT complexes do not appear to be highly enriched for specific transcripts, we cannot rule out the possibility that sequestration by IFITs may have varying effects on different targets, affording a degree of specificity. Variation in the effects of IFIT binding to different targets could arise from intrinsic differences in translatability or stability of individual target mRNAs, especially lytic herpes virus transcripts, which are primarily intronless [62, 63].
In summary we have shown that IFITs exert an antiviral effect on a herpesvirus which does not express any of the putative pathogen associated RNA signatures expressed by RNA viruses. In addition, in infected cells, IFITs do associate with canonically capped viral and cellular mRNAs that are not known to be possess cap0 structures. Further, IFIT depletion led to decreases in IFNβ as well as several other antiviral effectors of the interferon pathway, suggesting that IFITs may possess broad antiviral effects by virtue of their ability to amplify the interferon response. Finally, by virtue of the IFITs ability to interact with canonically capped cellular and viral transcripts, they may also affect both host cell and viral gene expression by direct effects on mRNA.
293T cells (kind gift of Lori Frappier, University of Toronto) were grown at 37°C in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% fetal bovine serum (FBS) and glutamine. iSLK cells [58] (gift of Don Ganem, UCSF) were maintained in DMEM containing 10% charcoal stripped FBS (Sigma) and 1% glutamine with 250μg/ml G-418 and 1μg/ml puromycin. iSLK cells were infected with WT KSHV derived from bacmid BAC16, expressing eGFP and hygromycin resistance [28]. Bac16 KSHV infected iSLK cells were maintained in 1.2 mg/ml hygromycin, 250μg/ml G-418 and 1μg/ml puromycin. TRExBCBL1-Rta (kindly provided by Prof. J. Jung) were cultured in RPMI 1640 10% Tet System approved FBS (Clontech) and 1% glutamine with 50μg/ml hygromycin.
IFIT1 (L-019616-00-0005), IFIT2 (L-012582-02-0005), IFIT3 (L-017691-00-0005) and negative control On-target plus Smart Pool siRNAs (D-001810-03) were purchased from Thermo Scientific. Each siRNA was transfected into iSLK cells using Lipofectamine RNAiMAX (Invitrogen) according to the manufacturer’s protocol. For KD of all three IFITs, each siRNA was used at 10 nM final concentration and NC siRNA was used at 30 nM. For individual IFIT KD, each siRNA or NC siRNA was used at 10 nM. RT-qPCR or immunoblotting was performed to verify knockdown of the relevant protein.
IFIT1 GIPZ shRNA clones (RHS4531-EG3434) were purchased from Dharmacon. Lentiviruses were prepared by transient transfection of 293T cells with a three-plasmid system (a GIPZ plasmid expressing shRNA against cellular IFIT1; pMD2.G [envelope plasmid expressing vesicular stomatitis virus glycoprotein]; and psPAX2 [packaging plasmid]). Viral supernatant was harvested at 48hr post transfection with 0.45μm syringe filter. Lentivirus were concentrated at 10000g in 10% sucrose buffer for 3.5hr as described [64] and immediately used to infect iSLK/Bac16 or TREx BCBL1-Rta cells. iSLK/Bac16 cells were infected with each lentivirus at an MOI of 15 and then induced to permit KSHV lytic replication. Cells were harvested 48hr post-induction and Western blotting of IFIT1 was performed. Lentiviruses containing three independent shRNAs were used to infect TRExBCBL1-Rta cells at an MOI of 15. For shRNA knockdown experiments, TRExBCBL1-Rta cells were infected twice with concentrated lentiviruses within two days. 2 days after the second infection, cells were seeded at 1 × 106/ml and induced with 1μg/ml doxycycline. Cells were harvested or sorted by flow cytometry at 48 post induction.
Total cellular RNA was isolated from washed cell pellets using Qiazol and Qiagen miRNeasy columns according to the manufacturer’s protocols. mRNA was purified from 6 μg cellular RNA using Qiagen Oligotex mRNA Minikit (Qiagen). cDNA libraries were prepared using the ABI high Capacity cDNA Reverse Transcription Kit with RNase inhibitor (Applied Biosystems). Real-time quantitative PCR (qPCR) was performed with SYBR green PCR Master Mix (Applied Biosystems) according to the manufacturer’s protocol. Each sample was analyzed in triplicate with gene specific primers and β-actin was used as the endogenous control. The gene-specific primers were as follows:
IFIT1 Q1F: 5’-ggaatacacaacctactagcc-3’;
IFIT1 Q1R: 5’-ccaggtcaccagactcctca-3’;
IFIT2 Q1F: 5’-gggaaactatgcctgggtc-3’;
IFIT2 Q1R: 5’-ccttcgctctttcattttggtttc-3’;
IFIT3 Q1F: 5’-tgaggaagggtggacacaactgaa-3’;
IFIT3 Q1R: 5’-aggagaattctgggttgttgggct-3’
OAS1 Q1F: 5’-gcgccccaccaagctcaaga-3’
OAS1 Q1R: 5’-gctccctcgctcccaagcat-3’
OAS2 Q1F: acccgaacagttccccctggt-3’
OAS2 Q1R: 5’-acaagggtaccatcggagttgcc-3’
OAS3 Q1F: 5’-tgctgccagcctttgacgcc-3’
OAS3 Q1R: 5’-tcgcccgcattgctgtagctg-3’
OASL Q1F: 5’-gcggagcccatcacggtcac-3’
OASL Q1R: 5’-agcaccaccgcaggccttga-3’
ORF6 Q1F: 5’-ctgccataggagggatgtttg-3’
ORF6 Q1R: 5’-ccatgagcattgctctggct-3’
ORF47 Q1F: 5’-agcctctaccctgccgttgttct-3’;
ORF47 Q1R 5’-acgaccgcgactaaaaatgacct-3’;
ORF54 Q1F: 5’-gtagccgcatatgccagattgtg-3’
ORF54 Q1R: 5’-ttttgaagcccttgaggatgtgtc-3’
ORF56 Q1F: 5’-cacagattcccgtcaatacaaa-3’;
ORF56 Q1R, 5’-gtatcttcagtaggcggcagag-3’;
ORF57 Q1-5: 5’-gcagaacaacacggggcgga-3’
ORF57 Q2-3: 5’-gtcgtcgaagcgggggctct-3’
ORF70 Q1F: 5’-gactatacaggccaggggtttgac-3’
ORF70 Q1R: 5’-ggcgggttccacgcacac-3’
K4 Q1F: 5’-gtttgcaatctggggacacg-3’
K4 Q1R: 5’-tggtaaccgagacagcacttg-3’
β-actin Q1F: 5’-tcaagatcattgctcctcctgag-3’
β-actin Q1R: 5’-acatctgctggaaggtggaca-3’
High-throughput deep sequencing of RNA was performed as previously described [29] with some modifications. Briefly, iSLK cells were transfected with a mixture of 10 nM final concentration of each IFIT siRNA or negative control siRNA (30nM final concentration) and were treated with 1 μg/ml doxycycline after 6 hrs. Cells were harvested at 48 hr post induction for RNA isolation. RNA samples from iSLK cells were prepared using Qiagen miRNeasy kits according to the manufacturer’s protocols. cDNA libraries were prepared from poly(A) RNA and were sequenced on a HiSeq2000 instrument with 50 cycle single end reads. Sequenced reads obtained from Bac16 KSHV-infected iSLK cells were aligned to the KSHV Bac16 (GenBank accession no. GQ994935.1) and Hg19. Differential gene expression was measured using USeq’s Defined Region Differential Seq application as described previously [29].
Protein samples were analyzed by sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) and immunoblotted with rabbit polyclonal anti-IFIT1 antibody (PA3-848, ThermoScience), rabbit polyclonal anti-ORF57 antibody [65], mouse anti-K8.1 monoclonal antibody (kind gift from Bala Chandran), anti-α tubulin polyclonal antibody (Sigma) or mouse anti-actin monoclonal antibody (MA5-15739, Thermo Science) and horseradish peroxidase-conjugated anti-rabbit secondary antibody (NA934V, GE Healthcare) or anti-mouse secondary antibody (NA931V, GE Healthcare), followed by visualization with a Prometheus ProSignal Peroxide Kit (Genesee Scientific). Image capture and densitometry quantitation of IFIT1 and IFIT3 proteins was performed with a BioRad GelDoc system and Image Lab software (V5.1). Final depletion was calculated as percentage of IFIT KD from NC control and normalized to tubulin.
iSLK cells were grown on glass coverslips plated at 120,000 cells per well in six-well dishes and treated with 1 μg/ml doxycycline to induce virus lytic replication. Cells were fixed at 48hr, 72hr and 96hr post induction, washed with 1x PBS, fixed and permeabilized with PBS containing 4% paraformaldehyde and 0.2% Triton X-100 for 15–20 minutes at room temperature, and then washed two times with 1x PBS followed by incubation with blocking buffer (20% goat serum in PBS) for 30 mins at room temperature. Cells were incubated with anti-IFIT3 polyclonal antibody (Thermo Science) or anti-IFIT1 polyclonal antibody (Thermo Science) at a dilution of 1:500, at 37°C for 1 hour. The slides were washed three times with 1x PBS and incubated with Alexa Fluor 594 goat anti-rabbit IgG (A11072, Invitrogen) for 1 hour at 37°C (in the dark). Nuclear staining was performed with 4',6- diamidino-2-phenylindole (DAPI) (Invitrogen). Images were collected and analyzed with a Zeiss Imager M2 microscope system. To determine cells staining positive for IFIT3, more than 6 fields in which each field included more than 2000 cells (totally at least 12000 cells) were counted with a 20x objective. Statistical testing for comparison of proportions and p values was performed using MedCalc software, which uses the "N-1" Chi-squared test (MedCalc software, Ostend Belgium)
To induce KSHV lytic gene expression or virus replication, iSLK cells were treated with 1 μg/ml doxycycline. Where KD of IFITs was performed, siRNA transfections were performed 6 hours prior to induction. Cells were harvested at 48 hr post induction for RNA preparation. At 48 hrs post-induction, cell viability was greater than 94% by vital dye staining. For virus production, supernatants of the cells were harvested 5 days post induction, cleared by centrifugation twice, and filtered through a 0.80 μm pore size cellulose acetate filter. Serial dilutions of supernatants were used to infect 293T cells. 48 hours after infection, flow cytometry was performed to measure the number of GFP positive cells, each representing a cell infected by a GFP expressing KSHV virion [29]. Each infection was done in triplicate and each infected cell sample was assayed by flow cytometry in technical triplicates. Based on the dilution factor, infectious virus titers in the iSLK cell supernatant were calculated. Pellets of the cells from which supernatant was harvested were processed for DNA isolation using Qiagen DNeasy Blood and Tissue kit. 50 ng of each DNA were used for qPCR using primers specific for ORF59 (see above) and SYBR green PCR MasterMix (ABI).
RtcB expression plasmid (kind gift from Dr. Alexei V. Korennykh), contains the RtcB gene from E. coli cloned into pGEX-6P and expressed with an N-terminal GST-tag. Protein isolation and purification were performed as previously described [40] with some modifications. Briefly, E. coli BL21-CodonPlus (DE3) carrying the RtcB expression plasmid was grown at 37°C in LB medium containing 50 μg/ml ampicillin until the A600 reached 0.6 to 0.7. The culture was chilled down to 20°C and protein expression was induced by IPTG (0.25mM final concentration). Incubation was continued at 20°C for 16 hr with constant shaking. Cells were harvested by centrifugation and stored at −80°C. All subsequent procedures were performed at 4°C. The cell pellet was suspended in 50 ml buffer A (20 mM HEPES pH 7.5, 300 mM NaCl, 10% glycerol, 2 mM DTT, 0.1 mM EDTA, 1% Triton X-100) with proteinase inhibitor (0.3μg/ml aprotinin, 0.5μg/ml leupeptin and 0.7μg/ml pepstatin A), 10μg/ml DNase and 10μg/ml lysozyme. After mixing for 30 min, cells were sonicated and insoluble material was removed by centrifugation. The clarified lysate was added to a column with 10ml glutathione resin washed with buffer A. The column was rotated for 1h at 4°C to bind GST-RtcB to the resin. Washing was performed with buffer A and the resin was resuspended in low salt buffer A (100mM NaCl without Triton X-100). PreScission Protease (GE Healthcare) was added and the column was gently rotated overnight at 4°C. The cleaved RtcB was then eluted from the column in 10ml buffer A. Ion exchange (MonoQ) purification (S7 Fig) followed by S200 gel filtration (S7 Fig) was performed with unsalted buffer (20 mM HEPES pH 7.5), high salt buffer (20 mM HEPES pH 7.5, 1M NaCl) and buffer B (buffer A without 1% Triton X-100). Purified RtcB was eluted and diluted to 100μM in buffer B with 0.5% Triton X-100, aliquoted and stored at -80°C.
We measured several specific cleavage sites generated by RNase L to analyze its activity as described [40]. Briefly, RtcB ligase and RtcB ligation adaptor (5’-OH-GAUCGUCGGACTGTAGAACTCTGAAC-3’) were added to cellular RNA (500ng) to ligate 2′,3′-cyclic phosphate containing RNAs to the adapter. The underlined bases in the ligation primer were RNA and the remainder were DNA. EDTA-quenched ligation reaction (1μL) was used as a template for reverse transcription with Multiscribe reverse transcriptase (ThermoFisher) and RT primer (5′-TCCCTATCAGTGATAGAGAGTTCA GAGTTCTACAGTCCG-3′). SYBR-green based qPCR was conducted using a universal rev qPCR primer (5’-TCCCTATCAGTGATAGAGAG-3’) and cleavage site-specific forward primers designed for each RNA target: tRNA His-36 (5’-GTTAGTACTCTGCGTTGTGGA-3’), RNY4-27 (5’-GATGGTAGTGGGTTATCAGAT-3’) and RNY5-30 (5’-GTGTTGTGGGTTATTGTTAGA-3’). U6, which has a naturally occurring 2′,3′-cyclic phosphate and an RNase L independent cleavage site, was used as endogenous control. Primers for the U6 site were U6 Q1F: 5’-GCTTCGGCAGCACATATACTA-3’ and U6 Q1R: 5’-CGAATTTGCGTGTCATCCTTG-3’, qPCR was carried out for 60 cycles using 62°C annealing/extension for 1 min.
iSLK/Bac16 cells were treated with 1 μg/ml doxycycline to induce KSHV replication. Cells were harvested at 48 hr post induction for immunoprecipitation. Cells were lysed by freeze-thawing in hypotonic buffer containing 20mM HEPES, pH7.3, 2mM MgCl2, 10% glycerol, 0.2mM EGTA, 1mM DTT, 1x protease inhibitor cocktail (Sigma) and RNasin (Promega). All subsequent steps were performed at 4°C. Lysates were clarified by centrifugation and precleared with normal rabbit IgG (Bethyl) and protein A-agarose beads, followed by immunoprecipitation with anti-IFIT1 plus anti-IFIT3 Rabbit polyclonal antibody (ThermoScience), or normal rabbit IgG overnight, followed by incubation with protein A-agarose beads. The beads were washed four times in IP washing solution (500 mM NaCl, 0.25% NP-40, 0.25% Triton X-100, 0.5% CHAPS). Coimmunoprecipitated RNA was isolated from the immunoprecipitates using Qiazol with additional glycogen and Qiagen miRNeasy columns according to the manufacturer’s protocols, with an on-column DNase treatment (Qiagen). Immunoprecipitated viral and cellular gene mRNA was quantitated by Real-time Quantitative PCR (qPCR) with SYBR green PCR Master Mix (Applied Biosystems) according to the manufacturer’s protocol. Each sample was analyzed in triplicate with gene specific primers. The gene-specific primers were as follows:
GAPDH Q1F, 5’-agggtcatcatctctgccccctc-3’;
GAPDH Q1R, 5’-tgtggtcatgagtccttccacgat-3’
MT-ATP6 Q1F: 5’-gtatgagcgggcgcagtgatt-3’
MT-ATP6 Q1R: 5’-atggggataaggggtgtaggtgtg-3’
U1 Q1F: 5’-ccatgatcacgaaggtggttt-3’
U1 Q1R: 5’-atgcagtcgagtttcccacat-3’
U2 Q1F: 5’-ctcggccttttggctaagat-3’
U2 Q1R: 5’-cgttcctggaggtactgcaa-3’
U5 Q1F: 5’-ctctggtttctcttcagatcgc-3’
U5 Q1R: 5’-ccaaggcaaggctcaaaaaat-3’
U6 Q1F: 5’-gcttcggcagcacatatactaaaattgga-3’
U6 Q1R: 5’-ataggaacgcttcacgaatttgcg-3’
U15 Q1F: 5’-ggtcacgtcctgctcttggtc-3’
U15 Q1R: 5’-atgcctctaaatcgatcaataaat-3’
U16 Q1F: 5’-atgatgtcgtaatttgcgtctt-3’
U16 Q1R: 5’-ctcagtaagaattttcgtcaacc-3’
KSHV gene specific primers are listed above.
|
10.1371/journal.pcbi.1005583 | Linking structure and activity in nonlinear spiking networks | Recent experimental advances are producing an avalanche of data on both neural connectivity and neural activity. To take full advantage of these two emerging datasets we need a framework that links them, revealing how collective neural activity arises from the structure of neural connectivity and intrinsic neural dynamics. This problem of structure-driven activity has drawn major interest in computational neuroscience. Existing methods for relating activity and architecture in spiking networks rely on linearizing activity around a central operating point and thus fail to capture the nonlinear responses of individual neurons that are the hallmark of neural information processing. Here, we overcome this limitation and present a new relationship between connectivity and activity in networks of nonlinear spiking neurons by developing a diagrammatic fluctuation expansion based on statistical field theory. We explicitly show how recurrent network structure produces pairwise and higher-order correlated activity, and how nonlinearities impact the networks’ spiking activity. Our findings open new avenues to investigating how single-neuron nonlinearities—including those of different cell types—combine with connectivity to shape population activity and function.
| Neuronal networks, like many biological systems, exhibit variable activity. This activity is shaped by both the underlying biology of the component neurons and the structure of their interactions. How can we combine knowledge of these two things—that is, models of individual neurons and of their interactions—to predict the statistics of single- and multi-neuron activity? Current approaches rely on linearizing neural activity around a stationary state. In the face of neural nonlinearities, however, these linear methods can fail to predict spiking statistics and even fail to correctly predict whether activity is stable or pathological. Here, we show how to calculate any spike train cumulant in a broad class of models, while systematically accounting for nonlinear effects. We then study a fundamental effect of nonlinear input-rate transfer–coupling between different orders of spiking statistic–and how this depends on single-neuron and network properties.
| A fundamental goal in computational neuroscience is to understand how network connectivity and intrinsic neuronal dynamics relate to collective neural activity, and in turn drive neural computation. Experimental advances are vastly expanding both the scale and the resolution with which we can measure both neural connectivity and neural activity. Simultaneously, a wealth of new data suggests a possible partitioning of neurons into cell types with both distinct dynamical properties and distinct patterns of connectivity. What is needed is a way to link these three types of data; how is it that patterns of connectivity are translated into patterns of activity through neuronal dynamics?
Any model of neural activity should also capture the often-strong variability in spike trains across time or experimental trials. This variablity in spiking is often coordinated (correlated) across cells, which has a variety of implications. First, correlations play an essential role in plasticity of network structure [1–4]. Theories that describe spiking correlations allow for a self-consistent description of the coevolution of recurrent network structure and activity [5, 6]. Second, correlations between synaptic inputs control their effect on postsynaptic neurons: inputs that arrive simultaneously can produce stronger responses than those arriving separately. This has been referred to as “synergy” or “synchronous gain” in early work [7], and the magnitude of this synergy has been measured in the LGN by Usrey, Reppas & Reid [8] and cortex by Bruno & Sakmann [9] (but see [10]). Indeed, the level of correlation in an upstream population has been shown to act as a gain knob for firing rates downstream [11]. Finally, correlated fluctuations in activity can impact the fidelity with which populations can encode information [12, 13]. Importantly, the coding impact depends on a subtle interplay of how signals impact firing rates in a neural population and of how noise correlations occur across the population [14–18]. An accurate description of how network connectivity determines the individual and joint activity of neural populations is thus important for the understanding of neural activity, plasticity and coding.
Many studies of collective activity in spiking systems can be traced to the early work of Hawkes on self- or mutually-exciting point processes [19, 20]. The Hawkes model is also closely related to the linear response theory that can be used to describe correlations in integrate-and-fire networks [21, 22]. Here, each neuron and synapse is linearized around a central operating point and modes of collective activity are computed around that point [23–25]. Including a nonlinear transfer of inputs to rates in the Hawkes model gives a generalized linear model, which has been applied with considerable success to multi-neuron spike train data [26].
Analyses based on computing modes of collecting activity based on linearized dynamics have led to significant insights, but they also impose a limitation. While shifts of the operating point can modulate the linearized dynamics of biophysical models [27], this approach cannot capture the impact of nonlinear neural dynamics at the operating point.
Here, we present a systematic method for computing correlations of any order for nonlinear networks of excitatory and inhibitory neurons. Nonlinear input-rate transfer couples higher-order spike train statistics to lower-order ones in a manner that depends on the order of the nonlinearity. In its simplest form, this coupling shows how pairwise–correlated inputs modulate output firing rates. This generalizes the effects of pairwise correlations on neural gain in single-neurons [7, 8, 11] and feedforward circuits [28–31] to networks with high levels of recurrence and feedback.
We begin with simple models and progress to nonlinearly interacting networks of spiking neurons. Our method is diagrammatic, in the sense that the interplay of network connectivity and neural dynamics in determining network statistics is expressed and understood via a systematic series of graphical terms. Such graphs are commonly referred to as “Feynman diagrams” after Richard Feynman, who invented them. We use this diagrammatic expansion to make and explain three main scientific points. First, we show how neural dynamics lead to spike correlations modulating firing rates in a recurrent, nonlinear network. Second, we illustrate an additional role of the prominent “heavy-tailed” feature of neural connectivity where some neurons have many more connections than others, and some connections are much stronger than others. We show how this feature interacts with nonlinearities to control network activity. And third, we show how different single-neuron nonlinearities affect the dependence of firing rates on correlations.
We will show that any coupled point process model, even one with nonlinearities or negative interactions, has an associated expansion for all spike train cumulants organized by the strength of coupling of higher statistical moments with lower ones (e.g., the influence of the two-point correlation on the mean). The full model we aim to describe is one where each neuron generates a spike train which is conditionally renewal with intensity:
r i ( t ) = ϕ i (∑ j (g i j * d N j d t) ( t ) + λ i ( t )) . (1)
Here gij(t) is a matrix of interaction filters, λi(t) is the baseline point process drive to neuron i and * denotes convolution: ( g * f ) ( t ) = ∫ t 0 ∞ d t ′ g ( t - t ′ ) f ( t ′ ) (with the integral starting at the initial time for the realization). ϕi is the transfer function of neuron i. Neuron j’s spike train is d N j d t = ∑ k δ ( t - t j k ), a sum over Dirac deltas at each of the k spike times. We will take the spike trains to be conditionally Poisson given the input, so that in each time window (t, t + dt), the probability of neuron i generating m spikes is (ri(t)dt)m/m! exp (−ri(t)dt). This corresponds to a point process generalized linear model (GLM), or nonlinear multivariate Hawkes process [32]. In contrast to biophysical or integrate-and-fire models in which spike trains are generated deterministically given the membrane potential (which might, however, depend on noisy input), this model with escape noise generates spike trains stochastically with a rate that depends on the “free” membrane potential (i.e., with action potentials removed) [33].
Current methods for the analysis of single-neuron and joint spiking statistics rely on linear response techniques: using a self-consistent mean field theory to compute mean firing rates and then linearizing the spiking around those rates to determine the stability and correlations. We begin with a simple example highlighting the need to account for nonlinear effects. We take an excitatory-inhibitory network of NE = 200 excitatory (E) neurons and NI = 40 inhibitory (I) neurons, all with threshold-quadratic transfer functions ϕ i ( x ) ≡ ϕ ( x ) = α ⌊ x ⌋ + 2. In this example we took network connectivity to be totally random (Erdös-Rényi), with connection probabilities pEE = 0.2 and pEI = pIE = pII = 0.5. For simplicity, we took the magnitude of all connections of a given type (E − E, etc.) was taken to be the same. Furthermore, the time course of all network interactions is governed by the same filter g ( t ) = t τ 2 exp ( - t / τ ) (with τ = 10 ms), so that gij(t) = Wijg(t). W is a matrix of synaptic weights with units of mV, so that the input to ϕ can be interpreted as the free membrane potential. We set the strength of interactions such that the net inhibitory input weight on to a neuron was, on average, twice that of the net excitatory input weight so that for sufficiently strong interactions, the network was in an inhibitory-stabilized regime [34].
We examined the magnitude and stability of firing rates as we increased the strength of synaptic coupling. We used mean field theory to predict the firing rates, and predicted their linear stability by the standard measure of the spectral radius of the stability matrix Ψ i j = ϕ i ( 1 ) g i j. (ϕ i ( 1 ) denotes the first derivative of neuron i’s transfer function with respect to its input.) As the strength of interactions increases, the mean field prediction for the firing rates loses accuracy (Fig 1A). This occurs well before the mean field theory crosses the stability boundary |Ψ| = 1 (Fig 1B). Examining simulations as the weights are increased reveals an even more fundamental failure of the theory: before the synaptic weights are strong enough for the mean field theory to be unstable, the simulations reveal divergent firing rates (Fig 1C; the raster stops when the instantaneous rates diverge).
Rather than restricting theoretical analysis to regimes of extremely low coupling strength or linear models, we here develop a framework for activity cumulants that can apply to models with nonlinear input-spike transfer. This will allow us to properly predict both the spiking cumulants and the location of the rate instability in the nonlinear network above. Thus, we develop a framework for activity cumulants that can apply to models with nonlinear input-spike transfer for strongly coupled networks. The mean field and linear response predictions for spiking cumulants correspond to the lowest order terms of this expansion and are good whenever that lowest-order truncation is valid.
We will build this framework systematically: We begin with statistics of the drive λi(t), then consider a filtered point process, g * λ(t). In these simple models we will introduce the method and terminology that we will use to understand the more complicated models. We continue by considering the linearly self-exciting Hawkes process, taking a single neuron so g = g and ϕ(x) = x, before proceeding to arbitrary nonlinearities ϕ. Finally, we introduce an arbitrary network structure g. This model is intimately related to the popular GLMs for spiking activity, where the nonlinearity ϕ is commonly taken to be exponential, refractory dynamics can be embedded in the diagonal elements of g, and λ corresponds to the filtered stimulus [26]. A use of GLMs it to fit them to recorded spike trains and then ask about the structure of the inferred functional connectivity g. In contrast, we interpret g as reflecting the structural connectivity and synaptic and membrane dynamics of a specified model network and aim to compute statistics of its neurons’ spike trains. The derivation given here will be heuristic. A more rigorous derivation of the expansion is given in Methods: Path integral representation.
An inhomogeneous Poisson process generates counts within a window dt independently with a Poisson distribution at rate λ(t). A spike train produced by this process is
d N d t ( t ) = ∑ k δ ( t - t k ) (2)
where tk is the kth spike time, and N(t) is the spike count. The mean and autocovariance for this process are given by the familiar formulas:
⟨d N d t ( t )⟩ = λ ( t ) (3)
⟨d N d t ( t ) d N d t ( t ′ )⟩ c = λ ( t ) δ ( t - t ′ ) (4)
where angular brackets denote the expectation across realizations and the subscript c denotes a cumulant, not the moment (i.e., we have subtracted all terms which factor into products of lower moments) [35]. The delta function arises because the process is independent at each time step, so that there is no correlation between events from one time t and any other time t′. In fact, because the events are generated independently at each time point, all of the cumulants of this process can be written as
⟨∏ i d N d t ( t i )⟩ c = ∫ d t λ ( t ) ∏ i δ ( t i - t ) (5)
where integrating out one of the delta functions puts the second cumulant in the above form. We can interpret this equation as describing a source of events appearing at rate λ(t) at time t that propagate to times ti. In this case, because the events are generated independently at each t, the events only propagate to the same time t. For a general point process, cumulants measured at the collection of times {ti} could be affected by events occurring at any past time, so that we would have to account for how events propagate across time.
The expansion for cumulants we will develop has a natural construction in terms of graphs (Feynman diagrams), wherein components of the graph represent factors in each term. A set of defined rules dictate how terms in the expansion are formed from those graphs. While this graphical representation is not necessary to understand the inhomogeneous Poisson process, we describe it in detail in this simple case to develop intuition and introduce terminology. We use cumulants in this construction because they provide the fundamental building blocks of moments; any n-th order moment can be constructed from up to n-th order cumulants). This also simplifies the developed expansion.
To begin, the nth cumulant has n arguments {ti, i = 1, …, n}, one for each factor of d N d t in Eq 5. We represent each of these by an open white vertex in the graph labeled with the time point, ti, and we represent the source term λ(t) with a gray vertex. The white vertices are called “external” vertices whereas the gray vertex is called “internal.”
The internal gray vertex represents the intensity of the underlying stochastic process generating events with rate λ(t). The white external vertices represent the spike trains whose statistics we measure at times {ti}. For each delta function, δ(t − ti), in Eq 5, we place an edge drawn as a dotted line from the gray vertex to the white vertex, i.e., from the event-generating process to the spike train’s measurement times (Fig 2). More generally, events generated by the source propagate through the graph to affect the external vertices.
In order to construct the term associated with each diagram, we multiply the factors corresponding to edges (delta functions linking t and ti) or the internal vertex (λ(t)), and then integrate over the time t associated with the internal vertex. This links the generation of events by λ(t) to their joint measurement at times {ti} through the propagator (here δ(t − ti)). For the diagrams shown in Fig 2, these rules reproduce the cumulant terms in Eq 5. Note that these graphs are directed, since we only consider causal systems where measured cumulants are only influenced by past events.
In general, a given moment will be the sum of terms associated with many different graphs. For example the second moment is given by
⟨d N d t ( t 1 ) d N d t ( t 2 )⟩ = ⟨d N d t ( t 1 ) d N d t ( t 2 )⟩ c + ⟨d N d t ( t 1 )⟩ ⟨d N d t ( t 2 )⟩ . (6)
Each term on the right-hand side will have a corresponding graph. Moreover, the graph for the second term will include two disconnected components, one for each factor of the mean rate, which appears as in Fig 2. The graphs for the cumulant will always be described by connected graphs.
We proceed to a simple model of synaptic input: a presynaptic Poisson spike train, with count N(t) and intensity λ(t), drives postsynaptic potentials with shape g(t):
ν ( t ) = ϵ (g * d N d t) ( t ) (7)
where * denotes convolution: ( g * f ) ( t ) = ∫ t 0 ∞ d t ′ g ( t - t ′ ) f ( t ′ ) (with the integral starting at the initial time for the realization). We assume g is normalized, ∫ - ∞ ∞ g ( t ) d t = 1, so that ϵ gives the magnitude of the filtering.
The cumulants of the postsynaptic potential ν(t) can be calculated directly. In general, they are given by:
⟨∏ i ν ( t i )⟩ c = ∫ d t λ ( t ) ∏ i ϵ (g * δ) ( t i - t ) (8)
where the input spikes are generated at times t, arrive at times given by the delta functions and influence the cumulant measured at {ti} through g. Eq 8 is the same as that for the inhomogeneous Poisson process but with factors of g*. This provides a simple interpretation of Eq 8: cumulants of the filtered Poisson process are given by taking the cumulants of the underlying Poisson process and examining how they can be filtered through the system at hand.
Similarly to the case for the Poisson process, we can represent the cumulants graphically. We again represent each measurement time on the left-hand side of Eq 8 by an external vertex (Fig 3a). The convolution of δ and g in Eq 8 corresponds to an internal time point which we integrate over (denoted by primes). We also represent these internal time points with white vertices that carry a factor of ϵ, the magnitude of the filter. We represent the source term λ(t) with a gray vertex. All vertices that are not “external” are again called internal. Every internal vertex also carries its own time index, t′.
The internal gray vertex again represents the intensity of the underlying stochastic process, λ(t). The white external vertices represent the processes whose statistics we measure at times {ti}. For each delta function, δ(t′ − t), in Eq 8, we place an edge drawn as a dotted line from the gray vertex to the white vertex, i.e., from the event-generating process to the arrival time of the event t′. In this example an event “arrives” as soon as it is generated. A wavy edge corresponds to the filter, g, and represents the effect of a spike arriving at time t′ on the output process measured at time ti (Fig 3b). Events generated by the source thus propagate through the graph to affect the observed, external vertices.
In order to construct the expression associated with each diagram, we again multiply the factors corresponding to each edge in the diagram (e.g., δ(t′ − t) or g(ti − t′)) or internal vertex (ϵ or λ(t)) and then integrate over the times associated with the internal vertices. Note that integration over the internal times t′, t′′, etc. results in the convolutions ϵ(g * δ)(ti − t). Integration over the time t associated with the source term corresponds to the outermost integral in Eq 8 This links the generation of events by λ(t) to their joint measurement at times {ti} through their arrival times (via δ(t − t′)) and temporal filtering (g(ti − t′)). For the diagrams shown in Fig 3, these rules reproduce the cumulant terms in Eq 8. Note that the graphs are directed, as for the expansion we describe the “propagator” term will be causal.
We can simplify the cumulants of this process (and the corresponding diagrammatic representations) by considering the propagator of ν(t) (also known as the linear response or impulse response). The propagator measures the change in 〈ν(t)〉 in response to the addition of one input spike in N(t). We can compute it by taking a functional derivative with respect to the input intensity λ(t):
Δ ( t , t ′ ) = δ δ λ ( t ′ ) ⟨ν ( t )⟩ = δ δ λ ( t ′ ) (ϵ ∫ t 0 ∞ d t ′ ′ g ( t - t ′ ′ ) ⟨d N ( t ′ ′ ) d t⟩) = ϵ ∫ t 0 ∞ d t ′ ′ g ( t - t ′ ′ ) δ λ ( t ′ ′ ) δ λ ( t ′ ) = ϵ ( g * δ ) ( t - t ′ ) (9)
Since the dynamics are linear, this is also equivalent to the change of the expected rate with the addition of one spike to the input, i.e., taking λ(t) ← λ(t) + δ(t′ − t) and 〈ν(t)〉c ← 〈ν(t)〉c + Δ * δ(t) (or equivalently the Green’s function of the expected rate). This allows us to rewrite the cumulants in terms of the input rate and the propagator:
⟨∏ i ν ( t i )⟩ c = ∫ d t λ ( t ) ∏ i Δ ( t i , t ) (10)
which can be represented graphically by introducing a solid, directed edge for Δ(t, t′) (Fig 3c). The propagator will be a central feature of the expansion for cumulants in more complicated models involving connections among neurons.
In order to generalize the graphical representation of Poisson cumulants, we begin with a linearly self-exciting process as considered by Hawkes [19]. Let the rate be a linear function of the instantaneous event rate (that is to say the firing rate conditioned on a particular realization of the event history)
r ( t ) = ϵ (g * d N d t) ( t ) + λ ( t ) . (11)
We assume that g(τ) and λ(t) are such that r(t) > 0, and ∫ - ∞ ∞ d τ g ( τ ) = 1. If ϵ < 1, then an event will generate less than one event on average, and the rate will not diverge. The history dependence of the firing rate will now enter into our calculations. We can compute the expected rate using the self-consistency equation:
r ¯ ( t ) ≡ ⟨d N d t ( t )⟩ = ϵ (g * ⟨d N d t⟩) ( t ) + λ ( t ) = ϵ ( g * r ¯ ) ( t ) + λ ( t ) (12)
We provide an alternate derivation of this result that will prove useful below: We construct a perturbative expansion of the mean firing rate and show how this expansion can be re-summed to yield the full rate of the self-exciting process. This procedure can also be applied to obtain cumulants of arbitrary order for this process.
We will begin with a recursive formulation of the self-exciting process. In contrast to the filtered Poisson process of the previous section, here the process with count N generates events, which then influence its own rate, dN/dt. Each event can thus generate others in turn. In the case of a linear filter, g, the following approach is equivalent to the Poisson cluster expansion [36–38] and similar to the construction of previous linear response theories for spike train covariances [22]. Define the nth order self-exciting process, Nn(t), to be the inhomogeneous Poisson process given by:
d N n ( t ) d t = d N 0 ( t ) d t + d M n - 1 ( t ) d t , (13)
where N0(t) and Mn(t) are inhomogeneous Poisson processes with rates λ(t) and νn(t), respectively, where
ν n ( t ) = ϵ (g * d N n d t) ( t ) (14)
so ν 0 ( t ) = ϵ ( g * d N 0 d t ) ( t ). Mn(t) is a process with intensity that depends on a stochastic realization of Nn(t), making M0(t) a “doubly stochastic” process. We can generate these processes recursively: To generate Nn(t), we use a realization of Nn−1(t) to compute the rate νn−1 and generate a realization of Mn−1(t). These are added to events generated from the original inhomogeneous Poisson process with rate λ(t) to produce Nn(t). We can use this recursive procedure to develop an expansion for the cumulants of the process at a given order in ϵ (thus a given order in the self-convolution of g).
Let us compute the value of 〈 d N d t ( t ) 〉 in powers of ϵ using our recursive approach. The zeroth order solution, 〈 d N 0 d t ( t ) 〉, is the rate of the inhomogeneous Poisson process λ(t). At order n, we compute 〈 d N n d t ( t ) 〉 using the (n − 1)st order solution in the right-hand side of Eq 13. At first order, using the Poisson solution for 〈 d N 0 ( t ) d t 〉 we get
⟨d N 1 d t ( t )⟩ = ⟨d N 0 d t ( t )⟩ + ⟨d M 0 d t ( t )⟩ (15)
= λ ( t ) + ϵ (g * ⟨d N 0 d t⟩) ( t ) (16)
= λ ( t ) + ϵ ( g * λ ) ( t ) (17)
= λ ( t ) + ϵ ∫ t 0 ∞ d t ′ g ( t - t ′ ) λ ( t ′ ) (18)
At second order we similarly arrive at
⟨d N 2 d t ( t )⟩ = ⟨d N 0 d t ( t )⟩ + ⟨d M 1 d t ( t )⟩ (19)
= λ ( t ) + ϵ (g * ⟨d N 1 d t⟩) ( t ) (20)
= λ ( t ) + ϵ ∫ t 0 ∞ d t ′ g ( t - t ′ ) λ ( t ′ ) + ϵ 2 ∫ - ∞ ∞ d t ′ g ( t - t ′ ) ∫ t 0 ∞ d t ′ ′ g ( t ′ - t ′ ′ ) λ ( t ′ ′ ) (21)
At higher orders we would obtain further terms with additional convolutions with g.
It will be useful to write these expansions in another way, which will allow their form to generalize to nonlinear processes: we will construct the cumulants from the baseline rate and the propagator. We can always replace
λ ( t ) = ∫ t 0 ∞ d t ′ δ ( t - t ′ ) λ ( t ′ ) (22)
resulting in
⟨d N 1 d t ( t )⟩ = ∫ t 0 ∞ d t ′ δ ( t - t ′ ) λ ( t ′ ) + ϵ ∫ t 0 ∞ d t ′ g ( t - t ′ ) ∫ t 0 ∞ d t ′ ′ δ ( t ′ - t ′ ′ ) λ ( t ′ ′ ) (23)
Fig 4a shows the graphical representation of this expansion. As before, the order of the moment is given by the number of external vertices and each external vertex carries a measurement time ti. We have three types of internal vertices: two open white vertices that carry factors of ϵ (one type has one wavy incoming and one wavy outgoing line; the other has one incoming dotted line and one wavy outgoing line) and one gray vertex (that has one outgoing dotted line). As before, each gray internal vertex corresponds to the source term, and thus represents the factor λ(t). The white internal vertices and their edges represent how the events generated by the source are propagated through the filter g. Each white vertex corresponds to a possible past event time, t′. To construct the cumulant corresponding to a diagram, we integrate over all these possible internal times, weighting each by their influence on the current spiking rate. These weights are given by the filters, g, represented by the wavy edges. The graphical representation of 〈 d N 1 d t ( t ) 〉 (using the delta function as in Eq 23) is shown in Fig 4a.
We can compute the firing rate of the self-exciting process r ¯ ( t ) as the limit of the nth order self-exciting processes, continuing the process outlined for Eq 13:
r ¯ ( t ) = ∑ n = 0 ∞ ϵ n ( g ( n ) * δ * λ ) ( t ) , (24)
where g(n) is the n-fold convolution of g with itself and g(0)(t) = δ(t). Indeed, we can see that this expression for r ¯ ( t ) yields the same recursive self-consistency condition as above:
r ¯ ( t ) = ϵ 0 ( g ( 0 ) * δ * λ ) ( t ) + ∑ n = 1 ∞ ϵ n ( g ( n ) * δ * λ ) ( t ) = λ ( t ) + ϵ (g * ∑ n = 1 ∞ ϵ n - 1 g ( n - 1 ) * δ * λ) ( t ) = λ ( t ) + ϵ ( g * r ¯ ) ( t ) . (25)
We can also represent this recursive relation graphically as in Fig 4b, using a black vertex to denote the mean-field rate r ¯ ( t ). The infinite sum defined by Eq 13 has a specific graphical representation: the leftmost vertex and wavy line in the right-hand side of Fig 4b (top) can be detached and factored, with the remaining series of diagrams corresponding exactly to those of the mean. This series of subgraphs on the right-hand side sums to 〈 d N d t ( t ) 〉, leading to the recursion relation in Eq 25 (Fig 4b). This graphical representation is equivalent to the recursion relation.
The propagator, Δ(t, t′), measures the fluctuation in the expected rate (around the mean-field value) in response to the addition of one spike at time t′ to the drive λ(t). Setting λ(t) ← λ(t) + δ(t − t′) and r ¯ ( t ) ← r ¯ ( t ) + ( Δ * δ ) ( t , t ′ ) in Eq 12 gives:
r ¯ ( t ) + ( Δ * δ ) ( t , t ′ ) = ϵ (( g * r ¯ ) ( t ) + ( g * Δ * δ ) ( t , t ′ )) + λ ( t ) + δ ( t - t ′ ) Δ ( t , t ′ ) = ϵ ( g * Δ ) ( t , t ′ ) + δ ( t - t ′ ) (26)
where for convolutions involving Δ(t, t′), we use the notation (f * Δ)(t, t′) = ∫ dt′′ f(t − t′′)Δ(t′′, t′) and (Δ * f)(t, t′) = ∫ dt′′ Δ(t, t′′)f(t′′ − t′)
As with the expected rate r ¯ ( t ), we can examine the propagators of the n-th order self-exciting processes. For the first-order process N1(t),
Δ 1 ( t , t ′ ) = δ ( t - t ′ ) + ϵ ( g * δ ) ( t - t ′ ) (27)
The first term is the propagator of the inhomogeneous Poisson process. The second term of Δ1 is the propagator of the filtered Poisson process, Eq 9. This equation can be represented by the same type of graphs as for the expected rate (Fig 4c top), but stand for functions between two time points: the initial time t′ of the perturbation, and the later time t, at which we are computing the rate of the process. We don’t represent these initial and final points as vertices, because the propagator is a function that connects two vertices. However, we still integrate over the times corresponding to the internal vertices since the propagator accounts for the total effect of a perturbation of the source on the observed activity.
In general, the propagator for the nth-order self-exciting process can be computed by taking a functional derivate of the rate with respect to the input rate λ:
Δ n ( t , t ′ ) = δ δ λ ( t ′ ) (λ ( t ) + ϵ g * r ¯ n - 1 ( t )) = δ ( t - t ′ ) + ϵ δ δ λ ( t ′ ) (g * r ¯ n - 1) ( t ) = δ ( t - t ′ ) + ϵ (g * ∑ k = 0 n - 1 ϵ k g ( k ) * δ) ( t , t ′ ) (28)
This recursion relation can be expressed graphically just as for the mean rate (Fig 4c, top). Factoring out ϵg* corresponds to popping off an edge and vertex from the series (Fig 4c, middle). Taking the limit n → ∞ in Eq 28 yields the self-consistency condition for the full propagator Δ(t, t′) given by Eq 26, and indicated by the solid black line in Fig 4c (bottom).
These diagrammatic expansions may seem cumbersome for so simple a model. Even for the self-exciting Hawkes process, however, they allow the fast calculation of any order of spike train cumulant. Let us begin with the second cumulant of the instantaneous firing rate. Again we will construct an expansion in ϵ, i.e., powers of g. To zeroth order, this is the inhomogeneous Poisson solution. To first order in ϵ we have
⟨ dN1dt(t)dN1dt(t′) ⟩c=⟨ (dN0dt(t)+dM0dt(t))(dN0dt(t′)+dM0dt(t′)) ⟩c=⟨ dN0dt(t)dN0dt(t′) ⟩c+⟨ dN0dt(t)dM0dt(t′) ⟩c+⟨ dM0dt(t)dN0dt(t′) ⟩c+⟨ dM0dt(t)dM0dt(t′) ⟩c=∫t0∞dsδ(t−s)δ(t′−s)λ(s)+ϵ∫t0∞dsδ(t−s)(g*δ)(t′−s)λ(s)+ϵ∫t0∞dsδ(t′−s)(g*δ)(t−s)λ(s)+ϵ∫t0∞dsδ(t−s)δ(t′−s)(g*∫t0∞ds′δ(s−s′)λ(s′))(s). (29)
The first term on the second line is the second cumulant of the inhomogenous Poisson process. The other terms arise from the dependency of the processes M0(t) and N0(t). The expectation over M0(t) must be performed first, followed by that over N0(t), because the process M0(t) is conditionally dependent on the realization of N0(t), having intensity ϵ ( g * d N 0 d t ) ( t ) (Eq 13). This decomposition relies on the linearity of the expectation operator.
We can construct diagrams for each of these terms using the same rules as before, with the addition of two new internal vertices (Fig 5a). These new vertices are distinguished by their edges. The first has two outgoing dotted lines representing the zeroth-order propagator δ(t − t′), as in the second cumulant of the inhomogeneous Poisson process. It represents events that are generated by the drive λ(t) and propagate jointly to the two measurement time points. The second new vertex has the same two outgoing lines and one incoming wavy line for the filter g(t, t′)–it represents the fourth term on the right-hand side of Eq 29. This vertex carries a factor of ϵ and represents the filtering of past events that then propagate to the two measurement time points.
Continuing the computation of the second cumulant to any order in ϵ will result in higher order terms of the linear response and expected rate being added to the corresponding legs of the graph. At a given order n, one leg of each diagram will be associated with a particular term in the expansion, to order n, of the expected rate or the linear response. The second cumulant of dN2/dt would thus add diagrams with two filter edges to the diagrams of Fig 5a, either both on the same leg of the graph or distributed among the graph’s three legs.
As with the filtered Poisson process, we can simplify this sum of four diagrams for the second cumulant of the first-order self-exciting process. Examining subgraphs of each term on the right-hand side of Fig 5A reveals a connection to the linear response and mean rate of the first-order self-exciting processes. On each leg emanating from the internal branching vertex, the four terms sum to the product of two copies of the linear responses of the first-order self-exciting process N1(t) (compare subgraphs on the top branch of the diagrams in Fig 5a with Fig 4a). Similarly, the sum of the legs coming into the branching vertex is the firing rate of N1(t) (compare to Fig 4b). So, we will group the terms on the legs of the graph into contributions to the linear response and the mean (Fig 5b middle).
When we add the diagrams of up to order n together, we can separately re-sum each of these expansions because of the distributivity of the expectation. So, we can replace the entire series to all orders in ϵ with simpler diagrams using the full representations for the linear response and expected rate (Fig 5b). This can be proved inductively, or by rigorously deriving the Feynman rules from the cumulant generating functional (Methods: Path integral representation). This yields the following result for the second cumulant, which corresponds to the final graph at the bottom far right of Fig 5b:
⟨d N d t ( t ) d N d t ( t ′ )⟩ c = ∫ t 0 ∞ d s Δ ( t - s ) Δ ( t ′ - s ) r ¯ ( s ) (30)
This is the complete analytic result for the second cumulant of the self-exciting process for fluctuations around the mean field solution r ¯ ( t ) [19]. It can be represented by the single term on the right-hand side of Eq 30 and the corresponding single diagram (Fig 5b, right). Compare this with the filtered Poisson process, which has a diagram of the same topology but with different constituent factors (Fig 3C, middle row). The Feynman diagrams capture the form of the re-summed perturbative expansions for the cumulants, while the definitions of the vertices and edges capture the model-specific rate, r ¯ ( t ), and propagator, Δ(t, t′).
One might think that the higher cumulants are generated as simply by replacing each leg of the filtered inhomogeneous Poisson process with the correct propagator, along with the rate r ¯ ( t ). This would mean that the general cumulant term would be given by:
⟨∏ i d N d t ( t i )⟩ c = ∫ t 0 ∞ d t r ¯ ( t ) ∏ i Δ ( t i , t ) . (31)
This is incorrect, as many important terms arising from the self-interaction would be lost.
The reason this naive generalization fails is that it neglects the higher-order responses to perturbations in the event rate. For example, the second cumulant responds to perturbations in the rate; this quadratic response impacts the third cumulant. We can see this in the third cumulant of the first-order self-exciting process:
⟨d N 1 d t ( t ) d N 1 d t ( t ′ ) d N 1 d t ( t ′ ′ )⟩ c = ⟨(d N 0 d t ( t ) + d M 0 d t ( t )) (d N 0 d t ( t ′ ) + d M 0 d t ( t ′ )) (d N 0 d t ( t ′ ′ ′ ) + d M 0 d t ( t ′ ′ ′ ))⟩ c = ∫ t 0 ∞ d s δ ( t - s ) δ ( t ′ - s ) δ ( t ′ ′ - s ) λ ( s ) + ϵ ∫ t 0 ∞ d s δ ( t - s ) δ ( t ′ - s ) ( g * δ ) ( t ′ ′ - s ) λ ( s ) + ( t ↔ t ′ ↔ t ′ ′ ) + ϵ ∫ t 0 ∞ d s δ ( t - s ) δ ( t ′ - s ) δ ( t ′ ′ - s ) (g * ∫ t 0 ∞ d s ′ δ ( s - s ′ ) λ ( s ′ )) ( s ) + ϵ ∫ t 0 ∞ d s ′ δ ( t ′ ′ - s ′ ) ∫ t 0 ∞ d s δ ( t - s ) δ ( t ′ - s ) ( g * δ ) ( s - s ′ ) λ ( s ′ ) + ( t ↔ t ′ ↔ t ′ ′ ) . (32)
The first term is the third cumulant of the inhomogeneous Poisson process. The second and third are generalizations of the terms found in the second cumulant (we have used (t ↔ t′ ↔ t′′) to denote “all other permutations of t, t′, t′′”). These terms are part of the naive expression in Eq 31. The last term is the novel one that arises due to the “quadratic response.” It appears when we compute
⟨d N 0 d t ( t ) d M 0 d t ( t ′ ) d M 0 d t ( t ′ ′ )⟩ c = ϵ ∫ t 0 ∞ d s δ ( t ′ - s ) δ ( t ′ ′ - s ) ⟨d N 0 d t ( t ) (g * d N 0 d t) ( s )⟩ c (33)
We have to take into account that the process d N 0 d t ( t ) is correlated with the rate of the process d M 0 d t ( t ) (since one is a linear function of the other!). This produces a “cascade” effect that results in the quadratic response. For the first-order process, only one step in the cascade is allowed. By introducing branching internal vertices similar to those in Fig 5, we can express these somewhat unwieldy terms with diagrams. These are shown in Fig 6. The cascade of one source spike producing three spikes in the first-order process is represented by the second diagram of Fig 6a and the cascade of one source spike producing two spikes, one of which then produces another two spikes in the first-order process, is represented by the last diagram of Fig 6a.
As before, continuing to higher orders in the recursively self-exciting process would add diagrams with additional filter edges along the legs of the graphs in Fig 6a, corresponding to additional steps in the potential cascades of induced spikes. For example, the third cumulant of the second-order process, ⟨ d N 2 d t ( t ) d N 2 d t ( t ′ ) d N 2 d t ( t ′ ′ ) ⟩ c, would add diagrams with two filter edges to those of Fig 6a, with those two filter edges appearing either sequentially on the same leg of the graph or distributed among the legs of the graph. We can then use the same ideas that allowed us to re-sum the graphs representing the second cumulant. As before, we identify the expansions of the mean-field rate, r ¯, and the linear response, Δ, along individual legs of the graph and use the multilinearity of cumulants to resum those expansions to give the diagrams at the bottom of Fig 6.
Considering the re-summed graphs, we have the following result for the third cumulant:
⟨d N 1 d t ( t ) d N 1 d t ( t ′ ) d N 1 d t ( t ′ ′ )⟩ c = ∫ t 0 ∞ d s Δ ( t , s ) Δ ( t ′ , s ) Δ ( t ′ ′ , s ) r ¯ ( s ) + ∫ t 0 ∞ d s ′ Δ ( t ′ ′ , s ′ ) ∫ t 0 ∞ d s Δ ( t , s ) Δ ( t ′ , s ) ( g * Δ ) ( s , s ′ ) r ¯ ( s ′ ) + ( t ↔ t ′ ↔ t ′ ′ ) (34)
The types of diagram developed for up to the third cumulant encompass all the features that occur in the diagrammatic computations of cumulants of linearly self-exciting processes. The general rules for diagrammatically computing cumulants of this process are given in Fig 7. They are derived in general in Methods: Path integral representation. The graphs generated with this algorithm correspond to the re-summed diagrams we computed above.
For the nth cumulant, ⟨ ∏ i d N d t ( t i ) ⟩ c, begin with n white external vertices labelled ti for each i. Construct all fully connected, directed graphs with the vertex and edge elements shown in Fig 7. For each such fully connected directed graph constructed with the component vertices and edges, the associated mathematical term is constructed by taking the product of each associated factor, then integrating over the time points of internal vertices. The nth cumulant is the sum of these terms. This produces cumulants of up to third order, as recently shown by Jovanović, Hertz & Rotter [38], as well as cumulants of any order. As we show next, this procedure can also be generalized to calculate cumulants in the presence of a nonlinearity, including both thresholds enforcing positive activity (as commonly disregarded in studies of the Hawkes process) and any nonlinear input-rate transfer function.
Now we include a nonlinearity in the firing rate, so that the process produces events dN/dt with a rate given by
r ( t ) = ϕ ((g * d N d t) ( t ) + λ ( t )) (35)
We begin by considering the mean-field solution r ¯, which, if it exists, is self-consistently given by r ¯ ( t ) = ϕ ( ( g * r ¯ ) ( t ) + λ ( t ) ). Thus, as always the mean-field solution is given by neglecting second and higher-order cumulants of the spiking process. Next, we consider the propagator, which as above is the linear response of the rate around the mean field, given by expanding Eq 35 around the mean-field solution r ¯ ( t ) and examining the gain with respect to a perturbation of the rate. This propagator obeys:
Δ ( t , t ′ ) = ϕ ( 1 ) · (( g * Δ ) ( t , t ′ )) + δ ( t - t ′ ) (36)
where ϕ(1) is the first derivative of ϕ with respect to the input, evaluated at g * r ¯ + λ. We will first develop a recursive formulation of the mean-field rate and propagator, which will be required for calculating cumulants of the full process. For an arbitrary nonlinearity ϕ, we would begin by Taylor expanding around 0. For simplicity, we here consider a quadratic ϕ so that:
r ( t ) = λ ( t ) + ϵ 1 ((g * d N d t) ( t ) + λ ( t )) + ϵ 2 ((g * d N d t) ( t ) + λ ( t )) 2 . (37)
We now develop the point process dN/dt recursively at each order of the nonlinearity:
d N m , n d t ( t ) = d N 0 , 0 d t ( t ) + d M m - 1 , n d t ( t ) + d P m , n - 1 d t ( t ) (38)
where Mm,n is an inhomogeneous Poisson process with rate ϵ 1 ( g * d N m , n d t ) ( t ) and Pm,n is an inhomogeneous Poisson process with rate ϵ 2 ( g * d N m , n d t ) 2 ( t ) and N0,0(t) is an inhomogeneous Poisson process with rate λ(t). To generate a set of events in N at order m in the linear term of ϕ and order n in the quadratic term of ϕ, we take events at each previous order, (m − 1, n) and (m, n − 1) and use those to develop Mm−1,n(t) and Pm,n−1(t). These, together with N0,0(t), give d N m , n d t ( t ). In contrast to the linear self-exciting process, the quadratic process here is recursively defined on a lattice.
Similar to the case of the linearly self-exciting process, we can use this recursive definition to develop an expansion for the mean-field rate and propagator in powers of ϵ1 and ϵ2. When we calculate higher-order cumulants, we will identify the expansions of the mean-field firing rate and propagator which will allow us to use them to simplify the resulting diagrams. The mean-field rate to finite order in m, n is once again given by neglecting second and higher-order cumulants of Nn,m, which allows us to take an expectation inside the quadratic term of Eq 38. Taking the expectation of both sides of this equation in the mean field approach then yields:
r ¯ m , n ( t ) = λ ( t ) + ϵ 1 ( g * r ¯ m - 1 , n ) ( t ) + ϵ 2 (g * r ¯ m , n - 1) 2 ( t ) . (39)
For example,
r ¯ 1 , 1 = λ ( t ) + ϵ 1 ( g * r ¯ 0 , 1 ) ( t ) + ϵ 2 ( g * r ¯ 1 , 0 ) 2 ( t ) (40)
where
r ¯ 1 , 0 ( t ) = λ ( t ) + ϵ 1 ( g * λ ) ( t ) (41)
r ¯ 0 , 1 ( t ) = λ ( t ) + ϵ 2 (g * λ) 2 ( t ) . (42)
Similarly, the propagator (for the dynamics of the recursive process, linearized around zero) is, to finite order in m, n:
Δ m , n ( t , t ′ ) = δ ( t - t ′ ) + ϵ 1 ( g * Δ m - 1 , n ) ( t , t ′ ) + 2 ϵ 2 (g * r ¯ m , n - 1) ( t ) (g * Δ m , n - 1) ( t , t ′ ) . (43)
To zeroth order in ϵ2, this yields an expansion of the mean-field rate r ¯ ( t ), which takes the same form as the expansion of the rate of the linearly self-exciting process (Eq 13) and admits the same graphical representation (Fig 4b). Similarly, a perturbative expansion of the linear response about the mean-field rate to zeroth order in ϵ2 takes the same form as for the linearly self-exciting process (Eq 28) and admits the same graphical representation (Fig 4c).
To account for the nonlinear terms arising at first order and greater in ϵ2, we will need to add another type of internal vertex in diagrammatic descriptions of the cumulants. These vertices, carrying factors of ϵ2, will have two incoming edges and any number of outgoing edges. Each incoming edge carries the operator g* and the number of incoming edges corresponds to the order in the Taylor expansion of the nonlinearity. (It also corresponds to the order of cumulant influencing that vertex’s activity. The number of outgoing edges corresponds to the order of cumulant being influenced, locally in that subgraph.) The factor of r ¯ ( t ) that appears in other vertices is modified to be consistent with the mean firing rate under the quadratic nonlinearity, and will thus obey Eq 35 above.
The mean-field rate and propagator, to first order and greater in ϵ2, can be represented diagrammatically using the new vertex (e.g., Fig 8a and 8b). Notice that these directed graphs are treelike, but with their leaves in the past. Repeating these calculations to the next order in ϵ2 can be accomplished by taking the basic structure of Fig 8 and, along each leg entering the new vertex for ϵ2, inserting the previous-order graphs (Figs 4a and 8a). Including higher-order terms in ϵ1 would require inserting those graphs along the ϵ-carrying vertices of Fig 4a.
In addition to expanding the mean field rate and propagator, we can use Eq 38 to calculate cumulants of d N d t to finite order in ϵ1 and ϵ2. The first nonlinear correction to the firing rate appears at first order in ϵ2:
⟨d N 0 , 1 d t ( t )⟩ c = λ ( t ) + ϵ 2 ⟨(g * d N 0 , 0 d t) 2 ( t )⟩ c (44)
which can be represented diagrammatically using the new vertex (Fig 9a). Notice that in contrast to the corresponding graph for the mean field expansion (Fig 8a), this diagram has a “loop” (a cycle were it an undirected graph). This reflects the dependence of the rate on the second cumulant of the baseline process N0,0. This dependence of the firing rate on higher-order spiking cumulants is a fundamental feature of nonlinearities.
Proceeding beyond the first order in both ϵ1 and ϵ2, we see that the expansion of each term of the nonlinearity depends on the other:
⟨d N 1 , 1 ( t ) d t⟩ c = λ ( t ) + ϵ 1 g * ⟨d N 0 , 1 d t ( t )⟩ c + ϵ 2 ⟨(g * d N 1 , 0 d t) 2 ( t )⟩ c (45)
so that at each order in ϵ2 we must insert the solution at the previous order in ϵ2 and the same order in ϵ1 (and vice versa). This recursive mixing of expansions between the linear and nonlinear terms of ϕ seems intractable. However, this joint expansion can be re-summed; for a more formal derivation, see Methods: Path integral representation. For the linear model, this re-summing left us with simple expressions for the firing rate and propagator. For the nonlinear model, re-summing leaves us with a new expansion organized by the number of loops appearing the Feynman diagrams, so it is called a loop expansion. The Feynman rules for the re-summed diagrams of the nonlinearly self-exciting process are given in Fig 10. For a quadratic nonlinearity, these rules allow us to write the one-loop contribution to the firing rate (Fig 9b):
r 1 ( t ) = ∫ t 0 t d t 1 ∫ t 0 t d t 2 Δ ( t , t 1 ) ϕ ( 2 ) 2 (( g * Δ ) ( t 1 , t 2 )) 2 r ¯ ( t 2 ) (46)
where ϕ(2) is evaluated at ( g * r ¯ ) ( t 1 ) + λ ( t 1 ). These Feynman rules also allow us write down the one-loop correction to the propagator (Fig 9c):
Δ 1 ( t , t ′ ) = ∫ t 0 t d t 1 ∫ t 0 t d t 2 Δ ( t , t 1 ) ϕ ( 2 ) 2 (( g * Δ ) ( t 1 , t 2 )) 2 (g * Δ) ( t 2 , t ′ ) . (47)
The appearance of a loop in the Feynman diagram for the mean rate of the quadratically self-exciting process is a general feature of nonlinearities. It represents the influence of higher-order spike train cumulants on lower order ones. In order to measure that dependency, we can count the number of loops in the graphs. To do this, we add a bookkeeping variable, h. We count the number of loops by multiplying factors of h and 1/h. Each internal vertex adds a factor of 1/h and each outgoing edge a factor of h. In this way, every vertex with more than one outgoing edge will contribute a factor of h for every edge beyond the first. h thus effectively counts the order of fluctuations contributed by the vertex. For example, the mean for the linear self-exciting process has a graph with a single internal vertex and a single internal edge, so it is zeroth order in h (Fig 4b). The two-point function, however, having two edges and one internal vertex (Fig 5b), is first order in h. Similarly, the tree-level diagrams will always contribute a total of hn−1, where n is the order of the cumulant.
In terms of powers of h, a graph for a nth order cumulant with one loop will be equivalent to a graph for a n + 1st order cumulant with one less loop. Consider cutting one of the lines that form the loop in Fig 9b at the internal vertex and leaving it hanging. Now the graph for the one-loop correction to the mean rate appears to be a graph for a second cumulant–it has two endpoints. The power counting in terms of h, however, has not changed. The one-loop correction to the mean is of the same order in h as the tree-level second cumulant. In general, we will have that the order hm will be given by
m = n + l - 1 (48)
where n is the number of external vertices and l is the number of internal loops. The number of loops thus tracks the successive contributions of the higher order fluctuations. This expansion is called the “loop” expansion and is equivalent to a small-fluctuation expansion. If one can control the size of the fluctuations, one can truncate the loop expansion as an approximation for the statistics of the system. One way of doing this with networks is to insure that the interactions are O ( 1 / N ) so that h ∝ 1/N and the expansion becomes a system size expansion.
The Taylor expansion of an arbitrary nonlinearity ϕ could have infinitely many terms. This would lead, in the recursive formulation, to infinitely many processes {M, P, …}. Even after re-summing the recursive formulation, this would leave an infinite number of diagrams corresponding to any given cumulant. There are two ways to approach this. The first is to insist on a perturbative expansion in the nonlinear terms, e.g., only consider terms up to a fixed order in the Taylor expansion of the nonlinearity ϕ.
The second approach to controlling the order of the loop expansion is to consider a regime in which mean field theory is stable as this will also control the fluctuations, limiting the magnitude of the loop contributions [39]. We expect that the bookkeeping variable h could be related to the largest eigenvalue of the tree-level propagator, λ1, so that m-loop corrections would scale as λ 1 m. The expansion then breaks down in the regime of a bifurcation or “critical point.” In this case, the spectrum of the linear response diverges, causing all loop diagrams to similarly diverge. This is a fluctuation-dominated regime in which mean field theory, along with the fluctuation expansion around it, fails. In that case, renormalization arguments can allow discussion of the scaling behavior of correlations [40].
No new concepts are required in moving from a nonlinear self-exciting process to a network of interacting units. Each external and internal vertex must now be associated with a unique neuron index i and the integrations over time for the internal vertices must now be accompanied by summations over the indices of the internal vertices. In addition, the filter g(τ) must be expanded to include coupling across units. In general, this is given by gij(τ) for the coupling from neuron j to neuron i. We will consider the general model of a network of units that generate conditionally Poisson-distributed events, given an input variable. The conditional rate for unit i is given by
r i ( t ) = ϕ i (∑ j (g i j * d N j d t) ( t ) + λ i ( t )) . (49)
Similarly, the propagator now obeys
Δ i j ( t , t ′ ) = ϕ i ( 1 ) · (∑ k ( g i k * Δ k j ) ( t , t ′ )) + δ i j δ ( t - t ′ ) . (50)
These dynamics are qualitatively the same as those of the nonlinearly self-exciting process (Eq 35 but replace the neuron’s own rate with the sum over its presynaptic inputs). Introducing these sums over neuron indices yields the complete set of rules for generating Feynman diagrams for the cumulants of this model (Fig 11).
Our methods predict how spike time statistics of all orders emerge from the interplay of single-neuron input-output properties and the structure of network connectivity. Here we demonstrate how these methods can be used to predict key phenomena in recurrent spiking networks: the fluctuations and stability of population activity. First, we isolate the contributions of nonlinearities in single-neuron dynamics to network activity and coding as a whole. We do so by computing “one-loop” correction terms; these correspond to the first structures in our diagrammatic expansion that arise from nonlinear neural transfer. The one-loop corrections provide for the dependence of nth order spiking cumulants on n + 1st order cumulants. Predictions that would be made by linearizing neural dynamics, as in classic approaches for predicting pairwise correlations [19, 22, 41] and recent ones for higher-order correlations [38], are described as “tree-level.” We show how these one-loop corrections, which give new, explicit links between network structure and dynamics (Fig 12), predict spiking statistics and stability in recurrent networks.
In our analysis of the impact of nonlinear neural transfer on network dynamics, a principal finding was that spike correlations could affect firing rates, as described by the one-loop correction to the mean-field firing rates. In this section we illustrate the importance of this effect in a class of networks under intensive study in neuroscience: randomly connected networks of excitatory and inhibitory cells. We began with a network for which we expect classical theoretical tools to work well, taking the neurons to have threshold-linear transfer functions ϕ(x) = α⌊x⌋. Here, as long as the neurons do not receive input fluctuations that push their rates below this threshold, the tree-level’ theory that takes transfer to be entirely linear should work well. We then move on to consider nonlinear effects.
As in our original motivational example, we took network connectivity to be totally random (Erdös-Rényi), with pEE = 0.2 and pEI = pIE = pII = 0.5. The magnitude of all connections of a given type (E − E, etc.) was taken to be the same and the time course of all network interactions was governed by the same filter g ( t ) = t τ 2 exp ( - t / τ ) (with τ = 10 ms), so that gij(t) = Wij g(t). (The matrix W contains synaptic weights.) The net inhibitory input weight on to a neuron was, on average, twice that of the net excitatory input weight.
We examined the spiking dynamics as the strength of synaptic weights proportionally increased (Fig 14a) and studied network activity with using both theory and direct simulation. Due to the high relative strength of inhibitory synapses in the network, firing rates decreased with synaptic weight (Fig 14d). The magnitude of spike train covariances (reflected by the integrated autocovariance of the summed excitatory population spike train) increased (Fig 14e). These changes were also visible in raster plots of the network’s activity (Fig 14b and 14c).
At a critical value of the synaptic weights, the mean field theory for the firing rates loses stability (Fig 14f). The location of this critical point is predicted by the linear stability of the dynamics around the mean-field rate; the spectrum of the propagator Δ(ω) = (I − ϕ(1) g(ω))−1 diverges when the spectral radius of ϕ(1) g is ≥1. (This is also the point where the spectral radius of the inverse propagator crosses zero.) Until that critical point, however, the tree-level predictions for both firing rates and spike train covariances (i.e., mean field theory and linear response theory) provided accurate predictions (Fig 14d and 14e).
We next give a simple example of how nonlinearities in neural transfer cause this standard tree-level theory (mean-field formulas for rates and linear response theory for covariances) to fail–and how tractable one-loop corrections from our theory give a much-improved description of network dynamics. We take the same network as above, but replace neurons’ threshold-linear transfer functions with a rectified power law ϕ(x) = α⌊x⌋p (Fig 15a). This has been suggested as a good description of neural transfer near threshold [42–46]. For simplicity, we take the power law to be quadratic (p = 2). As we increased synaptic weights, the tree-level theory qualitatively failed to predict the magnitude of spike train covariances and firing rates (Fig 15d and 15e black curve vs. dots). This occurred well before the mean-field firing rates lost stability (Fig 15f, black).
Higher-order terms of the loop expansion described above (Nonlinearities impose bidirectional coupling between different orders of activity) provide corrections to mean field theory for both firing rates and spike train correlations. These corrections represent coupling of higher-order spike train cumulants to lower order cumulants. In the presence of an input-rate nonlinearity, for example, synchronous (correlated) presynaptic spike trains will more effectively drive postsynaptic activity [7, 47]. This effect is described by the one-loop correction to the firing rates (Fig 9).
The one-loop correction for the mean field rate of neuron i in a network is given by the same diagram as the one-loop correction for the nonlinearly self-exciting process, Fig 9, but interpreted using the network Feynman rules (Fig 11). This yields:
r i , 1 = ∫ t 0 t d t 1 ∫ t 0 t d t 2 ∑ j , k Δ i j ( t , t 1 ) 1 2 ϕ j ( 2 ) (∑ l g j l * Δ l k ( t 1 , t 2 )) 2 r ¯ k ( t 2 ) (52)
where t0 is the initial time. This correction was, on average, positive (Fig 15d for excitatory neurons; also true for inhibitory neurons). Similarly to firing rates, the loop expansion provides corrections to higher-order spike train cumulants. The one-loop correction to the spike train covariances (Eq 51, derived in Fig 13) accounts for the impact of triplet correlations (third joint spike train cumulants) on pairwise correlations and provided an improved prediction of the variance of the population spike train as synaptic weights increased (Fig 15e).
Since the one-loop correction to the firing rates could be large, we also asked whether it could impact the stability of the firing rates–that is, whether pairwise correlations could, through their influence on firing rates through the nonlinear transfer function, induce an instability. This is a question of when the eigenvalues of the propagator diverge—or equivalently, when the eigenvalues of the inverse propagator cross zero. The one-loop correction to the inverse propagator is given by the “proper vertex” obtained by amputating the outside propagator edges of the one-loop correction to the propagator; or equivalently, calculated from the Legendre transform of the cumulant-generating functional [39]. We can heuristically derive the one-loop stability correction as follows.
The full propagator, Δ, obeys the expansion
Δ = Δ ¯ + Δ 1 + Δ 2 + … (53)
where Δ ¯ is the tree-level propagator, Δ1 is the one-loop correction, two-loop corrections are collected in Δ2, and so on (Fig 16a). Notice from the diagram (Fig 16a, second diagram on the right-hand side) that the one-loop correction to the propagator begins and ends with the tree-level propagator, Δ ¯. We will label the bubble in the middle of the diagram Γ1. The first two-loop correction is a chain of loops (Fig 16a), and so can also be factored as Δ ¯ Γ 1 Δ ¯ Γ 1 Δ ¯. We can represent this factorization diagrammatically by pulling out the tree-level propagator and the loop Γ1 (Fig 16b). Just as at two-loop order we were able to factor out a factor Δ ¯ Γ 1 and obtain the expansion of the propagator to one loop, continuing to higher-orders in the loop expansion of the full propagator would all the rest of the full propagator with factors of Δ ¯ Γ 1 in front. The remaining terms would have factors starting with the two-loop correction, and so forth.
Pulling out all terms of Eq 53 that begin with Δ ¯ Γ 1 and summing them allows us to write (Fig 16b):
Δ = Δ ¯ + Δ ¯ Γ 1 Δ + O ( 2 loops ) (54)
We now truncate at one loop and operate on both sides with the inverse of the tree-level propagator:
Γ 0 Δ ≈ Γ 0 Δ ¯ + Γ 0 Δ ¯ Γ 1 Δ (55)
= I δ + Γ 1 Δ , (56)
revealing that −Γ1 is the one-loop correction to the inverse propagator. From the Feynman rules (Fig 11), that factor is:
Γ j m , 1 = ∫ t 0 t d t 1 ∫ t 0 t d t 2 ∑ k 1 2 ϕ j ( 2 ) (∑ l g j l * Δ l k ( t 1 , t 2 )) 2 ϕ k ( 1 ) g k m (57)
where ϕ j ( 2 ) denotes the second derivative of the transfer function of neuron j, evaluated at its mean-field input (and similar for ϕ k ( 1 )). The eigenvalues of this provide a correction to the stability analysis based on the tree-level propagator. This predicts that the firing rates should lose stability significantly before the bifurcation of the mean field theory (Fig 15f, red vs. black). Indeed, we saw in extended simulations that the spiking network could exhibit divergent activity even with synaptic weights that mean field theory predicted should be stable (Fig 1c). In summary, mean field theory can mis-predict the bifurcation of the rate of spiking models since it fails to capture the impact of correlations on firing rates through nonlinear transfer functions.
Recent work has shown that cortical networks are more structured than simple Erdős-Rényi networks (e.g., [48–53]). One feature of cortical networks is a broad spread of neurons’ in- and out-degree distributions (i.e., the distributions of the number of synaptic inputs that each neuron receives or sends); another is broadly spread synaptic weights. These network properties, in turn, can have a strong impact on population activity [54–58]. Here, we illustrate the link between network structure and activity in the presence of nonlinear neural transfer. To generate structured networks, we began with the type of excitatory-inhibitory networks discussed in the previous section, but took the excitatory-excitatory coupling to have both heavy-tailed degree and weight distributions. Specifically, we took it to have truncated, correlated power law in- and out-degree distributions (Methods: non-Erdős-Rényi network model). We then took to the synaptic weights to be log-normally distributed [48, 59]. For simplicity, we took the location and scale parameters of the weight distribution to be the same.
We then examined the network dynamics as the location and scale of the excitatory-excitatory synaptic weights increased. For each mean weight, we sampled the excitatory-excitatory weights from a lognormal distribution with that mean and variance. The excitatory-inhibitory, inhibitory-excitatory and inhibitory-inhibitory weights remained delta-distributed. Each such network specified a weight matrix W, which allowed the methods described previously for computing tree-level and one-loop rates, covariances and stability to be straightforwardly applied. For strong and broadly distributed synaptic weights (Fig 17b), the network exhibited a similar correlation-induced instability as observed in the Erdős-Rényi network (Fig 17c) even though mean field theory predicted that the firing rates should be stable (Fig 17f, black vs. red curves). As synaptic weights increased from zero, the mean field theory for firing rates provided a misprediction (Fig 17d, black line vs. dots) and the linear response prediction for the variance of the population spike train also broke down (Fig 17e, black line vs. dots). The one-loop corrections, accounting for the impact of pairwise correlations on mean rates and of triplet correlations on pairwise correlations, yielded improved predictions (Fig 17d and 17e red lines) and a much more accurate prediction for when firing rates would lose stability (Fig 17c and 17f). These effects were similar to those seen in Erdős-Rényi networks (Fig 15), but the transition of the firing rates occurred sooner, both for the mean field (because of the effect of the weight and degree distributions on the eigenvalues of the weight matrix) and one-loop theories (because of the impact of the correlations on the firing rates).
In the previous section, we investigated how a non-Erdős-Rènyi network structure could amplify the one-loop corrections by increasing spike train correlations. We now examine a different single-neuron nonlinearity: ϕ(x) = αex, which is the canonical link function commonly used to fit GLM point process models to spiking data [26].
As before, we take the mean synaptic weight onto each neuron in the network to be 0. First, we take excitatory and inhibitory neurons to have the same baseline drive, λE = λI = −1.5. As we scale synaptic weights, we see that the one-loop correction is small compared to the tree-level theory for the firing rates, population variances and stability analysis (Fig 18a–18c, red vs. black lines). It nevertheless provides an improved correction for the variance of the excitatory population spike train (Fig 18b, between 1.5 and 2 mV synaptic weights). The bifurcation of the one-loop theory is close to the bifurcation of the mean field theory, and before that point the mean field theory and one-loop corrections both lose accuracy (Fig 18a and 18b). This makes sense: when the mean field theory fails, the only reason that the one-loop correction to the rates would be accurate is if all third- and higher-order spike train cumulants were small. Those higher-order correlations are not small near the instability.
Next, we broke the symmetry between excitatory and inhibitory neurons in the network by giving inhibitory neurons a lower baseline drive (λI = −2.). This shifted the bifurcation of the mean field theory and the one-loop correction to much higher synaptic weights (Fig 19c). For intermediate synaptic weights, we saw that the one-loop correction provided a better match to simulations than the tree-level theory (Fig 19a and 19b, between 1 and 1.5 mV synaptic weights). For stronger synapses, however, the simulations diverged strongly from the tree-level and one-loop predictions (Fig 19a and 19b, around 1.5 mV synaptic weights). In principle, we could continue to calculate additional loop corrections in an attempt to control this phenomenon. The exponential has arbitrary-order derivatives, however, preventing a perturbative expansion of the nonlinear terms—suggesting a renormalization approach [39], which is beyond the scope of this article. In sum, with an exponential transfer function, we saw that for intermediate synaptic weights, the one-loop correction improved on the tree-level theory. For strong enough synaptic weights, however, both failed to predict the simulations. How soon before the mean-field bifurcation this failure occurred depended on the specific model.
Joint spiking activity between groups of neurons can control population coding and controls the evolution of network structure through plasticity. Theories for predicting the joint statistics of activity in model networks have been locally linear so far. We present a systematic and diagrammatic fluctuation expansion (or, in reference to those diagrams, loop expansion) for spike-train cumulants, which relies on a stochastic field theory for networks of stochastically spiking neurons. It allows the computation of arbitrary order joint cumulant functions of spiking activity and dynamical response functions, which provide a window into the linear stability of the activity, as well as nonlinear corrections to all of those quantities.
Using this expansion, we investigated how nonlinear transfer can affect firing rates and fluctuations in population activity, imposing a dependence of rates on higher-order spiking cumulants. This coupling could significantly limit how strong synaptic interactions could become before firing rates lost stability.
Dynamical mean field theory is a classic tool for analyzing the rate models with Gaussian-distributed synaptic weights, which reveals a transition to chaotic rate fluctuations with strong connectivity [60]. This proceeds, briefly, by taking the limit of large networks and replacing interactions through the quenched heterogeneity of the synaptic weights by an effective Gaussian process mimicking their statistics. Recent extensions have incorporated a number of simple biological constraints, including non-zero firing rates [61–63] and certain forms of cell type-specific connectivity [61, 64–66]. In this framework, spiking is usually only described in the limit of slow synapses as additive noise in the rates, which can shift the transition to chaotic rate fluctuations to higher coupling strengths and smooth the dynamics near the transition [61, 67].
An alternative approach is to start from the bottom up: to posit an inherently stochastic dynamics of single neurons and specify a finite-size network model, and from these derive a set of equations for statistics of the activity [68, 69]. This approach provides a rigorous derivation of a finite-size rate model as the mean field theory of the underlying stochastic activity, as well as the opportunity to calculate higher-order activity statistics for the activity of a particular network [40, 70–72]. This is the approach taken here with the popular and biologically motivated class of linear-nonlinear-Poisson models. A similar approach underlies linear response theory for computing spike train covariances [73], which corresponds to the tree level of the loop expansion presented here. For integrate-and-fire neuron models receiving Poisson inputs, Fokker-Planck theory for the membrane potential distributions can be used to calculate the linear response function of an isolated neuron [74], which together with the synaptic filter and weight matrix determines the propagator Δ.
The loop expansion is organized by the dependence of lower-order activity cumulants to higher-order ones. The first-order (tree level) description of the nth activity cumulant does not depend on higher-order cumulants. One-loop corrections correspond to dependence of the order n cumulants on the tree-level n + 1-order cumulants, two-loop corrections correspond to dependence of the order n cumulant on the tree-level n + 1 and n + 2 order cumulants and so on. This coupling arises from the nonlinearity of the single-neuron transfer function ϕ (Results: Nonlinearities impose bidirectional coupling between different orders of activity: nonlinearly self-exciting process; Methods: Path integral representation). When the transfer function is linear at the mean-field rates, the tree-level theory provides an accurate description of activity (Fig 14). This corresponds to the second- and higher-order derivatives of the transfer function ϕ with respect to the total input, evaluated at the mean-field rates, being zero. When ϕ has non-zero second- or higher derivatives at the mean-field rates, higher orders of the loop expansion can be important. The magnitude of n-loop corrections depends on two things: the magnitude of the n + 1-order tree-level activity cumulants and the magnitude of the n + 1st derivative of ϕ at the mean-field rates (i.e., the strength of the coupling to that cumulant).
Recent work has shown that the the magnitude of spike train correlations depend on the motif structure of the network (Fig 17; [41, 55, 56, 75]), as well as on the correlation structure of the inputs it receives. For a particular model network of interest, the accuracy of a truncated loop expansion can be evaluated empirically by comparing against simulations. Computing higher-order loop corrections can quickly become unwieldy. The ratio of the n-loop correction to a particular spike train cumulant to the n + 1 loop correction can provide an estimate (but not a guarantee) of whether truncating at n loops is sufficient. If the loop contributions scale with a small parameter such as the inverse system size, this provides a natural justification for truncation. This can occur if the tree-level cumulants scale with 1/N, which happens if synaptic weights scale as 1/N [23, 71], or (in the thermodynamic limit) if connections are strong but sparse [76], or if strong, dense, balanced excitation and inhibition actively cancel correlations [77]. Finally, if the system lies close to a bifurcation of the mean field theory so that the eigenvalues of the propagator diverge, then the mean field theory and this expansion around it can also fail. In that case, renormalization arguments can allow the discussion of the scaling behavior of correlations [40].
For Fig 17, we generated the excitatory-excitatory connectivity with a truncated power-law degree distribution. The marginal distributions of the number of excitatory-excitatory synaptic inputs (in-degree) or outputs (out-degree) obeyed:
p ( d ) = { C 1 d γ 1 , 0 ≤ d ≤ L 1 C 2 d γ 2 , L 1 < d ≤ L 2 0 , else (58)
where d is the in- or out-degree. Parameter values are contained in Table 1; C1 and C2 are normalization constants to make the degree distribution continuous at the cutoff L1. The in- and out-degree distributions were then coupled by a Gaussian copula with correlation coefficient ρ to generate in- and out-degree lists. These lists generated likelihoods for each possible connection proportional to the in-degree of the postsynaptic neuron and the out-degree of the presynaptic neuron. We then sampled the excitatory-excitatory connectivity according to those likelihoods.
Here we outline the derivation of a path integral formalism for a network of processes with nonlinear input-rate transfer, following methods developed in nonequilibrium statistical mechanics [86–91]. We will begin by developing the formalism for a simple model, where a spike train is generated stochastically with intensity given by some input ν(t). We will specify the cumulant generating functional of the spike train given ν(t) below.
The general strategy is to introduce an auxiliary variable, called the “response variable,” whose dynamics will determine how information from a given configuration (e.g. the spike counts n(t)) at one time will effect future configurations of the dynamics. Introducing the response variable allows us to write the probability density functional for the process in an exponential form. The integrand of that exponential is called the “action” for the model, which can then be split into a “free” action (the piece bilinear in the configuration and response variables) and an “interacting” one (the remainder). Cumulants of the process can then be computed, in brief, by taking expectations of the configuration and response variables and the interacting action against the probability distribution given by the free action.
Let n(t) be the number of spike events recorded since some fiducial time t0. In a time bin dt, Δn events are generated with some distribution p(Δn) and added to n(t). Let the events generated in any two time bins be conditionally independent given some inhomogeneous rate ν(t), so that p(Δn) = p(Δn|ν). So, assuming that initially n(t0) = 0, the probability density functional of the vector of events over M time bins is:
p [ Δ n ( s ) : s ≤ t ] = ∏ i = 1 M p ( Δ n i | ν i ) = ∏ i = 1 M ∫ d n ˜ i 2 π i e - n ˜ i Δ n i P ( n ˜ i | ν i ) = ∏ i = 1 M ∫ d n ˜ i 2 π i e - n ˜ i Δ n i + W [ n ˜ i | ν i ] (59)
where P ( n ˜ i | ν i ) is the Laplace transform of p(Δni|νi) and W [ n ˜ i | ν i ] is the cumulant generating functional for the auxiliary variable. In the third step we have written the distribution of p(Δni) as the inverse Laplace transform of the Laplace transform. The Laplace transform variable n ˜ i is our auxiliary response variable. In the fourth step we identified the Laplace transform of the probability density functional as the moment-generating functional, so that W [ n ˜ i | ν i ] is the cumulant generating functional of the spike count. Note that these are complex integrals. The contour for the integration over n ˜ i is parallel to the imaginary axis.
Taking the continuum limit M → ∞, dt → 0 then yields the probability density functional of the spike train process n ˙:
p [ n ˙ ] = ∫ D n ˜ ( t ) e - ∫ d t ( n ˜ ( t ) n ˙ ( t ) - W [ n ˜ ( t ) ] ) (60)
where D n ˜ ( t ) = lim M → ∞ ∏ i = 1 M d n ˜ i 2 π i and n ˙ = d n d t and we suppress the conditional dependence of n ˜ ( t ) on ν(t). In the continuum limit the integral is a functional or path integral over realizations of n ˜ ( t ). We will call the negative exponent of the integrand in Eq 60 the action:
S [ n ˜ , n ˙ ] = ∫ d t (n ˜ n ˙ - W [ n ˜ ]) . (61)
We have slightly abused notation here in that a factor of 1/dt has been absorbed into W [ n ˜ ]. We will justify this below.
We have not yet specified the conditional distribution of the events given the input ν(t), leaving W [ n ˜ ( t ) ] unspecified. Here, we will take the events to be conditionally Poisson [92], so that
W[ n˜]=(en˜−1) ν(t) (62)
(In the continuum limit, the rate ν(t) allowed us to absorb the factor of 1/dt into W. A finite size time bin would produce ν(t)dt events in bin dt.)
This representation of the probability density functional yields the joint moment-generating functional (MGF) of n ˙ and n ˜:
Z [ J , J ˜ ] = ∫ D n ˙ ( t ) ∫ D n ˜ ( t ) e - S [ n ˜ , n ˙ ] + J n ˜ + J ˜ n ˙ (63)
and the moment-generating functional of n ˙:
Z [ J ˜ ] = ∫ D n ˙ ( t ) ∫ D n ˜ ( t ) e - S [ n ˜ , n ˙ ] + J ˜ n ˙ (64)
The above strictly applies only to the inhomogeneous Poisson process. This formalism is adapted to the self-exciting process by introducing conditional dependence of the rate ν(t) on the previous spiking history. In the discrete case, before taking the limit M → ∞, we say that the rate νi = ϕ[ni−], where ϕ is some positive function and ni− indicates all spiking activity up to but not including bin i. This requirement is equivalent to an Ito interpretation for the measure on the stochastic process n ˙ ( t ). Because of this assumption, the previous derivation holds and we can write
W [ n ˜ ] = (e n ˜ ( t ) - 1) ϕ ( n ˙ ( < t ) ) (65)
where n ˙ ( < t ) = n ˙ ( s ) : s < t. In the continuum limit, there is an ambiguity introduced by the appearance of the time variable t in both n ˜ ( t ) and n ˙ ( t ). This is resolved in the definition of the measure for the functional integral and affects the definition of the linear response (below). Again, this is a manifestation of the Ito assumption for our process.
The specific model used in this paper assumes a particular form for the argument of ϕ. We assume that the input is given by
ν ( t ) = ϕ ( ( g * n ˙ ) ( t ) + λ ( t ) ) (66)
where g(t) is a filter that defines the dynamics of the process in question and λ(t) is an inhomogeneous rate function. The result is that the action for nonlinearly self-exciting process is given by
S [ n ˜ , n ˙ ] = ∫ d t (n ˜ n ˙ - (e n ˜ ( t ) - 1) ϕ (( g * n ˙ ) ( t ) + λ ( t ))) . (67)
The only extension required to move from the above action to the network model is to introduce indices labelling the neurons and couplings specific for each neuron pair. Nothing of substance is altered in the above derivation and we are left with
S [ n ˜ , n ˙ ] = ∑ i ∫ d t (n ˜ i n ˙ i - (e n ˜ i ( t ) - 1) ϕ (∑ j ( g i j * n ˙ j ) ( t ) + λ i ( t ))) . (68)
|
10.1371/journal.pbio.1000025 | Neuropilin-1/GIPC1 Signaling Regulates α5β1 Integrin Traffic and Function in Endothelial Cells | Neuropilin 1 (Nrp1) is a coreceptor for vascular endothelial growth factor A165 (VEGF-A165, VEGF-A164 in mice) and semaphorin 3A (SEMA3A). Nevertheless, Nrp1 null embryos display vascular defects that differ from those of mice lacking either VEGF-A164 or Sema3A proteins. Furthermore, it has been recently reported that Nrp1 is required for endothelial cell (EC) response to both VEGF-A165 and VEGF-A121 isoforms, the latter being incapable of binding Nrp1 on the EC surface. Taken together, these data suggest that the vascular phenotype caused by the loss of Nrp1 could be due to a VEGF-A164/SEMA3A-independent function of Nrp1 in ECs, such as adhesion to the extracellular matrix. By using RNA interference and rescue with wild-type and mutant constructs, we show here that Nrp1 through its cytoplasmic SEA motif and independently of VEGF-A165 and SEMA3A specifically promotes α5β1-integrin-mediated EC adhesion to fibronectin that is crucial for vascular development. We provide evidence that Nrp1, while not directly mediating cell spreading on fibronectin, interacts with α5β1 at adhesion sites. Binding of the homomultimeric endocytic adaptor GAIP interacting protein C terminus, member 1 (GIPC1), to the SEA motif of Nrp1 selectively stimulates the internalization of active α5β1 in Rab5-positive early endosomes. Accordingly, GIPC1, which also interacts with α5β1, and the associated motor myosin VI (Myo6) support active α5β1 endocytosis and EC adhesion to fibronectin. In conclusion, we propose that Nrp1, in addition to and independently of its role as coreceptor for VEGF-A165 and SEMA3A, stimulates through its cytoplasmic domain the spreading of ECs on fibronectin by increasing the Rab5/GIPC1/Myo6-dependent internalization of active α5β1. Nrp1 modulation of α5β1 integrin function can play a causal role in the generation of angiogenesis defects observed in Nrp1 null mice.
| The vascular system is a hierarchical network of blood vessels lined by endothelial cells that, by means of the transmembrane integrin proteins, bind to the surrounding proteinaceous extracellular matrix (ECM). Integrins are required for proper cardiovascular development and exist in bent (inactive) and extended (active) shapes that are correspondingly unable and able to attach to the ECM. Extracellular guidance cues, such as vascular endothelial growth factor and semaphorins, bind the transmembrane protein neuropilin-1 (Nrp1) and then activate biochemical signals that, respectively, activate or inactivate endothelial integrins. Here, we show that Nrp1, via its short cytoplasmic domain and independently of vascular endothelial growth factor and semaphorins, specifically promotes endothelial cell attachment to the ECM protein fibronectin, which is known to be crucial for vascular development. Notably, Nrp1 favors cell adhesion by associating with fibronectin-binding integrins and promoting the fast vesicular traffic of their extended form back and forth from the endothelial cell-to-ECM contacts. Binding of the Nrp1 cytoplasmic domain with the adaptor protein GIPC1, which in turn associates with proteins required for integrin internalization and vesicle motility, is required as well. It is likely that such an integrin treadmill could act as a major regulator of cell adhesion in general.
| In vertebrates, the development of a hierarchically organized and functional vascular tree relies on the dynamic interaction of endothelial cells (ECs) with the surrounding extracellular matrix (ECM), which is mediated by heterodimeric αβ integrin adhesive receptors [1]. During evolution, vertebrates have acquired an additional set of adhesion-related genes that regulate blood vessel assembly and function [2]. Among these genes, the ECM protein fibronectin (FN) and α5β1 integrin, the predominant FN receptor, have proven to be essential for embryonic vascular development and tumor angiogenesis [3]. Indeed, in vertebrate embryos FN is the earliest and most abundantly expressed subendothelial matrix molecule [3,4]. Endothelial α5β1 mediates cell adhesion to FN and the assembly of soluble FN dimers (sFN) into a fibrillar network [3], which has also been implicated in branching morphogenesis [5].
The biological activities of integrins depend on the dynamic regulation of their adhesive function in space and time. In cells, integrins exist in different conformations that determine their affinities for ECM proteins [6] and are continuously endocytosed, trafficked through endosomal compartments, and recycled back to the plasma membrane [7,8]. Therefore, during vascular morphogenesis, real-time modulation of EC–ECM adhesion can result from two interconnected phenomena: the regulation of integrin conformation and traffic in response to extracellular stimuli [8,9]. Indeed, there is mounting evidence that pro- and antiangiogenic cues regulate blood vessel formation by modulating integrin function [1]. In this respect, the transmembrane glycoprotein neuropilin 1 (Nrp1), which is expressed in both neurons and ECs [10], is remarkable because it was originally identified as a surface protein mediating cell adhesion [11] and then found to also act as a coreceptor for both pro- and antiangiogenic factors, such as vascular endothelial growth factor A 165 (VEGF-A165, VEGF-A164 in mice) [12,13] and semaphorin 3A (SEMA3A) [14–20], respectively.
The extracellular region of Nrp1 contains two repeated complement-binding domains (CUB domains; a1-a2 domains), two coagulation-factor-like domains (b1-b2 domains), and a juxtamembrane meprin/A5/μ-phosphatase (MAM; c) homology domain. The Nrp1 intracellular region is only 50 amino acids in length, and its function is poorly characterized [21]. Through its b1-b2 domains, Nrp1 binds and potentiates the proangiogenic activity of VEGF-A165, which contains the heparin-binding peptide encoded by exon 7 [13]. In addition, Nrp1 acts as the ligand-binding subunit of the receptor complex for the antiangiogenic SEMA3A [14–20], whose sema and immunoglobulin-basic domains, respectively, bind the a1-a2 and b1-b2 domains of Nrp1 [21]. The MAM/c domain instead mediates the SEMA3A-elicited Nrp1 oligomerization that is required for SEMA3A biological activity [21]. Interestingly, the short cytoplasmic domain of Nrp1 is not required for SEMA3A signaling in neurons [22]. In addition, the extracellular b1-b2 domains of Nrp1 mediate heterophilic cell adhesion independently of VEGF-A165 and SEMA3A [11].
Nrp1 null mice display an embryonic lethal phenotype, characterized by dramatic vascular defects ascribed to impaired angiogenic sprouting [23], branching [24], or arterialization [25] that is significantly more severe than and/or qualitatively different from that of mice lacking either VEGF-A164 (VEGF120/120 mice) [26] or Sema3A [16]. Indeed, although Nrp1–/– embryos die in utero by 13.5 days postcopulation [23], VEGF120/120 pups are recovered at birth at a normal Mendelian frequency [27]. Moreover, a major feature of Nrp1 null mutants, i.e., the severe impairment of neural tube vascularization [23], is not phenocopied by VEGF120/120 mouse embryos [26]. In addition, differently from Nrp1 null mice [23,24], the vascular phenotype of Sema3a null mice is significantly influenced by the genetic background [16,28–30]. These findings suggest that the vascular defects caused by the loss of Nrp1 could be due to a VEGF-A164/SEMA3A-independent function of Nrp1 in vascular cells, such as adhesion to the ECM [31,32]. However, how Nrp1 regulates integrin-dependent EC linkages to the surrounding matrix is still obscure. Here, we shed light on the molecular mechanisms by which Nrp1, via its short cytoplasmic domain and independently of VEGF-A165 and SEMA3A, specifically controls a biological function that is crucial for vascular development [3], namely, α5β1-mediated EC adhesion to FN.
To understand the mechanisms by which Nrp1 modulates EC adhesion to different ECM proteins, we silenced the expression of Nrp1 in human umbilical artery ECs by RNA interference (RNAi). Parenthetically, Nrp1 has been found to be expressed at higher levels in arteries than in veins [33]. Endothelial cells were transfected twice with either a pool of three different small interfering RNAs (siRNAs) targeting human Nrp1 (sihNrp1) or control nontargeting siRNA (siCtl). Twenty-four hours after the second transfection, Western blot analysis revealed that, in comparison with control cells, Nrp1 protein, but neither β-tubulin nor the Nrp1 interactor GAIP interacting protein C terminus, member 1 (GIPC1), was successfully silenced in sihNrp1 ECs (Figure 1A).
Next, we investigated the effect of Nrp1 silencing on EC adhesion to different ECM proteins. Fibronectin, vitronectin (VN), and type I collagen (COLL-I) are typical constituents of the provisional angiogenic ECM [1,3], whereas laminin (LN) isoforms are major components of the vascular basement membrane surrounding both immature and mature blood vessels [34]. Short-term (15 min) adhesion assays showed that loss of Nrp1 greatly reduced EC adhesion to FN but not to VN, COLL-I, or LN (Figure 1B–E), suggesting that positive modulation of cell adhesion by Nrp1 is not a general phenomenon [31] but rather a function restricted to specific ECM proteins, such as FN.
Because FN polymerization by ECs has been suggested to participate in vascular morphogenesis [3], we next examined the role of Nrp1 in the fibrillogenesis of endogenous FN. During FN matrix assembly, current models envisage the binding of sFN to surface integrins, thus causing the conversion of FN to a conformational form that favors fibril formation through interactions with other integrin-bound FN dimers [3]. Endothelial cells were cultured in a medium containing FN-depleted fetal calf serum, and accumulation of endogenous FN into fibrils was then detected by confocal immunofluorescence analysis. In comparison with control cells, sihNrp1 ECs were impaired in their ability to incorporate endogenous sFN into a dense fibrillar network 3 h after plating (Figure 1F and 1G). Time-course real-time reverse transcription PCR (RT-PCR) and Western blot analyses revealed that the endogenous FN fibrillogenesis defect observed in sihNrp1 ECs was not due to a reduction in FN mRNA (Figure S1A) or protein (Figure S1D). Hence, Nrp1 specifically promotes EC adhesion to FN and FN matrix formation.
To start dissecting the mechanisms by which Nrp1 controls the interaction of human ECs with FN, we sought to compare the abilities of full-length and deletion constructs of mouse Nrp1 (mNrp1) to rescue the adhesion and fibrillogenesis defects of sihNrp1 ECs (Figure 2). In particular, we investigated the role played by the extracellular and cytoplasmic moieties of Nrp1. Indeed, the Nrp1 cytodomain, although dispensable for SEMA3A collapsing activity in neurons [22], could signal in cultured ECs [35]. Moreover, the C-terminal SEA sequence of Nrp1 interacts with the PDZ domain of the endocytic adaptor protein GIPC1 [36], whose knockdown during development results in altered arterial branching [37]. Therefore, we transduced sihNrp1 ECs with retroviral vectors carrying the hemagglutinin (HA)-tagged full-length (mNrp1) and deletion mutants of murine Nrp1 (Figure 2A), lacking either the C-terminal SEA amino acids (mNrp1dSEA) or the whole cytoplasmic domain (mNrp1dCy). The sihNrp1 pool did not target any of the mNrp1 constructs, and immunoprecipitation experiments on membrane-biotinylated cell monolayers revealed that all three transmembrane proteins were efficiently exposed on the cell surface (Figure 2B).
In comparison to wild-type mNrp1, both mNrp1dSEA and mNrp1dCy constructs were severely impaired in their abilities to rescue sihNrp1 EC defects in adhesion to FN (Figure 2C) and endogenous FN fibrillogenesis (Figure 2D–F). Accordingly, only mNrp1 overexpression stimulated the adhesion of NIH 3T3 fibroblasts to FN, whereas neither mNrp1dSEA nor mNrp1dCy were active in this respect (Figure 2G). Moreover, mNrp1 overexpression did not promote NIH 3T3 adhesion to VN (Figure S2), further supporting the concept that Nrp1 behaves as a substrate-specific enhancer of cell adhesion. Hence, it appears that the cytoplasmic domain of Nrp1, in particular its SEA motif, which interacts with the endocytic adaptor GIPC1 [36], is required for Nrp1 stimulation of EC spreading on FN and polymerization of endogenous FN.
Opposing autocrine loops of VEGF-A [38–41] and SEMA3A [16,19,42,43] have been found in ECs both in vitro and in vivo. Therefore, we investigated whether the SEA motif and the full cytoplasmic domain of Nrp1 could be required for the modulation of EC adhesion to FN by VEGF-A165 and SEMA3A. Consistent with previous observations [16,18,20,44], silencing Nrp1 completely blocked VEGF-A165-dependent stimulation (Figure 2H) and SEMA3A-dependent inhibition (Figure 2I) of human EC adhesion to FN. As expected, inhibition of cell adhesion to FN by SEMA3F, which signals through Nrp2 [20,21], was not affected by Nrp1 knockdown (Figure 2J). Moreover, similarly to what was observed for SEMA3A in neurons [22], we found that the cytoplasmic domain of Nrp1 is entirely dispensable for both VEGF-A165 (Figure 2H) and SEMA3A (Figure 2I) activity on EC adhesion to FN, because all three mNrp1 constructs rescued sihNrp1 EC response to these factors with a similar efficiency. Thus, the Nrp1 SEA motif and cytodomain are required for Nrp1 modulation of EC adhesion to FN and sFN incorporation into fibrils but not for Nrp1 activity as a VEGF-A165 and SEMA3A coreceptor.
α5β1 Integrin is the main FN receptor in ECs [1,3], and by transmitting the actin-dependent tension to sFN, it triggers FN fibrillogenesis [45]. To elucidate whether Nrp1 stimulation of cell adhesion to FN was directly mediated by Nrp1 or was dependent on α5β1 integrin, CHO cells lacking (CHO B2) or expressing (CHO B2α27) the α5 integrin subunit were transfected with mNrp1 and allowed to adhere to FN. Overexpression of mNrp1 stimulated CHO cell adhesion to FN in the presence (CHO B2α27; Figure 3A) but not in the absence (CHO B2; Figure 3B) of α5β1 integrin. Therefore, Nrp1′s proadhesive activity on FN is nonautonomous and mediated by α5β1 integrin.
We then examined whether in ECs Nrp1 could interact physically with α5β1 integrin. Lysates from ECs adhering on endogenous ECM were immunoprecipitated with an antibody (Ab) recognizing the FN receptor α5β1 and then blotted with anti-Nrp1 Ab. Nrp1 coimmunoprecipitated with α5β1, and blotting Nrp1 immunoprecipitates with anti-α5β1-integrin Ab further confirmed the association between endogenous hNrp1 and α5β1 integrin in ECs (Figure 3C). To better understand whether the Nrp1 cytoplasmic domain was required for the interaction with α5β1 integrin, lysates of NIH 3T3 fibroblasts overexpressing HA-tagged full-length or deletion constructs of mNrp1 and green fluorescent protein (GFP)-tagged α5 integrin subunit (α5-GFP) [46] were immunoprecipitated with anti-GFP Ab and then blotted with anti-HA Ab (Figure 3D). We found that both the C-terminal SEA and the cytoplasmic domain of Nrp1 were fully dispensable for its interaction with α5β1 integrin.
To understand the spatial and functional relationships between Nrp1 and α5β1 integrin in ECs, we first generated a monomeric red fluorescent protein (mRFP)-tagged mNrp1 construct (mNrp1-mRFP) that was then cotransfected with α5-GFP in ECs. Fluorescent confocal microscopy showed that at the plasma membrane of ECs adhering on FN mNrp1-mRFP was enriched in close proximity to, or even tightly intermingled with, α5-GFP-containing adhesion sites (Figure 4A, arrows). Moreover, mNrp1-mRFP and α5-GFP fully colocalized in intracellular vesicles (Figure 4A, arrowheads). Notably, immunofluorescence analysis of endogenous endothelial proteins confirmed the spatial links between hNrp1 and vinculin (Figure 4B) or α5β1 integrin (Figure 4C) at either adhesion sites (Figure 4B and 4C, arrows) or vesicular structures located in their proximity (Figure 4C, arrowheads). The observation that Nrp1 and α5β1 colocalization was particularly apparent in intracellular vesicles indicated that these two molecules may associate at or near the point of endocytosis and that they may be internalized as a complex, which is then subsequently disassembled upon recycling to the plasma membrane. We have previously found that endosomal integrin complexes can be preserved by treating the cell with primaquine (PMQ), a receptor recycling inhibitor, prior to lysis [47]. Therefore, we immunoprecipitated α5β1 integrin or Nrp1 from cells that had been treated with PMQ for 10 min and probed for the presence of the α5β1/Nrp1 complex by Western blotting. Pretreatment of the cells with PMQ greatly increased the coprecipitation of α5β1 integrin with Nrp1 and vice versa (Figure 3C), indicating the likelihood that this complex is more stable in endosomes than at the plasma membrane.
To further characterize the interaction between Nrp1 and α5β1 integrin, we measured fluorescence resonance energy transfer (FRET) in live NIH 3T3 cells transfected with α5-GFP alone or cotransfected with α5-GFP and mNrp1 tagged with the fluorescent protein Cherry, an improved version of mRFP (mNrp1-Cherry). Total internal reflection fluorescence (TIRF) illumination [48] was used to selectively excite α5-GFP at the basal cell plasma membrane where ECM adhesions lie. Fluorescence resonance energy transfer was measured by fluorescence lifetime imaging microscopy (FLIM) [49] and was read out as a decrease in donor (GFP) fluorescence lifetime. We found that the α5-GFP fluorescence lifetime was significantly reduced in cells that coexpressed mNrp1-Cherry, indicating that FRET, and thus a close physical interaction, was occurring between α5β1 and Nrp1 at adhesion sites with an 11.5% FRET efficiency (Figure 5).
Taken together, these data indicate that in living cells Nrp1 physically associates with α5β1 at or near sites of cell–ECM contact and that this interaction is likely maintained following internalization of the complex.
The efficiency of cell adhesion and spreading on ECM is generally thought to be proportional to the amount of either active or total (i.e., active and inactive) integrin at the cell surface [1,6]. We found that lack of Nrp1 did not alter the global amount of either total (Figure 1A), as already reported [31], or active α5β1 integrin, as recognized by the mouse monoclonal Ab (mAb) SNAKA51 [45] (Figure S3). Then, we analyzed whether Nrp1 could influence the amount of α5β1 integrin on the endothelial surface. Biotinylation experiments revealed that knocking down human Nrp1 did not diminish the surface levels of either total or active α5β1 integrin in sihNrp1 ECs (Figure 6A), thus suggesting that a mechanism alternative to the control of integrin conformation should be responsible for Nrp1-dependent activation of α5β1 integrin function in ECs.
On the basis of our observations that Nrp1 and α5β1 integrin colocalize in intracellular vesicles (Figure 4A and 4C) and that inhibition of recycling by PMQ increased the association of Nrp1 with α5β1 integrin (Figure 3C), we decided to monitor the effect of Nrp1 knockdown on the internalization of total and active surface α5β1 integrin. Endothelial cells were surface-labeled with cleavable biotin at 4 °C and incubated at 37 °C for different times to allow internalization, and then biotin remaining on cell-surface proteins was cleaved at 4 °C [50]. Integrin internalization was quantified by immunoprecipitation of either total (Figure 6A and 6B) or active (Figure 6A and 6C) α5β1 integrin, followed by Western blot analysis with streptavidin. Notably, although endocytosis of the cell-surface pool of total α5β1 integrin (i.e., active plus inactive heterodimers) was not detectably altered in sihNrp1 cells (Figure 6A and 6B), knockdown of Nrp1 markedly reduced the quantity of active (SNAKA51-positive) α5β1 heterodimers internalized by ECs (Figure 6A and 6C). Taken together, these data indicate that on the cell surface Nrp1 interacts with active α5β1 heterodimers at adhesion sites (Figure 4A and 4C, arrows) and acts to promote their internalization and localization to intracellular vesicles (Figure 4A and 4C, arrowheads).
To visualize the internalization and postendocytic trafficking of the α5β1/Nrp1 complex, we deployed the photoactivatable (PA) α5-GFP (α5-PA-GFP) probe that we had previously used to monitor α5β1 trafficking in human ovarian carcinoma A2780 cells [51]. However, the multitude of fluorescent vesicles travelling to and from the cell surface made it difficult to track the progress of individual α5β1 integrin transport vesicles. Therefore, we used TIRF to restrict the plane of activating fluorescence, such that only α5-PA-GFP present at or near the cell surface became photoactivated. Then we tracked the movement of this photoactivated fraction of α5β1 integrin using time-lapse epifluorecence microscopy. With this novel technique, α5β1 integrin was photoactivated almost exclusively at adhesion sites (mostly fibrillar adhesions), where it colocalized with mNrp1-Cherry (Figure 7, arrows). Photoactivated α5β1 was then rapidly (<6 s) internalized and cotransported with mNrp1-Cherry in small endocytic vesicles (Figure 7, empty arrowheads, and Video S1) that moved away from the fibrillar adhesions. In addition, we found that α5β1 integrin turnover in ECM adhesions was unexpectedly very rapid (Figure S4 and Video S2), with the α5-PA-GFP signal leaving the adhesive sites, accumulating in vesicles, and disappearing by ∼45 s after photoactivation in approximately 50% of the adhesion sites and by ∼115 s in the remaining ones (Video S2).
Having established that α5β1 integrin and Nrp1 are cointernalized at fibrillar adhesions, we wished to determine whether the integrin was then recycled from Nrp1-positive vesicles back to the plasma membrane. To address this, we aimed a pulse of 405-nm laser light at a “single point” corresponding to Nrp1-positive vesicles, leading to the immediate photoactivation of α5-PA-GFP integrin largely within the confines of these structures (Figure 8 and Video S3). During the following 80 s, fluorescence was lost from the photoactivated vesicle, and this was accompanied by a corresponding increase in integrin fluorescence at peripheral elongated structures that look like adhesion sites (Figure 8, red circle and white arrows). On the contrary, when the activating laser was aimed at a cell region devoid of Nrp1-positive vesicles, little or no photoactivation occurred (Figure S5 and Video S4), indicating that the α5-PA-GFP fluorescence detected in Figure 8 was indeed at mNrp1-Cherry vesicles and not at the plasma membrane above and below them.
Taken together, these data indicate that α5β1 integrin and Nrp1 are cointernalized into intracellular vesicles, which are then rapidly returned or recycled to the plasma membrane. Interestingly, both the internalization and the recycling of Nrp1-associated α5β1 integrin occur at the site of adhesion to the ECM.
In the eukaryotic early endocytic pathway, the small GTPase Rab5 is a rate-limiting component that regulates the entry of cargoes from the plasma membrane into the early endosome [52]. Hence, we analyzed the early endocytic steps of active α5β1 integrin in ECs cotransfected with mNrp1-mRFP and Rab5-GFP, which were incubated with the α5β1 integrin activation reporter mAb SNAKA51 for 30 min at 4 °C and then at 37 °C for different time points. Fluorescent confocal analysis indicated that, after 1–3 min of internalization at 37 °C, Nrp1 and active α5β1 integrin colocalized in early Rab5-positive vesicles near the EC plasma membrane (Figure 9A). Accordingly, immunofluorescence analysis of endogenous endothelial proteins confirmed that hNrp1 and Rab5 colocalized in vesicles, many of which were located near adhesion sites (Figure 9B, empty arrowheads), further supporting the view that Nrp1 can induce α5β1-mediated adhesion by promoting the preferential internalization of its active conformation into Rab5-positive early endosomes and the ensuing recycling to newly forming cell–ECM contacts.
Next, to characterize the molecular mechanisms by which Nrp1 regulates the traffic of active α5β1 integrin, we evaluated the abilities of mNrp1 full-length and mutant constructs to rescue the integrin internalization defects that we observed in sihNrp1 ECs. Remarkably, only wild-type mNrp1, but neither mNrp1dSEA nor mNrp1dCy construct, was able to rescue the sihNrp1 EC defects in the endocytosis of active α5β1 integrin (Figure 6D). Therefore, in ECs the SEA motif of Nrp1, which binds the endocytic adaptor GIPC1 [36], is mandatory for Nrp1 stimulation of cell adhesion to FN (Figure 2C), endogenous FN fibrillogenesis (Figure 2D–F), and active α5β1 integrin endocytosis (Figure 6D).
The N-terminal portion of GIPC1 mediates its oligomerization, whereas its central PDZ domain can bind the C-terminal consensus S/T-X-Φ sequence of Nrp1 [36], the α5 integrin subunit [53], and the Rab5/Rab21 interactor protein APPL1 [54,55]. Thus, we theorized that as a result GIPC1 could support the Rab5-dependent early internalization of α5β1 integrin. To test this hypothesis, we silenced the expression of GIPC1 in human umbilical artery ECs by RNAi and examined its effect on α5β1 integrin traffic. Western blot analysis showed that, 96 h after the second transfection, GIPC1 protein, but not β-tubulin, was successfully silenced in sihGIPC1 ECs in comparison with control cells (Figure 10A). Knockdown of GIPC1 in ECs dramatically reduced the amount of internalized total (Figure 10C and 10D) and active (Figure 10C and 10E) α5β1 integrin by ∼70% throughout the whole internalization assay, suggesting that indeed the interaction of α5β1 integrin with GIPC1 is crucial for the endocytosis and the proper functioning of this integrin. Accordingly, short-term adhesion assays showed that, in comparison with control cells, sihGIPC1 ECs adhered poorly to FN (Figure 10B) and much less efficiently assembled endogenous sFN into a fibrillar network (Figure S1H) in comparison with cells transfected with siCtl (Figure S1G). The latter defect was not due to a reduction in FN mRNA or protein levels as demonstrated by real-time RT-PCR (Figure S1B) and Western blotting (Figure S1E). Hence, within Nrp1 the extracellular domain mediates the association with α5β1 integrin, and the C-terminal SEA sequence allows the binding to the endocytic adaptor GIPC1 that stimulates the internalization and traffic of active α5β1 integrin, finally promoting EC adhesion to FN and FN fibrillogenesis.
Because the C terminus of GIPC1 binds to the minus-end-directed motor myosin VI (Myo6) that has also been involved in endocytosis [56], we considered the hypothesis that Myo6 could cooperate with GIPC1 in promoting α5β1 integrin internalization. Interestingly, RNAi-mediated knockdown of Myo6 in human umbilical artery ECs (Figure 10A) resulted in a significant (∼70%) impairment of active α5β1 integrin internalization (Figure 10F and 10H), whereas the total integrin pool was only mildly affected (∼25%; Figure 10F and 10G). These data, together with the fact that sihMyo6 EC adhesion to FN was severely hampered (Figure 10B), indicate that Myo6 cooperates with GIPC1 in the regulation of active α5β1 integrin endocytosis.
Similarly to what we noticed after Nrp1 and GIPC1 knockdown, ECs in which Myo6 was silenced did not efficiently assemble an endogenous FN fibrillar network (Figure S1I) in comparison with cells transfected with siCtl (Figure S1G). However, differently from what we observed in sihNrp1 and sihGIPC1 ECs, the endogenous FN fibrillogenesis defect seen in sihMyo6 ECs was due to an inhibition of FN1 gene transcription mRNA (Figure S1C), which associated to a significant reduction of FN protein levels as well (Figure S1F). Indeed, in addition to its role in cytoplasmic transporting and anchoring, Myo6 is also present in the nucleus, where it promotes the RNA-polymerase-II-dependent transcription of active genes [57]. Here we identify the FN1 gene as a new Myo6 transcriptional target and downstream effector that can bolster EC adhesion and motility.
Defects of developing blood vessels caused by Nrp1 gene knockdown in mice [23,24] are different from vascular malformations displayed by mice lacking either SEMA3A [16] or VEGF-A165 (Vegf-a120/120 mice) [26]. Furthermore, it has been recently reported that Nrp1 is required for EC responses to both VEGF-A165 and VEGF-A121 isoforms, the latter being incapable of binding Nrp1 on the EC surface [58,59]. Therefore, it is conceivable that the vascular abnormalities of Nrp1–/– mice could be due at least in part to the disruption of a VEGF-A165/SEMA3A-independent Nrp1 function. α5β1 Integrin and its ligand FN are key players in vascular development [3]. The data reported here support a model in which Nrp1, through its cytoplasmic domain and independently of its activity as a SEMA3A and VEGF-A165 coreceptor, stimulates GIPC1/Myo6-dependent endocytosis and traffic of active α5β1 integrin, thus promoting EC adhesion to FN and FN fibrillogenesis.
In rescue experiments, where we reintroduced full-length and mutant murine Nrp1 constructs in human ECs in which endogenous hNrp1 was simultaneously knocked down by RNAi, we showed that EC adhesion to FN and polymerization of endogenous sFN into fibrils depend on the cytoplasmic domain of Nrp1, the C-terminal SEA motif representing the minimal sequence required to exert these functions. Importantly, as already shown for SEMA3A-elicited growth cone collapse in neurons [22], we found that the cytoplasmic domain of Nrp1 is instead dispensable for VEGF-A165 stimulation and SEMA3A inhibition of EC adhesion to FN. Moreover, by using two CHO cell clones differing in the expression of α5β1 integrin, we demonstrated that Nrp1 alone does not directly mediate adhesion to FN and that it requires α5β1 integrin. Therefore, we conclude that in ECs, independently of VEGF-A165 and SEMA3A, Nrp1 stimulates α5β1-mediated adhesion to FN and endogenous FN fibrillogenesis via its cytoplasmic SEA motif [36]. This motif, similar to the C-terminal SDA sequence of the α5 integrin subunit [53], selectively and specifically binds the PDZ domain of the homomultimeric endocytic adaptor GIPC1.
It is known that conformational activation of cell-surface integrins supports cell adhesion and spreading, whereas transition of integrins toward an inactive bent conformation causes cell de-adhesion and rounding up [1,60,61]. However, we observed that lack of Nrp1 does not result in the reduction of either active or total α5β1 integrin either at the cell surface or intracellularly. Rather, by combining biochemical analysis with conventional and TIRF/FLIM confocal microscopy, we found that at the plasma membrane Nrp1 is tightly associated with adhesion sites, where it physically interacts with α5β1 integrin. The complex formed between active α5β1 integrin and Nrp1 is then rapidly internalized into Rab5-positive endosomes in an Nrp1-dependent fashion. Interestingly, the integrin is then returned to the plasma membrane from Nrp1-containing vesicles, and this recycling event appears to be targeted to adhesive structures. In addition, although the extracellular domain of Nrp1 is sufficient for its interaction with α5β1 integrin, the C-terminal GIPC1-binding SEA sequence of Nrp1 is necessary for stimulating EC adhesion to FN. Accordingly, knocking down either GIPC1 or its interacting motor Myo6 results in a significant impairment of active α5β1 integrin endocytosis and EC adhesion to FN.
Taken together, our data indicate that, during EC adhesion and spreading on FN, Nrp1, through its extracellular domain, transiently interacts with active α5β1 integrin at adhesive sites and, via its cytoplasmic association with GIPC1, enhances the early endocytosis and the ensuing recycling of active α5β1 integrin to newly forming adhesion sites (Figure 11A). It is therefore likely that fast cycles of endocytosis from and recycling to ECM adhesions of active α5β1 integrin could allow real-time optimization of adhesion during EC spreading on FN. These conclusions are in line with the recent findings by Ivaska and colleagues [8,9] that found how endocytosis of β1 integrins, in addition to their established role in directional migration [7], regulates cell adhesion and spreading as well. In particular, they reported that class V Rab GTPases (for review, see [52]) Rab21 and Rab5 directly bind to several integrin α subunits, α5 included, by interacting with the conserved membrane proximal region GFFKR, which interestingly has been previously implicated in conformational integrin activation [61]. It is thus conceivable that GIPC1 oligomers could favor α5β1 integrin endocytosis by bridging the α5 integrin subunit and the Rab5/Rab21 interactor APPL1, finally stabilizing the interaction between these small GTPases and α5β1 integrin. This could represent a main functional feature distinguishing α5β1 from other integrin heterodimers not interacting with GIPC1. Finally, the fact that by 2 min after activation α5-PA-GFP disappeared from preexisting adhesion sites into vesicles without a concomitant cell retraction suggests the existence of a steady endo-exocytic flow of (active) α5β1 integrins from and toward existing ECM adhesions as well (Figure 11B). This mechanism could allow adherent cells to be always ready to rapidly exchange integrins among cell–ECM contacts in response to extracellular stimuli. Such a scenario is also compatible with a previous study by Ezratty and colleagues [62] and implies that disassembly of ECM adhesions could depend on an imbalance of endocytosis over recycling.
Our observation that Myo6 siRNA severely impairs EC adhesion to FN and results in a significant reduction in the internalization of active α5β1 integrin suggests that Myo6 cooperates with GIPC1 (Figure 11A and 11B) and is compatible with the notion that Myo6 plays a role in the formation and transport of endocytic vesicles along F-actin microfilaments [56]. The decrease in FN mRNA that we noticed in sihMyo6 ECs is likely due to the lack of the transcriptional activity displayed by Myo6 in the nucleus [57] that could depend on a still not fully characterized actin–myosin-based mechanism of transcription [63,64]. Therefore, Myo6 can support EC adhesion and motility by promoting both active α5β1 integrin traffic (Figure 11A and 11B) and FN1 gene transcription (Figure 11C). Additionally, these findings can have significant implications for the biology of α5β1-expressing human carcinomas [51,65], in which Myo6 can be overexpressed and promote metastatic invasion [66–68].
In conclusion, we propose here that Nrp1, in addition to and independently of its role as coreceptor for VEGF-A165 and SEMA3A, stimulates through its cytoplasmic domain the spreading of ECs on FN by increasing the Rab5/GIPC1/Myo6-dependent internalization of active α5β1 integrin. Nrp1 modulation of α5β1-mediated adhesion can play a causal role in the generation of angiogenesis defects observed in Nrp1 null mice. We anticipate that signaling pathways controlling Nrp1 expression in ECs could ultimately modulate the activity of α5β1 integrin. In particular, Nrp1 is a major target of the inhibitory Delta-like 4–Notch signaling pathway [69] that negatively regulates the formation of endothelial tip cells [10]. Higher expression of Nrp1 in tip ECs compared with that in stalk ECs of angiogenic sprouts could differentially modulate α5β1 integrin traffic, thus favoring tip cell adhesion and spreading on FN. Finally, both Nrp1 [70] and α5β1 integrin [71,72] are expressed in pericytes and vascular smooth muscle cells, which have been implicated in vascular remodeling by intussusceptive angiogenesis [73]. Further work is needed to assess whether Nrp1 is regulating α5β1 integrin function not only in ECs but also in pericytes and vascular smooth muscle cells.
Goat polyclonal anti-Nrp1 (C-19) and rabbit polyclonal anti-β-tubulin (H-235) were from Santa Cruz Biotechnology. Mouse monoclonal anti-human-Nrp1 (MAB 3870) was from R&D Systems. Mouse monoclonal anti-FN (MAB88904) and anti-αvβ3-integrin (MAB1976), goat polyclonal anti-α5β1-integrin (AB1950), rabbit polyclonal anti-α5-integrin (AB1928), rabbit polyclonal anti-α2-integrin (AB1936), and anti-α3-integrin (AB1920) were from Chemicon. Mouse monoclonal anti-human-vinculin (V9131) and rabbit polyclonal anti-Rab5 (R4654) were from Sigma-Aldrich. Rat monoclonal anti-HA (3F10) was from Roche. Rabbit polyclonal anti-GFP (A11122) and 4′,6-diamidino-2-phenylindole (DAPI) were from Molecular Probes. Goat polyclonal anti-GIPC1 (ab5951) and rabbit polyclonal anti-Myo6 (ab11096) were from Abcam. Streptavidin–horseradish peroxidase was from Amersham. Mouse monoclonal anti-active-α5-integrin, SNAKA51, was previously described [45].
Human plasma FN was from Tebu-bio. Human plasma vitronectin, Engelbreth-Holm-Swarm murine sarcoma laminin, and calf skin collagen type I were from Sigma-Aldrich. Recombinant human VEGF-A165 was from Invitrogen. Recombinant human SEMA3A and mouse Sema3F were from R&D Systems. Sulfo-NHS-SS-Biotin was from Pierce.
Hemagglutinin-tagged mNrp1 deletion constructs were generated by standard PCR protocols according to the Taq polymerase manufacturer's instructions (Fynnzymes) and using an HA-tagged version of full-length mNrp1 kindly donated by A. Püschel (Westfälische Wilhelms-Universität, Münster, Germany) as template. Cytoplasmic domains and the last three amino acids SEA were deleted using the following oligonucleotide primers: (i) 5′-cgccatggagagggggctgccgttg-3′ (Fw); (ii) 5′-ccaacaggcacagtacag-3′ (Re1) to amplify the mNrp1 deleted of the cytoplasmic domain; (iii) 5′-gtaattactctgtgggttc-3′ (Re2) to amplify the mNrp1 deleted of the three amino acids SEA. The corresponding PCR product was first thymidine–adenine (TA)-cloned into pCR2.1TOPO (Invitrogen) and subsequently subcloned in PINCO retrovirus or pAcGFP-N1 Vector (BD Bioscience) whose GFP coding sequence was previously substituted with the cDNA of mRFP, a kind gift of R. Tsien (University of California, San Diego, CA). α5-GFP and Rab5-GFP constructs were kindly provided, respectively, by A.F. Horwitz (University of Virginia, Charlottesville, VA) and M. Zerial (Max Plank Institute of Molecular Cell Biology and Genetics, Dresden, Germany).The α5-PA-GFP construct was previously described [51].
The day before oligofection, ECs were seeded in six-well dishes at a concentration of 10 × 104 cells/well. Oligofection of siRNA duplexes was performed according to the manufacturer's protocols. Briefly, human ECs were transfected twice (at 0 and 24 h) with 200 pmol of siCONTROL nontargeting siRNA (as control), siGENOME SMART pools (in the case of hGIPC1 and hMyo6), or a mix of three (in the case of hNrp1) siRNA oligonucletides (Dharmacon). After 24 h (in the case of hNrp1) or 96 h (in the case of hGIPC1 or hMyo6) had passed since the second oligofection, ECs were lysed or tested in functional assays. In the case of hNrp1, the single oligonucleotide sequences were: (1) 5′-AAUCAGAGUUUCCAACAUA-3′; (2) 5′-GAAGGAAGGGCGUGUCUUG-3′; (3) 5′-GUGGAUGACAUUAGUAUUA-3′.
Six-thousand ECs were resuspended in 0.1 ml of EBM-2 (Clonetics) with or without appropriate stimuli (50 ng/ml VEGF-A, 200 ng/ml SEMA3A, and 400 ng/ml SEMA3F) and plated on 96-well microtiter plates (Costar) that were previously coated with ECM proteins at different concentrations and then saturated with 3% bovine serum albumin. After 15 min of incubation at 37 °C, cells were fixed in 8% glutaraldehyde and then stained with 0.1% crystal violet in 20% methanol. Cells were photographed with a QICAM Fast 1394 digital color camera (QImaging) and counted by means of Image-ProPlus 6.2 software (Media Cybernetics). In adhesion assays with NIH 3T3 fibroblasts or CHO cells, Dulbecco's modified Eagle's medium was used.
Endothelial cells were lysed in buffer containing 25 mM Tris-HCl, pH 7.6, 100 mM NaCl, 0.15% Tween-20, 5% glycerol, 0.5 mM ethylene glycol tetraacetic acid (EGTA), and protease inhibitors (50 mg ml−1 pepstatin; 50 mg ml−1 leupeptin; 10 mg ml−1 aprotinin; 2 mM phenylmethanesulfonylfluoride (PMSF); 2 mM MgCl2). Cells were lysed in buffer, incubated for 20 min on wet ice, and then centrifuged at 15,000g, 20 min, at 4 °C. The total protein amount was determined using the bicinchoninic acid (BCA) protein assay reagent (Pierce). Equivalent amounts (1,200 μg) of protein were immunoprecipitated for 1 h with the antibody of interest, and immune complexes were recovered on Protein G-Sepharose (GE Healthcare). Immunoprecipitates were washed four times with lysis buffer, twice with the same buffer without Tween-20, and then separated by SDS-PAGE. Proteins were then transferred to a Hybond-C extra nitrocellulose membrane (Amersham), probed with antibodies of interest, and detected by an enhanced chemiluminescence technique (PerkinElmer).
Integrin traffic assays were performed as previously described by Roberts et al. [50] with minor modifications. Cells were transferred to ice, washed twice in cold phosphate-buffered saline (PBS), and surface-labeled at 4 °C with 0.2 mg/ml sulfo-NHS-SS-biotin (Pierce) in PBS for 30 min. Labeled cells were washed in cold PBS and transferred to prewarmed EGM-2 at 37 °C. At the indicated times, the medium was aspirated, and dishes were rapidly transferred to ice and washed twice with ice-cold PBS. Biotin was removed from proteins remaining at the cell surface by incubation with a solution containing 20 mM sodium 2-mercaptoethanesulfonate (MesNa) in 50 mM Tris-HCl (pH 8.6), 100 mM NaCl for 1 h at 4 °C. MesNa was quenched by the addition of 20 mM iodoacetamide (IAA) for 10 min, and after other two further washes in PBS, the cells were lysed in 25 mM Tris-HCl, pH 7.4, 100 mM NaCl, 2 mM MgCl2, 1 mM Na3VO4, 0.5 mM EGTA, 1% Triton X-100, 5% glycerol, protease mix (Sigma), and 1 mM PMSF. Lysates were cleared by centrifugation at 12,000g for 20 min. Supernatants were corrected to equivalent protein concentrations by BCA assay, and integrins were isolated by immunoprecipitation and analyzed by SDS-PAGE.
Cells were placed in RNAlater solution (Ambion), kept at 4 °C for 24 h, and frozen at −80 °C. After the cells were thawed on ice, total RNA was extracted following the manufacturer's recommended protocol (SV Total RNA Isolation System, Promega). The quality and integrity of the total RNA were quantified by means of the RNA 6000 Nano Assay kit in an Agilent 2100 bioanalyzer (Agilent Technologies). cDNAs were generated from 1 μg of total RNA using the High Capacity cDNA Reverse Transcription Kit (Applied Biosystems).
mRNA expression of FN and endogenous control genes, i.e., 18S rRNA, glyceraldehyde 3-phosphate dehydrogenase (GAPDH), and TATA binding protein (TBP), was measured in the samples by real-time RT-PCR using TaqMan Gene Expression Assays run on an ABI PRISM 7900HT Fast Real-Time PCR System (Applied Biosystems). The following assays were used: Hs00365058_m1 (FN), Hs99999901_s1 (18S rRNA), Hs99999905_m1 (GAPDH), and Hs00427620_m1 (TBP). Three replicates were run for each gene for each sample in a 384-well format plate (cDNA concentration 20 ng/well) according the manufacturer's protocol. Between the three measured endogenous control genes, we chose TBP for normalization, identified by geNorm [74]. The experimental threshold (Ct) was calculated using the algorithm provided by the SDS 1.9.1 software (Applied Biosystems). Ct values were converted into relative quantities using the method described here [75]. The amplification efficiency of each gene was calculated using a dilution curve and the slope calculation method [75].
Cells were plated on glass coverslips coated with 1 μg/ml FN (TebuBio) and allowed to adhere for 3 h. In addition, ECs cotransfected with mNrp1-mRFP and Rab5-GFP were then washed in PBS, incubated with 10 μg/ml SNAKA51 Ab in EBM-2 for 30 min at 4 °C, washed 3 times in PBS, transferred to prewarmed EGM-2, and allowed to recover at 37 °C for 2 min to induce endocytosis. Cells were washed in PBS, fixed in 4% paraformaldehyde, permeabilized in 0.01% saponin for 10 min on ice, and incubated or not with the Alexa-Fluor-405-conjugated secondary antibody (Molecular Probes) for 1 h at room temperature. Cells were analyzed by using a Leica TCS SP2 AOBS confocal laser-scanning microscope (Leica Microsystems). Immunofluorescence analysis was performed as previously described [16].
Small interfering RNA silencing was performed, and after the second oligofection, cells were seeded onto glass coverslips in six-well dishes at a concentration of 20 × 104 cells/well and left to adhere for 3 h in EGM-2 medium (Clonetics) containing FN-depleted serum. Cells were then washed with PBS and fixed with 3.7% paraformaldehyde for 20 min at room temperature. Next, cells were permeabilized in PBS containing 0.1% Triton X-100 on wet ice for 2 min and incubated with anti-FN Ab for 1 h at room temperature. After three washes, cells were incubated with anti-mouse Alexa Fluor 555 for 45 min at room temperature and subsequently with DAPI. Cells were finally examined using a Leica TCS SP2 AOBS confocal laser-scanning microscope (Leica Microsystems).
Fluorescence resonance energy transfer was detected using a Lambert Instruments fluorescence attachment (LIFA) on a Nikon Eclipse TE 2000-U microscope, with the same changes to the condenser as described above and a filter block consisting of a Z473/10 excitation filter, a Z 488 RDC dichroic mirror, and a HQ 525/50M emission filter. The light source was a modulated 473-nm laser diode, which allows, in combination with the modulated intensifier from the LIFA system, measurement of fluorescence lifetimes using frequency domain. The laser was brought into TIRF mode before acquiring the images for the lifetime analysis.
Donor (D) lifetime, τ, was analyzed either in the presence or in the absence of the acceptor (A), in adhesion sites, characterized by high donor concentrations, using the FLIM software (version 1.2.1.1.130; Lambert Instruments, The Netherlands). Fluorescence resonance energy transfer efficiency (E) was calculated as E = 1 – (τDA/τD). Lifetime τ was evaluated in four different areas (12 × 12 pixels) of seven α5-GFP and seven α5-GFP/mNrp1-Cherry transfected NIH 3T3 cells. For statistical evaluation, results were analyzed with Student's t test.
Total internal reflection fluorescence experiments have been performed on a Nikon Eclipse TE 2000-U microscope equipped with 60× and 100× 1.45 NA Nikon TIRF oil immersion objectives. The Nikon Epi-fluorescence condenser was replaced with a custom condenser in which laser light was introduced into the illumination pathway directly from the optical fiber output oriented parallel to the optical axis of the microscope. The light source for evanescent wave illumination was either a 473-nm diode, a 405-nm diode, or a 561-nm laser (Omicron), with each laser line coupled into the condenser separately to allow individual TIRF angle adjustments. Each laser was controlled separately by a DAC 2000 card or a uniblitz shutter operated by MetaMorph (Molecular Devices). A filter block consisting of an E480SPX excitation filter, a FF 495 dichroic mirror, and an ET 525/50M emission filter was used for activation of α5-PA-GFP with the 405-nm laser. After activation the filter was manually changed to a green/red dual filter block (ET-GFP/mCherry from AHF Analysentechnik, Germany) to allow simultaneous time-lapse acquisition of activated α5-PA-GFP and mNrp1-Cherry using 473- and 561-nm excitation. A Multi-Spec dual emission splitter (Optical Insights, NM) with a 595-nm dichroic and two bandpass filters (510–565 nm for green and 605–655 nm for red) was used to separate both emissions. All cell imaging was performed with a Cascade 512F EMCCD camera (Photometrics UK).
Localized activation of α5-PA-GFP in mNrp1-Cherry-positive vesicles was done on a FV 1000 Olympus confocal microscope, using two-channel imaging and a separate SIM scanner for 405-nm activation [51].
|
10.1371/journal.pntd.0004901 | Zika Virus, a New Threat for Europe? | Since its emergence in 2007 in Micronesia and Polynesia, the arthropod-borne flavivirus Zika virus (ZIKV) has spread in the Americas and the Caribbean, following first detection in Brazil in May 2015. The risk of ZIKV emergence in Europe increases as imported cases are repeatedly reported. Together with chikungunya virus (CHIKV) and dengue virus (DENV), ZIKV is transmitted by Aedes mosquitoes. Any countries where these mosquitoes are present could be potential sites for future ZIKV outbreak. We assessed the vector competence of European Aedes mosquitoes (Aedes aegypti and Aedes albopictus) for the currently circulating Asian genotype of ZIKV.
Two populations of Ae. aegypti from the island of Madeira (Funchal and Paul do Mar) and two populations of Ae. albopictus from France (Nice and Bar-sur-Loup) were challenged with an Asian genotype of ZIKV isolated from a patient in April 2014 in New Caledonia. Fully engorged mosquitoes were then maintained in insectary conditions (28°±1°C, 16h:8h light:dark cycle and 80% humidity). 16–24 mosquitoes from each population were examined at 3, 6, 9 and 14 days post-infection to estimate the infection rate, disseminated infection rate and transmission efficiency. Based on these experimental infections, we demonstrated that Ae. albopictus from France were not very susceptible to ZIKV.
In combination with the restricted distribution of European Ae. albopictus, our results on vector competence corroborate the low risk for ZIKV to expand into most parts of Europe with the possible exception of the warmest regions bordering the Mediterranean coastline.
| In May 2015, local transmission of Zika virus (ZIKV) was reported in Brazil and since then, more than 1.5 million human cases have been reported in Latin America and the Caribbean. This arbovirus, primarily found in Africa and Asia, is mainly transmitted by Aedes mosquitoes, Aedes aegypti and Aedes albopictus. Viremic travelers returning from America to European countries where Ae. albopictus is established could become the source for local transmission of ZIKV. In order to estimate the risk of seeding ZIKV into local mosquito populations, the susceptibility of European Ae. aegypti and Ae. albopictus to ZIKV was measured using experimental infections. We demonstrated that Ae. albopictus and Ae. aegypti from Europe were not very susceptible to ZIKV. The threat for a Zika outbreak in Europe should be limited.
| Zika virus (ZIKV) (genus Flavivirus, family Flaviviridae) is an emerging arthropod-borne virus transmitted to humans by Aedes mosquitoes. ZIKV infection in humans was first observed in Africa in 1952 [1], and can cause a broad range of clinical symptoms presenting as a “dengue-like” syndrome: headache, rash, fever, and arthralgia. In 2007, an outbreak of ZIKV on Yap Island resulted in 73% of the total population becoming infected [2]. Following this, ZIKV continued to spread rapidly with outbreaks in French Polynesia in October 2013 [3], New Caledonia in 2015 [4], and subsequently, Brazil in May 2015 [5, 6]. During this expansion period, the primary transmission vector is considered to have been Aedes aegypti, although Aedes albopictus could potentially serve as a secondary transmission vector [7] as ZIKV detection has been reported in field-collected Ae. albopictus in Central Africa [8]. As Musso et al. [9] observed, the pattern of ZIKV emergence from Africa, throughout Asia, to its subsequent arrival in South America and the Caribbean closely resembles the emergence of Chikungunya virus (CHIKV). In Europe, returning ZIKV-viremic travelers may become a source of local transmission in the presence of Aedes mosquitoes, Ae. albopictus in Continental Europe and Ae. aegypti in the Portuguese island of Madeira. Ae. albopictus originated from Asia was recorded for the first time in Europe in Albania in 1979 [10], then in Italy in 1990 [11]. It is now present in all European countries around the Mediterranean Sea [12]. This mosquito was implicated as a vector of CHIKV and DENV in Europe [13]. On the other hand, Ae. aegypti disappeared after the 1950s with the improvement of hygiene and anti-malaria vector control. This mosquito reinvaded European territory, Madeira island, in 2005 [14], and around the Black Sea in southern Russia, Abkhazia, and Georgia in 2004 [12]. The species was responsible for outbreaks of yellow fever in Italy in 1804 [15] and dengue in Greece in 1927–1928 [16]. To assess the possible risk of ZIKV transmission in Europe, we compared the relative vector competence of European Ae. aegypti and Ae. albopictus populations to the Asian genotype of ZIKV.
The Institut Pasteur animal facility has received accreditation from the French Ministry of Agriculture to perform experiments on live animals in compliance with the French and European regulations on care and protection of laboratory animals. This study was approved by the Institutional Animal Care and Use Committee (IACUC) at the Institut Pasteur. No specific permits were required for the described field studies in locations that are not protected in any way and did not involve endangered or protected species.
Four populations of mosquitoes (two populations of Ae. aegypti: Funchal (32°40’N, 16°55’W) and Paul do Mar (32°45’N, 17°13’W), collected on island of Madeira and two populations of Ae. albopictus: Nice (43°42’N, 7°15’E) and Bar-sur-Loup (43°42’N, 6°59’E) in France) were collected using ovitraps. Eggs were immersed in dechlorinated tap water for hatching. Larvae were distributed in pans of 150–200 individuals and supplied with 1 yeast tablet dissolved in 1L of water every 48 hours. All immature stages were maintained at 28°C ± 1°C. After emergence, adults were given free access to a 10% sucrose solution and maintained at 28°C ± 1°C with 70% relative humidity and a 16:8 light/dark cycle. The F1 generation of Ae. aegypti from Madeira and F7-8 generation of Ae. albopictus from France were used for experimental infections.
The ZIKV strain (NC-2014-5132) originally isolated from a patient in April 2014 in New Caledonia was used to infect mosquitoes. The viral stock used was subcultured five times on Vero cells prior to the infectious blood-meal. The NC-2014-5132 strain is phylogenetically closely related to the ZIKV strains circulating in the South Pacific region, Brazil [5] and French Guiana [17].
Infectious blood-meals were provided using a titer of 107 TCID50/mL. Seven-day old mosquitoes were fed on blood-meals containing two parts washed rabbit erythrocytes to one part viral suspension supplemented with ATP at a final concentration of 5 mM. Rabbit arterial blood was collected and erythrocytes were washed five times with Phosphate buffered saline (PBS) 24 h before the infectious blood-meal. Engorged females were transferred to cardboard containers with free access to 10% sucrose solution and maintained at 28°C and 70% relative humidity with a 16:8 light/dark cycle. 16–24 female mosquitoes from each population were analyzed at 3, 6, 9, and 14 days post-infection (dpi) to estimate the infection rate, disseminated infection rate and transmission efficiency. Briefly, legs and wings were removed from each mosquito followed by insertion of the proboscis into a 20 μL tip containing 5 μL FBS for 20 minutes. The saliva-containing FBS was expelled into 45 μμL serum free L-15 media (Gibco), and stored at -80°C. Following salivation, mosquitoes were decapitated and head and body (thorax and abdomen) were homogenized separately in 300 μL L-15 media supplemented with 3% FBS using a Precellys homogenizer (Bertin Technologies) then stored at -80°C. Infection rate was measured as the percentage of mosquitoes with infected bodies among the total number of analyzed mosquitoes. Disseminated infection rate was estimated as the percentage of mosquitoes with infected heads (i.e., the virus had successfully crossed the midgut barrier to reach the mosquito hemocoel) among the total number of mosquitoes with infected bodies. Transmission efficiency was calculated as the overall proportion of females with infectious saliva among the total number of tested mosquitoes. Samples were titrated by plaque assay in Vero cells.
For head/body homogenates and saliva samples, Vero E6 cell monolayers were inoculated with serial 10-fold dilutions of virus-containing samples and incubated for 1 hour at 37°C followed by an overlay consisting of DMEM 2X, 2% FBS, antibiotics and 1% agarose. At 7 dpi, overlay was removed and cells were fixed with crystal violet (0.2% Crystal Violet, 10% Formaldehyde, 20% ethanol) and positive/negative screening was performed for cytopathic effect (body and head homogenates) or plaques were enumerated (head and saliva samples). Vero E6 cells (ATCC CRL-1586) were maintained in DMEM (Gibco) supplemented with 10% fetal bovine serum (Eurobio), Penicillin and Streptomycin, and 0.29 mg/mL l-glutamine.
All statistical tests were conducted with the STATA software (StataCorp LP, Texas, USA) using 1-sided Fisher’s exact test and P-values>0·05 were considered non-significant.
To test whether Ae. aegypti from a European territory were able to transmit ZIKV, we analyzed the vector competence of two Ae. aegypti populations collected on the island of Madeira based on three parameters: viral infection of the mosquito midgut, viral dissemination to secondary organs, and transmission potential, analyzed at 3, 6, 9, and 14 dpi. Only mosquitoes presenting an infection (i.e. infected midgut) were analyzed for viral dissemination. The two populations presented similar infection (P = 0.50 (3 dpi), 0.17 (6), 0.36 (9), 0.50 (14); Fig 1) and disseminated infection (P = 0.59 (3 dpi), 0.63 (6), 0.43 (9), 0.06 (14); Fig 1) with the highest rates measured at 9 dpi and 9–14 dpi, respectively. When examining transmission efficiency, only Ae. aegypti Funchal were able to transmit ZIKV at 9 (1 individual among 20 tested) and 14 dpi (1 among 20) (Fig 1). When considering the number of viral particles in heads, no significant difference was detected between Ae. aegypti Funchal and Ae. aegypti Paul do Mar (P = 1 (3 dpi), 0.22 (6), 0.60 (9), 0.38 (14); Fig 2). When examining viral loads in saliva, only Ae. aegypti Funchal exhibited 1550 particles at 9 dpi and 50 at 14 dpi (Fig 2).
To determine if Ae. albopictus present in continental Europe were able to sustain local transmission of ZIKV as previously observed with CHIKV and DENV, we evaluated the vector competence of two Ae. albopictus populations collected in Nice and Bar-sur-Loup in the South of France. When compared with Ae. aegypti, the two Ae. albopictus populations showed similar infection rates at 3 dpi (P = 0.08) and 6 dpi (P = 0.11) and disseminated infection rates at 9 dpi (P = 0.62) and 14 dpi (P = 0.10) (Fig 1). Only one individual among 24 Ae. albopictus Bar-sur-Loup tested at 14 dpi was able to transmit ZIKV (Fig 1). When analyzing the number of viral particles in heads, only few mosquitoes were infected (Fig 2). When examining saliva, one Ae. albopictus Bar-sur-Loup exhibited 2 viral particles at 14 dpi (Fig 2).
In summary, ZIKV dissemination through Ae. aegypti was noticeably superior and the virus in saliva was detected earlier in Ae. aegypti than in Ae. albopictus. However both mosquito species showed similar transmission efficiencies at 9–14 dpi.
ZIKV could be transmitted, spread and maintained in Europe either via (i) Madeira where the main vector Ae. aegypti has been established since 2005 or (ii) Continental Europe where Ae. albopictus is known to have been present since 1979 [12]. We demonstrated that ZIKV was amplified and expectorated efficiently in saliva by European Ae. aegypti from Madeira. This contrasts with the lower vector competence for ZIKV of French Ae. albopictus. Taking these observations and the overall average lower temperatures of most regions of Europe into account, the risk of major outbreaks of Zika fever in most areas of Europe, at least for the immediate future, appears to be relatively low.
Our results highlight the potential risk for ZIKV transmission on Madeira where two main factors are present: the presence of the main vector, Ae. aegypti introduced in 2005 [18] and imported cases from Brazil with which Madeira, an autonomous region of Portugal, maintains active exchanges of goods and people sharing the same language. Thus Madeira Island could be considered as a stepping stone for an introduction of ZIKV into Europe.
Autochthonous cases of CHIKV and DENV have been reported in Europe since 2007: CHIKV in Italy in 2007, South France in 2010, 2014, and DENV in South France in 2010, 2013, 2015, and Croatia in 2010 [19]. The invasive species Ae. albopictus first detected in Europe in 1979 [10] has played a central role in this transmission [19]. Thus, there might be a risk of a similar establishment of ZIKV in Europe upon the return of viremic travelers [20, 21]. We showed that Ae. albopictus from South France were less competent for ZIKV infection requiring 14 days to be expectorated in the mosquito saliva after infection. Therefore, we can suggest that the Asian tiger mosquito from Southern France and more widely, Europe, are less suitable to sustain local transmission of ZIKV compared to CHIKV and perhaps, DENV. Ae. albopictus Nice were not able to expectorate ZIKV in saliva at day 14 post-infection like Ae. albopictus Bar-sur-Loup suggesting two populations genetically differentiated.
Considering the extensive airline travel between Latin America and Europe, the risk for local transmission of ZIKV in the European area where the mosquito Ae. albopictus is widely distributed, is assumed to be minimal based on our studies of vector competence. Nevertheless, reinforcement of surveillance and control of mosquitoes should remain a strong priority in Europe since Aedes mosquitoes also transmit DENV and CHIKV and virus adaptation to new vectors cannot be excluded, as previously observed with CHIKV in La Reunion [22, 23].
|
10.1371/journal.pgen.1004918 | Combining Natural Sequence Variation with High Throughput Mutational Data to Reveal Protein Interaction Sites | Many protein interactions are conserved among organisms despite changes in the amino acid sequences that comprise their contact sites, a property that has been used to infer the location of these sites from protein homology. In an inter-species complementation experiment, a sequence present in a homologue is substituted into a protein and tested for its ability to support function. Therefore, substitutions that inhibit function can identify interaction sites that changed over evolution. However, most of the sequence differences within a protein family remain unexplored because of the small-scale nature of these complementation approaches. Here we use existing high throughput mutational data on the in vivo function of the RRM2 domain of the Saccharomyces cerevisiae poly(A)-binding protein, Pab1, to analyze its sites of interaction. Of 197 single amino acid differences in 52 Pab1 homologues, 17 reduce the function of Pab1 when substituted into the yeast protein. The majority of these deleterious mutations interfere with the binding of the RRM2 domain to eIF4G1 and eIF4G2, isoforms of a translation initiation factor. A large-scale mutational analysis of the RRM2 domain in a two-hybrid assay for eIF4G1 binding supports these findings and identifies peripheral residues that make a smaller contribution to eIF4G1 binding. Three single amino acid substitutions in yeast Pab1 corresponding to residues from the human orthologue are deleterious and eliminate binding to the yeast eIF4G isoforms. We create a triple mutant that carries these substitutions and other humanizing substitutions that collectively support a switch in binding specificity of RRM2 from the yeast eIF4G1 to its human orthologue. Finally, we map other deleterious substitutions in Pab1 to inter-domain (RRM2–RRM1) or protein-RNA (RRM2–poly(A)) interaction sites. Thus, the combined approach of large-scale mutational data and evolutionary conservation can be used to characterize interaction sites at single amino acid resolution.
| The interactions of proteins with each other are essential for almost all biological processes. Many of the sites of protein contact have evolved to maintain these interactions, but use different sets of amino acid residues. As a result, the residues at a contact site in a protein from one species might not allow a protein interaction when they are tested in a second species. This property underlies the idea of inter-species complementation assays, which test the effect of replacing protein segments from one species by their equivalents from another species. However, this approach has been highly limited in the number of changes that could be analyzed in a single study. Here, we present a novel approach that combines a high-throughput analysis of mutations in a single protein with the set of natural sequences corresponding to evolutionarily divergent variants of this protein. This integration step allows us to map at high resolution both sites of inter-protein interaction as well as intra-protein interaction. Our approach can be used with proteins that have limited functional and structural data, and it can be applied to improve the performance of computational tools that use sequence homology to predict function.
| Protein activity, folding and stability are regulated by the interactions of proteins with other macromolecules. Thus, the identification of sites on a protein where these interactions occur is a critical but difficult undertaking. In some cases, structural analyses provide these sites at high resolution. In other cases, combinations of biochemical, biophysical and genetic methods with mutagenesis strategies have delineated specific residues that contribute to physical interactions. However, the vast number of protein-protein interactions and the low throughput and robustness of approaches to identify interaction sites have led to the limited and often imprecise characterization of only a tiny fraction of the contact sites. Sequence-based computational methods offer an alternative and cost-effective approach that can predict interacting positions by making use of homologous sequences. For example, the evolutionary trace method [1] assumes that the locations of interaction sites are conserved over evolution, and that sequence variation within these sites occurs in response to changes in evolutionary constraints to allow the protein to maintain its activity. Other computational methods are based on the idea that physical interaction between two proteins leads to linked evolutionary changes between their contact sites [2,3,4]. Thus, the correlated changes between pairs of positions in multiple sequence alignments of two interacting proteins can identify binding sites [2]. However, despite improvements in the construction of multiple sequence alignments and phylogenetic trees, and the huge increase in the number of homologous sequences, the accuracy of these methods remains challenged by fundamental problems [5,6]. For example, transient interactions often yield poor evolutionary signals due to increased rates of substitutions at contact sites [7]. In consequence, these contact sites resemble other, less critical residues in the protein that also tolerate multiple substitutions.
We begin with the idea that substitutions tolerated in nature usually cause only minor changes in structure [8]. Thus, if a position in a protein is substituted with an amino acid that is found at that position in homologous proteins, the resulting protein is likely still to function in its native organism. However, when such a substitution has a detrimental effect, it may have affected a functional site that has changed over evolution [9]. For a protein contact site, such a detrimental effect is likely due to the lack of other compensating substitutions also present in the homologous protein that have co-evolved to support its binding to a partner protein. Alternatively, compensatory substitutions might be present in the homologue of the protein partner. Complementation assays using a protein with such natural substitutions have been used to characterize binding site residues [10,11,12,13]. However, the utility of this approach has been limited by the lack of large-scale assays that can test a protein’s activity when it carries all the possible substitutions that occur in homologous sequences.
Recently, a method known as deep mutational scanning was developed to assess the functional consequences of up to hundreds of thousands of variants of a protein in a single experiment [14,15]. This method combines next generation sequencing with a functional selection, using the change in frequency for each variant over the course of the selection as a proxy for the variant’s activity. We previously applied this method to study the in vivo function of an RNA recognition motif (RRM) of the Saccharomyces cerevisiae poly(A)-binding protein, Pab1 [16].
The eukaryotic poly(A)-binding protein regulates mRNA translation and decay [17,18,19] by binding to the poly(A) tail of an mRNA via its four RRMs [20,21]. This binding leads to an interaction between RRM2 and the translation initiation factor eIF4G, a constituent of the mRNA cap-binding complex, eIF4F [22], which is assumed to enhance the rate of translation by supporting the establishment of a closed loop structure of the mRNA [23,24,25]. Yeast encode two eIF4G paralogues, eIF4G1 and eIF4G2 [26], which both interact with Pab1 [12]. Complementation assays by Otero et al. [12] with yeast Pab1 containing residues from the human orthologue mapped the binding site for the two eIF4G isoforms to five amino acids on the surface of Pab1 RRM2 [12]. However, this study addressed only the 25 Pab1 residues in the RRM2 domain that vary between human and yeast, and thus the contribution of the other 50 RRM2 residues and the precise Pab1 contact sites for the two isoforms of eIF4G were not determined.
We analyzed deep mutational scanning data for the RRM2 domain of yeast Pab1 to examine the functional consequences in yeast of single amino acid substitutions that differentiate the yeast domain from its homologues. This large-scale inter-species complementation data allowed us to characterize the eIF4G1 and eIF4G2 binding sites on the RRM2 surface at single amino acid resolution and to identify residues associated with the RRM2–poly(A) and RRM2–RRM1 interactions. By combining epistasis data for double mutants with natural variation data, we identify a humanizing substitution that promotes a change in binding specificity of the yeast Pab1 RRM2 from the yeast to the human eIF4G1 protein. Taken together, in vivo deep mutational scanning data integrated with evolutionary variation can be used to characterize interaction sites with high resolution and to predict epistatically interacting residues in natural homologues of a protein.
We recently scored the in vivo function of more than 100,000 variants of the RRM2 domain of the yeast Pab1 [16]. The assay was based on turning off the expression of a wild-type copy of the PAB1 coding sequence and assaying growth of yeast dependent on mutated versions of a C-terminally truncated form (Pab1-343) that includes the first three RRM domains and a small portion of RRM4. For each variant, we assigned an enrichment score that represents the ratio between the fractions of its sequence read counts after and before selection, normalized to the wild-type enrichment score. Hence, enrichment scores serve as indirect readouts for the effects of mutations on growth rate. We obtained scores for 1246 single amino acid substitutions, including 1190 missense mutations and 56 nonsense mutations (∼83% of all possible single amino acid substitutions in the 75 amino-acid long sequence that covers most of this domain) [16].
We realized that the scores of variants with amino acid substitutions present in Pab1 homologues might provide insight into functional sites that diverged in sequence throughout the evolution of this protein. To this end, we collected sequences of 52 poly(A)-binding proteins that represent all Pab1 homologues in the UniProtKB/Swiss-Prot database. The 52 homologous sequences include both orthologues and paralogues of the poly(A)-binding protein and are derived from eukaryotic species including fungi, plants and mammals. All 52 proteins carry four tandem RRM domains, allowing us to align the Pab1 RRM2 against all its corresponding domains. The multiple sequence alignment showed conservation between the homologous RRM2 sequences and the yeast Pab1 RRM2 ranging from 88% identity for Candida glabrata to 55% identity for Encephalitozoon cuniculi. The alignment revealed 210 single amino acid differences (“natural substitutions”) with respect to the yeast Pab1 RRM2 sequence. The in vivo deep mutational scanning data from our previous study [16] provide functional scores for 197 of these 210 substitutions (Fig. 1A).
Most of these natural substitutions resulted in small effects (Fig. 1B), with a median score of −0.07 relative to the wild-type (the score, in log2 scale, is comparable to ∼5% reduction from the wild-type score) and narrow upper and lower quartiles. On the contrary, substitutions that do not appear in Pab1 homologues (“non-natural substitutions”) displayed a much larger range and more negative effects, with a median score of −0.53 (comparable to ∼30% reduction from the wild-type score). That most natural changes result in small effects suggests that the functional constraints on the poly(A)-binding protein remained largely constant throughout its evolution. However, a few natural substitutions showed low enrichment scores that correspond to poor Pab1 performance in S. cerevisiae. In particular, enrichment scores of 45 natural substitutions ranged between −0.15 and −0.5 (a range that we term mildly deleterious, comparable to ∼10–30% reduction from the wild-type score) and enrichment scores of 17 other natural substitutions were lower than −0.5 (a range that we term strongly deleterious, comparable to more than 30% reduction from the wild-type score) (Fig. 1A). We further compared the score distribution of natural variants to the score distribution of synonymous variants which serve as a proxy for non-deleterious variants, as previously described [16]. This comparison allowed us to assess the contamination of the mildly and the strongly deleterious groups by variants that carry non-deleterious mutations (S1 Fig.). Based on this analysis, we estimated that the natural substitutions in the mildly deleterious group are contaminated by 35% non-deleterious variants, while the natural substitutions in the strongly deleterious group are contaminated by only 8% non-deleterious variants. Given these results, we further analyzed only mutations classified as strongly deleterious.
The solvent accessibility of residues in the structure of a human orthologue of Pab1 reveals that both natural non-deleterious and natural strongly deleterious substitutions, compared to all other non-natural substitutions, occur preferentially at solvent-exposed areas (Fig. 1C). However, an evaluation of the conservation of each substitution using its Blosum62 score revealed a significant difference between the natural non-deleterious and the natural strongly deleterious groups (Fig. 1D). Though both groups showed high conservation compared to non-natural substitutions, the natural strongly deleterious substitutions displayed a lower conservation score (median of −1) than the natural non-deleterious substitutions (median of 0). The differences in Blosum62 score distributions of the two groups suggests that natural deleterious effects in general are due to substitutions to amino acids that display physicochemical properties that are neither as disruptive as non-natural substitutions nor as subtle as natural non-deleterious ones. Nonetheless, a few natural-deleterious substitutions resulted from replacements by highly similar amino acids (e.g. L186M and L153V), indicating that sometimes the exact identity of the Pab1 residue is of crucial importance.
Of the 25 single amino acid substitutions that differentiate the yeast Pab1 RRM2 domain from its human orthologue (Fig. 2A), 24 have enrichment scores in our dataset. Three of these mutations (E181R, A185K and L186M) are strongly deleterious (Fig. 2B). These three substitutions occur in two short stretches of the yeast Pab1, 180-KE-181 and 184-DAL-186, that when replaced with the corresponding human stretches to create 180-ER-181 and 184-EKM-186 interfere with in vitro binding to ∼100 amino acid fragments of yeast eIF4G1 and eIF4G2 [12,22]. The large-scale mutational data indicate that the other two mutations in these short stretches, K180E and D184E, cause no measurable effect on function (Fig. 2B).
To test whether the in vivo effects on Pab1 performance correlate with eIF4G1 and eIF4G2 binding, we established a two-hybrid assay between yeast Pab1 and the N-terminal 341 amino acids of yeast eIF4G1 or eIF4G2, which contain the binding sites for Pab1 [12,22]. The full-length Pab1 tested with the eIF4G1 or eIF4G2 fragment did not activate HIS3 reporter gene expression (Fig. 2C). However, as some protein-protein interactions can be detected by the yeast two-hybrid system only when parts of the proteins are removed [27], we tested various truncation products of Pab1 for eIF4G1 and eIF4G2 association. Indeed, RRM2 alone produced a positive interaction signal with both isoforms (Fig. 2C).
In agreement with Otero et al. [12], the replacement of residues 184–186 with those from human resulted in complete loss of binding to both eIF4G1 and eIF4G2 (Fig. 2D). When tested individually, A185K and L186M did not bind eIF4G1 or eIF4G2, while D184E showed wild-type binding. The replacement of residues 180–181 with those from human abolished eIF4G1 binding and reduced eIF4G2 binding. This residual binding to eIF4G2 may reflect the greater sensitivity of the two-hybrid assay compared to the in vitro assay [12]. When tested individually, E181R resulted in loss of eIF4G1 and eIF4G2 binding, while K180E had no effect (Fig. 2D). Since the E181R effect on eIF4G2 binding was more severe in the absence of the K180E substitution, K180E might suppress the negative effect of the E181R mutation on eIF4G2 binding by decreasing the local positive charge. Overall, the in vivo function of Pab1 carrying any of the five single amino acid substitutions correlates with the ability of Pab1 to support eIF4G1 and eIF4G2 binding.
We hypothesized that the deleterious effects of some of the other natural substitutions might be due to a loss of eIF4G1 and eIF4G2 binding. We therefore tested in the two-hybrid assay the 17 substitutions that cause a strongly deleterious effect, as well as A185K and D184W, which score similarly but had lower sequence read coverage in the original experiments [16]. Of these 19 mutations, 10 (occurring in 8 different residues) impaired the ability of RRM2 to bind eIF4G1, with I137F, T145H, T145L, V148K, E181R, A185H, A185K and L186M showing the most severe effects (Fig. 3A, left). D138T and A141D resulted in modest effects on eIF4G1 binding (Fig. 3A, left). The same Pab1 variants assayed against eIF4G2 revealed similar effects (Fig. 3A, right), suggesting that eIF4G1 and eIF4G2 use the same set of Pab1 RRM2 residues for binding. However, eIF4G1 binding was more sensitive to A141D and T145L compared to eIF4G2. Based on the effects of the natural amino acid substitutions on binding, we set the boundaries of eIF4G recognition site to the upper surface of RRM2 (Fig. 3B), a region much wider than previously identified [12].
While combining natural variation with in vivo deep mutational scanning highlights the contribution to protein-protein interactions of residues that change over evolutionary time, it overlooks highly conserved residues and ignores the effects of substitutions to amino acids that do not appear in homologues. We therefore sought to study the effects of mutations on Pab1 RRM2–eIF4G1 association by an alternative approach. To this end, we performed a large-scale two-hybrid analysis. We expressed each of three libraries of RRM2 as a DNA-binding domain hybrid, with mutations covering Pab1 positions 131–150, 151–175 or 176–197, and tested for the binding of these hybrids to the yeast eIF4G1 expressed as an activation domain hybrid. Samples were collected before (input) and after (selected) two-hybrid selection, and the library segments were recovered and sequenced. For each variant, the change in its frequency from input to selected pool (i.e. its enrichment score) was determined as previously described [16]. We were able to extract enrichment scores for 802 single amino acid substitutions across the three library segments, which comprise 60% of all possible substitutions (S1 Table). While mutations that disrupt RRM2 structure caused fortuitous activation of the yeast two-hybrid reporter gene, positions that were shown to be sensitive to natural substitutions when tested individually showed similar sensitivities to mutation in this large-scale assay, suggesting that the enrichment scores for mutations that specifically affect the contact site for eIF4G1 were valid (S2 Fig.). In particular, of the 44 mutations that reduced the enrichment score by more than 50% (log2 enrichment score < −1), 22 mutations occur at the eight positions that were found by our natural variation analysis to be involved in eIF4G1 binding (I137, D138, A141, T145, V148, E181, A185 and L186); eight mutations occur at the immediate sequence neighbors of these positions (D136, S147, F149 and D184); and 11 mutations occur at residues that show physical but not immediate sequence proximity to these contact site residues (G150, G188, M189, L190 and N192). Overall, in addition to identifying eIF4G1 contact site residues that were elucidated by the combined approach of the in vivo mutational data and the natural variation data, the large-scale two-hybrid results highlighted the contribution of residues at the periphery of this site (Fig. 4A). To understand why mutations at these positions were not discovered using our combined approach, we examined the level of natural variation at these sites. While F149 and G150 are fully conserved, the other residues show some degree of variation in Pab1 homologues. Though some of these natural changes interfered with eIF4G1 binding in the two-hybrid assay, none of them showed a strongly deleterious effect in vivo (Fig. 4B), suggesting that the central residues of the eIF4G1 binding site are more sensitive to natural variation substitutions in vivo than the peripheral ones.
To understand how incompatible Pab1 variants have evolved in different lineages, we constructed a maximum likelihood tree from the 52 Pab1 homologues. In agreement with theoretical expectations [28], we found that the number of substitutions in Pab1 that were strongly deleterious in S. cerevisiae increases with evolutionary distance (Fig. 5A). Specifically, while closely related fungi provide zero or one strongly deleterious substitution, the microsporidian Encephalitozoon cuniculi, which carries the most diverse PABP sequence, contributes six deleterious substitutions. The deep divergence of E. cuniculi PABP, likely due to rapid evolution of microsporidia after branching off the fungal lineage [29], provides a unique set of mutations (I137F, D138T and A141D) that interfered with eIF4G1 binding. However, unlike the metazoan substitutions that interfered with this binding, the E. cuniculi substitutions localize to helix α1 (Fig. 3B), which suggests two alternative paths of eIF4G-binding site evolution. In addition, the deleterious effects of substitutions T145L and T145H, from the non-yeast paralogues of the poly(A) binding protein (PABP5 and PABP4L), reveal the critical function of T145 in eIF4G binding. Taken together, these results highlight the need to analyze evolutionarily remote sequences in order to obtain a detailed map of functional sites in proteins.
The functional scores of the natural substitutions that occurred throughout evolution suggest ancestral states that were likely to promote the divergence of the eIF4G1-binding site. In particular, for position 185, we observe a stepwise decrease in charge in the S. cerevisiae lineage, from lysine through histidine and asparagine to alanine (Fig. 5B, middle). Both A185K and A185H were strongly deleterious in yeast, suggesting that the lack of positive charge in yeast was accompanied by other changes in eIF4G or in Pab1 orthologues that are no longer compatible with the ancestral state of this position. At positions 181 and 186, substitutions matching variation within the S. cerevisiae lineage were mildly deleterious or non-deleterious, while substitutions matching variation that occurred after the fungal–metazoan divergence were strongly deleterious. Therefore, changes in eIF4G or in Pab1 orthologues that compensate for the otherwise detrimental effects of these mutations are likely to be conserved along the metazoan branch of the tree.
We asked whether the yeast Pab1 and eIF4G protein sequences might enable us to infer the compensatory changes that allowed the establishment of the strongly deleterious substitutions E181R, A185K and L186M in the human orthologue of Pab1. For instance, a pair of mutations comprising one humanizing substitution in yeast Pab1 that interferes with yeast eIF4G1 association and a compensating, second humanizing mutation in the yeast eIF4G1 might restore binding. However, the identification of candidate humanizing substitutions in the yeast eIF4G1 that may form deleterious–compensatory clusters with humanizing mutations in Pab1 is challenging due to the extreme diversification of eIF4G1 and its contact site residues throughout evolution (Fig. 6A). Thus, we decided to explore the inter-protein interactions in Pab1 that underpin the binding of the RRM2 domain to either the yeast or human eIF4G1. While the human and yeast RRM2 domains interacted with their cognate eIF4G1 fragment, neither bound to its non-cognate eIF4G1 fragment (Fig. 6B), suggesting that eIF4G1 binding specificity is dependent on the 25 positions that differ between the yeast and the human RRM2 domains.
We tested a few humanizing mutations in Pab1 RRM2 for their ability to change the binding specificity towards human eIF4G1. Though there are many possible combinations of humanizing substitutions, we used the deep mutational scanning results to narrow down the list of candidate residues. We first evaluated the ability of Pab1 RRM2 fragments that carry each of the three humanizing substitutions (E181R, A185K and L186M) that abolished binding to the yeast eIF4G1 to bind the human eIF4G1 fragment. The E181R variant activated the two-hybrid reporter gene (Fig. 6B), indicating that despite other sequence differences, elements within the yeast Pab1 RRM2 domain support this change in binding specificity. Unlike E181R, A185K and L186M did not bind to human eIF4G1, suggesting that these two substitutions require other humanizing changes in Pab1 RRM2 to function. Combining A185K and L186M with E181R to form a triple mutant did not enable binding of yeast Pab1 to human eIF4G1 (Fig. 6B). Because this triple mutant carries all of the strongly deleterious substitutions that differ between the human and the yeast Pab1 RRM2 domain, this finding suggests that some of the remaining mildly deleterious or non-deleterious substitutions are necessary to compensate for the detrimental effects of A185K and L186M on eIF4G1 binding.
Because the deep mutational scanning of Pab1 RRM2 provided functional scores for multiple variants that change two amino acids [16], we realized that the contribution of other humanizing substitutions to the function of contact site residues might be inferred from the epistasis scores of such variants. We calculated epistasis by taking the enrichment score of a double mutant and subtracting the product of the scores of the component single mutants. Humanizing substitutions that compensate for the deleterious effects of E181R, A185K or L186M are likely to show positive epistasis (i.e. the double variant functions better than predicted) while humanizing substitutions that do not should display no epistasis. We extracted the epistasis scores for 866 double mutants ([16], S2 Table), each carrying two substitutions that are found in one of the 52 homologues of Pab1. Comparing the epistasis score distribution of these variants to that of 38,742 double mutants that carry pairs of mutations that do not occur in any of the individual homologues of Pab1 that were sampled in our analysis revealed a small yet significant increase (Wilcoxon rank sum test p-value = 3.712e-10) in epistatic interactions between substitutions that are present together in natural variants (Fig. 6C). Thus, two mutations found in a natural protein variant are more likely to interact positively, either by synergistic or compensatory mechanisms.
Of the 866 double mutants with two substitutions found in Pab1 homologues, eight carry one of the strongly deleterious humanizing substitutions together with a second humanizing mutation (S2 Table). Of these, a double mutant carrying the deleterious substitution L186M together with the non-deleterious substitution G177E had a high epistasis score (Fig. 6C). Specifically, while L186M alone resulted in ∼30% loss of in vivo function, addition of the non-deleterious G177E substitution restored Pab1 function to the wild-type level (S2 Table). G177E was able to partly restore eIF4G2 binding of an RRM2 mutant that carries the L186M substitution (Fig. 6D), suggesting that the positive epistasis of G177E and L186M is at least in part due to an improved association of the double mutant with eIF4G2. While adding G177E to the triple mutant did not shift the binding specificity towards the human eIF4G1, humanizing its adjacent residue by E176Q substitution supported this switch (Fig. 6D), suggesting that the local humanized environment of G177E is important for its function. The contribution of E176Q and G177E to human eIF4G1 binding is specific, as other groups of humanizing substitutions, found either at a distance or in close physical proximity to the three deleterious substitutions, were not able to promote this shift in binding specificity (S3A Fig.). Thus, despite the lack of measurable effects of single amino acid substitutions at position 177 of yeast Pab1 (Fig. 1A), the amino acid at this position is important for Pab1 binding to the human eIF4G1. The ancestral state of position 177 in the Pab1 lineage was glutamic acid, which was replaced by glycine in the recent ancestor of S. cerevisiae (S3B Fig.). Therefore, it is likely that the pre-establishment of glutamic acid at position 177 compensated in human for the detrimental effects of at least one of the three deleterious substitutions, while becoming dispensable in the evolutionary path that was taken by S. cerevisiae.
Of the other nine natural and strongly deleterious substitutions in Pab1, five (K140A, L153V, S155V, K156N and A158E) map to the interface between RRM1 and RRM2. In particular, L153 and K156, present in the human orthologue, are key residues in the interaction between RRM1 and RRM2 that allow for efficient poly(A) binding [30]. In addition, an allosteric change in the RRM1 and RRM2 interface upon poly(A) binding is suggested to facilitate the association of RRM2 with eIF4G [31]. Therefore, deleterious substitutions in the RRM1–RRM2 contact site are likely to result from loss of either poly(A) or eIF4G binding activity, or both.
Three other substitutions (Y197N, Y197V and A199E) map to the poly(A)-binding site [30]. Residue 197 is the only RNA-binding residue that is highly divergent, as all the other residues that bind RNA are either identical across the 52 homologues or display a small variation that is highly tolerated by the yeast protein. It is likely that the structure of the poly(A) forces extreme conservation on the RNA-binding residues, similar to enzyme-substrate binding sites [32], in a way that prevents useful characterization by natural substitutions.
We used the deep mutational scanning data on the yeast Pab1 RRM2 domain to delineate the functional consequences of 197 single amino acid substitutions to residues that occur in Pab1 homologues. As expected [33,34], the great majority of these natural substitutions had a minor effect on Pab1 activity, indicating that the primary constraints on poly(A)-binding protein function remain the same among different organisms. Of the 17 strongly deleterious substitutions, nearly all mapped to either protein-protein (RRM2–eIF4G), inter-domain (RRM2–RRM1) or protein-RNA (RRM2–poly(A)) interaction sites, suggesting that all known ligand-binding sites in Pab1 RRM2 experienced some degree of divergence over evolutionary time.
We characterized the eIF4G-binding site in Pab1 at single amino acid resolution, demonstrating that integrating results from mutagenesis with natural variation data provides a compact list of mutations that are likely to interfere with protein-ligand interaction. The rapid generation of this list overcomes limitations of other mutagenesis approaches. In particular, deletion experiments are unlikely to provide an accurate map of the eIF4G contact site in Pab1, as its critical residues span most of the primary sequence of RRM2 and are brought together by the three-dimensional structure. An alanine scan, which tests the effects of substituting single amino acids to alanine, would prove insufficient to identify the involvement of T145, V148 and E181 in eIF4G binding, as judged by the minor effects of these alanine changes on the in vivo function of Pab1 [16].
Although our combined approach delineated eight Pab1 RRM2 surface residues that are associated with eIF4G1 binding, the large-scale two-hybrid assay identified nine additional residues, located mostly at the periphery of the contact site. The higher sensitivity of the RRM2 domain to mutations at the eIF4G1 contact site in the two-hybrid assay is likely due to the higher selection pressure in this assay, than in vivo, for the RRM2 eIF4G1 interaction, as mutations that disrupt this interaction are not lethal [12,16,22]. Nonetheless, the differences between the regions that were highlighted by the two approaches point to the central residues, discovered by our combined approach, as more important to the in vivo function of the eIF4G1 binding site than the peripheral residues, added by the yeast two-hybrid assay. This difference in the importance of residues to the interaction is likely to mirror the higher evolutionary conservation of the central binding site residues compared to the peripheral ones [35].
Despite the greater ability of the two-hybrid assay to uncover positions on the RRM2 surface that associate with eIF4G1 binding, our combined approach of using natural variation to filter the deep mutational scanning results on the in vivo function of RRM2 yields an increased fraction of mutations that interfere with eIF4G1 binding, Because large-scale mutational data are usually not available for a protein interaction, these results emphasize the advantage of this combined approach to identify the effects of mutations on protein interactions. In addition, while mutations that damage the structure of a protein can affect the two-hybrid readout, the short list of candidate mutations created by the in vivo approach is likely to exclude such indirect effects [8]
Elucidating contact sites with high resolution is important to clarify how proteins exert their functions. With respect to Pab1, we found that the binding sites for eIF4G1 and eIF4G2 extend beyond the helix α2 element [12] to include part of helix α1. The inclusion of this helix provides a plausible explanation for the molecular mechanism that couples poly(A) and eIF4G binding by Pab1. In yeast, binding of eIF4G to Pab1 requires the prior association of Pab1 with poly(A) in order to promote translation [22]. In human, these sequential steps are separated by inter-domain allostery of RRM2 and RRM1, allowing PABP1 to adopt a more extended conformation in the presence of RNA [31]. Since the association of RRM2 and RRM1 involves direct interactions between helix α1 of RRM2 and helix α2 of RRM1 [30,31], conformational changes of the two domains might make helix α1 of RRM2 and its surrounding residues available for eIF4G association upon poly(A) binding. Our finding that a Pab1 fragment consisting only of RRM1–RRM2 was unable to bind eIF4G supports the regulatory role of this inter-domain interaction in this function.
eIF4G1 and eIF4G2 are functionally interchangeable under optimal growth conditions [36]. However, differences in eIF4E co-purification and in vitro translation efficiencies suggest that each of the two isoforms possesses unique roles in translation under non-optimal conditions [37,38]. Despite the overlap in location and similar mutational sensitivity of the binding sites for eIF4G1 and eIF4G2, a few Pab1 RRM2 substitutions resulted in differential sensitivities to binding. Whether this difference in Pab1 RRM2 binding points to altered mechanisms of action is a matter of further studies. T145L, which bound only to eIF4G2, might be useful in clarifying specific roles for the isoforms in translation.
We identified three substitutions (E181R, A185K and L186M), corresponding to the residues present in the human PABP1, each of which when introduced into the yeast Pab1 eliminated its binding to yeast eIF4G1. We tested whether these substitutions might switch the specificity of Pab1 to bind human eIF4G1. The single humanizing substitution, E181R, allowed the yeast Pab1 RRM2 to bind to human eIF4G1, demonstrating that in spite of sequence diversification, the human and yeast orthologues of eIF4G1 and Pab1 share similarities with respect to their physical association. However, Pab1 carrying A185K and L186M did not bind to the human eIF4G1, even after humanizing the contact site by other substitutions. Thus, this shared similarity in binding activity is likely to be maintained by other intra-protein interactions in Pab1 that compensate for the otherwise deleterious effect of A185K and L186M.
Our finding that two substitutions that are both present in an individual homologue are more likely to display positive epistatic interactions suggests that compensating mutations reconstruct functional modules that are conserved between organisms despite changes in the amino acid sequence that comprise these modules. Indeed, that the addition of the G177E substitution repairs the binding of an RRM2 L186M mutant to the yeast eIF4G2 suggests that the two humanizing substitutions restore a functional binding site for the yeast eIF4G2. Additional studies will be required to determine whether the tendency for positive epistasis of two substitutions present in a homologous sequence is a universal property of proteins or a specific feature of Pab1. Nonetheless, it is likely that substitutions from paralogues of closely related species are more prone to this type of epistasis than substitutions from other homologous sequences, given the functional conservation and the small number of amino acid changes in these paralogues.
Additionally, G177E together with E176Q combined with the three deleterious substitutions E181R, A185K and L186M to allow yeast Pab1 binding to the human eIF4G1. This finding supports the use of epistatic interactions between two natural substitutions tested in a model organism to infer similar interactions between those residues in their natural context. We suggest that systematic integration of large-scale epistasis data with bioinformatic tools that use sequence homology might improve prediction accuracies of co-evolutionary relationships and functional association between residues.
Approximately 20% of S. cerevisiae genes are essential for growth on rich glucose medium [39], with many of the remaining genes required upon environmental or genetic perturbations. Therefore, growth selections compatible with deep mutational scanning can be used to study the in vivo function of a large fraction of yeast proteins. This experimental strategy can also be applied to cross-species complementation assays to analyze human proteins in yeast [9,40,41]. However, the score assigned to each protein variant reflects the consequence of mutation only on growth rate. Therefore, inferring the direct impact of mutations on an in vivo activity such as ligand binding remains challenging. Here we show that integrating deep mutational scanning results with natural variation data provides a high throughput inter-species complementation assay that can be used to identify and characterize functional regions in proteins, including protein-protein contact sites. In addition, the large-scale analysis of natural amino acid substitutions can provide an experimental platform to evaluate the performance of computational tools that use protein homology to predict function and co-evolutionary relationships.
The procedures for Pab1 RRM2 deep mutational scanning, including establishment of the experimental platform, construction of mutant libraries, sequencing of RRM2 DNA fragments and data analysis were previously described [16]. Unless otherwise indicated, only variants with input-read counts greater than 40 were used for the analysis.
pOBD2 and pOAD were used to test the interactions between Pab1 and eIF4G isoforms in the yeast two-hybrid system. Full length PAB1 encoding amino acids 1–578 (DMP87) was PCR amplified from pCM188-Pab1 [16] and cloned into the NcoI and SalI sites of pOBD2. The following PAB1 truncations, encoding amino acids 1–343 (DMP88), 1–204 (DMP183), 123–204 (DMP180) and 1–120 (DMP179) were PCR amplified from p415GPD-Pab1-343BX [16] and cloned into the NcoI and SalI sites of pOBD2. PAB1 fragments encoding amino acids 123–204 (RRM2) with the point mutations I137F (DMP201), D138T (DMP202), K140A (DMP203), A141D (DMP230), T145H (DMP204), T145L (DMP189), V148K (DMP205), L153V (DMP206), S155V (DMP207), K156N (DMP208), A158E (DMP209), K180E (DMP197), E181R (DMP193), D184E (DMP188), D184W (DMP210), A185H (DMP211), A185K (DMP186), L186M (DMP185), Y197N (DMP191), Y197V (DMP194), A199E (DMP192), [K180E, E181R] (DMP198), [D184E, A185K, L186M] (DMP190), [E181R, A185K, L186M] (DMP235), [E181R, A185K, L186M, A158V, T159C] (DMP286), [E181R, A185K, L186M, E176Q, G177E] (DMP287), [E181R, A185K, L186M, V148A, K180E, D184E] (DMP291), [E181R, A185K, L186M, P135K, Q194R] (DMP292), G177E (DMP293), [G177E, L186M] (DMP297) and [E181R, A185K, L186M, G177E] (DMP298) were created by PCR using the same p415GPD-Pab1-343BX plasmid as a template and cloned into the NcoI and SalI sites of pOBD2, C-terminal and in-frame with the Gal4 DNA binding domain. eIF4G1 and eIF4G2 fragments encoding amino acids 1–341 were amplified from yeast genomic DNA (strain W-303) and cloned into the EcoRI and SalI sites of pOAD, C-terminal and in-frame with the Gal4 activation domain (DMP92 and DMP212, respectively). The human PABP1 fragment encoding amino acids 95–176 was amplified from HsCD00042197 (PlasmidID) and cloned into the NcoI and SalI sites of pOBD2 (DMP264). The human eIF4G1 fragment encoding amino acids 1–260 was amplified from HsCD00342900 (PlasmidID) and cloned into the NcoI and SalI sites of pOAD (DMP265)
Yeast strain PJ694a (MATa trp1-901 leu2-3,112 ura3-52 his3-200 gal4Δ gal80Δ LYS2::GAL1-HIS3 GAL2-ADE2 met2::GAL7-lacZ) carrying pOBD2- and pOAD-based vectors were grown overnight, at 30°C in synthetic complete media lacking leucine and tryptophan. To test for activation of the HIS3 reporter gene, cells were spotted in a dilution series on synthetic complete plates lacking leucine and tryptophan, with or without histidine and grown at 30°C for three days.
We collected 52 Pab1 homologues (see S1 File for sequences and accession numbers), representing sequences of all poly(A)-binding proteins that carry four consecutive RRM domains that can be found in the UniProtKB/SwissProt database [42], which contains high quality annotations of protein sequences. Multiple sequence alignment (MSA) was performed using Clustal Omega [43] with default parameters (S2 File). Enrichment scores for natural and non-natural single amino acid substitutions were obtained from Supplementary Table 2 of Melamed et al [16].
To assess the fraction of natural substitutions that result in impaired function, enrichment score distributions of 160 natural single amino acid substitutions, 539 non-natural single amino acid substitutions and 229 synonymous variants with input read counts greater than 500 were determined. The stringent input read count threshold was set to minimize fluctuations of enrichment scores due to low representation of variants in the library pools. The enrichment scores distribution of the synonymous variants was used as a proxy for the enrichment scores distribution of non-deleterious variants in the dataset. To estimate the fraction of deleterious substitutions within the natural substitutions, for each enrichment score bin shown in S1B Fig., we subtracted the estimated fraction of non-deleterious variants from the fraction of the natural variants.
For each single amino acid substitution, the fractional accessible surface area (ASA) was obtained for the side chains of the wild-type residue in the human PABP1 RRM2 structure (PDB ID 2K8G) using VADAR server, version 1.8 [44]. Data for K164 residue was omitted, as this residue is absent from the human RRM2 (see Fig. 2A). The Blosum62 matrix was used to score each substitution to determine the degree of conservation. Box plots were generated using R-studio software.
RRM2 sequences containing one of the three library segments were PCR amplified from the library plasmids that were previously described [16] and cloned into the NcoI and SalI sites of pOBD2. Yeast expressing the RRM2 hybrid containing one of the three libraries were grown to log phase in SC medium lacking leucine and tryptophan, supplemented with 2% glucose, and diluted into fresh medium lacking leucine, tryptophan and histidine to a final concentration of 4 × 104 cells/mL. Selection was carried out for 21 h with the culture growing to a density of 5 × 106–1 × 107 cells/mL. 2.5 × 108 cells from each culture were collected before (“input”) and after selection (“selected”). Library preparation for high throughput sequencing was carried as previously described [16]. Amplicons were created with internal primers that flanked each library segment and carried at their 5’ end common sequencing targets for Illumina read1, read2 and index primers (11 PCR cycles) and with external primers that added Illumina adapter sequences (8 PCR cycles). Amplicons were sequenced by an Illumina NextSeq500 using paired-end reads. We used the Enrich software package (Fowler et al. 2011) to filter for high quality reads (base Q score >20). Based on the variance of enrichment scores of 2423 synonymous variants (i.e. variants that encode the wild-type Pab1 RRM2 protein sequence and carry at least one synonymous mutation), we selected variants with at least 20 input read counts for further analysis (synonymous variance <0.4 for all three libraries). Enrichment scores of single amino acid substitutions were log2 transformed and visualized using Java TreeView 1.1.6r2 [45]. Average linkage hierarchical clustering with a Euclidean distance similarity metric for both RRM2 residues and substituting amino acids was performed using Gene Cluster 3.0 [46].
A maximum likelihood tree was constructed using the Phylogeny.fr tool [47] using default parameters. Probabilities for ancestral states were calculated using the JTT model of substitution by the FastML tool [48]. Ancestral amino acids were considered “true” if their reconstruction probabilities were greater than 0.7 (the sum of probabilities for all amino acids equals 1.0). Otherwise, the two most probable amino acids with a minimal probability of 0.3 for each, and sum of probabilities greater than 0.75 were considered.
The human PABP1 RRM1-RRM2 structure (PDB ID 1CVJ) was visualized using PyMol software (version 1.5.0.5).
|
10.1371/journal.pntd.0005485 | Genetic diversity and population structure of the tsetse fly Glossina fuscipes fuscipes (Diptera: Glossinidae) in Northern Uganda: Implications for vector control | Uganda is the only country where the chronic and acute forms of human African Trypanosomiasis (HAT) or sleeping sickness both occur and are separated by < 100 km in areas north of Lake Kyoga. In Uganda, Glossina fuscipes fuscipes is the main vector of the Trypanosoma parasites responsible for these diseases as well for the animal African Trypanosomiasis (AAT), or Nagana. We used highly polymorphic microsatellite loci and a mitochondrial DNA (mtDNA) marker to provide fine scale spatial resolution of genetic structure of G. f. fuscipes from 42 sampling sites from the northern region of Uganda where a merger of the two disease belts is feared. Based on microsatellite analyses, we found that G. f. fuscipes in northern Uganda are structured into three distinct genetic clusters with varying degrees of interconnectivity among them. Based on genetic assignment and spatial location, we grouped the sampling sites into four genetic units corresponding to northwestern Uganda in the Albert Nile drainage, northeastern Uganda in the Lake Kyoga drainage, western Uganda in the Victoria Nile drainage, and a transition zone between the two northern genetic clusters characterized by high level of genetic admixture. An analysis using HYBRIDLAB supported a hybrid swarm model as most consistent with tsetse genotypes in these admixed samples. Results of mtDNA analyses revealed the presence of 30 haplotypes representing three main haplogroups, whose location broadly overlaps with the microsatellite defined clusters. Migration analyses based on microsatellites point to moderate migration among the northern units located in the Albert Nile, Achwa River, Okole River, and Lake Kyoga drainages, but not between the northern units and the Victoria Nile drainage in the west. Effective population size estimates were variable with low to moderate sizes in most populations and with evidence of recent population bottlenecks, especially in the northeast unit of the Lake Kyoga drainage. Our microsatellite and mtDNA based analyses indicate that G. f. fuscipes movement along the Achwa and Okole rivers may facilitate northwest expansion of the Rhodesiense disease belt in Uganda. We identified tsetse migration corridors and recommend a rolling carpet approach from south of Lake Kyoga northward to minimize disease dispersal and prevent vector re-colonization. Additionally, our findings highlight the need for continuing tsetse monitoring efforts during and after control.
| Northern Uganda is an epidemiologically important region affected by human African trypanosomiasis (HAT) because it harbors both forms of the HAT disease (T. b. gambiense and T. b. rhodesiense). The geographic location of this region creates the risk that these distinct foci could merge, which would complicate diagnosis and treatment, and may result in recombination between the two parasite strains with as yet unknown consequences. Both strains require a tsetse fly vector for transmission, and in Uganda, G. f. fuscipes is the major vector of HAT. Controlling the vector remains one of the most effective strategies for controlling trypanosome parasites. However, vector control efforts may not be sustainable in terms of long term reduction in G. f. fuscipes populations due to population rebounds. Population genetics data can allow us to determine the likely source of population rebounds and to establish a more robust control strategy. In this study, we build on a previous broad spatial survey of G. f. fuscipes genetic structure in Uganda by adding more than 30 novel sampling sites that are strategically spaced across a region of northern Uganda that, for historical and political reasons, was severely understudied and faces particularly high disease risk. We identify natural population breaks, migration corridors and a hybrid zone with evidence of free interbreeding of G. f. fuscipes across the geographic region that spans the two HAT disease foci. We also find evidence of low effective population sizes and population bottlenecks in some areas that have been subjects of past control but remain regions of high tsetse density, which stresses the risk of population rebounds if monitoring is not explicitly incorporated into the control strategy. We use these results to make suggestions that will enhance the design and implementation of control activities in northern Uganda.
| The tsetse fly (genus Glossina) is the major vector of human African trypanosomiasis (HAT) and animal African trypanosomiasis (AAT). The diseases occur throughout sub-Saharan Africa, causing extensive morbidity and mortality in humans and livestock [1][2]. Human disease is caused by two different subspecies of the flagellated protozoa Trypanosoma brucei; T. b. rhodesiense in eastern and southern Africa, and T. b. gambiense in west and central Africa. The two HAT diseases are separated geographically more or less along the line of the Great Rift Valley [3]. Although the animal disease (or Nagana) is caused by different trypanosome subspecies; T. b. brucei, T. congolense and T. vivax, animals are also known to be reservoirs of the human infective T. b. rhodesiense. Thus, while AAT is a problem in its own right because of economic losses and reduced availability of nutrients [4][5][6], the animals also act as important reservoirs for human infective T. b. rhodesiense.
Although the human diseases have been on a decline[7], they still put 60 million people at risk in 37 countries covering about 40% of Africa [8][9]. The human disease T. b. gambiense is near elimination while control of T. b. rhodesiense remains more complicated because of animal reservoirs. For both T. b. gambiense and T. b. rhodesiense, there are no prophylactic drugs or vaccines available. Furthermore, the drugs for treatment are expensive, can cause severe side effects, and are difficult to administer in remote villages [10][11]. Although AAT can be prevented with prophylactic drugs and effectively treated with trypanocidal drugs, progress towards elimination of the animal disease has been slow because of the high cost of drug administration and repeated emergence of drug resistance [12]. Thus, AAT instances remain high and continue to burden livestock farmers [13] and provide animal reservoirs of T. b. rhodesiense. As a consequence, the most effective way to control both AAT and HAT is control of the tsetse vector [14].
Uganda is in the precarious position of being the only country that harbors both forms of HAT, with T. b. gambiense present in the northwestern corner of the country and T. b. rhodesiense found in the center and southeast [15]. There is a significant risk that the two sleeping sickness subspecies will merge in the north-central districts of Uganda, a region already burdened by political and social instability [16]. Merging of the two disease belts would complicate treatment and diagnosis [17], and may lead to the emergence of unforeseen pathologies if there is recombination between the T. b. gambiense and T. b. rhodesiense trypanosomes [18][19][20].
Glossina fuscipes fuscipes is a member of the palpalis group of tsetse and is the main vector implicated in the transmission of both AAT and HAT in Uganda. The vector is distributed over vast regions of sub-Saharan Africa (Fig 1), where it occupies discrete patches of riverine and lacustrine habitats distributed among pasture and agricultural land. Assessing the population structure and the extent to which these apparently discrete populations are connected by dispersal and migration patterns is central to defining the most effective scale for vector control [21][22]. For example, the major challenge that faces most control efforts is tsetse rebound following short-term control efforts. The source of rebounding populations could be residual pockets of surviving individuals or migrant flies coming from neighboring untreated regions, or both [23]. Increased knowledge of vector population dynamics through application of population genetics can help in assessing the suitability of the operational units selected for vector control and result in more effective tsetse control.
Although previous studies have made great strides towards understanding the population biology of G. f. fuscipes in Uganda [24][25][26][27][28], regions north of Lake Kyoga remain a high priority for additional study. Northern Uganda harbors both forms of HAT in close geographic proximity [7][20]. Identifying the precise extent of the two disease belts and possible risk of merger has been difficult until recently because of social and political upheaval experienced in these regions [29][30]. Our previous population genetic studies have identified three major genetic units present in north and south of Lake Kyoga and in western Uganda [24][25][31]. Each of these units consists of genetically distinct populations with high differentiation between sampling sites and evidence of further sub-structuring [22][25][31][24].
A more detailed understanding of the genetic structure and population dynamics of G. f. fuscipes in northern Uganda will help estimate the likelihood of the merger of the two HAT disease forms, and identify the best tsetse control strategies for the region. In this study, we comprehensively sampled G. f. fuscipes from 42 sites in areas north of Lake Kyoga and assessed variation in 16 nuclear microsatellites and over a 570 bp region of mitochondrial DNA (mtDNA) to understand both short time scale resolution of demographic events [32][33] and inference of phylogeographic events dating further back in time [34][35]. We compared this new knowledge of fine scale population structure, migration patterns and population dynamics in northern Uganda with previous studies that concentrated on the southern and central regions of the country. This comparative approach allowed us deeper insights into the evolutionary forces at play and enriched our ability to make recommendations for G. f. fuscipes control strategies in northern Uganda.
The map in Fig 1 shows the sampling sites. We used biconical traps [36] to collect from 30 sites between January 2014 and April 2015, and also included 12 collections from a previous study between January 2008 and January 2012 [31][25]. Sampling sites were chosen to detect fine spatial scale population structure. To do this, we collected from multiple sites separated by just over 5 km, which is the smallest unit area for which genetic differentiation has been observed in G. f. fuscipes in Uganda [25]. At each site, we placed an average of 6 traps at least 100 m from each other and collected an average of 18 flies per trap over a period of 3–4 days. Flies were stored individually in 95% ethanol and information on sex, collection date, trap number, and geographic coordinates of each trap was recorded. The genotypic data included from previous studies [25][29] were separated from our samples by a time span of 3–7 years (approx. 22–52 generations), which opened up the possibility of genetic change. However, a previous study showed no evidence of large demographic changes between the temporal collections [37], justifying the combined analyses of 42 sampling sites spanning these time points (details in S1 Table).
We extensively sampled in areas north of Lake Kyoga, which includes tsetse flies that in a previous study were grouped into two genetic clusters [25], with an effort to sample major water drainage systems. [25] described one genetic cluster north of the Lake Kyoga and the Victoria Nile, and one in western Uganda. In the northern genetic cluster, we sampled the Albert Nile, the Achwa River, the Okole River, and Lake Kyoga drainages (see Table 1). The Albert Nile basin is in the far northwest corner of Uganda, a region known to be an active focus of T. b. gambiense sleeping sickness [20][38]. The Albert Nile is bordered to the east and eventually joined by the Achwa River, and both drainage systems generally consist of patchy habitat suitable for G. f. fuscipes, characterized by lowland woodland near semi-permanent water bodies [2]. Habitat patches are surrounded by unsuitable savannah, agricultural and pastoral lands. Although the district of Arua was recently included in a pilot vector control program in 2011–2013 [[39]], our samples from 2014 (DUK, AIN and GAN) that may have been impacted did not overlap spatially with the program. Further south, we sampled the Okole River and Lake Kyoga basins (Fig 1, Table 1). These regions form vast areas of marsh and swampland, and are the most northerly geographical extent of the T. b. rhodesiense HAT disease belt [40]. Some districts in the Lake Kyoga drainage, such as Dokolo and K’maido, were targets of the Stamp Out Sleeping Sickness (SOS) campaign of 2006–2009 [41], which may have impacted our samples from this region from 2009 (OC) and 2014 (OCU, AMI and KAG). Finally, in the western genetic cluster described by [25], we collected samples along and south of the Victoria Nile (Fig 1, Table 1), which flows northwest from Lake Kyoga into Lake Albert on the edge of the western rift valley. Here we sampled sites along and on minor tributaries of the Victoria Nile in the districts of Masindi and Kiryandongo (Table 1). This region is characterized by lowland woodland and the Uganda Wildlife Authority protects much of the region as part of the Murchison Falls National Park.
DNA was extracted from two to three legs per sample, using PrepGEM Insect DNA Extraction kit (ZyGEM New Zealand, 2013), following the manufacturer’s protocols and stored at -20°C. We collected genotypic data from 16 microsatellite loci (details in S2 Table). Amplifications were performed with fluorescently labeled forward primers (6-FAM, HEX and NED) using a touchdown PCR (10 cycles of annealing at progressively lower temperatures from 60°C to 51°C, followed by 35 cycles at 50°C) in 13.0μl reaction volumes containing 2.6 μl of 5X PCR buffer, 1.1 μl of 10 mM dNTPs, 1.1 μl of 25mM MgCl2 and 0.1 μl of 5 units/μl GoTaq (Promega, USA), 0.1 μl of 100X BSA (New England Biolabs, USA), 0.5 μl of 10 μM fluorescently-labeled M13 primer, 0.5 of μl 10 μM reverse primer, and 0.3 μl of 2 μM M13-tailed forward primer. For loci C7b and GmL11, 0.5 units of Taq Gold polymerase (Life Technologies, USA) were used instead of Promega GoTaq. PCR products were multiplexed in groups of two or three and genotyped on an ABI 3730xL Automated Sequencer (Life Technologies, USA) at the DNA Analysis Facility on Science Hill at Yale University (http://dna-analysis.yale.edu/). Alleles were scored using the program GENEMARKER v2.4.0 (Soft Genetics, State College, PA, USA) with manual editing of the automatically scored peaks.
We followed the protocol designed by [25] to sequence a 570 bp fragment of mtDNA that spans the COI and COII genes. Briefly, we used primers COIF1 (5’–CCT CAA CAC TTT TTA GGT TTA G– 3’) and COIIR1 (5’–GGT TCT CTA ATT TCA TCA AGT A– 3’) to amplify 570 bp with an initial denaturation step at 95°C for 5 min, followed by 40 cycles of annealing at 50°C, and a final extension step at 72°C for 20 min. We used a reaction volume of 13.0 μl containing 1 μl of template genomic DNA, 2.6 μl of 5X PCR buffer, 1.1 μl of 10 mM dNTPs, 0.5 μl of 10mM primers, 1.1 μl of 25 mM MgCl2, and 0.1 μl (U/μL) of GoTaq polymerase (Promega, USA). The PCR products were purified using ExoSAP-IT (Affymetrix Inc., USA) as per the manufacturer’s protocol. Sequencing was carried out for both forward and reverse strands on the ABI 3730xL automated sequencer at the DNA Analysis Facility on Science Hill at Yale University (http://dna-analysis.research.yale.edu/). Sequence chromatograms were inspected by eye and sequences trimmed to remove poor quality data using GENEIOUS v6.1.8 (Biomatters, New Zealand). The forward and reverse strands were used to create a consensus sequence for each sample, and the sequences trimmed to a length of 490 bp. Only a subset of the samples screened for microsatellite variation was also sequenced at the mtDNA locus (Table 1).
For nuclear microsatellite marker validation, we tested for neutrality and independence with GENEPOP v4.2 [42]. We tested for departures from Hardy-Weinberg (HW) proportions in each sample and microsatellite locus using an approximation of an exact test based on a Markov chain iteration (10,000 dememorization steps, 1000 batches, 10,000 iterations per batch in the Markov chain); significance values were obtained following the Fisher’s method that combines probabilities of exact tests [43]. We tested for genotypic linkage disequilibrium (LD) among each pair of loci using the Guo and Thompson method [44]. To correct for false assignments of significance by chance alone for all simultaneous statistical tests and comparisons, we used the Benjamini-Hochberg False Discovery Rate (FDR) method [45]as opposed to the Bonferroni correction, because of lower incidence of false negatives[45][46].
For nuclear microsatellite data, we assigned individuals to genetic units without prior information on sampling locality with STRUCTURE v2.3.4 [47][48]. STRUCTURE simultaneously identifies unique genetic units and provides a probability of assignment (q-value, ranging from 0 to 1) for each individual. Twenty independent replicate runs for each K = 1–10 were carried out with an admixture model, independent allele frequencies, and a burn-in value of 50,000 steps followed by 250,000 iterations. The optimal value of K was determined using STRUCTURE HARVESTER v0.6 [49] to calculate the ad hoc statistic “ΔK” [50], and independent replicates were aligned with CLUMPP v1.1.2 [51].
In addition to STRUCTURE, we performed Discriminant Analysis of Principal Components (DAPC) with the "adegenet" package v1.4–2 [52] in the R version 3.0.2 environment [53]. The DAPC is a multivariate, model-free method that makes no assumptions about deviations from Hardy Weinberg and linkage disequilibrium, designed to describe patterns of genetic clustering among groups of individuals [54]. In this analysis, we grouped samples by their site of origin and used the cross-validation formula available to choose number of principal components (PCs) to retain. To understand the partitioning of microsatellite variance within and between genetic units, we performed an analysis of molecular variance (AMOVA) in ARLEQUIN v3.5 [55].
Genetic diversity indices including observed heterozygosity (HO), expected heterozygosity (HE), number of alleles, allelic richness (AR) and [56] estimator of inbreeding coefficient (FIS) were calculated in GENALEX v6.501 [57]. Pairwise differentiations at different hierarchical levels were estimated with two F-statistics. For comparison with previous G. f. fuscipes studies, we calculated Wright’s F-statistics [58], following the variance method developed by [56] using 10,000 permutations in ARLEQUIN. For accuracy with highly polymorphic markers [59], we estimated Jost’s DEST with the R package DEMEtics [60][53], where p-values and confidence intervals were calculated based on 1000 bootstrap resamplings. With the resulting F-statistics, we tested for isolation by distance (IBD) using Rousset’s procedure [61] implemented in the “isolation by distance” v3.23 web service [62]. Geographic distances were generated using the web-based “geographic matrix generator” v1.2.3 [63]. The significance of the regression was tested by a Mantel test with 10,000 randomizations [64].
Using microsatellite data, we estimated effective population size (Ne) for each sampling site independently. We did not group sites based on assignment to genetic clusters because strong evidence of substructure within clusters would violate assumptions. We estimated Ne using two methods implemented in NEESTIMATOR v2.01 [65]: the modified two-sample temporal method [66] based on [67] for sites with multiple temporal samples, and the one-sample linkage disequilibrium (LD) method [68] for all 42 sites. We used two methods to estimate Ne because they each have different strengths and weaknesses [66,69–71]. The two-sample temporal method [64] is useful because it is robust when there are overlapping generations [71], but only provides an average estimate across two time points assuming a closed population, so cannot be used to assess the impact of control efforts. On the other hand, the LD method [66] is useful because it can provide an estimate for each sampling point and employs the bias corrections by [72], but is influenced by bias associated with non-overlapping generations and is not powerful enough to distinguish from infinite population sizes when there are insufficient polymorphisms and numbers of markers to detect patterns of LD [67, 71].
We tested for population bottlenecks using two methods implemented in the program BOTTLENECK v1.2.02 [73]. The first method tested for an excess of heterozygosity relative to observed allelic diversity [74]. We used the two-phase mutation model (TPM), the most appropriate for microsatellites [75], with 70% single-step mutations and 30% of multiple-step mutation. Significance was assessed using Wilcoxon’s signed rank test, as is recommended when fewer than 20 loci are used [73]. The second method tested for a bottleneck-induced mode shift in allele frequency distributions that is usually evident in recently bottlenecked populations [76].
We investigated the mixed ancestry suggested by STRUCTURE analysis in the Achwa and Okole River regions. These sampling sites displayed an average probability of assignment (q-values) of less than 0.8 (See Table 1), which could either be caused by methodological shortcomings (i.e. low genetic distance and inability of the markers to distinguish clear genetic clusters), or by accurate detection of interbreeding of two distinct lineages. Following [77], we tested for accurate detection of interbreeding by comparing observed admixture data against two alternative admixture models; a pure mechanical mixing model representing a scenario of strong reproductive barriers and free migration, and a hybrid swarm model representing a scenario of free hybridization and admixture using HYBRIDLAB v1.0 [78]. For the mechanical mixing model, we simulated individual admixture proportions by randomly drawing alleles from the observed allele frequency distribution of 'pure' samples where the average probability of assignment (q-values) were greater than 0.8 to a single STRUCTURE cluster (Table 1). For the hybrid swarm model, we simulated individual admixture proportions from the observed allele frequency distribution of 'admixed' samples where the average probability of assignment (q-values) were less than 0.8 to any single STRUCTURE cluster (Table 1). We chose localities from the geographic extremes of the northwest and northeast units to represent 'pure' samples, and regions with generally uncertain assignment from the Achwa and Okole Rivers to represent 'admixed' samples. Then, we used STRUCTURE to estimate individual probability of assignment (q-value) with all three datasets; the true observed genotypes, the simulated genotypes under a mechanical mixing model, and the simulated genotypes under a hybrid swarm model. Finally, we used a Wilcoxon signed rank test to assess differences between observed and simulated distributions. We interpret significant differences between simulated mechanical mixing and hybrid swarm datasets as evidence that uncertain STRUCTURE assignments do not represent a methodological shortcoming. Likewise, we interpret non-significant differences between the observed data and the simulated hybrid swarm data as evidence for interbreeding.
Using microsatellite data, we determined if patterns of observed genetic structure could be attributable to sampling related individuals, by testing for relatedness between individuals using the program ML-Relate [79]. We assigned pairwise relationships within each genetic unit into one of four relationship categories: unrelated (U), half siblings (HS), full siblings (FS) or Parents/offspring (PO).
Detection of first generation migrants and progeny of successful mating of very recent migrants between genetic regions was done using GENECLASS v2.0 [80], and using FLOCK v3.1 [81], a program that provides accurate assignment of individuals to genetic units of origin even in the absence of pure genotypes. In GENECLASS, we used the "detect migrant function", which calculates the likelihood of finding an individual in the locality in which it was sampled (Lh), the greatest likelihood among all sampled localities (Lmax), and their ratio (Lh/max) to identify migrants. To distinguish true from statistical migrants (type I error), we selected the Rannala and Mountain criterion [82], and the Monte Carlo resampling algorithm of [83] (n = 1000) to determine the critical value of the test statistics, Lh/Lmax. Individuals were considered immigrants when the probability of being assigned to the reference population was lower than 0.05. In FLOCK, we used a K value of 4, starting partitions chosen by location of origin, ran 500 iterations and used a log-likelihood difference threshold (LLOD) value of 1.
For the mtDNA sequence data, all statistical parameters and tests were calculated using the program ARLEQUIN [55]. Genetic diversity within populations was estimated by computing haplotype diversity (Hd) and nucleotide diversity (Nd) [84] in DnaSP v5.0 [85]. Relationships among haplotype lineages were inferred by constructing a parsimony network using TCS [86] implemented in the free, open source population genetics software PopART (http://otago.ac.nz). We used nucleotide diversity to estimate genetic differentiation (ΦST) and performed an analysis of molecular variance (AMOVA) in ARLEQUIN. We tested for IBD with the same methods described above for the microsatellites.
Finally, we compared mtDNA haplogroup assignment with the Microsatellite STRUCTURE assignment, and tallied the percent mismatch. Individuals were considered a mismatch if they displayed a high q-value (probability of assignment) score to one microsatellite based cluster but low frequencies of the haplogroup generally found in the same geographic region as the microsatellite based cluster.
Of the 17 microsatellite markers considered, most were under HW equilibrium in the majority of populations with the exception of pg17, which was dropped from the analyses, as it showed significant departures from HWE at P<0.05. All remaining loci were polymorphic in all populations analyzed except D101, which was monomorphic in one population (OC). The most polymorphic locus was GpB20b (24 alleles), while the least polymorphic was B05 (5 alleles; details in S3 Table). None of the LD tests on pairs of microsatellite loci gave a significant result after the Benjamini-Hochberg correction, confirming neutrality and independence of markers.
Fig 2A shows the results from STRUCTURE analyses. In this analysis, individuals fell into three genetic clusters with clear geographic variation in probability of assignment (q-value) (Table 1, S4 Table). The DAPC multivariate analysis (S1 Fig) corroborated the results of STRUCTURE.
Based on the results of the STRUCTURE and DAPC analyses and their geographic locations, we grouped sampling sites into four units: West, Northwest, Transition Zone, and Northeast. Sampling sites west of lake Kyoga along the Victoria Nile (i.e. UWA, KR, KF, MS) had average q-values > 0.8 to a single cluster (blue in Fig 2). The samples north of Lake Kyoga (Fig 1) belong to two genetic clusters (gray and orange, Fig 2) and were grouped into three units. The “Northwest” unit comprises samples from the Albert River drainage (e.g. DUK, GAN, and AIN) and the most northerly Achwa River sites (i.e. NGO, LAG and PAW) with average q-values > 0.8 to a single cluster (gray in Fig 2). The “Northeast unit” comprises samples from north of Lake Kyoga (e.g. KAG, AM, OCU and AMI, Table 1) with average q-values > 0.8 to one cluster (orange in Fig 2). Sampling sites between the Northwest and Northeast units in the Achwa and Okole River basins (Fig 1) had a much lower average q-value (0.54) than the other three units, and moving west to east, probability of assignment to one cluster (gray in Fig 2) progressively decreased while it increased for the other cluster (orange in Fig 2). We refer to this region between the Northwest and Northeast units as the "Transition Zone".
Microsatellite based FST between sampling sites either within or between the Structure-defined clusters ranged from 0 to 0.229 with most comparisons being statistically significant (S5A Table). Table 2 reports average FST between the four units (Northwest, Transition Zone, Northeast, and West). The West unit is the most genetically distinct from the other three (FST = 0.162, 0,163, and 0.218 for Northwest, Transition Zone, and Northeast, respectively). Lower but still statistically significant FST values were estimated between the Northwest and Northeast units (FST = 0.064) and even lower values between these units and the Transition Zone (FST = 0.021 and 0.035, respectively). DEST values showed the same trend as FST, except with overall higher estimates (S6 Table). Isolation by distance (IBD) analyses (S7 Table) showed a significant correlation between genetic distance and geographic distance for all sampling sites combined (R2 = 0.438, p = 0.0001) and for sampling sites within the Northwest (R2 = 0.259, p = 0.00) and Transition Zone (R2 = 0.216; p = 0.001). No significant IBD was detected among sampling sites in the Northeast and West units.
AMOVA results using microsatellites showed that most of the variation was between individuals within sampling sites (89.63%) but differences were statistically significant at all levels of comparison, including between sampling sites and among the four units (Table 3).
Overall, all sites showed moderate to high levels of genetic variability (S1 Table). HO ranged from 0.461 in LAG to 0.690 in OM and HE ranged from 0.537 in KAG to 0.678. For most of the sites, HO and HE microsatellite diversities were similar, indicating random mating within sites. Averaged over all samples and loci, the inbreeding coefficient (FIS) were generally low with an overall grand mean of 0.048±0.008, and with significant heterozygote excess in 7 out of 42 populations (S1 Table). Allelic richness ranged from a high of 7.785 in KR to a low of 4.188 in AMI, with an overall mean of 5.186 (S1 Table). Generally, microsatellite diversity was highest in flies sampled in the Northwest and the Transition Zone sites (Table 1; S1 Table) and lowest in flies from the Northeast (e.g. in sites KAG, AM, OCU and AMI). The trend of decline in diversity from the Northwest to the Northeast is apparent and significant when allelic richness values were linear-regressed over longitude (R2 = 0.121; p = 0.032; S2 Fig). Flies in the West unit had microsatellite diversity values similar to or on par with the Northwest unit.
S8 Table shows the results of the Ne estimates based on microsatellite data using the LD and the temporal methods. Estimates using the heterozygote excess method were infinite for all sites tested. Using the one-sample LD method, Ne estimates ranged widely from 101.6 (36.4-infinite 95% confidence interval [CI]) in OC to 1685.7 (234.2-infinite CI) in UGT and were all bound by a CI that included infinity (S8 Table). Ne estimates using the two-sample temporal method ranged from 103 (73–138 CI) in KTC to 962 (669–1309 CI) in OCU (S8 Table). Where a comparison between the two methods was possible, estimates were largely congruent except for one site (OCU), where Ne using the temporal method was 962 (669–1309 CI), and using the LD method was 112 (47.7-infinite CI; S8 Table). Results based on the TPM model indicate a genetic bottleneck in 5 sampling sites (NGO from the Northwest, OCA from the Transition zone, AMI and OC from the Northeast, and MS from the West; S8 Table). Results based on allele frequency distributions showed a genetic bottleneck in only one sample (AMI from the Northeast; S8 Table).
Fig 3 shows the results of the HYBRIDLAB analyses. The distribution of STRUCTURE assignments from the simulated hybrid swarm and mechanical mixing datasets are clearly distinct (Fig 3) with a Wilcoxon two-tailed p-value of 0.002 (S9 Table). This indicates power to detect interbreeding in the transition zone and thus evidence of hybridization. Comparisons of these models with the observed data (S9 Table) indicate that the observed data (Fig 3A) matches most closely with the hybrid swarm model (Fig 3B) than the mechanical mixing (Fig 3C) from which it is statistically distinct.
Relatedness analyses showed that the majority (>86%) of the individuals in all units are unrelated (Table 4). The percentage of individuals that were full siblings was very low, ranging between 0.33% and 0.91% for all units. An even lower number of individuals had parent-offspring relationships ranging from 0% in the Transition Zone to 1.04% in the Northeast.
Microsatellite-based migrant detection using GENECLASS and FLOCK showed a higher number of migrants between the Northwest, the Transition Zone, and the Northeast than between these areas and the West (Fig 4). GENECLASS indicated slight asymmetry in migration into the Northwest. There are 20 migrants from the Transition Zone into the Northwest and 10 migrants in the reverse direction, with both sexes almost equally represented (10 and 2 male migrants vs. 10 and 6 female migrants; S10 Table). We also detected two first generation female migrants from the Northeast to the Northwest. In contrast, migration between the Transition Zone and the Northeast is symmetrical with 8 migrants from the Northeast into the Transition Zone and 9 migrants in the opposite direction. Both sexes are moving in both directions, although the low sample sizes (2 and 3 male migrants vs. 5 and 1 female migrants; S10 Table) precludes any strong conclusion. We also detect two migrants between the Northwest and West, one in each direction. FLOCK analysis provided similar migration rates between regions (Fig 4), but showed less asymmetry from the Transition Zone into the Northwest (23 from the Transition Zone into the Northwest, and 17 in the opposite direction), and more asymmetry from the Northeast into the Transition Zone (13 from Northeast into the Transition Zone, two in the opposite direction; S10 Table). FLOCK showed no migration between the West and any other region (S10 Table).
The mtDNA dataset consisted of 481 sequences (490 bp long), which included 289 sequences from individuals sampled for this study (a subset of the ones screened for microsatellite loci variation, Table 1) plus 192 sequences from individuals from previous ones [25][31], Table 1). Sequences could be grouped into 30 haplotypes (S11 Table), displayed as a TCS network (Fig 2B). There are three major haplogroups (groups of related haplotypes); Haplogroup A, Haplogroup B, and Haplogroup C (Fig 2B). Table 1 reports haplogroup frequencies for each site and for the four units. Haplogroup A occurs in all studied regions, but is most frequent in the Northwest and West units (75.1% and 74.4%, respectively). It occurs less commonly in the Transition Zone (24.0%) and only rarely in the Northeast (4.8%). Haplogroup B occurs most commonly in the Northeast unit (95.2%) and it is less common going from Northeast unit to the Transition Zone (76.0%) and to the Northwest unit (24.9%). Haplogroup B does not occur in the West unit and Haplogroup C occurs only in this unit (25.6%, Table 1, Fig 2C).
The number of haplotypes at each sampling site ranged from 1 to 6 (S1 Table). Haplotype 1 is the most frequent (186 individuals) and occurs in all units except the West, and falls into Haplogroup B (S11 Table). Haplotype 2 from Haplogroup A is the second most common (140 individuals) and it is found in all units (S11 Table). The third and fourth most common haplotypes fall in Haplogroup A and C, and only occur in the West (41 individuals and 19 individuals, respectively). Thirteen haplotypes were singletons (observed once in the sample) and fall into a mix of haplogroups, eight of which were from the Northwest, four from the Transition Zone and one from the West (S11 Table). Nucleotide diversity averaged 0.002 and ranged from 0 (OSG and KF) to 0.008 (UWA; S1 Table). Likewise, average haplotype diversity was 0.757, and ranged from 0 (KF and OSG) to 0.836 (UWA; S1 Table). There was no apparent difference in haplotype diversity from Northwest to Northeast units.
S6B Table shows estimates of genetic differentiation (ΦST) between sampling sites. ΦST ranged from 0 to 1; with some sampling sites showing no evidence of differentiation (e.g. PD in the Transition Zone vs AMI in the Northeast), while reached 1 for pairs that did not share haplotypes at all (e.g. OSG in the Northwest vs KF in the West). S7 Table shows the results of the IBD analyses using mtDNA-based ΦST. Like in the microsatellite-based test, the correlation between genetic distance and geographic distance was significant for all sampling sites combined (R2 = 0.490, p = 0.001) and for samples within the Northwest unit (R2 = 0.425, p = 0.001), but non-significant for the Northeast and West units. Unlike in the microsatellite-based IBD tests, the correlation between geographic and genetic distance in the Transition Zone was non-significant (R2 = 0.002; p = 0.374).
AMOVA results based on mtDNA agree with the Microsatellite (Table 3), with most of the variation between individuals within sites (59.84%) and significant values at all levels of comparison, including between the four units (Northwest, Transition Zone, Northeast, and West; Table 3).
To evaluate the possible role of differential introgression of nuclear vs mitochondrial markers we assessed levels of mismatches by comparing individual assignments for each marker type (S4 Table), and calculated percent individuals with discordant nuclear vs mitochondrial assignment (Table 1). This analysis could only be done for the three northern units because the common microsatellite based cluster (blue in Fig 2) in the West was not clearly associated with a single mitochondrial haplogroup, as both Haplogroup A and Haplogroup C occur there. On the contrary, the Northeast and Northwest were clearly associated each with a single haplogroup, so we scored any individual from the north with a microsatellite-based q-value greater than 0.9 as a “match” if both nuclear and mitochondrial assignments were associated with the same region (grey/grey or orange/orange in Fig 2), or a “mismatch” if assignments were associated with different regions (grey/orange or orange/grey in Fig 2). The highest percentage of mismatches were found in the Northwest unit (22.8%), followed by the Transition Zone (20.03%), and then the Northeast unit (4.0%) (Table 1).
We evaluated the fine scale genetic structure of G. f. fuscipes populations north of Lake Kyoga in Uganda, a region that is of special interest due to the impending risk of merger of the two forms of HAT disease that G. f. fuscipes transmits in Uganda. Our sampling scheme targeted the fine spatial scale resolution of genetic structure so as to provide the most accurate information available on genetic connectivity and population dynamics in the region spanning the two HAT disease foci. This kind of information is necessary to inform vector control program design [87–91]. Findings indicate two strong genetic breaks in northern Uganda and determine that hybridization is occurring freely across the contact zone between the Northwest and Northeast. We explored underlying mechanisms of population dynamics in northern Uganda and found that large influence has been imposed by (i) sustained connectivity of the Northwest with the rest of the G. f. fuscipes species range, (ii) past geologic events associated with the opening of the great rift valley during the last ~35 ka, and (iii) vector control programs that have caused population bottlenecks but have not always sustainably controlled tsetse populations. We also identified a general pattern of isolation by distance and moderate migration within interconnected regions. Findings suggest that population rebounds may have occurred from very close by populations soon after vector control efforts, and that interbreeding across a hybrid zone that spans the two disease foci could promote recolonization from different genetic units across further distances. These results support the need for long-term monitoring and a design that mitigates recolonization from neighboring regions, especially within the hybrid zone that spans the two disease foci.
Genetic diversity at both microsatellite and mtDNA markers (Table 1) were generally low compared to many Diptera and Coleoptera species, which is consistent with reproductive limits imposed by the tsetse's viviparous life history [92]. The mtDNA haplotype network (Fig 2B) was congruent with the network published by [24] with more haplotypes because of the higher spatial resolution of this study. Levels of diversity in both markers (Table 1) were similar to previous estimates for sampling sites north of Lake Kyoga [25][31], but higher than estimates of southern Uganda populations [27]. We found a subtle decline in genetic diversity from west to east (S1 Fig) in northern Uganda similar to the pattern previously observed in central and southern Uganda [25]. [25] suggested that this gradient reflected sequential founder events originating from the main tsetse belt in the Northwest and moving eastward. Conversely though, our results for northern Uganda are not consistent with a single genetic origin from the main tsetse belt because we found two distinct genetic backgrounds (Fig 2A) and two distinct mtDNA haplogroups (Fig 2B). This apparent inconsistence between past and current results could be due to the inability of previous studies to pick up the spatial differentiation and admixture patterns that we detected because of their much sparser sampling than in this study. Rather than sequential founder events pushing for a northwest to northeast range expansion, our results suggest that sustained connectivity to the greater G. f. fuscipes distribution and recent human induced population processes, such as vector control and habitat destruction, may account for the higher genetic diversity in the Northwest vs the Northeast.
The G. f. fuscipes range extends continuously westward as far as Cameroon and Gabon (Fig 1; [93][94]) and has been sustained since the last glacial maximum ~15–20 ka [95][2][96], with the Uganda sites being at the extreme northeast of G. f. fuscipes’ contiguous distribution. The size of this range and its temporal stability suggest that populations from the main part of its distribution are likely to be interconnected and old enough to harbor the high levels of genetic diversity characteristic of large and stable populations. This may have facilitated intermittent gene flow and can be a factor in explaining the higher genetic diversity in the Northwest than in Northeast of Uganda. In contrast, populations to the east and south of Lake Kyoga are bordered by unsuitable habitat to the east [93], and have experienced recent arid periods during the last glacial maximum ~15–20 ka, and again during the latest Pleistocene ~14 ka, when the lakes in Uganda completely desiccated multiple times [97][95]. These climate events could have led to contractions and expansions of populations, accentuating the effects of genetic drift and creating isolated populations with low genetic diversity such as that found in UGT, AMI, AM, OC, KAG, and OCU (Table 1; S2 Fig). Despite high genetic diversity in some localities in the West such as UWA (Table 1; S2 Fig), which conforms with the general pattern of high to low diversity from west to east (S2 Fig), connectivity with the rest of the G. f. fuscipes distribution in the West is limited by Lake Albert and the less suitable habitat in the bordering Blue Mountains. Thus, we suggest that the high allelic richness and haplotype diversity in the West was created by contact between the distinct genetic lineages at the Victoria Nile with a small amount of asymmetrical introgression (see below) rather than through connectivity with the central part of the species range.
As expected, the gradient from higher to lower effective population size estimates (Ne) from the northwest to the southeast parallels the results on genetic diversity, and is likely caused by similar evolutionary forces, as Ne calculations are based on diversity estimates. Our interpretation of Ne was somewhat limited because we were only able to draw inference from the two-sample temporal method. As expected, the one-sample LD based Ne estimates yielded high confidence intervals that overlapped with infinity (S8 Table). Improved Ne estimates based on a larger number of nuclear markers will be an important focus of future research using Single Nucleotide Polymorphisms (SNPs). Despite uncertainty in Ne estimates from the LD method [67], the temporal method [64] provided estimates that ranged from 100 to 1000, had low confidence intervals (S8 Table), and showed higher estimates in the Northwest. This result is in line with the distinct life-history traits of tsetse flies such as lower population sizes, reproductive outputs, and longer generation times than other insects. Ne and genetic diversity results that we report for the Northwest were similar to what has been reported in G. f. fuscipes sampled from northern Uganda [31][25] and in G. palpalis, another riverine species [1]. However, estimates were higher than reported in populations from southern Uganda [27]. This suggests that Northwest populations are influenced by either high connectivity with the rest of the G. f. fuscipes range, or by lower levels of vector control in the Northwest as compared to regions impacted by the SOS campaign in the Northeast.
Detection of recent bottlenecks (S8 Table) provides further evidence that Ne has been influenced by vector control campaigns. The bottleneck analysis we used can detect extreme reductions in population sizes that occurred more recently than 2–4 Ne generations [67][98], which corresponds to 25–500 years in G. f. fuscipes, depending on the exact Ne and generation time of the population in question [99][25][31][27]. Signals of bottlenecks from these short time scales can be due to natural or human induced changes in population size. Both of these causes may be at play given G. f. fuscipes’ patchy distribution, unique life history traits, and history as the target of intense, even if somewhat irregular, vector control campaigns. Bottlenecks in OC and AMI can be attributed to the SOS campaign of 2009 [100][41]. Similar tsetse control projects, like the Farming in Tsetse Controlled Areas (FITCA) in southeastern Uganda in places not included in this study like Okame, Otuboi, and Bunghazi, resulted in detection of bottlenecks in these areas in previous studies [26][25]. On the other hand, we found no evidence of bottlenecks in the 2014 samples most proximal to the location of the pilot vector control program conducted by [39] in the district of Arua (DUK, AIN and GAN). This may have been because the location of sampling was too spatially distant (minimum of ~20 km) to influence the population sampled, or because the time of sampling was too temporally near (~6 months) for a genetic signal to propagate. Survey data indicates that some control efforts resulted in long-term reduction in tsetse census [26,31], while other control efforts such as the SOS campaign in the Northeast resulted in only short-term population size reductions. Population rebound is evidenced by the similar number of flies caught per trap at sites in the SOS region and at sites where no control activities have been carried out. For example, during sampling in 2014, traps set at two sites in the SOS region (OCU and AMI) caught an average of 67 and 14 flies per trap, which are numbers that are similar to or higher than the average catch of 18 flies per trap. This underscores the importance of long-term control and monitoring campaigns following tsetse control. A population bottleneck in GOR (S8 Table) remains unexplained by known vector control campaigns, which may indicate that some bottlenecks are induced by natural evolutionary processes such as weather events or changes in ecological interactions. Thus, there is evidence that the joint influence of natural processes such as the greater connectivity to the rest of the G. f. fuscipes distribution in the Northwest, the dramatic climate change including arid periods in central and southern Uganda in the last ~35 ka years, and recent vector control programs have determined the west to east gradient in G. f. fuscipes genetic diversity and population dynamics in northern Uganda.
Clustering (Fig 2A) and multivariate (S1 Fig) analyses detected three distinct genetic clusters each composed of multiple sampling sites and broadly corresponding to the Northwest, Northeast, and West units (Fig 2C). mtDNA haplotypes also clustered into three haplogroups (Fig 2B), which approximately correspond to these same three regions (Fig 2C). Two of these genetic lineages have been described by previous research as Northern and Western clusters [24][25][31], and we find evidence of a previously undescribed divergence in the Northern cluster, which is now partitioned into Northeast, Transition Zone, and Northwest units.
Our data confirm deep genetic divergence between the G. f. fuscipes nuclear lineages found at the Victoria Nile, which harbor a mix of mtDNA haplotypes of both northern and southern associated lineages (Fig 2). STRUCTURE clustering showed close geographic proximity of distinct clusters at the confluence of the Okole River and the Victoria Nile (Fig 2). This stark genetic break in the nuclear genetic make-up may be due in part to insufficient sampling between UWA in the West and AKA and OLE in the North for accurate description of the shape and geographic span of the genetic divergence between these regions. Future sampling efforts should encompass detailed sampling in this region to determine if there is indeed another transition zone between the West and units identified in this study (i.e. Northwest, Transition Zone, and Northeast), and between the West and previously described units [24][25][31]. Divergence between the North and West is thought to have originated during past allopatry more than 100,000 years ago [25]. Subsequent changes in the river systems associated with the opening of the great rift valley 13,000–35,000 years ago [95] created the modern outflow from Lake Victoria into Lake Kyoga and the reversal of the Kafu river to meet the Victoria Nile before flowing into Lake Albert [97]. These changes may have shifted the range of the Western G. f. fuscipes populations into contact with the Northern units at the Victoria Nile.
We find mixed mtDNA ancestry in the West, which [24] described and suggested indicates recent rare female dispersal from the north and chance amplification of northern haplotypes by drift. Another possible explanation is the preferential introgression of organelle DNA from the resident population into the colonizing genetic background when two divergent lineages come into secondary contact during range expansion [101]. This scenario is supported by changes in the river systems and multiple drying cycles of the lakebeds [85] that would have promoted repeated retraction and expansion of G. f. fuscipes in central Uganda. If northern lineages had recolonized central Uganda before a northward shift of the southwestern lineage, the result would be a large number of northern mtDNA haplotypes in a Western nuclear genetic background. There are also possible ecological interactions at play because this region is unique in the co-distribution of other tsetse species, G. morsitans submorsitans and G. pallidipes [2] especially in the large protected area of the Murchison Falls National Park (Fig 1). Thus, evidence supports that strong evolutionary and ecological forces maintain genetic distinctiveness between the West and the other genetic units, but the details remain unclear and an important focus for future work. More fine scale sampling of all tsetse species across the North/West genetic break as well as experiments that test mating compatibility would help shed light on the mechanism(s) that maintain genetic discontinuity.
Both microsatellite based FST and mtDNA based ɸST showed significant differentiation despite our fine scale sampling effort, which aligns with previous studies that have found significant differentiation across small geographic scales of as little as 1–5 km2 in Uganda [102][27][25]. Tsetse flies are known to be sensitive to environmental conditions and exist in discrete patches [2]. We suggest that low connectivity between adjacent habitat patches coupled with small Ne has allowed genetic drift to create significant differentiation at small spatial scales in G. f. fuscipes in northern Uganda. High signals of isolation by distance we detected in both microsatellites and mtDNA (S7 Table) further support the idea that population structure is maintained by the dual action of migration and genetic drift.
The genetic break between the Northwest and Northeast forms a broad region of mixed microsatellite and mtDNA assignment along the Achwa and Okole rivers, in what we call the Transition Zone. The genetic break between the Northwest and Northeast and the one between the broad northern and southern clusters described by [25] are both characterized by what we think are secondary contact with admixed individuals and introgression of mtDNA haplotypes. However, the width of the transition zone, the level of differentiation, and the patterns and levels of admixture, is different across these two contact zones, with a broader, less differentiated, and more gradual pattern of admixture in the Transition Zone than in the North/South contact. The Transition Zone extends more than 200 km (Fig 2), while the secondary contact zone between the North and South clusters extends less than 75 km [25][31]. This difference in width may have been facilitated by colonization patterns and the distinct geographical break imposed by the swampy upper reaches (southern extent) of Lake Kyoga at the contact zone between the North and South, while less conspicuous physical breaks only partially limit movement of flies to and from the Transition Zone (Fig 1). The Transition Zone is characterized by uninterrupted suitable habitat along the entire length of the Achwa River, with only short distances of less than 15 km between the Achwa and Okole Rivers and neighboring drainage basins of Lake Kyoga and the Albert Nile (Fig 1).
The levels of differentiation are also different between these two contact zones. In the Transition Zone, microsatellite-based FST estimates are lower (average FST = 0.064, Table 2) than the comparable values for the North and South clusters (average FST = 0.236; [25]). This pattern was even more extreme in mtDNA ΦST estimates, with an average ΦST of 0.080 between the Northwest and Northeast (Table 2) and 0.535 between the North and South [25][103][31].
Similarly, the patterns of admixture are distinct between the two contact zones. In the North/South contact zone, there is a dramatic increase in mismatched individuals that assign with high frequency (>90%) to one nuclear based genetic cluster but with mtDNA haplotypes found in another [31] at the zone of contact, with 16.98% in the contact zone vs 0–2% on either side. The pattern of mismatch in the North/South contact zone suggest differential introgression of mtDNA and nuclear loci, which could be due to Wolbachia infections [25][104][103][31], given its maternal inheritance and ability to induce cytoplasmic incompatibility in G. morsitans [105] and other insects [106]. In contrast, in northern Uganda, the Transition Zone does not display an increase in mismatches, with 19.9% in the contact zone vs 26.2% to the north, which leaves no evidence of differential levels of introgressions of the two markers (Table 1) or asymmetrical introgression. The match of observed data with the hybrid swarm model in the HYBRIDLAB analysis (Fig 3) provides further evidence of relatively free and symmetrical interbreeding in the Transition Zone over multiple generations.
Taken together, our results suggest that for the northern secondary contact area, isolation by distance and genetic drift are the two most likely processes that have shaped the distribution of the nuclear and mtDNA polymorphisms, rather than Wolbachia infections. Nonetheless, symmetrical interbreeding in the Transition Zone of this study remains tentative without the ability to classify hybrid classes because of wide and overlapping 95% confidence intervals around expected q-values (Fig 3A). Further genetic characterization of the northern hybrid zone as well as characterization of the circulating Wolbachia strain(s) in the North would improve understanding of the forces shaping the genetic cline that lies between the disease belts of the two forms of HAT in northern Uganda.
The methods we used to detect migrants reflects both first generation migrants and progeny of successful mating of very recent migrants rather than dispersal, and thus allowed us to assess recent gene flow across the full geographic range of our study [82][81]. We detect comparatively high migration rates among the northern clusters and low migration between these units and the West. High gene flow between the three northern units supports the assertion by [91] and others that waterways, in this case the Achwa River, maintain connectivity in tsetse populations. The vast majority of the migrants were a result of short-range dispersal from geographically proximate sampling sites connected by rivers. GENECLASS detected only two long-range migrants from the Northwest into the Northeast, which would not be expected with available ecological and physiological data that indicate tsetse cannot disperse over long distances [107]. Thus, it is likely these long-range migrants are offspring of assortative mating between first generation migrants found in geographically intermediate locations rather than first generation migrants.
The overall direction of migration we detected was slightly asymmetrical towards the Northwest from the Northeast. However, we found no evidence of sex-bias (S10 Table). These findings agree with previous studies which detected similar movement rates for the two sexes for G. f. fuscipes from the southeast into the northeast [25][31]. [25] suggested that movement along riverine habitats might be linked to passive dispersal of pupae via seasonally flooded river systems. Transportation of adults and pupae downstream may also be aided by large floating islands with dry substrate that form in backwaters and eddies and move northwards for sometimes hundreds of km along the major rivers in the region such as the Nile and its tributaries, and potentially, the Achwa river [108] [109]. Nonetheless, this hypothesis remains to be tested, and alternatives include the movement of flies with livestock [3][110], shifting distribution of suitable habitat with human activities, and ongoing migration along corridors of suitable habitat that connect the north and south of Uganda.
The observations from this study have important implications on the epidemiology of the two HAT diseases, as well as on future vector control and monitoring efforts in this region. A dense sampling scheme across a relatively small geographic area allowed an unprecedented spatial resolution of genetic structure in this region. Our results point to the presence of four genetic units, three of which have high levels of gene flow among them. The genetic distinctiveness of the West from the other three units suggests that this unit could be treated as a separate entity from the Northern ones. However, when planning control and monitoring strategies, it is opportune to look at the patterns and levels of genetic discontinuities between West vs. South and West vs. North genetic backgrounds in more detail to more precisely define the boundaries of each genetic unit at a country-wide scale. Given the results of this work, for sampling sites North of Lake Kyoga, control efforts undertaken at the unit level are unlikely to produce long-lasting results due to re-invasion from adjacent units, unless physical barriers are incorporated to avoid re-invasion from adjacent units. The best strategy would be a “rolling-carpet” initiative where control is initiated from the Northeast through the Transition Zone into the Northwest followed by intense monitoring and additional control to manage fly migration from previously cleared sites due to population recrudescence after control.
Our results suggest that ecological and geographic features, especially the river systems in northern Uganda, play a major role in keeping G. f. fuscipes populations connected–a fact that should be taken advantage of when designing control. The genetic connectivity we found along waterways provides further support for a vector control strategy that incorporates targets along waterways and barriers to recolonization from adjacent stretches of riverbanks. This idea is also supported by a recent study that comprehensively evaluated a “tiny targets” vector control strategy along riverine savannah and found that a target density of 20 per linear km can achieve >90% tsetse control [39].
Our data also suggest that there is current movement of flies from the Northeast and Northwest into the Transition Zone but with a slight asymmetry towards the Northwest. Given that previous studies also demonstrated northward migration from the east [25][31], it is possible that tsetse, besides livestock movement, is contributing to the northwards expansion of the T. b. rhodesiense sleeping sickness.
Of major relevance for disease control is the finding of high levels of genetic intermixing and gene flow in the Transition Zone, which implies that a fusion of the two diseases (T. b. gambiense and T. b. rhodesiense) is unlikely to be prevented by an incompatibility between vector populations in the region of contact. Given the extent of connectivity in the three northern genetic units and the apparent genetic stability of G. f. fuscipes populations in the region [37], ongoing monitoring following control would be paramount if interventions are to be sustainable. Monitoring programs should involve a combination of both ecological and genetic surveys to check on changes in population density and re-emergence either from residual pockets of tsetse or dispersal from proximal locations. For example, our results from the Northeast highlight the risk of population rebound following control. In this region, we found strong evidence of genetic bottlenecks indicating initial success of the SOS campaign in reducing tsetse density. However, our 2014–2015 surveys in the same sampling sites returned some of the highest tsetse trap densities. It appears, therefore, that when control activities were relaxed at the end of the SOS campaign, tsetse populations recovered to high densities. Focused monitoring could provide early detection of such population rebound and allow for identification of the source and proper mitigation of the recolonizing tsetse.
|
10.1371/journal.pntd.0003262 | Evidence for Co-evolution of West Nile Virus and House Sparrows in North America | West Nile virus (WNV) has been maintained in North America in enzootic cycles between mosquitoes and birds since it was first described in North America in 1999. House sparrows (HOSPs; Passer domesticus) are a highly competent host for WNV that have contributed to the rapid spread of WNV across the U.S.; however, their competence has been evaluated primarily using an early WNV strain (NY99) that is no longer circulating. Herein, we report that the competence of wild HOSPs for the NY99 strain has decreased significantly over time, suggesting that HOSPs may have developed resistance to this early WNV strain. Moreover, recently isolated WNV strains generate higher peak viremias and mortality in contemporary HOSPs compared to NY99. These data indicate that opposing selective pressures in both the virus and avian host have resulted in a net increase in the level of host competence of North American HOSPs for currently circulating WNV strains.
| West Nile virus (WNV) emerged in North America in 1999 and rapidly spread across the U.S. due to the presence of highly susceptible mosquito vectors and avian hosts. One of the major avian reservoirs for WNV in the U.S. is the house sparrow (HOSP), which has low mortality during WNV infection. Here, we investigate how the response of wild HOSPs to WNV infection has changed as a result of the 15-year history of WNV circulation in the U.S. In addition, we evaluated the impact of WNV evolution on viral infection in HOSPs and report that WNV has become increasingly pathogenic to HOSPs over time, while HOSPs may have developed resistance to early WNV strains. Thus, HOSPs are still likely to be an important avian reservoir for WNV in the U.S., and WNV has adapted to its avian hosts during emergence in North America.
| West Nile virus (WNV; Flaviviridae) is an arbovirus that was first reported in North America in 1999 in New York. By 2003, the virus had spread to the West Coast. WNV has remained endemic in the U.S. due to the high prevalence of competent Culex spp. mosquito vectors and avian hosts [1]–[4]. The birds considered to be the most important WNV reservoirs are passerines, which are highly susceptible and maintain high viremias for several days during infection [1], [5]. Because infection of Culex vectors is dose dependent, the magnitude of serum viremia in a bird determines its host competence [1], [4], [6].
Resident birds are considered to be more important for the spread of WNV across the U.S. than migratory birds [7]. The house sparrow (HOSP; Passer domesticus) is a resident passerine and is highly competent for WNV [1], [5]. Additionally, HOSPs are ubiquitous across North America in urban, suburban, and rural landscapes and are a frequent bloodmeal source for Culex mosquitoes [8]–[10]. Unlike infection in American crows (Corvus brachyrhynchos), infected HOSPs sustain viral titers above the threshold required for mosquitoes to become infected but exhibit a low mortality rate [5]. However, reports have suggested that WNV causes enough mortality to contribute to a declining population of HOSPs in the U.S. [11]–[13]. The WNV seroprevalence of HOSPs has been estimated to fluctuate annually and locally [14], [15], with levels reaching as high as 40% during outbreak years [16].
Since the first identification of WNV in North America, the virus has diverged into 3 described genotypes. By 2003, the original East Coast genotype was replaced by the North American WN02 genotype, defined by a valine-to-alanine amino acid substitution at codon 159 in the envelope protein (E-V159A) [17], [18]. There are reports that suggest viral isolates containing this mutation may increase the rate of WNV dissemination in Culex mosquitoes [17], [19], [20]. A third genotype, SW03, was first described for WNV isolates collected in the southwest U.S. in 2003 [21]. The SW03 genotype is characterized by the E-V159A substitution in conjunction with an alanine-to-threonine substitution at codon 85 in the NS4A protein (NS4A-A85T). The NS4A-A85T mutation has not been specifically assessed for differential viral phenotypic effects in either avian hosts or mosquito vectors. Isolates obtained during routine surveillance since 2003 have largely been limited to WN02 and SW03 genotypes that have been found co-circulating in the U.S. as recently as 2012 [22], [23].
Dual-host viruses such as WNV have many constraints on viral evolution. Due to the necessity for replication in birds and mosquitoes for its enzootic maintenance in North America, WNV has been subject to widespread purifying selection [21], [23], [24] to maintain efficient replication in two disparate hosts [25], [26]. However, WNV has adaptively evolved at discrete loci during its 15 years of circulation in North America [21], [23], and whether or not this evolution has been driven by passerines is unknown. In order to assess the possibility that transmissibility of WNV could be a driving force for the fixation of the E-V159A substitution and alternative genotype-specific amino acid substitutions in North America, the competence of HOSPs for East Coast, WN02, and SW03 genotype viruses were compared. Furthermore, the competence of North American HOSPs for the same founding East Coast strain, NY99, was also evaluated over 14 years to identify potential co-evolutionary signatures in an avian host.
A protein alignment of 132 WNV isolates was performed using Clustal Omega [27]. Twelve of these isolates were used for experimental inoculation of HOSPs in this study. A maximum likelihood phylogeny was constructed with 1,000 bootstrap replicates using PhyML [28]. Non-synonymous diversity and divergence calculations were performed using DNAsp v5 [29]. Sequences for isolates TX8759 and TX8779 were determined as described previously and have been assigned GenBank accession numbers KJ786936 and KJ786935, respectively [30].
Wild HOSPs were trapped in Larimer County, CO, in 2012–2013 using mist nets. Serum from each bird was tested for WNV neutralizing antibodies using a 90% plaque reduction neutralizing test as reported previously [31]. Groups of 5–8 seronegative birds were inoculated subcutaneously with 1,500 PFU of WNV. Blood was collected daily by jugular venipuncture for 7 days post-inoculation. Blood was immediately diluted 1∶10, coagulated for 30 minutes at room temperature, and spun for 10 minutes at 2500× g. Serum viral titers were quantified using Vero cell plaque assay as reported previously [32]. The lower limit of detection for this assay was 1.7 log10 PFU/mL.
Reservoir competence was calculated as the product of HOSP susceptibility, mean daily HOSP infectiousness, and duration of infectiousness for mosquitoes, as previously reported [5]. A value of 1.0 for HOSP susceptibility was used for all WNV isolates, as 100% of challenged birds demonstrated viremias. The lower threshold of HOSP serum viremia considered infectious to mosquitoes was 4.7 log10 PFU/mL [4]. Infectiousness was calculated based on a linear regression analysis as the proportion of mosquitoes predicted to become infected after feeding on a host with known viremia [4], [6], [33], [34].
Statistical significance of differences in peak viremia and reservoir competence was calculated using ANOVA. A Mantel-Cox log-rank test was used to compare survival curves. For regression analyses, r2 values were used to determine the best model, and a linear model was used. All calculations were performed using GraphPad Prism 6 (San Diego, CA) or R (www.R-project.org).
This work was performed under approved institutional animal care guidelines. Protocols were approved by the Institutional Animal Care and Use Committees at the Division of Vector-borne Diseases, Centers for Disease Control and Prevention (approval number 13-009), the University of California, Davis (approval numbers 12874 and 15895), and Colorado State University (approval number 10-2078A).
Previous studies of WNV evolution in North America have identified 3 major genotypes: East Coast, which includes the prototypic NY99 strain that was the first isolate sequenced during the U.S. epidemic but is no longer known to be in circulation; WN02, characterized by a valine-to-alanine mutation at E-159; and SW03, characterized by the E-V159A substitution and an alanine-to-threonine substitution at NS4A-85 (Table 1). These broad groups form three clusters in a phylogeny of North American isolates (Fig. 1). However, the SW03 genotype also includes some isolates that cluster within the WN02 genotype, such as isolate 12 (TX2689; Fig. 1). This suggests that the NS4A-A85T substitution has occurred independently on multiple occasions, and the SW03 genotype encompasses a group of viruses with variable genetic backgrounds.
In general, the East Coast genotype contains lower genetic diversity compared to the WN02 and SW03 groups (Table 2). Fewer East Coast genotype isolates are available because it was circulating for only a few years, compared to nearly a decade of circulation and diversification for WN02 and SW03 genotypes [35]. This is reflected in the WNV phylogeny, where the relative sizes of each genotypic population are emphasized, and in the proportions of the isolates we chose to test (Fig. 1). In order to examine the fitness effects of WNV diversity, we selected 12 isolates collected between 1999 and 2012 (Table 1) that recapitulate the genetic variation and divergence of WNV in the U.S (Tables 2–4). We chose two East Coast isolates from New York, six WN02 isolates from Texas, and four SW03 isolates from Texas and Mexico, identified in Figure 1 by numbers 1–12.
To determine whether the WN02 displacement of the East Coast genotype was the result of viral adaptation to North American avian hosts, groups of HOSPs collected in 2012 and 2013 were inoculated with 12 WNV isolates representing the three North American genotypes: East Coast, WN02, and SW03 (Table 1). In total, seventy-two birds were collected and inoculated with WNV in 2012–2013, and viremias were measured daily for 7 days. As expected for wild-caught birds, there was considerable variability in viral titers among replicates within groups (Fig. 2a). While the peak viral titer was generally observed on day 3 for HOSPs inoculated with any WNV genotype, peak viremias of individual birds occurred on different days. To determine the overall peak titer for each virus, the peak viral titer for individual HOSPs was determined irrespective of the day post-inoculation and then averaged. The peak titers also were averaged by viral genotype, and peak viral titer varied significantly by genotype. WN02 viruses induced a mean peak titer in HOSPs that was 10-fold greater than East Coast viruses (Fig. 2a, p<0.05). SW03 viruses produced a similar 10-fold increase in mean peak viral titer over East Coast viruses, though this difference was not significant (p = 0.09). This is likely due to the large amount of variation in viral titers observed from inoculated HOSPs (Fig. 2a).
To investigate whether viral adaptation to HOSPs has occurred over time, peak viral titers were analyzed by year of viral isolate collection. Linear regression analysis indicated that peak viral titer increased at a significant rate (Fig. 2b, p<0.05) with an average increase in peak titer of 0.10 log10 PFU/mL sera per year (95% CI: 0.04 to 0.16). The mean peak viral titer induced in HOSPs collected in 2012–2013 by WN02 and SW03 viruses isolated in 2012 was 1.2 log10 PFU/mL sera higher than the peak titer generated by NY99 in HOSPs collected in 2012–2013. This analysis is consistent with the corresponding chronologic appearance of East Coast vs. WN02/SW03 genotypes and indicates that WNV has adapted to HOSPs over time.
Moderate mortality is characteristic of WNV infection in the HOSP. The percentage of surviving birds was calculated for each virus for 7 days post-infection. On average, HOSPs collected in 2012 and inoculated with East Coast, WN02, and SW03 isolates had similar survival curves, with a mean mortality of 15–30% by 7 days post-infection (Fig. 2c). However, three viruses had significantly different survival profiles: TX2689 (SW03), TX8759 (WN02), and TX8779 (WN02), which resulted in 65–85% HOSP mortality by 7 days post-infection (p<0.05) and induced the highest peak viral titers among the 12 WNV strains tested in HOSPs (8.9, 8.3, and 7.9 log10 PFU/mL sera, respectively; Fig. 2a). Interestingly, these 3 viruses were isolated in 2012, indicating that some WNV strains circulating in 2012 may have been more pathogenic to HOSPs than those isolated in previous years. Significantly, the two viruses with the highest mortality and peak viral titers (TX2689 and TX8759; Fig. 2b and 2c) share a common amino acid substitution, NS2A-R188K (Tables 3 and 4), that emerged in North America as early as 2008 [23].
Infectiousness of WNV-infected HOSPs for mosquitoes is a combination of both the magnitude and duration of viremia. In the absence of performing vector competence studies for all of the viruses assessed, the mean reservoir competence index for each viral isolate in HOSPs was generalized by predicting the proportion of mosquitoes likely to become infected using linear regression analysis based on previously published data [4], [6], [33], [34]. With these calculations, an index value of 1.0 would indicate that 100% of mosquitoes feeding on a host for 1 day would be predicted to become infected by the host, though it does not predict the number of mosquitoes that would transmit WNV or the effects on mosquito survival. The mean competence index for HOSPs infected with WNV isolates from the WN02 genotype was 2.4, compared to 1.1 for the East Coast genotype (Fig. 3a), indicating that 120% more mosquitoes would be predicted to become infected after feeding on HOSPs infected with a WN02 isolate compared to mosquitoes feeding on HOSPs infected with an East Coast isolate. The mean HOSP competence index for the SW03 genotype was 1.9 (Fig. 3a), meaning that 73% more mosquitoes would become infected after feeding on HOSPs infected with a SW03 isolate than mosquitoes feeding on HOSPs infected with an East Coast isolate, and 26% more mosquitoes would become infected from feeding on HOSPs infected with a WN02 isolate compared to feeding on HOSPs infected with a SW03 isolate. These results were compared to previously published WNV competence indices for birds inoculated with the NY99 strain of WNV. Species of the avian order Anseriformes, such as the Canada goose (Branta canadensis), have WNV competence indices close to 0 and are considered non-competent hosts [5], [6]. Estimates for passerines suggest competence indices of at least 1, with HOSPs having values between 1 and 1.5, and members of the Corvidae family, such as the blue jay (Cyanocitta cristata), having values between 1.5 and 2.5 [5], [6]. HOSPs inoculated with 6 of the 12 tested WNV isolates, including NY99, had competence indices within the typical range of HOSPs (NY99, NY2001, TX114, TX2600, TX6115, M19433; Fig. 3a). However, HOSPs inoculated with 6 other isolates had competence indices greater than 2, which is more similar to the range for corvids (AR7465, TX8759, TX7191, TX8779, AR6572, TX2689; Fig. 3a). When the reservoir competence indices were stratified by year of viral isolate collection, a significant association between year and index value was observed (Fig. 3b, p<0.05) with an average increase in reservoir competence of 0.09 per year (95% CI: 0.05 to 0.15). The WNV competence of HOSPs trapped in 2012–2013 for viruses collected over 13 years varied between 1.1 and 2.6, or a 140% increase in predicted mosquito infectivity.
Given that WNV induces mortality in HOSPs and that mortality is likely related to the magnitude of viremia induced [36], it is likely that WNV infection has imposed a selective pressure on HOSPs for reduced infection-related mortality by reducing peak viremias. To determine whether HOSPs have modulated their ability to sustain WNV replication over time, results from similar experimental inoculations with the NY99 strain in HOSPs trapped between 2000 and 2014 were compared to data from this study using HOSPs trapped in 2012–2013. Peak viral titers were calculated for individual birds in 7 previous experiments, including 3 published studies [1], [5], [31], and analyzed by year of HOSP collection. Four of the previous experiments were performed at the Centers for Disease Control and Prevention using HOSPs trapped in Larimer County, CO, and 3 experiments were performed at the University of California-Davis using HOSPs trapped in Kern County, CA. These 2 geographically distinct populations of HOSPs showed no difference in peak viral titer over time; therefore, they were analyzed together. Using combined data from HOSPs trapped between 2000 and 2014, the peak viral titer for infected HOSPs was found to be significantly negatively associated with year of HOSP collection (Fig. 4a, p<0.05) with an average decrease in peak titer of 0.11 log10 PFU/mL sera per year (95% CI: 0.03 to 0.18). Overall, the mean peak viremia elicited by the NY99 strain in HOSPs has decreased by 1.0 log10 PFU/mL sera from 2000 to 2014. As expected, the 7-day survival of HOSPs inoculated with NY99 has increased from 75% in 2002 to 100% in recent years, though this trend is not significant (Fig. 4b). Accordingly, the mean host competency index also demonstrated a negative correlation with the year of HOSP collection (Fig. 4c, p<0.05) with an average decrease in reservoir competence of 0.10 per year (95% CI: 0.04 to 0.15). The mean competence index value for HOSPs inoculated with NY99 has decreased from 1.6 in 2000 to an estimated 0.9 in 2014. This difference would be expected to correlate with a decrease in mosquito infection of 44% for HOSPs inoculated with NY99.
WNV has evolved to replicate to higher peak titers in HOSPs (Fig. 2) since WNV emergence in North America in the late 1990s. Conversely, the founding East Coast genotype (NY99) has demonstrated a reduced capacity for eliciting infectious titers in HOSPs over time (Fig. 4). Taken together, these observations indicate a cyclic pattern of adaptive selection acting on WNV and avian hosts, suggestive of the ‘Red Queen’ hypothesis of evolution [37]. Mortality and fitness effects of high replication of the founding strain of WNV in HOSPs in North America may have served as a significant selective pressure for increased control of WNV replication in HOSPs that, in turn, may have selected for viral adaptations to increase viremia and therefore transmissibility to mosquitoes. Ultimately, because HOSPs inoculated with WNV only have decreased viremias in response to East Coast viruses that are no longer circulating, the consequence of viral evolution has been an increase in reservoir competence of HOSPs from 1.6 in 2000 to 2.6 in 2013 for extant WNV genotypes (Fig. 3). The Red Queen hypothesis would predict that HOSPs will further adapt to sustain lower viremias in response to WN02 and SW03 genotypes, with corresponding viral mutations selected in order to offset avian antiviral effects.
The variation in titers that we observed in HOSPs may be unrelated to selective pressures acting on the virus and host. However, WNV has been a significant cause of death for HOSPs in the U.S., and HOSP abundance has decreased in response to WNV infection [16]. Overall, HOSP abundance in the rural U.S. has decreased significantly by an average of 3.1% per year from 1999 to 2012 (p<0.05) [38]. The proportion of population decrease that is due to WNV infection is unknown, but the results from this study suggest WNV may have contributed to population decline recently due to higher WNV-induced mortality. Accordingly, selection acting on HOSPs to mediate lower WNV-induced mortality is a plausible explanation. One potential mechanism of increased survival in HOSPs is a better regulation of viral titers, modulated by a change in the innate immune response to WNV. As WNV is known to antagonize the host interferon response [39]–[44], the host may be able to modulate viral titers by evasion of viral antagonism. Sequencing of HOSP innate immune genes from archival samples may reveal genetic differences between individuals demonstrating variable viremias and mortality profiles.
Based on the subsequent increase in viremias induced by contemporary WNV strains in HOSPs, it is possible that HOSPs have exerted a selective pressure on WNV that contributed to the emergence of the WN02 and SW03 genotypes. The observed increase in viremia in HOSPs also may be the consequence of a general viral adaptation to mosquitoes, multiple avian species, or a specific viral adaptation to another avian species in North America, such as the American robin (AMRO; Turdus migratorius), that is thought to be the most preferred host for mosquito bloodmeals [45]. To test whether AMROs may have driven the evolution of WN02 and SW03 genotypes, similar experimental inoculations with viral isolates collected during different years would need to be performed. However, it is unlikely that WNV has adapted to American crows (AMCRs), as all North American WNV strains are uniformly pathogenic to AMCRs due to the conserved proline at NS3-249 in North American isolates [46]. Although pathogens are generally assumed to evolve towards decreased pathogenicity in a susceptible host, there are examples of short-term increases in pathogen virulence in birds, such as the emerging bacterium Mycoplasma gallisepticum in wild house finches (Haemorhous mexicanus) [47]. For example, higher host mortality, which increases mosquito transmissibility of WNV, may increase viral spread by reducing flock immunity [48].
HOSPs were introduced into North America in the 1850s [49]. Thus, divergence between New and Old World HOSP populations is likely, and these experiments with North American HOSPs may not be consistent with other geographically distinct HOSP populations. Interestingly, amino acid variation at the E-159 locus has been observed in Old World Lineage 1A WNV isolates several times prior to the introduction of WNV to North America [50], as well as Lineage 2 WNV strains [51], but no other lineage has acquired an alanine at this position. Since HOSPs are prevalent in Europe and Africa and would likely serve as important avian hosts, it is possible that other substitutions at E-159 are beneficial to viral replication in distinct HOSP populations. The NS4A-85 locus is also hyper-variable among WNV strains (Table 4), with a threonine, valine, or isoleucine present in other Lineage 1 and 2 viruses. The NS2A-R188K mutation that was associated with higher peak viral titers and mortality in inoculated HOSPs presented herein (Fig. 2, Tables 3–4) is also found in Lineage 2 viruses. Although statistical analyses of WNV evolution do not identify these sites as the targets of diversifying selection, the variability of these sites across lineages combined with observed phenotypic effects in HOSPs suggests they may be adaptive changes.
The evolution of WNV strains that have increased the magnitude and duration of viremias in HOSPs highlights the potential importance of HOSPs for the maintenance of WNV in North America. Furthermore, declining WNV-induced mortality in HOSPs infected with an early WNV strain suggests that WNV has imparted a significant selective pressure on wild bird populations. This evidence of virus-host co-evolution suggests that the competence of North American birds for WNV is likely to continue to change.
|
10.1371/journal.pntd.0007068 | Very severe tungiasis in Amerindians in the Amazon lowland of Colombia: A case series | Tungiasis is a parasitic skin disease caused by penetrating female sand fleas. By nature, tungiasis is a self-limiting infection. However, in endemic settings re-infection is the rule and parasite load gradually accumulates over time. Intensity of infection and degree of morbidity are closely related.
This case series describes the medical history, the clinical pathology, the socio-economic and the environmental characteristics of very severe tungiasis in five patients living in traditional Amerindian communities in the Amazon lowland of Colombia. Patients had between 400 and 1,300 penetrated sand fleas. The feet were predominantly affected, but clusters of embedded sand fleas also occurred at the ankles, the knees, the elbows, the hands, the fingers and around the anus. The patients were partially or totally immobile. Patients 1 and 3 were cachectic, patient 2 presented severe malnutrition. Patient 3 needed a blood transfusion due to severe anemia. All patients showed a characteristic pattern of pre-existing medical conditions and culture-dependent behavior facilitating continuous re-infection. In all cases intradomiciliary transmission was very likely.
Although completely ignored in the literature, very severe tungiasis occurs in settings where patients do not have access to health care and are stricken in a web of pre-existing illness, poverty and neglect. If not treated, very severe tungiasis may end in a fatal disease course.
| Tungiasis (also called sand flea disease) is a neglected tropical disease (NTD) caused by the penetration of female sand fleas in the skin, typically at the toes, the sole or the heel. Once embedded in the upper strata of the skin, the parasite hypertrophies, enlarging its body size by a factor of 2000 within ten days. This causes intense inflammation with pain and itching, eventually leading to impaired mobility. During a period of three weeks, eggs are expelled through a tiny opening in the skin. When the last egg has been released into the environment, the parasite shrinks and eventually dies. Hence, by nature tungiasis is a self-limited infection. However, in endemic settings re-infection is the rule and parasite load gradually accumulates over time. Here we report five cases with extremely severe tungiasis in patients with 400 to 1,300 embedded sand fleas. Not only the feet were affected, but clusters of parasites also occurred at the ankles, the knees, the elbows, the hand, the fingers and around the anus. The patients were partially or totally immobile. Two patients were cachectic and one required a blood transfusion. All patients showed a characteristic pattern of pre-existing medical conditions and culture-related behaviour facilitating continuous re-infection.
| Tungiasis (sand flea disease), one of the most neglected tropical diseases (NTDs), is caused by female sand fleas (Tunga penetrans and more rarely T. trimamillata) penetrated into the skin. The disease is prevalent in resource-poor communities in South America, the Caribbean, sub-Saharan Africa and Madagascar [1–5]. Children, the elderly and persons with disabilities bear the highest disease burden [6, 7]. Intensity of infection varies widely and correlates to severity of disease [1, 8]. Tetanus is a known sequel in individuals with insufficient immunization [3, 9].
In endemic communities light infections with 5 to 10 embedded sand fleas predominate. [8]. Reports on very severe tungiasis with a hundred or more embedded sand fleas are scanty in the current literature [10–12]. In contrast, in publications from the end of the 19th to the middle of the 20th century, very severe tungiasis was frequently reported [13–26]. Here we report five cases of very severe tungiasis in inhabitants of four Amerindian communities in Vaupés Department, in the Amazon lowland of Colombia. The patients shared a spectrum of pre-existing medical conditions facilitating constant re-infection and presented socio-economic and environmental characteristics which together influenced the development of tungiasis into a life-threatening condition.
Vaupés Department is situated in the Southeast part of Colombia and covers an area of 54,134 km2. Geographically, it belongs to the Amazon basin and is almost completely covered with dense rain forest. The Río Vaupés is an affluent of the Río Negro, a major tributary to the Amazon River. Vaupés Department is inhabited by about 35,000 people of whom 12,000 live in the capital Mitú. At least 220 Amerindian communities are spread all over the department but only a minority can be accessed by boat or small aircraft. Rarely, communities have more than 200 inhabitants. People live from fishing, hunting and collection of edible plants in the forest and subsistence cultivation of cassava (Manihot esculenta).
The only hospital of the department is located in Mitú. A couple of primary health care centers are dispersed in the municipalities of Carurú and Taraira. They serve communities which can be reached by forest tracks or boat.
During a period of 12 weeks, five patients with very severe tungiasis were observed. Four patients were seen at the emergency unit of Mitú hospital, one in the community she was living in. Patient 1 and 2 were from Wacará (N01°14ʹ45.19ʺ, W07°00ʹ37.20ʺ, patient 3 from Nuevo Pueblo (N00°51ʹ55.01ʺ, W69°33ʹ52.01ʺ), patient 4 from Los Angeles (N0°34ʹ26.31ʺ, W70°07ʹ28.98ʺ), and patient 5 from Puerto Pinilla (N0°55ʹ39.30ʺ, W69°57ʹ33.15ʺ).
Patient 1 was completely immobile and had to be carried in a hammock from her community to the Vaupés River for six hours and from there she was transported by boat to Mitú. Patient 2 had considerable pain while walking and was hobbling slowly on the lateral rim of his feet. Patient 3 and 4 were completely immobile. They were carried in a hammock to a small airstrip by community members and then transported by aircraft to Mitú. Patient 5 was found living in an isolated place at a small tributary of the Vaupés River. As the access was extremely difficult and her condition was relatively good, she was examined and treated at home.
Patients were undressed, the whole body was washed and carefully examined for the presence of embedded sand fleas. The skin was also inspected for signs of bacterial and fungal infection. Severity of tungiasis was determined using a previously established score [27]. Staging was performed according to the Fortaleza classification [28].
Immediately after the examination, the patients were treated topically with NYDA, a formula containing two dimeticone (silicone) oils with low viscosity (Pohl-Boskamp GmbH & Co. KG, Hohenlockstedt, Germany) [29]. Due to its physical mode of action, NYDA is registered as a class II medical device [29]. Treatment with dimeticones is considered as the reference treatment of tungiasis by the Ministry of Health and Social Protection of Colombia.
Affected body areas were carefully wetted with the dimeticone. In areas with hyperkeratotic skin and several layers of sand fleas situated on top of each other, the oil was vigorously rubbed into the skin. The treatment was repeated after 24 hours. In patient 4, an additional application was made 1 week after the initial treatment.
The tetanus-vaccination status was verified and patients were vaccinated against tetanus if necessary. Because of severe anemia, patient 3 received 2 × 250 ml red blood cell concentrate. Patient 1 was treated with oxacillin intravenously (twice one gram per day for nine consecutive days) due to severe bacterial superinfection of lesions at the feet. Patient 2, 4 and 5 were treated with albendazole (400 mg per day for three days) and tinidazole (three tablets of 500 mg per day for two days) to eliminate intestinal helminths. Patient 3 was treated with metronidazole (500 mg every 12 hours for five days), oxacillin (twice one gram per day for nine days), gentamycin intravenously because an infection with gram-negative bacteria was suspected (160 mg every six hours for seven days) and albendazole (400 mg per day for three days). Patients were monitored for up to 15 days and changes in their clinical condition were documented.
Since the patients did not speak Spanish, questions were translated by health assistants speaking the same language as the patients.
The study was performed as part of routine health care provided by public health personnel of Vaupés Health Department in Mitú Hospital. The examination of the skin for the diagnosis of tungiasis is part of the routine health care and was carried out with the aim to cure patients from a life-threatening condition. All patients provided oral consent. The objective of the examination as well as the risks and benefits of treatment were explained to each patient and relatives/caregivers present. The examination of minors was made with the authorization and in the presence of at least one of their parents. In accordance with Resolution 008430 of 1993, of the Ministry of Health and Social Protection of Colombia, which regulates research in humans, the study is classified as a low risk study.
Demographic, socio-economic and clinical characteristics are depicted in Table 1. Patient 1 (female, 72 years) lived together with patient 2, her grandson, in a small hut without a solid floor. She was suffering from gonarthrosis for long. Since she could not work anymore, she rested day and night in her hammock. Food was provided by her son once a day, but food shortage was common. Several dogs belonged to the household.
Patient 2 (male, 16 years) was the grandson of patient 1. He suffered from bilateral deafness and mental retardation since birth and had never left the village he was born in. He was unable to take care for himself or to accompany his father for hunting or gathering food. Instead he waited the whole day inside the hut, crouching on his heels or directly squatting on the ground next to his grandmother. The only piece of clothing he had were torn shorts.
Patient 3 (male, 69 years) lived in a very remote community located near the frontier with Brazil. He belonged to the Iupdah-Maku ethnicity, a group of Amerindians who only recently became sedentary. Iupdah people entirely live from hunting and edible fruits they collect in the rain forest. The patient was suffering from gonarthrosis and therefore was unable to walk. He entirely depended on food provided by relatives. His daughter, who had taken care of him, had moved away to another community a couple of months ago. The patient was left alone in a shelter without walls, where he spent the whole day in the hammock. Two dogs were his only companions.
Patient 4 (male, 81 years) lived in a small community located at the Brazilian border. He belonged to the Tuyuca ethnic group. He suffered from gonarthrosis since long and was cared for by his daughter. However, the daughter had moved away some time ago. The eldest son should have taken care of the patient, but was unable to provide sufficient food even for his own family.
Patient 5 (female, 94 years) lived in an isolated dwelling at a small tributary of the river Vaupés. She belonged to the Siriana ethnic group and lived with her oldest son, who had no wife. Her mobility and vision were restricted. A week before the patient was identified by the medical team, she had been treated by a relative who had applied a plant extract of unknown origin on the feet. This had reduced the number of viable lesion from around 1000 to about 50.
Patient 1 and 3 had an extremely severe form of tungiasis with approximately 1,000–1,300 embedded sand fleas in all stages of development. The soles and the lateral rims of both feet were covered with several layers of embedded sand fleas on top of each other and closely-packed (Figs 1 and 2). Clusters of embedded sand fleas existed at the ankles, lower legs, knees, at the elbows and around the anus (Figs 3 and 4). The palm, the back of the hand and the fingers were also affected (Figs 5 and 6). The feet emitted a strong odor of necrotizing flesh. Patient 1 and 3 were anemic, dehydrated and cachectic. Their weight was 35 kg and 39 kg, respectively. Patient 1 was intensely infested with head lice and also had myiasis at the right foot.
Patient 2 had approximately 250 embedded sand fleas in all stages of development of which the great majority had penetrated at the feet. The density of parasites was particularly high in the interdigital area of the soles (Fig 7). A small cluster of embedded sand fleas was detected around the anus. Most of the sand fleas were viable. The patient weighed only 23 kg.
Patient 4 had approximately 400 embedded sand fleas in all stages of development. Lesions were located on the soles, the lateral rim of the feet, ankles, knees, elbows and hands. On the soles lesions occurred in three layers on top of each other.
Patient 5 had approximately 1,000 lesions, of which 950 consisted of decaying or dead sand fleas at the time of examination.
The laboratory findings are summarized in Table 2. Patient 1 and 3 had a severe anemia. Patient 2 had a leukocytosis of 12,700 cells/μl. Patient 2 and 4 showed a hypereosinophilia (2,540 and 2,520 eosinophils/μl, respectively).
After two applications of the dimeticone oil, patients recovered rapidly. After three to four days, inflammation of the skin had regressed considerably (Fig 8) and patients could place their feet on the ground without feeling pain. After one week all lesions had developed into crusts and patients were transferred to a rehabilitation centre for Amerindians at the periphery of Mitú. At the end of the rehabilitation, i.e. 15 to 20 days after the first treatment with dimeticone, the patients had increased their weight Patient 1: from 35 to 41 kg; patient 2: from 23 to 28 kg; patient 3: from 39 to 45 kg; patient 4: from 33 to 41 kg (not data available from patient 5).
Literature from the middle of the 19th to the middle of the 20th century abounds with descriptions of very severe cases of tungiasis, in Latin America, the Caribbean as well as in sub-Saharan Africa [13–26, 30–34]. Since the 1950ties descriptions of very severe tungiasis became scanty. The few published cases concerned patients with alcohol addiction or psychiatric disorders who–at least from time to time–had been laying on the ground for hours [10, 12, 35, 36]. In current textbooks, very severe tungiasis is not mentioned at all.
This case series shows that very severe tungiasis still occurs, but goes unnoticed by health care providers because the patients are living in remote settings in the hinterland and do not have access to health care. The short period in which the five patients were identified indicates that very severe tungiasis is expected to occur in many Amerindian communities. Although this case series is small, it identified a pattern of characteristics which together determine that a self-limiting skin infection develops into a life threatening disease. Understanding of these determinants of very severe tungiasis allowed the development of a surveillance scheme by the Ministry of Health and Social Protection of Colombia.
It has been shown that age-specific prevalence curves show a peak in children and adults > 60 years and that the elderly always bear a high disease burden [1, 37]. Due to age-related poor sight elder people are unable to identify exactly where the parasite is located, and by consequence need help from family members to remove embedded sand fleas with a sharp instrument, the traditional treatment still in use in Amerindian communities [11]. Even if vision is adequate, elder people will usually be unable to bend down to the feet. Besides, elder people are less mobile and tend to stay at intradomiciliary transmission sites for many hours of the day. Another characteristic is that elder people frequently live alone and need to be cared for by relatives.
Patient 1 and 2 were wholly dependent on food provided by the only healthy adult person in the household. Patient 3, 4 and 5 were taken care for by their daughters. When the daughter left the community, the elder sons barely managed to provide food for their own family and usually nothing was left for the patients.
In Amerindians of the Amazon basin it is a question of survival for the family that old or handicapped people, who cannot provide food for themselves, become gradually separated from their families and are no longer cared for. This explains the malnutrition and cachexia observed in patients 1, 2 and 3.
Medical conditions which cause patients to involuntarily spend many hours in direct contact with the soil, such as sleeping sickness, mental disorders, alcoholism or Klippel-Trenaunay-Syndrome, are known since long as factors predisposing to very severe tungiasis [38, 39]. The same holds true when patients do not perceive pain or itch as in leprosy [38].
Patient 2 exposed skin of the buttocks for many hours of the day when crouching on his heels or squatting directly on a dirt floor in his ragged shorts. Persistently exposing skin at other areas than the feet facilitates the penetration of sand fleas at ectopic sites, such as around the anus and the inguinal area [13, 14, 22, 40].
Four patients had a pre-existing medical condition which restricted their mobility and left them laying in their hammock most time of the day. In the Amazon lowlands hammocks are installed such that they swing 20–40 cm above the floor. People rest in the hammock with a hand, an elbow, the lateral rim of the foot and/or the knee outside very close to the floor and place these body areas from time to time on the ground. By consequence, rather large areas of the skin are exposed to sand fleas which, in turn, explains the high parasite burden and the ectopic localizations (Figs 1B, 1C, 2B and 3B) [41, 42]. Since tungiasis of the feet impairs mobility due to severe pain, a vicious cycle develops in embedded sand fleas accumulating over time eventually resulting in very severe tungiasis and total immobility, if patients do not have access to treatment.
Another predisposing factor for a continuous accumulation of a high parasite load is intradomiciliary transmission. In the setting of the five patients, intradomiciliary transmission was very likely: In Amerindian communities, a fire is lit just below the hammock to warm up the sleeping person the whole night. This makes the surrounding earth floor dry and causes cracks in the soil, a perfect niche for the completion of the off-host cycle of T. penetrans. Dogs are frequently infected with T. penetrans and usually spend the night inside the dwelling next to the fire place [11, 13, 36]. Thus, eggs expelled from embedded sand fleas develop into adults in the area directly below the hammock or next to it. Hence, the probability to get infected is high as soon as the feet of a person are placed on the ground.
In Vaupés Amerindians strongly believe that severe tungiasis is due to an oath (maldad or incantation) and, hence, they are convinced that the disease cannot be cured. By consequence, people think that the afflicted person will be devoured by the parasite sooner or later and that only a traditional healer (payé) can interrupt this process. This explains the attitude that community members think that it is better to stay away from the affected person and why the patients of this study were secluded by the community.
The web-of-causation making tungiasis a life-threatening condition is depicted in Fig 9.
Lack of knowledge of the pathogenesis of tungiasis was identified as the leading cause of very severe disease and death in colonial times when Spanish and Portuguese conquistadors penetrated into the interior of the yet unknown continent and were confronted with pathogens they had never met before [43]. The scenario was similar in Africa, after T. penetrans arrived in Angola in 1875 and then rapidly spread along trading routes and with military missions [14, 17, 21, 23, 24, 33]; in natives of Madagascar after the disease all of a sudden appeared on the island [18, 25]; in troops deployed for the first time in an endemic area, such as French soldiers in Mexico in 1862 [16, 31].
That the lack of knowledge of tungiasis rapidly leads to very severe disease and eventually to death is best demonstrated by the history of 100 Irish settlers who moved from the Coast of French Guyana to the interior of the country in 1852. They became so heavily infected with sand fleas that 70 settlers died within a year and the remaining 30 came back to the coast in a very debilitated condition [25]. Lack of knowledge also seemed to play a role when tungiasis spread in Central Africa in the 1920s [33, 34]. However, lack of knowledge does not seem to play a role in the five patients. Tungiasis is known since pre-colonial times in Amerindians and local people clearly understand the pathogenesis of the disease [44].
The spectrum of pathology associated with very severe tungiasis and its consequences are known since long. Already in 1900 L.L. Decle, a physician of the British army noted: “Ulcers caused by Tunga penetrans were the most frequent ailment treated and second only to smallpox as a cause of death in Luanda’s hospital between July and October 1877…” “Never in my life have I seen such awful ulcers. Some of the men had their bone of their big toe protruding fleshless for more than an inch; others had quite a square inch of the bone of the heel exposed. Even when tungiasis did not cause death, the parasite paralyzed movement.” The author continued:”In some villages of Uduhu, I found the people starving, as they were so rotten with ulcers from jiggers that they had been unable to work in their fields, and could not even go to cut the few bananas that had been growing.” [20]
Morbidity associated with tungiasis and its sequels is depicted in Table 3.
The patients in this study showed a few peculiar findings. Patient 2 and 3 had a cluster of lesions around the anus, an ectopic localization which has never been described before.
Patient 1 and 3 showed a severe anemia which required immediate blood transfusion in patient 3. There is reason to believe that the anemia was the consequence of the high parasite load. First, severe infestation with Ctenocephalides felis in animals causes a significant anemia [45, 46]. Second, in contrast to most other Siphonaptera, which are only temporary ectoparasites, T. penetrans sucks blood almost permanently. Third, significant anemia was present in a very severe case of tungiasis from Tanzania with about 1,100 lesions [42]. With a persistent high parasite burden, the chronic blood loss will be substantial over time and eventually results in life-threatening anemia. Of course, the co-existing hookworm infection also may have contributed to anemia.
Hitherto, the only available treatment of tungiasis in Amerindian communities is surgical removal of embedded sand fleas using inappropriate instruments such as thorns, sharpened wooden sticks, knives etc. It goes without saying that this procedure is painful and always bears the risk of bacterial or fungal superinfection of the sore. Even in hospital settings, surgical extraction is virtually impossible if a great number of sand fleas are located in clusters or on top of each other in hyperkeratotic skin as in our patients.
The medical device NYDA contains two dimeticones with different viscosity and rapidly enters into tiny openings and covers microscopic surfaces [47]. The formula is used for the treatment of headlice in more than 20 countries. Its mode of action is purely physical. The topical application of this product has been proved highly effective in randomized control trials in Kenya and Uganda in patients with up to 30 embedded sand fleas [29, 48]. Here we show that the formula is also effective in patients with several hundred of embedded sand fleas located in clusters and in several layers on top of each other in hyperkeratotic or partially necrotizing skin. While in simple tungiasis an application of a few drops of the dimeticone—targeted to the abdominal cone which protrudes through the skin–are sufficient [48], in very severe tungiasis the skin needs to be intensively wetted and treatment should be repeated after 24 hours.
Taken together, this case-series shows that very severe tungiasis still occurs in Amerindian communities. The true frequency of this devastating condition is probably underestimated. A characteristic pattern of pre-existing medical conditions and socio-economic and environmental factors determines whether tungiasis develops into a life-threatening condition. Obviously, most of these factors are related to extreme poverty.
Our findings are also a good argument to make a call for action for those countries in which tungiasis occurs in remote settings and where health coverage is poor. Dimeticone should be made available to treat patients in an early stage of disease to avoid life-threatening sequels.
|
10.1371/journal.pcbi.1004273 | Evolution of Self-Organized Task Specialization in Robot Swarms | Division of labor is ubiquitous in biological systems, as evidenced by various forms of complex task specialization observed in both animal societies and multicellular organisms. Although clearly adaptive, the way in which division of labor first evolved remains enigmatic, as it requires the simultaneous co-occurrence of several complex traits to achieve the required degree of coordination. Recently, evolutionary swarm robotics has emerged as an excellent test bed to study the evolution of coordinated group-level behavior. Here we use this framework for the first time to study the evolutionary origin of behavioral task specialization among groups of identical robots. The scenario we study involves an advanced form of division of labor, common in insect societies and known as “task partitioning”, whereby two sets of tasks have to be carried out in sequence by different individuals. Our results show that task partitioning is favored whenever the environment has features that, when exploited, reduce switching costs and increase the net efficiency of the group, and that an optimal mix of task specialists is achieved most readily when the behavioral repertoires aimed at carrying out the different subtasks are available as pre-adapted building blocks. Nevertheless, we also show for the first time that self-organized task specialization could be evolved entirely from scratch, starting only from basic, low-level behavioral primitives, using a nature-inspired evolutionary method known as Grammatical Evolution. Remarkably, division of labor was achieved merely by selecting on overall group performance, and without providing any prior information on how the global object retrieval task was best divided into smaller subtasks. We discuss the potential of our method for engineering adaptively behaving robot swarms and interpret our results in relation to the likely path that nature took to evolve complex sociality and task specialization.
| Many biological systems execute tasks by dividing them into finer sub-tasks first. This is seen for example in the advanced division of labor of social insects like ants, bees or termites. One of the unsolved mysteries in biology is how a blind process of Darwinian selection could have led to such highly complex forms of sociality. To answer this question, we used simulated teams of robots and artificially evolved them to achieve maximum performance in a foraging task. We find that, as in social insects, this favored controllers that caused the robots to display a self-organized division of labor in which the different robots automatically specialized into carrying out different subtasks in the group. Remarkably, such a division of labor could be achieved even if the robots were not told beforehand how the global task of retrieving items back to their base could best be divided into smaller subtasks. This is the first time that a self-organized division of labor mechanism could be evolved entirely de-novo. In addition, these findings shed significant new light on the question of how natural systems managed to evolve complex sociality and division of labor.
| The “major transitions in evolution”, whereby cells teamed up to form multicellular organisms or some animals went on to live in societies, are among the keys to the ecological success of much life on earth [1]. The efficiency of both organisms and animal societies frequently depends on the presence of an advanced division of labor among their constituent units [2–4]. The most celebrated examples can be found in social insects, which exhibit astonishing levels of social organization and are ecologically dominant in many natural ecosystems [5,6]. Through division of labor, social insects can perform complex tasks by dividing them up into smaller sub-tasks carried out by different sets of individuals [7–10]. Although the adaptive benefits of division of labor are evident, the way in which it can evolve is more enigmatic, since an effective division of labor requires the simultaneous co-occurrence of several complex traits, including self-organized mechanisms to decompose complex tasks into simpler subtasks, mechanisms to coordinate the execution of these tasks, mechanisms to allocate an appropriate number of individuals to each task, and the ability of individuals to effectively carry out each of the subtasks [4]. The complexity of this co-evolutionary problem is further exacerbated by the fact that division of labor should also be flexible to be able to cope with changing environmental conditions [4,10,11].
To date, most analytical and individual-based simulation models of division of labor [4,9,10,12–16] have focused merely on determining the optimal proportion of individuals engaging in different tasks [12] or on determining optimal task allocation mechanisms [4,9,10,13,16], sometimes in relation to particular levels of intragroup genetic variation [14,15]. These studies implicitly assume that pre-optimized behaviors to carry out each of the different subtasks, which we refer to as “pre-adapted behavioral building blocks”, are already present in nonsocial ancestors [17], and that division of labor merely involves the rewiring of these behaviors. Empirical support for this hypothesis can be found for example in the somatic cell differentiation in multicellular organisms, which is derived from a genetic switch involved in the induction of diapause during stress periods in unicellular ancestors [2,18]. Similarly, in insect societies, worker brood care is thought to be derived from ancestral parental care [19], and reproductive division of labor as well as worker task specialization may be derived from mechanisms that initially regulated reproduction and foraging in solitary ancestors [17,20–22].
A limitation of traditional analytical modeling approaches to division of labor [4,10], however, is that they can only consider a finite and pre-specified number of behavioral strategies. In recent years, artificial evolution of teams of embodied agents has been used to enable the study of social traits in more detail, taking into account more realistic physical constraints and a much larger set of allowable behaviors and strategies [23–25]. In evolutionary swarm robotics, for example, this framework has been used to study the evolution of the origin of communication [26,27], collective transport [28], collective motion [29], aggregation [30–32] and chain formation [33] (reviewed in [23,24,34–37]). Nevertheless, to date, no study in evolutionary swarm robotics has succeeded in evolving complex, self-organized division of labor entirely de novo [38,39]. This may be due to the fact that most evolutionary robotics studies have made use of neural network-based approaches [23–25,36], which have been shown to scale badly to more complex problems [38,40].
The main aim of our study was to test if other nature-inspired evolutionary methods than traditionally used in evolutionary swarm robotics would be able to achieve complex task specialization in social groups. Analogously to the situation in nature where subtask behaviors may or may not be recycled from pre-adapted behavioral building blocks, we do this using one of two approaches, in which we either do or do not pre-specify the behaviors required for carrying out the different subtasks. Evidently, we expected that task specialization could evolve much more easily when pre-adapted behavioral building blocks were present, but we were also interested to see if a self-organized mechanism of task specialization could be evolved entirely de-novo using our recently developed method of Grammatical Evolution [41]. This nature-inspired evolutionary method allows a set of low-level behavioral primitives to be recombined and evolved into complex, adaptive behavioral strategies through the use of a generative encoding scheme that is coupled with an evolutionary process of mutation, crossover and selection [41].
The type of division of labor we consider in our set-up is known as “task partitioning”, and requires different tasks to be carried out in sequence by different sets of individuals [7]. In particular, our experimental scenario was inspired by a spectacular form of task partitioning found in some leafcutter ants, whereby some ants (“droppers”) cut and drop leaf fragments into a temporary leaf storage cache and others (“collectors”) specialize in collecting and retrieving the fragments back to the nest [42,43] (Fig 1). In our analogous robotics setup, we used a team of robots [44] simulated in-silico using an embodied swarm robotics simulator [45] (Fig 2) and required the robots to collect items and bring them back to the nest in either a flat or sloped environment (see Fig 1 and Fig 2B and Material and Methods). In this setup, task specialization should be favored whenever some features of the environment (in our case, the presence of a slope) can be exploited by the robots to achieve faster foraging (“economic transport”, [46]) and reduce switching costs between different locations [9,47]. The results of these experiments show for the first time that complex, self-organized task specialization and task allocation could be evolved in teams of robots. Nevertheless, a fitness landscape analysis also demonstrates that task specialization was much easier to evolve when pre-evolved behavioral building blocks were present. We use these findings as a starting point to speculate about the likely path that nature took to evolve complex sociality and division of labor. Furthermore, we discuss the potential of our nature-inspired evolutionary method for the automated design of swarms of robots displaying complex forms of coordinated, social behavior.
Our experimental setup is inspired by the type of task partitioning observed in Atta leafcutter ants [42,43], that collect leaves and other plant material as a substrate for a fungus that is farmed as food (Fig 1A). In these insects, particularly in species that harvest leaves from trees, leaf fragments are retrieved in a task partitioned way, whereby some ants (“droppers”) specialize in cutting and dropping leaf fragments to the ground, thereby forming a leaf cache, and others specialize in collecting leaves from the cache to bring them back to the nest (“collectors”) [42,43]. In addition, another strategy is known whereby the whole leaf cutting and retrieval task is carried out by single individuals (“generalists”), without any task partitioning [42,43]. Task partitioning in this scenario is thought to be favored particularly in situations where the ants forage on leaves from trees, due to the fact that the leaf fragments can then be transported purely by gravity, which saves the ants the time to climb up and down the tree, and the fact that there are few or no costs associated with material loss thanks to the large supply of leaves [7,43,48] (Fig 1A). This theory is supported by the fact that species living in more homogeneous grassland usually retrieve leaf fragments in an unpartitioned way, without first dropping the leaves (Fig 1C), particularly at close range to the nest [43,49].
In the corresponding robotic setup, we substituted the tree with a slope area and leaves with cylindrical items. A team of robots then had to collect these items from what we call the source area and bring them back to what we refer to as the nest area (Fig 1B). Simulations were carried out using the realistic, physics-based simulator ARGoS [45]. As demonstrated in the past, controllers developed within ARGoS can be directly transferred to real robots with minimal or no intervention [50,51]. The robots involved in the experiments were a simulated version of the foot-bot robot, a version of the MarXbot robot [44], which is a differential-drive, non-holonomic, mobile robot (Fig 2A). A screen-shot of a simulation instant is shown in Fig 2B. We used a setup whereby 5 items were always present in the source area. The 5 items were replaced and put in a random position within the source area each time a robot picked up one of them. This is justified by the fact that leaf availability in the natural environment is often virtually unlimited. A light source was placed at a height of 500 m, 500 m away from the nest, in the direction of the source area. The light allowed the robots to navigate in the environment, since phototaxis allowed them to go towards the item source, whereas anti-phototaxis allowed them to return to the nest. The slope area had an inclination of about 8 degrees. The linear velocity of the robots on the flat part of the arena was 0.15 m/s, but this reduced to a maximum speed of 0.015 m/s when they had to climb up the slope, and increased to 0.23 m/s when they came down from the slope. If an item was dropped in the slope area, it slid down the slope at a speed of 1 m/s until it reached the cache area, where it was stopped due to friction and to the impact with other items in the cache. This was done to simulate leaves being dropped from the tree, as in Fig 1A. In addition, in some of the experiments, we considered a flat environment of the same length and width as the one described above (Fig 1D), to mirror the case in nature where ants forage in a flat, homogeneous environment (Fig 1C).
In a first set of experiments, we assumed that the behavioral strategies required to carry out each of the subtasks (dropper or collector behavior, as well as generalist, solitary foraging) were available to the robots as pre-adapted behavioral building blocks and then determined the optimal mix of each of the strategies [12]. This setup, therefore, matched some evolutionary scenarios proposed for the origin of division of labor in biological systems based on co-opting pre-adapted behavioral patterns [2,17–22]. In addition, this scenario allowed us to determine under which environmental conditions task partitioning is favored, and provided a fitness benchmark for the second scenario below, where task partitioning was evolved entirely de-novo.
In this first set of experiments, dropper, collector and generalist foraging strategies were implemented as follows:
Dropper strategy: A dropper robot is a robot that climbs the slope area and never descends it again, continuously collecting items from the source area and dropping them to the slope area.
Collector strategy: A collector robot is a robot that never climbs the slope area. Instead, it continuously collects items from the cache (when present) and brings them back to the nest. If it cannot find any items, the collector robot keeps exploring the cache area by performing random walk, until an item is found.
Generalist strategy: A generalist robot is a robot that performs a standard foraging task. It climbs the slope and explores the source area, collects items, and brings them all the way back to the nest. The generalist robot does not explore the cache area, but in case it finds an item at the cache while going towards the source, it collects it and brings it back to the nest.
The rules that we employed to implement these strategies are shown in S1 Table. We also assumed that the robots would specialize for life in each of these available strategies according to a particular evolved allocation ratio. This was equivalent to assuming that in nature, these behavioral strategies would already have evolved due to selection in their ancestral environment, and that natural selection would favor a particular hard-wired individual allocation between the different sets of tasks, e.g. through fine-tuning of the probability of expression of the gene-regulatory networks coding for the different behavioral patterns. For these experiments, we used teams of 4 robots, to match the evolutionary experiments with fine-grained building blocks (cf. next section). Subsequently, a fitness landscape analysis was used to determine the optimal mix between the three strategies in one of two possible environments, a flat or a sloped one (Fig 1B and 1D). This was done via exhaustive search, that is, by testing all possible ratio combinations and determining the corresponding fitness values in the two environments, rather than using an evolutionary algorithm. This was possible due to the relatively small search space, which gave access to the full fitness landscape. Group performance, measured by the total number of items retrieved to the nest over a period of 5,000 simulated seconds, for each possible mix of the three strategies, was measured in 10 simulated runs and then averaged.
In a second set of experiments, we considered an alternative scenario where both task specialization and task allocation could evolve entirely de-novo, starting only from basic, low-level behavioral primitives. These primitives were simply navigational behaviors allowing robots to either go towards the source or towards the nest, as well as a random walk behavior:
PHOTOTAXIS: uses the light sensor to make the robot go towards the direction with the highest perceived light intensity.
ANTI-PHOTOTAXIS: uses the light sensor to make the robot go towards the lowest perceived light intensity.
RANDOM WALK: makes the robot move forward for a random amount of time and then turn to a random angle, repeating this process while the block is activated, without using any sensors.
In addition, a mechanism of obstacle avoidance, based on the robot’s range and bearing and proximity sensors, was switched on at all times to avoid that the robots would drive into each other or into the walls of the foraging arena. Finally, two instantaneous actions were allowed, namely picking up and dropping an item. To be able to evolve adequate behavioral switching mechanisms, we allowed the robots to perceive their position in space, that is, whether they were in the source, slope, cache or nest, based on sensorial input from the ground and light sensors, as well as perceive whether or not they were currently holding an item.
The fine-grained behavioral building blocks were combined together using a method known as grammatical evolution [52] as implemented in GESwarm [41], in order to evolve rule-based behaviors representing more complex strategies. GESwarm was developed for the automatic synthesis of individual behaviors consisting of rules leading to the desired collective behavior in swarm robotics. These rules were represented by strings, which in turn were generated by a formal grammar. The space of strings of such a formal grammar was used as a behavioral search space, and mutation, crossover and selection were then used to favor controllers that displayed high group performance.
The individual behavior of a given robot was expressed by a set
R
composed of an arbitrary number nR of rules Ri:
R={Ri},i∈{1,…,nR}.
Each rule was composed of three components:
Ri=Pi×Bi×Ai,
where
Bi
denotes a subset of all possible fine-grained behavioral building blocks (phototaxis, anti-phototaxis and random walk),
Ai
denotes a subset of all possible instantaneous actions (pickup, drop, change behavior or change an internal state variable) and
Pi
denotes a subset of all possible preconditions. The preconditions were specified as logical conditions with respect to the current value of a number of state variables, which included both sensorial input (the environment they were in and whether or not they were carrying an item) and internal state variables (a state variable that specified whether they wanted to pick up an item or not and two memory state variables, with evolvable meaning).
If all the preconditions in
Pi
were met, and if a given robot was executing any of the low-level behaviors present in
Bi
, all actions contained in
Ai
were executed with evolvable probability pl. In this way, we could allow the evolution of probabilistic behaviors, which have been extensively used both in the swarm robotics literature [53,54] and as microscopic models of the behavior of some social animals [55,56]. Finally, each robot executed all rules and actions in their order of occurrence.
To be able to generate the rules above, we devised a grammar using the Extended Backus-Naur Form notation [57]. Within the framework of grammatical evolution [41,52], a genotype represented a sequence of production rules to be followed to produce a valid string (in our case a set of rules) starting from that grammar. Mutation and crossover acted at the level of this genotype, modifying the sequence of production rules. The full grammar of GESwarm is described in [41].
Biologically speaking, our GESwarm rule-based controllers can be considered analogous to gene-regulatory networks or to logic circuits in the brain, and the internal memory state variables in our model can be seen as analogous to epigenetic states or memory states in the brain. Furthermore, as in biological systems, we use a generative encoding (a string coding for a series of conditional rules, similar to a DNA sequence coding for conditionally expressed gene regulatory networks) and evolve our system using mutation and crossover. One departure in our setup from biological reality, however, was that we used genetically homogeneous teams, as is common in evolutionary swarm robotics [58], but different from the situation in most social insects, where sexual reproduction tends to be the norm. This choice was made because homogeneous groups combined with team-level selection has been shown to be the most efficient approach to evolve tasks that require coordination [28]. Nevertheless, this setup can still be considered analogous to the genetically identical cells of multicellular organisms [59] or the clonal societies of some asexually reproducing ants [60] that both display complex forms of division of labor.
We executed a total of 22 evolutionary runs on a computer cluster, of which we used 100 to 200 nodes in parallel. The number 22 was chosen to meet the total amount of computational resources we had at our disposal (3 months of cluster time) and was statistically speaking more than adequate. All evolutionary runs were carried out for 2,000 generations using 100 groups of 4 robots and were each evaluated 3 times. This relatively small number of robots was chosen to limit the computational burden of the evolutionary runs. Nevertheless, we also verified if the evolved controllers could be scaled to larger teams of 20 robots each. In this case, the foraging arena was scaled in direct proportion with the number of robots. We used single-point crossover with crossover probability 0.3 and mutation probability 0.05. We chose a generational replacement with 5% elitism, in order to exploit parallel evaluation of multiple individuals on a computer cluster. We used roulette-wheel selection, that is, the probability to select a given genotype was proportional to its fitness relative to the average fitness of all genotypes in the population. As fitness criterion we used group performance, measured as the total number of items retrieved to the nest over a period of 5,000 seconds. During post-evaluation, this same fitness criterion was used to evaluate the evolved controllers. We also assessed the average absolute linear speed of the robots along the long axis of the arena, measured as a percentage of the theoretical maximum speed, and the degree of task specialization, measured as the proportion of items that were retrieved through the action of multiple robots (i.e. by task specialists).
In the first set of simulations, we assumed that each robot could specialize for life to one among the three possible preexisting behavioral strategies required for task partitioning, dropper, collector and generalist, and determined the optimal mix between the three strategies based on an exhaustive search of the full fitness landscape (Fig 1B and 1D). These simulations were performed both in a flat and a sloped environment. As proposed for natural systems [7,43,48], our a priori hypothesis was that task partitioning would be favored particularly in the sloped environment, and that maximal group performance would be achieved when some robots would specialize in dropping items in a cache and others in collecting items from the cache. This is because, in such an environment, some of the robots would be able to avoid the costly traversal of the slope area (i.e. avoid switching costs) and because gravity could also help to move items across the slope, thereby resulting in more economical transport (Fig 1).
Examination of the obtained fitness landscapes reveals that there was one globally attracting optimum in each of the two environments considered (Fig 3A and 3B). As expected, this optimum involved task partitioning in the sloped environment (Fig 3B), with a mix of 50% droppers and 50% collectors being most efficient, but only generalist foraging in the flat environment (Fig 3A, S1 and S2 Videos). In addition, our fitness landscape analysis showed that when pre-adapted behavioral building blocks can be used in the evolutionary process, the fitness landscape tends to be very smooth, thereby making task specialization easily evolvable, without the risk of the system getting trapped in suboptimal local optima. It should also be noted that in our setup, the absolute group performance was significantly higher (t-test, t = -16.6, d.f. = 18, p<10−11) in the sloped environment (144.1 ± 4.3 S.D. items collected in 5,000 s, n = 10) than in the flat one (120.2 ± 1.4 S.D. items collected in 5,000 s), due to the fact that in the first case, gravity helped to move the items towards the source.
In a second set of experiments, we used GESwarm [41] to evolve task specialization and task allocation entirely de-novo, starting only from basic, low-level behavioral primitives (see Materials and Methods). Surprisingly enough, these evolutionary experiments demonstrated that task partitioning and fully self-organized task specialization and task allocation could also emerge entirely from scratch by selecting purely on overall group performance (number of items retrieved to the nest). In particular, our experiments show that in 59% (13 out of 22) of the runs, the majority of the items were retrieved by the robots in a task-partitioned way in the final evolved controller obtained after 2,000 generations (Fig 4, S3 and S4 and S5 Videos). In these cases, most of the items were first dropped by one robot and later picked up by another one. In contrast to the case with predefined behavioral strategies, however, the task specialization that was seen in these controllers did not entail fixed roles, but instead was characterized by a dynamic allocation in response to the size of the cache. An example of a controller (nr. 2) displaying such behavior is shown in S3 Video, where the majority of the robots first exploit the source to act as droppers, but then move down the slope as the cache fills up to act as collectors (the evolved rules of this controller are shown in S2 Table). The robots shown in these simulations used simple probabilistic rules to switch from the source to the cache area, while the cache itself was exploited to switch from the cache area back to the source area. We observed that the latter mechanism was also very simple and based on stigmergy, i.e. robots would collect from the cache whenever objects were found on the way, but would continue all the way to the source when cache items were not encountered. Thanks to these simple mechanisms, the robots could dynamically switch roles in response to the size of the cache. The same adaptive specialization dynamics are apparent in Fig 5A, where the density of the robot positions across the arena is shown across the 30 runs used for post-evaluation of the same controller, and in Fig 5B, which displays the individual trajectories of the four robots in a typical evaluation run (the spatial segregation and robot trajectories for all other evolved controllers are shown in S1 Fig).
That such self-organized task specialization and task allocation could evolve from first principles by selecting purely on group performance is significant, given that we started from a random controller that barely achieved any foraging during the first few generations (Fig 4, S1 Video). As in the case without pre-adapted building blocks that we considered in the previous section, also here the presence of a slope was sufficient for the evolution of task partitioning. Indeed, when we conducted the very same experiments in a flat environment, none of the controllers evolved task partitioning and generalist foraging was the favored strategy [41].
Significantly, the evolved rules for both generalist foraging [41] and task partitioned object retrieval scaled very well also to larger teams of robots. An example is shown in S1 Video, where one of the evolved controllers from a 4 robot team is implemented in a team of 20 robots. In this case, the achieved group performance scaled almost perfectly with the increase in group size (457 ± 72 S.D. in the 20 robot team vs. 103 ± 24 S.D. in the 4 robot one). Excellent scalability properties were also shown by the fact that for the 8 best evolved controllers, the performance ratio of the rules when implemented in the 20 robot teams relative to that in the 4 robot ones in which the rules were first evolved was very close to the expected linear scaling factor of 5 (4.4, S.D. 0.14, see S3 Table).
Although the lack of fixed roles precluded an analysis in terms of behavioral roles similar to that presented in the section above, it turned out that both increased amounts of task partitioning and higher average linear speeds significantly increased group fitness (multiple regression analysis, p<0.01 and p<10−5, respectively, n = 22, Fig 6). In fact, all 8 evolved controllers displaying a high group performance (top 35%, >ca. 100 items collected) had very high levels of task partitioning (92% ± 0.08 S.D. of all items retrieved in a task partitioned way) and achieved a high average linear speed (31% ± 0.6% S.D. of the theoretical maximum). Significantly, out of these 8, the performance of the best evolved controller (135 ± 14 S.D., n = 30 items retrieved) was not significantly different from the optimal 2 dropper-2 collector mix obtained in the experiment using hand-coded behavioral strategies above (144.1 ± 4.3 S.D., t-test, t = 2.01, d.f. = 38, p > 0.05). Among these 8 best controllers, between 4 and 11 rules were used to switch between the different allowed behaviors and instantaneous actions (cf. evolved rules shown in S2 Table). Interestingly, in 3 of these best controllers, the rules employed as a precondition a memory state variable that was increased or decreased as a result of some of the actions performed in other rules. In principle, the use of these state variables could have allowed for the evolution of mechanisms akin to the response threshold model, which has been extensively used in studies on division of labor [4,9,10,16]. Nevertheless, none of our controllers succeeded in evolving this particular mechanism, and task allocation instead appeared to be based purely on probabilistic and stigmergic switching, as explained above.
A detailed analysis of the fitness and behavior of the final evolved controllers demonstrated that there was one global optimum characterized by a high level of task partitioning and high linear speed (Fig 6). Nevertheless, some runs were trapped in suboptimal regions of the search space. For example, some controllers merely displayed generalist foraging, which was suboptimal in our setup (Fig 6, bottom right points). Similarly, other controllers were characterized by defective locomotory skills, even if some actually achieved task partitioning (Fig 6, left blue points). Finally, two evolved controllers were characterized by a high degree of task partitioning and a decent speed, but nevertheless had low overall performance due to the use of a suboptimal dropping strategy, characterized by a continuous dropping and picking up in all the regions of the environment, which affected performance but not speed and degree of task partitioning (Fig 6, two blue points in the upper-right corner). These outliers, however, did not change the fact that fitness was strongly correlated with both the degree of task specialization and the linear speed of the robots.
Despite the variation in performance of the final evolved controllers, an analysis of fitness and degree of task partitioning over the course of the evolutionary runs (Fig 4) clearly demonstrates that high task partitioning was evolutionarily stable, since any transition to high task partitioning never reverted back to generalist foraging in later generations.
One of the unsolved mysteries in biology is how a blind process of Darwinian selection could have led to the hugely complex forms of sociality and division of labor as observed in insect societies [4]. In the present paper, we used simulated teams of robots and artificially evolved them to achieve maximum team performance in a foraging task. Remarkably, we found that, as in social insects, this could favor the evolution of a self-organized division of labor, in which the different robots automatically specialized into carrying out different subtasks in the group. Furthermore, such a division of labor could be achieved merely by selecting on overall group performance and without pre-specifying how the global task of retrieving items would best be divided into smaller subtasks. This is the first time that a fully self-organized division of labor mechanism could be evolved entirely de-novo. Overall, these findings have several important implications. First, from a biological perspective, they yield novel evidence for the adaptive benefits of division of labor and the environmental conditions that select for it [4], provide a possible mechanistic underpinning to achieve effective task specialization and task allocation [4] and provide possible evolutionary pathways to complex sociality. Second, from an engineering perspective, our nature-inspired evolutionary method of Grammatical Evolution clearly has significant potential as a method for the automated design of adaptively behaving teams of robots.
In terms of the adaptive benefits of division of labor and the environmental conditions that select for it, our results demonstrated that task partitioning was favored only when features in the environment (in our case a slope) could be exploited to achieve more economic transport and reduce switching costs, thereby causing specialization to increase the net efficiency of the group. Previous theoretical work has attributed the evolution of task specialization to several ultimate factors, some of which are hard to test empirically [61]. Duarte et al. [4], for example, reviewed modeling studies that showed that the adaptive benefits of a behaviorally-defined division of labor could be linked to reduced switching costs between different tasks or locations in the environment, increased individual efficiency due to specialization, increased behavioral flexibility or reduced mortality in case only older individuals engage in more risky tasks (“age polyethism”). Out of these, there is widespread agreement on the role of switching costs and positional effects as key factors in promoting task specialization [4,10,47,62], and our work confirms this hypothesis. Indeed, in our set-up, task partitioning greatly reduced the amount of costly switching required between environmental locations. Furthermore, our work also confirms the economic transport hypothesis, i.e. that task partitioning results in more economical transport, which in our case was due to the fact that gravity acted as a helping hand to transport the items. Previously, this hypothesis had also found significant empirical support [7,43,46,48], e.g. by the fact that in leafcutter ants, species that collect leaves from trees tend to engage in task partitioned leaf retrieval, whereas species living in more homogeneous grassland usually retrieve leaf fragments in an unpartitioned way, without first dropping the leaves, particularly at close range to the nest [43,49].
A surprising result in our evolutionary experiments was that adaptive task specialization was achieved despite the fact that the robots in each team all had identical controllers encoded by the same genotype. This implies that a combination of individual experience, stigmergy and stochastic switching alone were able to generate adaptive task specialization, akin to some of the documented mechanisms involved in behavioral task specialization in some asexually reproducing ants [63] and in cell differentiation in multicellular organisms and clonal bacterial lineages [59,64,65]. The choice of using homogeneous, clonal groups of robots with an identical morphology precluded other mechanisms of division of labor observed in nature from evolving, based, for instance, on morphological [4,12] or genetic [4] role specialization. Such mechanisms, however, could be considered in the future if one allowed for genetically heterogeneous robot teams [28] or evolvable robot morphologies. Lastly, the grammar we used did not specifically allow for recruitment signals to evolve, such as those observed in leafcutting ants, where both trail pheromones and stridulation are used as mechanisms to recruit leaf cutters [66,67], or the ones in honeybees, where the tremble dance is used to regulate the balance between number of foragers and nectar receivers inside the colony [68,69]. Nevertheless, including low-level primitives for communication behavior into the grammar, which we plan to do in future work, would readily allow for the evolution of such mechanisms, and would likely boost the performance of the evolved controllers even further (cf. [26,27]).
In terms of the mechanisms of task specialization and task allocation evolved, our work is important in that it alleviates one of the limitations of existing models on the evolution of task specialization, namely, that they normally take pre-specified subtasks and an existing task allocation model (e.g. the response threshold model) as point of departure [4], thereby greatly constraining the path of evolution. Our work is an important cornerstone in establishing, to the best of our knowledge, the first model that bridges the gap between self-organization and evolution without significantly constraining the behavioral strategies and coordination mechanisms that can be obtained to achieve optimal task specialization and task allocation. In fact, compared to other previous studies on evolution of task specialization [47,62,70–72], our work is the first to consider non-predefined sub-tasks that could evolve de-novo and combine into complex individual behavioral patterns.
Although our experiments demonstrate that division of labor and behavioral specialization in teams of identical robots could evolve in both the scenarios we considered, fitness landscape analyses showed that optimal task allocation could be achieved more easily if optimized behaviors capable of carrying out the different subtasks were available as pre-adapted behavioral building blocks. This leads us to suggest that when building blocks are solidified in earlier stages of the evolution, complex coordination strategies such as task specialization are more likely to evolve as the fitness landscape becomes smoother and also easier to explore due to its greatly reduced size. In addition, it brings further support for the hypothesis that, in nature, the evolution of division of labor in social groups and other transitions in the evolution of sociality also tends to be based on the co-option of pre-existing behavioral patterns, as opposed to requiring the de-novo evolution of many entirely new social traits [17]. Our results, therefore, match and can be integrated with available evidence with respect to the importance of preadaptations in the origin of advanced forms of sociality [2,17–22,73]. For example, reproductive division of labor and worker task specialization are thought to be derived from mechanisms that initially regulated reproduction and foraging in solitary ancestors [17,20–22], sibling care is thought to be derived from ancestral parental care [19], and reproductive altruism (i.e., a sterile soma) in some multicellular organisms evolved via the co-option of a reproduction-inhibiting gene expressed under adverse environmental conditions [73]. Furthermore, it confirms other studies that have examined the building block hypothesis with various digital systems, for example in the context of genetic algorithms [74], evolution of single robot morphologies [75] and the open-ended evolution of simple computer programs [76].
From an engineering perspective our study is the first to achieve a complex form of division of labor using an evolutionary swarm robotics approach, and the first to use the method of Grammatical Evolution to evolve complex, non-trivial behavioral patterns. This result is novel in the field of evolutionary swarm robotics, where, few exceptions aside, most studies have used non-incremental and non-modular approaches, e.g. based on monolithic neural networks [38,77]. In fact, previously, the only other studies which evolved a rudimentary task allocation in swarms of robots were those of Tuci et al. [78], which used a neural network controller combined with a fitness function favoring a required preset task allocation [78], of Duarte et al. [40], which used evolved neural network controllers capable of carrying out particular subtasks, which were then combined with a manually engineered decision tree, and the work of refs. [79–81], which used open-ended evolution and a simplified robotic scenario to evolve heterogeneous behaviors for collective construction [79,80] and pursuit [81] tasks in presence of a pre-specified set of three sub-tasks. Typically, the behavioral complexity that could be reached in these artificial neural network-based studies was quite limited, making the evolution of self-organized task specialization in homogeneous groups out of reach for these methods. In fact, the evolution of self-organized task specialization would clearly require a non-standard neural network approach, involving recurrent neural connections to keep track of the internal state (e.g. the current direction of motion to be able to perform phototaxis), a mechanism to achieve modularity and a mechanism to switch stochastically between these modules. Extending the neural network approach used in evolutionary swarm robotics to this level of complexity would be an interesting task for the future. Other studies on task allocation and task partitioning in swarm robotics typically used traditional, manually engineered approaches [82–88] (reviewed in [89]). All these methods are significantly less general than ours, given that we used a nature-inspired automatic design method with a single fitness criterion, group performance, without any pre-engineered decision-making mechanisms, and simultaneously evolved a self-organized task decomposition and task allocation mechanism as well as optimized behaviors to carry out each of the evolved subtasks. We therefore believe that GESwarm and grammatical evolution will play a key role in the future of evolutionary swarm robotics.
In conclusion, our work and the results we obtained are therefore important both to explain the origin of division of labor and complex social traits in nature, as well as to advance the field of evolutionary swarm robotics, as we showed that the novel methodological and experimental tools we developed were able to synthetize controllers that were beyond the level of complexity achieved to date in the field.
|
10.1371/journal.pgen.1000415 | Measures of Autozygosity in Decline: Globalization, Urbanization, and Its Implications for Medical Genetics | This research investigates the influence of demographic factors on human genetic sub-structure. In our discovery cohort, we show significant demographic trends for decreasing autozygosity associated with population variation in chronological age. Autozygosity, the genomic signature of consanguinity, is identifiable on a genome-wide level as extended tracts of homozygosity. We identified an average of 28.6 tracts of extended homozygosity greater than 1 Mb in length in a representative population of 809 unrelated North Americans of European descent ranging in chronological age from 19–99 years old. These homozygous tracts made up a population average of 42 Mb of the genome corresponding to 1.6% of the entire genome, with each homozygous tract an average of 1.5 Mb in length. Runs of homozygosity are steadily decreasing in size and frequency as time progresses (linear regression, p<0.05). We also calculated inbreeding coefficients and showed a significant trend for population-wide increasing heterozygosity outside of linkage disequilibrium. We successfully replicated these associations in a demographically similar cohort comprised of a subgroup of 477 Baltimore Longitudinal Study of Aging participants. We also constructed statistical models showing predicted declining rates of autozygosity spanning the 20th century. These predictive models suggest a 14.0% decrease in the frequency of these runs of homozygosity and a 24.3% decrease in the percent of the genome in runs of homozygosity, as well as a 30.5% decrease in excess homozygosity based on the linkage pruned inbreeding coefficients. The trend for decreasing autozygosity due to panmixia and larger effective population sizes will likely affect the frequency of rare recessive genetic diseases in the future. Autozygosity has declined, and it seems it will continue doing so.
| Population geneticists use genetic markers to quantify and compare levels of inbreeding in populations and identify disease-associated loci; epidemiologists utilize demographic factors to quantify disease risk modifiers. Our research group sought to investigate the intersection of these two disciplines and examine the way in which demographic trends associated with decreasing levels of inbreeding may influence genomic structure and how this may affect medical genetics research. By examining two age-heterogeneous populations of outbred North Americans, we were able to ascertain genetic changes occurring over the past century that have been likely brought about by recent increases in mobility, urbanization, and population admixture. Using multiple measures of the genomic manifestations of distant consanguinity, we showed significant trends towards decreasing levels of autozygosity and more marginal inbreeding coefficients as study participant birth years neared the chronological present day. We believe this finding is particularly important, as decreasing autozygosity and less homozygosity genome-wide may help to slightly reduce the burden of rare recessive diseases in the future.
| Rates of travel and migration within North America have increased substantially over the past century due to advancements in infrastructure and technology. It has been hypothesized that this ease of travel and globalization has shaped the demographic structure of both North American and world populations in recent generations. Within the past two centuries, population growth, admixture and expansion have been rapid, causing increasing genetic variation in many populations [1],[2].
In our research, we have investigated how demographic trends in the past century have been recapitulated in quantifiable genetic changes, and how this may impact medical genetics and genetic diseases. We focused our study on genome-wide rates of autozygosity and measured tracts of extended homozygosity in two age-heterogeneous samples of North Americans to estimate the genetic effect possibly attributable to demographic change.
We measured autozygosity in the form of runs of extended homozygosity (ROHs). We analyzed these extended tracts of homozygosity genome-wide, showing a strong positive association between increasing chronological age and increasing rates of autozygosity. These homozygous runs were used to quantify consanguinity in our analysis populations. We also utilized a modified inbreeding coefficient to quantify decline in the proportion of excess homozygosity outside of linkage disequilibrium. This modified inbreeding coefficient, has been calculated using data that has had all neighboring single nucleotide polymorphisms (SNPs) that are in linkage disequilibrium (LD) with each other removed (LD pruned data), and is referred to as Fld. Our results show that older members of the population have a tendency to possess more homozygous runs, which comprise a higher percentage of the total genomic length, than those found in younger participants. These older participants also exhibit a larger proportion of excess homozygosity based on estimates from inbreeding coefficients.
The homozygous runs representing autozygosity in relatively outbred populations may also be highly relevant in disease gene discovery. The mapping of these regions on a genome-wide scale could help to identify low-frequency variants associated with complex disease [3],[4]; therefore helping to alleviate the methodological constraints of the common-disease/common-variant mode of inheritance that is generally utilized in whole-genome associations studies [5]. Genetic effects of consanguinity have been shown to be associated with epistatic effects at disease susceptibility loci causing reduced resistance to environmental risk factors and infectious diseases [6],[7]. Epidemiological studies, and animal models, have provided empirical evidence that consanguinity is a risk for complex diseases such as high blood pressure, cancers, osteoporosis, schizophrenia, epilepsy and depression [1], [6], [8]–[10].
We first measured the runs of homozygosity in our discovery population comprised of a cohort of controls compiled by the Coriell Institute, to be a representative sample of neurologically normal North Americans of European descent. We generated summary descriptive statistics for all 809 individuals aged 19–99 that quantified mean and standard deviations for number of ROHs, total percentage of the genome in ROHs (%ROH) and average ROH length (Table 1). We then compared these measures across strata of ∼20 year age groups. In comparisons of the youngest and oldest age strata, significant differences exist in all measures of homozygosity. The oldest age group (estimated current age ≥80 years) presented larger, more frequent ROHs, causing more of the genome to be comprised of ROHs than the youngest age group (estimated current age ≤39 years). These differences were significant (|t|>2.5, p-value≤0.01) for all ROH measures, although the greatest difference appeared in comparing %ROH between the two age groups (|t| = 3.53, p-value = 0.0005). Differences in Fld were suggestive (|t| = 1.91, p-value = 0.056). This illustrates possible generational differences in autozygosity in an outbred population of unrelated individuals (Table 2).
The trend of increasing autozygosity associated with chronological age remained significant in a number of linear regression models, each model incorporating different covariates (Table 3). In multivariate regression models following up the initial results, the associations of chronological age and both the number of ROHs and the %ROH were unattenuated by the introduction of statistical adjustments for either observed or expected homozygosity outside of linkage disequilibrium, Fld (standardized β≥0.10, t-statistic≥2.5, p-values≤0.01). Trends showing a positive association between Fld and chronological age were also significant (standardized β≥0.08, t-statistics≥2.14, p-value≤0.033). This association between Fld and chronological age was relatively unattenuated in models adjusted for average ROH length. The weakest associations with chronological age occurred when examining the linear association with average ROH length. The effect size of this association with average ROH length was small, with standardized beta-coefficients varying between 0.06 and 0.07 (t-statistics between 1.77 and 2.12), although still below our a priori significance threshold of p<0.05 for two of the models. All models for the association between chronological age and average ROH length were at least borderline significant (p-value<0.10).
The predictive models based on the results of our regression analyses forecasts a multi-faceted trend for increasing autozygosity as years since birth increases (from the temporal present). These models, summarized in Figure 1, describe patterns of autozygosity and excess homozygosity having decreased over the twentieth century. Based on these hypothetical models that include data imputed to additional years outside of the 80 birth year range for participants in this study, the number of ROHs has decreased by 14.0% over the past 100 years; while the %ROH and average run lengths have also decreased by similar factors, 24.3% for the former and 10.5% for the latter. Fld has decreased by a factor of 30.5% in our models.
A subset of participants from the Baltimore Longitudinal Study of Aging (BLSA), were selected to replicate these associations from our analyses of the Coriell Control cohort. These participants were selected based on their similarity to the Coriell cohort used in the initial analyses. The results from the BLSA cohort supported all of our hypotheses tested in the initial discovery cohort.
In the BLSA cohort, significant differences in rates of homozygosity exist when comparing participants between the standardized chronological ages of 40–59 years to those aged 80–99 years. These two, more demographically similar, age groups were compared to stringently test replication due to the relatively few BLSA samples in the 19–39 year old cohort (N = 19) that would have limited statistical power for comparisons. Results were consistent to those in the initial analyses of the Coriell Control cohort, with t-tests showing significant differences between age strata for number of ROHs (|t| = 2.8, p-value = 0.0051), %ROH (|t| = 3.1, p-value = 0.0022), and average ROH length (|t| = 2.0, p-value = 0.0485). The younger participants were generally less homozygous than the older strata, with significant differences in Fld (|t| = 3.15, p-value = 0.0018).
Linear regression models of age-associated decrease in measures of autozygosity were also utilized to test the validity of our initial results. Identical regression models as those used in the Coriell cohort were used in the BLSA cohort. The smaller BLSA cohort successfully replicated all trends found to be significant in the analyses of the Coriell cohort. These models were generally more significant in the replication analyses than in the initial discovery cohort (Table 3).
This research has definitively shown the existence of a trend for decreasing autozygosity with younger chronological age in the North American population of European ancestry. The ROHs we identified, larger than 1 Mb, are clearly representative of autozygosity due to distant consanguinity in our outbred populations, and not chromosomal abnormalities or common copy number variants [3],[11],[12]. Using our predictive models of decreasing Fld, we show a quantifiable decrease in consanguinity over the twentieth century. Based on data provided in Carothers et al [13], this decrease in Fld found in our discovery population is on the order of individuals transitioning from a single inbreeding loop 4–5 generations prior, to no inbreeding loops within <6 generations. We postulate that the increased mobility, urbanization and outbreeding in North America in the last century has led to less consanguinity (and thus less homozygosity and homogeneity) in younger individuals [1],[2].
We have shown a weaker association with chronological age for the measured average ROH length. It is possible that since populations that are becoming more outbred, less consanguinous and more heterogeneous recombination could fracture ROHs into smaller segments, for which a robust measure such as %ROH would be less affected than average ROH size. This could contribute to higher variation in average run length measurements, resulting in increased variation in the measure causing more possible type I error and decreased statistical power when compared to %ROH.
With extended regions of homozygosity decreasing in size and becoming less frequent, this structural genomic trend may have some latent effect on public health, as well as the recently developed methods for genome-wide association studies (GWAS). Theoretically, an excess of ROHs and excess homozygosity (identified using the linkage pruned inbreeding coefficient, Fld) may increase the chances of rare recessive genetic diseases. The trends shown in this research may have a larger impact in modifying the epigenetic, epistatic and polygenic pathways that influence many complex traits [6],[7]. This is particularly of interest when considering the rates at which partially recessive alleles may decline in frequency and reduce phenotypic variation in complex polygenic traits.
Our results show that if demographic trends continue towards a globalized, urbanized and more freely mobile world, populations will become less consanguineous.
This initial genome-wide analysis was undertaken on a subset of the 828 unrelated clinical controls from the National Institute of Neurological Disorders and Stroke (NINDS) funded Neurogenetics repository at the Coriell Institute. These samples were collected by Coriell to be used as convenience genetic controls, representative of the North American population of varied European descent. The replication cohort for this study was taken from the Baltimore Longitudinal Study of Aging, a community based longitudinal study of aging currently in its 50th year of follow up.
Individuals from both cohorts were genotyped concurrently at the National Institute on Aging's Laboratory of Neurogenetics (LNG) using the Illumina Infinium technology (Illumina Inc., San Diego, CA). The assays used for genotyping included were the Infinium II HumanHap550 v. 1, Infinium II HumanHap550 v. 3, or a composite of Infinium HumanHap300 and Infinium II 240S. By combining the genotype data from Infinium HumanHap300 and Infinium II 240S assayed participants, to an equivalent level of genomic coverage as the Infinium II HumanHap550 assays, we were able to standardize participant data across a total of 545,066 single nucleotide polymorphisms genotyped on the Illumina platforms. The raw genotype data were stored and quality controlled using GERON genotyping (http://neurogenetics.nia.nih.gov), an intranet repository for genotype data created on the Illumina platform.
All samples from both cohorts were first quality controlled for a minimum of a 97% successful genotype call rate. Any samples failing this initial quality control step were re-genotyped using a new DNA aliquot until a 95% successful call rate was achieved. 13 of the initial DNA samples from our neurological control population were ultimately excluded due to consistent call rates below our inclusion threshold of 97%. Of the 848 participants from BLSA that were genotyped who based on available data were not self-reported African American, 34 samples had call rates below 97%. SNPs with minor allele frequencies less than 5% and departures from Hardy-Weinberg equilibrium (HWE test, p<0.01) and missingness per SNP greater than 5%, were excluded from further analyses. PLINKv1.0.1 was used to carry out sex-checks based on heterozygosity of the X chromosome genotypes were used to exclude 4 participants from the Coriell cohort whose self-reported sex did not match that presented in the genotypic data ([14], http://pngu.mgh.harvard.edu/~Purcell/plink/). 14 of the samples from the BLSA cohort were excluded due to X chromosome heterozygosity inconsistent with self-reported sex. Blinded sex-checks were carried out for all samples using the default function for genotypic estimates of gender in the Illumina Beadstudio package (Illumina Inc., San Diego, Caliornia). The only conflicting genders reported by the Beadstudio results were identical to those reported by PLINK.
All samples utilized from both studies underwent further quality control procedures to check for any indications of population stratification or substructure. Identical by descent (IBD) analyses were used to identify repeated samples or distantly related individuals, with an apriori exclusionary criteria of sharing at most a 2.0% proportion of the genotyped SNPs as identical by descent in any pair-wise combinations of samples. This strict exclusionary criteria is necessary, as including any participants that are related within recent generations could bias our analyses and lead to a non-independence of measures of autozygosity. Probands from groups of related individuals were randomly selected for inclusion in the analyses. This IBD filtering called for the removal of one duplicate sample from the Coriell population. This filter eliminated 101 cryptically related individuals from the BLSA. The high number of related individuals in the BLSA is to be expected as a result of recruitment patterns in the study. The BLSA itself is a cohort of volunteers initially recruited from a group of retired Federal scientists, who subsequently recruited friends and family members. Family members of early participants in the BLSA were given priority enrollment during the course of the study.
Identical by state distances were generated using multi-dimensional scaling for the remaining samples in an attempt to identify population outliers not already considered during the admission/adjudication of the participants. The distribution was standardized on HapMap samples used to aid in the detection of stratification and outliers. One participant was removed from the Coriell cohort of neurological controls due to ancestry consistent with African samples. There were no identifiable sub-population clusters or outliers apparent within the European American samples in the multidimensional scaling analyses of the Coriell dataset. When the Coriell cohort, BLSA cohort, and HapMap samples underwent combined multidimensional scaling analyses (using 410,834 shared quality controlled SNPs as a basis for comparison), additional outliers for possible stratification effects were removed from the BLSA cohort, with 222 samples being more than two standard deviations from the combined population mean for any of the four components of the MDS model. BLSA is a relatively ethnically diverse population, compared to the Coriell samples. Beginning in 1990, a conscious effort was made to recruit African Americans, while Asians, Latin Americans and participants from other ethnic groups were also recruited in recent decades. This deviation from mean component measures in BLSA during the quality control process may be due to some slight level of additional cryptic relatedness or population admixture/stratification not seen in the discovery cohort samples. All genotyping quality assessments, sex-checking, IBS and IBD calculations were carried out using PLINK v1.0.1. A summary of quality control results can be found in Table S1 in the supplemental materials.
After the quality control process, we were left with an analytic population of 809 neurologically normal participants genotyped at 476,962 SNPs. These participants were sampled between the ages of 15 and 95 years old (age range = 80 years, mean age at sampling = 61.7 years old, standard deviation = 16.7 years). However, records show a 6 year period over which these controls were adjudicated and sampled. There was no statistical correlation between the dates of sample collection and participant age (Pearson correlation, p-value>0.05), this suggests no sampling bias with regard to age at collection. This data allowed for the calculation of the participants' estimated current chronological age (within less than1 year) standardized to the year 2008. Chronological age refers to the participant's calculated current age, regardless of death, and ranges between 19–99 years of age (mean age of 61.7 years±16.8). The population is comprised of 57.9% female participants.
477 participants from the BLSA were selected for replication purposes after passing similar quality control to the Coriell cohort, including the removal of genotyping failures, population outliers and cryptically related samples. These samples were individually standardized to 450,364 quality controlled SNPs. BLSA samples in the replication population possessed a population standardized mean chronological age of 68.3 (S.D. = 13.7). With regard to autozygosity measures, the BLSA samples were slightly lower than those of the Coriell cohort in both the entire population and the four age strata (Tables 1 and 2). The BLSA samples also exhibited ∼1% more excess heterozygosity than the Coriell cohort, based on mean Fld calculations. This may be due to the fact that the BLSA cohort is derived from an urban dwelling population based in Baltimore, MD. The BLSA cohort was comprised of slightly more males than the Coriell cohort, but this should not be a factor in the replication as all analyses were confined only to the 22 autosomes.
We utilized the PLINK v1.0.1 toolkit to identify runs of homozygosity. Primary criteria for inclusion of a genomic region into a homozygous run are the region must be at least 1 megabase in length and contain at least 50 adjacent SNPs (per Mb) with homozygous genotype calls. This robust size and SNP density threshold for inclusion into ROHs allows for the algorithmic exclusion of copy number variants, centromeric and SNP-poor regions. The density requirement of at least 50 SNPs per Mb is based on an apriori genome-wide coverage target of ∼500,000 quality controlled SNPs in analytic populations. This requirement of at least 50 SNPs per Mb is similar to the requirements for ROHs found in the Gibson et al., 2005 analysis of runs of homozygosity in HapMap Phase II data [15]. A sliding window of 50 SNPs was used to identify these runs, and included no more than 2 SNPs with missing genotypes and 1 possible heterozygous genotype. These analyses were limited to the 22 autosomal chromosomes. Identical parameters were used to generate these measures in the BLSA cohort as were used in the Coriell cohort.
Our metrics for comparing rates of autozygosity among the participants in this study were able to be calculated after identifying the ROHs. Our primary measures of autozygosity include: total percentage of the genome included in ROHs and the average length of ROHs. The total percentage of the genome included in ROHs was calculated by summing the length of each individual ROH per participant. This summed length of identified ROHs was then divided by a factor of 2,645 and subsequently converted to a percent by multiplying the dividend by 100. The division by 2,645 in the generation of the %ROH measure is based on the number of megabases covered by SNPs included in the Infinium HumanHap 550v.1 and 550v.3 assays used to generate our genome-wide datasets. This estimate of coverage of 2,645 Mb was calculated by summing the distance between the first and last available SNP of each chromosomal arm for each of the 22 autosomes. The average length of ROHs was calculated by dividing the total length by the number of ROH segments per participant. Both of these measures are expressed in Mb.
As ROHs are associated with regions of high linkage disequilibium, we also attempted to examine rates of homozygosity outside of LD. To calculate the additional measure of Fld, we created LD pruned versions of each of the genome-wide datasets. We accomplished this by algorithmically excluding SNPs in LD with neighboring SNPs to create the LD pruned datasets, and then calculating inbreeding coefficients based on the remaining data.
Using data from all 809 participants in our analysis population (and a separate-identical analysis for all 477 participants in our replication study), we calculated variance inflation factors (VIF) for each of the possible pairwise combinations of SNPs within a sliding window of 50 SNPs (with 5 SNP overlaps per window). We excluded all SNPs with a VIF>1.05 within each sliding window. This VIF threshold corresponds to a maximum multiple correlation coefficient representative of ∼1% co-linearity of genotype calls with any other SNP in the sliding window of analysis. This stringent variance threshold allowed us to trim the genome-wide dataset for the Coriell cohort to 48,902 SNPs dispersed relatively evenly across the 22 autosomes, and to 34,307 autosomal SNPs in the BLSA cohort.
We calculated expected rates of homozygous calls per participant based on HWE expectations of genotype frequencies using the LD-pruned datasets. We then calculated cohort specific observed rates of homozygosity within the LD-pruned datasets, expressed as a summed count of homozygous genotypic calls per participant. Using PLINK, we then calculated single population inbreeding coefficients (F statistics) to summarize the proportion of homozygous genotypes differing from our HWE based expectations per participant. These F statistics, based on calculations carried out on the datasets containing only SNPs not in LD with each other, comprise the summary measure we refer to as the LD pruned inbreeding coefficient (Fld). We converted this to a percentage in our tables for ease of comparison, although actual Fld coefficients are used in the predictive models. In our randomly selected populations of unrelated individuals, the Fld values we calculated are a proxy for the occurrence of excess homozygosity on a genome-wide level. The Fld statistic allows for an accurate assessment of autozygosity outside of linkage disequilibrium, without being adversely affected by low SNP density after removing SNPs in regions of LD. The linkage pruned sub-sets of the genotypic data was used for the calculation of observed and expected rates of homozygosity outside of LD as well of the generation of the inbreeding coefficients that comprise the Fld statistic. A discussion of the use of the LD-pruned data to construct ROHs may be found in the supplemental materials in Text S1.
Descriptive statistics were generated for all variables involved in analyses. These include counts, means and standard deviations for all three measures of autozygosity (number, %ROH and average run length) and Fld to be used as dependent variables, as well as for the primary predictor variable of estimated chronological age. These measures were all relatively normally distributed in our population of neurological controls.
Generational differences were estimated by sorting participants into 4 age strata based on 20 year intervals. Descriptive statistics were calculated again to compare mean variation in measures of autozygosity and Fld. Differences in mean measures of autozygosity and Fld between the oldest (estimated current age of 80–99 years) and youngest (estimated current age of 19–39 years in the Coriell cohort) generations were compared using a basic two-way t-test for each of the autozygosity measures.
Linear regression models were constructed in order to investigate possible associations between chronological age and autozygosity measures. Similar models were used to evaluate the association between Fld and chronological age. Separate regression models were constructed for dependent variables of number of ROHs, %ROH, average ROH length and Fld. These models were initially adjusted for gender only (gender was not a statistically significant term in any models, p-value>0.05), although to create more parsimonious models, gender was not included. Additional covariates of observed and expected rates of homozygosity outside of LD were added to the second and third sets of models respectively to further scrutinize and follow-up the initial results. Fld was used as a covariate in the fourth model set to account for the possible confounding effect of chance homozygosity outside of LD in the examinations of trends involving measures of autozygosity derived from ROHs. Subsequent regression models evaluating the trend for increasing Fld with chronological age were adjusted for average ROH length. Additional regression models investigating associations in combined cohorts are described in the supplemental materials in Text S1 and Figure S1.
A second set of linear regression models were created to investigate the possible multiplicative effect of age, by using age2 as the primary predictor of increasing autozygosity measures (gender adjusted) or Fld. However, none of these models that incorporated age2 showed a stronger association with the measures of autozygosity, as the standardized-beta-coefficients and r2 values were actually smaller than those in the previous models of linear age. These models are not included in the manuscript as they add no additional pertinent information, but are available upon request.
Linear predictive models of autozygosity decrease over the twentieth century were extrapolated from regression models based on the Coriell discovery cohort. These models estimate decreasing rates of autozygosity and excess homozygosity as time progresses. These are based on the regression coefficients from the original un-adjusted models of chronological age predicting demographic change in the total number of ROHs, %ROH, average ROH lengthand Fld. These models provide estimates of time dependent means and confidence intervals for both measures. Percent change over 100 years was estimated for each measure using these models. All estimates of percent change were based on a minimum value of zero except for Fld, when a scalar minimum for the calculation based on the lowest value within the 95% confidence interval of the predictive model (Fld = −0.0031) was used.
|
10.1371/journal.ppat.1000244 | Dissecting the Cell Entry Pathway of Dengue Virus by Single-Particle Tracking in Living Cells | Dengue virus (DENV) is an enveloped RNA virus that causes the most common arthropod-borne infection worldwide. The mechanism by which DENV infects the host cell remains unclear. In this work, we used live-cell imaging and single-virus tracking to investigate the cell entry, endocytic trafficking, and fusion behavior of DENV. Simultaneous tracking of DENV particles and various endocytic markers revealed that DENV enters cells exclusively via clathrin-mediated endocytosis. The virus particles move along the cell surface in a diffusive manner before being captured by a pre-existing clathrin-coated pit. Upon clathrin-mediated entry, DENV particles are transported to Rab5-positive endosomes, which subsequently mature into late endosomes through acquisition of Rab7 and loss of Rab5. Fusion of the viral membrane with the endosomal membrane was primarily detected in late endosomal compartments.
| Dengue virus (DENV) is the most common arthropod-borne infection worldwide with 50–100 million cases annually. Despite its high clinical impact, little is known about the infectious cell entry pathway of the virus. Previous studies have shown conflicting evidence about whether the virus fuses directly with the cell plasma membrane or enters cells by receptor-mediated endocytosis. In this manuscript, we dissect the cell entry pathway of DENV by tracking single fluorescently labeled DENV particles in living cells expressing various fluorescent cellular markers, using real-time multi-color fluorescence microscopy. We show that DENV particles are delivered to pre-existing clathrin-coated pits by diffusion along the cell surface. Following clathrin-mediated uptake, the majority of DENV particles are transported to early endosomes, which mature into late endosomes, where membrane fusion occurs. This is the first study that describes the cell entry process of DENV at the single particle level and therefore provides unique mechanistic and kinetic insights into the route of entry, endocytic trafficking behavior, and membrane fusion properties of individual DENV particles in living cells. This paper opens new avenues in flavivirus biology and will lead toward a better understanding of the critical determinants in DENV infection.
| Dengue virus (DENV) is a mosquito-transmitted, enveloped RNA virus that belongs to the family Flaviviridae. This family also includes West-Nile virus (WNV) and tick-borne encephalitis virus (TBEV). DENV causes the most common arthropod-borne infection worldwide with 50–100 million cases annually [1]–[3]. Despite its threat to human health, there are presently neither vaccines nor antiviral drugs to prevent or treat dengue infection. The development of novel therapies requires insight into the viral life cycle. A potential target for intervention strategies is the infectious cell entry pathway.
The infectious entry of DENV is mediated by the viral envelope glycoprotein E, which is organized in 90 homodimers on the surface of the virion [4],[5]. The E glycoprotein is involved in interaction with cellular receptors as well as the subsequent membrane fusion process [6]–[8]. In vitro studies with TBEV indicate that membrane fusion is triggered upon exposure of the virus to low pH [8]. At low pH, the E proteins undergo a dramatic re-organization which leads to the formation of E trimers [9]. The crystal structure of the E protein has been solved in its dimeric pre-fusion, and trimeric post-fusion configurations [10],[11]. Although much is known about the molecular mechanisms involved in the membrane fusion process, many critical questions regarding the cell entry pathway of flaviviruses remain unanswered.
The cell entry mechanism of DENV remains controversial. Early electron microscopy studies provided evidence for direct fusion with the plasma membrane [12],[13], whereas a recent study indicates that DENV enters cells via clathrin-mediated endocytosis [14]. Clathrin-mediated endocytosis involves internalization of ligands and receptors through a clathrin-coated pit, which buds into the cell cytosol and delivers its cargo to early endosomes and subsequently to late endosomes and lysosomes [15]–[17]. Other flaviviruses have also been described to infect their host cell via clathrin-mediated endocytosis [18]–[21]. Evidence for flavivirus entry via this pathway is based on the use of inhibitors of clathrin-mediated uptake, such as chlorpromazine and dominant-negative mutants of Eps15 [18],[20],[22]. Furthermore, addition of acidotropic reagents to cells has been observed to dramatically reduce viral infectivity and membrane fusion activity, suggesting that flaviviruses mediate membrane fusion from within acidic endosomes [23]–[26]. A recent study on the entry of WNV particles demonstrates that WNV colocalizes with the early endosome marker EEA-1 (Early Endosome Antigen-1), and at later time points with the late endosome/lysosome marker LAMP-1 (Lysosome-Associated Membrane Protein-1) [27]. Taken together, these studies suggest clathrin-mediated endocytosis as a viable pathway for flavivirus entry, but the exact manner in which DENV virus particles enter cells and traffic through the endocytic network remains unclear, as does the identity of the organelle in which viral fusion occurs.
In this study, we dissected the cell entry pathway of DENV by tracking fluorescently labeled DENV particles in living cells expressing various fluorescent cellular markers using real-time multi-color fluorescence microscopy. These experiments demonstrate that DENV infects its host cell via clathrin-mediated endocytosis. DENV particles move on the cell surface in a diffusive manner until they join a pre-existing clathrin-coated pit. Following clathrin-mediated uptake, the majority of DENV particles enter early endosomes that progress to late endosomes, where membrane fusion occurs.
In order to visualize single DENV particles in living cells, we labeled the virus with the lipophilic fluorescent probe DiD. The concentration of the DiD dye in the viral membrane is sufficiently high so that its fluorescence is largely quenched, but still allows single DiD-labeled virions to be detected. Membrane fusion can be observed as fluorescence dequenching. We have shown previously that this labeling procedure does not affect the infectious properties of DENV [26]. The tracking experiments were performed in African green monkey kidney cells (BS-C-1), which are highly permissive to DENV infection [26],[28]. To test whether DENV is internalized through clathrin-mediated endocytosis, BS-C-1 cells stably expressing enhanced yellow fluorescent protein (eYFP) fused to the light chain of clathrin (LCa-eYFP) were used. We and others have previously shown that LCa-eYFP highlights more than 95% of the coated pits and vesicles in living cells and that this fusion protein does not disturb the functional integrity of clathrin molecules [29],[30]. DiD-labeled DENV was added in situ to these cells at 37°C and fluorescent images were recorded at 2 frames per second for 25 min. Figure 1A shows the distribution of LCa-eYFP (green) and DiD-labeled DENV particles (red) in a cell. The LCa-eYFP signal appeared as discrete structures in cells. A typical example of a DENV entry event is shown in Figure 1B. In this example, the virus particle first binds to and moves along the cell surface. Forty-eight seconds post-binding, the virus particle associates with a discrete spot containing the LCa-eYFP signal. Thereafter, the clathrin signal around the virus particle increases, indicating maturation towards a clathrin-coated vesicle. At 94 seconds, the clathrin signal rapidly disappears, presumably due to uncoating of the clathrin-coated vesicle. Membrane fusion eventually occurs at 512 seconds post-infection. Quantitative analysis of 47 virus trajectories revealed that 98% of the DENV particles that fused with endosomes entered through LCa-eYFP positive clathrin-coated pits. On average, the clathrin signal colocalized with the virus particle for 83 seconds (Figure 1C), which is consistent with previously observed dynamics of clathrin-mediated endocytosis [30],[31].
To confirm that DENV specifically enters cells via clathrin-mediated endocytosis, we investigated the effects of chlorpromazine, a cationic amphiphilic drug that inhibits the formation of clathrin-coated pits [32], and of a dominant-negative mutant form of Eps15 (E95/295), a protein required for clathrin-dependent uptake [33], on DENV infectivity. Viral infectivity was severely impaired in cells treated with chlorpromazine (Figure 1D) and significantly reduced in cells expressing dominant-negative Eps15 (Figure 1E). Furthermore, no membrane fusion events were seen in real-time virus tracking experiments in chlorpromazine-treated cells (results not shown). Taken together, these results indicate that DENV requires clathrin-mediated endocytosis for its infectious entry.
Tracking individual particles also allowed us to determine how DENV particles recruit clathrin-coated pits. A detailed characterization of the individual trajectories showed that virus particles associate with clathrin on average at 111 s post-attachment to the cell surface. Nearly all particles (92%) were observed to move along the cell surface and join a pre-existing clathrin-coated pit. The remaining minor fraction either appeared to land directly on a pre-existing clathrin-coated pit or a clathrin-coated pit was formed directly at the site of the virus particle. An example of the surface motion of DENV towards a clathrin-coated pit is depicted in Figure 2A and Video S1, which is published as supporting information on the web site.
Subsequently, we investigated whether the surface motion of DENV is characterized by random diffusion or directed movement. To this end, we plotted the mean-square displacement (MSD) of each particle prior to association with a clathrin-coated pit as a function of time. A linear relationship between MSD and time would indicate simple diffusion, an upward curvature designates directed motion, and a downward curvature implies diffusion within a confined region. Figure 2B gives the MSD plot for 5 typical virus trajectories. The apparent linear relationship between the MSD and time indicates that DENV moved on the cell surface in a diffusive manner. During the tracking experiments, we noticed that the mobility of the virus drops when the particle overlaps with a clathrin-coated pit. To obtain a quantitative insight into this behavior, we calculated the diffusion constants from the MSD plots for each particle prior to or during colocalization with clathrin and used that as a measure for surface mobility of the virus. The results show that many DENV particles that were associated with the cell surface were quite mobile, but once they were captured by a clathrin-coated pit their mobility was highly reduced (Figure 2C). Furthermore, treatment of cells with chlorpromazine revealed that DENV particles remained migrating along the cell surface throughout the duration of the experiment in a manner similar to that seen for particles prior to clathrin-mediated entry in untreated cells (Figure 2C).
Following clathrin-mediated internalization, virus particles are typically trafficked along an endocytic pathway, which comprises a network of highly dynamic vesicles and endosomes. Endocytic trafficking is regulated by a large family of small Rab GTPases [34]–[38]. Specific Rab GTPases are often enriched in distinct intracellular vesicles and may be used to identify endocytic vesicles and endosomes. For example, Rab5 and Rab7 primarily decorate early and late endosomes, respectively [39]–[41]. Recent live-cell imaging studies have also revealed a small fraction of the endosomes containing both Rab5 and Rab7, which likely indicates intermediate endosomes that are maturing towards late endosomes [34],[35],[37].
To study the itinerary of endosomal compartments visited by DENV, we tracked single DiD-labeled virus particles in BS-C-1 cells co-expressing Rab5 fused to enhanced cyan fluorescent protein (Rab5-eCFP) and Rab7 fused to enhanced yellow fluorescent protein (Rab7-eYFP). We have used this approach before and observed that low level expression of Rab5-eCFP and Rab7-eYFP in cells does not adversely affect endocytic trafficking inside the cell [34]. A typical example of a cell co-transfected with Rab5-eCFP and Rab7-eYFP early after infection with DiD-labeled DENV particles is depicted in Figure 3A. Rab5- and Rab7-positive endosomes can be observed as clear distinct spots that are localized in the cell periphery as well as in the perinuclear region of the cell. We analyzed 51 virus trajectories in total and observed that 86% of the particles first enter Rab5-positive early endosomes; the other 14% of the virions are directly delivered to Rab5/Rab7-positive intermediate endosomes. An example of DENV endocytic trafficking is shown in Figure 3B and Video S2, which is published as supporting information on the web site. At 42 seconds post-binding, this particular virus particle moves with a velocity of 0.42 µm/s towards an intermediate endosome, as shown by the colocalization with Rab5-eCFP and Rab7-eYFP. Subsequently, the intermediate endosome enclosing the virus particle matures into a late endosome as detected by the disappearance of the Rab5 signal. At 342 seconds, the virus particle resides in a Rab7-positive late endosome and induces membrane fusion from within this organelle at 508 seconds post-infection as detected by a five-fold increase in the DiD-intensity. An example of DENV intracellular trafficking via Rab5-positive endosomes is shown in Video S3, which is published as supporting information on the website.
We observed different modes of Rab7 accumulation and Rab5 dissociation during endosome maturation. Rab5-positive early endosomes carrying DENV particles were found to mature either through a gradual appearance of Rab7 (55%) or by merging with an existing Rab7-positive endosome (45%). Typical examples of these modes of Rab7 accumulation are depicted in Figure 3C, and Videos S4 and S5. Likewise, the exit of the Rab5 signal also appears to take place in different modes. About 70% of the intermediate endosomes complete the maturation process by a gradual release of Rab5 (Figure 3D, Video S6). In the remaining cases, the virus particle appeared to be sequestered into a Rab7-enriched domain of the intermediate endosome, which subsequently pinched off and moved away as a late endosomal compartment (Figure 3D, Video S7).
We have previously identified two distinct populations of Rab5-positive early endosomes [34]. A group of dynamic early endosomes are transported on microtubules and rapidly mature towards late endosomes, while the remaining are relatively static and mature much more slowly. Influenza virus, low density lipoproteins, and epidermal growth factors were previously found to be preferentially targeted to the dynamic, rapidly maturing population, whereas transferrin is non-selectively delivered to both populations. DENV was non-selectively delivered to both endosome populations (data not shown).
Enveloped viruses escape from the endocytic pathway by a membrane fusion reaction. In our experimental set-up, membrane fusion can be detected as a sudden increase of DiD fluorescence due to the dilution of the DiD-probes from the viral membrane into the endosomal membrane. This assay allowed us to directly examine the nature of the endosomes from which DENV mediates membrane fusion. Individual virus trajectories showed that the majority of the virus particles first joined a Rab5-positive endosome, which then matured through the Rab5/Rab7-copositive intermediate stage into a Rab7-positive late endosome, were membrane fusion was observed. Figure 4A gives a quantitative kinetic analysis of the endocytic trafficking behavior and membrane fusion events of all analyzed DENV particles. Internalization of DENV particles appeared to be relatively quick, since 50% of the particles localized to early endosomes at 3.5 minutes post-attachment to the cell surface. Thereafter, the particles started to associate with Rab7-positive endosomes. The first membrane fusion events were detected at 5 minutes post-infection, and nearly all fusion events occurred within 17 minutes post-infection. The average time point of membrane fusion was 12.5 min, which is in agreement with our previous results [26].
The vast majority (80%) of particles induced membrane fusion from within Rab7-positive late endosomes devoid of any detectable Rab5 signal, while 15% of the particles fused from within Rab5/Rab7-copositive intermediate endosomes. Only 5% of the virus particles fused from within Rab5-positive early endosomes that lack Rab7 (Figure 4B). Membrane fusion was initiated at on average 5.5 minutes post-entry of DENV into the Rab7-positive endosomes (Figure 4C).
Our observation that DENV (serotype 2, strain S1) primarily fuses from within Rab7-positive late endosomes is somewhat surprising, since a recent report showed that expression of dominant-negative Rab7T22N did not affect DENV (serotype 2, strain New Guinea C) infectivity [22]. To investigate whether this discrepancy is related to the different virus strains used, we analyzed the infectious properties of both viruses on HeLa cells expressing dominant-negative Rab5S34N and Rab7T22N. In agreement with the above results, viral infectivity of S1 was severely impaired in cells expressing dominant-negative Rab7, whereas the infectious properties of NGC were unaffected under the conditions of the experiments (Figure S1). To investigate the requirement for Rab7 of S1 infectivity in more detail, we performed single-particle tracking experiments in cells transiently expressing the dominant-negative Rab7T22N mutant [42],[43]. Figure 4D shows that the number of membrane fusion events was significantly reduced by a factor of 4 in these cells (T-test: P<0.001), which indicates that S1 needs to travel to Rab7-positive endosomes to undergo membrane fusion.
Despite the medical importance of DENV, little information is available about the infectious cell entry pathway of the virus. In this study, we investigated the cell entry process of single DENV particles in real-time by simultaneous tracking of fluorescently labeled DENV particles and endocytic structures in cells. This approach allowed us to obtain mechanistic and kinetic insights into the route of internalization and endocytic trafficking behavior of individual DENV particles in living cells.
Previous electron-microscopy studies suggested that DENV penetrates both mammalian and insect cells by direct fusion with the plasma membrane [12],[13]. In contrast, this report shows that DENV enters cells via clathrin-mediated endocytosis and fuses from within late endosomes. We observed that more than 98% of the particles that underwent membrane fusion, first associated with a clathrin-coated structure for a substantial time period. Furthermore, treatment of cells with chlorpromazine as well as expression of a dominant-negative Eps15 mutant significantly suppressed the number of DENV-infected cells. It is not clear what the explanation is for the discrepancy, but it might be related to the methodology that was used to investigate the cell entry process of the virus. The conclusion that DENV utilizes clathrin-mediated endocytosis for internalization is in agreement with recent observations of Acosta and co-workers [14]. During the course of this study, these investigators published that DENV infectivity in C6/36 mosquito cells is severely inhibited after treatment of the cells with a variety of chemical and molecular inhibitors of clathrin-mediated endocytosis.
Real-time imaging studies showed that macromolecules either induce de novo formation of clathrin-coated pits or are recruited to pre-existing clathrin-coated pits [31],[44],[45]. For example, influenza virus particles land on the cell surface and induce de novo formation of clathrin-coated pits at the site of binding [30]. This study indicates that DENV particles first diffuse along the cell surface before they encounter pre-existing clathrin-coated pits. After the virus associates with the pit, the clathrin signal around the virus particle increases, which implies maturation of the clathrin-coated pit and formation of a clathrin-coated vesicle. Thereafter, the clathrin signal rapidly disappears again, typically within a time scale of a few seconds. This behavior is similar to that of reoviruses, which have been shown to stabilize and induce maturation of pre-existing clathrin-coated pits [31].
Recently, several modes of endosome maturation have been described. Rink et al. showed that Rab5-positive vesicles, which have split off from a dynamic early endosomal network, accumulate Rab7 and subsequently gradually lose Rab5 [35]. Vonderheit et al. found that Rab5-positive endosomes, containing Semliki Forest Virus (SFV) particles, gradually acquire Rab7 in a separate domain. The SFV particles are sequestered into this Rab7 domain, which pinches off as a Rab7-positive late endosome, leaving a Rab5-positive endosome behind [37]. We observed both modes of endosome maturation. Most DENV particles progressed from early to late endosomes by gradual appearance of Rab7 and a gradual loss of Rab5. In addition, 45% of Rab5-positive endosomes carrying DENV merged with existing Rab7-positive endosomes. Occasionally, we observed that DENV particles sequestered into a distinct Rab7 domain, similar to the behavior observed from endosomes containing SFV [37].
DENV particles predominantly fused from within Rab7-positive endosomes. Furthermore, the membrane fusion activity was significantly impaired in cells expressing dominant-negative forms of Rab7, which indicates that progression of DENV to Rab7-positive endosomes is important for its infectious entry. In contrast, Krishnan et al. have recently demonstrated that the infectivity of DENV-2 strain NGC was not affected by dominant-negative Rab7, while ablation of Rab5 severely reduced the number of infected cells [22]. A direct comparison between both virus strains revealed that viral infectivity of S1 was severely impaired in cells expressing dominant-negative Rab7, whereas the infectivity of NGC was unaffected. These results suggest that both virus strains have distinct entry characteristics. In this respect it is interesting to note that DENV-2 strain NGC induces syncytium formation in a fusion from without assay at pH 6.4, whereas the pH threshold for the DENV-2 S1 strain is around pH 5.8 (personal communication, Dr. P. Young, University of Queensland, Australia). The different pH-dependent properties of these virus strains may therefore reflect the distinct requirements for functional endocytic trafficking in cells. Future experiments should reveal whether the pH threshold determines in which organelle membrane fusion occurs.
DENV particles reside on average for 5.5 min in Rab7-positive endosomes prior to the onset of membrane fusion. This result is surprising considering that TBEV efficiently fuses with liposomes in a model system in a time scale of seconds after low-pH exposure [46]. Pre-exposure of TBEV to low pH for 10–20 seconds in the absence of liposomes completely abolishes the membrane fusion activity of the virus [46]. Similar results were obtained for WNV (unpublished results, J. Wilschut and J. M. Smit). Our finding that DENV fuses several minutes after entering a late endosome might therefore suggest that, in addition to exposure to the acidic lumen of the late endosome, other cellular factors are involved in the activation of the membrane fusion machinery of DENV. Another possibility is that the accumulation of Rab7 significantly precedes acidification to the fusion pH.
Taken together, we propose the following model for cell entry of DENV S1 strain. First, the virus particle binds to a cellular receptor. Subsequently, DENV diffuses as a virus-receptor complex or rolls over multiple receptors along the cell surface towards a clathrin-coated pit. Upon capture by a pre-existing clathrin-coated pit, the virus particles loses its mobility. Then, the clathrin-coated pit matures and pinches off into the cell cytoplasm to deliver the particles to Rab5-positive early endosomes. In general, the early endosome carrying the virus matures into a late endosome by gradual accumulation of Rab7, followed by a gradual loss of Rab5. Finally, the DENV particles localize to Rab7-positive late endosomes and move through the cytoplasm of the cell until the onset of membrane fusion allows the genetic material of the virus to be delivered into the cytoplasm.
Single-particle tracking has substantially enriched our knowledge on viral cell entry mechanisms and has revealed previously unknown aspects of virus-host interactions [30],[47],[48]. The mechanistic and kinetic insights offered by this technique provide a better understanding of disease pathogenesis and may lead to a rational design of antiviral drugs and vaccines. This is the first study that describes the cell entry pathway of DENV at a single-particle level. The parameters obtained in this study will serve as a framework for our current study on the fate of individual antibody-opsonized DENV particles into Fc receptor-bearing to elucidate the molecular basis of antibody-dependent enhancement of DENV infection.
Aedes albopictus C6/36 cells were maintained in Minimal Essential Medium (MEM; Life Technologies, Breda, The Netherlands) supplemented with 10% fetal bovine serum, 25 mM HEPES, 7.5% sodium bicarbonate, 200 mM glutamine, 100 µM non-essential amino acids, penicillin (100 U/ml), and streptomycin (100 µg/ml) at 30°C, 5% CO2. HeLa cells were cultured in a 1∶1 mix of DMEM (Life Technologies) and HAM (Life Technologies) supplemented with 10% fetal bovine serum, 25 mM HEPES, penicillin (100 U/ml), and streptomycin (100 µg/ml) at 37°C, 5% CO2. BS-C-1 cells were maintained in MEM (Invitrogen, Carlsbad, CA, USA) supplemented with 10% fetal bovine serum at 37°C, 5% CO2. BS-C-1 cells stably expressing LCa-YFP were created by use of the RetroMax retroviral expression system (Imgenex, San Diego, CA, USA) and cultured in BS-C-1 medium [30]. BS-C-1 cells were grown on glass coverslips (MatTek, Ashland, MA, USA), and prior to the tracking experiments washed with serum-free, phenol red-free medium.
The plasmid encoding Rab5-eCFP was a gift from Dr. M. Zerial (Max Planck Institute, Dresden, Germany). The Rab7-eYFP plasmid was previously constructed by Dr. M. Lakadamyali [34]. The GFP-tagged dominant-negative Eps15 mutant E95/295 and its empty vector D3Δ2 were kindly provided by Dr. A. Benmerah and Dr. A. Dautry-Varsat (Institute Pasteur, Paris, France). The plasmids encoding GFP-tagged dominant-negative Rab5 mutant Rab5S34N, wild-type Rab5-GFP, myc-tagged dominant-negative Rab7 mutant Rab7T22N, and wild-type Rab7-myc were gifts from Dr. P. van der Sluijs (University Medical Center, Utrecht, The Netherlands). Cells were transfected with the plasmids using the transfection reagent FuGENE, according to the manufacturer's protocol (Roche, Nutley, NJ, USA).
To analyze the route of DENV cell entry, viral infectivity was measured in HeLa cells expressing dominant-negative Eps15 mutants. At 30 hours post-transfection, cells were infected at MOI 5. At 21 hours post-infection, cells were washed with PBS, trypsinized, fixed with 4% paraformaldehyde, and permeabilized with 0.5% saponin in PBS containing 2% FBS. Expression of the myc-tagged plasmids was detected with monoclonal antibody A-14 (Santa Cruz Biotechnology, Santa Cruz, CA, USA). DENV infection was measured using the monoclonal antibody MAB8702 directed against the viral E protein (Chemicon, Hampshire, United Kingdom). Cells were analyzed on a FACS Calibur cytometer. The effect of chlorpromazine on DENV infectivity was determined by an infectious center assay in BS-C-1 cells as described before [25]. Chlorpromazine (15 µM) was added to the cells 30 min prior to addition of the virus. At 1 hour post-infection, cells were washed and fresh medium containing 20 mM ammonium chloride was added. At 24 hours post-infection, cells were fixed and stained intracellularly with MAB8702 to measure infection [26].
DENV serotype 2 strain PR159 S1, generously provided by Dr. Richard Kuhn (Purdue University, Lafayette, IN, USA), was produced and labeled with DiD as described previously [26]. Briefly, monolayers of C6/36 cells were inoculated with DENV at MOI 0.1. At 72 hours post-infection, the progeny virions were harvested, purified by ultracentrifugation, and cleared from tartrate using 100 kD filter devices (Millipore, Amsterdam, The Netherlands). Subsequently, 2 nmol DiD (Molecular Probes, Eugene, OR, USA) dissolved in dimethyl sulfoxide (DMSO) was mixed with approximately 5×109 genome-containing DENV particles while vortexing in a total DMSO concentration of less than 2.5%. After 10 min, the unincorporated dye was removed by gel filtration. DiD-labeled virus was stored at 4°C and used within 2 days. Virus preparations were analyzed with respect to the infectious titer and the number of physical particles, as described previously [26].
Tracking experiments were carried out 24 to 48 hours post-transfection as described previously [34]. Briefly, fluorescent images were recorded by exciting CFP with a 454 nm Argon laser (Melles-Griot, Carlsbad, CA, USA), YFP with a 532 nm Nd∶YAG laser (Crystalaser, Reno, NV, USA), and DiD with a 633 nm helium-neon laser (Melles-Griot). For the clathrin experiments, simultaneous images were recorded of DiD-labeled virions and LCa-eYFP at 2 frames per second. In case of the Rab5/Rab7 experiments, the excitation of DiD was continuous, whereas the excitation of CFP and YFP were alternated at 0.5 Hz. The fluorescent emission was spectrally separated by 650 nm long-pass dichroic mirrors (Chroma, Rockingham, VT, USA) and imaged onto two separate areas of charge-coupled device camera (CoolSNAP HQ, Roper Scientific). A 665 nm long-pass filter was used for the emission of DiD. For the emission of CFP and YFP, bandpass filters of 480/40 nm and 585/35 nm were used, respectively. The CFP and YFP filters were toggled by a motorized wheel at 0.5 Hz synchronically with the 454 nm and 532 nm lasers. Image analysis and single-particle tracking was performed using custom-written IDL software as described previously [26],[49]. Briefly, background and noise were reduced by convolution with a Gaussian spatial filter. Viral trajectories were generated by pairing virus spots in each frame according to proximity and similarity in intensity. Colocalization of viruses with fluorescent cellular markers was identified with an automated program and confirmed by eye, the criteria for colocalization being that the objects move together and have at least partial overlap. Only those particles that moved roughly within the focal plane and showed more than a fivefold increase in fluorescence intensity after membrane fusion were used for image analysis. Characterization of the movement of DENV particles on the cell surface was done by generating MSD-plots. The MSD at time interval τ is the average of all squared displacements throughout the virus trajectory prior to or during association with clathrin. The diffusion constants were calculated from the slope of the MSD-plot.
|
10.1371/journal.pntd.0003844 | Diverse Genotypes of Yersinia pestis Caused Plague in Madagascar in 2007 | Yersinia pestis is the causative agent of human plague and is endemic in various African, Asian and American countries. In Madagascar, the disease represents a significant public health problem with hundreds of human cases a year. Unfortunately, poor infrastructure makes outbreak investigations challenging.
DNA was extracted directly from 93 clinical samples from patients with a clinical diagnosis of plague in Madagascar in 2007. The extracted DNAs were then genotyped using three molecular genotyping methods, including, single nucleotide polymorphism (SNP) typing, multi-locus variable-number tandem repeat analysis (MLVA), and Clustered Regularly Interspaced Short Palindromic Repeats (CRISPR) analysis. These methods provided increasing resolution, respectively. The results of these analyses revealed that, in 2007, ten molecular groups, two newly described here and eight previously identified, were responsible for causing human plague in geographically distinct areas of Madagascar.
Plague in Madagascar is caused by numerous distinct types of Y. pestis. Genotyping method choice should be based upon the discriminatory power needed, expense, and available data for any desired comparisons. We conclude that genotyping should be a standard tool used in epidemiological investigations of plague outbreaks.
| Yersinia pestis is a highly pathogenic bacterium and the causative agent of human plague. It has caused three recognized pandemics and is a current human health problem in many countries of Africa, Asia and the Americas, including Madagascar. The pathogen cannot be eradicated from natural plague foci as it persists in various known and cryptic rodent reservoir species. Genotyping is a critical tool in understanding the molecular epidemiology and possible kinetics of plague. In the present study, we succeeded in extracting DNA and genotyping directly from human clinical samples from Madagascar. We applied three different methods, including single nucleotide polymorphism (SNP) typing, multi-locus variable-number tandem repeat (VNTR) analysis (MLVA), and Clustered Regularly Interspaced Short Palindromic Repeats (CRISPR) analysis. Relative to their discriminatory power, all three methods provided important genotype information useful for understanding the molecular epidemiology of the disease, revealing that multiple, distinct genotypes caused human plague in Madagascar within one year, 2007.
| Yersinia pestis, the causative agent of plague, is one of the most deadly zoonotic pathogens on record, with hundreds of millions of human deaths attributed to it over the course of three historical pandemics [1]. Human cases typically present in one of three forms, including bubonic, septicemic, and the contagious pneumonic form [1], and are notifiable to the World Health Organization [2,3]. Since its registration as a notifiable disease in 1954, plague has been reported in various Asian, American and African countries [2,3]. Plague persists in these countries in various known and other cryptic rodent reservoir species in multiple established foci [1,4,5]. Understanding the epidemiology of this pathogen in these natural reservoirs and in human outbreaks requires the development and implementation of effective molecular genotyping tools that can successfully identify and characterize Y. pestis, preferably from a wide variety of sample types.
Plague was first introduced to Madagascar in 1898 during the third pandemic. The disease then spread to the capital city of Antananarivo in 1921 and became established in the surrounding highlands while disappearing from the coastal areas [6,7]. Plague currently persists in two large foci above 800 m in elevation in the central and northern highlands. It remains a significant public health concern, with hundreds of human cases reported annually [7]. Malagasy plague cases are categorized as confirmed (isolation of Y. pestis), presumptive (positive by microscopy but no strain isolation), or suspected (negative test results or no tests performed, but clinical symptoms). Due to logistical difficulties, the frequency of biological case confirmation (confirmed and presumptive cases) is very low in Madagascar (21.4% of suspected cases from 1957–2001) [8], although this can be increased using F1 antigen detection [9]. Historically, genotyping of Y. pestis in Madagascar has been limited to confirmed cases [10–12]. However, recently developed molecular assays provide the opportunity to directly investigate clinical samples without the need for strain isolation.
In the present study, we investigated clinical samples by extracting DNA and then using three different genotyping methods. Each method could successfully be applied despite the background of human DNA and revealed important genotype information useful for understanding the molecular epidemiology of Y. pestis in Madagascar.
Samples were de-linked from the originating patients and analyzed anonymously. All adult subjects provided informed consent, and a parent or guardian of any child participant provided informed consent on their behalf. The consent was approved by signature. Data collection and investigation on human samples were finally approved by the Ethical Committee of the Ministry of Health of Madagascar.
In 2007, 99 human clinical samples were collected from 21 districts in Madagascar (S1 Table). They originated from suspected and confirmed bubonic and pneumonic human plague cases. Tested clinical material included bubo aspirates or sputum collected by the Malagasy Central Laboratory for plague and the Institut Pasteur de Madagascar (provided by Lila Rahalison). All 99 cases from which the clinical specimens were collected were F1 antigen positive and 93 were culture positive (S1 Table). DNA was extracted from inactivated clinical samples using the QIAamp DNA Mini Kit (Qiagen, Hilden, Germany).
DNAs were screened, as previously described, across assorted SNPs—Mad-08 through Mad-49 from reference [12] and Mad-57 through Mad-78 from reference [11]—in a hierarchical fashion. Specifically, SNP Mad-43 was screened first to determine whether a sample belonged in Group I or II, two previously described major groups in Madagascar [12,13]. Then, additional Group I or II SNPs were screened to determine what SNP-determined group (i.e., node) each sample belonged in. Sequencing was performed to determine SNP states for some samples that yielded ambiguous results using the melt mismatch amplification mutation assays (Melt-MAMA) (contact corresponding author for specific methods). Further genotyping was attempted on all 99 samples using a 43-locus MLVA system [14] and by sequencing three CRISPR loci [15–18], as previously described.
Clinical samples that were successfully genotyped using MLVA were analyzed in conjunction with data from 262 previously published samples in a neighbor-joining analysis to determine MLVA subclades [12]. Phylogenies were then constructed using the SNP, MLVA, and CRISPR data for the successfully genotyped clinical samples, with separate phylogenies generated for Groups I and II (as determined by SNP data) for the MLVA and CRISPR based phylogenies (Fig 1 and S1 Table). The MLVA based phylogenies were generated in MEGA6 [19] as neighbor-joining dendrograms using mean character based distance matrices and include bootstrap values ≥50 generated in PAUP 4.0b10 (D. Swofford, Sinauer Associates, Inc., Sunderland, MA) based upon 1,000 simulations. The geographic distributions of the identified MLVA subclades were mapped using ArcGIS 10.2.1 for Desktop (ESRI, Redlands, CA) (Fig 2).
SNP, MLVA, and CRISPR typing all provided robust genotyping overall, despite being used on DNA extracted from clinical samples that included serous, bloody, mucous, or putrified materials. Only four samples amplified very poorly in the SNP and MLVA assays and so were not included in the phylogenetic or geographic analyses. These failures appeared unrelated to culturing success (S1 Table). Two other samples displayed mixed genotypes when analyzed using MLVA and so were also excluded, leaving 93 samples in the phylogenetic and geographic analyses (S1 Table).
The MLVA, SNP, and CRISPR phylogenies demonstrated remarkable congruence. Ten MLVA subclades were identified, eight of which corresponded to previously described subclades [12] and two of which were new (Fig 1 and S1 Table). The ten identified subclades were found in geographically distinct areas (Fig 2) and most corresponded to a SNP phylogeny node/lineage and/or to a CRISPR group (Fig 1 and S1 Table). Although all three methods were able to identify major groups, they differed in their discriminatory power and general applicability to clinical samples. MLVA provided the greatest discriminatory power with 81 unique MLVA genotypes among the 93 samples. However, this method was expensive to run (ten multiplexed PCR reactions and six capillary electrophoresis runs per sample [14]) and was not successful for 4 of the 99 samples (S1 Table). In contrast, there were 22 CRISPR and 13 SNP genotypes among the 93 samples (Fig 1). CRISPR required only three sequencing runs and was successful for all 99 samples, identifying 17 new and, to our knowledge, Madagascar-specific CRISPR spacers. These new spacers were named in accordance with previously published Y. pestis CRISPR spacers [15,16,18] (S2 Table), with new consecutive numbers assigned to new spacers within each of the three loci. This is in contrast to the CRISPR naming strategy recently published for Y. pseudotuberculosis, in which CRISPR spacers were assigned consecutive numbers without regard to the specific locus [20]. The Melt-MAMA assays predominantly used to genotype the SNPs were very economical but demonstrated a higher failure/ambiguous call rate than either MLVA or CRISPR on these clinical samples (S1 Table).
Multiple genotypes continue to cause human plague in Madagascar. Within this single plague season, a total of ten MLVA subclades were identified to cause disease, eight of which were previously described (Fig 1). These subclades showed geographic distributions consistent with earlier observations (Fig 2) [12]. The geographic distributions of the SNP genotypes in lineages q and s (i.e., MLVA subclades I.B and I.A, respectively) were similarly compatible with previous reports [11,12]. Some expansions in the known geographic distributions of several subclades were observed in this study, the most significant of which involved the observation of several subclade I.B samples in three additional northeastern central highlands districts (Fig 2). Whether these are true expansions or if this is due to a lack of samples from these geographic areas in the previous study is unknown. Of seven other previously described MLVA subclades not seen here, two were speculated to be currently extinct (I.I and I.K) and three were restricted to geographic areas not sampled in this study (I.C and I.G in the northern highlands and II.C in the Betafo district) [12]. The failure to observe the remaining two previously described subclades (I.F and II.D), despite sampling within their known geographic distributions, could be due to chance or may indicate the extinction of these two subclades as of 2007.
In addition to the above, two new MLVA subclades were also identified (Fig 1). Subclade II.A.2 appears to be a previously unrecognized subdivision within the previously identified subclade II.A based upon a neighbor-joining analysis of the 93 samples from this study and 262 previously published samples [Fig 2, S1 Table in 12]. Subclade II.A was not statistically supported in the previous analysis and so may not be a robust group [12]. Subclade I.L appears entirely new and was geographically restricted to district Ambalavao in the southern central highlands (Fig 2). This subclade may be newly emerged or may not have been previously observed due to the very low sampling in this district in the previous study [12].
Using CRISPR, a total of 18 new and, apparently, Madagascar-specific spacers have been identified, 17 here and one previously [17]. Of these, 13 belong to the A locus, two to the B locus, and three to the C locus (S2 Table). Compared to a total of 140 Y. pestis specific spacers published worldwide [15,16,18], this is a fairly high number for a geographic area the size of Madagascar, although it is consistent with the high plague activity in this endemic country and the similarly large numbers of SNP and MLVA genotypes that have been reported from Madagascar [11–13].
Of the CRISPR genotypes observed in Madagascar, the a1-a2-a3-a4-a5-a6-a7-a8 b1-b2-b3-b4-b5 c1-c2-c3 CRISPR genotype may represent the root CRISPR genotype in Madagascar, as it was shared by samples in Groups I (CRISPR genotype I.1, n = 14) and II (CRISPR genotype II.1, n = 1) (Fig 1 and S1 Table). This CRISPR genotype was also observed in the majority of the genotyped samples in a recent study of a 2011 pneumonic plague outbreak that occurred outside the recognized northern highlands plague focus in northern Madagascar. These samples all belonged to SNP node k [17], placing them in CRISPR genotype I.1 described here. Two other genotyped samples from that study, one from a rat trapped in one of the outbreak areas, commune Ambarakaraka, and one from a reference sample from the northern Bealanana district, possessed another CRISPR genotype seen here, genotype I.15 [17] (Fig 1 and S1 Table). This is interesting, given the association of this CRISPR genotype with MLVA subclade I.A (Fig 1), as this is the only indication of this MLVA subclade in northern Madagascar. This hypothesis could be confirmed by genotyping these samples with either SNP Mad-57, marking the s lineage [11], or with MLVA, but is strongly suggested by the congruence observed here between MLVA and CRISPR (Fig 1). The spread of MLVA subclade I.A to northern Madagascar would not be unexpected, given the previous success of this genotype in spreading throughout the central highlands and to the port city of Mahajanga [11,12]. Previously published SNP and MLVA data for samples from northern Madagascar are limited [11,12] and a more extensive analysis using a larger set of samples from this region could determine the timing and extent of the spread of this MLVA subclade to northern Madagascar.
In this clinical sample analysis, SNPs provided less discriminatory power than MLVA or CRISPR and the Melt-MAMA SNP genotyping assays demonstrated a higher failure/ambiguous call rate. However, additional whole genome sequencing and SNP discovery, particularly of strains from the d or k nodes, will likely lead to additional SNP genotypes and could potentially allow identification of all currently described MLVA subclades in Madagascar. Also, SNP analysis remains the only reliable method for accurately differentiating between Groups I and II `12] (Fig 1). In addition, the failure/ambiguous call rate could likely be improved using a different SNP genotyping method and, given the hierarchical nature of the SNP analysis, could still be relatively economical even if a more expensive method were used.
The MLVA, SNP, and CRISPR results reported here indicate that direct DNA extraction and genotyping of clinical samples is possible using these methods and may be used for epidemiological investigations, sidestepping the need for obtaining bacterial cultures. Method choice should be based upon the discriminatory power needed, expense, and available data for any desired comparisons. Multiple genotypes continue to be responsible for causing human plague in Madagascar and continue to be observed in geographically distinct areas.
|
10.1371/journal.ppat.1002706 | Plasticity of the β-Trefoil Protein Fold in the Recognition and Control of Invertebrate Predators and Parasites by a Fungal Defence System | Discrimination between self and non-self is a prerequisite for any defence mechanism; in innate defence, this discrimination is often mediated by lectins recognizing non-self carbohydrate structures and so relies on an arsenal of host lectins with different specificities towards target organism carbohydrate structures. Recently, cytoplasmic lectins isolated from fungal fruiting bodies have been shown to play a role in the defence of multicellular fungi against predators and parasites. Here, we present a novel fruiting body lectin, CCL2, from the ink cap mushroom Coprinopsis cinerea. We demonstrate the toxicity of the lectin towards Caenorhabditis elegans and Drosophila melanogaster and present its NMR solution structure in complex with the trisaccharide, GlcNAcβ1,4[Fucα1,3]GlcNAc, to which it binds with high specificity and affinity in vitro. The structure reveals that the monomeric CCL2 adopts a β-trefoil fold and recognizes the trisaccharide by a single, topologically novel carbohydrate-binding site. Site-directed mutagenesis of CCL2 and identification of C. elegans mutants resistant to this lectin show that its nematotoxicity is mediated by binding to α1,3-fucosylated N-glycan core structures of nematode glycoproteins; feeding with fluorescently labeled CCL2 demonstrates that these target glycoproteins localize to the C. elegans intestine. Since the identified glycoepitope is characteristic for invertebrates but absent from fungi, our data show that the defence function of fruiting body lectins is based on the specific recognition of non-self carbohydrate structures. The trisaccharide specifically recognized by CCL2 is a key carbohydrate determinant of pollen and insect venom allergens implying this particular glycoepitope is targeted by both fungal defence and mammalian immune systems. In summary, our results demonstrate how the plasticity of a common protein fold can contribute to the recognition and control of antagonists by an innate defence mechanism, whereby the monovalency of the lectin for its ligand implies a novel mechanism of lectin-mediated toxicity.
| All multicellular organisms have developed mechanisms to defend themselves against predators, parasites and pathogens. As a common mechanism, animals, plants and fungi use a large arsenal of carbohydrate-binding proteins (lectins) to protect themselves from predation and parasitism. The success of this type of innate defence mechanism critically depends on the diversity of specific recognition of foreign carbohydrate structures by the host lectins. In this study, we use NMR structure determination to show that part of this diversity is created by the plasticity of common protein folds. The identified fungal lectin that is toxic to nematodes and insects, adopts a common lectin fold but is remarkable in terms of its specificity and affinity for the recognized foreign carbohydrate structure, the number and location of the carbohydrate binding sites on the protein and the degree of oligomerization. Since the identified in vivo target of the fungal lectin is characteristic for invertebrates, our results may be exploited to develop novel approaches for the control of animal and human parasites.
| Adequate and efficient defence mechanisms to protect an organism's integrity and survival have been essential for the evolution of multicellularity since loss of individual cells may be detrimental for a multicellular organism. Any defence mechanism thereby critically relies on the ability to discriminate between self and non-self. Since all living cells display specific carbohydrate structures on their surface [1], glycans have been used for the recognition of non-self since the beginning of multicellular life [2]. Accordingly, many of the proteins that are able bind to specific carbohydrate structures, commonly referred to as lectins, have been implicated in defence, mainly in the innate immune systems of animals which is considered an ancestral defence mechanism and a first and immediate line of defence against potentially harmful microorganisms [3]. These lectins are either membrane-bound or secreted and localize to the interface between the host and the environment where they bind to microorganism-associated carbohydrates and function either as receptors triggering the expression of host immune effectors, by opsonizing the microorganisms for host immune effectors or immune cells (reviewed in [4]) or as direct immune effectors by killing the microorganism upon binding [5]–[8]. In analogy to latter function of combining non-self recognition and killing, plants use insecticidal lectins to defend themselves against herbivorous insects [9]. Recently, a group of fungal lectins, commonly referred to as fruiting body lectins, has been shown to play a role in the defence of multicellular fungi against predators and parasites based on their toxicity to various model organisms [10]–[15]. According to the above role of lectins in defence, most defence lectins should be specific for carbohydrate structures that do not exist in the host (are non-self) and are characteristic for the target organism. To date, only very few target carbohydrate structures or glycoconjugates of such lectins involved in innate defence mechanisms have been identified and their recognition by the lectin investigated at molecular level [5], [8], [10], [16], [17].
In organisms lacking an antibody-based adaptive immunity, such a lectin-based defence strategy critically relies on a large diversity in carbohydrate specificities. This diversity can be achieved either by diversification on the level of lectin folds and/or by the plasticity of a common lectin fold. The known fruiting body lectins belong to six structural families [14] of which the β-propeller-fold lectins, actinoporin-like lectins, galectins and β-trefoil (ricin B or R-type) lectins [18] are the most prominent ones. Some of these lectins are multidomain proteins harbouring in addition a cysteine protease/dimerization domain (R-type Marasmius oreades agglutinin [MOA] and Polyporus squamosus lectin [PSL]) [19], [20] or a pore-forming module (R-type Laetiporus sulphureus lectin [LSL]) [21]. In the first case, it was demonstrated that both domains are required for toxicity [10] suggesting that the lectin domain guides the catalytic domain to specific target structures. However, most lectins implicated in the defence of plants and fungi are composed just of lectin domains and contain multiple binding sites for either the same or different carbohydrate structures. For some of these lectins it has been demonstrated that this multivalency is essential for their toxicity [22]. These results suggest that lectin-mediated toxicity involves crosslinking of glycoconjugates but the exact mechanism remains unclear.
We describe the identification and characterization of a novel, monovalent lectin, CCL2, from fruiting bodies of the ink cap mushroom Coprinopsis cinerea and present the NMR structure of CCL2 in ligand-free form and in complex with its in vivo ligand. The lectin was found to bind specifically and with an atypical high affinity to Fucα1,3-modified core N-glycans in vitro, using a single, topologically novel binding site on its β-trefoil fold. N-glycans carrying such a modification are characteristic for invertebrates but absent from fungi. We applied biotoxicity assays to demonstrate toxicity towards two model invertebrates. In accordance with the in vitro binding data, the nematotoxicity of CCL2 was dependent on core α1,3-fucosylation of C. elegans N-glycans on intestinal proteins of the nematode. These results show how multicellular organisms exploit the plasticity of a common protein fold to create a novel lectin specificity and an alternative mechanism of lectin-mediated toxicity for defence.
We detected a soluble 15 kDa protein from fruiting bodies of the model mushroom C. cinerea by virtue of its binding to horseradish peroxidase (HRP) in immunoblots. The protein was present in extracts from fruiting bodies but not from vegetative mycelium, indicating a fruiting body-specific expression. We isolated the protein using HRP-affinity chromatography (Figure 1A) and identified it as hypothetical protein CC1G_11781 of C. cinerea strain Okayama7 by MALDI-MS/MS. Since the protein, termed CCL2 (Coprinopsis cinerea lectin 2), was extracted from fruiting bodies of the C. cinerea strain AmutBmut (Swamy et al 1984), the respective genomic locus of strain AmutBmut was cloned and sequenced. This sequence served as a basis for the cloning of the respective cDNA from total RNA isolated from AmutBmut fruiting bodies. A second cDNA, coding for an isoprotein (52% identity; Table S1), termed CCL1 (Coprinopsis cinerea lectin 1) (CC1G_11778), was cloned and sequenced accordingly. The two proteins are predicted to contain neither a signal sequence for classical secretion nor N-glycosylation sites. The cDNAs coding for CCL1 and CCL2 were cloned in pET expression vectors and the proteins were expressed in the cytoplasm of E. coli BL21(DE3). The recombinant proteins were highly expressed and soluble (Figure S1) and versions containing eight N-terminal His-residues were purified using metal-affinity chromatography. Size exclusion chromatography of the purified CCL2 showed that the protein exists as a monomer in solution (Figure S2).
Immunoblots using a CCL2-specific antiserum confirmed that CCL2 is abundant in fruiting bodies and absent from vegetative mycelium (Figure 1B). The differential expression of both CCL2 and CCL1 was quantified at the transcript level by qRT-PCR (Figure 1C). The results indicate that the mRNA levels of CCL1 and CCL2 are more than 1000-fold and 60,000-fold, respectively, higher in fruiting bodies than in the vegetative mycelium.
Based on the binding to the plant glycoprotein HRP and a similar expression pattern as previously characterized lectins from this organism [23], [24], we hypothesized that CCL2 is a lectin. Fluorescently labeled CCL2 was used to probe a glycan array offered by the Consortium of Functional Glycomics (CFG) (Figure 2 and Table S2), confirming that CCL2 is a lectin that binds specifically to carbohydrate structures containing the Fucα1,3GlcNAc motif e.g. the LewisX antigen (Galβ1,4[Fucα1,3]GlcNAc; Glycan structure #133/134 on the array). The disaccharide Fucα1,3GlcNAc alone, however, showed a very low fluorescence, suggesting that at least a trisaccharide was required for efficient binding. Glycan array analysis with purified CCL1 (Figure S3 and Table S3) yielded almost the same results as with CCL2. The binding specificity of CCL2 was further studied with several carbohydrates in vitro by NMR spectroscopy and isothermal titration calorimetry (ITC) as summarized in Table 1. The trisaccharide LewisX bound with a moderate KD of 456 µM and the NMR spectra displayed intermediate to slow exchange behavior during the titration, whereas the binding of sialylated LewisX, was slightly better by a factor of ∼3. However, fucosylated chitobiose (GlcNAcβ1,4[Fucα1,3]GlcNAc-spacer; Figure 3A), absent on the glycan array, had by far the highest affinity among the tested oligosaccharides with a KD of 1.4 µM (Table 1 and Figure S4). Monitoring the binding by NMR spectroscopy revealed large chemical shift changes under the slow exchange regime (Figures 3B and C). Binding occurs with a stoichiometry of 1∶1 and no further changes were observed by adding an excess of ligand (1∶50). The largest chemical shift deviations occurred at residues W78, N90-T95, G108 and K109 (Figure 3D).
Since CCL2 did not show sequence similarity to any known structure we determined the 3D structure of CCL2 by NMR spectroscopy (Figure 4). CCL2 adopts a β-trefoil fold consisting of three β-β-β-β repeats with a pseudo C3 symmetry. β1 and β4 of each repeat form together a β-barrel whereas β2 and β3 adopt a β-hairpin that usually harbors the carbohydrate-binding site [25]. The β-trefoil structure can be compared to a tree [26] in which the trunk is represented by the β-barrel (β1 and β4, β5 and β8, β9 and β12), the roots are formed by the N- and C-terminus together with the two loops β4–β5 and β8–β9, the upper crown is formed by the three β-hairpins (β2 and β3, β6 and β7, β10 and β11) and the lower crown by the loops connecting the β-barrel with the β-hairpin loops. As can be seen from Figures 4B and D, the loops β6–β7 and β7–β8 in subdomain β are shorter than in the other subdomains. In addition, subdomain β shows a deviation from the most characteristic feature of β-trefoil proteins, the QxW motif in each subdomain [25]. Subdomain β contains a YxW instead. A search for structurally similar proteins revealed a large number of bacterial, fungal and plant toxins displaying high structural similarity but low sequence identity (Table S4).
The 3D structure was used to visualize the largest chemical shift deviations from the titration experiment with GlcNAcβ1,4[Fucα1,3]GlcNAc (from Figure 3D) in Figure 4C. The largest deviations occur at the interface between subdomain β and γ, mainly on strand β8 and its unusually short preceding loop β7–β8 (β subdomain) and in the β9–β10 loop (γ subdomain). This arrangement does not correspond to the typical binding interface of β-trefoil lectins and therefore we decided to investigate this new binding mode.
We solved the 3D structure of the complex between CCL2 and fucosylated chitobiose (GlcNAcβ1,4[Fucα1,3]GlcNAcβ–sp) by NMR spectroscopy. 82 intermolecular distance restraints that are well distributed over the binding interface (Figure 5A) were derived from a 3D 13C F1-edited F3-filtered HSQC-NOESY [27] spectrum (Figure S5). A precise structural ensemble of the complex was obtained (Figure 5B and Table 2). The carbohydrate is bound at the interface of the subdomains β and γ in the lower crown (Figure 5C), in particular between the β-strands β6 and β8 and the linker β7–β8 of the β subdomain and the loop between β9–β10 of the γ subdomain. Compared to the canonical binding sites (Figure 5G and Figure S6) this is a very unusual binding location for ricin B type lectins.
The well-defined trisaccharide is oriented such that GlcNAc2 (see Figure 3A for nomenclature of the individual sugars in the trisaccharide) stacks on top of Fuc2′ thereby locking the conformational freedom of the glycan resulting in a narrow clustering of the glycosidic angles (Figure S7). The hydrophobic B-face of Fuc2′ is oriented towards the protein (bottom) and the hydrophobic B-face of GlcNAc2 towards the solution (top). In this orientation GlcNAc1 is tilted horizontally such that its B-face is located on the back contacting the protein. Contacts to all three sugar units are mediated by a large number of potential H-bonds and hydrophobic interactions (Figures 5D–F and Table 3).
The specific recognition of each sugar unit can be described as follows: Fuc2′ approaches the edge of β-strand β8 and the tip of loop β9–β10 with its b-face and bridges subdomain β and γ in this way (Figures 5C–F). In this orientation O4 and O5 face down and are specifically recognized by H-bonds to the main chain (V93 HN and O) of the unusually short loop between strands β7 and β8 (Figure 5E). The equatorial hydroxyl groups of O3 and O2 form H-bonds to G108 HN (second largest chemical shift deviation, Figure 3D) and Lys109 NH3+. In addition the hydrophobic methyl group and the axial H2, both facing downwards, form hydrophobic contacts with Trp94/Trp95 and Val93, respectively. The methyl group is located above the ring of W94 enabling favorable Me-π interactions that are supported by an upfield shift of the H6 resonance (−0.18 ppm; Table S5). In total all characteristic groups of Fuc2′ are specifically recognized by 4 H-bonds, hydrophobic and π interactions. Both the location of Fuc2′ at the subdomain interface and the recognition by three H-bonds to the main-chain are unprecedented in all ricin B type lectin complex structures. GlcNAc1 is specifically recognized at its equatorial acetamido group by a H-bond of its HN to Asn91 O, and at O6 by an H-bond to the side chain of Asn90. The acetamido group forms hydrophobic interactions to Val93 and Me-π interactions with Tyr57 which is supported by an upfield shift of the methyl 1H resonance (−0.24 ppm). Its hydrophobic b-face packs to the Tyr92 side chain. Only a GlcNAc would be recognized at this position since the equatorial orientation of the acetamido and the CH2OH group are necessary for their recognition by H-bonds and the equatorial positioning of O3 and O4 is required for the stacking between Fuc2′ and GlcNAc2. GlcNAc2 is mainly recognized via its acetamido group by an H-bond to Trp78 HN (supported by the largest HN chemical shift deviation, Figure 3D), hydrophobic interactions of the methyl with Leu87 and a stacking of the entire acetamido group to the ring of Tyr92 (Figure 5D). Me-π interactions to Y92 are supported by an upfield shift (−0.24 ppm). GlcNAc2 that stacks on top of Fuc2′ is slightly laterally shifted exposing the hydrophobic H4 facing downwards. H4 is located on top of the Trp94 ring and favorable H-π interactions are supported by an upfield shift of its resonance (−0.39 ppm; Table S5). Two additional potential H-bonds are observed in some structures of the ensemble: between the carbonyl of W78 and O3 of GlcNAc2 and between K109 NH3+ and GlcNAc2 O6. In summary, GlcNAc1 and Fuc2′ are specifically recognized by interactions to almost all of their functional groups whereas the recognition of GlcNAc2 is more relaxed. It is mainly recognized at its equatorial acetamido group attached to C2. This residue must be able to stack to Fuc2′ in order to properly position the acetamido group; both GlcNAc and GalNAc fulfill this requirement and will be recognized in this position. Accordingly, CCL2 binds to fucosylated LacdiNAc (GalNAcβ1,4[Fucα1,3]GlcNAc; Glycan structure #89) on the array.
The large number of H-bonds to the main chain is remarkable. The unusually short β7–β8 loop contributes three and the β9–β10 loop one such H-bonds. Since the protein main chain does not change upon binding, part of the recognition pattern on the protein is preformed. However, the lengths and conformations of these loops are a special feature of CCL2 homologues as illustrated on a structure-based alignment (Figure 6) and are not conserved in the β-trefoil fold. Note also that the short β7–β8 loop lacks the typical 310 helix segment as seen for example in the structurally most closely related R-type lectin MOA (Figure 5G) which would clash with the carbohydrate.
The interaction of CCL2 with GlcNAcβ1,4[Fucα1,3]GlcNAc is governed by a large ΔH gain of −50 kJ mol−1 at the expense of 16 kJ mol−1 for −TΔS (Figure 7A). The thermodynamicbinding parameters are comparable to those of other high affinity lectins in Figure 7B (Table S6). In contrast to typical lectin interactions with medium affinity CCL2 uses an unusually large number of H-bonds (5–7 to backbone, 5 to side chain) and hydrophobic contacts (Trp78, Tyr92 and Trp94) for recognition of its target. A comparable number and kind of contacts is only found for few high affinity lectin interactions with a comparable KD∼1 µM. Interestingly, the calreticulin interaction with Glcα1,3Manα1,2Manα1,2Man with a KD of 0.77 µM is governed by almost identical thermodynamic values [28], whereas the structurally closely related R-type lectin PSL [20] that binds to 6'sialyl lactose with a KD of 1.3 µM [29] displays a moderately favored enthalpy but almost no entropic penalty. Both lectins use a similar number of direct H-bonds for their target recognition as CCL2 does: 10 (2 to backbone, 8 to side chain) and 9 (4 to the backbone, 5 to side chains), respectively, and a comparable amount of hydrophobic interactions.
We tested the toxicity of CCL2 against four model organisms: the nematode Caenorhabditis elegans, the insects Aedes aegypti and Drosophila melanogaster, and the amoeba Acanthamoeba castellanii. The biotoxicity assays were performed either by feeding the test organisms with E. coli expressing the recombinant lectin as described previously [30], or by adding the purified lectin to the food source of the organisms.
These experiments showed a toxicity of CCL2 for C. elegans and D. melanogaster (Figure 8) but not for A. aegypti or A. castellanii (Figure S8). In the case of C. elegans, feeding on CCL2-expressing E. coli stopped the development of all wildtype (N2) L1 larvae in the assay (Figure 8B). This toxicity was dose-dependent and the presence of 30% of CCL2-expressing E. coli among the fed bacteria was sufficient to reduce the development of more than 95% of the L1 larvae (Figure S9). In the case of D. melanogaster, CCL2 caused a significant delay in development of both pupae and flies by 4- and 10-fold, respectively, relative to the control (Figure 8D). The toxicity of CCL1 towards C. elegans (Figure S10) was found to be similar to that of CCL2.
The observed toxicity was likely to be mediated by binding of CCL2 to the N-glycan cores of glycoproteins in the susceptible organisms since α1,3-fucosylation of N-glycan cores was demonstrated both for C. elegans and D. melanogaster and caused cross-reactivity of anti-HRP antisera with these organisms [31]. Therefore, C. elegans mutants impaired in either fucose biosynthesis (bre-1) or a variety of fucosyltransferases were tested for their susceptibility to CCL2-mediated toxicity (see scheme in Figures 8A and B; Figure S11). In agreement with our predictions, the bre-1(ye4) mutant that is unable to synthesize GDP-fucose was completely resistant to CCL2 intoxication. In addition, the fut-1(ok892) mutant lacking Fucα1,3 at the proximal GlcNAc of the chitobiose core [32] was partially resistant, as most of the worms survived and developed, but just half of the larvae reached L4 stage after 48 hours. On the other hand, a deletion in the fut-6 gene, which results in loss of tetrafucosylated N-glycans in C. elegans, as does a deletion in the fut-1 gene [32], was as sensitive as N2 (wildtype) to CCL2. In order to further explore these results, a fut-6(ok475)fut-1(ok892) double mutant was constructed and found to be completely resistant to CCL2 (Figure 8B). As nematodes are able to α1,3-fucosylate both GlcNAc residues of the core region of some N-glycans [33] and both fut-1 and fut-6 are required for the full fucosylation of this core region (see scheme in Figure 8A; Yan, Paschinger and Wilson, personal communication), our results suggest that the α1,3-fucosylated chitobiose core of N-glycans is the ligand of CCL2 in C. elegans. The partial resistance of the fut-1 mutant can be explained by binding of CCL1/2 to N-glycan cores carrying a single fucose on the distal GlcNAc (Manβ1,4[Fucα1,3]GlcNAc). We hypothesize that this is a less favorable ligand due to the lack of an acetamido group on the mannose.
To study the phenotype of CCL2-mediated intoxication and to follow the fate of the toxic lectin in the worms, different C. elegans strains were fed with E. coli cells producing an N-terminal fusion of CCL2 to the red fluorescent dTomato protein [34]. As can be observed in Figure 8C, a strong fluorescence was observed in the upper intestine of the completely susceptible worms N2 and fut-6(ok475) as a result of CCL2-binding to the intestinal epithelium. This fluorescence was accompanied by an evident damage of intestinal cells which resulted in a massive expansion of the intestinal lumen. In agreement with the effects on larval development (Figure 8B), the fut-6(ok475)fut-1(ok892) double mutant that is resistant to CCL2-mediated intoxication, showed neither red fluorescence nor cell damage or expansion of the intestinal lumen. These results suggest that, in the absence of binding to the intestinal epithelium, the ingested lectin is completely cleared from the lumen after 1 hour. Accordingly, an intermediate phenotype, with some staining and cell damage, mostly in the upper part of the intestinal epithelium, was observed in the partially resistant fut-1(ok892) mutant.
We evaluated the contribution of individual amino acid side chains on the carbohydrate-binding affinity by introducing several point mutations at the binding interface followed by ITC measurements. All variants expressed well (except N91A) and folded properly as judged from 15N-HSQC spectra (Figure S12). Significant decreases in affinity were observed for all mutants except N91A (Table 1 and Figure S13). The Y92A mutation decreased the affinity beyond the detection limit. The second largest affinity decreases are observed for W94A and W78A, indicating that the aromatic side chains provide the largest contribution to carbohydrate-binding affinity. A significant decrease in affinity was also observed for Y57A, L87A, N90A and V93A point mutants (4- to 17-fold).
CCL2 variants were also tested in vivo for toxicity towards C. elegans. Remarkably, those mutants that retained carbohydrate binding with high affinity (KD<30 µM) in vitro were as toxic as wild type CCL2. Mutants with lower in vitro affinity, however, showed a decreased toxicity towards C. elegans (Figure 8E). In summary, these results confirm the carbohydrate-coordinating residues of CCL2 that were identified by NMR spectroscopy and suggest that high carbohydrate-binding affinity of the lectin is required for toxicity.
Our results strongly suggest that the newly identified lectins play a role in fungal defence. The lack of motility and the resulting inability of multicellular fungi and plants to escape from predators and parasites has led to the development of very similar defence strategies. In the absence of adaptive immune mechanisms and circulating immune cells, both types of organisms solely rely on innate defence. Whereas plant defence has already been intensively studied [9], [35]–[37], fungal defence has only recently been explored. It appears that, similar to plants, in addition to small molecules [38], proteins play a key role in the defence of multicellular fungi, in particular against predators and parasites [39]. Among the different types of potential fungal defence proteins identified [13], [26], [40], [41] the number and diversity of lectins is remarkably high, in accordance with the suitability of glycoepitopes for non-self recognition in innate defence mechanisms. Most fungal defence lectins are highly abundant in reproductive and long-term survival structures such as fruiting bodies and sclerotia, respectively, which require special protection [13]. This expression pattern, also found for CCL1 and CCL2 (Figures 1B–C), is analogous to plants where the expression of many lectins is confined to seeds.
The strong and specific toxicity of CCL1 and CCL2 towards D. melanogaster and C. elegans is in accordance with the prevalence of the phyla Arthropoda and Nematoda as predators of mushrooms both in nature [42], [43] and in mushroom farms [44], [45]. In addition, this specificity of CCL1/CCL2-mediated toxicity correlates with the identification of α1,3 fucosylated N-glycan cores as target structures of these lectins in vivo (Figure 8), since this epitope is present exclusively in plant and invertebrate N-glycans [31]. The NMR structure revealed that CCL2 recognizes the fucose-containing trisaccharide, GlcNAcβ1,4[Fucα1,3]GlcNAc, as part of this epitope with high specificity. Within this trisaccharide, almost all functional groups of Fucα1,3GlcNAc and the acetamido group of the distal GlcNAc2 are recognized. The recognition of the distal saccharide is more relaxed, a GalNAc with an acetamido at the same position will be equally well recognized. Accordingly, among the glycans of the mammalian glycan array, GalNAcβ1,4[Fucα1,3]GlcNAc (fucosylated LacdiNAc = LDN-F) was one of the best binders. Since there is space for extensions at O6 of GlcNAc1 and O4 of GlcNAc2 (Figure 5E) we can derive the following recognition sequence: X-1,4GalNAc/GlcNAcβ1,4[Fucα1,3][Y-1,6]GlcNAc in which X and Y are tolerated extensions. In addition, binding of substituted LewisX structures on the glycan array (Figure 2) suggests that substitutions at O3 and O6 of the galactose (corresponding to the distal GlcNAc in α1,3 fucosylated chitobiose) and at O6 of GlcNAc (corresponding to the proximal GlcNAc in α1,3 fucosylated chitobiose) are allowed. Accordingly, we would expect specific binding of CCL2 to paucimannose-type N-glycans carrying both α1,6 and α1,3-linked fucose on the proximal and possibly α1,3-linked fucose on the distal GlcNAc (Figure 8A). The GlcNAcβ1,4[Fucα1,3]GlcNAc motif is also a central part of the anti-HRP epitope that it is recognized by antisera raised against HRP in agreement with the isolation of CCL2 as HRP-binding lectin (Figure 1A). Since this epitope is also a key carbohydrate determinant of pollen and insect venom allergens [46], it appears that the same glycoepitope has been selected as target by the antibody-mediated mammalian adaptive immune system and a lectin-mediated fungal defence system.
The high affinity of CCL2 to the recognized trisaccharide determined by ITC is remarkable. Typically, individual carbohydrate binding sites of lectins have a rather low affinity to their ligands and this low affinity is usually compensated by multivalency achieved either by multiple binding sites on the same polypeptide chain or by oligomerization of polypeptide chains with one or few binding sites which leads to a high avidity towards multivalent ligands [47]. However, high affinity carbohydrate binding sites of lectins have been described and they differ from low affinity binding sites by their degree of specificity [48]: whereas low affinity binding sites often have a broad specificity towards terminal mono- or disaccharides present on many different glycans, high affinity sites recognize distinct oligosaccharides that are characteristic for specific glycans and glycoconjugates. The high affinity and specificity of the carbohydrate binding site in CCL2 towards the recognized trisaccharide is achieved by H-bonds and key hydrophobic contacts to almost all functional groups of Fucα1,3GlcNAc as well as the acetamido group of the distal GlcNAc2. The ladder interaction is central for the high affinity, the absence of the distal acetamido group as in Galβ1,4[Fucα1,3]GlcNAc (LewisX) leads to a drop in affinity by ∼300 fold (in Table 1). To our knowledge, CCL2 is the only lectin that binds GlcNAcβ1,4[Fucα1,3]GlcNAc with such a high specificity and affinity, making CCL2 superior to anti-HRP for detection of this glycoepitope. Since this and the other recognized glycoepitope, GalNAcβ1,4[Fucα1,3]GlcNAc (LDN-F), are present in parasitic helminths [49]–[51], CCL2 may be used for the diagnostics of parasitic infections in animals and humans. The toxicity of CCL2-binding to at least one of these epitopes in vivo, may be exploited to develop novel approaches for the prevention or therapy of these infections. Another application could be the use of CCL2 on lectin microarrays for differential glycan profiling [52] or cellular glycomics [53].
The NMR solution structure of CCL2 in complex with its ligand demonstrates the versatility and plasticity of the β-trefoil fold with regard to carbohydrate binding. First, the carbohydrate specificity of CCL2 is very different from other β-trefoil lectins which recognize terminal galactose epitopes like Galα1,3Gal [19], Galβ1,3GalNAc [54] or Galβ1,3GlcNAc [21], rather than an epitope with a terminal fucose. Second, unlike most β-trefoil lectins which utilize three almost identical binding sites per monomer, CCL2 recognizes the identified carbohydrate ligand via a single binding site. This binding site of CCL2 is located at a very unusual site of the β-trefoil fold, the interface between subdomains β and γ. This stands in contrast to the fungal β-trefoil lectin SSA that also uses a single but canonical binding site [54]. None of the typical carbohydrate binding residues present in other β-trefoil lectins are found in CCL2 emphasizing the uniqueness of this non canonical binding site (Figure 6). Based on few β-trefoil complexes in which the binding site is slightly shifted from the canonical towards the CCL2 location [55]–[58] we speculate that this non-canonical binding site might have arisen from a previous recognition of other parts of the invertebrate N-glycan by the canonical binding site β (Figure 5G) and then have changed to recognize another epitope of the same glycan by the non-canonical binding site. The key residues of the CCL2 binding site are highly conserved in CCL2 homologs of other fungi (Figure 6 and Table S1), but highly variable in other β-trefoil lectins. The unusual carbohydrate specificity is mainly based on H-bonds from the protein main chain which requires the proper arrangement of three main chain sections: most importantly the characteristically short β7–β8 loop, strand β6 and the β9–β10 loop. In particular, the short β7–β8 loop is conserved in all CCL2 homologues with a consensus sequence LPxxYVW, a signature we propose for the identification of lectins with a similar target specificity. In summary, based on sequence alignment we predict that the homologous CCL2 like genes of basidiomycetes have the same unusual binding location and the same target specificity as CCL2 (except LB_L2 that lacks the crucial Y93). As we do not have any evidence for a difference in regulation, specificity or function between the different paralogs, e.g. CCL1 and CCL2, we speculate that this redundancy is a strategy to avoid loss of specific defense effectors by individual gene mutations.
The strong toxicity of CCL2 towards C. elegans and D. melanogaster is surprising in the light of the monomeric state of the lectin in solution and the consequential lack of multivalency for the identified ligand since clustering of glycoconjugates on cell surfaces is generally regarded as a prerequisite for lectin-mediated toxicity [59]. CCL2 mutant proteins unable to bind the HRP epitope are not able to bind anymore to the C. elegans intestinal epithelium which rules out the presence of an additional binding site on CCL2 with different specificity for this tissue (A. Butschi, unpublished results). Thus, we hypothesize that the high affinity of the single carbohydrate-binding site of CCL2 compensates for the lack of multivalency and that CCL2 acts by a novel toxicity mechanism that does not seem to involve clustering. Accordingly, CCL2 variants with a lower affinity in vitro showed a reduced toxicity in C. elegans. Remarkably, the consequences of intoxication of C. elegans by CCL1/2 and the multivalent fruiting body lectins MOA and CGL2 are very similar, all of them leading to disintegration of the intestinal epithelium and a substantial enlargement of the intestinal lumen (Figure 8C) [10], [16]. In addition, experiments aiming at the localization of the target glycoconjugates using fluorescently labeled CCL2 and CGL2 gave very similar results (Figure 8C) [16]. Interestingly, disintegration of the intestinal epithelium and enlargement of the intestinal lumen were also observed with the nematode-specific Cry toxins from Bacillus thuringiensis where carbohydrate-dependent binding to the intestinal epithelium appears to trigger expulsion of microvilli from the apical side of the intestinal epithelial cells [60]. In any case, interference with carbohydrate binding by the lectin, either by mutating genes involved in the biosynthesis of the identified target glycans in C. elegans or altering the identified carbohydrate binding sites in the lectin, abolished toxicity and binding of the fluorescently labeled lectin to the intestinal epithelium (Figure 8C) [10], [16]. It should be noted, however, that not all variants of CCL2 were tested for toxicity towards C. elegans and none was tested for toxicity towards D. melanogaster. Thus, although we can show that the recognition of specific glycans is a crucial part of lectin-mediated defence mechanisms, the exact mechanisms of toxicity remain to be elucidated. Possible mechanisms are direct membrane damage or the interference with cellular signaling pathways, recycling of cell surface receptors, cell-cell or cell-matrix interactions. In order to distinguish between these possibilities and to find potential targets of novel antihelminthics, we are currently in the process of identifying the glycoprotein(s) targeted by CCL2 and CGL2 in C. elegans.
LewisX trisaccharide methyl glycoside, 3′-Sialyl-LewisX tetrasaccharide methyl glycoside and Fucα1,3GlcNAc-OMe were purchased from Carbosynth, UK. The chemically synthesized fucosylated chitobiose GlcNAcβ1,4[Fucα1,3]GlcNAcβ-O(CH2)5COONa [61] was a kind gift of Mayeul Collot, ENS, France. LewisX tetrasaccharide and 3′-Sialyl-lactose were a kind gift of Eric Samain, CERMAV, France. The identity and purity of the carbohydrates was checked using 2D NMR spectroscopy.
Detailed information of the strains used in this study can be found in Table S7. Escherichia coli strain DH5α was used for cloning and amplification of plasmids, strains BL21(DE3) and BL21(DE3)/pLysS were used for bacterial expression of proteins and biotoxicity assays and strain OP50 was used to feed C. elegans during regular breeding. Cultivation conditions of the various organisms are described in Text S1.
CCL2 was isolated and purified from C. cinerea as described in Text S1.
Purified CCL2 was separated by SDS-PAGE, excised from the gel and identified by MALDI-MS/MS. Details of the procedure are described in Text S1.
Details of the quantification are described in Text S1.
The PCR-based cloning strategies for the various CCL1- and CCL2-encoding genes are described in Text S1.
Protein expression of CCL2 was evaluated by immunoblotting. Soluble protein extracts of vegetative mycelium and fruiting bodies from C. cinerea were obtained as described above and separated on a 12% SDS-PAGE and probed with specific antiserum raised in rabbits against purified recombinant CCL2 (Pineda Antikörper-Sevice, Berlin, Germany) and detected with HRP-conjugated secondary antibodies. Transcription levels of both genes were assessed by quantitative real-time PCR (qRT-PCR) as described in Text S1.
Purified CCL1 and CCL2 were fluorescently labeled with Alexa Fluor 488 (Invitrogen) according to the manufacturer's protocol and used (at a final concentration of 200 µg/ml) to probe versions 4.2 and 3.1, respectively, of the mammalian glycan array offered by Core H of the Consortium for Functional Glycomics (CFG).
Unlabelled and uniformly 15N or 13C/15N labeled proteins were overexpressed in E. coli as His8-fusions and purified with affinity chromatography (see Text S1). Samples were dialyzed against NMR buffer (50 mM KH2PO4, pH 5.7, 150 mM NaCl). Complexes of CCL2 with GlcNAcβ1,4[Fucα1,3]GlcNAcβ-O(CH2)5COONa were prepared by titrating the concentrated carbohydrate solution of typically 10 mM into a ∼1 mM solution of CCL2 in NMR buffer until a 1∶1 stoichiometry was reached. Subsequently, the pH was lowered to 4.7 using 10% deuterated acetic acid to avoid precipitation.
NMR spectra were acquired on Avance III 500, 600, 700, 750 and Avance 900 Bruker spectrometers at 310 K. NMR data were processed using Topspin 2.1 (Bruker) and analyzed with Sparky (Goddard, T.D. & Kneller, D.G. SPARKY 3. University of California, San Francisco). The 1H,13C,15N chemical shifts of the protein, free and in complex, were assigned by standard methods [62]. Assignment of carbohydrate resonances of the complex was achieved using NOE correlations and exchange peaks with signals of the free carbohydrate since neither TOCSY based spectra nor a natural abundance 13C-HSQC showed bound carbohydrate signals. The following spectra were used for this purpose 2D 1H-1H NOESY, 2D 13C/15N F1-filtered NOESY and 2D 13C F1-filtered F2-filtered NOESY [63]. The assignments of intermolecular NOEs were derived from 3D 13C F1-edited, F3-filtered NOESY-HSQC [27] spectra of the protein-carbohydrate complex. More details are found in the Text S1.
The AtnosCandid software package [64], [65] was used to generate initial CCL2 structures (free and bound) using three 3D NOESY spectra (13Cali-edited, 13Caro-edited and 15N-edited) and one 2D NOESY spectrum. The automatically generated upper limit restraints file was used as a starting point for the first level of manually refining the protein structures by a simulated annealing protocol using the Cyana package [64]. Preliminary structures of the CCL2-carbohydrate complex were generated using the Cyana package with the above mentioned restraints and manually assigned intermolecular and intra-carbohydrate NOE distance constraints. To create the topology of the carbohydrate for the Cyana library file an initial model was generated by SWEET [66]. The carbohydrate spacer was truncated to a methyl group. 300 structures were generated by CYANA starting from random carbohydrate and protein starting structures. Ensemble of 30 structures of CCL2 free and in complex were refined with AMBER 9.0 [67].in implicit solvent using NOE-derived distances, torsion angles and hydrogen bond restraints as summarized in Table 2. For more details see Text S1. The Ramachandran statistics of CCL2 free and in complex, respectively, show 79.9% and 80.2% in the most favored regions, 18.0% and 18.7% in the additionally allowed regions, 1.5% and 1.0% in the generously allowed regions and 0.6% and 0.2% in the disallowed regions.
Biotoxicity assays for A. aegypti and A. castellanii were performed with recombinant E. coli as previously described [30]. For C. elegans, a liquid toxicity assay was performed as follows: a synchronous population of L1 larvae as well as a bacterial culture of recombinant E. coli expressing CCL2 or containing a vector control were obtained as described [22]. E. coli cells were pelleted and re-suspended in sterile PBS to an OD600 = 2. The assay was set up in 96-well plates (TPP) by mixing 80 µl of the bacterial suspension and 20 µl of L1 larvae containing approximately 30 individuals. Each treatment (different bacterial and/or worm strain combinations) was done in 5 replicates. The worms were allowed to feed on the suspended bacteria at 20°C in the dark. The total number of animals and the percentage of individuals reaching L4 stage were quantified after 48 h. The biotoxicity assay with D. melanogaster was performed adding purified protein to the rearing medium as previously described [68] using 20 eggs.
For the statistical analysis of the toxicity assays, pairwise comparisons were done using the non-parametric Kolmogorov-Smirnov test in the case of C. elegans, A. castellanii and D. melanogaster and the parametric T-student test for A. castellanii. The response variables (development, survival and clearing area) were compared between the tested lectin and the control or between mutant and wildtype.
Details are described in Text S1.
More information is found in Text S1.
ITC experiments were performed on a VP-ITC instrument (MicroCal). The calorimeter was calibrated according to the manufacturer's instructions. Protein and carbohydrate samples were dialyzed against NMR buffer at room temperature using a 3.5 kDa membrane (Spectra/Por) and Micro DispoDialyzer (100 Da cutoff; Harvard Apparatus), respectively. The disaccharide Fucα1,3GlcNAc-OMe was not dialyzed but dissolved in NMR buffer. The sample cell (1.4 mL) was loaded with 70 µM protein; carbohydrate concentration in the syringe was 2–4 mM. A titration experiment typically consisted of 30–50 injections, each of 3 µL volume and 6 s duration, with a 6.7 min interval between additions. Stirring rate was 307 rpm. Raw data were integrated, corrected for nonspecific heats, normalized for the molar concentration, and analyzed according to a 1∶1 binding model.
The atomic coordinates of the structures of CCL2 free and in complex with the fucosylated chitobiose (GlcNAcβ1,4[Fucα1,3]GlcNAcβ-OMe) have been deposited in the Protein Data Bank with accession codes 2LIE and 2LIQ, respectively. The chemical shifts of the free protein and in complex were deposited in the BioMagResBank (BMRB) under the accession numbers 17890 and 17902, respectively. The cDNA sequences of CCL1 and CCL2 from C. cinerea strain AmutBmut were deposited in GenBank under accession number ADO87036 and ACD88750, respectively.
|
10.1371/journal.pgen.1005398 | The Relationship between Gene Network Structure and Expression Variation among Individuals and Species | Variation among individuals is a prerequisite of evolution by natural selection. As such, identifying the origins of variation is a fundamental goal of biology. We investigated the link between gene interactions and variation in gene expression among individuals and species using the mammalian limb as a model system. We first built interaction networks for key genes regulating early (outgrowth; E9.5–11) and late (expansion and elongation; E11-13) limb development in mouse. This resulted in an Early (ESN) and Late (LSN) Stage Network. Computational perturbations of these networks suggest that the ESN is more robust. We then quantified levels of the same key genes among mouse individuals and found that they vary less at earlier limb stages and that variation in gene expression is heritable. Finally, we quantified variation in gene expression levels among four mammals with divergent limbs (bat, opossum, mouse and pig) and found that levels vary less among species at earlier limb stages. We also found that variation in gene expression levels among individuals and species are correlated for earlier and later limb development. In conclusion, results are consistent with the robustness of the ESN buffering among-individual variation in gene expression levels early in mammalian limb development, and constraining the evolution of early limb development among mammalian species.
| The variation generating mechanisms of development interact with the variation sorting mechanism of natural selection to produce organismal diversity. While the impacts of natural selection on existing variation have received much study, those of development on the generation of this variation remain less understood. This fundamental gap in our knowledge restricts our understanding of the key processes shaping evolution. In this study, we combine mathematical modeling, and population-level and cross-species assays of gene expression to investigate the relationship between the structure of the gene interactions regulating limb development and variation in the expression of limb genes among individuals and species. Results suggest that the way in which genes interact (i.e., development) biases the distribution of variation in gene expression among individuals, and that this in turn biases the distribution of variation among species.
| Phenotypic variation within populations is a prerequisite of evolution by natural selection, and in theory has the potential to bias the trajectory and rate of evolutionary change [1–6]. As such, identifying the processes that shape phenotypic variation has long been a fundamental pursuit of evolutionary biologists. Historically, evolutionary biologists have tended to focus on the sorting of population-level variation by selective processes, rather than on the production of that variation by developmental processes [7]. As a result, the effect of developmental processes on the distribution and magnitude of phenotypic variation among individuals and species remains unclear for most systems. In this study we use the mammalian limb as a study system to investigate two questions that address the relationship between developmental processes and phenotypic variation at the level of gene expression dynamics: (1) Does the structure of the gene network affect the distribution of variation in gene expression among individuals?, and (2) Is the distribution of variation in gene expression among individuals correlated with the evolutionary divergence in gene expression among species?
The mammalian limb is an ideal system for examining these questions because its development is well characterized, its morphology diverse, and since its form is central to many mammalian behaviors, its morphology is certainly under selection [8–12]. Many of the critical gene interactions that regulate limb outgrowth and patterning in mouse, the traditional mammal model, have been identified [9,10,13]. Initial budding of the limb from the body and limb outgrowth (embryonic day [E] 9.5 –E11) are regulated by interactions between several genes, including Bmp4, Gli3, Grem1, Shh, AER-Fgf’s (e.g., Fgf4, Fgf8), Fgf10, and Hox genes (Fig 1A). Knockouts of these genes result in pathological phenotypes ranging from severe (e.g., complete limb agenesis; AER-Fgf’s, Fgf10) to moderate (e.g., limb truncations; Bmp4) to mild (e.g., malformed digits; Shh, Gli3, Grem1) [13–18]. Most of these genes (e.g., Bmp4, Gli3, Grem1, Shh, AER-Fgf’s, and Hox genes) are also involved in later limb outgrowth and patterning (E11 –E13), but some of their interactions differ (e.g., Hox genes and Gli3, Shh, AER-Fgf’s; Fig 1B). As a result, the structure of the gene regulatory network differs for earlier (E9.5 –E11) and later (E11 –E13) limb development. This structural difference provides two opportunities to investigate the relationship between network structure and gene expression variation among individuals.
This structural difference also provides an opportunity to contrast earlier and later limb development. Research suggests that the main segments of the limb (e.g., stylopod, zeugopod, and autopod) are specified by or during the time of initial limb outgrowth [19,20]. As a result, disruption of early limb development could have potentially catastrophic effects on limb formation that are not likely to be selectively advantageous (e.g., limb agenesis). In contrast, disruptions of later limb development are less likely to have as severe an impact on the overall limb structure. While later disruptions might impact the relative size of limb segments, they are less likely to result in no limb at all. Following this logic, we might hypothesize that genes regulating early limb development generally exhibit less variation in expression among individuals than those regulating later limb development [21–33]. Additionally, it is possible that select early limb genes might vary at a level equal to or greater than that of individual later genes, but that this variation is dampened at the system level by the interactions among genes that characterize the gene network (i.e., developmental buffering) [34–37]. As population-level variation provides the raw material upon which natural selection acts, we can further hypothesize that the genes regulating early limb development also exhibit less variation in expression among species [38]. Support for these hypotheses would reinforce the importance of network structure (i.e., development) in shaping variation in mammalian limbs among individuals and over evolutionary time, while failure to support these hypotheses would suggest that network structure does not play a critical role in the generation of limb variation.
To test these hypotheses, we computationally modeled the gene networks regulating mouse limb development, and determined the sensitivity of network genes to system perturbation and the ability of network genes to perturb the system when altered. We also assessed the sensitivity of the system as a whole to perturbations in gene interactions and expression. We experimentally quantified naturally occurring variation in the expression of several network genes within a population of mouse individuals. To compare variation among species, we used transcriptomic data (RNASeq) from four mammals with divergent limb morphologies (bat, opossum, pig and mouse). We then assessed the relationship between gene and network sensitivity and gene expression variation among mouse individuals and among mammalian species. Our results suggest that the gene network that regulates early limb development is more robust than that regulating later limb development, and that this robustness buffers variation in early limb gene expression among individuals, and constrains the evolution of early limb development among species.
We assembled early (ESN) and late (LSN) stage networks for key genes regulating limb development from previously published experimental studies [9,13] (Fig 1). The ESN regulates initial limb outgrowth and the initiation of the epithelial-mesenchymal interactions that are critical to continued limb development. These events occur from embryonic days (E) 9.5 to E11 in mouse. The LSN, in contrast, regulates the limb’s differentiation along its anterior-posterior (i.e., thumb to pinky) axis and elongation along its proximal-distal (shoulder to fingertips) axis from E11 to E13 in mouse. To describe the temporal behavior of the activity (i.e., expression) levels of genes in these networks we built mathematical models (see S1 Methods).
After building the models, we ran a series of simulations in which we computationally interrupted interactions between genes and compared the resulting expression levels with those of the unaltered, default model (Table 1).
Within the ESN, removal of the Hox to Grem1 or Gli3R to Grem1 link affects Grem1 and Bmp4 expression levels (i.e., alters expression by 10% or more), but does not affect the expression levels of other genes. Removal of the AER-Fgf’s to Shh or Hox to Shh links only affects the expression levels of Gli3R and Shh, removal of the Fgf10 to AER-Fgf’s or Bmp4 to AER-Fgf’s links only affects the expression level of the AER-Fgf’s, and removal of the Hox to Fgf10 or AER-Fgf’s to Fgf10 links only affects the expression level of Fgf10. Removal of the Shh to Gli3R link affects the expression level of Gli3R, while removal of the Grem1 to Bmp4 affects the expression level of Bmp4. Removal of the Bmp4 to Grem1 link results in no significant change in expression levels. In total, 14 of 77 possible interactions are affected (i.e. expression levels change by 10% or more) by alterations in the ESN (18%) (Table 1).
For the LSN, removal of the Shh to Gli3R link affects Gli3R expression levels, but does not affect the expression levels of other genes. Removal of the AER-Fgf’s to Shh and Hox to Shh links affects only Shh and Gli3R expression levels. Removal of the Repressor X to Grem1 or Grem1 to Bmp4 link disrupts the expression levels of Bmp4, Grem1, and the AER-Fgf’s. Removal of the AER-Fgf’s to Repressor X link also disrupts the expression levels of Bmp4, Grem1, and the AER-Fgf’s, but also affects Repressor X expression levels. Removal of the Hox to Grem1 link affects the expression levels of Bmp4, Gli3R, Grem1, Shh, and the AER-Fgf’s. Removal of the Gli3R to Hox, AER-Fgf’s to Hox, or Gli3R to Grem1 links disrupts the expression levels of all genes save Repressor X. Finally, when the Bmp4 to Grem1, Bmp4 to AER-Fgf’s link is removed, expression levels of all genes are affected. In total, 52 of 84 possible interactions are affected (i.e. expression levels change by 10% or more) by alterations in the LSN (62%) (Table 1).
For each gene in each model, we then determined the number of genes whose removal alters expression of the gene in question (i.e. expression levels change by 10% or more), and the number of genes that exhibit expression changes when the gene in question is removed. We used the resulting values to generate a simulation space that was used to evaluate the ability of genes to affect other genes, and to be affected themselves by network perturbations (Fig 2C and 2D). Simulation spaces for the ESN and LSN were generated using the same scales to allow comparisons. For the ESN (Fig 2C), all genes fall in the lower left quadrant of the space, suggesting that they do not greatly affect expression of other genes, and are not greatly affected by others. In contrast, most LSN genes (Fig 2D) fall in the right upper and lower quadrants of simulation space. Genes in both the upper (e.g., Bmp4, Gli3R) and lower (e.g., AER-Fgf’s, Grem1, Shh) right quadrants and their boundaries are affected by perturbations in other genes, but genes in the upper right quadrant also affect the expression of other genes while genes in the lower right quadrant do not. Hox A/D falls in near the middle of the simulation space, suggesting that it moderately affects others and is affected by them. Only Repressor X falls in the lower left quadrant for the LSN, suggesting that it does not affect others and is not affected itself.
We next varied the parameter values used in the models and compared the resulting gene expression levels to those of the unaltered, default model (S1 Table). Results indicate that the ESN is most sensitive to changes in Hox A/D parameters (39% of Hox A/D parameter changes result in a ≥10% change in the expression level of another gene), followed by Shh (20%) and Bmp4 (18%), Gli3R (18%), and Grem1 (17%). The ESN is less sensitive to changes in AER-Fgf (10%) and Fgf10 (9%) parameters. Bmp4 is the most sensitive of the ESN genes to changes in parameters of other genes (32% of parameter changes result in a ≥10% change in the expression level of Bmp4), followed by Gli3R (26%), Grem1 (19%), Fgf10 (17%), and Shh (13%). Hox A/D (9%) and Fgf8 (7%) expression levels are less sensitive to ESN parameter changes.
The LSN is more sensitive to changes in Repressor X (73%), Bmp4 (60%), and the AER-Fgf’s (53%), and less sensitive to changes in Hox A/D (39%), Grem1 (39%), Shh (25%) and Gli3R (18%). Within the LSN, the AER-Fgf’s (61%) and Gli3R (63%) are the most sensitive to parameter changes in other genes, followed by Bmp4 (51%), Grem1 (47%), and Shh (40%), while Hox A/D (30%) and Repressor X (17%) are less sensitive.
The percentages listed above were used to generate a sensitivity space, similar to the simulation space described above (Fig 2A and 2B). ESN and LSN sensitivity spaces were generated using the same scales to facilitate comparisons. Similar to the simulation results, all ESN genes group within or on the boundary of the lower left quadrant of the space, suggesting that alteration of the values of their related-parameters does not greatly affect expression of other genes, and that their expression levels are not greatly affected by alterations in the values of the related-parameters of other genes. LSN genes are more distributed in the sensitivity space. Bmp4 and the AER-Fgf’s lie in the upper right quadrant, similar to their location in the simulation space (Fig 2C). Grem1, Shh, Gli3R and fall in or on the boundary of the lower right quadrant, indicating that they do not affect others but are affected themselves. Of these, Grem1 and Shh also fall within the lower right quadrant of the simulation space (Fig 2C). Repressor X lies within the upper left quadrant, suggesting that alteration of the values of its related-parameters affects the expression of other genes but that its expression is not greatly affected alterations in the values of the related-parameters of other genes. Repressor X also falls on the left side of the simulation space, but in the lower quadrant. Similar to its location in the simulation space, Hox A/D falls near the center of the plot.
We performed a series of real-time quantitative PCR (qPCR) assays to quantify the expression levels of genes that appear in both the ESN and LSN models (Bmp4, Gli3, Grem1, Shh, and the AER-Fgf Fgf8) in mouse embryos. For the early developmental stages (ES), the averaged, scaled expression level was 2.02 for Bmp4, 2.58 for Fgf8, 2.19 for Grem1, 1.68 for Shh, and 0.02 for Gli3. For the later developmental stages (LS), the average, scaled expression level was 2.29 for Bmp4, 0.83 for Fgf8, 1.92 for Grem1, 2.32 for Shh, and 0.02 for Gli3.
Statistical tests reveal that the mean-standardized variances of expression levels significantly differ among genes in the earlier (ES, E10-E11; Bartlett’s Test, F-ratio = 8.614, DF = 4, P < 0.001*) and later (LS, E11-E13; Bartlett’s Test, F-ratio = 5.823, DF = 4, P < 0.001*) stages of development. In the ES, Shh displays the highest average mean-standardized variance (coefficient of variation, CoV) (0.847), followed by Bmp4 (0.567), Grem1 (0.531), Fgf8 (0.474), and Gli3 (0.380). Fgf8 displays the highest average CoV in the LS (1.701), followed by Shh (1.523), Bmp4 (1.037), Gli3 (0.910), and Grem1 (0.703).
Litter membership also has the power to significantly explain the variance in expression levels in a given gene (e.g., Bmp4) that are observed among individuals (ANOVA; Bmp4 F-ratio = 2.379, DF = 8, P = 0.026*; Gli3, F-ratio = 5.742, DF = 8, P = < 0.001*; Grem1, F-ratio = 4.412, DF = 8, P = < 0.001*; Fgf8, F-ratio = 7.097, DF = 8, P < 0.001*; Shh F-ratio = 2.162, DF = 8, P = 0.043*).
We next compared the among-individual, standardized variation in the expression level of a gene in vivo (CoV, from qPCR) with the: (1) number of genes whose removal alters expression of the gene in question (i.e., alters expression level by 10% or more), and (2) number of genes that exhibit expression changes when the gene in question is removed (from the simulation analyses).
For both the ES and the LS, neither the relationships between the number of genes whose removal alters expression of the gene in question (#1) and the CoV (ES—Least-Squares Regression, R2 = 0.202, P = 0.448; LS—R2 = 0.214, P = 0.433), nor the relationships between the number of genes that exhibit expression changes when the gene in question is removed (#2) and CoV (ES—R2 = 0.503, P = 0.180; LS—R2 = 0.142, P = 0.532) are significant (Fig 3).
For both the ES and LS, Shh is among the genes that are the least sensitive to perturbations in other genes, has a relatively low impact on the expression of other genes when altered, and displays a relatively high CoV. The opposite is true for Gli3R during the ES. Fgf8 displays the highest CoV during the LS and is highly sensitive to perturbations in other genes, like all LS genes.
Given the large difference in the percentage of possible interactions that are affected by computational alterations in the ESN and LSN (18% and 62%, respectively), we next compared the level of variation in measured gene expression during early (ES) and later (LS) development (from qPCR). For every model gene (5 of 5), the average CoV is greater later than earlier in development (P = 0.031*) (Fig 3). When the variation around the averages is taken into account using a resampling technique, the average CoV remains significantly greater later than earlier in development for four of the five genes (Bmp4, P = 0.028*; Gli3, P < 0.001*; Grem1, P = 0.024*; Fgf8, P < 0.001*). Only for Shh, a gene with among the highest CoV in both the early and later stages, does the average CoV does not remain significantly greater later than earlier in development (P = 0.092). The average CoV of the housekeeping gene β-actin does not significantly differ during earlier (0.832) and later (0.929) development (P < 0.217).
To calculate gene expression variation among species, we first generated transcriptomic libraries for bat (Carollia perspicillata), opossum (Monodelphis domestica), pig (Sus scrofa) and mouse (Mus musculus) forelimbs for early (ES; early limb bud) and late (LS; paddle) limb stages. We then used a set of 6,583 genes orthologous to all four species (see S1 Methods) to calculate the among-species conservation of gene expression at each developmental stage, using the mean of all species pairwise Spearman coefficients. All resulting pairwise Spearman coefficients are positive and > 0.50, suggesting that the orthologous genes might perform similar functions between species (Fig 4A and 4B). However, the degree of gene expression conservation decreases from 0.5667 at ES to 0.5612 at LS of forelimb development. To test the robustness of this difference with respect to the selection of orthologous genes, we randomly sub-sampled 500 sets of orthologous genes at early and late stages at intensities ranging from 50 to 100% of all orthologous genes (Fig 4C). For each intensity, the distributions of gene expression conservation levels between early and late stages were significantly different (T-test, P-value < 0.05*). These results suggest that gene expression patterns vary more among species during the LS than ES of limb development, consistent with patterns of variation among individuals.
We also calculated the among-species conservation of mean-standardized expression at each developmental stage for the five genes that appear in both the ESN and LSN models (Bmp4, Gli3, Grem1, Fgf8, and Shh). At the ES, Gli3 (standard deviation [SD] = 0.177) falls within the top 25% of conserved genes while Shh (SD = 0.975) and Fgf8 (SD = 0.475) fall within the top 25% of divergent genes. Bmp4 (SD = 0.257) and Grem1 (SD = 0.260) fall near the middle of the range of genes. Fgf8 (SD = 0.971) and Shh (SD = 1.196) are also among within the top 25% of divergent genes at the LS. However, Grem1 (SD = 0.123) is among the 25% most conserved genes at this stage, and Bmp4 (SD = 0.253) and Gli3 (SD = 0.235) fall near the middle of the range of genes.
Average divergence level and average variation in expression levels among mouse individuals (CoV, as measured with qPCR) are positively correlated for the ES (R = 0.906) and LS (R = 0.854) (Fig 5), and the correlation between divergence level and CoV is significant for the ES (P = 0.019*) and LS (P = 0.016*) after bootstrapping.
The results reported here suggest that the structure of the early stage network (ESN) renders it more robust to perturbation than the later stage network (LSN). Findings also suggest that among individual variation in expression levels is lower for genes regulating early (ES) than late (LS) limb development, and that gene expression levels are heritable. Results of this study also suggest that the expression levels of genes at early stages generally vary less among species. Additionally, results suggest that among individual and among species variation in the expression levels of several model genes are significantly correlated for the early and later stages of limb development. Taken together, these findings suggest a scenario in which a robust ESN buffers among individual variation in gene expression early in limb development, and, as variation is a prerequisite of evolution by natural selection, limits the evolution of early limb development among species. The findings of this study are therefore consistent with the hypotheses that: (1) the structure of the early limb gene network influences the distribution of variation in mammalian limbs among individuals and over evolutionary time [21–23,38–43], and, more generally, (2) the process of early limb development generally precludes the random accumulation of variation in gene expression across the network [21,44]. Results of this study are also consistent with a scenario in which species-specific differences accumulate as development progresses. However, it is important to note that the results of this study are based on a limited number of RNASeq samples per species. Analyses of additional samples are needed to determine the degree to which the RNASeq-driven results of this study are robust to experimental variation among samples. Furthermore, while the network models used in this study are based on solid experimental data [9, 13], the results of this study are only as accurate as the network models being used.
This study did not find a significant correlation between individual gene’s sensitivity to network perturbation or impact on the network when perturbed and variation in expression among individuals. This result could stem from the lack of a relationship between these variables, or from the incompleteness of the limb gene network used in this study. However, this study did find evidence that variation in expression level significantly differs among genes, with some genes being more variable among individuals in a population, and others less so. Shh displays among the greatest variation in expression levels during both the early and later stages of limb development among individuals and species, and has among the least impact on the system when altered. During the early stages of limb development, Gli3 displays the least variation in expression levels and has among the greatest effect on the system when altered.
As population-level variation provides a necessary prerequisite for evolution by natural selection, we might expect genes with the greatest expression variation among individuals to also display the most variation in expression among species. In this study this would be late stage Shh, Bmp4, and Fgf8. This study did find a significant correlation between the variation in the expression of these and the other key model genes among individuals and species during late limb development. Furthermore, of the few studies that have compared gene expression among mammalian limbs, a disproportionally high number have found differences in later stage Shh, Bmp4, and Fgf8 expression among species. Evolutionary changes in Bmp4 expression contribute to digit reduction in horses and jerboa, while evolutionary changes in Shh signaling contribute to digit reduction in pigs [45]. A broader initial range and secondary redeployment of Shh signaling helps generate the unique phenotype of the bat wing [46], and hind limb loss in dolphins is initiated by a disruption in Shh signaling [47]. Shh signaling is also activated exceptionally early during the rapid outgrowth of opossum forelimbs [48,49]. Fgf8 expression is higher in the AER of bat wings than mouse limbs and is also secondarily redeployed later in bat wing development [50]. However, genes beyond Shh, Bmp4, and Fgf8 also display expression differences in mammalian limbs. For example, the expression of 5’ Hox A/D genes differs in bat and kangaroo limbs [51,52], compared to mouse [53]. Clearly more studies of limb development in diverse mammalian species are needed to resolve this issue, but results to date are consistent with alterations in the expression of genes acting during late limb development (E11 to E13 in mouse) including Shh, Bmp4, and Fgf8 frequently contributing to mammalian limb evolution.
In line with the proposed predominance of changes in late limb development in limb evolution, this study found no evidence for the existence of significant variation in the expression of early limb genes that is masked by systems-level processes. However, the expression-based findings of this study do not rule out the existence of cryptic genetic variation in genes with roles in early limb development. Cryptic genetic variation can provide a source of evolutionary potential when uncovered by environmental or genetic perturbations [54,55], and thereby can expedite evolutionary change. Thus, if the genes regulating early limb development possess significant cryptic genetic variation that has been uncovered over evolutionary time, we would expect early limb development to vary among species. Early limb development, as defined in this paper, encompasses establishment of the limb field and the initial outgrowth of the limb. The primary limb segments (e.g., stylopod, zeugopod and autopod) are also likely specified during this time [19,20]. Initial limb outgrowth appears to be generally conserved in tetrapods across large phylogenetic distances [56,57]. Initial limb outgrowth is even conserved in some tetrapods that do not possess limbs in their adult form (e.g., boas, dolphins) [47,58]. The primary segments of the limb are also broadly conserved across limbed tetrapods [8,59]. These observations together with the findings of this study suggest that the genes regulating early limb development either do not possess significant cryptic genetic variation, or that the robustness of the ESN has inhibited the ability of environmental or genetic perturbations to uncover this variation. Whichever the case, the early development of the limb appears to have been relatively conserved over the evolutionary history of tetrapods.
All animal work was conducted according to relevant national and international guidelines. Animals were euthanized using CO2 inhalation followed by cervical dislocation. The University of Illinois IACUC approved this research (protocols #13128, 14159, 14199, 14209).
The starting point for the mathematical models used in this work was the seminal paper by Bénazet et al. 2009 [13]. These models were designed to pertain to the entire limb. Two interconnected feedback loops were incorporated into their model: a fast loop between Grem1 and Bmp4 and a slower loop between Shh, Grem1 and the AER-Fgf’s. Following the findings reported in Sheth et al. 2013 [9], this network was divided into an Early (ESN) and Late (LSN) Stage Network and augmented to also include Hox genes (specifically, Hox A and D genes), Gli3R, Fgf10, and an as yet unidentified repressor, Repressor X. It is important to note that the specific genes, gene interactions, and equations that we include in our models match those presented in Bénazet et al. 2009 and Sheth et al. 2013, which are well supported by experimental, biological evidence.
Mathematical models to describe gene interactions were constructed in MATLAB and were based on ordinary differential equations, which are outlined in S1 Methods, along with the model parameters.
We ran a series of simulations in MATLAB in which we removed interactions between genes. Only one interaction (i.e. link between two genes) was removed in each simulation, and removal simulations were performed for all interactions.
We ran a series of simulations in MATLAB in which we varied the parameter values used in the models. Only one parameter was modified in each analysis.
We performed a series of real-time quantitative PCR (qPCR) assays to quantify the expression levels of the genes that appear in both the ESN and LSN, namely Bmp4, Gli3, Grem1, Shh, and the AER-Fgf Fgf8, in the forelimbs of 71 mouse embryos (outbred ICR strain, Taconic) from 9 litters.
These litters ranged in age from E10 to E13. The limbs of the animals in these litters were staged according to Wanek’s staging guide, which divides mouse limb development into 15 stages [60]. Embryonic limbs from limb ridge (Wanek stage 1) through bud formation (Wanek stage 4; 4 stages total; E10 –E11) were grouped into an “early stage” (ES) for analyses, while limbs in the paddle stages of development (Wanek stages 5–8; 4 stages total; >E11 –E13) were grouped into a “late stage” (LS). Limb samples were evenly spread over all stages. The Coefficient of Variation (CoV), which is the standard deviation divided by the mean, was used to quantify variation in expression level for a given gene for early and later limb development [61]. Additional details for the qPCR analyses are in S1 Methods, and standard and dissociation curves for each gene are in S2 Methods.
Bartlett’s Test was used to compare the variance of expression levels in the ES and LS [61]. ANOVA was used to examine the contribution of litter membership (i.e., heredity) to observed patterns of gene expression [61].
The relationship between among individual variation in gene expression level with the number of genes whose removal alters expression of the gene in question by 10% or more from the default value (from simulation analyses) and whose values deviate by 10% or more from the default values when the gene in question is removed (from simulation analyses) was statistically assessed using Least-Squares Regression [61]. The average levels of variation in measured gene expression during early and later development (CoV) were statistically compared [61]. To determine the significance of the observed differences in CoV, we used a Monte Carlo approach, in which we shuffled the observed CoV from early and later development, to generate a null distribution of CoV differences between them. Specifically, we pooled all replicate CoV irrespective of developmental stage, then randomly drew, with replacement, two samples equal in size to the measured early and later samples, respectively. We used as a measure of significance the proportion of 10,000 replicates in which the difference between CoV’s of randomly shuffled samples was greater than or equal to the observed difference.
Embryonic mice, opossums, bats and pigs with early (ES) or late (LS) stage forelimbs were obtained from a variety of sources (see S1 Methods). Forelimbs for the ES were harvested at Stage 14 for bat, E11 for mouse, Stage 28 for opossum and E22 for pig [62–65]. For the LS, forelimbs were harvested at Stage 15 for bat, E12 for mouse, Stage 29 for opossum and E26 for pig.
Limbs were removed from embryos and stored in RNALater in -20°C until further processing. RNA was extracted from tissues using E.Z.N.A. Total RNA Kit I (OMEGA bio-tek #R6834), and converted into RNASeq libraries with the Illumina TruSeq RNA Sample Preparation Kit (Illumina RS-122-2001). Libraries were sequenced on an Illumina HiSeq 2500 housed in the Roy G. Carver Biotechnology Center at the University of Illinois. Resulting reads were processed, aligned to published genomes or de novo assemblies, and gene expression levels assessed (see S1 Methods). All data from this study have been deposited in the Gene Expression Omnibus (GEO) with the accession number GSE71390.
We analyzed the conservation of the gene expression profiles of bat, mouse, opossum, and pig across embryonic limb development, using the mean of all species pairwise Spearman coefficients (see S1 Methods). The relationship between these average species pairwise coefficients and among individual gene expression variation was assessed using Pearson Product-Moment Correlation [61]. To account for the variation among samples we used a bootstrap approach. Specifically, for the comparisons of the average species pairwise coefficients and among individual gene expression variation, we resampled, with replacement, the CoV’s (among individuals) and mean-standardized expression levels (among species) and recalculated the Pearson product-moment correlation coefficient (R) between the resampled CoV’s and mean-standardized expression levels. We used as a measure of significance the proportion of 10,000 replicates in which the calculated R was greater than or equal to zero. We also ordered the species pairwise Spearman coefficients for each individual gene from highest (most conserved) to lowest (most divergent).
|
10.1371/journal.pcbi.1002154 | Change in Allosteric Network Affects Binding Affinities of PDZ Domains: Analysis through Perturbation Response Scanning | The allosteric mechanism plays a key role in cellular functions of several PDZ domain proteins (PDZs) and is directly linked to pharmaceutical applications; however, it is a challenge to elaborate the nature and extent of these allosteric interactions. One solution to this problem is to explore the dynamics of PDZs, which may provide insights about how intramolecular communication occurs within a single domain. Here, we develop an advancement of perturbation response scanning (PRS) that couples elastic network models with linear response theory (LRT) to predict key residues in allosteric transitions of the two most studied PDZs (PSD-95 PDZ3 domain and hPTP1E PDZ2 domain). With PRS, we first identify the residues that give the highest mean square fluctuation response upon perturbing the binding sites. Strikingly, we observe that the residues with the highest mean square fluctuation response agree with experimentally determined residues involved in allosteric transitions. Second, we construct the allosteric pathways by linking the residues giving the same directional response upon perturbation of the binding sites. The predicted intramolecular communication pathways reveal that PSD-95 and hPTP1E have different pathways through the dynamic coupling of different residue pairs. Moreover, our analysis provides a molecular understanding of experimentally observed hidden allostery of PSD-95. We show that removing the distal third alpha helix from the binding site alters the allosteric pathway and decreases the binding affinity. Overall, these results indicate that (i) dynamics plays a key role in allosteric regulations of PDZs, (ii) the local changes in the residue interactions can lead to significant changes in the dynamics of allosteric regulations, and (iii) this might be the mechanism that each PDZ uses to tailor their binding specificities regulation.
| PDZ domain proteins (PDZs) act as adapters in organizing functional protein complexes. Through dynamic interactions, PDZs play a key role in mediating key cellular functions in the cell, and they are linked to currently challenging diseases including Alzheimer's, Parkinson's and cancer. Moreover, they are associated with allosteric regulations in mediating signaling. Therefore, it is critical to have knowledge of how the allosteric transition occurs in PDZs. We investigate the allosteric response of the two most studied PDZs, PSD-95 and hPTP1E, using the perturbation response scanning (PRS) approach. The method treats the protein as an elastic network and uses linear response theory (LRT) to obtain residue fluctuations upon exerting directed random forces on selected residues. With this efficient and fast approach, we identify the key residues that mediate long-range communication and find the allosteric pathways. Although the structures of PSD-95 and hPTP1E are very similar, our analysis predicts that their allosteric pathways are different. We also observe a significant change in allosteric pathways and a decrease in binding affinity upon removal of the distal α3 helix of PSD-95. This approach enables us to understand how dynamic interactions play an important role in allosteric regulations.
| Allosteric regulation orchestrates functional behaviors in biological networks through appropriate switches. From a biochemical perspective, allostery can be described as a perturbation at one place in a protein structure, such as the binding of a ligand that alters the binding affinity of a distant site or enzymatic activity [1]. Several models have been suggested for explaining the ‘allosteric mechanism’. Models of conformational transition between co-existing states such as the MWC model of Monod [2], and the ‘induced fit’ KNF model of Koshland [3] were the first views among them. They described allostery as a binding event that causes conformational change via a single propagation pathway [4]. A new view of allosteric transitions supported from NMR studies, referred to as the ‘population shift’ model, has replaced the MWC and KNF models [5]–[8]. The population shift models claim that a protein in the unliganded form exhibits an ensemble of conformational states and ligand binding leads to a redistribution of the population of these states. In this view, it is important to explore how protein dynamics might contribute to allostery and make communication possible within a protein. Unlike the classical allostery models, the population shift-models also suggest that allostery can be mediated without any significant conformational change [9]–[15] but rather from changes in dynamics.
Moreover, recent experimental and theoretical evidences indicate that allostery is not limited to multi-domain proteins or complexes [5] and it may even be a fundamental property of all proteins, even single domain proteins. In single domain proteins, it is evident that residues that are energetically connected through structural rearrangements and dynamics lead to allosteric regulation [6], [11], [15]–[17]. More importantly, studies on single domain protein PDZ (post-synaptic density-95/discs large/zonula occludens-1) have indicated that allostery can arise not only from large conformational changes, but also from changes in dynamics [12], [14].
Indeed, PDZ domain proteins (PDZs) are the most studied system for understanding single domain allostery [11], [16], [18]–[25]. PDZs are small protein-protein interaction modules and typically recognize specific amino acids in the C-terminal end of peptide motifs or proteins [26]–[28]. Various studies on several PDZs, including statistical coupling analysis (i.e. sites that have correlated mutation based on evolutionary information) [16], [29], [30], molecular dynamics [11], [22], [31]–[33], normal mode analysis [34], [35], NMR relaxation methods and site directed mutational analysis [12], [18]–[20], [25], [36] have shown that several PDZs exhibit allosteric behavior that appears to connect incoming signals, notably binding to recognition motifs present on an upstream partner, to downstream partners [11], [16], [18]–[25]. In many different cellular contexts, PDZs function to transduce these binding events into favorable domain-domain assembly of complexes [14]. Thus, it is critical to understand the residues involved in these allosteric pathways in order to modulate the PDZ mediated interaction in cell regulation especially those in disease pathways. Moreover, a recent experimental study by Petit et al. [12], has confirmed yet another strong allosteric power of one of the PDZs: the hidden dynamic allostery. The removal of the non-canonical third helix (α3) in PSD-95 (PDZ3), which lies outside of the binding pocket, reduces the binding affinity drastically due to a change in side chain dynamics upon truncation, indicating the role of entropy and dynamics in allosteric regulation. More interestingly, further investigation has shown that the removal of this distal α3 disrupts the communication between PDZ3 and SH3-GK, which modulates the binding of Disc large protein (Dlg) to the localization protein GukHolder [37]. Therefore, the hidden dynamic allostery related with α3 is indeed a regulatory module within the context of larger interdomain interactions.
In summary, PDZs do not solely act as simple scaffold proteins. On the contrary through dynamics, they propagate signals to functionally important distant sites for intramolecular and intermolecular interactions [16]. They all have the same conserved structure and similar sequences [16], yet different PDZs have evolved different dynamics properties tailored to mediate different functions in the cell [14]. Thus, it would be very important to understand how signals are passed from one residue to another within the network of PDZs and how the sequential and structural variations alter the allosteric pathways for those allosteric PDZs [11], [18], [20]–[24]. Here we would like to tackle this problem with our new method called perturbation response scanning (PRS) [38], [39]. PRS treats the protein as an elastic network and uses linear response theory (LRT) to obtain residue fluctuations upon external perturbation. By sequentially exerting directed random forces on single residues along the chain of the unbound form and recording the resulting relative changes in the residue coordinates using LRT, we can successfully reproduce the residue displacements from the experimental structures of bound and unbound forms. The method is well established and tested for 25 proteins that display a variety of conformational motions upon ligand binding, including shear, hinge, allosteric, and partial refolding as well as more complex protein motions [39].
In the present study, we investigate the allosteric transitions by analyzing response fluctuation profiles upon perturbation on binding site residues by PRS. We focus on two widely studied PDZs: the third PDZ from the post-synaptic-density-95 (PSD-95 PDZ3) and the second PDZ from the human tyrosine phosphates 1E (hPTP1E PDZ2). The results from our computationally inexpensive and effective approach successfully identify the dynamically linked allosteric residues obtained from experiments (NMR or mutagenesis techniques) [12], [18]–[20], [25], [36] as well as evolutionarily coupled residues from sequence-based statistical approaches [16],[29],[30] and key residues predicted from molecular dynamics, normal mode analysis and protein energy-based networks [11], [22], [31]–[35], [40]. As a further test, we construct the communication pathway between these residues that might be responsible in transmitting allosteric signals. We achieve this through linking residues that show similar directionality of motion upon perturbation of binding sites. Interestingly, the constructed allosteric pathway indicates a strong structural residue coupling network. Moreover, we observe that the two PDZs, PSD-95 and hPTP1E, have distinct allosteric pathways despite their structural similarity, indicating the role of dynamic coupling in these domains [14], [35], [41]. The residues in the allosteric pathway of PSD-95 are homogenously distributed along the secondary structural motifs while the allosteric pathway of hPTP1E shows more localization around in regions of β1–β2 loop, β2 and β3 strands and the region of β5 strand and the α2 helix, missing the region of the α1 helix. The differences in the allosteric pathways of these two PDZs indicate the critical of role of dynamic coupling in PDZ domains and that differences in residue sequences within the same fold can lead to different dynamic coupling. Indeed, PDZs master this to mediate different cellular functions in different parts of the cell [14]. In addition to that, our PRS analysis indicates that the allosteric pathway of PSD-95 significantly alters upon removal of the distal third helix (α3 helix). This indicates that local changes in the network alter the directionalities of correlated motion, which may lead to a change in binding affinity [35], [42]. Strikingly, when we incorporate the change in backbone dynamics into the docking computation through generating multiple conformations by PRS, we also observe an increase in binding energies upon removal of the third helix.
Our objective is to apply a computational approach, perturbation response scanning (PRS), to identify the network of dynamically important residues and propose a possible pathway responsible for intramolecular signaling. As we mentioned earlier, PRS combines the elastic network model with linear response theory to compute the residue fluctuation profile of an unbound conformation upon exerting a random external force on a residue, and it is shown to be very successful in capturing binding-induced conformational changes [39]. When a ligand approaches a receptor, it exerts forces around binding pockets, inducing certain dynamical changes. Here, we utilize PRS to mimic the nature of a binding event by exerting forces on the binding sites of an unbound conformation. Thus, we analyze the residue response fluctuation profile upon exerting random forces on binding sites of unbound conformations and identify the residues showing distinctive responses (i.e. higher fluctuation than the average fluctuation response) upon perturbing the residues at the binding sites. (See Materials and Methods for details.)
Elastic network models (ENMs) are utilized to explore allosteric behaviors in proteins [43]–[51]. ENMs are based on a purely mechanical approach, viewing a protein structure as an interconnected series of springs between interacting residue pairs. They provide information on equilibrium fluctuations and the various contributions to those fluctuations from different modes of motion. Moreover, by introducing a specific perturbation to the system and measuring its dynamic response, ENMs can provide detailed information about the energy landscapes, beyond the correlations between equilibrium fluctuations. To this aim, there are new modified ENMs developed whereby perturbations are introduced through modifying effective force constants [49], [50], distances between contacting pairs [52], or both [45]–[46]. Most of these analyses are focused on changes in the most functionally related mode (i.e. usually the slowest modes) upon perturbations. Although an ENM approach itself, our PRS model differs in two aspects. First, we introduce perturbations by inserting random external forces on the nodes of unbound conformations, (i.e. α-carbons) instead of modifying the distances between pairs of nodes or spring constants. This enables us to exert external forces on the binding sites (i.e., random Brownian kicks) and analyze the residues affected by the perturbation on the binding sites similar to the natural allosteric regulations where an approaching ligand induces certain dynamical changes in distal parts of the protein. Second, PRS uses the entire Hessian matrix to compute the residue displacement response upon exerting random forces on the selected residues. The allosteric regulation in small domain proteins like PDZs can arise through changes in dynamics [11], [14], unlike large conformational changes observed in large systems such as GroEL [47], [50] and myosin [53]. Therefore, more than one normal mode can contribute to allosteric regulations. In that respect, the advantage of using the full Hessian matrix in PRS can induce several related modes upon perturbation at the binding site.
Mutagenesis and NMR relaxation methods demonstrated that a network of residues exists that has a dynamic response upon ligand binding in both hPTP1E PDZ2 and PSD-95 PDZ3 [12], [19], [20], [25], [36]. Thus, we applied our approach to the unbound structures of two PDZ domain proteins: hPTP1E (PDB entry: 3LNX) and PSD-95 (PDB entry: 1BFE) and computed the allosteric response ratio χj for each residue, which is the normalized average mean square fluctuation response of residue j upon perturbing only the binding site residues over the mean square average response of the same residue j obtained by perturbations on all residues. Thus, the index of allosteric response ratio χ enables us to identify residues that are more sensitive to perturbation around the binding pocket. Figure 1 presents the allosteric response ratio profiles of (A) hPTP1E and (C) PSD-95 and the corresponding color-coded ribbon diagrams of these two proteins. Experimentally identified residues are marked with red dots. The ribbon diagrams of (B) hPTP1E and (D) PSD-95 are colored based on the allosteric response ratio, χj, using a spectrum of red (the highest mean square fluctuation response) to orange, yellow, green, cyan and blue (the lowest response). The residues with the highest allosteric response ratio (χj>1.00) are shown as stick representations. Particularly, those in agreement with the experimental analysis are labeled. Overall, there is a good agreement with experimentally identified allosteric residues and those predicted by our approach. Using χj>1.00 as a threshold value for the allosteric response ratio, we predicted 6 out of 10 experimentally identified allosteric residues for hPTP1E [25] and similarly 8 out of 11 for PSD-95 [19] (i.e. the predicted residues correspond to the peaks in the allosteric response ratio profiles). We would like to note that we also tested our approach in another allosteric PDZ domain, SAP97 (PDB entry: 2AWX) which shows slight conformational change upon binding [18]. Using the same threshold value for χj>1.00, we were able to distinguish not only the residues near canonical binding sites but also those distant from the binding site (Table S1), indicating the predictive power of PRS in identifying allosteric residues.
To our knowledge, all of previous computational studies including all-atom molecular dynamics [31], [32] and the rotamerically induced perturbation method (RIP) [11] identified certain critical residues using the previous NMR structure of hPTP1E (See Table S2 for predictions based on the previous NMR structure by different methods). Here, we apply our computational approach to the recently reported high-resolution crystal structure of hPTP1E PDZ2 [25], indicating that new bound and unbound structures deviate from previously determined NMR structures of hPTP1E and there are very minor structural changes in PDZ2 upon peptide binding.
The previous study of the RA-GEF2 peptide binding to hPTP1E PDZ2 using NMR relaxation technique identified residues that have significant changes in their side-chain dynamics upon peptide binding [20], [36]. This study also revealed that there are two distal surfaces physically linked to the peptide-binding site: (i) “distal surface 1 (DS1)”, which contains residues in the N terminal of β6 and the anti-parallel β strand formed by β4 and β5 (Val61, Val64, Leu66, Ala69, Thr81, and Val85), and (ii)”distal surface 2 (DS2)”, located next to helix α1, consisting of residues Ala39 and Val40. In the recent study Zhang et al. [25] identified 10 residues (Ile6, Ile20, Val22, Val26, Val30, Ile41, Val61, Val64, Val78, Val85) that have significant changes in side-chain dynamics upon binding both RA-GEF2 and APC peptides to PDZ2. These identified residues overlap with the findings of their previous study and they are located in the region of the binding site (Ile20, Val22, Val26 in the β2 strand, and Leu78 in helix α2), DS1 (Val61, Val64 and Val85), and in DS2 (Ile41). The highest allosteric response ratios obtained by PRS are also observed for the same residues except Val26 and Val64 (Figure 1A). Other residues that give high mean square fluctuation response (χj>1.0) are summarized in more detail in Table 1, and those which agree with the experimentally identified ones [25] are highlighted in boldface. We also construct a two-way contingency table that presents the pattern matching between the experimentally identified residues and our prediction by PRS using a Fisher's exact test. The resulting p-value of hPTP1E, 2.9E-2, from the test indicates that there is a statistically significant matching between experiment and our method (Table S4).
In addition, the residues critical in allosteric pathways are characterized via statistical coupling analysis (SCA) of an evolutionary network using a large and diverse multiple sequence alignment of the PDZ domain family. Using the SCA method, Lockless and Ranganathan [16] predicted a set of residues within the family of PDZ domains that communicate signals through the protein core. When we compare our predictions with those obtained from SCA, nine residues (Ser17, Ile20, Gly24, Gly25, Gly34, Ala46, Val61, His71 and Val85) emerge as the residues with high allosteric response ratio (χi) that are in agreement with the evolutionary network residues of hPTP1E [16], [30], [54]. The Fisher' exact test based on our method and SCA provides a p-value of 5.0E-4, indicating a high level of agreement. (Table S4).
The residues identified with high allosteric response ratios for PSD-95 PDZ3 are also in good agreement with double mutant cycle analysis [19]. The two-way contingency table based on experiment and method resulted in a high level of pattern matching, with a Fisher's exact test p-value of 1.5E-3 (Table S4). The mutational study of Chi et al. [19] indicates that the three positions Gly329, Val362, and Ala376 yield significant energetic coupling interactions with His372. In fact, among these coupling interactions the interaction between His372 and Val362 show long-range energetic coupling in the PSD-95 PDZ3 domain. As shown in Figure 1B, PRS analysis also captures the importance of the long-range energetic coupling interaction between His372 and Val362 of the PSD-95 PDZ3 domain. In this context, it is worth noting that studies based on a non-equilibrium perturbation-based molecular dynamics technique, called anisotropic thermal diffusion (ATD) [22], and the rotamerically induced perturbation method (RIP) [11], [41], also reported a complete signaling pathway of PDZs including PSD-95. ATD analysis proposed a signaling pathway between His372 and Ile335 that passed through Ile327 and Phe325 [22]. RIP analysis has also shown that some PDZs have more dynamic responses than the others and this was highly coupled with evolutionary SCA analysis [11]. The general pattern derived from both perturbation based MD analyses agreed with that obtained from PRS (See details for Table S4). The list of residues identified as allosteric residues with these different methods for these two PDZs is presented in Tables S2 and S3.
Furthermore, the energetic coupling residues (Gly329, Leu323, Ile327, His372, Ala376, Gln384) in PSD-95 were also successfully identified using an ENM-based structural perturbation (SPM) method [33], [47], [49], [55] based on exploring the propagation of the response of a local perturbation at a given residue to all other residues in a given structure. As we mentioned earlier, the basic premise behind SPM and PRS methods is similar except the harmonic springs connected to residues are changed by a small amount in SPM whereas the force is directly applied to residues in PRS. In addition to that, SPM focuses on changes in the single mode upon perturbation. It is usually the 1st slowest mode in large proteins [52]. However, in the case of the small domain protein of PSD-95, rather than the 1st mode, the 13th and 20th slowest modes significantly overlap with binding induced fluctuations [33]. On the other hand, PRS does not use the bound structure. PRS uses the Hessian of the whole unbound conformation and it automatically includes the modes that induce a response vector upon exerting forces on the binding site residues.
By linking the residues involved in allosteric regulations with respect to their response behavior, we can construct the allosteric pathways with PRS. PRS enables us to measure the relative directionality between the responses of a pair of neighboring residues to a perturbation. (i.e. the alignment of their response vectors). If the residues collectively move in line, their directionality should be parallel. After obtaining the directionality of different pairs of residues, we carry out a systematic analysis of the residues with the highest allosteric response ratio. For these residues, we search all possible interactions with a window size of 3 and identify residue pairs that collectively move in line together. A pathway is constructed by linking the sequential pairs showing similar directional response upon perturbation. Each constructed pathway is weighted based on alignment angles (i.e. directional similarity) between linking residues. Then we select the pathway with maximum total weight.
By this analysis, the allosteric pathway constructed for hPTP1E PDZ2 follows through the connections Ser 17 → Val22 → Gly25 → Arg31 → Ile35 → Val61 → Leu64 → Thr70 → Ala74 → Leu78 → Thr81 → Leu88 (Figure 2A). Interestingly, the residues Val22, Val61, and Leu78 are located at the critical regions determined by the mutational analysis [25]. Since the model in the present study is low-resolution, we identify the residue Val22 that is near residue Ile20. The experimental mutational analysis showed that a change at Ile20 resulted in extensive changes in side chain dynamics while mutations at residues Ile35 and His 71 had a limited response in dynamics. Thus it is concluded that Ile20 might act as a hub that is energetically and dynamically important for transmitting changes in dynamics throughout the PDZ domain [36]. When we analyze the directionality preference of this residue with each residue identified for the most highly weighted pathway, we find that Ile20 collectively moves together with each of them, indeed acting as a hub in our dynamic network analysis. Moreover, the PRS pathway shows a remarkably high similarity (Ser17, Gly25, Ile35, Val61, His71, and Val75) with the statistical coupling analysis obtained by Lockless and Ranganathan [16].
As shown in Figure 2B, the most highly weighted pathway for PSD-95 is obtained through connections Ile314 → Ile327 → Ile338 → Ala347 → Leu353 → Val362→ Leu367 → His372 → Lys380 → Val386 → Glu396. Interestingly, Val362 [16], [19], Lys380, and Val386 [16] yield significant energetic coupling interactions with His372 which are confirmed by mutagenesis studies. While the general pattern of signal propagation predicted from our method agrees with that inferred from the SCA analysis [16] there are some differences. The discrepancy between our model and the two proposed pathways by SCA may result because SCA analysis investigates the signaling pathway originating from a single residue, His372. However other residues at the binding pocket may be important for intramolecular signaling. Our analysis uses response profiles obtained by sequentially exerting a random force at a single residue along all the residues at the binding site. Thus, our approach might lead to the prediction of extra residues, such as Lys380, that interacts with the peptide and is near His372. Our model does not include Phe325 in the allosteric pathway, yet it finds Ile327, which is near residue 325. Moreover, MD analysis has shown that the mutation of Ile327 to Val leads to a dramatic signal reduction of Phe325, showing that position 327 is involved in mediating the signal pathway and highly linked with Phe325 [22].
Overall, when we compare the allosteric pathways of the two different PDZs, PSD-95 and hPTP1E, we see a clear difference (Figure 2C). There are some overlap regions between the two PDZ domains including residues in the β2 and β3 strands, the loop between β4 and β5 strands, and the C-terminal of the α2 helix. However, the predicted allosteric pathway of PSD-95 has a more homogeneous distribution through N-terminal to C-terminal, whereas the pathway of hPTP1E seems more localized, especially in regions of β1-β2 loop, β2 and β3 strands and the region of β5 strand and the α2 helix, missing the regions around the α1 helix. Indeed, the allosteric behavior of Ala347 in the α1 helix has also been found by SCA [16] and other MD analysis [22]. This comparison indicates that these two PDZs with similar sequences and structures have different allosteric behavior, indicating the role of dynamic coupling in single domain allostery. Thus, slight changes in the residue network changes dynamic coupling, which can lead to distinct allosteric paths.
A recent experimental study [12] provided further support that allosteric communication can be driven by the network of residue interactions of PSD-95 without any conformational change. To investigate this phenomenon, they removed the non-canonical C-terminal third helix (α3, residues 394-399). Strikingly, removal lowers the binding affinity 21-fold and has a significant effect on the internal dynamics of PDZ3, even though it lies outside of the binding site and does not make direct interactions with the binding C-terminal peptide (CRIPT) residues.
Using PRS, we also analyzed the truncated PSD-95 structure and investigated the impact of removal of helix α3 in the allosteric communication pathway. The most highly weighted pathway of the truncated structure is presented in Figure 3. Comparison of the pathway of PSD-95 (Figure 2B) and the truncated one (Figure 3) computed by PRS remarkably shows that the removal of the α3 helix significantly alters the allosteric pathway, indicating that the interactions responsible in transmitting intramolecular signals are being lost upon truncation of helix α3. For the truncated PSD-95 structure, the most highly weighted pathway has been identified through connections Ile314 → Ile 327 → Glu334 → His372 → Lys380 → Ile388, which is shown in Figure 3. Some of the interactions specifically located in the α1 helix and the loop between the β4 strand and the α2 helix predicted for the full PSD-95 were lost after removal of the α3 helix. Qian and Prehoda [37] showed that truncation of a portion of the α3 helix modulates and initiates the binding of Dlg to the localization protein GukHolder. Therefore, it is reasonable to say that this non-canonical α3 helix has a significant biological role in this allosteric regulation and the fact that the α3 helix is involved in the allosteric pathway obtained by PRS supports this.
In our recent work, we analyzed the dynamics of PDZs showing different binding specificities and showed that we can discriminate the binding specificity of PDZs based on their dynamics [35]. Within this picture, it is not surprising to see a change in binding affinity of PSD-95 upon truncation of the distal helix α3, because this leads to a change in dynamics. In order to investigate this any further, we also investigate the changes in the binding affinity upon removal of the α helix using docking techniques where we incorporate the changes in dynamics of PSD-95 into docking.
Computational docking methods are commonly used to identify the correct conformation of ligand-bound proteins along with their binding energy. However, docking algorithms predict incorrect binding modes or energies for about 50–70% of all ligands when the receptor is kept in a single conformation [56]. This is especially critical for PDZ whose dynamics play a key role in peptide binding specificity [35]. Some docking methods also incorporate the side chain flexibility of the receptor around binding pockets [57]–[60]. In our previous study [42], we incorporated the backbone flexibility of PDZs by generating multiple receptor conformations through restrained-replica exchange molecular dynamics (REMD) runs where the restraints are obtained by binding-induced elastic network modes. In this present study, we first generate multiple receptor conformations using the response vectors obtained upon perturbation of each residue via PRS. This provides us more computational efficiency in exploring conformational space. Then, we dock these multiple receptor conformations of PSD-95 and the truncated one against its native peptide (CRIPT) using RosettaLigand [58], [60]. RosettaLigand is docking software that computes the best-docked pose through a Monte Carlo minimization procedure in which the rigid body position and orientation of the small molecule and the protein side-chain conformations are optimized simultaneously. The lowest binding energy scores and corresponding peptide RMSDs of PSD-95 and the truncated third alpha helix of PSD-95 structures interacting with the CRIPT peptide are summarized in Table 2 for two different docking cases, (i) using only bound crystal structure (PDB code:1BE9) and (ii) using ensemble of structures obtained by applying PRS to the crystal structure. We cannot see this difference in binding affinities when we perform single receptor docking by using only the full and α3 helix truncated forms of the crystal structure. When we use PRS generated multiple receptor conformations to predict binding energies of PSD-95 and the truncated one, we find that the binding energy increases upon truncation of the C-terminal third alpha helix (α3 helix) as also observed experimentally [12]. This analysis indicates that the residue networks and their related dynamics indeed play a key role in binding affinities of PDZ. Our PRS analysis suggests that the significant change in the dynamics pathway of residue communication, caused by truncation of the α3 helix, leads to a change in binding affinity of its native peptide.
Allosteric responses in PDZs usually arise, because a perturbation at one site is transferred to the distal part of the protein through a network of residue communications. Here we investigate how the perturbation of a residue at the binding site is transferred through the dynamics of the residue network interactions. Thus we investigate the allosteric response of the two most investigated PDZs, PSD-95 and hPTP1E using our low resolution dynamics approach PRS. PRS is based on ENM where it uses only the topology of the given structure, and then using linear response theory, it computes the response fluctuation vector of each residue in the chain upon exerting a random force on a single residue. Using PRS, we compute the allosteric response ratio for each residue, which is the normalized average mean square fluctuation response upon perturbation. Most of the residues that are identified experimentally as residues in allosteric pathways indeed show high allosteric response ratios, indicating the consistency and usefulness of the PRS method for extracting the residues in the signaling pathway. Since PRS not only gives the mean square fluctuation of the response but also its directionality, we construct the allosteric pathway by linking the residues aligning in the same direction upon perturbations. Interestingly, our analysis has shown that the allosteric pathways of PSD-95 and hPTP1E are distinctively different from each other, despite the fact that they have similar structures. Likewise, we also observe a significant change in the allosteric pathway upon truncation of the distal α3 helix of PSD-95. Moreover, our flexible docking analysis where we generate an ensemble of multiple receptor conformations by PRS shows an increase in binding energy upon truncation. Overall, these results strongly suggest that local changes in residue network interactions can lead to changes in dynamics in allosteric regulations and various PDZs grasp to mediate different functions in the cell.
We analyze unbound structures of hPTP1E (3LNX) [25] and PSD-95 (1BFE) [61] in this study. The backbone root mean square deviation (RMSD) between hPTP1E and PSD-95 structures is 1.89 Å, while the sequence identity between pairs is only 36%. The all-atom RMSD between unbound and bound structures of PSD-95 is 1.13 Å (backbone RMSD = 0.73 Å) while that of hPTP1E is 1.03 Å (backbone RMSD = 0.46 Å).
PRS is based on sequentially exerting directed random forces on single-residues along the chain of the structure and recording the resulting relative displacements of all the residues using LRT. The model views a protein structure as a three-dimensional elastic network. The nodes of the elastic network are Cα atoms of each residue where identical springs connect the interacting α-carbons in their native fold. In all elastic network models (ENMs), all residue pairs are subject to a uniform, single-parameter harmonic potential if they are located within an interaction range, or cutoff distance, rc. The major drawbacks of using cutoff distances are: (i) they are generally taken arbitrarily and (ii) their optimal values vary for different proteins [62], [63]. Instead of using any arbitrary cutoff distance, the interaction strength between all residue pairs can be weighted by the inverse of the square distance of their separation [63], [64]. We modify PRS by applying the concept of inverse square dependence for the interactions between residue pairs [63], [64] and introducing specificity between bonded and non-bonded interactions [35]. We tested the modified version on previously analyzed [39] 25 unbound protein structures that make various conformational changes upon bindings, and the results showed that the modified version successfully captures these conformational changes.
The free-body diagram of the central Cα atom of each sphere exhibits all of the pairwise interaction forces generated by the coordinating Cα atoms as schematically illustrated in Figure 4A. Each Cα atom must be in equilibrium under the action of interaction forces in the absence of external forces. The sum of forces on residue i along the x-, y-, and z-directions must be equal to zero under native state conditions,(1)where fij is the internal force on site i due to its interaction site j, ,. is the angle between the x axis and the line of action of fij, rij is the instantaneous separation vector between sites i and j and Xi, Yi and Zi are the components of the instantaneous position, Ri. The force balance can be generalized to the complete set of N sites (i.e. sites are Cα atoms of a protein) and M interactions (i.e. an interaction between any two Cα atoms is determined if the distance between two Cα atoms is less than the cut-off distance) as(2)where B is the directional cosine matrix.
If there are external forces acting on a set of residues of the folded structure as shown in Figure 4B, the force balance of the complete set of N sites and M interactions takes the following form:(DOC)(3)where Δf is the residual interaction forces and ΔF is a 3Nx1 vector containing the external force components at each residue. The native structure may undergo conformational changes about the equilibrium state under the action of these forces. During this process, the positional displacements ΔR and the bond deformations Δr are geometrically compatible. The relation between the positional displacement vector and the bond distance is given by(4)where [B]T is the transpose of B.
Within the scope of an elastic network of residues that are connected to their neighbors with springs, the interaction forces, Δf, are related to the bond distance through Hooke's law by(5)where the coefficient matrix K is diagonal. Although the entries of K are taken to be equivalent in the original method [38], we introduce two different spring constants for the residue interaction network for bonded and non-bonded interactions, γb and γnb. The spring constant of the bonded part (γb) is taken as 1. For the non-bonded part (γnb), the interactions between residue pairs i and j are weighted by the inverse square of the distances, rij (as 8/rij2).
Moreover, the work done by the external forces ΔF is equal to the work done by the internal forces Δf so substituting Equations (4) and (5) into Eq. (3), we obtain(6)
Let's note that the term in Eq.(6) is also equivalent to the Hessian (H) [65].
On the other hand, one may choose to perturb a single residue or a set of residues, and calculate the response of the residue network through, (7)orwhere the ΔF vector contains the components of the externally applied force vectors on the selected residues.
In this study, first we apply a force as a unit vector on residue i along 7 directions (i.e. in x-, y-, z-, both x- and y-, both x- and z-, both y- and z-, all x, y, z directions. Then, we build a perturbation response matrix that includes average displacement ΔR for each residue j due to a force applied on residue i, (8)where the magnitude of positional displacements for residue j in response to a perturbation at residue i is defined as,(9)
In order to predict which residues are critical in allosteric pathways, we distinguish the residues exhibiting significant fluctuation upon perturbation on binding site residues. Therefore, we define an index called the allosteric response ratio, χj for each residue, which is the ratio of average fluctuation response of the residue j upon perturbations placed on binding site residues to average response of residue j upon perturbations on all residues, shown as:(10)where Aij is the response fluctuation profile of residue j upon perturbation of residue i. The numerator is the average mean square fluctuation response obtained over the perturbation of the binding pocket (BP) residues, whereas denominator is the average mean square fluctuation response over all residue perturbation. Thus, NBP is the number of residues in the binding pocket and NBP1 and NBPm correspond to residue indexes in the binding pocket (residues 320-328 and 371-380 for PSD-95 and residues 16-23 and 70-79 for hPTP1E). To identify the critical residues in the allosteric pathway, for each residues we compute χj in each perturbed direction and take into account of the maximum value of χj. Then, we sort out all χj and select the residue positions by setting a threshold of 1.0 or better. To understand how the sensitivity and specificity change, we predict the allosteric residues by varying the threshold of response ratio lower or higher than 1.00. We found that taking a threshold value lower than 1.0 gives same experimentally identified allosteric residues to ones obtained by using χj>1.00 as a threshold value (Table S5).
We should note that the procedure has been also repeated using several random directions, rather than the 7 directions and we observed that our predictions do not change significantly. The schematic representation showing how we identify allosteric binding sites can be found in Figure 4C.
While PRS is a residue-based low-resolution approach, the essential dynamics analysis [66] is carried out on all-atom molecular dynamics (MD) trajectories to support the validity of the methodology. The details of the analysis are explained in Text S1. The comparison of residues that give the highest mean square fluctuation response (χj>1.00 for PSD-95) upon perturbation with respect to the coarse-grained approach and the essential dynamics analysis is presented in Table S6. Overall, 82% of predicted residues from the essential dynamics analysis of all-atom MD trajectories are the same as those obtained by our low-resolution model (see Text S1 for more details). Moreover, the residues found by the coarse-grained approach that do not overlap with those of the all-atom approach are sequentially in close proximity to the residues identified by both approaches.
PRS can be used to measure the degree of collectivity of the response of a group of neighboring residues to a perturbation on any residue. This enables us to construct an allosteric pathway through linking those residues showing similar response upon perturbations of the binding site.
To understand the nature of the response, the submatrix of residue k in response to perturbations in i from the inverse of the Hessian (See Equation 7) matrix can be decomposed into its eigenvalues and eigenvectors:(11)
If the residues collectively move in line they have a single dominant eigenvalue and their corresponding eigenvectors should be parallel, indicating that they move cooperatively in the same direction. Therefore, to compare if the responses of two residues are same, we check the dot product of their corresponding eigenvectors,(12)where θ is the angle between the two eigenvectors. After obtaining the directionality of different pairs of residues upon perturbations on the binding site, we carry out a systematic network analysis using only the residues that give the highest fluctuation response upon perturbation. For these identified residues, we use a window size of 3 (i.e. if the residue 320 shows the highest mean square fluctuation response, the residues 319, 320, and 321 are taken into account), and search extensively to find residue pairs in sequence that move collectively upon perturbation. To this aim, we first calculate the overlap coefficients of the residue pairs by using the dot product of response vectors (Eq. 12). Using a cut off value of 0.98, we find the residue pairs that move in the same direction. Importantly, this means we identify the residue pairs showing also a high allosteric response ratio. We then perform an extensive search by generating all possible pathways through connecting these identified residue pairs and weight each pathway with the product of overlap coefficients. As an example, the predicted allosteric residue containing 314 in PSD-95 has the highest overlap coefficient with residue 327 with a value of 0.99. Then residue 327 has also very high overlap coefficient (with a value of 0.98) with residue 338. We then construct a pathway Ile314→Ile327→Ile338 which gives a total weight of 0.99x0.98 = 0.97. After exhaustive construction of all possible pathways we select the pathway with maximum total weight.
|
10.1371/journal.pntd.0001505 | Association of Mast Cell-Derived VEGF and Proteases in Dengue Shock Syndrome | Recent in-vitro studies have suggested that mast cells are involved in Dengue virus infection. To clarify the role of mast cells in the development of clinical Dengue fever, we compared the plasma levels of several mast cell-derived mediators (vascular endothelial cell growth factor [VEGF], soluble VEGF receptors [sVEGFRs], tryptase, and chymase) and -related cytokines (IL-4, -9, and -17) between patients with differing severity of Dengue fever and healthy controls.
The study was performed at Children's Hospital No. 2, Ho Chi Minh City, and Vinh Long Province Hospital, Vietnam from 2002 to 2005. Study patients included 103 with Dengue fever (DF), Dengue hemorrhagic fever (DHF), and Dengue shock syndrome (DSS), as diagnosed by the World Health Organization criteria. There were 189 healthy subjects, and 19 febrile illness patients of the same Kinh ethnicity. The levels of mast cell-derived mediators and -related cytokines in plasma were measured by ELISA. VEGF and sVEGFR-1 levels were significantly increased in DHF and DSS compared with those of DF and controls, whereas sVEGFR-2 levels were significantly decreased in DHF and DSS. Significant increases in tryptase and chymase levels, which were accompanied by high IL-9 and -17 concentrations, were detected in DHF and DSS patients. By day 4 of admission, VEGF, sVEGFRs, and proteases levels had returned to similar levels as DF and controls. In-vitro VEGF production by mast cells was examined in KU812 and HMC-1 cells, and was found to be highest when the cells were inoculated with Dengue virus and human Dengue virus-immune serum in the presence of IL-9.
As mast cells are an important source of VEGF, tryptase, and chymase, our findings suggest that mast cell activation and mast cell-derived mediators participate in the development of DHF. The two proteases, particularly chymase, might serve as good predictive markers of Dengue disease severity.
| To clarify the involvement of mast cells in the development of severe Dengue diseases, plasma levels of mast cell-derived mediators, namely vascular endothelial cell growth factor (VEGF), tryptase, and chymase, were estimated in Dengue patients and control subjects in Vietnam. The levels of the mediators were significantly increased in Dengue hemorrhagic fever (DHF) and Dengue shock syndrome (DSS) patients compared with those of Dengue fever (DF) and control (febrile illness and healthy subjects) patients, and the soluble form of VEGF receptors (sVEGFR)-1 and -2 levels were significantly changed in the patients with severe disease. After 2–4 days of admission, the mediator levels had returned to similar levels as those of DF and control subjects. Furthermore, the levels of the Th17 cell-derived mast-cell activators IL-9 and -17 were increased in DHF and DSS. In-vitro production of VEGF in human mast cells was significantly enhanced in the presence of IL-9 when these cells were inoculated with Dengue virus in the presence of human Dengue virus-immune serum. As mast cells are an important source of VEGF, and tryptase and chymase are considered to be specific markers for mast cell activation, mast cells and mast cell-derived mediators might participate in the development of DHF/DSS.
| Dengue virus infection is associated with disease, ranging from Dengue fever (DF) to Dengue hemorrhagic fever (DHF) and/or Dengue shock syndrome (DSS). As severe diseases typically develop in individuals suffering secondary Dengue virus infection, host immunological factors appear to play a role in DHF and DSS [1]. DHF and DSS are characterized by increased vascular permeability and hemorrhagic manifestations [2], with the former phenotype recognized as the hallmark of these severe forms of Dengue. However, the cellular factors and immune molecules underling the development of DHF and DSS are not well understood.
Recent studies on Dengue virus infection have demonstrated that the serum levels of vascular endothelial cell growth factor (VEGF)-A (formerly VEGF) are elevated in DHF patients [3]. VEGF/vascular permeability factor (VPF) was first identified and characterized as a potent stimulator of endothelial permeability [4], and was shown to increase vascular permeability 50,000 fold more efficiently than histamine [5]. VEGF was subsequently reported to promote the proliferation, migration, and survival of endothelial cells [6]. VEGF is a member of a growing family of related proteins that includes VEGF-B, -C, -D, and placental growth factor [7]. A potential candidate for the VEGF-binding molecule is the soluble form of its receptor. At least two types of VEGF receptors are expressed on endothelial cells; both are transmembrane receptor tyrosine kinases, namely, VEGFR-1 or Fms-like tyrosine kinase 1 (Flt-1), and VEGFR-2 or kinase insert domain receptor (KDR) [8]. VEGFR-1 is expressed on monocyte-macrophage lineages other than endothelial cells, whereas VEGFR-2 is expressed primarily on endothelial cells and their progenitors [9], [10]. In addition to its role in promoting endothelial permeability and proliferation, VEGF may contribute to inflammation and coagulation. For example, under in-vitro conditions, VEGF induces the expression of several types of cell adhesion molecules, including E-selectin, intercellular adhesion molecule 1 (ICAM-1), and vascular cell adhesion molecule 1 (VCAM-1), in endothelial cells and promotes the adhesion of leukocytes [11], [12]. Moreover, VEGF signaling up-regulates tissue factor mRNA expression, and protein and procoagulant activities [13]. The proinflammatory/procoagulant effects of VEGF are mediated, at least in part, by the activation of the transcription factors NF-κB, Egr-1, and NFAT. VEGF has been implicated as a pathophysiological mediator in several human disease states, including rheumatoid arthritis, cancer, and inflammatory bowel disease [14]–[16].
Dengue patients typically exhibit increased levels of urinary histamine, which is a major granule product of mast cells and whose levels correlate with disease severity [17]. A large autopsy study of 100 DHF cases from Thailand found that mast cells in connective tissue around the thymus exhibited swelling, cytoplasmic vacuolation, and loss of granule integrity, which are suggestive of mast cell activation [18]. Although recent in-vitro studies have also reported the involvement of mast cells in Dengue virus infection [19], [20], the potential role of mast cells in severe Dengue disease has not yet been explored.
The activation of mast cells, which reside mainly in tissues and are associated closely with blood vessels and nerves [21], [22], is tightly linked with local increases in vascular permeability in allergic disease. Mast cells are key effector cells in IgE-dependent immune responses, such as those involved in the pathogenesis of allergic disorders or in certain instances of immunity to parasites [23]. Recent works have revealed another aspect of mast cell effector function, and mast cells play important roles in inflammation and host defenses against foreign pathogens [24], [25].
Mast cells synthesize and release a range of biologically active substances, including proteases, biogenic amines, cytokines, chemokines, and lipid mediators [26]. Mast cell proteases are key protein components of secretory mast cell granules and are essential for innate antimicrobial inflammatory responses [27]–[29]. It is estimated that mast cell proteases account for >25% of total mast cell protein [30] and that human skin mast cells contain a total of ∼16 µg tryptase and chymase per 106 cells [31]. Mast cell proteases, tryptase, and chymase are serine proteases with trypsin- or chymotrypsin-like substrate specificities, and are the major proteins stored and secreted by mast cells. Measurement of the serum (or plasma) levels of these proteases are recommended in the diagnostic evaluation of systemic anaphylaxis and mastocytosis, with total tryptase levels generally reflecting either the increased burden of mast cells in patients with all forms of systemic mastocytosis, or the decreased burden of mast cells associated with cytoreductive therapies in these disorders. Tryptase and chymase levels generally reflect the magnitude of mast cell activation and are typically elevated during systemic anaphylaxis. Secreted tryptase and chymase promote inflammation, matrix destruction, and tissue remodeling by several mechanisms, including the destruction of procoagulant, matrix, growth, and differentiation factors, and the activation of proteinase-activated receptors, urokinase, metalloproteinases, and angiotensin. In addition, these two serine proteases also modulate immune responses by hydrolyzing chemokines and cytokines, and can also suppress inflammation by inactivating allergens and neuropeptides responsible for inflammation and bronchoconstriction. Thus, similar to mast cells themselves, mast cell serine proteases play multiple roles in host defenses, which may be either beneficial or harmful depending on the specific conditions. As substantial levels of tryptase and chymase are only found in mast cells, these proteases are considered to be selective markers of mast cell activation [26]. The importance of cytokines and chemokines together with mast cells in the pathogenesis of Dengue virus infection has been demonstrated [19], [20], however, the roles of the mast cell-specific proteases, tryptase and chymase, remain unclear.
Here, to determine the roles of mast cells and mast cell-derived mediators in DHF and DSS, we first measured the levels of VEGF, soluble forms of VEGFR-1 and -2, tryptase, and chymase in the plasma of Dengue patients and healthy control subjects. Moreover, because IL-9 has been reported as a T cell-derived growth factor of mast cells [32]–[34] and more recently has been implicated as a Th17-derived cytokine that contributes to inflammatory diseases, the involvement of IL-9 and IL-17 in Dengue infection was also investigated.
The study was performed at two hospitals, Children's Hospital No. 2 in Ho Chi Minh City (HCMC) and the Center for Preventive Medicine in Vinh Long Province (VL), Vietnam. The enrolment was a consecutive sequence of hospitalized children at each hospital. The inclusion criteria on admission to the hospital were age (6 months to 15 years old) and ethnicity (Kinh race). A total of 103 subjects from HCMC and VL were enrolled in this study during 2002–2005 (Table 1). The patients were suspected to have Dengue virus infection based on clinical symptoms at admission. After hospitalization, the patients were diagnosed using standardized serology techniques, as described below, and the WHO (1997) classification criteria for Dengue virus infection [35]. It was reported that the sensitivity of WHO criteria for DSS in Vietnam was only 82%, mainly due to the lack of evidence for thrombocytopenia [36]. Therefore, we basically followed the WHO criteria, but included patients lacking a significant reduction of platelet count, which accounted for no more than 11% of all DHF/DSS cases. Our classification scheme met the requirements of the simplified Integrated Management of Childhood Illness (IMCI) classification system, which is based on plasma leakage as a hallmark of severe dengue disease (DHF/DSS) [37].
Plasma samples were obtained from the 19 DF, 43 DHF, and 41 DSS patients on the day of admission, and an additional 189 plasma samples from healthy, unrelated school children living in HCMC and VL who had no symptoms of Dengue virus infection were collected as control samples. Eighteen (male: 12, female: 6) plasma samples were also collected from school children with a febrile illness (38.9±0.9°C: mean±SD) without an obvious source of infection, including Dengue virus.
This study was approved by the institutional ethical review committees of the Institute of Tropical Medicine, Nagasaki University, Jikei University School of Medicine in Tokyo, and the Pasteur Institute in Ho Chi Minh City. Written informed consent was obtained from the parents or legal guardians of the subjects upon enrollment.
The sample collection and serological diagnosis performed in this cohort study were identical to those reported in our previous study [38]. Blood samples were collected from patients with suspected Dengue infection at the time of admission (day 0) and twice during the following four days (days 2 and 4). Plasma samples were used for the titration of anti-Dengue virus IgM and IgG antibodies, virus isolation, and RT-PCR for the determination of viral serotype. Dengue virus infection was determined by previously established serologic criteria for IgM/IgG ELISAs to Dengue virus (DEN 1–4) and Japanese encephalitis virus in paired plasma, collected with at least three-day intervals [39]. IgM and IgG ELISAs were performed using kits obtained from the Pasteur Institute, HCMC and were considered positive if the ratio of optical density (OD) of test sera to the OD of negative control plasma was ≥2.3 [35]. The cases were diagnosed as secondary infection when the DV IgM-to-IgG ratio was <1.8 [40].
Dengue virus serotyping was performed as previously reported [38]. Briefly, acute plasma samples were used to inoculate C6/36 (Aedes albopictus) cells, virus was obtained, and the Dengue virus serotype was then identified using either a direct or indirect fluorescent antibody technique with monoclonal antibodies supplied by the Centers for Disease Control and Prevention (Fort Collins, CO, USA) [41]. Viral RNA was also extracted from the acute plasma samples with the QIAamp Viral RNA Mini Kit (Qiagen, Hilden, Germany) for the molecular detection of Dengue virus and confirmation of its serotype, as previously described [41]. Briefly, cDNA from the Dengue virus genome RNA was amplified with the Ready-to-go RT-PCR test kit (Amersham, MA, USA) using a consensus primer set (D1 and D2) [31]. The serotype was then determined by semi-nested PCR using specific primer sets (TS1, TS2, TS3, and TS4) to amplify serotype-specific fragments from the regions encoding the capsid and membrane proteins of Dengue virus [39].
The plasma levels of VEGF (VEGF-A), sVEGFR-1, sVEGFR-2, IL-9, and IL-17 in samples from Dengue patient (DF, DHF, and DSS) and control groups (febrile illness and healthy subjects) were measured by ELISA kits (R&D Systems, Minneapolis, MN, or Peprotech Inc., Rocky Hill, NJ). The levels of tryptase or chymase in plasma from the Dengue patients and control groups, and the culture supernatants of mast cells were examined by ELISA kits (CSB, Newark, ED or Otsuka Pharmaceutical Co., Tokushima, Japan).
The human mast cell/basophil line KU812 [42] and human mast cell line HMC-1 [43] were maintained in RPMI 1640 medium or IMDM (Invitrogen, Grand Island, NY). In the infection experiments, Dengue virus 2 (DV16681 strain) was propagated in the C6/36 cell line, and virus titers were then determined by plaque assay using BHK-21 cells [44]. In control experiments, the virus was rendered nonreplicative by placing a sample aliquot under a germicidal lamp (125 mJ/10 min, UV irradiation at 254 nm) at a distance of 5–6 cm, followed by a 30-min incubation on ice [44]. For infection, HMC-1 or KU812 cell pellets were adsorbed at 4°C for 90 min with aliquots of Dengue virus or UV-inactivated virus, Dengue virus in combination with human dengue virus immune serum (1∶1,000 or 1∶10,000 final dilution), UV-inactivated virus in combination with human dengue virus immune serum (1∶1,000 or 1∶10,000 final dilution), or Dengue virus in combination with normal human serum (1∶1,000 final dilution) (premixed at 4°C for 90 min). Dengue virus 2 convalescent-phase sera were used in the antibody-dependent enhancement of Dengue virus infection. Mast cells were infected at a multiplicity of infection (MOI) of 3 plaque forming units (pfu)/cell. Following adsorption, cells were washed and plated in 96-well plates (0.25 mL/well) at 1×106 cells/mL and then incubated at 37°C in 5% CO2 for 24 h. To examine the effects of IL-9 on the production of VEGF from mast cells, mast cells treated with Dengue virus and antibody were incubated with or without recombinant human IL-9 (200 ng/mL, Peprotech, Rocky Hill, NJ). Activation of mast cells with Compound (C) 48/80 (300 µg/ml, Sigma-Aldrich, St. Louis, MO) was used as a positive control, and the culture supernatant of C6/36 cells was used as a negative control. For the measurement of VEGF levels in culture supernatants, culture supernatants were collected from each well after incubation and then stored at −80°C until being subjected to ELISA.
KU812 cells were inoculated with Dengue virus (MOI, 3) or Dengue virus-antiserum combinations. After incubation for 24 h, cells were fixed with 4% paraformaldehyde, washed, and then permeabilized with 0.1% saponin for 1 h at room temperature. Samples were then washed and incubated with mouse anti-Dengue virus monoclonal antibody 1B7 [45] and isotype-matched mouse IgG2a antibody (negative control, R&D Systems) on ice for 1 h, which were employed as primary antibodies. Subsequently, samples were washed and incubated with FITC-labeled anti-mouse IgG antibody (R&D Systems) for 1 h on ice. Cytospins were made for each sample, and positive cells were observed by fluorescence microscopy.
Plasma VEGF, sVEGFRs, IL-9, IL-17, tryptase, and chymase levels were compared between the Dengue (DF, DHF, or DSS) and control groups (febrile illness and healthy subjects) using the unpaired Student's t test. VEGF levels in the in-vitro experiments were also compared between the Dengue virus infection and control (UV-inactivated Dengue virus and Medium alone) samples using the unpaired Student's t test. A value of p<0.05 was considered statistically significant.
As mast cells are an important source of VEGF [46], [47], we first measured VEGF levels in plasma samples from the DF (n = 19), DHF (n = 43), and DSS (n = 41) patient groups, and the control group, which consisted of febrile illness and healthy subjects. On day 0 (admission), the VEGF plasma levels were significantly higher in DHF and DSS than those in DF, and febrile illness and healthy subjects (Fig. 1A). The sVEGFR-1 levels in plasma were higher in DSS than those in DF, DHF, febrile illness and healthy subjects (Fig. 1B). In contrast with sVEGR-1, the levels of sVEGFR-2 were dramatically decreased in DHF and DSS compared with DF or febrile illness and healthy subjects (Fig. 1C).
We next examined the levels of VEGF and sVEGFRs in DHF (n = 21) and DSS (n = 27) patients during the admission period (Fig. 2). The VEGF levels in DHF and DSS, and sVEGFR-1 levels in DSS were significantly higher than those of DF or healthy controls on the day of admission (day 0); however, 2–4 days later (convalescence), their levels had gradually declined to comparable levels with DF, febrile illness, and healthy subjects by the convalescent phase (day 4; VEGF DF: 0.61±+0.24 ng/ml, febrile illness: 0.57±0.11 ng/ml, and healthy subjects: 0.52±0.17 ng/ml; sVEGFR-1 DF: 180.9+55.3 pg/ml, DHF: 223.6±136 pg/ml, febrile illness: 201.3±167.1 pg/ml, and healthy subjects: 195.1±59.1 pg/ml). The plasma levels of sVEGFR-2 in DHF and DSS patients were significantly lower compared to those of DF, febrile illness, and healthy subjects; however, the levels were comparable between these groups by day 4 (Fig. 2). Taken together, these findings suggested the possibility that VEGF and sVEGFRs participated in severe Dengue virus infection.
We also measured the tryptase and chymase levels in plasma collected from the Dengue patients (day 0) and controls by ELISA. Plasma tryptase levels increased significantly in DHF and DSS compared with DF, febrile illness, and healthy subjects (Fig. 3). In contrast, the chymase levels were increased significantly in DSS compared with DF, DHF, febrile illness, and healthy subjects (Fig. 3). We next measured the plasma levels of tryptase and chymase in DHF (n = 21) and DSS (n = 27) patients during the admission period and found that the protease levels had gradually declined by days 2 and 4 to a comparable level with those of DF, febrile illness, and healthy subjects (chymase DF: 4.8±2 ng/ml, DHF: 6.7±2.4 ng/ml, febrile illness: 6.2±2 ng/ml, healthy subjects: 4.5±1.3 ng/ml, tryptase DF: 7.4±4.3 ng/ml, febrile illness: 6.7±2.6 ng/ml, healthy subjects: 7.7±3.4 ng/ml) (Fig. 2). These results suggested that mast cells and mast cell-derived proteases participated in the severe form of Dengue virus infection.
As IL-9 has been reported as a T cell-derived mast cell growth factor [32]–[34] and more recently, is implicated as a Th17-derived cytokine that can contribute to inflammatory diseases, we investigated the involvement of IL-9 and IL-17 in Dengue virus infection. The levels of IL-9 and IL-17 in Dengue patients on day 0, and those in blood samples collected from febrile illness and healthy subjects were measured by ELISA. The analysis showed that IL-9 and IL-17 levels were significantly increased in DHF and DSS compared with those in DF, febrile illness, and healthy subjects (Figs. 4A and B).
Although these results suggested that IL-9 and IL-17 participate in Dengue virus infection, IL-9 may act additively or synergistically with other factors, such as other Th2 cytokines, to induce optimal mast cell responses. To examine the possibility that Th2 cytokines affect mast cell responses in Dengue virus infection, IL-4 levels were also examined in plasma from Dengue patients and control groups (Fig. 4C). We found comparable levels of IL-4 between Dengue patients and control groups, suggesting the involvement of IL-9 and -17 in Dengue virus infection.
To investigate if Dengue virus induces VEGF production from mast cells, the in-vitro production of VEGF in the human mast cell lines KU812 and HMC-1 was examined. KU812 and HMC-1 cells were inoculated with Dengue virus in the presence of either human Dengue virus-immune or normal human serum, and VEGF levels in the culture medium were assessed 24 h after viral inoculation. As the antibody-dependent enhancement of infection in KU812 and HMC-1 cells was observed at 1∶1,000 and 1∶10,000 dilutions of human Dengue virus-immune serum in preliminary experiments (data not shown), a 1∶1,000 dilution was used in the in-vitro experiments in this study.
The production of VEGF was observed in both KU812 and HMC-1 cells after exposure to Dengue virus in the presence human Dengue virus-immune serum, however, VEGF levels were higher in KU812 cells (Table 2). No significant increase of VEGF level was observed when Dengue virus was inoculated with normal human serum (1∶1,000 final dilution) or when UV-inactivated Dengue virus was inoculated with human Dengue virus immune or normal human serum. In addition, no VEGF production by KU812 and HMC-1 cells was observed after mock-infection with human Dengue immune or normal human serum. These results suggested the importance of antibody to Dengue virus for mast cell secretion of VEGF in vitro.
As it is known that KU812 and HMC-1 cells are permissive to Dengue virus infection when the virus is inoculated together with human Dengue immune serum [20], the antibody-dependent infection of KU812 cells with Dengue virus was examined by immunofluorescence analysis in the presence and absence of human Dengue immune serum 24 h after the inoculation. Positive immunofluorescence was only observed in cells infected in the presence of human Dengue virus-immune serum, suggesting the occurrence of permissive infection of Dengue virus (Fig. 5). To determine the role of IL-9 in VEGF production by mast cells, KU-812 and HMC-1 cells were inoculated with Dengue virus and human Dengue virus-immune serum (1∶1,000 final dilution) in the presence and absence of IL-9. Although a low level of VEGF production by KU-812 and HMC-1 cells was observed without IL-9, VEGF levels were significantly increased in the presence of IL-9 (Table 3). The effect of IL-9 on VEGF production by KU812 and HMC-1 cells was not observed in the presence of normal human serum (data not shown). Taken together, these findings suggested the possibility that Dengue virus induces VEGF secretion from human mast cells during infection, and that IL-9 supports the production of VEGF in mast cells.
Recently, Srikiatkhachorn et al. [48] compared the plasma levels of VEGF-A and sVEFGR-1 and -2 between DHF and DF patients, and found a rise of VEGF-A and decline of sVEGFR-2 levels in DHF patients, with the severity of plasma leakage inversely correlating with sVEGFR-2 levels. These findings seemed to be consistent with our present results that VEGF and sVEGFR-2 were significantly increased and reduced, respectively, in DHF and DSS patients. Although the reason why sVEGFR-2 levels are decreased in DHF and DSS patients is not clear, as VEGF binding to VEGFR-2 on endothelial cells results in receptor phosphorylation, changes in endothelial cell morphology and proliferation, and maintenance of physiological condition of blood vessels, decreased sVEGFR-2 levels in severe Dengue patients might represent the dysfunction of homeostasis in vascular endothelial cells and correlate with increased plasma leakage [49]. We additionally observed a significant increase of sVEGFR-1 levels in DSS patients, which suggests that activation of monocytes/macrophages by Dengue virus leads to increased expression of soluble and surface VEGFR-1 on cells during severe Dengue infection, as was previously reported [49].
Regarding the relationship between VEGF level and severity of Dengue virus infection, Tseng et al. [3] observed the elevation of circulating VEGF levels in adult DHF patients during the early phases of Dengue infection, as compared to DF patients and healthy controls. In a study of a pediatric population with DHF, Srikiatkhachorn et al. [6] also detected a rise in circulating VEGF in the early febrile and defervescent stages of Dengue infection, but not during the later convalescent stage. However, subsequent studies reported contradictory findings, as increased circulating VEGF concentrations were not observed during the early febrile and toxic stages in DHF, but lower VEGF concentrations were detected in patients with more severe Dengue infection [50]–[52]. Several underlying reasons may explain these differences, such as poor study design, small sample size, and the lack of a standardized collection methodology and storage of blood samples used for the measurement of VEGF. In addition, VEGF is also expressed at low levels in a wide variety of normal adult human and animal tissues, and at higher levels in a few selected sites, namely, podocytes of the renal glomerulus, cardiac myocytes, prostatic epithelium and semen, and certain epithelial cells of the adrenal cortex and lung [53]. Dovrak et al. [54] reported that VEGF is substantially overexpressed at both the mRNA and protein levels in a high percentage of malignant animal and human tumors, as well as in many transformed cell lines. Thus, studies of VEGF production by mast cells during Dengue virus infection are complicated by these alternate sources of VEGF in human and animals, and may affect circulating VEGF levels.
Incubation of KU812 and HMC-1 cells with Dengue virus in the presence of human Dengue virus-immune serum resulted in enhanced VEGF production, which was not observed when UV-inactivated Dengue virus was incubated with human Dengue virus-immune serum or when Dengue virus was used alone to infect KU812 cells (Table 2). As the permissive infection of Dengue virus was observed in KU812 cells (Fig. 5), these findings suggest the critical importance of antibodies to Dengue virus for VEGF production by highly infected mast cells and indicate that infected mast cells can secrete VEGF without stimulation through FcεRI. Our results appear consistent with the findings that Dengue virus infection induces the production of chemokines by human mast cells without stimulation of FcεRI in the presence of human Dengue immune serum [22]. Brown et al. [55] reported that FcγRII plays a dominant role in antibody-enhanced Dengue virus infection of the human mast cell lines HMC-1 and KU812, and in the associated CCL5 release. In studies of DHF epidemics, Halstead et al. [56] and Guzman et al. [57] demonstrated that secondary infection is the most important host risk factor for DHF.
Boesiger et al. and Grützkau et al. [46], [47] reported that mouse and human mast cells produce and secrete VPF/VEGF, and release VEGF upon stimulation through FcεRI or after challenge with chemical mast cell activators. Notably, the FcεRI-dependent secretion of VEGF by either mouse or human mast cells is significantly increased in cells that have undergone upregulation of FcεRI surface expression by preincubation with IgE. As Koraka [58] reported that Dengue virus-specific IgE levels were significantly higher in DHF and DSS patients compared to those in DF and non-Dengue patients, FcεRI may be important for mast cell activation via IgE antibody in Dengue virus infection. However, we did not determine whether the patient sera collected in the present contained IgE antibody against Dengue virus. To clarify the importance of FcεRI for VEGF production by mast cells in Dengue virus infection, further studies are needed.
High levels of VEGF in culture supernatants were detected when KU812 and HMC-1 cells were cultured in the presence of IL-9 (Table 3). As IL-9 enhances the survival of mast cells and induces their production of proinflammatory cytokines, including Th1 and Th2 cytokines [59], it is possible that IL-9 primes HMC-1 and KU812 cells in vitro to respond to Dengue virus infection by promoting VEGF production. To evaluate the contribution of IL-9 and IL-17 to Dengue virus infection, we measured the plasma levels of these two cytokines in Dengue patients and found that both IL-9 and IL-17 were significantly increased in DHF and DSS compared with DF, febrile illness and healthy subjects. These findings suggest that Th9 and Th17 cells contribute to the inflammatory response to severe Dengue virus infection. It is possible that IL-9 may act additively or synergistically with other factors, such as additional Th2 cytokines, to induce the mast cell response observed in this study. However, as the level of IL-4 was not increased in the plasma of Dengue patients, our findings suggest the independent involvement of IL-9 secreted by Th2 cells in Dengue virus infection. Recently, IL-9-producing cells have been described as a new subset of the T helper cell population separate from Th2 that produces IL-9 in large quantities and contributes uniquely to immune responses [60], [61]. This cell population has been named ‘Th9’, and IL-9 secreted by T cells, particularly Th9 cells, may regulate chronic allergic inflammation [62]. Moreover, IL-9 has been recently proposed to function as a Th17-derived cytokine that contributes to inflammatory diseases [36].
Tryptase and chymase levels were significantly increased in DHF and DSS, and DSS, respectively, on admission compared with DF, febrile illness, and healthy subjects (Fig. 3). However, 2–4 days after admission, the levels of these proteases had returned to similar levels with the other patient groups (Fig. 2). These findings support the concept that mast cells and mast cell degranulation play important roles in the pathogenesis of DHF/DSS and might be suitable targets for new therapies and prevention of Dengue infection. However, it is presently unclear whether Dengue virus infection in mast cells directly induces chymase and tryptase production and secretion. Recently, Kitamura-Inenaga et al. [63] reported that encephalomyocarditis virus infection results in mast cell chymase and tryptase production in vivo, and additionally, viral infections have been shown to cause the accumulation of mast cells in the nasal mucosa during the first days of a symptomatic, naturally acquired respiratory infection [64]. However, the relevance and underlying mechanisms of mast cell infection and activation in the setting of viral infections remain to be characterized in detail.
Immunocytohistochemical studies in human tissues have identified two mast cell phenotypes distinguishable by their neutral protease content, namely the ‘mast cell-tryptase’ (MCT) phenotype and the ‘mast cell-tryptase-chymase’ (MCTC) phenotype [65]. MCT appears to be associated with immune system-related mast cells that play a primary role in host defenses and are preferentially located at mucosal surfaces. MCT mast cells are increased in number in areas of T lymphocyte infiltration and in allergic disease, and are reduced in number in acquired and chronic immunodeficiency syndromes [65]. In contrast, the MCTC phenotype appears to be associated with non-immune system-related mast cells that primarily function in angiogenesis and tissue remodeling, rather than immunologic protection, and are found predominantly in submucosal and connective tissues. In addition, MCTC mast cells are not increased in numbers in areas of heavy lymphocytic infiltration and are not decreased in number in immunodeficiency syndromes [65]. In the present study, a significant increase of chymase was observed in the plasma of DSS patients as compared with those of DF, DHF, and the control group, suggesting the possibility that MCTC mast cells contribute to the pathogenesis of severe forms of Dengue virus infection. However, further study is needed to clarify the roles of tryptase and chymase in severe Dengue virus infection.
Concerning the ability of mediators produced by mast cells other than VEGF to activate endothelial cells, King et al. [22] reported that Dengue virus plus Dengue virus-specific antibody treatment results in selective production of the T-cell chemoattractants RANTES, MIP-1α, and MIP-1β by KU812 and HMC-1 human mast cell-basophil lines. In addition, Brown et al. [66] demonstrated that antibody-enhanced Dengue virus infection of primary human cord blood-derived mast cells (CBMCs) and HMC-1 cells results in the release of ICAM-1 and VCAM-1, which subsequently activate human endothelial cells. St. John et al. [67] reported that the response to mast cell activation involves the de novo transcription of cytokines, including TNF-α and IFN-α, and chemokines, such as CCL5, CXCL12, and CX3CL1, which are well characterized to recruit immune effector cells, including cytotoxic lymphocytes, to sites of peripheral inflammation.
In conclusion, we found that mast cells and mast cell-derived mediators, namely VEGF, and the mast cell-specific proteases tryptase and chymase participate in the development of severe forms of Dengue virus infection, which is accompanied by elevated circulating levels of IL-9 and -17. As tryptase and chymase are known as selective markers of non-immune system-related activation of mast cells in submucosal and connective tissues, these two proteases, particularly chymase, might serve as good predictive markers of Dengue disease severity.
|
10.1371/journal.pbio.0050332 | Indirect Effects of Ploidy Suggest X Chromosome Dose, Not the X:A Ratio, Signals Sex in Drosophila | In the textbook view, the ratio of X chromosomes to autosome sets, X:A, is the primary signal specifying sexual fate in Drosophila. An alternative idea is that X chromosome number signals sex through the direct actions of several X-encoded signal element (XSE) proteins. In this alternative, the influence of autosome dose on X chromosome counting is largely indirect. Haploids (1X;1A), which possess the male number of X chromosomes but the female X:A of 1.0, and triploid intersexes (XX;AAA), which possess a female dose of two X chromosomes and the ambiguous X:A ratio of 0.67, represent critical tests of these hypotheses. To directly address the effects of ploidy in primary sex determination, we compared the responses of the signal target, the female-specific SxlPe promoter of the switch gene Sex-lethal, in haploid, diploid, and triploid embryos. We found that haploids activate SxlPe because an extra precellular nuclear division elevates total X chromosome numbers and XSE levels beyond those in diploid males. Conversely, triploid embryos cellularize one cycle earlier than diploids, causing premature cessation of SxlPe expression. This prevents XX;AAA embryos from fully engaging the autoregulatory mechanism that maintains subsequent Sxl expression, causing them to develop as sexual mosaics. We conclude that the X:A ratio predicts sexual fate, but does not actively specify it. Instead, the instructive X chromosome signal is more appropriately seen as collective XSE dose in the early embryo. Our findings reiterate that correlations between X:A ratios and cell fates in other organisms need not implicate the value of the ratio as an active signal.
| In the fruit fly, Drosophila, chromosomal signals determine sex. Diploid flies with two X chromosomes are female, whereas those with one X are male. Conventionally, it is thought that the ratio of the number of X chromosomes to autosomes (X:A) constitutes the signal, because triploid flies bearing two X chromosomes and three sets of autosomes (XX;AAA) are intersexual. Under this model, the X:A signal is defined as the balance between a set of X-linked “numerator” proteins that promote female development and autosomally encoded “denominator” proteins that counteract the numerator elements. Although the X:A signal is a textbook standard, only one strong denominator element exists, and it cannot account for the effects of altered chromosome number (ploidy) on sex. To understand how X and autosome doses influence sex, we examined haploids (1X;1A) and triploids during the brief embryonic period when sex is determined. We found that ploidy affects sex indirectly by increasing in haploids, or decreasing in triploids, the number of embryonic cell cycles in which chromosomal sex is assessed. Our findings indicate that the fly sex-determination signal is more accurately viewed as a function of the number of X chromosomes rather than as a value of the X:A ratio.
| Animals distinguish between numbers or kinds of sex chromosomes both to determine sex and to compensate for unequal gene expression between heterogametic (XY and ZW) and homogametic (XX and ZZ) sexes. In Drosophila and Caenorhabditis elegans, sex and dosage compensation are linked through genetic pathways that exploit transient differences in the expression of several dose-dependent X-linked genes to lock in developmentally stable regulatory states (reviewed in [1]). In mammals, sex is determined by the presence or absence of the Y chromosome, but X chromosome dosage compensation is initiated after a quantitative assessment of X chromosome dose (see [2]). It is thought, in all these cases, that X number is assessed in conjunction with overall ploidy, because changes in the number of autosomal sets relative to X chromosomes affects sexual development or dosage compensation.
The link between autosome dose and sex determination in Drosophila was established in the 1920s when Calvin Bridges showed that triploid flies bearing two X chromosomes and three sets of autosomes (XX;AAA) develop as sexual mosaics [3,4]. This led to the concept that the somatic sex-determination signal is not simply X dose, but rather, the ratio between the number of X chromosomes and the sets of autosomes in the zygote, the X:A ratio. Accordingly, in flies, an X:A of 0.5 (XY;AA) is said to signal male, and an X:A of 1.0 (XX;AA) to signal female, whereas the intermediate X:A of 0.67 (XX;AAA) is an ambiguous signal that some cells interpret as male and others as female. The conventional view for the fly is that each cell in the embryo reads the value of the X:A ratio by measuring the dose of X-linked “numerator” gene products with reference to autosomally encoded “denominator” proteins to set the appropriate on or off activity state of the master sex-determination gene Sex-lethal (Sxl) (see [5–7]). When the X:A equals 1.0, the numerator proteins activate the transiently acting establishment promoter, SxlPe, creating a pulse of SXL, an RNA binding protein [8]. In contrast, when the X:A is 0.5, the inhibitory effect of the denominator proteins predominates so that SxlPe is left inactive and no early SXL is made. Once the X:A ratio has been assessed, SxlPe is permanently inactivated, and the maintenance promoter, SxlPm, is turned on in both sexes; however, only in females is SXL present to bind the SxlPm-derived transcripts and direct them to be spliced into functional Sxl mRNA. Thereafter, Sxl is maintained in the on state by autoregulatory RNA splicing [9,10]. In males, where no early SXL is present, transcripts from SxlPm are spliced by default to a nonfunctional form and SXL is never produced. Once set in the stable, autoregulated, on (female) or off (male) state, Sxl controls all subsequent events in somatic sexual development through control of downstream effectors of sex determination and dosage compensation (reviewed in [1,5,11–13]).
Although four X-linked genes that fulfill all the requirements of X:A numerator elements and one autosomal gene that meets the definition of a denominator element have been identified (see [1,14]), the notion that the X:A ratio is the instructive sex-determining signal relies primarily on correlations between sexual phenotypes and X:A ratios in flies with abnormal ploidy (Table 1). Given modern understanding of the molecules involved and the fact that the system evolved to determine sex in diploid animals, where autosome dose never varies, some have argued that it makes more sense to consider primary sex determination as an X chromosome–counting process, rather than as an X:A sensing one [15–18]. In this alternate view, the male or female dose of X chromosomes is defined by the collective concentrations of four X-linked signal element (XSE) proteins: SisA, Scute, Unpaired, and Runt, that function to activate SxlPe. Proper assessment of XSE concentration by SxlPe depends on numerous protein cofactors present in equal amounts in XY and XX embryos. These cofactors include an autosomal gene product, Deadpan (Dpn) [17–19], but numerous maternally supplied proteins are thought to play the predominant quantitative roles in defining the effective XSE dose. The classic finding, that XX;AAA flies are intersexual, is explained by the XSE-sensing model as the consequence of triploidy affecting proper assessment of X dose and not as implying active participation of a set of autosomal factors analogous to the XSEs.
Haploids, 1X:1A, represent the most stringent test of the X:A model because they possess the male X number, but the female X:A ratio, and develop as females [20–24]. If the XSE-sensing alternative is indeed a more accurate representation of mechanism than is the X:A model, it must explain why the X:A ratio appears to be a better predictor of ultimate sexual phenotype than is X chromosome number.
To answer the question of whether the fly determines its sex by counting X chromosomes or by reading the X:A ratio, we reexamined sex determination in haploids and triploids. Because adult sexual phenotypes need not reflect the fidelity of sex-signal assessment [1,15], we monitored the transcriptional response of the direct sex-signal target, SxlPe, during the early embryonic period when chromosomal sex is assessed. Our results suggest that haploids become female, not because their X:A equals one, but rather because they undergo an extra nuclear division cycle that prolongs the period in which XSE genes are expressed. Remarkably, increased ploidy affects the sexual fate of triploids in a reciprocal manner. We found that triploid embryos cellularize one cell cycle earlier than diploids. The intersexual phenotype of XX;AAA flies thus appears to be, in part, a consequence of there being too little time to accumulate a sufficient concentration of XSE proteins to strongly activate SxlPe in all nuclei. Our findings provide direct experimental support for the notion that XSE gene dose, and not the value of the X:A ratio, is the molecular signal that determines sex in Drosophila.
In diploid flies, chromosomal sex is determined during the rapid syncytial nuclear divisions that precede formation of the cellular blastoderm [8,25,26]. The establishment phase of sex determination begins about 65–75 min after fertilization, during nuclear cycles 8 and 9, with somatic transcription of the XSE genes sisA and scute [16,27,28]. This first phase continues with female-specific activation of SxlPe at about 105 min, during cycle 12, and ends approximately 40 min later when SxlPe is shut off in the first minutes of cycle 14 [8,16,29]. The maintenance phase begins immediately thereafter with activation of SxlPm and the transition to the stable autoregulatory mRNA splicing mode of Sxl expression [8,17]. Thus, sexual fate has been determined well before the completion of somatic cellularization and the onset of gastrulation.
Early development of haploids mirrors that of diploids with an important exception. Haploid embryos undergo an extra syncytial division after cycle 13 and cellularize during nuclear cycle 15 [30–32]. We wondered whether this extra division cycle might provide an explanation for the female character of haploids. In essence, we asked whether haploids become female because their X:A ratio equals one, or because the extra cycle allows more time for XSE protein accumulation and SxlPe activation. The two hypotheses make different predictions. If the value of the X:A ratio is determining, SxlPe should be expressed at the same times in X;A haploid and XX;AA diploid embryos because the X:A ratio is the same in both cases. In contrast, if the extra haploid cycle is responsible for female development, SxlPe activation should be delayed in haploid embryos because they have fewer X chromosomes, and thus, lower amounts of XSE products than equivalently staged diploid females.
To generate haploid embryos, we used two different X-linked recessive maternal-effect mutations: maternal haploid (mh) and sesame (ssm) [30,33]. Homozygous mh or ssm females produced eggs in which the paternal genetic contribution is lost in the earliest divisions, resulting in the development of haploid embryos [34,35] (see Materials and Methods). Sibling females heterozygous for mh or ssm produced normal embryos that served as diploid controls. We used in situ hybridization to monitor SxlPe activity. Key to our analysis was the ability to see focused dots of nuclear staining representing the nascent SxlPe transcripts on the X chromosomes, as well as the accumulated cytoplasmic Sxl mRNA [17,27,29,36].
Haploid embryos exhibited a striking delay in the onset of SxlPe activity as compared to diploids (Figures 1 and 2). In diploid females, SxlPe was first activated during nuclear cycle 12. As diploids progressed through cycle 13, the nuclear dots stained more intensely and cytoplasmic Sxl mRNA was first seen. Strong Sxl expression continued during the first few minutes of cycle 14, with maximum nuclear and cytoplasmic Sxl RNA staining occurring before the formation of the membrane cleavage furrows (Figures 1 and 2). In contrast, in haploid embryos, SxlPe activation was delayed until cycle 14. No Sxl transcripts were seen in haploid cycles 12 or 13, and the pattern of Sxl expression in haploid cycle 14 resembled that seen in diploids during the onset of transcription. In diploid females, activation of SxlPe is a stochastic process occurring independently on each X in each nucleus during cycle 12 [27]. Like diploid cycle 12 females, early haploid cycle 14 embryos were mosaics with respect to the proportion of expressing nuclei, consistent with our observation that SxlPe expression is initiated during haploid cycle 14. As haploid cycle 14 progressed, a greater proportion of nuclei expressed SxlPe; nonetheless, all cycle 14 haploids contained some nuclei with no detectable Sxl expression (Figure 1 and unpublished data).
Whereas activation of SxlPe was delayed in haploids, peak expression and shutoff occurred at comparable phases during the cellularization cycles of haploids (cycle 15) and diploids (cycle 14) (Figures 1 and 2). In both cases, maximum nuclear dot staining intensity and peak accumulation of SxlPe mRNA occurred before the formation of the membrane cleavage furrow and thereafter declined rapidly in a somewhat nonuniform pattern. Based on the similar timing of the cellularization process in haploids and diploids [32], we estimate that SxlPe is expressed maximally during the first 5 to 10 min of the cellularization cycles and that it is shut off in nearly all nuclei approximately 10 min later.
The process of Sxl activation in haploids thus appears to fit the predictions of XSE-sensing models and contradict those of the X:A signal hypothesis. If the X:A ratio were the signal, SxlPe would have been expressed from cycle 12 until early in the cellularization cycles of both X;A and XX;AA embryos. Instead, SxlPe was active in haploids only in cycles 14 and 15. This suggests that it is the extra nuclear cycle that allows haploids to become female, presumably by allowing XSE products to rise above the levels found in diploid male embryos.
To determine whether the XSE genes are transcribed through the extra haploid cycle, and whether haploid XSE mRNA levels eventually exceed those found in diploid males, we analyzed in detail the expression of the key XSE gene, scute. The scute locus, also known as sisterlessB (sisB), encodes a transcriptional activator that dimerizes with maternally supplied daughterless protein to bind to and activate SxlPe [37]. Quantitatively, scute is the most important XSE gene and is needed to activate SxlPe in all regions of the embryo [14,18,38,39].
Consistent with earlier findings [16,28], low-level scute expression could be detected at nuclear cycle 9 in both diploid and haploid embryos, but cytoplasmic scute mRNA was first readily apparent in cycle 11 (unpublished data). In diploids, we could reliably distinguish sex-specific differences in scute mRNA from cycle 12 through the first minutes of cycle 14, with female embryos expressing approximately twice the amount of scute mRNA as equivalently staged males (Figure 3). As cycle 14 progressed beyond the point when SxlPe is active, scute mRNA staining rapidly declined, and it was no longer possible to discriminate between male and female embryos based on scute mRNA levels. In combination with previous reports for scute mRNA [16] and protein [39], our data confirm that scute is expressed in direct proportion to gene dose in the precellular embryo.
In haploid embryos, scute mRNA levels mimicked those seen in diploid males from cycle 12 until the beginning of cycle 14. However, instead of declining immediately thereafter, scute mRNA levels increased throughout haploid cycle 14, reaching a peak in the first minutes of cycle 15. Importantly, at the stage when SxlPe was active, in haploid cycles 14 and 15, the amount of scute mRNA in haploids appeared to surpass the maximum levels observed in diploid males (Figure 3). Thus, the expression pattern of scute fits the predictions of XSE-sensing models: expression begins at the correct stage and scute mRNA levels increase with time and nuclear number. The maximum scute mRNA levels observed in haploids exceed those present in diploid males and closely match the peak levels found in diploid females during early cycle 14 (Figure 3).
We interpret the sex-determining events occurring in haploid and diploid embryos as strongly supporting the hypothesis that X chromosome dose, as defined by threshold XSE protein concentrations, is the signal that directs sexual development. Our interpretation, however, leaves unexplained the phenomenon that started the notion of the X:A ratio as the sex signal: the mosaic intersexual phenotype of XX;AAA flies [3,4]. Simply put, if X number and XSE concentrations are paramount, why are XX;AAA flies intersexes rather than females?
To experimentally address the question of how triploidy impacts the initiation of sex determination, we examined the process of Sxl activation in triploid embryos. To generate the triploid embryos needed, we exploited the gynogenetic-2; gynogenetic-3 double mutant (gyn-2; gyn-3); so named because it can be used to produce diploid offspring with no paternal genetic contribution [40]. Gynogenetic progeny arise because gyn-2; gyn-3 females produce a small fraction of diploid eggs that, when fertilized by nonfunctional sperm from ms(3)K81 mutant males, develop as clones of their mothers. If these rare diploid eggs are, instead, fertilized by normal sperm, they initiate development as XXX;AAA or XXY;AAA triploids [40,41]. Experimentally, this has the advantage of generating triploid embryos without the extensive aneuploidy resulting from crosses with flies carrying compound autosomes (see [42]).
We first examined cellularizing embryos from gyn-2; gyn-3 mothers for nuclear and cell morphology and for the presence of Sxl protein. As expected, most embryos were indistinguishable from normal diploid females and males. They cellularized at cycle 14 nuclear density and either expressed SXL uniformly or not at all [43]. A small proportion of embryos from gyn-2; gyn-3 mothers; however, displayed unusual phenotypes. These rare progeny possessed relatively large nuclei and cellularized at cycle 13 nuclear density (Figure 4). These prematurely cellularizing embryos could be subdivided based on their pattern of Sxl protein staining. Half stained strongly and uniformly for SXL, whereas the other half exhibited weaker, nonuniform SXL staining, suggesting that they represented XXX;AAA and XXY;AAA embryos, respectively (Figure 4). Taken at face value, these data imply that triploids cellularize during nuclear cycle 13. This suggests that XX;AAA triploids may be sexual mosaics, not because of their intermediate X:A ratio, but rather because premature cessation of the X-counting process produces too low levels of XSE products to reliably activate SxlPe during the abbreviated syncytial blastoderm stage. Triploid XXX;AAA embryos would by this logic be female, because the three X chromosomes would supply sufficient XSE proteins to strongly activate SxlPe and the three copies of Sxl would produce enough SXL to reliably engage autoregulation.
To confirm that triploid embryos cellularize at cycle 13 nuclear density and to monitor the effect of premature cellularization on SxlPe, we examined embryos from gyn2; gyn3 mothers using in situ hybridization. Triploid XXX;AAA females were expected to display three nuclear dots indicating Sxl transcription from all three X chromosomes. We observed embryos with three nuclear dots and cycle 12 or 13 nuclear densities, but found none that had three dots and cycle 14 nuclear density. Many of those with three nuclear dots and cycle 13 density had begun to cellularize, confirming that triploid embryos undergo cellularization during nuclear cycle 13 (Figures 5 and 6). Examination embryos with two nuclear dots revealed they were of two kinds: normal diploid females with uniform SxlPe expression in cycle 13 and high levels of cytoplasmic Sxl mRNA in cycle 14, and the presumed XXY;AAA triploids, distinguishable by their weaker, nonuniform SxlPe expression and mRNA staining, and by their undergoing cellularization during cycle 13 (Figures 5 and 6).
Our findings with triploid embryos support the hypothesis that the flies determine their sex by measuring the concentration of XSE proteins in the precellular cycles, rather than by reading the value of the X:A ratio. The intersexuality of XX;AAA flies, traditionally attributed to a decrease in the ratio of female-determining to male-determining proteins, can be more accurately explained as an indirect effect of autosomal ploidy on the timing of embryonic cell cycles. In this view, XX;AAA embryos mimic diploid females until early cycle 13. From that point on, however, premature cessation of the X-counting process leads to less-efficient expression from SxlPe and the failure to reliably engage Sxl autoregulation, creating embryos that are mosaics for Sxl expression. A related phenomenon has been described for mutations affecting JAK/STAT signaling during sex determination. In that case, failure to maintain high-level SxlPe expression during diploid cycle 14 led to reductions in autoregulated Sxl expression, generating sexually mosaic embryos analogous to those described here [29].
Normal Drosophila are males if their cells possess one X chromosome and females if they have two Xs. Demonstration of this fact was central to Calvin Bridges' 1916 proof that genes are located on chromosomes [44]. The contemporary notion that sex is signaled not by X number, but by the value of the X:A ratio, stems from Bridges' work [3,4] showing that possession of two Xs was not sufficient to determine the female fate in triploid flies. Despite the long-standing acceptance of the X:A hypothesis, the evidence that the value of the X:A ratio determines sex is largely correlative and indirect. Fundamentally, the X:A model rests on correlations between adult sexual phenotypes and the value of the ratio in several karyotypes (Table 1). However, the inference that the value of the X:A ratio is instructive, and not merely predictive, as to sex, depends on the assumption that normal adult sexual phenotypes reflect normal operation of earlier regulatory events (see [1,14]). Our demonstration that changes in ploidy alter the temporal and developmental contexts in which sex is assessed shows this underlying assumption to be flawed. Haploidy alters sex determination by increasing the time when the sex signal is assessed; triploidy acts reciprocally, by compressing the time available. Both conditions alter the response of SxlPe to the sex signal in ways that suggest the promoter responds primarily to the concentrations of XSE products present in the embryo rather than to the particular value of the X:A ratio.
This revised view of sex determination is not entirely new. In 1934, Dobzhansky and Schultz [45] offered a cautionary alternative to the X:A hypothesis, warning that the influence of the autosomes on sex may be indirect. Their proposal, that “sex may be determined by the ratio between the number of X chromosomes present in the cell and the size of that cell” differs in specifics from our findings, but the fundamental logic is the same. In 1983, Baker and Belote [46] suggested that maternally contributed products, rather than autosomal factors, represented the key reference to which X dose is measured. More recently, Cline and colleagues [15–18], have pointed out the logical similarity of autosomal elements to maternal elements and highlighted the weak quantitative role of dpn, the sole autosomal element [17,18]. Our findings extend these critical evaluations of the X:A hypothesis, providing a mechanistic explanation for why the ratio may predict sex without specifying it, and offering experimental support for the idea that the primary sex-determination signal is better described as the dose of XSE genes than as the X:A ratio.
It could be argued that the differences between XSE-sensing and X:A-reading models are largely semantic. Both predict sex and both accommodate the same set of XSEs, maternal factors, and autosomal repressor. This, however, is more than an argument about the meaning of words. The XSE-sensing model has the advantages of clarifying terminology, erasing artificial distinctions between maternal and zygotic elements of the sex-signaling system, and providing a more concrete and accurate concept of mechanism. In the conventional X:A paradigm, autosomal “denominator elements” are necessarily core components of the X:A signal, whereas maternal factors are consigned to “X:A signal-transducing” roles [1,14]. This logical formalism creates a situation in which the sole denominator element, the relatively weak dpn, is a more central part of the sex signal than the more numerous and potent maternal signal-transducing factors including, Daughterless, the dimerization partner of the XSE protein Scute, Stat92E, the transcription factor signaled to bind SxlPe by the XSE unpaired, and Groucho, the corepressor needed for Dpn function at SxlPe. XSE-sensing mechanisms avoid such confusion by treating the maternal and autosomal signal element (MSE and ASE) proteins as parts of the cellular and biochemical context in which male and female XSE concentrations are assessed [1,15–18,37]. Reclassification of autosomal factors from X:A denominator elements to “context genes” [1] also serves to highlight the importance of the dynamic temporal and cellular milieu in which X chromosome counting occurs. Nuclear cycles 8–14 represent the transition from maternal to zygotic control of gene expression [47]. The timing of XSE gene and SxlPe activation, suggests that sex determination is directly connected to more-general events occurring at the mid-blastula transition. Factors such as changing chromatin environments and the timed onset of general zygotic gene expression mediated by Bicoid stability factor (BSF) [28,47], or other timing factors, seem likely to influence SxlPe's threshold response. Perhaps equally important, global events associated with the mid-blastula transition may couple the inactivation of XSE transcription and the rapid degradation of XSE and MSE mRNAs to the onset of cellularization [47,48]. If these events are responsible for the timely shut down of SxlPe (see [16,49]), they further highlight the role developmental context plays in preventing XY;AA diploids from activating SxlPe during cycle 14, and in explaining the sexual mosaicism of XXY;AAA animals.
In terms of transcriptional mechanisms, XSE-sensing schemes have the advantage of replacing the incorporeal concept of the value of the X:A ratio with the tangible notion that threshold concentrations of XSE proteins activate SxlPe. Models for how XSE thresholds are set need not invoke the conjectural titrations of XSE proteins by ASEs that seem inevitably to arise from the need to explain how the X:A ratio is read (see [5–7,49]). Instead, one can focus on how dose sensitivity might be explained by the known activators and repressors acting at SxlPe. The Drosophila dorsal–ventral and anterior–posterior patterning systems, in which enhancers integrate positive and negative inputs over narrow concentration ranges, provide precedents for how on or off decisions can be regulated by DNA-binding proteins (see [50–52]).
Although our findings, and those of others, suggest a more realistic approach to mechanism, our data on the correlations between XSE expression and timing of SxlPe activation raise something of a paradox. The modern form of Dobzhansky and Shultz's 1934 argument [45], that the changes in nuclear volume that accompany changes in ploidy might account for the predictive effects of the X:A ratio [15,45], would suggest that, for any given stage, the XSE concentrations in small 1X haploid nuclei should be similar to those found in larger 2X diploid nuclei, and thus, that Sxl expression should occur with similar timing in haploids and diploids. Given the absence of information on XSE protein concentrations, the apparent conflict between our observations and expectations based on nuclear volume is currently unresolvable; however, it cautions that factors in addition to relative XSE gene expression may influence the timing of SxlPe activation. Regardless, either view supports the argument that it is inappropriate to consider the value of the X:A ratio as a simple sex-determining signal [15,17,18,45,46]. Rather, both suggest that the sex-determination signal should be defined in the normal diploid context, in which differential X chromosome dose specifies sex by determining the concentrations of XSE products present in the embryo.
Looking beyond Drosophila, our reinterpretation of the effects of ploidy on primary sex determination has implications for other developmental systems that rely on differential doses of chromosomes to define sexual fates. These include systems thought to read X:A ratios, and one, that as traditionally viewed, cannot.
Haplodiploidy is a widespread means of sex determination in which haploids develop as males and diploids as females. The best understood example of haplodiploidy is complementary sex determination (CSD), known to occur in many bee and wasp species [53–55]. In CSD, females are heterozygous, and males hemizygous, for one sex-determining locus with multiple alleles. Although the CSD mechanism is unrelated to the X-counting process of the fruit fly [56], many haplodiploid species and genera lack CSD. The traditional X:A model of Drosophila is difficult to reconcile with haplodiploidy because the X:A balance is the same regardless of ploidy [57]. Our findings, however, suggest that a Drosophila-like chromosome-counting mechanism could operate in non-CSD haplodiploid species, if haploid and diploid zygotic gene doses were measured in similar cellular contexts. Presciently, Crozier [54,58] proposed that a variation of the Drosophila mechanism based on the chromosomal/cytoplasmic balance could distinguish haploid and diploid embryos. Such a chromosome-measuring system would have a strong maternal component [55,58] that could exhibit strain or species variation consistent with extensive involvement of the maternal genome in many insect sex-determining systems [53].
Mammalian sex depends on the Y chromosome, but a second aspect of sexual dimorphism, X chromosome inactivation, requires that X dose be assessed. It is thought that X counting in mammalian cells depends on the X:A ratio, because the number of active Xs increases with the number of autosome sets (see [2,59]). Recent models of the establishment of X inactivation incorporate the X:A concept; invoking titrations of autosome-encoded factors by X-linked sites [59–61] that bear remarkable similarity to early speculations as to how the fly X:A ratio might be read (see [46]). However, new findings suggesting a role for X chromosome pairing in X counting and choice [62,63], and indications that ploidy has less impact on X inactivation than generally thought [64] are difficult to reconcile with traditional notions of the X:A ratio. In this light, our findings caution that abnormal ploidy may also alter the cellular context in which mammalian X counting occurs. If so, the effects of altered ploidy may suggest only that autosomal products function in the X-counting process and not that they are a central part of a specific X:A signaling mechanism.
Of the well-known experimental systems said to depend on X:A ratios, it may be that only the nematode C. elegans actively consults the X:A balance when measuring X chromosome dose [65]. Why might assessment of the X:A ratio be central to worm sex, but only a minor aspect of the fruit fly mechanism? Perhaps the structures of the regulatory systems dictated their evolution. Superficially, the C. elegans mechanism resembles that of the fruit fly in that at least four XSE gene products regulate the expression state of a single sex-determining switch gene, xol-1 [66,67]. However, in C. elegans, the XSEs antagonize the actions of several discrete ASEs that function to activate xol-1 in males [65]; whereas in Drosophila, the XSEs activate their target, Sxl. For the fly, it is possible to envision how an ancestral X chromosome–counting mechanism, based on XSE dose and maternal factors, could have differentially expressed Sxl, and how the autosomal element, dpn, could later have been added to refine the regulation of Sxl [68]. In contrast, for C. elegans, autosomal elements must have been involved from the beginning, for without ASE-mediated activation of xol-1, the repressive sex-determining functions of the XSEs would have been moot. Whether C. elegans primary sex determination also relies on an extensive maternal contribution remains to be determined.
Haploid embryos were from females homozygous for the recessive X-linked maternal-effect mutations, maternal haploid (mh1) [30,35] or sesame (ssm185b), also known as Hira [33]. Diploid control embryos from sibling females heterozygous for mh (z1 w mh1/FM3 X z1 w mh1/Y) or ssm (w ssm185b/FM7 X w ssm185b/Y) were indistinguishable from embryos from wildtype stocks. Eggs from mh1/mh1 and ssm185b/ssm185b females develop as maternally derived (gynogenetic) haploids because, for mh1, the paternally derived sister chromatids fail to separate during the first embryonic mitosis, leading to their loss during the next three divisions [34], or for ssm185b, because the male pronucleus does not fully decondense, is arrested before the first S-phase, and fails to enter the first mitotic spindle [35]. Haploid embryos from mh1 or ssm185b mothers were indistinguishable with respect to Sxl and XSE gene expression (unpublished data). Our initial analysis of haploids used mh1, but most later experiments exploited ssm185b because we found that a fraction of embryos derived from mh1, but not ssm185b, mothers were partial diploids and that others appeared to have lost the X chromosome in some of their nuclei (unpublished data). Triploid embryos and sibling diploid controls were generated from mothers homozygous for two recessive maternal effect mutations gynogenetic-2 and -3 (gyn-2 and gyn-3) [40]. Most eggs laid by homozygous gyn-2; gyn-3 females are haploid and develop as normal diploid embryos when fertilized; however, gyn-2;gyn-3 mothers produce a small and variable percentage of diploid eggs that develop as XXX;AAA or XXY;AAA triploids depending on whether they are fertilized by an X-bearing or Y-bearing sperm [40,41]. The Drosophila Y chromosome does not influence sex determination. Stock z1 w mh1/FM3 was provided by M. Wolfner (Cornell University), w ssm185b/FM7 was from B. Loppin (Centre de Génétique Moléculaire et Cellulaire), and w1; gyn-2; gyn-3 was from the Bloomington Drosophila Stock Center.
Embryos were collected, and processed for immunocytochemistry according to Patel et al. [66]. Anti-Sxl mouse antibody (gift of T. Cline, University of California, Berkeley) was used at 1:300 dilution. Horseradish peroxidase secondary antibodies (Jackson ImmunoResearch) were used at a dilution of 1:300 and visualized with 3,3′diaminobenzidine. All embryos were stained with DAPI to visualize DNA. In situ hybridization was done using standard procedures as described [16,17,27,29,36,70]. Briefly, digoxygenin-labeled RNA probes complementary to Sxl exon E1 or the scute coding regions [16] were prepared using in vitro transcription of plasmid or PCR-derived templates. Sxl exon E1 probes detect both SxlPe-derived mRNA and Pe-derived nascent transcripts, the latter visible as focused dots of staining within nuclei. Sxl and scute are X-linked so the number of nuclear dots corresponds to the number of X chromosomes. In all cases, we analyzed expression of the endogenous loci. No transgenic promoter fusions were used. Because haploid embryos and their diploid controls derived from different females, embryos were collected, processed, and hybridized in parallel. Triploid embryos and their diploid control siblings were from the same egg collections. For embryo staging, cell cycle number was determined by nuclear density [30,32,71]. Nuclei change in size and appearance as they progress through the precellularization cycles [72], and we exploited this to stage embryos as closely as possible. Timing through the cellularization cycles was estimated by nuclear shape and length, by the distance from the base of the nucleus to the yolk, and by the extent of membrane furrow invagination [32,71]. Detailed comparisons of cell cycles and gene expression in haploid and diploid embryos have been published [30–32]. Time estimates for the cellularization cycle in triploids was by analogy to diploid and haploid embryos. An abundant literature (see [30,32]) has established that the timing of the mid-blastula transition is linked to the ratio of DNA to cytoplasm in the embryos of many species. Cleavage divisions stop and cellularization cycles begin when the nucleocytoplasmic ratio reaches a threshold value, explaining why haploids undergo one more cleavage division and tetraploids one fewer division than diploid embryos [32].
The figures summarize the results of many different experiments with haploid, diploid, and triploid embryos. Only some experiments were quantified by counting the number of embryos at specific stages, but the results were qualitatively assessed as the same for each repetition. The following represent numbers of embryos counted and recorded with respect to the listed conclusions, but many others were observed. Timing of SxlPe activation in haploids: 10 cycle: 11 embryos, 28 cycle: 12 embryos, 32 cycle: 13 embryos, 51 cycle: 14 embryos, and 42 cycle: 15 embryos. Timing of SxlPe activation in diploids has been established [16,17,27,29], but we note that about one fourth (11 of 39) of wild-type cycle 12 embryos exhibited detectable Sxl expression, consistent with the onset occurring in females during cycle 12. The time course of scute expression in Figure 3 was assembled from photographs of every cycle 12, 13, and 14 haploid embryo, every cycle 12 and 13 diploid embryo, and from 13 haploid and 10 diploid embryos in the cellularization cycles in the experiment. The embryos shown were judged as close in stage as possible based on the density, size, and morphology of DAPI-stained nuclei [32,72]. The percentage of triploids among diploid progeny of gyn-2, gyn-3 mothers was variable [40] for unknown reasons. The fraction of pre-germ band–extended triploids with mosaic SXL staining (presumed XXY;AAA) was about 50% in all experiments (21/39 counted). We counted 17 XXX;AAA and 14 presumed XXY;AAA embryos that expressed SxlPe, but observed numerous others.
|
10.1371/journal.pntd.0002413 | A Schistosoma haematobium-Specific Real-Time PCR for Diagnosis of Urogenital Schistosomiasis in Serum Samples of International Travelers and Migrants | Diagnosis of urogenital schistosomiasis by microscopy and serological tests may be elusive in travelers due to low egg load and the absence of seroconversion upon arrival. There is need for a more sensitive diagnostic test. Therefore, we developed a real-time PCR targeting the Schistosoma haematobium-specific Dra1 sequence.
The PCR was evaluated on urine (n = 111), stool (n = 84) and serum samples (n = 135), and one biopsy from travelers and migrants with confirmed or suspected schistosomiasis. PCR revealed a positive result in 7/7 urine samples, 11/11 stool samples and 1/1 biopsy containing S. haematobium eggs as demonstrated by microscopy and in 22/23 serum samples from patients with a parasitological confirmed S. haematobium infection. S. haematobium DNA was additionally detected by PCR in 7 urine, 3 stool and 5 serum samples of patients suspected of having schistosomiasis without egg excretion in urine and feces. None of these suspected patients demonstrated other parasitic infections except one with Blastocystis hominis and Entamoeba cyst in a fecal sample. The PCR was negative in all stool samples containing S. mansoni eggs (n = 21) and in all serum samples of patients with a microscopically confirmed S. mansoni (n = 22), Ascaris lumbricoides (n = 1), Ancylostomidae (n = 1), Strongyloides stercoralis (n = 1) or Trichuris trichuria infection (n = 1). The PCR demonstrated a high specificity, reproducibility and analytical sensitivity (0.5 eggs per gram of feces).
The real-time PCR targeting the Dra1 sequence for S. haematobium-specific detection in urine, feces, and particularly serum, is a promising tool to confirm the diagnosis, also during the acute phase of urogenital schistosomiasis.
| Schistosomiasis is a disease caused by parasitic worms of the genus Schistosoma. About 200 million people are affected worldwide. Also travelers are at risk as even a brief contact with infested freshwater can cause infection. S. mansoni and S. haematobium are the two main species that are identified in travelers and migrants. The eggs of these parasites are respectively excreted in feces and urine, and the diagnosis relies mostly on microscopy. In travelers, infections are easily missed due to low worm load or because egg excretion is not yet started upon arrival. Consequently, there is need for sensitive diagnostic tools that can be used in the early stage of infection. A previously published study reported the ability to detect S. mansoni DNA in serum by real-time PCR. To enable the diagnosis of urogenital schistosomiasis, we developed a PCR to detect S. haematobium DNA in serum. We demonstrated that the latter PCR is more sensitive than microscopy when applied on feces and urine, and, when performed on serum, particularly useful to confirm diagnosis during acute urogenital schistosomiasis. We comment on the plausible origin of parasite DNA in relation to the different life cycle stages present in the blood circulation.
| Urogenital schistosomiasis due to Schistosoma haematobium is a serious underestimated public health problem. It is endemic in 53 countries of the African continent and of the Middle East [1], [2]. Adult worms live in the capillary plexus of the bladder and other parts of the urino-genital system and eggs are excreted in the urine and occasionally found in feces.
Diagnosis of S. haematobium infections is traditionally done by microscopy but is often unreliable due to the circadian and day-to-day variations in egg excretion, and to low parasite load, especially in the traveler. Antibody-based assays are useful to confirm infection, but do not distinguish active infection from past exposure, and false-negative results occur, mainly in S. haematobium infections. Antibody tests are usually negative during acute symptomatic schistosomiasis. Assays that detect circulating antigens seem very promising in the early phase of infection but still lack sensitivity in the diagnosis of light infections [3], [4], [5], [6].
Recently, we developed a genus-specific real-time PCR (further called ‘genusPCR’) that sensitively detect all human infectious Schistosoma species in feces and urine [7]. The genusPCR was not able to detect schistosome DNA in serum although molecular analysis of serum is of interest in acute schistosomiasis before detectable levels of eggs are excreted [8]–[12]. In 2009, Wichmann and colleagues [10] described a real-time PCR, targeting a highly repeated 121-bp sequence of S. mansoni (named Sm1-7) to detect cell-free schistosome DNA in serum. This was proven successful in acute and chronic S. mansoni infection, but not so much in S. haematobium infection. To fill that gap, we developed a real-time PCR specific for the diagnosis of S. haematobium in serum samples. The real-time PCR targets Dra1, a S. haematobium-specific 121-bp repeat sequence originally described by Hamburger et al. and present in hundreds to thousands of copies and representing at least 15% of its genome [13]. We first tested this PCR (further called ‘draPCR’) on urine and feces samples to evaluate its species-specificity and its performance in comparison with microscopy, and then on serum samples to determine its potential as diagnostic tool for acute phase schistosomiasis.
The diagnostic procedures described in this manuscript are part of the standard diagnostic work-up of patients suspected of schistosomiasis. All samples were routine diagnostic samples from patients presenting at the Institute of Tropical Medicine (ITM, Antwerp, Belgium) policlinic and were stored after completion of the routine tests. The ITM has the policy that sample left-overs of patients presenting at the ITM policlinic can be used for research unless the patients explicitly state their objection. The Institutional Review Board of ITM approved the institutional policy of this presumed consent as long as patients' identity is not disclosed to third parties. All data have been analysed anonymously.
PCR analysis was retrospectively performed between January and October 2012 on samples that were stored at <−18°C between 2006 and 2012. In total, 330 clinical samples from 187 patients were analysed of which 110 urine, 84 stool, 126 serum, 9 blood samples, and one biopsy sample. The samples were from 145 travelers and 42 migrants that presented at the outpatient clinic of the ITM, Antwerp, Belgium. The patients travelled to Africa (92.5%), Asia (5.4%), the Middle East (1.6%) and South-America (0.5%). The median time interval between return and sample collection was 82 days (varying from 1 day to 1851 days).
Based on laboratory findings, the stool, urine and/or serum samples were selected from patients with confirmed (n = 47) or suspected (n = 140) intestinal or urogenital schistosomiasis. Confirmed cases were defined as patients with eggs of S. mansoni or S. haematobium as determined by microscopy. Individuals with a confirmed infection were treated with praziquantel after diagnosis. Serum of confirmed cases was collected before or at the time of egg detection, or after treatment. Suspected cases were patients who presented after a stay in an endemic region with clinical symptoms and/or with an increase in eosinophils (>0.45×109/L), a positive serology (IHA titer ≥1/160 or positive ELISA), the presence of Charcot-leyden crystals in feces or hematuria (>7 RBC/µL) or who travelled together with a confirmed case.
As previously described [7], [14], DNA was extracted with the QIAamp DNA stool mini kit (Qiagen Benelux, Venlo, The Netherlands) from 1 gram of feces that was dissolved into 5 mL ASL buffer (Qiagen). An 200 µl urine sediment was processed for DNA extraction with the QIAamp DNA mini kit (Qiagen) after centrifugation of 10 mL of urine and three wash/centrifugation steps. For DNA extraction of serum with the QIAamp DNA MIDI kit (Qiagen) or by phenol/chloroform, 1 to 2 mL of serum was used [7].
Positive control DNA of S. mansoni and S. haematobium were kindly provided by Dr. T. Huyse (ITM/KUL, Belgium). Positive control DNA of S. intercalatum and S. guineensis were kindly provided by Dr. F. Allan from SCAN at the Natural History Museum (London, UK) [15]. DNA was extracted from one adult worm and used in a 1/100 dilution (∼0.1 ng/µL). Positive control DNA of S. mekongi was obtained from a stool sample of a patient seen at ITM, containing 50 eggs per gram (EPG). Positive control DNA of S. japonicum derived from cercariae spotted on FTA filter paper kindly provided by Dr. J.P. Webster (London, UK).
The highly repetitive Dra1 sequence of S. haematobium (Accession number DQ157698.1) was selected as target and primers were identical to those described [13] (Sh-FW 5-gatctcacctatcagacgaaac-3′; Sh-RV 5′-tcacaacgatacgaccaac-3′). An additional fluorescent labeled hydrolysis probe was developed (Sh-probe 5′-tgttggtggaagtgcctgtttcgcaa-3′) for real-time monitoring of the PCR signal and was labeled with a 5′-FAM reporter, an internal ZEN quencher and a IowaBlack Fluorescent Quencher at the 3′-end (IDT, Leuven, Belgium). The amplicon size was 96 base pairs.
The draPCR was performed with a 25 µL reaction mix containing 5 µL DNA, 1× Perfecta qPCR Supermix (Quanta Biosciences), 500 nM of Sh-FW and Sh-RV primer, 250 nM of Sh-probe and 0.1 mg/mL bovine serum albumin. The program consisted of an initial step of 2 min at 95°C followed by 50 cycles of 15 s at 95°C and 60 s at 60°C.
The reaction was run on the SmartCycler II (Cepheid Benelux, Belgium). DNA detection was expressed by Cycle threshold (Ct)-values. In every run, the non-template control was negative (Ct = 0) and the S. haematobium control was positive.
To detect DNA of schistosome species other than S. haematobium, the Sm1-7PCR and genusPCR were performed as described before [7], [10].
The primer and probe design was verified with Integrated DNA Technology (IDT) Oligo Analyzer software (v3.1) (http://eu.idtdna.com/analyzer/Applications/OligoAnalyzer/). Primer and probe specificity was checked in silico by BLAST analysis (http://blast.ncbi.nlm.nih.gov/Blast.cgi) and by 2% agarose gel electrophoresis at 100 V for 35 minutes.
The analytical specificity of the PCR was tested on a panel of clinical control samples containing 23 different intestinal or blood parasites. The panel included stool samples (n = 14) from patients infected with protozoa (Giardia lamblia, Entamoeba dispar, E. histolytica, Blastocystis hominis, Enterocytozoon bieneusi, Encephalitozoon spp), nematodes (Ascaris lumbricoides, Strongyloides stercoralis, Trichostrongylus spp., Trichuris trichiura, or Ancylostomidae), trematodes (Clonorchis spp., Fasciola hepatica) or a cestode (Taenia saginata) and blood samples (n = 9) from patients infected with Plasmodium falciparum, P. vivax, P. ovale, P. malariae, Leishmania donovani, Loa loa, Onchocerca volvulus, Dirofilaria repens or Trypanosoma brucei rhodesiense.
The detection limit was determined on a 10-fold dilution series of a stool sample containing 580 EPG of S. haematobium. It was diluted in a negative stool sample which was dissolved in ASL buffer (Qiagen). DNA was extracted from each dilution and the highest dilution with a positive signal indicated the detection limit.
The variation in Ct-values was determined in a serum sample that was processed 8 times within the same run (repeatability) or 5 reactions that were run at different days (reproducibility). The coefficient of variation (CV, expressed as %) of the Ct-values was calculated.
Microscopy was performed on a single urine and/or fecal sample per patient at the time of presentation and in some cases, on a single follow-up sample one month after treatment. Diagnosis of schistosomiasis was confirmed when S. haematobium or S. mansoni eggs were detected in urine and/or feces.
Microscopic examination of urine samples was performed on the sediment of at least 20 mL end-stream urine and of stool samples following a concentration method on 3 grams of feces that had been homogenized in 42 mL of 10% formaldehyde-saline solution [16]. The infection intensity in stool was expressed by the number of EPG. The limit of detection was 10 EPG.
A combination of an in-house enzyme-linked immunosorbent assay (ELISA) using S. mansoni antigen (mixture of egg and adult worm extract) and an indirect hemagglutination inhibition assay (IHA), using a S. mansoni adult worm extract (ELI.H.A. Schistosoma, EliTech MICROBIO, France) with a cut-off at 1/160, were used to detect anti-schistosome antibodies.
IDT Oligo analysis approved no self- or heterodimerization between the primers and the probe. BLAST analysis with probe and primers indicated 100% query coverage and maximum identity with S. haematobium. No species other than Schistosoma were in silico recognised by the primers and probe. Gel electrophoresis obtained a single band of expected length for the amplicon of S. haematobium and no signal for the non-template control.
To determine the species-specificity, schistosome species of the three complexes were tested with the draPCR, the Sm1-7PCR and the genusPCR. The DNA controls of human species of the S. haematobium complex revealed a strong signal with the draPCR (Ct-values ranging from 15.21 to 16.65) and the cattle species S. bovis revealed a signal of medium intensity (Ct 29.68). DNA controls of other Schistosoma species gave no (S. japonicum, S. mekongi) or a very weak signal (S. mansoni, Ct 41.93) (Table 1). In comparison, the genusPCR easily recognized all species of the three complexes while the Sm1-7PCR detected a strong signal for S. mansoni and S. bovis, a medium to weak signal for the human species of the S. haematobium complex and no signal for species of the S. japonicum complex (Table 1).
Of interest is the difference in Ct-values measured for S. haematobium (Ct 15.21) and S. mansoni (Ct 41.93) by the draPCR. Since the amount of amplicon doubles every PCR cycle (i.e. increase by one log2), the difference of 26 Ct's is equivalent to a more than 67 million times lower sensitivity to detect S. mansoni in comparison to S. haematobium. The same counts for the detection of the S. bovis species in comparison to the other S. haematobium complex species by the draPCR with a difference of 14 Ct's that accounts for a 16,000 times lower sensitivity (Table 1).
No cross-reaction was seen with the draPCR in the 23 control samples with intestinal and blood parasites other than Schistosoma.
The analytical sensitivity demonstrated a detection limit of 0.5 EPG.
Repeatability and reproducibility testing revealed a CV of 1.03% and 1.04% respectively.
A panel of 110 urine samples and one bladder wall biopsy was analysed with the draPCR. A positive PCR signal was obtained in 14 urine samples (Ct-values ranging from 16.77 to 32.40) of which seven were positive for S. haematobium ova by microscopy (Table 2). The other seven urine samples (Ct-values ranging from 30.74 to 46.63) were from patients treated for schistosomiasis two weeks or one month before (n = 2) or from patients without previous treatment but with anti-schistosome antibodies (n = 3) or with S. haematobium eggs in feces (n = 1) or in a urine sample obtained three days earlier (n = 1). The latter urine sample also contained Trichomonas vaginalis. The 96 urine samples that were negative for S. haematobium or any other parasite by microscopy, were also negative by PCR (Table 2).
The biopsy containing S. haematobium eggs was positive by PCR with a Ct-value of 17.08.
The draPCR was evaluated on a panel of stool samples in which eggs of S. haematobium (n = 11), of S. mansoni (n = 21) or no schistosome eggs (n = 52) were microscopically detected. All samples with eggs of S. haematobium were positive (Ct-values ranging from 20.35 to 37.87) and all samples with eggs of S. mansoni were negative (Table 3). In addition, the draPCR revealed a positive signal in three samples without eggs (Ct-values varied from 36.78 to 45.33), two of which were follow-up samples of confirmed patients one month after treatment and one from a clinically suspected patient with eosinophilia.
There is no reference method for schistosome DNA detection in serum. We therefore used the level of evidence of infection (confirmed S. haematobium (n = 12) or S. mansoni infection (n = 20) or suspected cases (n = 64)) as a reference (Table 4). Of all suspected cases, 8/64 were migrants and 56/64 travelers of whom serum was collected within 12 weeks upon return in 39 travelers and after more than 12 weeks (varying from 91 to 336 days) in 17 travelers.
In total, 135 serum and blood samples were analysed from 96 patients of whom 64 patients with a single serum sample and 32 patients with one or more follow-up samples. No discordant results were obtained in different samples from the same patient.
The draPCR was positive in 27 samples from 13 patients of which 11 patients (22 samples) with a microscopic confirmed S. haematobium infection and two patients (5 samples) with a clinical suspicion based on the presence of eosinophilia (0.88 and 2.13 10*9/L) and anti-schistosome antibodies (IHA 1/160 and 1/640) (Table 4). Of all 22 PCR positive samples of individuals with a confirmed S. haematobium infection, all serum samples collected at the same date of the parasitological confirmation (n = 11), were positive. All follow-up serum samples obtained 14 to 96 days after treatment (n = 10 from 8 patients), were also positive by PCR but demonstrated higher Ct-values. In two of the 8 patients, both serological tests remained negative 1 month and 2 months after treatment.
PCR was additionally positive on one serum collected about 5 weeks after exposure (n = 1) while at that moment the urine and feces were microscopically negative and no antibodies could be detected. Detection of ova 42 days later, confirmed the S. haematobium infection.
In one serum sample from a patient for which a single S. haematobium egg was detected in urine 514 days after treatment, no PCR signal was observed but the analysis was performed on an insufficient volume of serum (1 mL instead of 2 mL). All other serum samples with a negative PCR signal were from patients with a confirmed S. mansoni infection (20 patients, 22 samples) or from clinically suspected patients without schistosome eggs in urine and feces (62 patients, 85 samples) (Table 4). Four of these suspected patients had microscopically confirmed infections with Ascaris lumbricoides (n = 1), Ancylostomidae (n = 1), Strongyloides stercoralis (n = 1) or Trichuris trichuria (n = 1) and nine had been treated 6 months to 3 year prior to serum collection. The 85 PCR negative samples showed schistosomal antibodies by ELISA (n = 4), IHA (n = 10) or both (n = 15) and/or belonged to patients suspected because of recent freshwater exposure or eosinophilia.
Urogenital schistosomiasis remains an important public health problem affecting approximately 112 million people, about half of the worldwide schistosome infections [2]. Schistosomiasis imported by travelers, expatriates and migrants is often caused by S. haematobium, with a frequency in the same range to that of S. mansoni [6], [17]–[19].
The presently developed PCR was designed to be used as a highly sensitive diagnostic tool for urogenital schistosomiasis in travelers returning from endemic regions. We opted for a real-time PCR format which has a short turn-over time and is preferred over conventional PCR methods due to its lower risk of contamination and higher sensitivity [20], [21]. The latter is of particular interest in travelers as they have often low parasite loads in the acute phase [6], rendering confirmation by microscopy erratic. The draPCR was able to detect all microscopy-confirmed S. haematobium infections in urine, bladder wall biopsy and feces and demonstrated no cross-reaction in clinical samples with microscopy-confirmed S. mansoni and other intestinal or blood parasites. Moreover, the draPCR detected ten extra S. haematobium-positive samples (7 urine and 3 stool samples) from 8 non-egg excretor patients that were highly suspected for urogenital schistosomiasis based on recent freshwater exposure, a strong antibody response (IHA≥1/1280) and/or the presence of eosinophilia. This confirms previous findings that PCR is highly sensitive in urogenital schistosomiasis diagnosis [22]–[24]. The draPCR can be of value on urine and stool samples of suspected patients when no eggs can be demonstrated by microscopy, especially as sampling is not invasive. Alternatively, the genus-specific PCR [7] could be used on urine and feces enabling the detection of schistosome DNA, regardless the causal species.
Besides the excellent performance of the draPCR on urine and stool samples, the most striking result of this study is the specific detection of S. haematobium in serum. All but one of the serum samples from patients with a confirmed S. haematobium infection, and none of the serum samples from patients with a confirmed S. mansoni infection, were positive with the draPCR. Schistosomal DNA could additionally be detected in the serum of one patient about 5 weeks after freshwater exposure and 42 days before confirmation of the S. haematobium infection by microscopy. This clearly demonstrates the diagnostic potential of the draPCR to detect S. haematobium in serum during the acute phase of the infection.
One serum sample of a patient with a confirmed S. haematobium infection was negative, which could be explained by previous treatment or the rather low volume (1 mL) of serum analysed. Due to the retrospective design of this study, PCR could not always be performed on an adequate volume of serum. We consider 2 mL as the optimal volume required for analysis.
Two extra S. haematobium infections were detected by the draPCR in serum samples of suspected patients. False-positivity by PCR seems very unlikely as both patients had recent exposure to freshwater in Mali, developed typical severe symptoms related to Katayama syndrome with hypereosinophilia and had a positive serological response five to eight weeks post-exposure. Moreover, S. haematobium DNA was also detected in the follow-up serum samples while urine and feces remained negative after treatment. The decreasing PCR signal in follow-up samples demonstrates the PCR's potential to semi-quantitatively monitor treatment. Extra studies are required to confirm this. Further research could additionally compare the persistence of detectable levels of parasite DNA in serum with levels of circulating antigen that are more related to the actual worm burden and rapidly decrease after treatment [25]. Also, more scientific data is needed to assure when parasite DNA is cleared from the blood stream after treatment in order to determine the discriminating power of PCR between active or past present infections with the same Schistosoma species. PCR positivity was seen until at least 3 months after treatment in the present study, and up to more than one year after treatment in patients with S. mansoni infections [10].
A drawback of this study is that we had no genital samples available to test with the draPCR. S. haematobium parasites can cause genital schistosomiasis [1], [26] but in travelers, only few cases were reported [26], [27]. Only one bladder wall biopsy was tested, and although this does not allow drawing conclusions, it is worth to mention that examination of tissue samples by PCR might be helpful to diagnose schistosomiasis in non-egg excretor individuals. Another drawback is that due to the retrospective sample collection, early acute phase samples were not frequently available because antibody or egg detection followed weeks later. A prospective study is needed with selection of patients during the (early) acute phase in order to compare the performance of different tools for urogenital schistosomiasis diagnosis [24].
The PCR targets Dra1, a sequence specific for S. haematobium [13] and previously successfully used in a conventional PCR on cercaria and infested snails [28] and on urine [24]. Similar to the Sm1-7 121-bp tandem repeat sequence that comprises 11% of the S. mansoni genome [8], [10], [29] and to the multicopy retrotransposon gene representing 14% of the S. japonicum genome [30], Dra1 also has a highly repetitive nature and represents 15% of the S. haematobium genome [13]. The presence of multiple copies of this target sequence enables the highly sensitive detection of S. haematobium DNA in serum and probably explains why the single copy 28S gene used in the genusPCR [7] was not successful to detect schistosome DNA in this matrix.
The findings of this study demonstrate that the draPCR for detection of S. haematobium infections in serum is complementary to the Sm1-7PCR that is most sensitive to detect S. mansoni infections [9]–[10]. Furthermore, we demonstrated the species-specificity of both PCRs in control DNA of adult worm extracts. Apart from the strong signal for the human species of the S. haematobium complex group, we also observed a very weak signal for S. mansoni with the draPCR and a weak signal for S. haematobium with the Sm1-7PCR. This can be explained by the fact that the highly repetitive sequences of S. haematobium or S. mansoni respectively, are most likely present in at least a single copy in the genome of other Schistosoma species [13] and are only detectable when a huge amount of parasite DNA is present as in the case of the adult worm extracts. Since we did not detect a signal with the draPCR in all 43 clinical samples of patients with confirmed S. mansoni infections (egg load varying between 10 and 120 EPG), we conclude that the draPCR and Sm1-7PCR are suitable for analysis of serum of patients suspected for urogenital and intestinal schistosomiasis, respectively.
What do we detect by the draPCR in serum, urine and feces?
In analogy with prenatal diagnostics and oncology [31], [32], Wichmann et al [10] used the term ‘cell-free parasite DNA’ (CFPD) to comprise the DNA that was detected in serum. Indeed, due to the high parasite turnover, diverse stadia of the parasite might be present in the blood circulation and are detectable depending on the phase of the infection. Once penetrated through the human skin, schistosomules travel with the venous circulation to the lungs within 7 to 10 days and thereafter to the liver region for maturation [1]. In the acute phase of urogenital infections, the draPCR probably detects DNA of degrading schistosomules or juvenile worms that did not survive or mate. Schistosomes are complex multicellular eukaryotes, and the schistosome DNA in serum might also originate from rapid turn-over of the tegument during maturation of the worm [33]. In addition, after the upstream migration of mature worms to the venous plexus of the bladder and deposition of eggs, DNA of eggs that circulate into the systemic circulation due to retrograde venous flow could be detected. In chronic infections, DNA from desintegrated eggs, or from killed worms after treatment could also be a target for PCR in serum.
In urine samples, the PCR primarily detects DNA from S. haematobium eggs. It is not unlikely that also transrenal nucleic acids of breakdown products of the parasite are detectable in the urine as demonstrated before for S. mansoni [34], [35] and other parasitic infections [36], [37]. In feces, DNA of S. haematobium eggs can be found due to the atypical location of the worm in the colon or rectal wall, or due to contamination of the stools with urine in case of a high-intensity infection [1], [38].
So far, no S. haematobium-specific PCR has been described before to be used in human serum of recently infected travelers. Our findings suggest that the draPCR in serum is suitable for diagnosis of urogenital schistosomiasis in a non-endemic setting and might be of value in diagnosing travelers during the acute phase of infection (4 to 6 weeks after exposure to infested water) before eggs excretion and seroconversion, and in light infections. Serology tests turn positive only about 6 to 12 weeks after exposure [6]. In addition, weak positive serological reactions are difficult to interpret and false-negative tests occur, especially with S. haematobium [4], [39], [40]. Further prospective evaluation of the draPCR on serum samples is needed, to demonstrate its diagnostic role during the early acute phase of the infection.
|
10.1371/journal.pgen.1003427 | Evolution after Introduction of a Novel Metabolic Pathway Consistently Leads to Restoration of Wild-Type Physiology | Organisms cope with physiological stressors through acclimatizing mechanisms in the short-term and adaptive mechanisms over evolutionary timescales. During adaptation to an environmental or genetic perturbation, beneficial mutations can generate numerous physiological changes: some will be novel with respect to prior physiological states, while others might either restore acclimatizing responses to a wild-type state, reinforce them further, or leave them unchanged. We examined the interplay of acclimatizing and adaptive responses at the level of global gene expression in Methylobacterium extorquens AM1 engineered with a novel central metabolism. Replacing central metabolism with a distinct, foreign pathway resulted in much slower growth than wild-type. After 600 generations of adaptation, however, eight replicate populations founded from this engineered ancestor had improved up to 2.5-fold. A comparison of global gene expression in wild-type, engineered, and all eight evolved strains revealed that the vast majority of changes during physiological adaptation effectively restored acclimatizing processes to wild-type expression states. On average, 93% of expression perturbations from the engineered strain were restored, with 70% of these occurring in perfect parallel across all eight replicate populations. Novel changes were common but typically restricted to one or a few lineages, and reinforcing changes were quite rare. Despite this, cases in which expression was novel or reinforced in parallel were enriched for loci harboring beneficial mutations. One case of parallel, reinforced changes was the pntAB transhydrogenase that uses NADH to reduce NADP+ to NADPH. We show that PntAB activity was highly correlated with the restoration of NAD(H) and NADP(H) pools perturbed in the engineered strain to wild-type levels, and with improved growth. These results suggest that much of the evolved response to genetic perturbation was a consequence rather than a cause of adaptation and that physiology avoided “reinventing the wheel” by restoring acclimatizing processes to the pre-stressed state.
| Acclimatizing and adaptive (evolutionary) processes allow organisms to thrive despite cellular and environmental perturbations. Our work examined whether adaptation restores stress responses towards wild-type (pre-stressed) versus novel physiological states during adaptation by studying a bacterium (Methylobacterium extorquens AM1) that was experimentally engineered and evolved with a novel central metabolism. The engineered strain was much slower and less fit than wild-type, but eight replicate populations evolved for six hundred generations showed substantial improvements. We found that changes in gene expression during adaptation consistently restored acclimatizing processes to the wild-type state, often in 8/8 evolved lines. Novel changes were common and largely restricted to one lineage; however, highly parallel novel changes revealed loci harboring beneficial mutations. Even rarer were reinforced changes, such as pntAB transhydrogenase, which increased beyond immediate acclimation during evolution to restore NAD(P)(H) metabolism and improve growth. Overall, a few novel or reinforcing changes drove the mass-restoration of physiology back to wild-type.
| Physiological stressors affect organisms across individual and evolutionary timescales: they invoke in individuals processes that work to restore homeostasis, and become over evolutionary timescales the selective pressures that drive adaptation in populations. How organisms generate innate and evolved responses to stressors – often termed physiological acclimation (or phenotypic plasticity) and adaptation, respectively – is a driving question today in many different fields of science, from the origins of drug resistance to the effects of global climate change. A common goal in many of these areas is to move from case-by-case studies towards a predictive understanding of how organisms will adapt to future stressors. However, whereas acclimatizing responses are generally “prewired” and relatively uniform between individuals of a population, the paths and outcomes of adaptation can be many and varied. Even under a simplified scenario of consistent selective pressures across replicate populations, evolution is not deterministic. There are many potential explanations for this variability - such as the randomness of mutations, escaping drift, epistasis, and clonal interference - all of which can give rise to multiple and sometimes quite disparate evolutionary outcomes [1]–[7]; yet, in other instances, adaptation is remarkably parallel between independently-evolved lineages, even down to the genetic level [8]–[10].
In replicate populations of laboratory-evolved organisms, parallelism is commonly interpreted as a sign of selection in either genetic [8] or phenotypic [11] data. Most studies determine the basis and parallelism of adaptation by comparing ancestral versus evolved states. However, in cases of adaptation to an environmental or genetic perturbation, there exists a third “wild-type” state that existed prior to the exposure to stressors that is often ignored. Exposure to genetic or environmental stressors invokes numerous processes that shift organisms from a wild-type to a perturbed physiological state, and it is this perturbed physiological state that is optimized over evolutionary timescales by natural selection. Thus, during experimental evolution all evolved strains share an initial set of acclimatizing responses that could be resolved differently by natural selection across replicate lineages. Given only a comparison of the ancestral (perturbed) and evolved states, it would be unclear how much of parallel adaptation represents convergent evolution of truly novel physiology, versus a wholesale restoration of cellular function to the pre-perturbed state. This has the potential to greatly conflate which physiological changes are likely causes versus consequences of improved fitness, and falsely identify highly parallel instances of adaptive evolution. To our knowledge, only one other study [12] has explicitly addressed the extent to which organisms adopt novel versus restored physiological states during adaptation to an environmental stressor, versus a genetic alteration.
By including data on the wild-type state prior to an environmental or genetic perturbation, it becomes possible to distinguish which evolved changes were truly novel versus simply altering the acclimatized state. This allows physiological changes to be categorized into four patterns (Figure 1A): restored, unrestored, or reinforced refer to whether acclimatizing responses were reversed, left unchanged, or enhanced through evolution, whereas novel changes did not manifest during initial acclimation, but appeared only later during evolution. Importantly, these classifications can be applied to various levels of physiological processes – from alterations in gene expression, to protein activity, metabolite concentrations and flux, and even higher-order properties such as growth rate or fitness – and could conceivably differ between levels. Ultimately, this framework provides a “direction” to orient the interpretation of physiological changes that occurred during adaptation, revealing the level and degree to which adaptation either restores prior cellular states or finds novel solutions to improve growth or fitness. We hypothesized that physiological changes that are simply restorative would occur commonly, and would frequently arise in parallel between replicate evolved lineages. By sorting out these restorative changes, the novel and reinforcing changes that remain should more clearly reflect the physiological bases of adaptation. Particularly when these novel or reinforcing changes occur in parallel, they may identify loci in which the causative, beneficial mutations occurred.
As a model system in which to examine the interaction between acclimation and adaptation to perturbations, we employed a combination of metabolic engineering plus experimental evolution to study physiological and evolutionary responses to a novel, sub-optimal central metabolism in Methylobacterium extorquens AM1. As a facultative methylotroph, M. extorquens AM1 is capable of utilizing one-carbon (C1) compounds like methanol as a sole source of carbon and energy, as well as other multi-carbon compounds like succinate [13]. Its metabolism of C1 compounds is a complex process that requires over 100 different genes [14], many of which were acquired via horizontal gene transfer [15], [16]. C1 substrates such as methanol or methylamine are oxidized first to formaldehyde, and in wild-type (WT), this toxic intermediate is then oxidized to formate using a pathway linked to tetrahydromethanopterin (H4MPT), an analog of folate [15], [17] (Figure 1B). From formate, C1 units can be further oxidized into CO2 for the production of NADH, or assimilated into biomass [18], [19]. To create an engineered Methylobacterium (EM) strain, the native H4MPT-based pathway of formaldehyde oxidation was disabled and replaced by a functionally analogous, yet non-homologous C1 pathway. Two genetic changes were required to make EM: (1) the deletion of the mptG locus, which encodes the enzyme that drives the first committed step in H4MPT biosynthesis and is necessary for growth or survival in the presence of methanol [17], and (2) the introduction of an expression plasmid with two genes – flhA and fghA, both from Paracoccus dentrificans – that drive the oxidation of formaldehyde to formate using glutathione (GSH) as a C1 carrier [20]. The introduction of the engineered GSH-dependent pathway restores the ability of the ΔmptG strain to grow on methanol, however this EM strain is approximately 3-times slower growing than WT. Furthermore, the EM strain exhibits morphological abnormalities that arose from overexpression of the foreign GSH pathway [20].
Eight replicate populations (F1–F8) were founded from an EM ancestor and propagated on methanol for over 600 generations in batch culture to study adaptation to a novel metabolic module. Adaptation in the F populations was substantial, rapid, and largely methanol-specific [20]. The cellular abnormalities that emerged as a consequence of introducing the foreign pathway were also eliminated, representing an example of a restored (morphological) change. Several beneficial mutations have been identified in these evolved lines, including four from an isolate from the population with the highest fitness gains (F4). Notably, all four of these beneficial mutations altered gene expression. The targets and apparent physiological pressures acting upon these beneficial mutations are as follows. (1) Increased pntAB expression: switching from the native to the engineered pathway of formaldehyde oxidation eliminated the cell's only direct source of NADPH production, and a transhydrogenase encoded by pntAB can overcome this limitation by reducing NADP+ to NADPH using NADH [21]. (2) Increased gshA expression, which encodes an enzyme in GSH biosynthesis: GSH is needed to react with formaldehyde in the engineered pathway, and its recruitment into central metabolism might dilute GSH away from its native functions to protect against oxidative stress [22]. (3) Increased icuAB, which encodes a cobalt transporter: this mutation allowed cells to overcome metal limitation in the medium [10]. And (4), decreased expression of the introduced GSH pathway (i.e. flhA and fghA) [20], [23]: Foreign genes and plasmids introduced through engineering or natural gene transfers are often sub-optimal in terms of their sequence, expression, or activity for their new host and function [23]–[25]. Correspondingly, mutations that decreased expression of flhA and fghA balanced the benefits of formaldehyde oxidation with the costs of gene expression, and these occurred in all eight evolved populations through a variety of genetic mechanisms [23]. While these mutations to pntAB, gshA, icuAB, and the foreign pathway are known to have improved fitness in one or more lineages, many other changes in cellular physiology were also altered as a consequence. It is unclear whether these mutations produced novel or restorative physiological states, nor the extent to which these changes occurred in parallel across replicate populations.
Here, we sought to examine the extent to which evolution creates truly novel physiological states in the F lines, versus simply restoring acclimatizing processes towards WT-like levels. To this end, we used DNA microarrays to analyze changes in global gene expression from WT, to EM, to each of the eight F strains, and classified significant changes into patterns of restored, unrestored, reinforced, and novel gene expression as in Figure 1A. Without knowledge of acclimatizing processes, the substantial transcriptional changes observed in the evolved lineages would have been perceived as novel; however, our analysis revealed an overwhelming trend towards restoring gene expression to the WT state. Furthermore, whereas over 300 genes restored expression in parallel across all eight replicates, novel or reinforced changes tended to be unique to one or a few populations. Rare examples of parallelism amongst the novel or reinforced changes were particularly enriched for the loci with known beneficial mutations described above, or other probable candidates. One example of a highly parallel and beneficial reinforced change – PntAB transhydrogenase – translated into a restorative change in physiology, as it appeared to return NAD(P)(H) pools perturbed in EM back toward WT levels. Thus, incorporating information from physiological acclimation to a genetic or environmental perturbation can “orient” the interpretation of evolutionary adaptation, thereby distinguishing restorations from novelty, and greatly enriching for physiological changes that were causes, and not just consequences of increased fitness.
To interpret gene expression and other physiological data, we first characterized the growth rate and competitive fitness of WT, EM, and isolates of each F population after 600 generations of evolution. Relative growth rates measured using a high-throughput, automated robotic system [26]–[28] indicated the evolved isolates were now 1.95 to 2.5 times that of their EM ancestor on methanol, while WT was 3 times as fast (Figure 1C). Improvements in the F isolates were similar to the gains previously measured at the population level [20], suggesting that our isolates were representative of their respective evolved populations. Furthermore, these improvements in growth rate correlated well with increases in relative fitness, as determined by head-to-head competition [29], confirming that selection focused largely on exponential growth (Figure 1C).
Given that the sole difference between the slow EM strain and WT was the replacement of the formaldehyde oxidation pathway, we hypothesized that adaptation in the F lines would largely focus upon this stage of methanol metabolism. To test this hypothesis, we determined the specific growth rate of strains on two additional C1 substrates: methylamine, and formate. Growth on methylamine is nearly identical to growth on methanol, except that formaldehyde enters C1 metabolism by way of methylamine dehydrogenase [30]; while, in contrast, growth on formate skips the steps of formaldehyde oxidation altogether (Figure 1B). Relative to EM, the improvement of strains on methylamine was nearly comparable to their respective gains on methanol, while there were much smaller gains on formate (Figure 1D). The large difference between improvement in the selective environment (with methanol) versus formate, or succinate [20], contrasts with the generic improvements across substrates that was observed after adaptation of WT on methanol [29]. Overall, these data suggest that selection in the evolved lineages was focused predominantly focused on the formaldehyde oxidation pathway of C1 metabolism required for both methanol and methylamine growth.
To investigate large-scale changes in physiology arising due to the replacement (acclimation) and subsequent evolution (adaptation) of the formaldehyde oxidation pathway, we used DNA microarrays to examine differences in global gene expression. We identified 878 genes that were differentially expressed relative to EM: 455 of which arose as acclimatizing responses to metabolic engineering, while the remaining 423 genes appeared only in the evolved isolates. Patterns of restored, unrestored, reinforced, or novel gene expression were categorized by following the fate of EM perturbations (if present) into each of the evolved lineages. Due either to experimental noise or an intermediate reversal in gene expression, a significant number of genes fell in-between our criteria for restored and unrestored, and were thus classified as a fifth group of “partially restored” changes. We present changes in gene expression in two ways (Figure 2A). First is a scatter plot that depicts both the changes that occurred during acclimation to the introduced pathway (WT vs. EM, x axis), versus those that occurred during adaptive evolution (EVO vs. EM, y axis). Second, we present a histogram (grey box) that compiles these data solely in terms of the changes that occurred during adaptation.
The majority of gene expression changes that occurred in the evolved strains were not novel, but restored perturbations to a WT-like state. Genes whose expression was fully or partially restored greatly outnumber the other categories. This is apparent by the large number of restored and partially restored genes in the scatter plot (Figure 2A), as well as by tabulating the number of genes satisfying each category across independent evolved isolates (Figure 2B, blue and purple). The next most numerous category was novel changes, followed by unrestored and then reinforced. As an additional method to explore similarity between transcriptional profiles, we used principal component analysis (PCA). Including all significant expression changes, PC1 clearly separats EM from WT and the evolved isolates. In contrast, PC2 distinguishes three evolved isolates from the remainder: separating F4 from F1 and F8 (Figure 2C). This highlights that, despite the great degree of parallelism in restorative gene expression, the transcriptomes of a few F lines appear to be quite distinct. Considering just those genes that are perturbed in EM (i.e., acclimatizing responses, with no novel changes), all evolved isolates cleanly fall between EM and WT, while F4, F1, and F8 remain quite distinct (Figure S1A). Considering just novel changes, only F4 and the pair of F1 and F8 are distinct from the rest (Figure S1B).
Given that most of the 455 genes with perturbed expression in EM were restored during adaptation, we hypothesized that this class of changes may be particularly likely to occur in parallel across the F lines. For each gene with significant expression changes, we tabulated how many instances of each class occurred across the F lines (Figure 3). Only about 10% (46/455) of perturbed genes satisfied the strict criterion for restoration in all eight populations (solid blue). By grouping strictly restored genes with cases of partial restoration, 72% (328/455) perturbed genes were restored across all eight populations (dashed-blue), and 98% (444/455) moved toward WT in at least four populations. In contrast, partially restored changes had little affect when combined with fully unrestored genes. This tremendous degree of parallelism was not observed for novel expression changes. Over 70% (330/483) of these occurred in just one strain, and of those that occurred in two populations, 81% (83/102) of these were specific to the F1 and F8 isolates that exhibited a particularly distinct transcriptional pattern.
Gene expression that was restored or unrestored in perfect parallel highlights the major acclimatizing responses of EM to perturbations invoked by metabolic engineering. The 46 genes that were restored across all lineages function in heat shock and stress responses (including recA), C1 metabolism (components of methanol dehydrogenase), chemotaxis response regulators, and various genes putatively of phage origin. Conversely, only two genes – a glycine riboswitch (META1_misc_RNA_19) and a conserved hypothetical protein (META2_0338) were never restored in any lineage. In the adaptation of the F lines, we hypothesized that a greater number of restored genes would indicate a closer return to the wild-type state, and thus faster growth. This could occur either directly through mutations to pathways controlling the expression of perturbed genes, or secondarily as other physiological processes are restored (e.g., increase in growth rate or decrease in stress). Contrary to this hypothesis, however, growth rate did not correlate with either increased instances of restored or partially restored gene expression, or with a decrease in unrestored expression (Figure S2A–S2C). Furthermore, none of the four aforementioned loci known to have experienced beneficial mutations fell into this class. This suggests that the overall extent to which expression of evolved isolates returned to the WT-like state is not a good indicator of growth improvement.
Many if not most evolution experiments focus only on changes that are novel with respect to their ancestor. Provided with only knowledge of the EM state and the commonly-used threshold of 2-fold differential expression, our analysis would have wrongly classified many instances of fully or partially restored and reinforced expression as being novel in the evolved lineages (see histogram in Figure 2A). However, by incorporating WT physiology, we were able to identify 423 genes whose expression is wholly novel in at least one evolved lineage. The number of genes with novel expression varied between the F strains from only 12 in F2, to 217 in F4 (Figure 2B). Most instances of novel expression were unique to one or a few evolved lineages, with the exception of a few loci. One might therefore expect that an increased number of novel changes would be correlated with higher fitness, however no correlation was found between growth rate and increased instances of novel gene expression (Figure S2D). In fact, the F1 and F8 strains share a large number of uniquely derived changes in gene expression - with functions in DNA transcription and translation, DNA synthesis, and a number of C1-related genes – yet are amongst the least improved lineages. In F4, many novel down-regulated genes (Figure 2A) are in fact instances of gene loss from a previously identified deletion on the M. extorquens AM1 megaplasmid [20], [31] that has been shown to be beneficial and recurring across experiments [32]. While individual cases of novel gene expression are no doubt important to growth and fitness gains in the F isolates, we found that these are in general less frequent than restorative changes, mostly restricted to one or a few strains, and on the whole a poor indicator of improvements gained in the F lines.
The rarest, and perhaps most interesting class of gene expression changes were those that were reinforced, in which the acclimatizing response of EM to metabolic engineering was augmented through the evolutionary process. We identified only 30 genes with reinforced expression in at least one evolved isolate (7% of perturbed genes), which include the increased expression of the pntAB operon, the up-regulation of two genes with putative functions in cobalamin biosynthesis, the down-regulation of genes with predicted functions in fatty acid metabolism, and other genes with poorly-annotated functions that were down-regulated. Most genes with reinforced expression were unique to one or a few F strains, and remained unrestored or were restored in the other isolates (Figure 3). Unlike the above tests, instances of reinforced expression were strongly correlated with improvements in growth rate (Figure S2E; R2 = 0.87, p = 0.005), however the sample size of reinforced changes is small. As described above, pntAB was known to contain a beneficial mutation in its promoter in F4 [20], and these data now show that increased expression at this locus was not novel, but rather a response that arose first in the acclimation of EM and was reinforced through evolution.
We hypothesized that highly parallel instances of novel or reinforcing changes in gene expression might be enriched for loci with beneficial mutations. Although 306 genes showed parallel changes in expression across at least six populations, and 453 genes were either novel or reinforcing, only 5 instances were observed that satisfied both criteria, and they were all novel. We identifed those loci with known beneficial mutations (gshA and icuAB); one other gene with parallel increases (META1_0936, a putative type I secretion membrane fusion protein); and two genes with parallel decreases (META1_2657, a putative soxC sulfite oxidase; and META2_1007, a putative beta-lactamase). Regarding the parallelism of reinforcing changes, the three loci composing the pntAB operon were increased in 8/8 lineages relative to EM, but only significantly so in 4/8 lineages. This suggests that beneficial mutations may be particularly common in expression changes that are both parallel and buck the trend of restoration to the WT-like state.
PntAB transhydrogenase functions in redox homeostasis, and thus it was intriguing to find that perturbed pntAB expression was not restored but actually increased further away from WT levels. We hypothesized that the consequence of increased pntAB could actually be restorative to the levels of pyridine nucleotides NAD(H) and NADP(H), despite its enhanced expression. We first examined the evolved isolates for mutations in the pntAB locus beyond that known for F4, and only one other strain – F3 – had a similar mutation in the upstream region (Figure 4A). Next, we investigated whether increased expression of pntAB equated to increased enzyme function. Transhydrogenase activity measured in WT, EM, and each of the evolved isolates closely mirrored changes in the expression of pntAB measured in the microarray analysis (Figure 4B). Our data suggest that, outside of F3 and F4, increased transhydrogenase in the F strains occurs either as a consequence of outside physiological changes (e.g., via allostery) or through trans-acting factors that drive increased expression in these lineages. We further examined the relationship between transhydrogenase levels and growth rate, and found a highly correlated positive relationship amongst the evolved F isolates (Figure 4C).
To determine whether the effect of increased transhydrogenase activity in the evolved strains was to restore the redox balance of pyridine nucleotides, we examined strain differences in the ratios of NADPH/NADP+ and NADH/NAD+. Interpreting changes in the steady-state concentrations of metabolites such as NADPH is complicated by the fact that these values represent a balance between production (such as by transhydrogenase) and consumption via biosynthesis. As there is relatively little degeneracy in the network of biosynthetic reactions, the rate of NADPH use should be nearly directly proportional to growth rate, such that mutations that increase the cell's capacity to grow can actually decrease the steady-state concentration of currency metabolites. Indeed, data from a variety of other organisms, such as Escherichia coli [33] and Lactococcus lactis [34] grown at different rates in chemostats have confirmed this intuition.
Consistent with the above expectations, the slow-growing EM strain possessed a much higher ratio of both NADPH/NADP+ and NADH/NAD+ than WT. Including the evolved isolates, the ratios (or redox state) of reduced to oxidized NADP(H) and NAD(H) were both highly negatively correlated with growth rate, such that faster-growing strains possess substantially lower ratios for each (Figure 5A and 5B, respectively). Variation in levels of NADPH/NADP+ also correlated well with changes in pntAB expression: strains with significant increases in pntAB (n = 4 strains) showed significantly lower NADPH/NADP+ ratios than those with marginal increases (p<0.05, Welch two-sample t-test); however, the same was not true for NADH/NAD+ ratios. Even amongst strains with significantly increased pntAB expression, those with cis-acting mutations (F3 and F4) were significantly faster and had lower NADPH/NADP+ ratios than strains with significant increases apparently driven in trans (F2 and F7). Importantly, almost all strains statistically significantly restore the redox states of NADP(H) and NAD(H) towards WT-like levels. Overall, these data suggest that the reinforcement of transhydrogenase activity increased the rate of NADPH production and drove the restoration of pyridine nucleotide metabolism back toward a WT-like state through an apparent variety of adaptive mechanisms.
Organisms are constantly pressured by ever-changing and potentially disruptive cellular and environmental conditions. Large-scale changes in physiology can occur due to ecological or environmental transitions, or upon sudden changes in genomic composition due to mutation, horizontal gene transfer, or genetic engineering in the laboratory. When a perturbed, sub-optimal physiology persists over multiple generations, transient acclimatizing responses begin to overlap with responses from evolutionary adaptation. Conceptually, processes of physiological acclimation and adaptation are intimately linked: as beneficial mutations should revert many acclimatizing processes from a perturbed to a baseline physiological state. In practice, the interplay between acclimatizing and adaptive responses to perturbations has often been ignored, leading to the scenario where large-scale, parallel restoration of physiology to a pre-stress state will appear as novel. We argue that a proper interpretation of evolved physiological states is only possible given knowledge of the initial acclimation to a new environment or genomic composition.
Our work sought to determine the extent to which cells adopt novel versus restorative physiological states by examining acclimatizing and adaptive responses to a novel central metabolism. We utilized a strain of M. extorquens AM1 (EM) that was metabolically engineered to utilize a foreign, GSH-based central pathway to oxidize formaldehyde during growth on C1 compounds, and was subsequently propagated in eight replicate F populations for over 600 generations of evolution to optimize growth using the engineered pathway. The physiology of the EM ancestor was perturbed in many ways: it was three-fold slower; adopted an elongated, curved or branched cell morphology [20], and exhibited a unique density-threshold for growth on methanol [35]. Here we document two additional levels of physiological perturbation: microarray analyses revealed 455 genes with altered expression from WT to EM, as well as perturbations in the central redox cofactors, NAD(H) and NADP(H). By orienting our analyses based upon the initial acclimation from WT to EM, we categorized evolved changes as restored, unrestored, reinforced, or novel as in Figure 1A. Given the particularly interesting connection between acclimatizing and adaptive processes in reinforcement, we further examined the systems-level consequences of enhanced PntAB activity.
The major pattern seen for evolved changes in physiology was an overwhelming trend to return to a wild-type state. Our work highlights a few general trends to be explored in other systems. First, the majority of gene expression differences distinguishing the ancestor and the evolved isolates were not novel, but instead restorative. Most restored genes were not themselves targets of beneficial mutations, but altered in response to other changes such things as NAD(P)(H) levels, or indirectly, improved growth rate (increased methanol dehydrogenase), or reduced stress (decreased recA and heat shock proteins). So much of gene expression was restorative that it outweighed instances of novel expression in all evolved strains. Similarly, PCA analysis confirmed that expression in the evolved isolates was more like WT than their common EM ancestor. One interesting future direction would be to examine the temporal component of adaptation, studying the degree to which physiology is restored as populations acquire sequential beneficial mutations. Second, the restoration of WT physiology occurred highly in parallel. Indeed, the vast majority of genes were restored, at least partially so, in all eight lineages. This is perhaps intuitive as a shared set of acclimatizing processes from EM were simply “turned off” in the case of stress-related responses, or “turned up” in the case of growth related genes, in the evolved lines. Without specific knowledge of these acclimatizing processes, however, most of these restorative changes would be wrongly classified as novel (Figure 2A, histogram). Third, some acclimatizing processes were left unrestored because physiological adaptation cannot, or has not yet, addressed these perturbations. These may represent fundamental and perhaps inescapable differences separating WT and EM physiologies. And finally, changes that are both highly parallel and either novel or reinforced are potentially enriched for loci targeted by beneficial mutations, and thus causal changes during adaptation. Increases in expression in gshA (6/8 novel), icuAB (6/8 novel), and pntAB (4/8 reinforced) are all outliers when comparing the parallelism of changes in each category across genes (Figure 3). In fact, by filtering out (highly parallel) restorative changes, we find only 19 genes (out of 878) that are novel or reinforced changes in half or more of the evolved strains. Including the parallel, beneficial decreases in the expression and/or activity of the foreign pathway that occurred in 8/8 strains [23], parallel changes in gene expression that are not restorative appear to be particularly enriched for beneficial mutations that drove adaptation.
Looking closer, we did find variation in how the various F lines adapted to an engineered C1 metabolism. Novel expression of genes very rarely occurred in more than one strain, and where observed, it was nearly always to the F1 and F8 strains. These isolates consistently showed different transcriptional profiles than the other F isolates, not only amongst novel genes, but also in the number, types, and degree to which genes are restored. Interestingly, both F1 and F8 are also amongst the slowest growing of the F strains, suggesting perhaps the presence of a multi-peaked fitness landscape in which these strains have found a local optimum. While it appears that the F populations restored many genes in parallel, and share at least a few common molecular and physiological mechanisms, additional work is needed to understand the full extent to which these strains found parallel versus divergent paths to optimize growth using an engineered central metabolism.
Reinforcing changes to physiology, while rare in our system, are an important link between processes of physiological acclimation and adaptation. We focused on one particular instance of reinforcement – the up-regulation of pntAB transhydrogenase – to investigate both the genetic basis for enhancing expression beyond acclimation and to uncover its physiological consequences. Normally, pntAB is expressed during multi-C and not C1 growth [36], however in EM, the only direct source of NAD(P)(H) production was lost with the deletion of the native pathway of formaldehyde oxidation. This perturbation might invoke increased pntAB to maintain NAD(P)(H) homeostasis during growth on methanol. Supporting this hypothesis, the deletion of pntAB was found to be neutral for C1 growth in WT but lethal in the EM strain (H.-H. Chou, data not shown), and the mutation in the F4 lineage that drives increased pntAB expression provides a 10% selective benefit in the ancestral background [20]. These results demonstrate the irreplaceable role of pntAB as an acclimatizing response in EM, and the benefit of reinforcing this function even further through adaptation.
While the increased expression and activity of PntAB transhydrogenase was reinforcing, this translated into a restorative effect upon metabolism. All eight F strains increased transhydrogenase activity significantly, despite significant increases in expression for only half of these. Upon sequencing the genomic neighborhood of pntAB, we identified only two strains – F4 and F3 – that possess known or candidate mutations to drive increased expression. As for the physiological consequence of increased transhydrogenase activity, all evolved strains tend to restore NAD(P)(H) metabolism, and strains with greater increases to pntAB – particularly the pair with mutations in the upstream region – have levels of NAD(H) and NADP(H) that are the closest to WT. The reinforcement of pntAB expression and transhydrogenase activity, as well as the restoration of NAD(P)(H) levels, are both well correlated with increased growth rate in the F populations. By increasing activity during acclimation, and reinforcing this response further during adaptation, transhydrogenase activity appears to have been critical in maintaining and improving growth in the EM strain.
Information on acclimatizing and adaptive responses in the engineering and evolution of EM allowed us to develop a framework in which to examine the true nature of evolved physiological change. We defined four basic patterns to describe not only novel changes to physiology, but also changes that restore, disregard, or reinforce the initial acclimatizing responses to perturbations (Figure 1A). To our knowledge, this linkage between immediate physiological acclimation and subsequent adaptation has explicitly been explored only once before [12]. This paper described a large number of “compensatory” changes in gene expression that effectively restored the wild-type (glucose-grown) state during the acclimation and experimental evolution of E. coli to sub-optimal carbon sources. The commonalities between adaptation to a poor environment versus a novel, suboptimal metabolic pathway are remarkable: whereas their study showed that 87% of genes were restored after adaptation to a poor substrate, we found that on average that 93% of genes were restored after adaptation to the foreign pathway; their change was environmental, ours genetic. Furthermore, reinforcing changes are reminiscent of the fixation of traits via genetic accommodation or assimilation [37]–[39], in that both processes stem from exposure to genetic or environmental stressors to reveal beneficial phenotypes that are “assimilated” and possibly reinforced by positive selection. However, in genetic assimilation, stress-induced phenotypes arise from cryptic genetic variation in populations [40] while, at least for the reinforced up-regulation of pntAB expression, the initial response required no standing genetic variation at all. In fact, the initial acclimatizing response of EM to increase pntAB was merely a generic response to NADPH shortage typically experienced during growth on multi-carbon substrates such as succinate [36], that was co-opted for methanol growth in EM, and further increased and optimized by selection during adaption of the F lines.
Overall, our results suggest that much of evolutionary adaptation effectively relieves processes of physiological acclimation. Rather than “reinvent the wheel” of C1 metabolism, a few causal mutations in the adaptation of the F populations propagated through physiology to restore WT homeostasis. In fact, more changes in gene expression occurred as a result of acclimation to genetic engineering (n = 455) than novel changes seen in any of the isolates after 600 generations of experimental evolution (12 to 217). Beneficial mutations were enriched toward novel and reinforcing changes that occurred in parallel. By distinguishing acclimatizing versus adaptive processes, a more accurate depiction on the nature and parallelism of physiological evolution is revealed.
All growth was performed using a modified “Hypho” minimal medium as in [20]. One liter of Hypho was prepared from 799 mL of deionized water, 100 mL phosphate salts (25.3 g of K2HPO4 plus 22.5 g NaH2PO4 in 1 L deionized water), 100 mL sulfate salts (5 g of (NH4)2SO4 plus 0.98 g MgSO4 in 1 L deionized water), and 1 mL of modified, high-iron “Vishniac” trace metal solution [10], [20]. All solutions were autoclaved separately and combined under sterile conditions, and the final medium was stored in the dark. Carbon substrates added just prior to inoculation consisted of: 20 mM methanol, 3.5 mM sodium succinate, 15 mM methylamine hydrochloride, or 20 mM sodium formate. Growth experiments were initiated by inoculating 10 µL of freezer stock into 9.6 mL Hypho in a 50 mL Erlenmeyer flask containing 10 mM methanol and 1.75 mM succinate plus 50 µg/mL kanamycin. Flasks were grown at 225 rpm in a 30°C shaker-incubator until reaching stationary phase (2–4 days). A second acclimation cycle was accomplished by transferring 150 µL of saturated culture into 9.45 mL fresh medium with 0.5× kanamycin plus the carbon substrate to be tested; then transferred again into the same conditions for experimental (measured) growth. A 1∶64 dilution of cultures with the given substrate concentrations allowed for six doublings per growth cycle.
All physiological assays (e.g., microarray analyses, enzyme assays, metabolite concentrations) were performed using cells that had reached half-maximal density following transfer from acclimation cultures also grown on methanol. This protocol results in eleven doublings of growth in a consistent environment while ensuring cells were still growing exponentially at the time of harvest. This gave the maximal possible time to approach steady-state physiology while staying within the constraints of the selective conditions. Furthermore, since it was previously found that the EM ancestor exhibits a unique cell-density threshold for growth [35], it would not have been possible to have diluted the cultures much more than the 1/64 used here.
Specific growth rates were determined in 48-well plates using a high-throughput, robotic system that automates measurements of optical density (i.e., OD600) in growing cultures at timed intervals [27]. This system consists of a plate-shaking tower, a plate reader, a robotic arm, and de-lidding station that transfer cultures between growth and measurements, all of which is scheduled with an open software manager program [26]. Strains for growth measurements were inoculated first into flasks, transferred to plates with for an acclimation phase, and transferred once more for measurement during the third cycle. All growth was performed in 640 µL total medium and were transferred in a 1/64 dilution (10 µL culture into 630 µL medium). To limit clumping and reduce noise in OD600 measurements in growing cultures, 0.1 mg/mL of prepared cellulase enzyme (Sigma-Aldrich, St. Louis, MO) was added to the growth medium (SMC, unpublished). The specific growth rate was calculated from the log-linear phase of growth for at least triplicate cultures of each strain using an open software analysis package [28].
Strains and plasmids relevant to this study are listed in Table S1 and were generated previously, unless otherwise noted. The ancestral strains for the F populations were described previously [20]. Briefly, they derive from two WT M. extorquens AM1 strains - one that is naturally pink (CM501), and another that is white (CM502) due to a neutral mutation in carotenoid biosynthesis [41] – to limit contamination between cultures. The EM strain was constructed in two steps: 1) the H4MPT-dependent pathway was disabled by deleting the mptG locus (encoding β-ribofuranosylaminobenzene 5′-phosphate synthase), the product of which drives the first committed step in the H4MPT biosynthesis [42]; and 2) the introduction of a GSH-dependent formaldehyde oxidation pathway on the plasmid pCM410 – which expresses the genes flhA (encoding S-hydroxymethyl-GSH dehydrogenase) and fghA (encoding S-formyl-GSH hydrolase) from Paracoccus denitrificans –into the ΔmptG backgrounds, generating completed pink (CM701) and white (CM702) EM strains [20].
Eight replicate populations were founded from either the pink (odd populations; CM701) or white (even populations; CM702) EM strains and evolved for over 600 generations in 9.6 mL Hypho medium plus 15 mM methanol in batch culture with transfers of 1/64 of the volume every four days for the first 300 generations, and every two days thereafter. These evolved “F” populations (F1-8) were streaked at generation 600 onto Hypho agar plates to isolate colonies for further characterization. In addition to the previously characterized isolate from the F4 population, CM1145 [20], we chose for this study the second of three random isolates from each of the other F populations for further investigation (Table S1).
Other strains relevant to this study were as follows. Fluorescence-based fitness assays required an EM reference strain (CM1232) that had been generated by replacing the katA locus with mCherry driven by a constitutive Ptac promoter [20]. To standardize the use of kanamycin in all cultures, we used a WT strain in which the kan resistance marker was inserted into katA (CM611) [29].
The relative fitness of WT and evolved strains was assessed in a head-to-head competition of co-cultures with a fluorescently-labeled reference as in [29]. Briefly, fully-grown cultures of WT and each evolved isolate were mixed in roughly equal optical densities with an EM strain expressing mCherry (CM1232). A sample of this mixture (T0) was diluted with Hypho plus 8% DMSO and stored at −80°C in 96-well plates; the rest was diluted 1∶64 into 640 µL of Hypho methanol medium in a 48-well plate and incubated with shaking at 30°C for 4 days, after which samples of the co-culture after competition (T1) were frozen for later analysis using flow cytometry. Because of the 4-day growth cycle, this amortizes fitness over all growth phases (i.e., lag, exponential, and stationary).
The ratio of labeled to unlabeled cells before and after co-culture growths was measured using a BD LSR Fortessa flow cytometer with an HTS attachment for 96-well plates (BD Biosciences, San Jose, CA). Recently it was found that the forward scatter (FSC) and side scatter (SSC) settings used in earlier work [20] systematically underestimated fitness increases relative to EM because of the cells' larger size. Here we set both scatter measurements set to 300 V to accommodate small bacterial cell sizes [23], and the flow-rate was adjusted to the lowest setting to produce reliable measurements of labeled and unlabeled events in dilute co-cultures. The ratio of nonfluorescent to fluorescent cells before (R0, from T0) and after (R1, from T1) competition were used to calculate the fitness (W) of strains relative to the EM reference (CM1232) using the following formula, assuming a 64-fold expansion of cells from six doublings per growth cycle:
Triplicate cultures of strains were grown to half-maximal OD600 in 15 mM methanol before harvesting and total RNA extraction using the RNeasy kit (Qiagen, Valencia, CA). Genomic DNA was removed using the TURBO DNA-free kit (Ambion, Austin, TX) and the RNA samples were concentrated using Amicon Ultra centrifugal filters (Millipore, Billerica, MA). Microarray analyses for all (n = 30) samples were performed by MOgene, Inc (St. Louis, MO) using one-color cDNA labeling and hybridization. The array probes and platform were designed previously [43] to include 60-mer oligonucleotides that provide two or more probes for confirmed and predicted ORFs in the Methylobacterium genome [31]. Raw and normalized expression data are available from the Gene Expression Omnibus, accession GSE42116. Pre-processing, normalization, and analysis of expression data was performed using the limma package [44], [45] with Bioconductor [46] and R [47]. Differentially expressed genes were identified by the proportion of differentially expressed probes in a limma contrast given: 1) at least two-thirds probes significant at p<0.05 in the moderated t-statistic, 2) at least one-half of probes significant at p<0.01, and 3) all significant probes with uniform changes either up or down. Probes that met these criteria were averaged in each strain to estimate the log2 difference in expression relative to EM. Genes differentially expressed in both EVO:EM and EVO:WT contrasts, and in the same direction, were classified as novel. Expression perturbations from acclimation were identified in a WT:EM contrast and further partitioned given information from EVO:EM and EVO:WT contrasts to define patterns of: restored expression, given an EVO:EM change back in the direction of WT expression; unrestored expression, given no EVO:EM difference but a significant EVO:WT difference; and reinforced expression, having an EVO:EM difference in the same direction (up or down) as the change from WT to EM (Figure 1A). Partially restored genes showed no EVO:EM difference and were not significant in a EVO:WT contrast. Principal component analysis was used to cluster and contrast the expression profile of WT, EM, and evolved strains, and was calculated using the prcomp function in R with scaling to account for large variance of expression changes between genes.
Cultures for the determination of TH activity and NAD(P)(H) ratios (below) were grown to half-maximal OD600 on methanol, spiked with another 15 mM methanol, and allowed to return to mid-exponential growth for approximately 16 hours to increase yield. Cultures for transhydrogenase activity measurements were pelleted and washed with 50 mM Tris-HCl (pH 7.5) before storage at −80°C. Upon thawing, cells were re-suspended in 2 mL Tris buffer and lysed by bead beating (MP Biomedicals, Solon, OH). Cell extracts were centrifuged for less than 15 s to collect the beads. The supernatant was removed and combined with a reaction mix consisting of: 20 µL of 40 mM MgCl2 (10×), 20 µL 5 mM NADPH (10×), 20 µL 10 mM 3-acetylpyridine adenine dinucleotide (10×), plus Tris buffer to equal 200 µL, total, in a 96-well plate. The increase in absorbance at 375 nm was measured immediately after addition of the reaction mix and the slope of the linear regression was used to calculate transhydrogenase activity (µmole of 3-acetylpyridine adenine dinucleotide reduced sec−1 mg−1) as follows: TH activity (µmole sec−1 mg−1) = slope (sec−1)×1/exctinction coefficient (0.0051 mol cm L−1)×1/path length (0.42 cm−1)×reaction volume (0.2 mL)×1/cell protein (mg)×1000 (conversion to µmole L−1).
Cell extracts for the measurement of pyridine nucleotide concentrations were prepared as follows. Metabolism in mid-exponential cells was quenched using vacuum-filtration and rapid immersion into hot extraction solutions. Oxidized pyridine nucleotides (NAD+ and NADP+) were selectively preserved in an acidic extraction solution consisting of 100 mM HCl plus 500 mM NaCl; reduced species (NADH and NADPH) were extracted using a basic solution of 100 mM NaOH plus 500 mM NaCl. For both acidic and basic extractions, 750 µL of culture was vacuum-filtered onto 0.45 µm nylon membranes (Millipore, Billerica, MA), immediately immersed into the appropriate extraction solution, briefly vortexed, and heated to 95°C for 5 m. Extracts were again briefly vortexed, centrifuged at maximum speed for 30 s, and the supernatant removed, flash frozen, and stored at −80°C for later use. Three biological replicates stemming from separate inoculations were extracted for each strain.
Pyridine nucleotides in cell extracts were quantified using enzymatic cycling [48] with alcohol dehydrogenase (ADH) or glucose-6-phosphate dehydrogenase (G6PDH) to measure NAD(H) and NADP(H), respectively. Each assay was performed using 20 µL of either acidic extraction solutions for oxidized species, basic solutions for reduced, or a serial dilution of (reduced) standards. For NAD(H), to 20 µL of cell extract or standard was added 180 µL of master solution consisting of: 20 µL 1 M bicine (pH 8.0) plus 40 mM EDTA (10×), 20 µL of 16.6 mM phenazine ethosulfate (10×), 20 µL of 4.2 mM thiazolyl blue tetrazolium bromide (10×), 20 µL of 100% ethanol, 2 µL of ADH (Sigma-Aldrich, St. Louis, MO) at 0.1 U/µL, and 98 µL water. The same mixes were used for the determination of NADP(H) with G6PDH, except that ethanol and ADH were replaced by 20 µL of 50 mM glucose-6-phosphate (10×) and 2 µL of G6PDH at 0.1 U/µL. Assays were conducted in 96-well plate format and measured in a Safire2 spectrophotometer (Tecan, Morrisville, NC) at 30°C by following the increase in absorbance at 550 nm over time.
|
10.1371/journal.pntd.0004868 | Visceral Leishmaniasis on the Indian Subcontinent: Modelling the Dynamic Relationship between Vector Control Schemes and Vector Life Cycles | Visceral leishmaniasis (VL) is a disease caused by two known vector-borne parasite species (Leishmania donovani, L. infantum), transmitted to man by phlebotomine sand flies (species: Phlebotomus and Lutzomyia), resulting in ≈50,000 human fatalities annually, ≈67% occurring on the Indian subcontinent. Indoor residual spraying is the current method of sand fly control in India, but alternative means of vector control, such as the treatment of livestock with systemic insecticide-based drugs, are being evaluated. We describe an individual-based, stochastic, life-stage-structured model that represents a sand fly vector population within a village in India and simulates the effects of vector control via fipronil-based drugs orally administered to cattle, which target both blood-feeding adults and larvae that feed on host feces.
Simulation results indicated efficacy of fipronil-based control schemes in reducing sand fly abundance depended on timing of drug applications relative to seasonality of the sand fly life cycle. Taking into account cost-effectiveness and logistical feasibility, two of the most efficacious treatment schemes reduced population peaks occurring from April through August by ≈90% (applications 3 times per year at 2-month intervals initiated in March) and >95% (applications 6 times per year at 2-month intervals initiated in January) relative to no control, with the cumulative number of sand fly days occurring April-August reduced by ≈83% and ≈97%, respectively, and more specifically during the summer months of peak human exposure (June-August) by ≈85% and ≈97%, respectively.
Our model should prove useful in a priori evaluation of the efficacy of fipronil-based drugs in controlling leishmaniasis on the Indian subcontinent and beyond.
| Visceral leishmaniasis is a disease caused by a virulent vector-borne parasite transmitted to man by phlebotomine sand flies. Fipronil-based drugs, administered to cattle orally, provide a potential means of sand fly control by permeating in cattle blood and being excreted in cattle feces, targeting adult females feeding on cattle blood and larvae feeding on cattle feces, respectively. An agent-based, stochastic simulation model was developed to represent sand fly population dynamics in a village in Bihar, India, at all developmental stages, with the goal of predicting the impact of various vector control strategies, utilizing drug treated cattle, on vector population numbers. Results indicate that success of treatment is dependent on the number of treatments applied annually and the seasonality of the sand fly lifecycle. Results further suggest that treatment schemes are most effective in reducing vector populations when high drug efficacy is maintained in cattle feces during periods of high larval density. Our approach incorporates detailed representation of the vector population and provides an explicit representation of the effects of insecticide application on adult and larval sand flies. Hence, this model predicts treatment schemes that may have the greatest potential to reduce sand fly numbers.
| The deadliest form of leishmaniasis, visceral leishmaniasis (VL), is vector-transmitted through the bite of phlebotomine sand flies in the Phlebotomus and Lutzomyia genera. This protozoan parasite results in an estimated 500,000 human infections and 50,000 human fatalities annually, making it the second most prevalent parasitic killer on Earth, behind only malaria [1,2]. The highest global rate of occurrence is on the Indian subcontinent with approximately 67% of all human instances occurring in India, Bangladesh and Nepal in areas of extreme poverty and high population density [3]. Bihar is the most impoverished, most densely populated, and most VL-endemic state in India, with 90% of the Indian VL cases reported there [4]. The VL pathogen, Leishmania donovani, is described as anthroponotic on the Indian subcontinent with humans acting as the only known reservoir for infection [5].
The known VL vector on the Indian subcontinent is the sand fly species Phlebotomus argentipes [6]. Phlebotomine sand flies are small Diptera, rarely exceeding a length of 3 mm, in the family Psychodidae and subfamily Phlebotominae [7] and are holometabolous consisting of four life stages: eggs, larvae, pupae, and adults. Sand flies are active primarily at night and are regarded as silent feeders [8]. P. argentipes females host blood feed primarily on cattle and humans within rural villages [9–12]. The blood meal is required in order to complete the oviposition process. Immature sand flies in Bihar have been found largely in areas within and surrounding cattle sheds [13–15], suggesting cattle feces may serve as a food source for larvae which feed on organic matter. Results of several laboratory experiments have found sand fly processes such as development, mortality and reproduction to be temperature-dependent with many of these processes occurring more rapidly at higher temperatures [16–20]. Nightly air temperatures in Bihar will exceed 20°C between March-October and are highest during the summer (June-August) and the observed sand fly population is small in January and February when minimum temperatures are lowest [21].
It has been suggested that further research regarding alternative or integrated vector control approaches should be examined to supplement the current practice [22]. Vector control in India comes in the form of indoor residual spraying (IRS) performed historically with DDT and more recently with synthetic pyrethroids. IRS controls endophilic sand flies, but blood-fed sand flies have been collected outdoors and indoors [9,21,23]. A survey concluded that roughly 95% of Bihari villager households have family members that sleep outdoors at least part of the year [24]. Logically, these villagers are therefore not protected by IRS and are potentially exposed to exophilic sand flies.
Fipronil-based drugs, orally administered to cattle and rodents, have been successful in killing laboratory-reared sand flies under controlled conditions, targeting blood-feeding adults and larvae that feed on host feces [25–27]. Orally applied fipronil can remain in the system of animals for several weeks to several months, dependent on the concentration administered (mg/kg body weight) and fipronil has a lengthy half-life of approximately 128 days [28], meaning that sand fly control can potentially be maintained for several months following a single treatment. With this form of treatment, the success of vector control could be independent of exophilic or endophilic feeding behavior and be dependent on host and oviposition site preferences. Hence, this form of treatment could potentially supplement the current practice of IRS by targeting exophilic, cattle-feeding adult sand flies and larval sand flies feeding on organic matter in the form of cattle feces.
A reduction in vector density should lead to a reduction in the transmission rate of VL as suggested by a recent VL model which predicted that either reducing vector density >67% through application of adulticides or >79% through breeding site destruction could eliminate the ability of the VL pathogen to persist [29]. Vector and pathogen seasonality in addition to social practice should be taken into consideration when developing a control plan. Not only should overall vector density be considered, but one also should consider vector density during spring/summer months (April-August) when villagers could potentially be at greatest risk of exposure to infected sand flies. Clinical VL in Bihar is commonly reported between the months of April and August [30]. Proper bed net usage during the warmer months of the year has been found to be strongly protective against VL [31]. However, several publications suggest that bed net usage in Asia and Africa declines in response to increased temperature [32–37].
Susceptible-Infected-Recovered (SIR) compartment models, and variants of this, have been developed in the past to represent VL epidemiology within human populations on the Indian subcontinent. The first such model examined three historical VL epidemic peaks in Assam, India which occurred between 1875 and 1950 and concluded that intrinsic processes related to host and vector dynamics, rather than extrinsic factors such as earthquakes or influenza outbreaks, provided the simplest explanation of the timing of the peaks [38]. More recent models representing VL epidemiology within human populations in Bihar have examined VL underreporting [39], antimony resistant VL [40], VL treatment, prevention, and control [41], and more specific vector control strategies, namely the application of adulticides and destruction of sand fly breeding sites [29]. The latter model represented the application of adulticides and the destruction of sand fly breeding sites via variables that reduced sand fly life expectancy and breeding site capacity, respectively, and predicted the impact of reducing vector density on the ability of the pathogen to persist (as indicated by the basic reproduction number Ro [42]).
The model originally published by [40] and subsequently used in [39] and [29] is a deterministic SIR-type model that focuses heavily on the natural history of VL infection within human populations, represented by 11 distinct stages. However, the vector (sand fly) population is represented by only three stages: the susceptible, latent, and infectious, with abundance of the latter used to calculate the VL transmission rate to humans. Emergence rate of susceptibles and mortality rates of each stage are held constant. The egg, larval, and pupal stages of the sand fly life cycle are not represented in this model, or in any other SIR-type VL model to the best of our knowledge. These limitations to current SIR-type models have been recognized and the exploration of individual-based, stage-structured, stochastic modelling approaches has been recommended [29], which could allow explicit evaluation of stage-specific impacts of vector control strategies on sand fly populations in Bihar.
In this paper we describe an individual-based, stochastic, stage-structured model that represents a temperature-driven sand fly vector population within a village in Bihar, India and simulates the effects of vector control through the use of fipronil-based drugs orally administered to cattle. The model does not include a human population or VL pathogen, but rather focuses on the effects of fipronil-induced mortality of larval and adult life stages on sand fly population dynamics. We first describe the model and evaluate its performance. We then use the model to simulate several fipronil-based control schemes in which we vary treatment frequency and timing of treatment application, focusing on resulting reductions in sand fly populations during spring/summer and especially during the period of peak human exposure (June-August). We also examine sensitivity of model predictions of treatment efficacy to parametric uncertainty.
The model represents the lifecycle of sand flies as they develop from eggs to larvae to pupae to pre-reproductive adults to pre-oviposition adults to reproductive adults to post-reproductive adults, as well as fipronil-induced larval and adult mortality (Fig 1). Rates of development, natural mortality, and reproduction depend on the environmental temperatures to which the sand flies are exposed. Eggs, larvae, and pupae are exposed to temperatures of the organic matter in which they develop, whereas adults are exposed to ambient temperatures. Natural mortality of larvae also depends on the density of larvae in the organic matter in which they are feeding. Fipronil-induced mortality occurs in adult flies that obtain a blood meal from fipronil-treated cattle, and in larvae that feed on feces from fipronil-treated cattle. Simulations are run on a daily time step, thus all rates and probabilities described below are calculated on a daily basis. Eggs, larvae, and pupae are represented as daily cohorts whereas adults are represented as individuals. That is, the size of each daily cohort of eggs that enter the system is monitored as these eggs develop into larvae and then into pupae. When a cohort of pupae develops to the adult stage, the resulting adults are represented as individual organisms and are followed through pre-reproductive, pre-oviposition, reproductive, and post-reproductive stages (only adult females are represented in the model).
Below we present the equations used in the model to represent the development, reproduction, natural mortality, and fipronil-induced mortality of sandflies.
To calculate rates of development of immature stages (eggs, larvae, pupae), we drew upon results of laboratory experiments conducted under constant temperatures [16,18,19] and then estimated temperature-dependent development under variable temperature regimes using the general equation described by [43]: 100/nl = K/[1+ exp(a − bx)]. This is a bisymmetrical, sigmoid curve with the distance between the lower and upper developmental temperature thresholds (K) estimated as K=[2C1C2C3−C22(C1+C3)]/(C1C3−C22), where C1, C2, and C3 are values for 100/nl on the curve at three temperatures on the abscissa. We represented the temperature-dependent development of eggs, larvae, and pupae as:
Ci,Eggs=0.5/[1+exp(−0.1601⋅Ti,O+5.6067)]
(1)
Ci,Larvae=0.0688052/[1+exp(−0.4754⋅Ti,O+11.298)]
(2)
Ci,Pupae=0.25/[1+exp(−0.2736⋅Ti,O+7.7067)]
(3)
where Ci,Eggs, Ci,Larvae, and Ci,Pupae represent the contributions of the current daily temperature on day i toward the development of eggs, larvae, and pupae, respectively, and Ti,O represents current temperature (°C) within the organic matter on day i (Fig 2a–2c). The model accumulates Ci over time separately for each cohort, and when ΣiCi = 1.0 for a given cohort, the organisms in that cohort advance to the next developmental stage (Fig 1).
After pupation, pre-reproductive adults must obtain a blood meal to advance to the pre-oviposition stage (Fig 1). We estimated the daily probability of obtaining a blood meal based on laboratory experiments in which 0, 3, 60, 85, 94, and 96% of flies obtained their first blood meal by the end of their first, second, third, fourth, fifth, and sixth day, respectively, as an adult [44], and used these results to develop the following curve:
Pi,Blood Meal=0.940321952 /[1+exp(−3.7061⋅Di,PE+10.551)]
(4)
where Pi,Blood Meal is the probability of a pre-reproductive adult obtaining a blood meal on day i and Di,PE is the number of days-post-emergence from pupation (Fig 3a).
We estimated the temperature-dependent development of adults from the pre-oviposition stage to the reproductive stage based on laboratory data collected by [19] in the same manner as described above for eggs, larvae, and pupae:
Ci,POAdults=0.363755/[1+exp(−0.1503⋅Ti,A+4.0206)]
(5)
where Ci,POAdults is defined and calculated in the same manner as the analogous terms in Eqs 1 through 3, except that Ti,A represents current air temperature (°C) on day i rather than temperature within organic matter (Fig 2d). When ΣiCi = 1.0, flies advance from the pre-oviposition to the reproductive stage (Fig 1).
Females lay eggs the day they advance from the pre-oviposition to the reproductive stage. We represented the number of eggs laid per reproductive female (Eqs 6 and 7) as a function of temperature (Fig 3b) based on laboratory observations [19]:
If Ti,A ≤ 28.5 then
Pi,OElAdults=25.1464684/[1+exp(−0.5238⋅Ti,A+12.441)]
(6)
If Ti,A > 28.5 then
Pi,OElAdults=25.1464684−25.1464684/[1+exp(−0.5238⋅Ti,A+12.441)]
(7)
where Pi,OElAdusts represents the number of eggs laid by a female on day i and Ti,A represents the current air temperature (°C) on day i. However, no eggs are laid if Ti,A ≤ 15C [19].
After oviposition, reproductive females have a 90% chance of becoming post-reproductive and a 10% chance of returning to the pre-reproductive stage [17]. If they return to the pre-reproductive stage, the daily probability of obtaining another blood meal is calculated using Eq 4, except Di,PE is redefined as the number of days since returning to the pre-reproductive stage.
Natural mortality of cohorts of eggs, larvae, and pupae depend on the temperature (Ti,O) of the organic matter (Fig 4a–4c) in which they are located, whereas natural mortality of adults depends on air temperatures (TiA) (Fig 4d). We represented the temperature-dependent natural mortality of eggs, larvae, pupae, and adults based on laboratory experiments conducted by [16,45]:
Pi,MEggs=0.00052737⋅Ti,O2−0.02872971⋅Ti,O+0.39946900
(8)
If Ti,A ≤ 28.5 then
Pi,MlLarvae=0.3898*exp(−0.156*Ti,O)
(9)
If Ti,A > 28.5 then
Pi,MuLarvae=0.0000000000144*exp (0.68195*Ti,O))
(10)
Pi,MPupae=0.00004973⋅Ti,O2−0.00261400⋅Ti,O+0.03635092
(11)
If Ti,A ≤ 10 then
Pi,Ml Adults=0.5556*exp(−0.239*TiA)
(12)
If Ti,A > 10 then
Pi,Mu Adults=0.0005*exp(0.1918*TiA)
(13)
where Pi,MEggs, Pi,MlLarvae, Pi,MuLarvae, Pi,MPupae represent the proportion of eggs, larvae, and pupae dying on day i and Pi,Ml Adults and Pi,Mu Adults represent the daily probability of dying for adults on day i. Independent of temperature, we estimated the maximum longevity of adults in the wild to be 30 days based on findings presented by [46].
We also represented density-dependent natural mortality of larvae based on rates of cannibalism observed in laboratory experiments conducted with different larval densities [47]:
Pi,MLarvae_C=(ri,Can)*Ni,Larvae
(14)
where Pi,MLarvae_c represents the proportion of larvae dying on day i due to cannibalism, Ni,Larvae represents the number of larvae in the system on day i, and ri,Can represents a proportional increase in cannibalism as the number of larvae increases. The data presented by [47] suggest an approximately linear relationship between larval density and rate of cannibalism, the slope of which (ri,Can) we calibrated, as described in the Model evaluation and Model calibration sections below.
In addition to depending on the frequency of treatment application and the proportion of the cattle treated, which we represented as management variables, rates of fipronil-induced mortality depend on (1) the proportion of adult sand flies that feed on cattle, (2) the proportion of larvae that feed in organic matter containing cattle feces, (3) the efficacy of fipronil contained in the blood of cattle, and (4) the efficacy of fipronil contained within cattle feces. We assumed that 50% of adult flies obtain their blood meal from cattle [9] and that 90% of eggs are laid on, and hence larvae develop in, organic matter containing cattle feces [13].
We represented the efficacy of fipronil within the blood of cattle as decreasing exponentially as a function of the number of days after fipronil application:
P′i,MAdults=0.515 ⋅exp(−0.094⋅Di,PT)
(15)
where P’i,MAdults represents the daily probability of dying for an adult fly that obtained a blood meal from treated cattle Di,PT days post-treatment (days after application of fipronil) [25] (Fig 5a). Once an adult obtains a blood meal from a treated cow, we assumed that its daily probability of dying due to fipronil did not change, that is, efficacy of the fipronil within the fly remained constant.
We represented the proportion of larvae dying due to fipronil within cattle feces as decreasing exponentially as a function of the number of days post-defecation [28,48]:
P′i,Larvae=Ej⋅exp(−0.00545⋅Di,PD)
(16)
where P’i,MLarvae represents the proportion of larvae dying on day i that are feeding on feces of treated cattle Di,PD days post-defecation (Di,PD days after the feces were deposited), assuming the feces were deposited j days after application of fipronil. The initial (maximum) efficacy of fipronil in cattle feces (Ej) (Eq 17) itself decreases exponentially over time [25]:
Ej=0.567⋅exp[−0.073(Di,PT−1)]
(17)
For example, fresh feces deposited 1 day after cattle are treated have a higher efficacy than fresh feces deposited 2 days after cattle are treated (Fig 5b). We assumed that fipronil-induced mortality and natural mortality were completely additive.
To evaluate the potential usefulness of the model in simulating the population-level response of sand flies to fipronil-induced mortality, we first verified that the model simulated adequately the rates of development, reproduction, natural mortality, and fipronil-induced mortality observed under laboratory conditions. That is, that the model code produced simulated data that mimicked the laboratory data from which it was parameterized when we simulated the laboratory experiments. We next calibrated the model to represent environmental conditions typical of Bihar, India such that the simulated population established a seasonally-varying, dynamic equilibrium under baseline conditions (without fipronil-induced mortality). We then evaluated performance of the baseline model by (1) assessing the ecological interpretability of seasonal trends in the simulated sand fly life cycle and (2) comparing simulated fluctuations in abundance of adult sand flies to fluctuations observed in each of three villages in Bihar over a 12-month period. We ran 10 replicate stochastic (Monte Carlo) simulations for each portion of the model evaluation procedure, except for the simulations required for verification of the temperature-dependent development and mortality rates of eggs, larvae, and pupae, which were deterministic.
We calibrated the model to represent environmental conditions typical of Bihar, India by representing annual fluctuations in (1) simulated air temperatures (Ti,A) with a time series of 365 minimum daily air temperatures recorded at a village in Bihar [21] and (2) simulated temperatures within organic matter (Ti,O) by fitting a cosine curve to a graphical representation of annual fluctuations in soil temperatures in West Bengal, India presented by [49] (Fig 10). We further calibrated the model by adjusting the parameter controlling the density-dependent mortality of larvae due to cannibalism (9.5 x 10−7) such that the simulated population established a seasonally-varying, dynamic equilibrium under baseline conditions (without fipronil-induced mortality) in which the mean annual abundance of adults was approximately equal to the breeding site capacity, or number of vectors (7,344) estimated by [29].
We evaluated the baseline model by (1) assessing the ecological interpretability of seasonal trends in the simulated sand fly life cycle and (2) comparing simulated fluctuations in relative abundance of adult sand flies to fluctuations in relative abundance of adults caught in light traps in each of three villages in Bihar over a 12-month period using a Sign Test. Simulated seasonal trends were representative of the general temperature-dependent trends characteristic of the sand fly life cycle in Bihar (Fig 11). Simulated oviposition did not occur until mid-February (day-of-year 42), when temperatures first exceeded the 15 C threshold suggested by [19], with the first mass emergence of adults occurring 85 days later during May (day-of-year 127), and the largest peak in adult abundance occurring during the latter portion of July (day-of-year 205), as observed by [21]. Simulated egg, larval, and pupal fluctuations were impossible to validate due to lack of field data, but showed similarity to simulated adult fluctuations (Fig 12a–12d). The fluctuations of egg and pupal population densities were more distinct because the developmental periods are considerably shorter than that of larvae and adults.
Simulated fluctuations in relative abundance of adults were not significantly different from the general trends in relative abundance of adults caught in the three villages in Bihar (sign test: p < 0.1263, p < 0.5000, and p < 0.0704, respectively), although, not surprisingly, trends in the field samples were less distinct [21] (Fig 13). Interestingly, trends in relative abundance at one village (Mohammadpur) were markedly different from both the simulated trends and the trends observed at the other two villages (p < 0.10), most likely due to markedly lower abundances during September, October, and November.
Simulation results suggest that the success of fipronil treatments in controlling sand flies depends not only on the frequency of applications but also on the timing of applications relative to the sand fly lifecycle. Synchronizing applications to maintain high efficacy of the drug in cattle feces during the period of high larval abundance seems particularly important. While more frequent applications obviously are more efficacious, they also are more expensive and more difficult logistically. Thus, the ability to assess not only efficacy of treatment schemes per se but also their cost-effectiveness and their logistical feasibility is of paramount importance. Adequate a priori assessment of novel control schemes targeted at specific aspects of a vector’s life cycle requires novel approaches, including models that explicitly represent key aspects of the processes by which the control method intervenes in the life cycle of the target species.
Several previous studies of VL epidemiology have focused on villages in Bihar and have included models with detailed representations of disease dynamics within human populations [29,39–41]. One study modeled the effect of specifically-targeted sand fly control strategies including the application of adulticides and the destruction of breeding sites, which were represented by reducing sand fly life expectancy and breeding site capacity, respectively [29]. Their model predicted that either reducing vector density >67% through application of adulticides or >79% through breeding site destruction could eliminate the ability of the pathogen to persist, as indicated by the value of the basic reproduction number (Ro < 1.0; [42]). Although providing a wealth of details concerning disease dynamics within humans and valuable information regarding the general magnitude of vector reduction required to control transmission to humans, environmental factors affecting the vector life cycle, and hence the transmission process, were not modeled explicitly [29]. These authors recognized the limitations this imposed on use of their model and provided appropriate caveats [29]. Other studies also have modeled the effect on VL control via direct manipulation of model parameters controlling mortality rate [50,51] or biting rate [52] of adult sand flies, again providing valuable information pertinent to objectives of their studies, but without explicit representation of environmental factors affecting the seasonality of such rates.
By explicitly representing the effects of seasonally-varying temperatures on development and survival of the various sand fly life stages, our model has allowed initial assessment of a novel control scheme targeted specifically at both larvae and adults. By specifically examining the relationship among the timing and frequency of treatment applications, the duration of drug efficacy, and the seasonality of the sand fly lifecycle, we can make initial assessments not only in terms of reducing average sand fly abundance, but also in terms of cost-effective reduction of human exposure to sand flies given local social practice and availability of alternative hosts.
As a common social practice, some family members of the vast majority of Bihari villager households sleep outdoors, particularly during the months with the hottest evening temperatures (June, July, August) [21,24]. Although indoor residual spraying, when properly applied [22], and bed net usage [31] offer protection against infected sand flies feeding indoors, outdoor sleeping places a sizeable portion of the population at risk of exposure to infected exophilic sand flies during periods of peak sand fly abundance. Thus reduction of sand fly abundance during the period when villagers are most likely to be exposed to outdoor-feeding sand flies is of particular importance when assessing the efficacy of control schemes. Considering the results from our model in terms of the efficacy of the various treatment schemes to reduce sand fly abundance during this critical period.
Economically, Bihar is the poorest state in India, with roughly $100 million gross domestic product compared to the national average of $274 million [53], and people within the VL-endemic zones are among the most impoverished people in the world [54]. Thus considerations of cost effectiveness become paramount in terms of the commercial feasibility of drug application by villagers in VL-endemic regions. The cost of treating with this form of drug is estimated at approximately $1 per cow per treatment, but milk production per cow is estimated to increase by $0.50 per day [55], offering incentive to villagers to pay to treat their animals. Our simulations assumed 100% of cattle were treated, as would occur in a field trial. However, likely <100% of cattle would be treated if this form of treatment became commercially available, since individual livestock owners would be responsible for its application. In this regard, the influence of regional differences in socio-economic conditions on the efficacy of alternative sand fly control schemes should be evaluated closely. Although beyond the scope of the present study, a future use of our model could include a sensitivity analysis aimed at assessing the impact of regional socio-economic differences on the efficacy of different treatment schemes under different hypotheses regarding local economic conditions and social practice.
Potential environmental and human health impacts, as well as effects on non-target species, always are a concern when evaluating new vector control methods. In this regard, treating cattle orally with fipronil-based drugs may have benefits over conventional IRS. IRS often involves the application of insecticides to the walls of homes and cattle sheds, thus exposing human inhabitants as well as non-target species coming into contact with the walls. DDT has known environmental consequences, but chronic exposure also could potentially be linked to human health concerns such as pancreatic cancer [56–58] making the choice to switch to synthetic pyrethroids logical. Fipronil-based drugs can be safely administered to cattle, given the acute oral LD₅₀ for fipronil-fed rats is ≈97mg/kg of body weight [28], and exposure of cattle to orally applied drugs with fipronil at a concentration of 0.5 mg/kg of body weight can provide effective control of adult and larval sand flies. We are not suggesting treatment with fipronil-based drugs as a replacement for IRS, but rather as a complimentary component. The impact of current practices of IRS and bed net administration on the vector population and human VL transmission has been inconclusive and thus there is much interest in alternative control methods and integrated control schemes [22]. By targeting sand flies feeding on cattle outdoors and larvae developing in cattle feces (areas not targeted by IRS), fipronil-based drug treatment may prove to be a potentially important component of an integrated pest management program.
Further evaluation of the effects of sand fly control through the use of fipronil-based drugs orally administered to cattle ideally would involve a field trial in Bihar. Among the most critical data obtained from such an experiment in terms of increasing confidence in our model predictions would be those shedding light on the proportion of adult sand flies that obtain their blood meal from cattle and the proportion of eggs oviposited in organic matter containing cattle feces. By far the most restrictive assumptions we have made in our model are that 50% of adult sand flies obtain their blood meal from cattle [9] and that 90% of sand fly eggs are laid on, and hence larvae develop in, organic matter containing cattle feces [13]. Although it would have been desirable to use sand fly field collections [25] and blood meal analysis of field caught sand flies [9] to parameterize equations representing adult and larval sand fly developmental processes, currently this is infeasible. Locations of immature sand flies in the field are largely unknown, making field observations difficult. It is also difficult to ascertain the age at which blood fed sand flies collected from the field acquire a blood meal. In contrast, daily immature P. argentipes and P. papatasi processes and adult blood feeding probability and oviposition can be observed under controlled conditions in the laboratory [16–20, 44].
The interaction of vector feeding tendencies and host availability on the success of vector control based on systemic insecticides is a topic of current investigation. A recent study modeled the effect of assuming different hypothetical functional relationships between biting behavior of mosquitos (e.g., indiscriminate, anthropophilic, zoophilic) and human host availability on subsequent predictions of malarial infections [59]. This author’s model results indicated that control efficacy was dependent on both intrinsic host preferences and variation in encounter rates with alternative hosts. Our model results assumed that a fixed 50% of sand flies blood fed on cattle treated with fipronil and that the remaining 50% fed on alternative hosts that did not contain fipronil in their blood. Previous blood meal analysis suggests opportunistic feeding behavior of P. argentipes. Research conducted in eight districts in the West Bengal, a state neighboring Bihar, by [10] and in a single district by [12] suggested that P. argentipes blood feeding was not driven by particular host preference but rather by host availability, as they fed primarily on humans in human dwellings and cattle in cattle sheds. Research conducted by [60] suggests P.argentipes feed on cattle primarily, but feed readily on humans in human dwellings, with the researchers referring to P. argentipes as “chance feeders.” In our model, the probability of a sand fly acquiring a blood meal from cattle implicitly represents relative host availability. The results of our sensitivity analysis in which the mean number of SFD decreased in response to an increased proportion of adults feeding on cattle are indicative of the predicted tendency of indiscriminately feeding mosquitoes presented by [59]. It goes without saying that this treatment is dependent on the presence of cattle. Although P. argentipes feed indiscriminately, they have been reported to be zoophilic in the past as well [11]. Therefore, explicit representation of sand fly feeding tendencies, not unlike that presented by [59], may be a beneficial addition to our model in the future.
Empirical evidence regarding the substrates in which oviposition occurs is sparse. Sensitivity analysis indicated the obvious importance of assumptions which directly affect the exposure of simulated sand flies to the drug. Immature sand flies are typically collected from the floors of cattle sheds and human dwellings in India, but often in small numbers [13–15]. Studies have uncovered sand fly larvae in larger numbers from diverse locations such as forest floors [61] and abandoned sheds [62–64], and adult sand flies have been collected from palm tree canopies in Bihar [23], suggesting oviposition may be occurring in a variety of microhabitats.
Other data limitations affecting parameterization of the present model included the need to use developmental data from laboratory studies of another species of Phlebotomus rather than field data on our target species. The use of P. papatasi, in addition to P. argentipes, data for model parameterization was necessary because laboratory data for P. papatasi are more abundant and the lifetables developed for them [18–19] are from studies conducted under a wider range of temperatures than those of P. argentipes [16]. The P. papatasi data provide upper and lower thermal limits [45], and thus provide at least a general basis for predicting how phlebotomine sand flies function at extreme temperatures. Although, similarities have been observed between these two species in terms of development [17], we suggest that laboratory studies be conducted to further study the temperature-dependence of the incriminated vector of VL on the Indian subcontinent, P. argentipes.
Notwithstanding the inevitable parametric uncertainties associated with the current model, we would suggest that our model structure might be adapted for initial evaluation of fipronil-based sand fly control under a range of different environmental conditions involving a variety of potential hosts. For instance, the VL vector in East Africa, Phlebotomus orientalis, is highly zoophilic in Ethiopia but feeds heavily on donkeys in addition to cattle [65,66], suggesting the need to include multiple host species. The hare is a potential reservoir for VL in Spain [67], suggesting a different route of administration may be required, such as a fipronil-based pour-on or a grain bait, rather than a bolus. Dogs are the primary reservoir for zoonotic VL in the Americas [68] and insecticide impregnated dog collars have shown promise in reducing the VL infection rate in dogs [69,70], suggesting that fipronil impregnated dog collars could provide a means of targeting sand flies in Latin America. Although the host species treated and route of drug administration would need to be modified depending on the situation, the current structure of our model should accommodate the necessary re-parameterizations.
While 90% of reported VL cases occur in six countries on the Indian Subcontinent: India, Bangladesh, South Sudan, Sudan, Ethiopia, and Brazil [71], both VL and cutaneous leishmaniasis (CL) also are present in Europe and climatic projections suggest that the Central European climate will become increasingly suitable for sand flies capable of vectoring VL and CL [72]. The large number of refugees fleeing to Europe from countries such as Afghanistan and Iraq, where CL is known to occur and where clinical VL cases have been documented [73], creates the potential for widespread leishmaniasis outbreaks. Our hope is that our model will prove useful in the a priori evaluation of the potential role of treatment schemes involving the use of fipronil-based drugs in the control of leishmaniasis on the Indian Subcontinent and beyond.
|
10.1371/journal.ppat.1000430 | Atg5-Independent Sequestration of Ubiquitinated Mycobacteria | Like several other intracellular pathogens, Mycobacterium marinum (Mm) escapes from phagosomes into the host cytosol where it can polymerize actin, leading to motility that promotes spread to neighboring cells. However, only ∼25% of internalized Mm form actin tails, and the fate of the remaining bacteria has been unknown. Here we show that cytosolic access results in a new and intricate host pathogen interaction: host macrophages ubiquitinate Mm, while Mm shed their ubiquitinated cell walls. Phagosomal escape and ubiquitination of Mm occured rapidly, prior to 3.5 hours post infection; at the same time, ubiquitinated Mm cell wall material mixed with host-derived dense membrane networks appeared in close proximity to cytosolic bacteria, suggesting cell wall shedding and association with remnants of the lysed phagosome. At 24 hours post-infection, Mm that polymerized actin were not ubiquitinated, whereas ubiquitinated Mm were found within LAMP-1–positive vacuoles resembling lysosomes. Though double membranes were observed which sequestered Mm away from the cytosol, targeting of Mm to the LAMP-1–positive vacuoles was independent of classical autophagy, as demonstrated by absence of LC3 association and by Atg5-independence of their formation. Further, ubiquitination and LAMP-1 association did not occur with mutant avirulent Mm lacking ESX-1 (type VII) secretion, which fail to escape the primary phagosome; apart from its function in phagosome escape, ESX-1 was not directly required for Mm ubiquitination in macrophages or in vitro. These data suggest that virulent Mm follow two distinct paths in the cytosol of infected host cells: bacterial ubiquitination is followed by sequestration into lysosome-like organelles via an autophagy-independent pathway, while cell wall shedding may allow escape from this fate to permit continued residence in the cytosol and formation of actin tails.
| M. tuberculosis is one of the world's most prevalent pathogens, infecting one third of humans and contributing to 2 million deaths each year. M. marinum (Mm) has been increasingly studied as a model of M. tuberculosis due to its relative safety and its shared mechanisms of pathogenesis; for example, previous studies have highlighted the importance of the secretion system ESX-1 in pathogenesis of both M. tuberculosis and Mm. Here, we show that the host's ubiquitin system, best known for tagging host proteins for degradation, also recognizes Mm that have escaped their original phagosomes via an ESX-1–dependent mechanism and entered into the cytosol. Ubiquitinated bacteria have two distinct fates; ubiquitin tagged Mm can be sequestered into lysosome-like compartments, but they also can shed cell wall, which may allow them to evade sequestration. Lysosomal sequestration is independent of autophagy, a response to starvation or infection that leads to degradation of organelles and certain other intracytosolic pathogens. These experiments reveal ubiquitination as a mechanism of host cell recognition of mycobacteria and the existence of a new interface between the host and invading microbe that may be amenable to therapeutic intervention.
| Mycobacterium marinum (Mm) is a close genetic relative of the important human pathogen M. tuberculosis, and shares with M. tuberculosis the ability to infect host macrophages, as well as to induce a similar pathologic response [1],[2]. As in mammalian models of infection with M. tuberculosis, loss of the ESX-1 secretion system greatly attenuates the virulence of Mm in its natural fish and amphibian hosts [3]–[5]. While the reason that ESX-1 is required for M. tuberculosis virulence remains unknown, there is some understanding of its role in pathogenesis of Mm infections. After Mm infection of host cells, ESX-1 is required for bacterial escape from the phagosome, recruitment of uninfected macrophages, and subsequent cell-to-cell spread [6]–[8]. Although whether or not M. tuberculosis ever escapes from phagosomes remains highly controversial [9],[10], loss of ESX-1 activity in this organism also leads to failure of cell-to-cell spread [11].
Some Mm polymerize actin, forming actin tails after entering the host macrophage's cytosol [12]. However, this represents only a fraction (∼25%) of the intracellular bacilli; the fate of the remaining 75% of infecting bacteria has yet to be established. To explore the fate of intracellular Mm, we developed an assay that accurately distinguishes intraphagosomal from cytosolic bacteria. Surprisingly, bacteria escaped from phagosomes to the cytosol many hours before actin polymerization was evident. A proportion of these newly-escaped cytosolic bacteria were ubiquitinated, as was bacteria-derived material that appeared to be shed from the mycobacterial surface and incorporated into host membranous structures.
Distantly related pathogens which can escape the phagosome, Salmonella typhimurium and Listeria monocytogenes, also associate with ubiquitin, but whether this represents direct ubiquitination of the bacterial surface or close apposition of ubiquitinated host proteins remains unproven, and the significance of the association is unclear [13]. Similar to ubiquitin associated Listeria, ubiquitinated Mm did not form actin tails; rather, they became associated with LAMP-1 positive macrophage lysosomes, suggesting a reuptake process from the cytosol.
The process of autophagy, which engulfs organelles and proteins into characteristic double membrane vacuoles during cell stresses including starvation and formation of intracytosolic protein aggregates [14],[15], is a mechanism for host defense against some intracellular pathogens escape primary phagosomes, including Salmonella, Listeria, Streptococcus pyogenes, and Shigella flexneri [16]–[19]. For each of these organisms, autophagic host defense is dependent on Atg5, a central component in the formation of autophagic membranes. Autophagy may also be important in host defense against Mycobacteria. Enhancement of autophagy in macrophages by starvation or rapamycin treatment leads to association of autophagy markers with M. tuberculosis var. bovis bacilli Calmette Guérin (BCG) phagosomes and overcomes the block in phagosome-lysosome fusion characteristic of infection with BCG and M. tuberculosis [20]. However, sequestration of ubiquitinated Mm in LAMP-1 positive compartments did not require induction of autophagy and was independent of Atg5, suggesting an alternative pathway for reuptake of these organisms from the cytosol. These studies demonstrate that Mm undergo a complex kinetic interaction within the macrophages that are their preferred replication niche. After escape from phagosomes, some bacteria polymerize actin to facilitate cell-to-cell spread, while others become ubiquitinated, leading to reuptake into a host vesicular compartment by a mechanism distinguishable from classic autophagy. These alternative fates may contribute to the wide variety of effects of ESX-1 on host-pathogen interactions during mycobacterial infection [21]–[23].
We found that digitonin could permeabilize macrophage plasma membranes, while leaving phagosome membranes intact, as previously described [24] (Figure S1A). We combined macrophage permeabilization with digitonin with an antibody to Mm cell walls to detect cytosolic mycobacteria. A proportion of wildtype (WT) Mm in infected macrophages could be stained after digitonin permeabilization, whereas Mm lacking region of difference 1 (ΔRD1), which have no functioning ESX-1 secretion system, failed to be stained by antibody to the bacterial cell wall under these conditions (Figure 1A, top two rows). This was not due to failure of the antibody to recognize the mutant bacteria, since all intracellular ΔRD1 Mm were stained after macrophage permeabilization with Triton X-100, which solubilized phagosome as well as plasma membranes (Figure 1A, third row). Thus, lack of staining after digitonin permeabilization was indicative of the presence of ΔRD1 Mm in phagosomes. All WT Mm were also stained with antibody under the same conditions of Triton X-100 permeabilization (Figure 1A, bottom row). Using this technique, we found that ∼21% of WT Mm had escaped from phagosomes by 3.5 hours post infection (HPI), a time before significant bacterial actin polymerization occurred, as detected by phalloidin staining (Figure 1A and 1B). Bacterial escape from phagosomes also was demonstrated by electron microscopy, which showed about 43% of WT Mm in the cytosol at 3.5 HPI, while 100% of ΔRD1 Mm were clearly enclosed within a membrane (Figure 1C, Figure S1B, S1C). As previously described [12], actin polymerization by WT Mm was not detected at 3.5 HPI and increased dramatically by 24 HPI (Figure 1B and 1D). Thus, WT Mm escape from macrophage phagosomes several hours prior to initiation of actin-based motility.
Because of the long delay between phagosome escape and initiation of actin-based motility, we hypothesized that cytosolic Mm might be recognized by ubiquitinating enzymes, similar to other bacterial species that escape the phagosome [13]. Polyubiquitin was found associated with WT Mm, since bacteria were stained by an antibody recognizing only polyubiquitin (FK1), as well as an antibody that recognizes monoubiquitin-derivatized surfaces as well as polyubiquitin chains (FK2) (Figure 2A). Approximately 9% of WT Mm taken up by macrophages stained with FK2 at all times from 3.5 to 48 h (Figure 2A). Comparison of this percentage with the 21% escape observed at 3.5 HPI suggests that about 40% of all bacteria in the cytosol associate with polyubiquitin at this time.
Often, the cell biologic effects of polyubiquitin are determined by whether the ubiquitin chains are assembled through K48 or K63 linkages. While K48 polyubiquitin is most often a signal for proteasomal or lysosomal degradation, K63 linkages have been associated with assembly of signaling scaffolds [25]. By using antibodies specific for K48 and K63 poyubiquitin linkages [26], we found that both chain types associated with WT Mm. Though K48-linked polyubiquitin association with Mm was more variable than K63, the proportions were not statistically distinct (Figure 2B). Specificity of the linkages was demonstrated by competition with K48 or K63 tetraubiquitin chains (Figure S2A).
To determine whether the ubiquitinated Mm were cytosolic or intraphagosomal, infected macrophages were permeabilized with digitonin and stained with both anti-ubiquitin and antibody to the Mm cell wall, which under these conditions recognizes cytosolic but not intraphagosomal bacteria (Figure 1A). This procedure revealed that virtually all ubiquitinated WT Mm were cytosolic at 3.5 HPI, as WT Mm that bound FK2 also were stained by the Mm cell wall antibody (Figure S2B). Quantitative immunoelectron (EM) microscopy confirmed the cytosolic localization of ubiquitinated WT Mm at 3.5 HPI, as 96% of ubiquitin associated WT Mm were in the cytosol (Figure 2C, left image and graph). In addition to direct association with intact bacteria, ubiquitin was frequently found associated with adjacent dense membrane structures (Figure 2C, right image), which appeared as networks with meshes and broadened parts, containing electron-dense material and occasionally small vesicles (and are designated further in the text as dense membrane networks).
These results suggested that WT Mm ubiquitination occurred upon direct contact of the bacteria with macrophage cytosol, a hypothesis supported by the lack of ubiquitin association with ΔRD1 Mm, which never become cytosolic (Figure 2A, Figure 3A, data not shown). In contrast, the attenuated iipA mutant, which escapes phagosomes but grows poorly in macrophages [27], did become polyubiquitinated, suggesting that lack of ubiquitin association with ΔRD1 Mm did not reflect simple loss of virulence or decreased intracellular growth (Figure S3A).
To determine whether defects in ΔRD1 Mm besides failure to escape phagosomes affected its ubiquitination, HeLa cell cytosol was incubated with WT and ΔRD1 Mm, as we found HeLa cells also ubiquitinated Mm during infection (data not shown). Deposited ubiquitin was detected with FK2. Immunofluorescence showed ubiquitin localization in bright patches on the surface of WT Mm, with little background staining from an isotype control antibody (Figure 3B). Quantification of ubiquitin deposition was done by flow cytometry (Figure 3C, top panel), which showed that bacterial ubiquitination was abrogated in the absence of HeLa cytosol, after chelation of divalent cations to remove Mg-ATP, which is required for activity of the E1 ubiquitin activating enzyme, and at 4°C (Figure 3C, bottom). Binding of ubiquitin to WT Mm likely was covalent, as washing multiple times with 8 M urea to disrupt noncovalent interactions did not inhibit staining with FK2 (Figure S3B). In this in vitro assay, ΔRD1 was ubiquitinated equivalently or at greater levels than WT, suggesting that the lack of ubiquitin association with ΔRD1 Mm during infection in macrophages was due to lack of phagosome escape rather than absence of the substrate for ubiquitination.
To test this hypothesis directly, we coinfected macrophages with ΔRD1 and WT Mm at a ratio of 1∶1. Under conditions of coinfection, ΔRD1 Mm escaped from phagosomes, as shown by staining of ΔRD1-RFP with Mm cell wall antibody after permeabilization with digitonin alone (Figure 3D, top panel), presumably because of phagosome lysis by WT. Under conditions of coinfection, ΔRD1 also became polyubiquitinated (Figure 3D, bottom panel). These results demonstrate that both in vitro and in cells, ΔRD1 Mm can be ubiquitinated when exposed to host cytosol. Thus, the failure of ΔRD1 to be ubiquitinated in monoinfection (i.e., when used alone to infect macrophages) results from its failure to escape from phagosomes.
Despite the association of ubiquitin with cytosolic bacteria at 3.5 HPI, we did not observe any ubiquitin-associated WT Mm that formed actin tails at 24 HPI (Figure 4A). Instead, at this time the ubiquitinated bacteria appeared to be reinternalized into membrane bound compartments, because the frequency of detection of ubiquitinated bacteria stained after digitonin permeabilization decreased, even though the total number of ubiquitinated bacteria detected after Triton X-100 permeabilization did not change over time (Figure 4B). At 24 and 48 HPI, fewer than 20 to 25% of ubiquitinated WT Mm were detected after plasma membrane permeabilization with digitonin alone. In support of the hypothesis that lack of accessibility to FK2 represented resequestration into a membrane-bound compartment, ubiquitinated WT co-localized with the late endosomal/lysosomal marker LAMP-1 (Figure 5A). Some Mm appeared with LAMP-1 in discrete patches (top row of Figure 5A) while the entire contour of others appeared to be surrounded with LAMP-1 (bottom row of Figure 5A); these patterns suggested recruitment and fusion of lysosomes to surround ubiquitinated Mm. By 24 HPI, 64% of ubiquitinated bacteria were associated with LAMP-1, a frequency that did not change at later times (Figure 5B). At all time points, almost all bacteria associated with LAMP-1 were ubiquitinated (Figure 5C). In addition, in three independent experiments, at 3.5 and 24 HPI, respectively, 5.7%±6.15% (SD) and 0% of ΔRD1 Mm co-localized with LAMP-1 when macrophages were infected with these ESX-1–deficient bacteria alone, implying that WT Mm in LAMP-1 positive compartments were derived from those that had escaped into the macrophage cytosol. To further characterize the nature of the LAMP-1 positive compartments, we next studied these at the ultrastructural level, using immunoEM. LAMP-1 positive membranes surrounded FK2 positive WT Mm and FK2 positive material close to the Mm (Figure 5D). The FK2 positive material was surrounded by two membranes enclosing a narrow, electron-lucent space, as in the case of an autophagosome, but containing LAMP-1, which is unusual for autophagosomal membranes (Figure 5D).
The morphology of the LAMP-1 positive compartment containing ubiquitinated WT Mm was difficult to define by immunoEM, as it was compromised by the pre-embedding labeling procedure, including permeabilization, required for FK2 labeling (see Materials and Methods). Therefore, to examine in more detail the nature of the membrane compartments enclosing the Mm, electron microscopic analysis was performed on cells directly embedded (without permeabilization) in Epon resin at various times after macrophage infection. At 3.5 HPI, 43% WT Mm appeared to be in the cytosol and 53% in phagosomes. 2% appeared in double membrane vacuoles, consistent in morphology with autophagosomes (Figure 6A, top panel), and 2% were in lysosomes containing heterogeneous material including organelles and membranes (Figure 6A, bottom panel). By 24 HPI the number of WT Mm in vesicles with the morphologic characteristics of autophagosomes and lysosomes increased to 9% and 27%, respectively (Figure 6A and 6B). 100% of ΔRD1 appeared to be in phagosomes at 3.5 and 24 HPI (N = 83 and 19 respectively for each time point, data not shown).
Because the double membrane structures into which ubiquitinated WT Mm were sequestered resemble autophagosomes, and M. tuberculosis has been reported to localize to autophagosomes and then to autolysosomes after induction of autophagy [20], we tested whether LC3, an early marker for autophagy [14], co-localized with ubiquitinated WT Mm at various times after infection. While LC3 occasionally appeared closely apposed to both FK2 stained and unstained WT Mm in both control and starved cells, (Figure S4A), fewer than 3% of ubiquitinated WT Mm co-localized with LC3 at 4 and 24 hours, and overall less than 1% of WT associated with LC3 (Figure S4C). As positive controls for LC3 staining, anti-LC3 detected LC3 aggregates in macrophages after induction of autophagy by starvation or after rapamycin treatment, associated with a conversion of LC3-I to the membrane bound form LC3-II, (Figure S4B, S4F). Thus, there was insignificant association of LC3 with ubiquitinated WT Mm at any point examined during macrophage infection.
To determine whether induction of autophagy could enhance association of ubiquitinated WT Mm with LAMP-1, macrophages were starved or treated with rapamycin 2 hours before infection, and co-localization of ubiquitinated bacteria with LAMP-1 positive vesicles was determined 4 and 24 HPI. Though LC3-I was converted to LC3-II under these conditions (Figure S4F), neither starvation nor rapamycin increased the rate of uptake of ubiquitinated WT into a LAMP-1 positive compartment (Figure 6C). Preincubation of macrophages with IFN-γ, which induces autophagy in the RAW264.7 macrophage cell line (20), also did not affect Mm targeting to LAMP-1 positive vacuoles (Figure S4D). In the absence of additional stimulation to autophagy, there was no increased conversion of LC3-I to LC3-II in infected macrophages between 6 and 24 HPI, while resequestration of cytosolic WT Mm was occurring, over the minimal levels in uninfected controls (Figure S4E), suggesting that bacterial infection does not induce an autophagy response under these conditions. Thus, our experiments provide no evidence that the sequestration of WT Mm into LAMP-1 positive compartments depends on conversion of LC3 or its association with Mm-containing vesicles.
These results raised the possibility that an autophagy-independent pathway led to the sequestration of ubiquitinated WT Mm into LAMP-1 positive membrane-bound vesicles. To test definitively whether autophagy was necessary for the targeting of ubiquitin-associated WT Mm to LAMP-1 positive vacuoles, we assessed infection of Atg5−/− mouse embryo fibroblasts, since Atg5 is a required component of the pathway for autophagosome formation [28]. As previously shown, Atg5−/− MEFs were unable to convert LC3-I to LC3-II after starvation and were thus incapable of activating the autophagy pathway (Figure S4G). During infection of MEFs, WT Mm escaped from phagosomes and polymerized actin [29], were ubiquitinated (Figure 6D), and ubiquitinated WT Mm were sequestered into LAMP-1 positive compartments (Figure 6E), thus recapitulating the major events in Mm infection of macrophages. Moreover, ubiquitinated mycobacteria associated with LAMP-1 in Atg5+/+ and Atg5−/− MEFs with similar kinetics, increasing over time from about 10% at 5 hours to about 60% at 24 hours (Figure 6D and 6E). Thus, the incorporation of cytosolic ubiquitinated WT Mm into a LAMP-1 positive compartment occurred independently of the events of classical autophagy. In addition, lack of Atg5 had little effect on the survival and replication of Mm at 24 HPI, with infected Atg5+/+ MEFs having 1.03±0.32 fold the colony forming units as Atg5−/− MEFs (average±SEM of three independent experiments).
Our EM images suggested that, in addition to the ubiquitin attached directly to the mycobacterial cell wall, there was a large amount of ubiquitin associated with dense membrane networks in close proximity to the cytosolic bacteria (Figure 2C, right image). These typical membrane networks were only seen upon macrophage infection with WT Mm and not in uninfected cells or during infection by ΔRD1 Mm (data not shown). We found that these membranes were at least in part Mm-derived because when infected macrophages were permeabilized with digitonin and stained with anti-Mm cell wall antibodies, bacterial cell wall material was often localized in small dots or vesicles near cytosolic bacteria by immunofluorescence (Figure 7A). Moreover, when bacteria were fluoresceinated prior to macrophage infection, following methods developed previously [30], shed cell wall material could be seen at 3.5 HPI both by electron microscopy using an anti-fluorescein antibody and by direct fluorescence microscopy (Figure 7B and 7C). This shed cell wall material was often ubiquitinated, and appeared to be incorporated in the dense membrane networks adjacent to cytosolic bacteria (Figure 7B–7D). Host membranes also contributed to this “peri-bacterial” ubiquitinated material: when cholera toxin B was used to label the macrophage lipid raft marker GM1 for 8 minutes before infection, it co-localized with these ubiquitinated membranes at 3.5 hours after infection (Figure S5). In addition, in immunoEM preparations, the fluorescein- and ubiquitin-positive membranes had the appearance of a bilayer and not of an Mm cell wall (Figure 2C, Figure 7C). These host- and Mm- derived dense membrane networks may represent residual phagosome membranes, with which mycobacterial cell wall molecules remain associated after Mm escape. Alternatively, released hydrophobic Mm cell wall molecules in the cytosol may mix nonspecifically with host membranes. Thus, it appears that Mm shed ubiquitinated cell wall material in macrophages. Since we observed only ubiquitinated mycobacteria in LAMP-1 positive vacuoles, shedding of the cell wall may represent a strategy to evade this fate. This view is supported by the EM observation of the dense membrane networks enclosed in a double membrane lining (data not shown).
To test whether any alterations in Mm cell wall antigenicity might accompany the shedding of ubiquitinated cell wall antigens, Mm were stained with anti-cell wall antibody at various times after infection, using TX-100 permeabilization to allow the antibody access to all intracellular bacteria. 100% of both WT and ΔRD1 Mm stained with the antibody at 3.5 hours, and ΔRD1 continued to be recognized by the cell wall antibody throughout the 48 h experiment (Figure 8, quantified in bottom right panel). In contrast, at 24 and 48 HPI, only about 60% of WT Mm stained with the antibody. The changes to the cell wall did not appear to be a direct mechanism of ubiquitin evasion, since at 24 and 48 hours only 71% and 28% of ubiquitinated WT Mm stained with the antibody, indicating that Mm which exhibited changes to their cell wall could still be ubiquitinated (data not shown). The ubiquitination of Mm exhibiting cell wall changes hints at a continuous process, since Mm exhibit these changes only at later time points of infection and thus must be ubiquitinated after many hours of residence within the macrophage, possibly after shedding their initial wall. In contrast to ubiquitin association, actin polymerization and staining with the anti-cell wall antibody were negatively correlated, since 20% of WT Mm with actin tails at 24 and 48 HPI stained with the Mm antibody (Figure 8, quantified in bottom right panel). This result suggests that cell wall alterations during residence in the cytosol may be required for WT Mm to gain the ability to polymerize actin.
Escape from phagosomes underlies the mechanism by which Mm and distantly related bacteria such as Shigella and Listeria enhance cell-to-cell spread of infection [31]. For Mm, phagosome escape requires a specialized secretion system, ESX-1 [6]. Cytosolic bacteria polymerize actin to induce motility, the mechanism underlying the increased cell-to-cell spread. Cytosolic bacterial products also can activate innate immunity through sensors such as NODs [32]. In this study we describe an entirely different fate for a subset of cytosolic bacteria. These bacteria become ubiquitinated and then are resegregated from the cytosol into a LAMP-1–positive compartment, which forms independently of Atg5, an essential component of classical autophagosome formation. Our data indicate that Mm may have evolved strategies to avoid sequestration, apparently involving cell wall shedding to deplete the ubiquitin signal and alterations in cell wall antigenicity that may promote actin polymerization, as illustrated in the model in Figure 9.
The observation that cytosolic bacteria were either ubiquitinated or motile (with actin tails), but not both, suggests that these are alternative fates. Since both actin polymerization and ubiquitination are self-reinforcing, there may be competition between the two reactions at the bacterial surface, with either outcome negatively feeding back on the other. If this is the case, the outcomes of actin polymerization or ubiquitination may be determined by specific characteristics of the bacteria's surface during residence in the cytosol.
Previous studies had determined that Mm do not gain the ability to polymerize actin until many hours after initial infection and had inferred from this fact that escape from phagosomes was likewise slow [12]. In the current study, we have developed methods to examine phagosome escape independent of actin polymerization and show, both by differential solubilization techniques and by electron microscopy, that bacteria enter the cytosol much earlier than previously thought. By 3.5 HPI, 20–40% of internalized bacteria have escaped from phagosomes to the cytosol. Early escape allows access of the mycobacterial wall to ubiquitination machinery; because we observed that a large amount of ubiquitinated cell wall was lost from the bacteria, Mm may continuously shed ubiquitinated wall molecules into the cytosol, allowing time for the underlying deubiquitinated surface to express surface characteristics required for actin polymerization (Figure 9). The ubiquitination of Mm with significant cell wall alterations, as demonstrated by loss of reactivity with antibody to in vitro grown bacteria, hints at an ongoing process, since these Mm must be ubiquitinated after this antigenic shift, possibly after shedding their original wall.
Ubiquitinated Mm were sequestered in LAMP-1 positive vacuoles. We considered the possibility that the membranes recapturing cytosolic Mm were generated through autophagy, as this process has been identified as an important pathway in innate defense against several intracellular pathogens, including both Gram positive or (L. monocytogenes and S. pyogenes) and Gram negative (S. typhimurium and S. flexneri) bacteria as well as M. tuberculosis [17],[18],[16],[19],[20]. However, our evidence suggests that resegregation of Mm from the cytosol by the LAMP-1–positive macrophage membranes is by a process distinct from classical autophagy, since the majority of bacteria-associated host membranes do not express LC3, a marker of autophagosomal membranes, and their formation is independent of Atg5. This is quite distinct from the other bacterial pathogens known to be incorporated into cytosolic vesicles, which have clearly associated LC3-containing membranes dependent on Atg5 for their formation. These results also differ from studies highlighting the importance of autophagy in targeting M. tuberculosis var. bovis bacilli Calmette Guérin (BCG) to lysosomes, since autophagy induction enhances BCG co-localization with lysosomal markers LAMP-1, cathepsin D, LBPA and the VoH+-ATPase [20]. Autophagy induction also decreased survival of BCG and M. tuberculosis H37Rv, while we observed no difference in Mm's survival in wild type or Atg5 deficient MEFs [20]. It is possible that the differences we observe between Mm and BCG are due to the latter bacteria's propensity to remain in the primary phagosome, while Mm clearly has an intracellular phase when it is not enclosed in a membrane-bound vacuole. BCG appears not to activate autophagy as a direct consequence of infection, but to become entrapped in an autophagosome after induction by another signal or after IFN-γ activation of macrophages [20],[33]. In contrast, ubiquitinated Mm become resequestered in lysosome-like vacuoles without any additional extrinsic signal. Possibly, autophagy engulfs BCG because of a damaged phagosomal membrane, similar to the autophagic clearance of Salmonella typhimurium in response to perforation of the Salmonella containing vacuole (SCV) rather than full escape of the microbe [16]. However, ubiquitination and non-autophagic sequestration may occur later during infection with M. tuberculosis, since there is evidence that after two days it escapes to the cytosol of human monocyte-derived dendritic cells and macrophages through an ESX-1 dependent mechanism [10].
We speculate that ubiquitination is required for Mm uptake into LAMP-1 positive compartments, since virtually all bacteria associated with LAMP-1 are ubiquitinated, but this hypothesis remains unproven. Our evidence suggests that the segregation of Mm that have escaped into the cytosol occurs by a process of membrane formation initiated by a biochemical pathway distinct from classical, Atg5 dependent autophagy. One possibility is that the reincorporation of ubiquitinated Mm into a membranous compartment intersects with the ESCRT pathway, which delivers ubiquitinated cargo to multivesicular bodies and finally to lysosomes [34]. The double membranes could point to elongated or cup-shaped lysosomes as have been described in liver parenchymal cells [35]. In addition, the function of these compartments needs to be further established. Clearly, the LAMP-1 positive membranes segregate Mm away from the cytosol, but whether they also act as degradative compartments remains to be established.
Ubiquitination is a signal for uptake of aggregated proteins into autophagosomes, a process that requires the ubiquitin binding protein p62 [36]. At this time it is not clear why ubiquitinated Mm do not trigger this autophagy response. In this regard, it is interesting that both Listeria and Shigella actively inhibit entrapment by autophagy; Listeria uses actin polymerizing ActA as well as other proteins to evade autophagy, while Shigella secretes IcsB, resulting in disruption of the interaction between Atg5 and VirG, another actin polymerizing protein [17],[19]. Our results showing no effect of deficiency of Atg5 on Mm growth are similar to experiments with wild type Listeria [17]. The low level of autophagy of Mm we observed, as well as its similar ability to polymerize actin, suggests the possibility of autophagy inhibition by Mm.
Ubiquitination of Mm may also have other functions, such as targeting of the bacteria or ubiquitinated membranes to the proteasome. Ubiquitinated Salmonella seems to be degraded by the proteasome, while Listeria is not [13]. While we observed little association of the 20 s proteasome with ubiquitinated Mm at 24 and 48 HPI, attempts to study involvement of the proteasome were limited since inhibition of proteasome activity with MG-132 led to loss of ubiquitination of Mm, most likely through depletion of free ubiquitin (data not shown). Regardless of whether the proteasome can degrade whole bacteria, its degradation of ubiquitinated bacterial membranes may be important in presentation of bacterial antigens via the MHC Class I pathway, a prominent feature of Mycobacterial infection [37],[38].
An growing list of bacterial effectors have been shown to interfere with the host's ubiquitination system, acting as ubiquitin ligases to target host proteins for degradation, dampening signaling through inflammatory pathways such as NF-κB [39]. However, the reciprocal bacterial surface molecules recognized by the host ubiquitination machinery have yet to be identified. Our results expand the list of known bacteria that are targeted by ubiquitin, as both Gram-positive Listeria monocytogenes and the Gram-negative Salmonella typhimurium associate with polyubiquitin during infection as discussed above. The mycobacterial cell wall has many distinct features, including a thick mycolic acid layer which distinguishes it from those of conventional Gram-positive and Gram-negative bacteria [40]. In addition, our in vitro ubiquitination experiments constitute the first evidence that bacterial association with ubiquitin is a true covalent association with the bacterial surface. A prior study demonstrated that ubiquitin can come into contact with Mycobacterium tuberculosis and BCG in the lysosome, where the ubiquitin peptides are toxic to the bacteria [41], but in the present study several lines of evidence demonstrate that Mm ubiquitination occurs in the cytosol. Most significantly, at early times after infection the vast majority of ubiquitin-associated Mm are in the cytosol, as judged both by their exposure to antibodies after digitonin permeabilization and by electron microscopy. In addition, the ΔRD1 mutant is never ubiquitinated when infecting macrophages alone, but can be ubiquitinated when allowed to come in contact with cytosol by coinfection with WT Mm, or by mixing with host cell cytosol. This demonstrates that ESX-1 is not required for expression of the targets of ubiquitination, but is necessary for exposure of Mm to cytosolic ubiquitinating enzymes. Ubiquitination may be a response of the host to bacterial cell wall proteins or other molecules that it detects as abnormal in some way, similar to the response to unfolded or modified proteins. The large number of mycobacterial cell wall lipopeptides and proteins with high proline content or hydrophobic sequences [42] are likely candidates for recognition by the host ubiquitination machinery [43]. However, the presence of K63-linked chains on the bacteria, usually associated with signal transduction pathways rather than protein degradation, suggests that marking bacterial proteins for destruction is not the only consequence of Mm ubiquitination [25].
In summary, Mm exhibits two apparently distinct fates upon prolonged residence in the cytosol of infected macrophages (Figure 9). On one path, ubiquitinated bacteria are sequestered into LAMP-1 positive compartments; this may limit spread of infection and lead to direct toxicity to the enclosed bacteria; on the other path, bacteria shed cell wall material and possibly escape ubiquitination and sequestration. This may permit further growth as well as dramatic changes to the mycobacterial cell wall, actin polymerization and spread of infection. We speculate that both fates may influence the outcome of infection, since sequestration may enhance antigen presentation and activate or modulate pathways for cytokine production. At the same time, these data hint at a novel pathway for delivery of cytosolic material into lysosomes, independent of classical autophagy. It will be important and interesting to determine the mechanisms involved and whether these processes are shared by other pathogenic mycobacteria, as well as other microbes that share Mm's life style of phagosome escape and actin polymerization.
Antibodies to M. marinum (Mm) were generated by injecting rabbits with cell wall fractions from ΔRD1 Mm grown in Sauton's media. Phalloidin-Alexa 350 and phalloidin-Alexa 594 were from Invitrogen (Carlsbad, CA). Mouse monoclonal anti-conjugated ubiquitin antibodies FK1 and FK2 were from Biomol (Plymouth Meeting, PA). K48, K63, anti-her2, and anti-HIV gp120 antibodies were produced at Genentech (South San Francisco, CA). The ID4B rat monoclonal LAMP-1 antibody was from BD Biosciences (San Jose, CA) and the LC3 antibody was from Novus Biologicals (Littleton, CO). The fluorescein antibody was from Zymed Laboratories (South San Francisco, CA). The GAPDH antibody was from Millipore (Temecula, CA). Cholera toxin B-Alexa 594 was from Invitrogen (Carlsbad, CA). Secondary antibodies used for EM were rabbit anti-mouse IgG and anti-rat IgG (Dako, Glostrup, Denmark).
WT Mm (strain M) and ΔRD1 Mm [44] were cultured in Middlebrook 7H9 (BD, Franklin Lakes, NJ) supplemented with 0.2% glycerol/0.05% Tween 80/10% ADC Enrichment (Fisher Scientific, Pittsburgh, PA). Bone marrow derived macrophages were harvested from 129/SvJ mice, cultured as previously described, and used between day 7 and 21 of growth [45]. Wild type and Atg5−/− MEFs, a kind gift of Noboru Mizushima, were maintained in DMEM containing 10% FBS.
2×105 macrophages or 1×105 MEFs were plated overnight in DMEM containing 10%FBS, 20 mM Hepes, 10% CMG14-12 supernatant [46] or DMEM with 10% FBS, respectively. 1 hour before infection, cells were washed with PBS and placed in DMEM containing 3% FBS and 3% CMG for macrophages or 3% FBS for MEFs. Mm were washed twice in PBS and passaged through a 26-gauge needle three times to disrupt aggregates. Mm were added to macrophages at MOI of 4 to 12 for 2 hours at 32°C, and macrophage monolayers then were incubated with 200 µg/ml amikacin for 2 hours, following PBS wash, to kill extracellular organisms. After a final PBS wash, macrophages were incubated in DMEM containing 3% FBS & 3% CMG14-12 supernatant at 32°C in a 5% CO2 humidified incubator. Alternatively, MEFs were infected for 3 hours at an MOI of 40, washed and incubated with 200 µg/ml amikacin for 2 hours, washed to remove the drug and incubated for the indicated time. To count colony forming units, MEFs were infected for 5 hours, washed with PBS and lysed with 0.25% Triton X-100 in PBS after 24 hours.
Infected macrophages were washed 2× with KHM buffer (110 mM potassium acetate, 20 mM Hepes and 2 mM MgCl2), incubated with KHM containing 25 ug/mL digitonin for 1 minute at room temperature and then washed 1× with detergent-free KHM buffer. Macrophages were then incubated with antibody (anti-Mm and/or FK2) in KHM containing 2% BSA or with buffer alone at 32°C for 12 minutes, washed with PBS and fixed with 4% PFA. Macrophages that had been incubated with buffer alone were then permeabilized with 0.2% Triton X-100 for 4 minutes and then stained with the appropriate primary antibodies, or the appropriate secondary antibodies if they had already received the primary. Bound antibody was visualized with Alexa 350-conjugated goat anti- mouse F(ab')2 (Invitrogen, Carlsbad, CA) for FK2 or Alexa 594-conjugated goat anti- rabbit F(ab')2 (Invitrogen, Carlsbad, CA) by incubating in 2% BSA in PBS for 30 minutes. Controls omitting primary antibodies showed minimal fluorescence from the secondary antibodies alone, minimal crossreactivity of the anti-mouse F(ab')2 or anti- rabbit F(ab')2 with bound antibody of the other species, and insignificant bleedthrough of green fluorescence into the red channel and vice versa. Stained samples were viewed using an Axioplan 2 light microscope (Carl Zeiss MicroImaging) with a Semrock filter set and images were recorded with a CoolSNAPHQ CCD camera (Photometrics).
Macrophages were fixed with 2% PFA and 0.2% glutaraldehyde in 0.1 M phosphate buffer, pH 7.4 for 2 hours at room temperature. Alternatively, they were fixed for 12 minutes with 4% PFA, permeabilized with 0.1% saponin for 6 minutes, incubated with FK2 or isotype control anti-gp120 antibody for 45 minutes, washed with PBS and postfixed with 2% PFA and 0.2% GA for 2 hours. After rinsing with PBS, the blocks were embedded in 12% gelatin, cryoprotected with 2.3 M sucrose, and frozen in liquid nitrogen as previously described [47]. Ultrathin cryosections were cut at −120°C, picked up with 1% methylcellulose, 1.2 M sucrose, thawed and collected on copper grids. After washing with PBS containing 0.02 M glycine, sections were incubated with rabbit anti mouse IgG followed by protein-A gold for the detection of FK2 in permeabilized cells, or single- or double-labeled with primary (and secondary) antibodies followed by protein-A gold as described earlier [47]. The sections were then contrasted with a 1.8% methylcellulose, 0.6% uranyl acetate mixture. Quantification of Mm ubiquitination was performed on minimally 200 randomly screened WT Mm from each time point (3.5 HPI, 24 HPI), from at least 2 experiments.
Cells were fixed either with 2% PFA and 0.2% glutaraldehyde in 0.1 M phosphate buffer, or with half-strength Karnovsky fixative (2% paraformaldehyde, 2.5% glutaraldehyde, and 0.1 M sodium cacodylate buffer) pH 7.4 for 2 hours at room temperature. After rinsing, the cells were in situ postfixed with 1% OsO4 and 1.5% K3Fe(CN)6, in 0.07 M Na-cacodylate stained en bloc with 0.5% uranyl acetate, dehydrated in ethanol and embedded in Epon. Ultrathin sections were cut parallel to the culture flask bottom, and stained with uranyl acetate and lead citrate. Mm were quantified as present in a phagosome, and autophagosome, or an autolysosome, if they were enclosed by a single membrane lining an otherwise ‘empty’ vacuole, or were surrounded together with some cytosol for at least three quarters by a double membrane lining, or were contained together with irregular vesicles and membrane sheets by a single membrane, respectively.
Cell free ubiquitination reactions were performed by adaptation of previously described methods [48]. After preparation as described above for macrophage infection, approximately 109 Mm were added to 200 µg S-100 HeLa cell lysate, 5 mM MG-132, 4 µM ubiquitin aldehyde to inhibit deubiquitinases, 5 µL of energy regenerating system including phosphocreatine and phosphocreatine kinase and 600 µM ubiquitin (Boston Biochem, Cambridge, MA). After 2 hours shaking at 500 rpm at 36°C, Mm were fixed with 4% PFA in PBS, and then stained with FK2 or anti-gp120 isotype control, followed by APC-conjugated goat anti mouse IgG (Jackson ImmunoResearch, West Grove, PA). To test whether urea could disrupt ubiquitin association, WT Mm were washed twice before PFA fixation with PBS or 8 M urea, fixed and stained as described above. Flow cytometry was performed on stained bacteria using a FACSCaliber (BD Biosciences, San Jose, CA), collecting between 10,000 and 100,000 events per sample. Data analysis was performed using FlowJo Flow Cytometry Analysis Software (Tree Star, Ashland, Oregon). The relative mean fluorescence intensity (MFI) for each sample was determined by dividing the MFI of the FK2- stained sample by the MFI of a ubiquitinated WT Mm sample stained with the anti-gp120 isotype control.
The surface of Mm was labeled with 6-(fluorescein-5-(and-6)-carboxyamido)hexanoic acid, succinimidyl ester 5 (Invitrogen, Carlsbad, CA) as previously described [30].
Infected macrophages were stained with monoclonal antibodies recognizing K48-linked polyubiquitin or K63-linked polyubiquitin [26] after fixation and permeabilization with 0.1% saponin. In some samples, antibodies were incubated at room temperature for 1 h with K48 or K63 tetraubiquitin chains in 10-fold excess (Boston Biochem, Cambridge, MA) prior to addition to the infected macrophages, as previously described [26]. Four independent competitions were performed in parallel. Bound antibody was visualized with goat anti-human IgG-Alexa 594 (Invitrogen, Carlsbad, CA). Comparison of the percent of Mm stained with K48 and K63 was made using a chi-square test for equality of proportions.
Macrophages were washed with PBS and incubated for 2 hours at 32°C with either EBSS (Sigma Aldrich, St. Louis, MO) or macrophage growth media containing 3% FBS & 3% CMG14-12 supernatant with 50 µg/mL rapamycin (Calbiochem, San Diego, CA) added. As controls, macrophages were incubated in identically constituted growth media without rapamycin, or in full growth media containing 10% FBS & 10% CMG14-12. Cells were then infected and incubated as described for 4 and 24 hours. For immunofluorescence, cells were permabilized with 0.1% saponin for 6 minutes and stained with anti-LC3 followed by goat anti rabbit IgG- F(ab')2 (Invitrogen, Carlsbad, CA). Murine IFN-γ was from Sigma Aldrich (St. Louis, MO).
Detection of LC3-I and LC3-II was performed essentially as previously described [49]. Briefly, after cell lysis, protein separation by SDS-PAGE, and transfer, blots were incubated with anti-LC3 overnight at 4°C. Bound LC3 antibody was detected with goat anti-rabbit conjugated to horse radish peroxidase (Jackson ImmunoResearch, West Grove, PA). Blots were stripped and reprobed for GAPDH as a loading control. Chemiluminescence was quantitated with a ChemiDoc XRS from and quantified using Quantity One Analysis Software (Bio-Rad Laboratories, Hercules, CA) by dividing the adjusted intensities for each sample to the adjusted intensity for GAPDH.
|
10.1371/journal.pntd.0001922 | Differential Anti-Glycan Antibody Responses in Schistosoma mansoni-Infected Children and Adults Studied by Shotgun Glycan Microarray | Schistosomiasis (bilharzia) is a chronic and potentially deadly parasitic disease that affects millions of people in (sub)tropical areas. An important partial immunity to Schistosoma infections does develop in disease endemic areas, but this takes many years of exposure and maturation of the immune system. Therefore, children are far more susceptible to re-infection after treatment than older children and adults. This age-dependent immunity or susceptibility to re-infection has been shown to be associated with specific antibody and T cell responses. Many antibodies generated during Schistosoma infection are directed against the numerous glycans expressed by Schistosoma. The nature of glycan epitopes recognized by antibodies in natural schistosomiasis infection serum is largely unknown.
The binding of serum antibodies to glycans can be analyzed efficiently and quantitatively using glycan microarray approaches. Very small amounts of a large number of glycans are presented on a solid surface allowing binding properties of various glycan binding proteins to be tested. We have generated a so-called shotgun glycan microarray containing natural N-glycan and lipid-glycan fractions derived from 4 different life stages of S. mansoni and applied this array to the analysis of IgG and IgM antibodies in sera from children and adults living in an endemic area. This resulted in the identification of differential glycan recognition profiles characteristic for the two different age groups, possibly reflecting differences in age or differences in length of exposure or infection.
Using the shotgun glycan microarray approach to study antibody response profiles against schistosome-derived glycan elements, we have defined groups of infected individuals as well as glycan element clusters to which antibody responses are directed in S. mansoni infections. These findings are significant for further exploration of Schistosoma glycan antigens in relation to immunity.
| Schistosomes are parasitic worms that cause chronic and potentially deadly disease in millions of people in (sub)tropical areas. An important partial immunity to infection does develop but this takes many years of exposure and multiple infections. Therefore, children are far more susceptible to re-infection after treatment than adults. This immunological protection is associated with specific antibody and T cell responses. Many antibodies generated during Schistosoma infection are directed against carbohydrate chains (glycans) expressed by the parasite. The nature of the glycan epitopes recognized by antibodies in natural schistosomiasis infection serum is largely unknown. We have used a so-called shotgun glycan microarray approach to study differences in anti-glycan antibody responses between S. mansoni-infected children and adults. This resulted in the identification of differential glycan recognition profiles characteristic for the two different age groups that may reflect differences in age or differences in length of exposure or infection in people living in an endemic area.
| Schistosomiasis (bilharzia) is a chronic and potentially deadly parasitic disease, and a major public health burden in (sub)tropical areas. An estimated 207 million people are affected and 779 million people are at risk of being infected with schistosomes [1], [2]. Schistosomiasis is caused by members of the helminth genus Schistosoma (S.) with S. haematobium, S. mansoni, and S. japonicum being the most widespread. Schistosomes have a complex life-cycle with larval, adult worm, and egg stages interacting with the human host, each playing a role in immunology, immunopathology and maintenance of infection. Schistosoma infection is commonly treated with Praziquantel (PZQ) [3], [4]. Although PZQ has proven to be very effective, concern has been raised about development of drug resistance upon the currently ongoing mass treatments in endemic areas [5], [6] and the need for an alternative anti-schistosomal drug is regularly emphasized [7]. Furthermore, drug treatment does not prevent reinfection and repeated treatments are essential for people living in endemic areas, resulting in high costs and requirements to infrastructure. Therefore it is of great importance that a vaccine inducing protection against schistosomiasis is developed.
Multiple longitudinal studies have shown that infected individuals do acquire significant levels of immunity after prolonged exposure to Schistosoma. The acquisition of immunity is age-dependent in human populations living in schistosomiasis endemic areas with children being far more susceptible to re-infection than older children and adults [8]–[12] indicating that it takes many years of exposure, multiple infections and treatments, and maturation of the immune system to acquire this type of immunity. Several immunological parameters, including specific antibody and T cell responses, are predictive of the age-dependent immunity or susceptibility to re-infection after treatment [8], [13], [14]. Especially high levels of IgE against adult worm antigens [15]–[20], but also IgG1, IgG3 and IgA [8], [14] levels have been associated with increased resistance to infection after treatment. IgM, IgG2 and IgG4, on the other hand, are blocking antibodies with possible detrimental consequences for the expression of protective immunity [14], [21]. IgM can block eosinophil-dependent killing mediated by IgG antibodies from the same or other sera [22], [23]. IgM was found to be more highly expressed in children than in adults and is therefore higher in the non-immune group compared to the more resistant people [24], [25].
Antibody responses in schistosomiasis have been mainly studied using soluble worm antigen (SWA) and soluble egg antigen (SEA), each consisting of complex mixtures of antigenic (glyco-)proteins, or using specific recombinant protein antigens. Most antibodies generated during Schistosoma infection are however directed against parasite glycans [26]–[30]. This is not surprising considering the fact that glycans are abundant in schistosomal secretions, decorate the outer surface of all Schistosoma stages, and are highly immunogenic [31], [32]. Schistosoma life stages each express a different glycan repertoire [31], [33], [34]. Elaborate studies on the glycome of the different Schistosoma life stages have indicated that hundreds of different glycan structures are present within the N- and O-linked glycans and the glycolipids [31]. So far, serum antibodies to only a small set of schistosome-related glycans have been determined in a limited number of studies [25], [29], [30]. The large gap in our knowledge about the contribution of anti-glycan antibodies to immunity to schistosomes may be overcome using a shotgun glycan array approach which allows the detection of serum antibodies to a large number of parasite-derived glycans simultaneously. In this glycan array technology, natural glycans isolated directly from relevant cells or organisms are presented on a surface to quantitatively measure the binding to complementary molecules at the whole natural glycome level thus including unique and unusual (e.g. pathogen-specific) glycans [1], [35]–[40]. We have generated such a shotgun glycan microarray containing natural N-glycan and lipid-glycan fractions derived from 4 different life stages of S. mansoni (male adult worm, female adult worm, cercariae, and eggs), and applied this array to the analysis of IgG and IgM serum antibodies in a selection of sera from an S. mansoni natural infection cohort. This resulted in the identification of antigenic glycans as well as differential glycan recognition profiles characteristic for different age groups and shows that the shotgun schistosome glycan microarray approach has discriminative properties to define groups of infected individuals.
Ethical approval for the Piida study was obtained from the Uganda National Council for Science and Technology (UNCST) and cleared by the Office of the President. The study was also supported by the Cambridge Local Research Ethics Committee. Prior to enrolment, the study was explained to each selected adult or parent/guardian of each selected child for the study and verbal consent obtained. Verbal informed consent was sought because of the high level of illiteracy in Piida and because Lougungu, the predominant language, is not a written language. This method was approved by the ethical review committee of the UNCST. Verbal consent was documented by recording the name of each individual providing consent.
S. mansoni adult worms, cercariae and eggs were obtained as reported previously (Robijn et al, 2005). BSA- and NH2-linked synthetic oligosaccharide conjugates were synthesized as described [35], [41]–[44]. Cy3 conjugated goat anti-human IgG (Fc-specific), BSA and ethanolamine were from Sigma (Zwijndrecht, the Netherlands). Alexa fluor 647 conjugated goat anti-human IgM (μ chain specific) was from Invitrogen (Breda, The Netherlands).
Human sera were obtained from S. mansoni infected individuals living in the Piida community, Butiaba, which is situated on the shore of Lake Albert in Uganda where S. mansoni is endemic with 72% prevalence [24], [45], [46]. The detection of S. mansoni eggs in the feces was used as an indicator of infection with S. mansoni. The study design, epidemiology, and sample collection have been described in detail previously [24]. In the current study, anti-glycan antibody responses were determined among two separate age-groups, 21 children aged 5–11 years (mean age: 9) and 20 adults aged 20–46 years (mean age: 29), non-randomly selected from the original Piida study cohort based on intensity of infection and sex. All subjects had patent S. mansoni infection and intensity of infection did not differ significantly between the two groups [P = 0.51, geometric mean (GM) infection intensity (epg) was 478.33 (CI95%: 260.90, 868.37) among children and 665.80 (CI95%: 278.39, 1592.36) among adults]. The two groups were comparable with respect to sex, with roughly 3 females: 2 males in both age-groups. Anti-SEA-IgG4 and -IgE and anti-SWA-IgG4 responses were comparable in the two age groups (P>0.20); anti-SEA-IgG1 responses were significantly greater among the children (P<0.001), whilst anti-SWA-IgG1 and -IgE were significantly greater among the adults (P≤0.01).
S. mansoni male and female worms, cercariae and eggs were homogenized in water (4 ml per g wet weight) and sequentially methanol and chloroform were added (7 and 13 volumes, respectively). The upper phase contains the glycolipids and the pellet the (glyco)proteins. Glycans were released from the different preparations of S. mansoni glycolipids and glycoproteins by ceramidase and PNGase F treatment, respectively. Released glycans were subsequently purified, labeled with 2-aminobenzoic acid (2-AA), and fractionated by hydrophobic interaction liquid chromatography with fluorescence detection, as described previously [35], [47].
Glycan fractions, (synthetic) glycoconjugates, and proteins were dissolved in 20 µl of 1× spotting buffer (Nexterion Spot, Schott Nexterion) with 10% DMSO in 384-wells V-bottom plates (Genetix, New Milton, UK). A total number of 1143 samples (192 from male worms, 192 from female worms, 384 from cercarial lipid glycans, 192 from cercarial N-glycans, 102 from egg N-glycans, and 81 (synthetic) glycoconjugates) were printed in triplicate on epoxysilane-coated glass slides (Slide E, Schott, Nexterion) by contact printing using the Omnigrid 100 microarrayer (Genomic Solutions, Ann Arbor, MI) equipped with SMP3 pins with uptake channels that deposit 0.7 nl at each contact. Each array was printed three times on each glass slide. Dot spacing was 290 µm (X) and 245 µm (Y), and array spacing was 6000 µm. Printed slides were incubated overnight at room temperature at sufficient humidity to prevent drying of the spots and to allow covalent binding of printed 2-AA-labeled glycans and glycoconjugates to the epoxysilane via reaction with primary or secondary amines [35].
Microarray slides were covered with a hand-cut silicone gasket creating barriers to separate the three printed arrays and to hold wash and incubation solutions within the individual array areas. To remove unbound compounds, the arrays were rinsed with 1 ml PBS. Remaining active epoxysilane groups were blocked with 2% BSA, 50 mM ethanolamine in PBS for 60 minutes at room temperature while shaking. Subsequently, the slides were rinsed with PBS. Each microarray was incubated with serum (diluted 1∶100 in PBS-0.01% Tween20 with 1% BSA) for 60 min at room temperature while shaking. After washing the slides with successive rinses of PBS-0.05% Tween20 and PBS, the slides were incubated with Cy3-labeled anti-human IgG and Alexa Fluor 647-labeled anti-human IgM (diluted 1∶1,000 in PBS-0.01 Tween20) for 30 minutes at room temperature while shaking and protected from exposure to light. After a final rinse with PBS-0.05% Tween20, PBS and water the slides were dried and kept in the dark until scanning.
A G2565BA scanner (Agilent Technologies, Santa Clara, CA) was used to scan the slides for fluorescence at 10 µm resolution using 2 lasers (532 nm and 633 nm). At these wavelengths the 2-AA label does not fluoresce. Data and image analysis was performed with GenePix Pro 6.0 software (Molecular Devices, Sunnyvale, CA). Spots were aligned and re-sized using round features with no CPI threshold. Background-subtracted median intensities were averaged and processed as described by Oyelaran et al. [48] and median values of negative controls included on each array were subtracted. Datasets were log2 transformed to remove the basic trends of variance and plotted against the sample numbers. Hierarchical clustering analysis (HCA, complete linkage clustering using Euclidean distance) and Principal component analysis (PCA) were performed to define associated groups of sera and glycan fractions using MultiExperiment Viewer v4.5 and Simca-P+ 12.0 (Umetrics), respectively. For HCA, non-parametric testing was used for comparisons and a p value <0.01 was used to identify glycan fractions that were differentially recognized by serum antibodies [49].
Glycan samples of interest were analyzed by matrix-assisted laser desorption ionization time of flight mass spectrometry (MALDI-TOF-MS) with an Ultraflex II mass spectrometer (Bruker Daltonics, Bremen, Germany) in the negative ion reflectron mode using 2,5-dihydroxybenzoic acid (DHB, Bruker Daltonics) (20 mg/ml in 30% ACN) as matrix. Glycopeakfinder (http://www.glyco-peakfinder.org) was used to define glycan composition.
Using the shotgun glycan microarray, anti-glycan IgG and IgM responses in sera from S. mansoni infected individuals were determined. First, the IgG and IgM responses against a set of BSA-conjugated synthetic glycan structures that were included in the glycan microarray were compared between the two age groups (<12 years vs >20 years). Overall, the IgG response was higher in the group of children compared to adults with significant differences between the groups in response to Fuc(α1–3)GalNAc(β1–4)GlcNAc (F-LDN) and Fuc(α1–3)GalNAc(β1–4)[Fuc(α1–3)]GlcNAc (F-LDN-F) (Figure 1). Also the IgM response was higher in children and differed significantly from that in adults for Gal(β1–4)[Fuc(α1–3)]GlcNAc (Lewis X, LeX) and F-LDN (Figure 1). When comparing IgG and IgM responses, IgG responses against F-LDN-F and GalNAc(β1–4)GlcNAc (LDN) were significantly higher than IgM in both age groups, while responses against LeX were dominated by IgM (Figure 1).
With respect to the numerous printed glycans isolated directly from the Schistosoma life stages, Figure 2 shows that overall the IgG and IgM response patterns against the different glycan fractions are similar between the two age groups, but with a higher anti-glycan response intensity in the age group <12 years. Examining the responses against individual glycan fractions printed, statistical analysis using a Mann Whitney U rank order test (p<0.01) revealed a significant difference between the two age groups for 14.5% and 13.4% of all glycan fractions present on the array for IgG and IgM respectively with all responses being higher in children than in adults. For IgG, this group of differentially recognized glycans mainly consisted of cercarial glycolipid glycans (n = 54), cercarial (n = 32) and egg N-glycans (n = 33), while for IgM the differentiating fractions contained glycans isolated from cercariae (N-glycans followed by lipid glycans, n = 78 and 32 respectively). Since the number of glycan fractions printed on the array was not equal for all sources, the numbers of differentially recognized glycan fractions were plotted as percentages of the total number of glycan fractions from each source (Figure 3E). This showed that almost one third (32.4%) of the total number of egg-derived N glycans were differentially recognized by IgG when comparing the responses between children and adults, while for IgM this was highest for the cercarial N-glycans (40.6%). To explore which glycan structures were differentially recognized between children and adults, the top 10 of glycan fractions with the biggest difference in response were analyzed by MALDI-ToF-MS (Tables S1 (IgG) and S2 (IgM)). Most of these fractions contained mixtures of glycans, and of potential antigenic glycan elements. The glycan fractions that were differentially recognized by IgM and were higher in children than in adults contained glycans with short fucosylated and/or xylosylated (truncated) core structures and a few more complex structures which contain both core fucose and xylose and LeX elements in the antennae. For IgG, the proposed glycan structures are more complex and may contain other types of glycan elements such as LeX-LeX (di-LeX) and GlcNAc-LeX (extended LeX). HCA and PCA of the subset of differentially recognized glycan fractions between the two age groups showed three clusters for IgG (high (red), intermediate (white), and low response (blue)) (Figures 3A and 3C). For the group <12 years, 11 individuals (52.4%) clustered together in the high response cluster, 4 children clustered in the intermediate (19%) and 6 (28.6%) in the low response group (Figure 3D). In contrast, only 3 adults showed a high or intermediate IgG response while the majority shows a low (85%) IgG mediated response (Figure 3D). Three clusters were observed for IgM (high, intermediate and low response) (Figures 3B and 3C). Most of the children (81%) clustered in the high response cluster while the majority of adults clustered in the low (45%) and intermediate (45%) response clusters (Figure 3D). These data indicate that IgG and IgM responses can be different for a selection of individuals since some of the high IgM responders did not show a high IgG mediated response. All of the children clustering in the high IgG response cluster also showed a high IgM response.
The results described above show that, although the responses are significantly different between the two age groups, the individuals do not cluster precisely according to the age groups. Especially the intermediate response clusters contain individuals from both age groups, indicating that factors other than age are responsible for the differential anti-glycan IgG and IgM responses. To explore this possibility further, we performed a non-supervised HCA to define other possible individual and glycan clusters.
Non-supervised HCA and PCA of IgG responses showed two main clusters of individuals with difference in anti-glycan responses (Figure 4A and 4B). The high response cluster 1 contains 14 individuals (12 children and 2 adults) of which 13 were also found in the high response cluster in the supervised age comparisons. The responses for the 27 individuals (9 children and 18 adults) in the other cluster are much lower (Figures 4A, 4B and 4C). With respect to the IgM responses, HCA and PCA also identify a high (red) and a low (blue) response cluster (Figures 4A and 4B). The high response cluster contains 19 individuals (17 children and 2 adults) and 22 individuals fall into the low response group (6 children and 16 adults) (Figure 4D). Although the high response cluster mainly contained children and the low response cluster mainly adults, the non-supervised clustering was different from the supervised clustering on age-dependent differentially expressed glycan fractions indicating that factors other than age play a role in IgM response clustering (data not shown).
From this non-supervised IgG and IgM response analysis for the entire array, four groups of individuals can be defined: group 1 with high IgG and high IgM responses, group 2 with high IgG and low IgM responses (mixed), group 3 with low IgG and high IgM responses (mixed), and group 4 with low IgG and low IgM responses. Group 1 consists of 10 children and 1 adult, while group 4 contains 4 children and 15 adults (Table 1). Interestingly, group 2 (2 children, 1 adult) and group 3 (5 children, 3 adults) do not seem to be biased in terms of age and show intermediate egg counts after treatment.
The grouping of individuals in the non-supervised HCA and PCA described above was mainly due to glycan clusters C1 and C3 (Figures 4A and 4B) together forming the majority of the glycans present on the shotgun glycan microarray. However, for both IgG and IgM an additional smaller glycan cluster (Figure 4, glycan clusters C2 and C4) was observed for which the grouping of individuals is different. For IgG, glycan cluster C2 mainly consisted of egg (n = 37) and worm N-glycans (n = 35), while IgM glycan cluster C4 mainly contained glycans isolated from cercarial N-glycans (n = 86). When plotting these numbers as percentages of the total number of glycan fractions from each source it was shown that more than one third (36.3%) of the total number of egg-derived N glycans were present in glycan cluster C2 and 44.8% of cercarial N-glycans are present in glycan cluster C4 (Figure 4D).
HCA on glycan cluster C2 (Figure 5A) revealed that all individuals from the original high response cluster also belong to the high response group when exploring responses to glycans in cluster C2 only (C1highC2high). Interestingly, a group of 10 individuals that belonged to the original low response cluster clustered differently from the rest with lower IgG responses against the subset of glycans in glycan cluster C2 (C1lowC2low) and thus differ from the other 17 present in this group of individuals that show an overall low response but show a high response for this selection of glycans in glycan cluster C2 (C1lowC2high). When comparing additional information for these subgroups of individuals from the original low response cluster it became clear that there were no differences in age, but egg counts post treatment (epg5) were lower for those individuals with the lowest IgG responses for glycan cluster C2 (Figure 5B). Strikingly, nine out of ten in the low response cluster were females. The response against the subset of synthetic glycan structures showed that the IgG response in the C1lowC2low is lower than for the C1lowC2high group for all glycan structures tested, but significantly lower for F-LDN and F-LDN-F only (Figure 5C).
Also IgM glycan cluster C4 showed a different grouping of individuals than for the complete glycan microarray (Figure 5D). As for IgG, all individuals from the original high response cluster also belong to the high response group when exploring responses to glycans in cluster C4 only (C3highC4high). However, eight individuals from the original low response cluster show a higher IgM response (C3lowC4high) than the other 14 individuals for the glycans in cluster C4 (C3lowC4low). For this group of 8 individuals the egg counts at 9 months post treatment (epg5) were higher than for the C3lowC4low group but this was not statistically significant (Figure 5E). In contrast to the clusters of individuals defined by anti-glycan IgG, no differences were observed for the IgM response when comparing C3lowC4low and C3lowC4high clusters (Figure 5F).
In particular for IgM responses against the glycans that make up cluster C4 it is clearly visible that the sera fall into three separate groups (Figure 5D), whereas only two groups are observed for cluster C3. This provides an important indication that different subsets of glycans give rise to antibodies which are discriminative for different groups of individuals.
To achieve more insight into the human immune response against Schistosoma-derived glycans we analyzed sera of infected individuals for antibody reactivity using a shotgun glycan microarray approach. In this study we selected sera from infected individuals from a larger study in Piida [24] to give two distinct age groups to be compared. In the larger study, S. mansoni was found to be highly endemic with an overall prevalence of 72% and with a peak in infection prevalence and intensity in children aged 10–14 years [24], [45], [46]. The selected sera that were chosen allow the exploration of differences in anti-glycan antibody responses between children and adults. In highly schistosomiasis endemic areas like Piida, young children are immunologically, and perhaps physiologically, more susceptible to reinfection after treatment than adults [11], [50] and immunological parameters, including specific antibody and T cell responses, are predictive of the age-dependent immunity or susceptibility to re-infection after treatment [8], [13], [15], [23].
First, we explored the IgG and IgM response to a limited set of synthetic glycoconjugates (Figure 1) to which antibody response profiles have been analyzed previously. In accordance with literature, schistosomiasis induced IgM responses to LeX were higher compared to IgG responses [25], [29], [51]. For GalNAc(β1-4)[Fuc(α1-3)]GlcNAc (LDN-F) high IgM and moderately high IgG responses have been reported [29]. In our glycan microarray analysis this was not the case for children, but when looking at adults only, the relative response to LDN-F is indeed slightly higher for IgM than for IgG (Figure 1). A study on chimpanzees experimentally infected with S. mansoni showed that responses to F-LDN and F-LDN-F are similar, and dominated by IgG [51]. Also in our glycan microarray analysis of naturally infected humans, the anti F-LDN-F response is clearly dominated by IgG, but with the response against F-LDN being higher than against F-LDN-F. Specific for the current group comparison of children and adults, significant differences were observed for F-LDN (IgG), F-LDN-F (IgG) and LeX (IgM), with in each case responses being higher in children. One previous study also indicated higher IgG response against F-LDN in children compared to adults living in an endemic area [30]. Just like in the glycan microarray in this study the IgM response against LDN-F was alike in children and adults [30]. Another study showed that median values for the IgM response against LeX were higher in children, but only slightly [25],
While the analysis of antibody responses to the limited set of synthetic conjugates yields some useful insights, the complete glycan microarray includes glycan fractions isolated directly from the schistosome providing the possibility to study numerous additional glycans. Focusing on these glycans, the IgG and IgM responses were also higher in children than adults for most fractions, similar to the observations for the synthetic glycoconjugates. The stronger antibody response against glycans in children and increased susceptibility to reinfection in this age group suggests that there is an inverse correlation between anti-glycan antibody titers and immunity. This would be in line with the smoke screen theory which reasons that high antibody responses towards glycans are beneficial for the parasite rather than the host by subverting the immune system away from epitopes that could provoke protective immune responses [26], [27]. However, anti-glycan antibodies responses cannot be generalized as there are hundreds of different, defined glycan antigens of schistosomes and it could also be hypothesized that while many are subversive, other antibody isotypes or responses to specific subsets of glycan elements may be linked to protective immunity. For example, it has been shown that IgM and IgG2 antibodies that reacted with schistosomula and egg carbohydrate epitopes are negatively associated [14] while IgE directed against glycolipids has been suggested to be positively associated with resistance to reinfection [52]. The glycan clusters in the unsupervised HCA (Figure 4A) also provided an important indication that different subsets of glycans give rise to antibodies which can be discriminative for different groups of individuals and clearly suggested that not all glycans show a similar antibody response.
In the currently used shotgun array, the glycan fractions together contain many different glycan elements expressed by one or more schistosome life stages. While some glycan fractions contain only a single glycan antigen, most fractions are formed of mixtures of glycans that were not separated by the chromatographic procedure used, or they contain glycans which display more than one antigenic glycan element, e.g. in different branches of a di-antennary N-glycan. Therefore, it would be too early to speculate which specific responses to each glycan element occur in the different groups of the cohort. To this end the fractions first need further sub-fractionation and structural analyses to improve separation and definition of the antigenic glycan elements present. What can already be learned from the stage-specific glycan fractions as a group is that for IgG most differentially recognized fractions were derived from cercarial lipid glycans, while IgM responses were clearly most dominant against cercarial N-glycans. Both cercarial lipid and N-glycan fractions contain glycans with LeX elements, however pseudo-Lewis Y elements are unique for cercarial lipid glycans while core β2-xylose occurs in cercarial N-glycans [31], [53], [54] possibly giving rise to differences in dominant responses observed for IgG and IgM. When analyzing the glycan structures in the top 10 of glycan fractions that were differentially recognized between children and adults differences were indeed observed for IgM and IgG. Differential IgM responses between children and adults seem to be mainly against fractions with short fucosylated and/or xylosylated (truncated) core structures and mono LeX elements while differential IgG responses were against more complex structures containing LeX-LeX (dimeric LeX) and LeX-(F-)GlcNAc elements.
The differences in anti-glycan antibody responses between children and adults for this selection of sera may reflect differences in age or differences in length of exposure or infection in an endemic area. The non-supervised HCA showed that the individuals did not cluster precisely according to age suggesting that other factors play a role in anti-glycan antibody response profiles. Also within one cluster of individuals the anti-glycan antibody responses varied for different glycan clusters as was shown for glycan clusters C2 and C4 (Figures 4A and 4B). One single glycan antigen is clearly not representative for the whole group and this stresses the need for screening antibody responses against multiple glycans and glycan elements. Shotgun glycan microarrays are valuable tools in this type of screening allowing the definition of groups of individuals as well as glycan element clusters to which similar antibody responses are generated in individuals. Having shown that the shotgun Schistosoma glycan microarray has discriminative power for studying differences in anti-glycan immune responses in different groups of individuals, this technique can now be applied to a randomly-selected epidemiological cohort to address whether anti-glycan antibody responses reflect differences in age, infection intensity or other factors that have not been explored yet.
|
10.1371/journal.pcbi.1004502 | The Encoding of Decision Difficulty and Movement Time in the Primate Premotor Cortex | Estimating the difficulty of a decision is a fundamental process to elaborate complex and adaptive behaviour. In this paper, we show that the movement time of behaving monkeys performing a decision-making task is correlated with decision difficulty and that the activity of a population of neurons in ventral Premotor cortex correlates with the movement time. Moreover, we found another population of neurons that encodes the discriminability of the stimulus, thereby supplying another source of information about the difficulty of the decision. The activity of neurons encoding the difficulty can be produced by very different computations. Therefore, we show that decision difficulty can be encoded through three different mechanisms: 1. Switch time coding, 2. rate coding and 3. binary coding. This rich representation reflects the basis of different functional aspects of difficulty in the making of a decision and the possible role of difficulty estimation in complex decision scenarios.
| Understanding how the brain produces complex cognitive functions has been a crucial question since ancient philosophical inquiries. The encoding of decision difficulty in the brain is fundamental for complex and adaptive behaviour, and can provide valuable information in uncertain environments where the future outcome of a choice must be evaluated beforehand. Here we show that neurons in premotor cortex represent the difficulty of a decision using at least three different variables: 1) the time of the neuronal response, 2) the intensity of the neuronal response, 3) the probability of switching from a low activity to a high activity profile. Moreover, we show that, by encoding the time elapsed from the end of the stimulus and commitment to a choice, another set of premotor neurons is able to provide information about the difficulty of the decision. These results show that the brain is implementing heterogeneous neural mechanisms to fulfill a complex cognitive function.
| The information about the difficulty of a decision can be very valuable to properly allocate cognitive resources or to develop complex plans. Indeed, not only humans but even very simple form of life like honey bees are able to selectively avoid difficult decisions [1]. Moreover the degree of difficulty in a decision can also serve as a building block for the construction of confidence and to predict the outcome of a course of action.
Despite the relevance of the representation of difficulty very few is known about how the brain encodes and manipulate this information. It was shown that the onset and steepness of ramping activity in LIP depends on the amount of evidence for the decision [2], hence on the difficulty. This result was confirmed and extended by Ponce-Alvarez and colleagues [3], who found that in neural ensembles showing abrupt changes of activity both the time of the change and its variability depend on the difficulty of the decision. Pardo-Vázquez and colleagues [4] showed evidence of the effect of difficulty in PMv neurons, where a population of decision selective neurons is reported to have higher firing rate (FR) in easy compared to difficult trials for preferred correct choices and lower FR for non preferred correct choices. Also neurons in rats orbitofrontal cortex were found to modulate stimulus difficulty [5]. This article is indeed part of a growing body of literature [5–9] suggesting that single neurons in different brain areas (orbitofrontal cortex, lateral intraparietal sulcus, pulvinar) are involved in decision confidence processing. Although doubts can be cast on the metacognitive nature of the tasks employed [1, 10–14], these studies present evidence that the activity of neurons at least correlates with the difficulty of the decision.
However all these results still leave open the question as to how the neural signals of difficulty can be encoded in single trials. As a working hypothesis we expect that, during the decision process, the difficulty can be encoded in a continuous way by some feature of the decision process itself or by another monitoring process. Nonetheless we often take decisions based on the perceived difficulty of another decision (e.g. if it is too hard to tell if somebody is bluffing at poker a player could decide to leave the trick). Therefore we also expect that the difficulty signal can be used by another decision process and in this case a discrete representation may arise. According to this general hypothesis both continuous and discrete representations could be implemented in the brain. In consequence, it remains unclear what kind of encoding is used in the brain to represent difficulty.
Therefore in the present study our aim is to shed light on the possible mechanisms used by the primate brain to represent decision difficulty. We have looked at how difficulty can be represented in single neurons recorded from ventral premotor cortex (PMv) while monkeys perform a visual discrimination task [4].
The correlation of reaction time (RT) and difficulty has been shown by many experimental [15–17] and explained by theoretical studies [18, 19]. In our experiment, given the experimental protocol, we can only record the movement time (MT) and not the RT. We hypothesize that MTs are related to the difficulty and tested this hypothesis. Indeed it was already shown [4] that MT is different in easy compared to difficult choices. Here we found that the MT correlates with the difficulty of the decision task. Therefore neurons encoding the MT could also bring information about the difficulty and the decision process itself. Indeed we found a neural correlate of MT in PMv. Moreover we report another population of neurons whose activity correlates with the discriminability of the stimulus and we investigate the computational schemes underlying this correlation. Our results suggest that both continuous and discrete coding schemes could be active in the brain.
We studied the decision-process in the primate brain during a simple binary decision task. Two male monkeys (Macaca mulatta) performed a two-interval two-alternative discrimination task. They were trained to compare the orientation of a reference bar (with variable orientation), presented during the first interval, with that of a test bar, presented during the second interval. Their task was to decide whether the test bar was tilted right or left as compared to the reference bar (see Fig 1A and Methods for details). The level of difficulty of the task was controlled by varying the difference between the orientation of the first and the second bar, i.e. the test bar’s relative orientation (TRO). The TRO was varied from one up to four degrees and in both directions. Consequently the choice of the subjects was affected and achieved almost perfect performance for TRO = 4° and TRO = −4°, as shown in Fig 1B.
Single cells from PMv were recorded while monkeys performed the task. For a more detailed description of the task, behavioral results and neural recordings see Methods and [4].
Our first objective was to find neural signatures of difficulty computations in PMv of the primate brain. It is plausible that these computations take place in the same area as where the decision is encoded. In addition, it has been shown evidence of different neural dynamics in a decision-related PMv population in easy compared to difficult choices [4]. We therefore analyzed the activity of PMv activity recorded during the decision task. Our analysis was restricted to a subset of the recorded neurons (324 neurons, see [4]), comprising the cells that were relevant to the decision task.
The correlation between difficulty and RT has been well established, both by experimental [15–17] and theoretical [18, 19] works. In our experiment we can only record the movement time (MT) and not the RT. We hypothesize that MTs are related to the difficulty and tested this hypothesis. The MT is the time from the end of the second interval (subjects were not allowed to chose one option before the end of the second interval) until the response of the subject. The MT must not be confused with the decision time, indeed the decision could be taken before the end of second interval (even though MT and decision time could be correlated). However, there is a big variability in MTs and they could therefore be informative about the difficulty of the decision. Here we want to test the hypothesis that MT are correlated with the difficulty of the decision.
As shown in Fig 1C, MTs decrease as a function of the ease of the trial in correct trials and increase in error trials (Pearson correlation coefficients: TRO< 0, correct: 0.11; TRO> 0, correct: -0.18; TRO< 0, error: -0.13; TRO> 0, error: 0.14; all p-values< 10−7). This X-shaped pattern has been previously associated to decision confidence [5, 7, 20–22]. However we note that this pattern is neither necessary [14, 23, 24] nor sufficient to define confidence, since it could also emerge from different computations. Therefore and since we don’t have a direct measure of confidence in our task we will state that the MT, like the RT or decision time, is informative about the difficulty, leaving open the question whether it could represent a signature of metacognitive processing.
We used a linear regression model (LMmt, see Methods) to test whether neurons in PMv encode the MT. In Fig 2 the linear analysis of a PMv single neuron is shown whose FRs encodes MT. Line in Fig 2A shows the slope of the regression line for a single neuron.
The significance of the slope of the regression line was assessed by Monte-Carlo random resampling (p < 0.05). The shaded area in Fig 2A represents the 95% of the distribution of the slope under the null hypothesis. Values of the slope falling in the range indicated by the shaded area considered as not significant since they are indistinguishable from values obtained by chance. Moreover a minimum number of consecutive significant bins is required in order to avoid false positives due to fluctuations. The minimum number of consecutive significant bins was chosen based on the distribution of consecutive significant bins under the null hypothesis. This probability distribution was estimated by Monte-Carlo random resampling. Fig 2B shows the distribution under the null hypothesis for the same single neuron of panel A (shown in ms for an easy comparison with panel A). We considered a neuron to have a significant slope of the regression line only if its longest interval of consecutive bins had a low probability under the null hypothesis. We believe that this strict requirement is necessary for this type of data, since the time correlation of FRs can affect the significance of results. Since we applied this test to the whole population we controlled the false discovery rate (FDR) with Benjamini-Hochberg procedure [25]. We show in S1 Fig the results of controlling FDR at varying values of Q. We observe that in a wide range of Q a large amount of true discoveries is made. For Q = 0.05, for example, we found 276 neurons (82% of the analysed population), such that the slope of the LMmt is significantly different from that given by chance. We note that this is a lower estimate of the number of neurons encoding MT in the dataset (as explained in S1 Text). We can choose such a small value in order to limit the number of false discoveries, however such a low value for Q can induce a large underestimate of the number of false null-hypotheses (as shown in S1 Text).
Even if we found many neurons with a significant slope, it is still possible that the LMmt is explaining a small portion of the variance of the data from those neurons. In order to rule out this possibility we analysed the coefficient of determination (R2) of the LMmt. Fig 2D shows the R2 and the 95% of its distribution under the null hypothesis for the same single neuron of panels B and C. We applied to R2 the same procedure used for the slope of the regression line (see Methods) and took a minimum number of consecutive significant bins (the distribution of consecutive significant bins under the null-hypothesis is shown in Fig 2D). We found that, for 248 neurons out of 276, the value of the R2 is significant, even though the values of the R2 for the 248 neurons are quite small (99% of the distribution is between 0 and 0.23). To our knowledge this method has not been used until now in the analysis of neurophysiological recordings and could set a new standard for analysing single neurons (see Methods for details). In summary we found 248 neurons in PMv, whose activity is informative about the MT.
We want to test here the idea that PMv neurons directly represent the discriminability of the stimulus and hence the difficulty of the decision. In the following single neurons analyses we used only correct trials, since error trials were not enough to produce clear results. We found a population of neurons whose FR was informative about the difficulty of the task for at least one of the two possible choices (as revealed by the linear regression model LMtro; same method as for the MT analysis above was applied, see Methods). Each of these neurons was able to encode the difficulty of the task only for one choice when isolated but their population signal can be integrated by downstream neurons.
Hence we identified among these difficulty neurons, a population whose activity was similar for right and left decisions. The FR of these neurons encoded the difficulty of the task, independently of the subject’s choice, as revealed by the linear model (LMdiff) (see Methods for details). Fig 3A shows the evolution in time of the slope of the regression line for a single neuron of this population. The shaded area represents the 95% of the distribution of the slope under the null-hypothesis (Monte-Carlo random resampling). The values of the slope of the regression line overlapping with the distribution of randomly resampled data are not significant, since they are indistinguishable from values given by chance. Values outside shaded area are considered significant. As shown in Fig 3B, in the time window where the slope is significant the FR of the neuron increases for both positive and negative values of TRO, producing a V-shaped pattern. Using this method we found 107 neurons for which the slope of the regression line was significant (Q = 0.1, Benjamini-Hochberg procedure, see Methods and S1 Text).
Fig 3C shows the R2 values and the 95% of its distribution under the null-hypothesis (Monte-Carlo random resampling). This result further constrains the encoding time window to regions where the R2 is significant. When we applied the R2 analysis only 66 neurons revealed to carry significantly more information than chance (Q = 0.1). This strong reduction in the number of informative neurons suggests that simple tests based on the slope of the regression line can be made more robust by applying also tests based on R2 to filter out elements that bring few linear information.
In summary we found 101 neurons encoding the difficulty of the task, or the TRO, for at least one of the two behavioural responses and 66 neurons encoding the difficulty for both behavioural responses (Q = 0.1).
In Fig 3D we show the population activity of these 66 neurons. When the behavioral response was incorrect (gray line) the FR of the population showed an inverse pattern compared to correct trials (black line). Overall, the normalized FR separated for correct and error trials formed an X-shaped pattern This neural pattern has already been described [5, 7] as a signature of decision confidence. However we are interpreting it here as the neural correlate of difficulty given the lack of a direct measure of confidence in the current task.
The V-shaped pattern shown above can arise from very different computational schemes. In the following, we will try to shed light on this matter. Again only correct trials will be analysed, although the proposed method could easily be applied to error trials. The increasing FR as a function of the absolute value of TRO, i.e., the V-shaped pattern, can arise from at least three distinct mechanisms (for a pictorial representation see Fig 4). 1) Switch time coding (panel A): Neurons increase the FR, switching from a low to a high activity state, with a different timing according to the discriminability, and with the average rate reflecting this timing. 2) Rate coding (panel B): Neurons increase the FR relative to the baseline in proportion to the discriminability. 3) Binary coding (panel C): Neurons have a binary response, i.e., they increase the FR with a probability that depends on the discriminability (e.g. in easy trials the activity is mainly high whilst in difficult trials the neuron mostly remains in a “down” state). In this last scenario mixing trials of high and low activity produces the V-shaped pattern of average FRs (a similar mechanism has been proposed for confidence encoding [21]).
In order to identify neurons implementing each of these mechanisms we used different statistical techniques. Although we present them here as separated mechanisms, we do not rule out the possibility that they could all appear at the same time.
We first verified whether the switch timing had any relevant effect in our data. To do so we used a Hidden Markov Model (HMM) (for its application with single neuron recordings see [26]) which is able to detect when a system switches from one state of activity to another (see Methods for details). In Fig 5 the analysis of a PMv single neuron is shown whose FR encodes difficulty with a switch time code. In Fig 5B we show a summary of the two-state HMM analysis for a single neuron (each row represents a trial). The color of the row changes from white to black when the neuron goes from a low to a high-activity state.
The separation of the two states is clearly visible also in the raster plot (Fig 5A, trials sorted according to the switch time estimated by HMM) and in the time averaged FR of single trials (Fig 5C). This neuron exhibits a lot of variability in the switch timing, changing state from just a few milliseconds up to 300 ms after stimulus onset. The timing of the change was correlated with the difficulty of the trial (Kendall correlation coefficient τ = 0.18, p < 0.05). Fig 5D represents the mean switch time as a function of TRO. Overall eight neurons showed a significant correlation (Q = 0.05) between the switch timing and the difficulty. Fig 5E represents the population average switch time as a function of TRO.
Once we had determined when a neuron changes its state we were then able to assess the relevance of the rate coding mechanism. In Fig 6 we show a PMv single neuron whose FR encodes difficulty with a rate code. To estimate whether the increase in FRs was proportional to the difficulty (i.e., the rate coding mechanism of Fig 4B), we first calculated the average FR from when 90% of the trials switched states (red vertical line in Fig 6B), until the coefficient of LMdiff had a significant value (according to the analysis explained above). Then we effectuated a correlation analysis between the level of difficulty and the average FR.
We obtained a significant correlation coefficient for the neuron in Fig 6 (τ = 0.24, Kendall correlation, p < 0.05), which suggests that it could be the FR of the neuron in the “up” state that encodes the trial’s level of difficulty. The FR of the neuron as a function of TRO is shown in Fig 6D. Overall eighteen neurons had a significant correlation (Q = 0.05) between the FR and the difficulty (Fig 6E shows the population average FR as a function of TRO).
To summarize, we found that eight neurons presented a significant impact on the timing in the formation of the pattern, while eighteen neurons increased the FR proportionally to the difficulty of the trial, thereby implementing the rate coding mechanism. There were also twelve neurons that presented both switch timing code and rate code (see Fig 8 for a graphical representation of all classes of neurons). We note that we could apply this method based on the HMM only to 56 neurons of the population (101 neurons) encoding the difficulty, as we considered the HMM analysis was only reliable under certain constraints (see Methods).
Both classes of neurons (that implementing a switch timing mechanisms and that implementing a rate coding mechanism) can be interpreted as continuously encoding the discriminability.
Conversely, the binary mechanism postulated above corresponds to a discrete encoding. Although we may expect a continuous representation, a discretization stage would be needed in order to take subsequent decisions based on the difficulty of a previous one.
An example of single neuron implementing the discrete mechanism (identified by the analysis detailed below) is shown in Fig 7.
Fig 7A shows the raster plot, where trials were ordered according to the time of state switch estimated by HMM analysis. We reasoned that the neurons showing a binary behavior should also lead to a characteristic pattern showing up in the HMM analysis: they should present a state switch only on a subset of trials. And indeed this pattern can be seen in Fig 7B (the pattern is only barely observable in the raster plot when trials are ordered according to HMM analysis, Fig 7A). The FR of the neuron after the state switch, as in Fig 7C, also shows a clear separation between the two states identified by HMM analysis. Comparing this figure with Figs 5C and 6C we can clearly see that this neuron switched state only in a subset of trials. The average FR of the neuron is shown in Fig 7D. The FR of the two states does not encode the difficulty of the task and the increasing activity as a function of the absolute value of TRO is due to the increasing proportion of trials in the high activity state as postulated by the mechanism represented in Fig 4C. The proportion of trials in the “high” state is shown for this neuron in Fig 7E.
In order to identify neurons with a discrete response we hypothesized that the distribution over trials of the mean FR as calculated during the test-bar presentation, has to consist of two different distributions. Note that the resulting distribution is not necessarily bimodal but it should differ substantially from the expected Poisson distribution [27–29]. For each trial and each neuron, therefore, we took the average FR in the time-window where neuron was found to be encoding the difficulty (according to the linear model analysis explained above and in the Methods section). S8 Fig shows the encoding window for each neuron. Then we fitted these mean FRs to a Gaussian mixture with two components. The resulting model is the following:
GM = 0 . 5 1 σ 1 2 π e - ( x - μ 1 ) 2 2 σ 1 2 + 0 . 5 1 σ 2 2 π e - ( x - μ 2 ) 2 2 σ 2 2 ,
where and μ1, 2 are the means of the each component and σ1, 2 the standard deviations, hence the model has four free parameters. We used an expectation maximization algorithm to obtain a maximum likelihood estimator of the parameters of the model (see Methods for details).
In order to rule out the possibility that a single Gaussian distribution model could fit the data better than the mixture model we used the Bayesian Information Criterion (BIC) that, while comparing the likelihood function of the two models, corrects the result by penalizing for the number of free parameters. Therefore, even if the likelihood of the single distribution model were equal to that of the mixture model, the BIC would always prefer the simpler model (or, conversely, a mixture model would be preferable only if it was able to explain much more than the single distribution model).
In conclusion, we consider a neuron to have a binary response only if the BIC was giving preference to the mixture model. In addition, to avoid cases where a small difference in the BIC score could favor the mixture model we discarded all those differences that were non significant under the null-hypothesis (H0: single Gaussian distribution, FDR controlled with Q = 0.05; see Methods for details). We found that sixty neurons displayed a binary mechanism in the case of at least one behavioral response (e.g., “left”) and seventeen of those for both behavioral responses (see Fig 8 for a graphical representation of all classes of neurons). In these neurons the V-shaped pattern of the FR can arise because the proportion of trials with high FR correlates with the difficulty of the trial.
In Fig 8 we show the different classes of neurons (neurons encoding MT, neurons encoding the difficulty, neurons using a rate code, switch timing code and binary code). The label and number in each rectangle indicate respectively the class and the number of neurons we found in that set. We also report in gray the number of neurons in the partial intersections, e.g. five neurons are in the intersection between the rate and switch time populations and three neurons use all three mechanisms (rate, time and binary).
In this study we tackled the following question: What are the mechanisms in the primate brain that encode the difficulty of a decision in single trials?
We have shown that the MT correlates with the discriminability of the stimulus and hence it could be used as a cue to infer the difficulty of a trial. The correlation of RT and difficulty has been proven by many experimental studies [15–17]. Theoretical studies [18, 19] suggest that this correlation is attributable to the decision time (more than movement time or sensory processing time). As in our task the subject is not allowed to respond until the end of the second stimulus, we were only able to record the MT. The decision process of the subject could in principle extend over the duration of the second stimulus, therefore, the MT could embrace part of the decision time and part of the motor preparation. In addition we cannot exclude a correlation between the time for motor execution and the difficulty. Our result is particularly important since it opens new questions about the distinct functions of decision time and MT in the formation of the RT. Moreover it shows that the common choice to model non-decision time as a fixed quantity [17, 19, 30, 31] could be not appropriate depending on the purposes of the model. In addition we found that neurons activity in PMv is informative about the MT. Moreover we have demonstrated that the FR of neurons in primate PMv encode stimulus discriminability.
The variability of neural responses could be explained by different computations performed by neurons in single trials that, once averaged, could produce the same pattern. We suggested three hypothesis for these computational mechanisms: 1) The switch time coding: when the activity of the neuron changes, the difficulty of the decision, is encoded in the timing of the change, 2) The rate coding: the difficulty is encoded in the FR, after the change has taken place; or 3) The binary coding: the neuron only switches between a high and a low activity state and the proportion of high activity trials depends on the discriminability of the stimulus. The first two alternatives correspond to a continuous encoding of difficulty, whereas the last one is a form of discrete encoding. We found, in fact, evidence for all three mechanisms in monkey PMv neurons. For certain neurons the timing and FR mechanisms work together, i.e., a neuron that changed state earlier on less difficult trials will also have a higher FR after the change. Other neurons present a binary response (increasing activity only in some trials), which suggests a possible role in more complex decision scenarios where decisions must be taken based on the difficulty of previous ones.
An important question is: why should neurons use only one mechanism to encode difficulty? Our hypothesis is that difficulty neurons carry-out more than one function in the sensory-motor path. It seems natural that difficulty may be encoded on a continuous scale, since we usually think about difficulty as a graded quantity. However, if difficulty is to have behavioral relevance, then, depending on the requested output, the information about difficulty may need to be discretized. Our hypothesis is that, while certain neurons use a continuous representation, other neurons read-out this scale and transform it into a discrete quantity in order to produce consistent behavior.
This hypothesis highlights the fact that all three proposed encoding mechanisms are not only evidenced by our decoding procedures but they stand as natural representations of difficulty that can easily be read out by higher processing brain areas. Indeed if a read out neuron could have access to the distribution of FR or to that of switch timing, this neuron could measure the difficulty of the decision. Binary neuron could implement for example a type of read out process where the difficulty information encoded by FR of switch timing statistics is used to give a binary classification of difficulty.
Most of our results depend on a linear model of the FR. But does this relation have to be linear (and not, for example, logarithmic or sigmoidal)? Firstly we note that linear functions have been extensively used to model the relation between the FRs of neurons and certain task features (e.g. [4, 32, 33]). Yet it is possible for the relation not to be linear. Indeed, we consider the linear function as a first probable approximation.
In order to assess the reliability of the linear model we also analyzed the R2 of the linear model and considered a neuron to carry information about the difficulty only if the R2 was significantly higher than chance (see Methods for details). Using the MT as a regressor for the linear model 90% of neurons with significant slope of the regression line showed significant values of the R2. Using the difficulty of the task as regressor, 61% of neurons with significant slope of the regression line had also significant values of R2 meaning that the linear model is able to explain enough variance of most neurons. However the remaining 39% of neurons had lower values of R2 suggesting that non-linear methods could maybe explain better those data. In general the applied method suggests that simple assessment of the statistical significance of the slope of regression line could be a weak control in this type of analysis and that the R2 can provide useful insight into the goodness of the linear model. We provide a non parametric test to address how much the R2 values are different from those produced by chance results.
The three mechanisms underlying the difficulty V-shaped pattern that we have suggested, raise the question of whether PMv neurons change their FR gradually, or whether they jump from a low to a high activity state. This question, that has often raised concerning the decision neurons of the lateral intraparietal sulcus (LIP), has been bothering the scientific community for some time now [3, 15, 34, 35]. Recently, [36] reliable evidence has been provided for the hypothesis that LIP neurons display a gradual ramp. Although our analysis was aimed at differentiating single trial mechanisms, we did not address this issue. We do note that all three proposed mechanisms are compatible with both a gradual and an abrupt transition of states.
Our results could also suggest another interpretation: that the PMv neurons are actually encoding the confidence or uncertainty in the decision. There is a growing body of research on the role of uncertainty estimation in perceptual decisions, on its neural representations and on its computational substrates [7, 9, 21, 24, 37, 38]. The difficulty of a decision is one of the main factors influencing the confidence in that decision as demonstrated by many experimental [20, 24, 39, 40] and theoretical [20, 21, 38, 41] studies. In our experimental setup the stimuli were well visible, hence the perceptual uncertainty of the stimuli was reduced compared to other situation (e.g. when the temporal or spatial integration of the signal is necessary to reduce uncertainty in the estimation of the relevant variable: random dots motion direction discrimination, numerosity estimation, etc.) and the uncertainty in the decision was mainly affected by the relative difference in orientation of test and reference bar. Confidence measures are related to the (objective) difficulty of task with a typical X-shaped pattern: The positive correlation with discriminability in correct trials is mirrored (i.e. negative correlation) in error trials [22]. The X-pattern may suggest a role of these neurons in metacognitive processing, nonetheless we note that this pattern is neither necessary (Kornell, 2013, Kornell et al., 2011, Kiani et al., 2014) nor sufficient to define confidence, since it could also emerge from different computations (Insabato et al. in preparation). Indeed when we represent the MT as a function of discriminability and separated for correct and error trials it shows the typical X-pattern (Fig 1C). It is indeed likely that the MT correlates with confidence, since it is well known that the decision time is related to decision uncertainty [20, 38, 42]. If the behaviour of subjects could then reflect the uncertainty or confidence in the decision, this may also be present in the neural recordings. Moreover the population FR of integrative neurons separated for correct and error trials formed the X-shaped pattern (Fig 3D). We cannot rule out that the population of integrative neurons is encoding the confidence in the decision and not only the difficulty. If this were true the proposed encoding mechanism for difficulty could actually serve as mechanisms for uncertainty coding. We could speculate that the continuous coding schemes proposed may serve as a representation for decision uncertainty, while the binary mechanism may form the basis of a classification of uncertainty for confidence rating or confidence guided behaviour. However the task we used has no direct confidence measurement and therefore reasonable doubts could be cast on this interpretation of the results.
Although our results do not directly support the interpretation of a neural representation of decision confidence in PMv, they demonstrate neurons in PMv involved in the encoding of the difficulty, which is a building-block for the construction of confidence.
Experiments were made using two male monkeys (Macaca mulatta). Animals (BM5, 8 kg; and BM6, 6 kg) were handled according to the standards of the European Union (86/609/EU), Spain (RD 1201/2005), and the Society for Neuroscience Policies and Use of Animals and Humans in Neuroscience Research. The experimental procedures were approved by the Bioethics Commission of the University of Santiago de Compostela (Spain).
The monkeys’ heads were immobilized during the task and looked binocularly at a monitor screen placed 114 cm away from their eyes (1 cm subtended 0.5 to the eye). The room was isolated and soundproofed. Two circles (1° in diameter) were horizontally displayed 6° at the right and 6° at the left of the fixation point (a vertical line; 0.5° length, 0.02° wide) displayed in the screen center. The monkeys used right and left circles to signal with an eye movement the orientation of visual stimuli to the right and to the left, respectively. Orientation Discriminations Task: the monkeys were trained to discriminate up to their psychophysical thresholds in the visual discrimination task sketched in Fig 1A (training lasted for approx. 11 months). The stimuli were presented in the center of the monitor screen and eye movements larger than 2.5° aborted the task. The orientation discrimination task was a two-interval, two-alternative forced-choice task. A masking white noise signaled the beginning of the trial and then the fixation target (FT) appeared in the center of the screen (Fig 1A). The monkey was required to fixate the FT. If fixation was maintained for 100 ms, the FT disappeared, and, after a variable pre-stimulus delay (100–300 ms), two stimuli (S1 and S2), each of 500 ms duration, were presented in sequence, with a fixed inter-stimulus interval (1 s). At the end of the second stimulus, the subject made a saccadic eye movement, in a 1200 ms time window, to one of the two circles, indicating whether the orientation of the second stimulus was clockwise or counterclockwise to the first. We also recorded the movement time (MT) of the subject, the time from the end of the second interval (S2) until the response. The orientation of the test bar relative to the reference (test relative orientation, TRO = S2 − S1) manipulated the difficulty of the task. Trials lasted approx. 3.5 s separated by a variable intertrial interval (1.5–3 s). Fifty milliseconds after the correct response, a drop of liquid was delivered as a reward. A modulation of the masking noise signaled the errors; the modulation started 50 ms after the incorrect response and lasted for 75 ms.
Monkeys’ weights were measured daily to control hydration, and once a week the animals had access to water ad libitum. The level of training was assessed by the psychometric functions. Once trained, the monkeys performed around 1000 trials per day. The lines were stationary, subtending 8° length and 0.15° wide. Three different S1 orientations were used for each monkey during the recordings: 87°, 90°, and 93° (BM5) and 84°, 90° and 96° (BM6); all angles referred to the horizontal axis. Different S2, eight per S1, were presented, four clockwise and four counterclockwise to S1 in steps of 1° (BM5) and 2° (BM6). More details can be found in [4].
Neuronal population: extracellular single-unit activity was recorded with tungsten micro-electrodes (epoxylite insulation, 1.5-3.5M, catalog # UEWMGCLMDNNF; FHC) in the posterior bank of the ventral arm of the sulcus arcuatus and adjacent surface in the ventral premotor cortex in the four hemispheres of the two monkeys (see [4], for a detailed description of the recording sites). In this work, we studied the responses of a subset (324) of the recorded neurons. This subset was selected with a ROC analysis of FR with respect to the choice (see [4] for details).
All analyses were performed using custom-made programs in Matlab. Unless noted otherwise, statistical analyses were applied to the FRs of single neurons during the 500 ms preceding the saccade. In fact, the second stimulus was presented during this period, and therefore the decision-making process was expected to take place during this time window. Our aim was to find any existing neurons whose activity relates to:
In order to accomplish this we used a linear regression analysis [43]. Of course, linearity is only one of numerous possible encoding mechanisms, even when we take only those concerning FRs into consideration. We decided on this for the sake of simplicity.
As experiments were done using animals that were awake it was very difficult to record single neurons over a long period, therefore the number of error trials for each neuron was very low and error trials were excluded from the linear analysis of difficulty as noted below.
The FR of the last 500 ms before the saccade was computed by averaging the spike count in a sliding window of 100 ms slided with a step of 20 ms. In this way we got for each trial and each neuron a time series r(t) of the FR, where t is time discretized in 25 time bins.
To individuate the neurons presenting a modulation of the movement time the following linear model (LMmt) analysis was used r(t) = d1(t) MT + d2(t), where d1, d2 are the parameters to be fitted. In order to assess the significance of the coefficients a Monte-Carlo random resampling method was used. This method allows to estimate the distribution of the parameters under the null hypothesis (no dependence between the FR and the MT). To this aim we built 100 surrogated data sets by randomly reassigning the labels (MT values), thus each surrogate was constructed by permuting the values of MT over all trials. This randomization destroys eventual correlation between the FR and the MT. We then applied the LMmt to each surrogate and obtained the estimated distribution of the coefficients under the null hypothesis. Neural activity in a bin t was considered linearly dependent on the MT if the coefficient d1(t) had a low probability (p < 0.05) under the null hypothesis. In addition we required a minimum number of consecutive significant bins in order to avoid false positive results due to fluctuations. The minimum number of significant bins was chosen by estimating the probability distribution of N consecutive significant bins under the null hypothesis. We used again a Monte-Carlo random resampling method to estimate this probability distribution. For each surrogate we marked all the bins with a low probability under the null hypothesis (p < 0.05) and extracted the maximum number of those bins that were consecutive. This procedure gave us an estimate of the number of consecutive significant bins under the null hypothesis. We considered the activity of a neuron to be dependent on the MT if the maximum number of consecutive significant bins had a low probability under the null hypothesis. In order to correct for multiple comparison (we applied the same test to all neurons) we used the Benjamini-Hochberg procedure [25] to control the False Discovery Rate to a value Q. We show in S1 Fig the results of using different values of Q. The number of total discoveries is represented by bars (errorbars represent standard deviation of bootstrapped data) as a function of the value of Q. The dashed red line represent the maximum number of accepted false discoveries. We can observe that there are many true discoveries in the set of accepted discoveries, indeed the number of total discoveries is much higher than the maximum number of accepted false discoveries for a wide range of Q. This shows that our findings are robust over a wide range of Q. However to further analyse the neurons found to be significant, e.g. to calculate the average population activity or to show the activity of one single neuron, we used Q = 0.05 in order to keep the number of false discoveries low. To give more insight on this method, appendix S1 Text presents a description of Benjamini-Hochberg procedure [25] for varying Q in the analysis of a synthetic dataset, where the ground truth is known.
It is possible that the slope of the regression line, d1, is significantly different than that obtained by chance under the null hypothesis but still the LMmt is explaining a small part of the variance of data. In order to assess the portion of explained variance we calculated the coefficient of determination (R2) of the LMmt. The values of the R2 were in the range between 0 and 0.75. To determine which values of the R2 were high enough, we estimated the distribution of the R2 under the null hypothesis, with the same Monte-Carlo random resampling methods explained above for d1, and applied the same constraints as above (p-value, minimum number of consecutive bins, multiple comparison correction). Finally we considered a neuron to be informative about the MT if both R2 and d1 were significant (FDR controlled with Benjamini-Hochberg procedure) in the same interval. This procedure was used for all linear models used in this study.
To individuate the neurons presenting a modulation of task difficulty only correct trials were used and the following linear model (LMtro) analysis was used independently on trials with positive and negative values of TRO: r(t) = d1(t) TRO + d2(t), where d1, d2 are the parameters to be fitted. The significance of the coefficients was assessed with a Monte-Carlo random resampling method was used as explained above for the LMmt. In addition a minimum number of consecutive significant bins was required (as explained above for the LMmt) in order to avoid false positive results due to fluctuations. Finally only neurons with significant R2 were considered to encode TRO, similar to the above explained analysis of LMmt. In S2 and S3 Figs, both for the R2 and for the slope of the linear model, we show the resulting number of discoveries as a function of Q (same conventions as in S1 Fig). It is easy to observe, in both figures, that in a wide range of Q the number of total discoveries is much higher than the maximum number of accepted false discoveries. For further analysis on this group of neurons we used Q = 0.1. Moreover we want to emphasize that the difference between the number of total discoveries and the maximum number of accepted false discoveries (i.e. the distance between bars height and the dashed red line) first increases, then stabilizes in a wide range of Q and then decreases towards Q = 1. For example, in S3 Fig, for Q = 0.1 44 total discoveries are made and 4 of them are expected to be false, hence 40 discoveries are true; for Q = 0.2, 88 total discoveries are made and 18 of them are expected to be false, hence 70 discoveries are true. In general, under these conditions, if one want to find more true discoveries (more power) a higher Q-value may be chosen, e.g. Q = 0.2, although one must be always aware that this higher number of true discoveries comes at the price of more false discoveries.
These neurons are not necessarily encoding TRO for both choices of the subject (left or right) since the LMtro was fitted separately for positive and negative values of TRO. However the whole ensemble of neurons encodes the information about the TRO. Then we looked for neurons, that can integrate the information encoded by this ensemble and represent the difficulty of the trial. Such neurons would present a V-pattern when the FR is plotted as a function of TRO (or a reversed V). In order to find this integrative difficulty neurons the following linear model (LMdiff) was used r(t) = d1(t) ∣TRO∣ + d2(t), where d1, d2 are the parameters to be fitted. The same statistical testing procedure was used as for the other linear models in order to assess the significance of results. In S4 Fig we show the total number of discoveries as a function of Q-value as for the other models described above. Here again we observe that our results are robust over a wide range of Q. However for further analysis on this group of neurons we used Q = 0.1.
In order to produce Fig 3D the FR of each neuron was normalized to its maximum value and then the activity of all neurons was averaged together.
We individuated three possible neural mechanisms responsible for the above mentioned modulation of the difficulty neurons. A simplified representation of these mechanisms is presented in Fig 4. In order to understand which difficulty neuron belongs to each of the three categories, we applied different methods:
In order to find neurons that switch states with a timing dependent on the difficulty, we used the Hidden Markov Model (HMM) analysis [44] to estimate the time of neural activity change due to the test bar. Indeed, the HMM was able to cluster the spiking activity of individual neurons into periods of ‘stationary’ activity (the states) within a single trial. Hence the switch time between states could be estimated. In order to find neurons whose activity after the change, as estimated by HMM, encoded the difficulty, we calculated the correlation between the mean activity and the difficulty of the task. The mean activity was calculated in the time window starting at the time bin where the 90% of the trials had passed from one state to the other and ending at the last significant time bin marked by the LMdiff or LMtro. In order to find the neurons whose activity could be explained as a compound of high and low FR states we fitted (with Expectation Maximization algorithm) the FR distribution to a Gaussian mixture model.
For each one of this method we controlled the FDR with Benjamini-Hochberg procedure [25]. Also in this case the number of true discoveries was quite stable over a wide range of Q; the reported results are for Q = 0.05.
To analyze the single-trial activity of the recorded neurons we used the Hidden Markov Model that clusters the spiking activity of individual neurons into periods of stationary activity within a single trial. The HMM technique has been successfully applied to characterize the single-trial activity of cortical neuronal ensembles during movement with holding and preparation [45, 46], taste processing [47], and perceptual decision making [3]. Here, we briefly review some aspects of the HMM analysis; more details about the algorithms can be found in previous works [3, 45, 47].
Within the HMM, the activity of a recorded neuron at time t is assumed to be in one of a (predetermined) number (Q) of hidden FR states. In each state q, the discharge of a neuron is assumed to be a Poisson process of intensity λq, which defines the instantaneous firing probability Eq, i.e., the probability of firing a spike within one time bin, equal to 2ms throughout this study. States are said to be hidden because they are not directly measured; instead, we observe the stochastic realizations of the state-dependent Poisson process (observation sequences). The state variable changes from state i to state j with fixed probabilities that defined a transition matrix A, given by Aij = P(qt+1 = j∣qt = i), where qt is the state at time t and i, j ∈ {1, …, Q}. The entire process is a Markov chain: the transition probabilities Aij are independent of time, i.e., they depend only on the identities of states i and j, which means that the state sequence at time t only depends on the state at time t − 1. In summary, for a single neuron the HMM is fully characterized by the spike-emission probabilities (E) and the transition matrix (A). These model parameters are estimated from the data, using a likelihood expectation-maximization algorithm [3, 45, 47].
Briefly explained, the procedure starts with random values for E and A and re-estimates the parameters to maximize the probability of observing the data given the model. After optimization of the model parameters, the Viterbi algorithm is used to find the most likely sequence of hidden states given, for each single trial, the model and the observation sequence [3]. In the present study we used the HMM to detect the transitions between a state of low and a state of high activity. For this reason, the number of states was set to Q = 2. For each neuron, the data was divided into two subsets, composed of trials corresponding to each behavioral response (left or right). For each subset, a HMM was estimated using the activity of 80% of the trials (randomly selected) during the period within the last 500 ms before the saccade. After optimization the most likely state sequence was stored for all trials.
Unfortunately, a HMM analysis was not reliable for all the neurons. We only considered the HMM reliable if a) the mean duration over trial of both states was at least 25 ms. (i.e., we do not take into account states with very brief duration), b) the number of state-switches per trial was three or less; or (i.e., we do not take into account bursting neurons), c) at least five of both the left and right oriented trials had a state-switch (i.e., we want neurons with 2 different states). We found 26 difficulty neurons (out of 101) whose HMM was interpretable. For this subset, we wanted to distinguish between the three V-shaped mechanisms, to do so we analyzed the state-switch time. For each trial the HMM gave the time in which it changed from a low to a high state (or vice versa).
We also performed the analysis with three hidden states (Q = 3). After applying the above mentioned constrains, we found only five neurons for which the three states HMM was reliable. For these neurons the third state was associated with the slowly increasing activity, which may be detected as an intermediate state. S5 Fig shows the dynamics of the hidden states for one of the three neurons. Since we are interested only in the timing of the change from inter-stimulus period activity to activity induced by the second stimulus we didn’t take into account the three states HMM. In addition the BIC selects the two states HMM as the better model for all these five neurons.
In S6 Fig we show the time course of state switch for the whole neural population. Overall the population switches from state one to state two gradually over time. However the curves in S6 Fig are not strictly monotonically increasing, meaning that some neuron present a second switch from state two to state one. Indeed we mainly found just one switch per neuron but some neurons presented also a return to state one: they respond rapidly to the test bar stimulus and then go back to a lower activity state. This dynamics is illustrated for an example neuron in S7 Fig.
Our aim was to investigate whether the FR distribution of neurons during correct trials was better described using a mixture model composed of two distributions rather than a single distribution. The procedure we applied was the following:
S9 Fig shows a visual summary of this analysis for a binary neuron, while S10 Fig show the results for a non-binary neuron.
|
10.1371/journal.pgen.1000406 | The Distribution of Fitness Effects of Beneficial Mutations in Pseudomonas aeruginosa | Understanding how beneficial mutations affect fitness is crucial to our understanding of adaptation by natural selection. Here, using adaptation to the antibiotic rifampicin in the opportunistic pathogen Pseudomonas aeruginosa as a model system, we investigate the underlying distribution of fitness effects of beneficial mutations on which natural selection acts. Consistent with theory, the effects of beneficial mutations are exponentially distributed where the fitness of the wild type is moderate to high. However, when the fitness of the wild type is low, the data no longer follow an exponential distribution, because many beneficial mutations have large effects on fitness. There is no existing population genetic theory to explain this bias towards mutations of large effects, but it can be readily explained by the underlying biochemistry of rifampicin–RNA polymerase interactions. These results demonstrate the limitations of current population genetic theory for predicting adaptation to severe sources of stress, such as antibiotics, and they highlight the utility of integrating statistical and biophysical approaches to adaptation.
| Adaptation by natural selection depends on the spread of novel beneficial mutations, and one of the most important challenges in our understanding of adaptation is to be able to predict how beneficial mutations impact fitness. Here, we investigate the underlying distribution of fitness effects of beneficial mutations that natural selection acts on during the evolution of antibiotic resistance in the opportunistic human pathogen P. aeruginosa. When the fitness of the wild type is high, most beneficial mutations have small effects. This finding is consistent with existing population genetic models of adaptation based on statistical theory. When the fitness of the wild type is low, most beneficial mutations have large effects. This distribution cannot be explained by population genetic theory, but it can be readily understood by considering the biochemical basis of resistance. This study confirms an important prediction of population genetic theory, and it highlights the need to integrate statistical and biochemical approaches to adaptation in order to understand evolution in stressful environments, such as those provided by antibiotics.
| Adaptation by natural selection ultimately depends on the spread of novel beneficial mutations that increase fitness. Can we predict the fitness effects of beneficial mutations? Gillespie[1],[2] argued that extreme value theory (EVT) provides a simple answer to this question: the tails of all-Gumbel type distributions (a very flexible type of distribution that includes many familiar distributions, including the normal) are exponential. As such, the fitness effects of beneficial mutations will be exponentially distributed provided that the fitness of the wild-type is high enough so that beneficial mutations are drawn from the extreme tail of the distribution of fitness effects of mutations. It is however unclear how robust this theory is with respect to the fitness of the population prior to selection. As the absolute fitness of the wild-type decreases (for example, because of environmental change), a larger proportion of single mutations will increase fitness. Beneficial mutations will therefore no longer be drawn exclusively from the tail of the distribution, hence the exponential distribution may no longer apply (Figure 1).
In this paper, we investigate how the distribution of fitness effects of beneficial mutations changes with the fitness of the wildtype population using the evolution of antibiotic resistance in the opportunistic pathogen Pseudomonas aeruginosa as a model system. Experimental studies of the underlying distribution of fitness effects of beneficial mutations[3],[4],[5],[6] have lagged behind the theory, both because beneficial mutations are exceedingly rare and because beneficial mutations of small effect are less likely to reach appreciable frequencies in populations because of the combined effects of drift[7] and competition between independent mutations[8]. To overcome these limitations, we used a fluctuation test to isolate clones of the bacterium Pseudomonas aeruginosa with mutations in the β-subunit of RNA polymerase (rpoB) that are beneficial in the presence of the drug rifampicin [9],[10],[11]. Our experimental design ensured that we obtained an unbiased sample of all beneficial mutations. First, we isolated mutants from populations that were propagated in culture media lacking rifampicin, implying that we isolated beneficial mutations prior to any selection for rifampicin resistance. Second, we experimentally prevented competition (ie clonal interference[8]) among independently derived beneficial mutations by randomly choosing independent mutants. Third, we ensured all mutations included in the analyses were unique by sequencing rpoB of the mutants. To test the hypothesis that the fitness effects of beneficial mutations are exponentially distributed, we used log-likelihood tests that have been specifically developed to test this hypothesis using this experimental design[12].
Using this approach, we show that the distribution of fitness effects of beneficial mutations is variable: under conditions where the fitness of the wild-type is high, the fitness effects of beneficial mutations are exponentially distributed, as predicted by theory. However, when the fitness of the wildtype is low, the data may no longer fit an exponential distribution because many beneficial mutations have large effects on fitness. We show that this non-exponential distribution of fitness effects emerges as a direct consequence of the molecular interactions that are under selection in this system and we argue that existing theory on the fitness effects of beneficial mutations cannot be applied to understand adaptation to novel stressful environments, such as those provided by antibiotics.
To investigate how the distribution of fitness effects of beneficial mutations changes with the fitness of the wild-type, we measured the fitness of beneficial mutations isolated at a high concentration of rifampicin (Table 1) and the wild-type across a gradient of rifampicin concentrations (Figure 2). At low concentrations of rifampicin (1–2 ug/mL), the fitness of the wildtype is high and we cannot reject the null hypothesis that the fitness effects of beneficial mutations are exponentially distributed, as determined by a likelihood-ratio test (Figure 3, Table 2). However, at high concentrations of rifampicin (>2 ug/mL), the fitness of the wild-type is low and the fitness effects of beneficial mutations are not exponentially distributed (Figure 3, Table 2).
One limitation of this study is that our power to test the null hypothesis is weakest under conditions where the fitness of the wild-type is high, because only half of the mutants that we isolated increase fitness at low concentrations of rifampicin. Given that we had to sample 80 mutants in order to identify 15 beneficial mutations (this saturation effect is in part attributable to a strong mutational bias towards two mutations), it unlikely that increasing the sample size of our study would have substantially increased the power of our analysis. It is also important to note that this limitation is not unique to this study: beneficial mutations are rare events, and all other comparable experimental evolution studies are based on a similar sample size of beneficial mutations[4],[13].
To gain insight into the mechanistic basis of fitness, we measured both the growth rate in the absence of antibiotics and the degree of rifampicin resistance for each beneficial mutation (Figure 4). At low concentrations of rifampicin (ie 1–2 ug/mL), selection for high levels of resistance is weak, and fitness is highly correlated with growth rate in the absence of antibiotics (r = .86–.9, P<.0001). The genetic variation in growth rate in the absence of antibiotics generated by spontaneous mutation is normally distributed (Figure 3; W = .92, P = .21), hence the fitness effects of beneficial mutations are exponentially distributed because only mutations in the right tail of the distribution (ie those mutations that are associated with a low cost of resistance[14]) are beneficial at low concentrations of rifampicin. At high concentrations of rifampicin, selection for resistance is strong and the fitness effects of beneficial mutations no longer fit an exponential distribution because most beneficial mutations had large effects on resistance and, therefore, fitness at high concentrations of rifampicin (Figure 4).
The large effect of beneficial mutations on resistance is consistent with the molecular interactions that occur between rifampicin and RNA polymerase. Structural studies have shown that rifampicin binds to a small, highly conserved pocket of the β-subunit of RNAP and only 12 amino acid residues are involved in direct interactions with rifampicin[9],[11]. Mutations at these residues cause a large increase in resistance (mean IC50 = 423 ug/mL, s.e = 25 ug/mL, n = 10). Residues that surround the binding pocket interact only indirectly with rifampicin, and it has been argued that resistance arises at these residues due to amino acid changes that alter the folding of the protein in the binding pocket. We identified only a small number of beneficial mutations (n = 4) in residues that are involved in indirect Rif-RNAP interactions and mutations at these residues give rise to intermediate levels of rifampicin resistance (mean IC50 = 197 ug/mL, s.e = 40 ug/mL, n = 4). This biophysical approach to understanding the effects of beneficial mutations suggests that the data may no longer fit an exponential distribution because of the high specificity of interactions between rifampicin and RNA polymerase: changes to the majority of amino acids that are involved in rifampicin-RNAP interactions results in large increases in resistance and, therefore, large increases in fitness at high concentrations of rifampicin. To test this hypothesis further, we assayed fitness in the presence of sorangicin[15], an antibiotic that has been shown to bind to the same domain of RNAP as rifampicin and share the same mode of action, inhibition of transcription initiation[16]. The biochemical difference between these antibiotics comes from the fact that sorangicin has a much higher conformational flexibility than rifampicin [16]. The fitness consequence of this difference is that many mutations that give a large increase in fitness under high concentrations of rifampicin give only a small increase in fitness in the presence of an equivalent dose of sorangicin (Figure 4), and the observed distribution of fitness effects of beneficial mutations does not differ significantly from the exponential (Figure 5, −2logΛ = 5.72, n = 9, P = .09).
It is important to note that these results do not necessarily imply that the distribution of fitness effects of mutations that are beneficial in the presence of sorangicin is universally exponential. Most mutations that increase sorangicin resistance do so by altering membrane permeability, instead of altering the structure of RNAP[16], but unfortunately the mutations that are responsible for this decrease in permeability to sorangicin are not known and we are therefore unable to measure the underlying distribution of fitness effects of mutations that are beneficial in the presence of sorangicin. Instead, our interpretation of this result is that it provides a clear demonstration that the high-affinity interactions that occur between rifampicin and RNAP are ultimately responsible for the non-exponential distribution of fitness effects of beneficial mutations at high concentrations of rifampicin.
The variability in the distribution of fitness effects of beneficial mutations in this study is consistent with population genetic theory. When the fitness of the wild-type is high, beneficial mutations can be viewed from a statistical perspective as representing draws from the extreme tail of the distribution of fitness effects of mutations, hence the fitness effects of beneficial mutations will be exponentially distributed. However, EVT does not specify how high fitness has to be in order for this theory to apply. In our experimental system, this distribution held over a wide range of parameter space: we failed to detect significant deviations from the exponential distribution when the fitness of the wild-type was reduced by 20–30%. When the fitness of the wild-type is low, statistical theory does not make any predictions regarding the form of the distribution of fitness effects of beneficial mutations and hence there is no reason to expect an exponential distribution. Comparable experimental studies in viral systems are in agreement with this idea: Sanjuan and colleagues [4] found that the fitness effects of beneficial mutations are exponentially distributed in VSV under conditions where the fitness of the wild-type is high, while Rokyta et al. [3] found that the fitness effects of beneficial mutations that allow phage to attack novel hosts (ie hosts that are inaccessible to a WT virus) are not exponentially distributed. Note that in these and our experiments, the data may stop fitting the exponential distribution under conditions of low wildtype fitness not only because EVT, by definition, no longer applies, but also because the underlying mutational distributions may vary with environmental conditions.
Despite our lack of certainty of the statistical explanation for the observed distribution when the fitness of the wild-type is low, we have a good molecular mechanistic explanation (in retrospect we may have been able to predict this distribution a priori). Specifically, the distribution is biased towards mutations of large effect as a result of the high specificity of interactions between rifampicin and RNA polymerase that arises from the low conformational flexibility of rifampicin. Antibacterial and antiviral drugs are usually involved in highly specific interactions with their target proteins [17], suggesting that a bias towards mutations of large effect may be a general feature of adaptation to antibiotics [17],[18] and other situations when high-specificity protein-ligand interactions are under strong selection, for example during host-parasite interactions[19] or when enzymes are selected to recognize novel substrates[20],[21].
Recently, there has been considerable interest among population geneticists in developing general models of adaptation based on the statistical properties of extreme events. Our work highlights both the strengths and limitations of this approach and we suggest that the development of a complete theory of adaptation will require integrating molecular biology, in order to be able to predict the impact of mutations on fitness, and statistical approaches to adaptation, to be able to understand how natural selection samples the distribution of fitness effects of beneficial mutations during adaptive walks.
A single clone of Pseudomonas aeruginosa PAO1 was inoculated into 5 ml of M9KB medium that was incubated overnight at 37 C with constant shaking (150 rpm). This overnight culture was diluted down 10−6 into fresh M9KB and 120 uL aliquots of this diluted culture were used to setup 480 cultures on 5 96 well microplates. These cultures were incubated overnight at 37 C without shaking. To isolate beneficial mutations, 5 uL of each of the 480 overnight cultures was plated out on M9KB supplemented with 62.11 ug/mL of rifampicin, the minimal concentration required for complete inhibition of growth of the wildtype strain. To isolate beneficial mutations, we isolated a single colony from each of the first 80 cultures that gave samples containing exactly 1 colony on agar plate containing rifampicin.
To determine the mutations underlying adaptation, we sequenced the rpoB gene in each of the 80 colonies that we isolated in our fluctuation test. Genomic DNA was isolated from each colony using a Wizard Genomic DNA extraction kit (Promega, UK) as per the manufacturer's instructions. Our sequencing strategy was to first sequence a highly-conserved domain of rpoB that is known to be important for rifampicin resistance in all 80 clones. This region was amplified with primers rpoB_fwd (5′-GTTCTTCAGCGCCGAGCG-3′) and rpoB_rev (5′-GCGATGACGTGGTCGGC-3′) that amplify the region of the rpoB gene between nucleotides 1178 and 1864. Reaction mixtures consisted of BIOTAQ polymerase (Bioline, UK), 1 mM dNTPs, 16 nM (NH4)2SO4, 62.5 mM Tris-HCL (pH 8.8), .01% Tween 20, 2 mM MgCl2, and each primer at a concentration of .2pM. Amplification reactions were carried out as follows: 94 C for 5 minutes, followed by 35 cycles of 94 C for 30 seconds, 60 C for 30 seconds, and 72 C for 1 minute, followed by a final incubation at 72 C for 10 minutes. PCR products were purified using a MultiScreen PCR96 filter plate (Millipore, UK) as per the manufacturer's instructions. Purified PCR products were sequenced with both forward and reverse primers using BigDye 3.1 sequencing (Applied Biosystems International) followed by ethanol/EDTA precipitation of sequencing products. In all cases, this strategy identified either a single mutation in this region of rpoB or no mutations.
Clones that lacked a mutation in the highly conserved domain of rpoB were subsequently sequenced for a second region that has previously been implicated in rifampicin resistance spanning nucleotides 1 to 1012 of the rpoB gene. This region was amplified and sequenced with primers rpoB_up (5′–ATGGCTTACTCATACACTGAG-3′) and rpo_B1 (5′-CTCGATGCG CACGACCTG-3′). The protocol was the same as described above, except that the annealing temperature used in the PCR reactions was 54 C instead of 60 C. We idenfied a single mutation in this region in all of the clones that did not contain a mutation in the highly conserved domain of rpoB. As a further control, we sequenced the entire rpoB gene in six randomly chosen clones. We failed to detect any second site mutations in rpoB using this approach.
To assay fitness, we estimated the growth rate, r, of each beneficial mutation and wildtype PAO1 at 4 different concentrations of rifampicin (0,1,2,8, and 64 mg/L). Pre-assay overnight cultures of each mutant were prepared by growth in M9KB. These cultures were then diluted 100 fold into fresh culture medium and we measured the growth rate of each culture using an automated microplate reader by taking hourly measurements of optical density at 600 nm (OD600) over a period of approximately 12 hours. All incubations were carried out at 37 C. Assays at low concentrations of rifampicin (0,1,2,8 mg/L) were carried out with 12 fold replication and assays at high concentrations of rifampicin (64 mg/L) were carried out with 18 fold replication. A further assay was carried out using the same method to measure fitness in the presence of sorangicin (20 ug/mL). OD600 is proportional to the log of cell density, and the slope of OD600 against time (mOD/min) in exponential growth phase therefore provides an estimate of r, the growth rate of the bacterial clone, such that ri,j is the growth rate to the ith genotype at the jth concentration of rifampicin.
To test the hypothesis that the fitness effects of beneficial mutations are exponentially distributed, we used a likelihood ratio test developed by Beisel and colleagues [12]. According to EVT, there are three limiting tail distributions, the Fréchet (which has a heavier-than-exponential tail), the Gumbel (which has an exponential tail), and the Weibull (right-truncated). The tails of all three of EVT domains can be described by the generalized Pareto distribution (GPD), which has a cumulative distribution function given by:with shape parameter κ and scale parameter τ. One very interesting property of the GPD is that the shape parameter is threshold-independent. This property of the GPD is critical, because it implies that it is possible to account for any potential bias against detecting mutations of small beneficial effect by simply re-scaling fitness data so that the fitness of beneficial mutations is expressed relative to the least-fit beneficial mutation instead of the wild-type. To take advantage of this property, we estimated the fitness of each beneficial mutation as as wi,j = ri,j−r1,j where wi,j is the fitness of the ith beneficial mutation in the jth environment, ri,j is the growth rate of the ith beneficial mutation in the jth environment, and r1,j is the growth rate of the least fit beneficial mutation in the the jth environment (ie the beneficial mutation that has the smallest increase in growth rate relative to the rifampicin sensitive clone in the jth environment).
The likelihood ratio test developed by Beisel and colleagues calculates −2logΛ, negative twice the difference in log-likelihood between two statistical models, one model in which the shape parameter of the GPD is set to 0 and the other in which the scale parameter of the GPD is free to vary (ie H0:κ = 0, HA: κ≠0). P values were calculated by performing 10000 parametric bootstrap replicates using a software package (EVDA) developed for R (software available at http://www.webpages.uidaho.edu/~joyce/Lab%20Page/Computer-Programs.html). This test is potentially sensitive to measurement error, but the accuracy of our fitness measurements was high enough (Average CV = 19.8%) that measurement error should not inflate the probability of making a type I error [12].
To measure the resistance conferred by rpoB mutations, we assayed growth in media containing rifampicin at different concentrations. Pre-assay cultures of rpoB mutants and PAO1 were prepared by overnight growth of freezer cultures in M9KB at 37 C. These cultures were then diluted 2.5×10−5 into M9KB, or M9KB supplemented with rifampicin at the following concentrations (all in ug/mL): 0, 3.9, 7.8, 15.6, 31.3, 62.5, 125, 250, 500, and 1000. Assay cultures were incubated at 37 C and we measured the optical density of cultures after exactly 24 hours of incubation (+/−10 minutes) at 600 nM using an automated microplate reader. We assayed the resistance of 12 cultures of each of the rpoB mutants that we identified and 12 replicates of PAO1. Resistance was calculated as IC50, the concentration of rifampicin necessary to cause a 50% reduction in optical density, using the following regression model: , where y is optical density, measured in absorbance units, x is the concentration of rifampicin, measured in ug/mL, and H is parameter that estimates the rate of decay in optical density with increasing rifampicin concentration.
|
10.1371/journal.pcbi.1006188 | Polymorphic sites preferentially avoid co-evolving residues in MHC class I proteins | Major histocompatibility complex class I (MHC-I) molecules are critical to adaptive immune defence mechanisms in vertebrate species and are encoded by highly polymorphic genes. Polymorphic sites are located close to the ligand-binding groove and entail MHC-I alleles with distinct binding specificities. Some efforts have been made to investigate the relationship between polymorphism and protein stability. However, less is known about the relationship between polymorphism and MHC-I co-evolutionary constraints. Using Direct Coupling Analysis (DCA) we found that co-evolution analysis accurately pinpoints structural contacts, although the protein family is restricted to vertebrates and comprises less than five hundred species, and that the co-evolutionary signal is mainly driven by inter-species changes, and not intra-species polymorphism. Moreover, we show that polymorphic sites in human preferentially avoid co-evolving residues, as well as residues involved in protein stability. These results suggest that sites displaying high polymorphism may have been selected during vertebrates’ evolution to avoid co-evolutionary constraints and thereby maximize their mutability.
| Amino acid co-evolution represents cases of simultaneous substitution of amino acids at distinct positions in protein sequences. In the MHC-I protein family, such co-evolution could result from either amino acid changes across species or changes within species due to the high polymorphism of MHC-I molecules. Here we show that signals captured by global methods such as Direct Coupling Analysis (DCA) to estimate co-evolution primarily result from changes across species. Moreover, our results indicate that polymorphic sites in MHC-I molecules tend to be decoupled from co-evolving ones. This could suggest that they have been selected to maximize their mutability, which is known to be functionally important to entail MHC-I molecules with a wide repertoire of binding specificities for antigen presentation.
| Major Histocompatibility Complex class I proteins (MHC-I), also referred to as Human Leukocyte Antigen class I (HLA-I) in human, are expressed on the surface of cells. MHC-I proteins form a complex with either ‘self’ ligands derived from the endogenous proteins or ‘foreign’ ligands (non-self) derived from invading pathogens or somatic alterations in cancer cells. Upon presentation of non-self ligands from inside the cytoplasm, the complex can be recognized by CD8 T-cells [1]. MHC-I proteins show a very high degree of polymorphism especially around the peptide-binding groove and tens of thousands of different alleles are reported in databases like PFAM [2] or IMGT/HLA [3]. Moreover, striking differences in binding specificity are observed between different alleles. Several evolutionary events contributed to MHC-I diversity in vertebrates. Duplication events occurred during the evolution of jawed vertebrate, which led to MHC-I polygenicity in many species [4,5]. Following the gene duplication events, the different gene copies diverged through separate evolutionary processes, which allowed some MHC-I genes to gain different functions, while others became dysfunctional or lost [6]. Consequently, the number of MHC-I loci differs between vertebrate species [7]. These duplication events produced 6 MHC-I genes in human all located on chromosome 6. Three of them (HLA-A, HLA-B and HLA-C) are broadly expressed in most cell types and are the main contributors to class I antigen presentation. The high level of allelic diversity of the MHC-I in vertebrate population is likely due to strong selection because of the exposure of vertebrate populations to various infections across the world [8] [9]. In particular, the polygenicity and polymorphism entails the immune system of each individual with the ability to present at the cell surface a wide range of peptides from foreign pathogens.
Despite their high polymorphism, MHC-I alleles share the same three-dimensional fold across vertebrates. In particular, the peptide-binding groove is composed of two almost parallel alpha helices and one beta sheet. This conserved structure across all MHC-I alleles suggests that they undergo molecular constraints. Molecular constraints can be predicted using stability models that investigate the impact of a mutation on the structure (e.g. alanine scanning) [10] or conservation [11]. Recent studies have also demonstrated that simultaneously evolving sites (also called co-evolving sites) can reveal structural contacts [12] folding intermediate [13], allosteric communication, core protein sites [14], or functionally important sites [15]. Several models are available in the literature to predict co-evolving sites. Most of the models evaluate a score to assess if a pair of sites simultaneously evolves regardless of the other residues. Some of these models use statistical formalisms such as Mutual Information [16], Statistical Coupling Analysis [17] or Coev [14,18] when others use combinatorial formalism [19,20]. The only model that investigates co-evolving residues in the light of global alignment is Direct Coupling Analysis (DCA) [12], also introduced in the EVfold suite [21]. This phylogeny-free method was shown to accurately identify sites in contact in protein structures, and because of this, DCA has been used to help predicting protein structures [21][22][23][24].
In this work, we study the co-evolving constraints on MHC-I across vertebrates’ species using DCA. Despite the low number of species (<500), we observed that DCA could accurately predict structural contacts directly from MHC-I protein sequence alignment. We then investigated the relationship between polymorphism and co-evolution constraints. Our work reveals that polymorphism within human does not contribute much to the observed co-evolution signal. Moreover polymorphic sites show little overlap with both co-evolving sites across vertebrates and sites predicted to be most important in protein structural stability. We further extended the DCA algorithmic framework to incorporate multiple MHC-I ligands per allele and observed the same uncoupling between co-evolving and polymorphic residues. These results suggest that polymorphic residues in MHC-I molecules preferentially avoid sites displaying strong stability or co-evolutionary constraints.
To investigate co-evolutionary constraints among MHC-I residues we retrieved all MHC-I protein sequences from the PFAM v30 database (PF00129) [2]. This domain family covers the MHC-I domains alpha1 and alpha2 (179 amino acid) and is present in 445 organisms [2]. We excluded from the dataset 117 sequences from 14 bacterial and viral species (see Materials and Methods). We ended up with 40’739 sequences, including 20’256 sequences from human MHC-I alleles where the MHC-I polymorphism has been most studied (Fig 1). We then applied DCA on the whole PFAM alignment. Considering pairs of residues that are distant along the protein sequence (more than 4 residues apart), we observed a very strong enrichment of structural contacts among pairs of residues with high DCA scores (Fig 2A). For instance, among the top 44 DCA predictions (25% of MHC-I PFAM domain length), 31 correspond to pairs of residues less than 8Å apart in crystal structures (see Fig 2A and Materials and Methods). For illustration the top 6 DCA predictions (pairs 3–29, 93–119, 47–60, 26–33, 148–154 and 36–43, with residue numbering as in X-ray structures) are shown in Fig 2B. Similar results were obtained using plmDCA [25][26](see S1 Fig). Overall, our results indicate that high enrichment in structurally interacting pairs of residues can be obtained with DCA even for a domain family spanning a relatively low number of species (in our case only vertebrates).
To assess the contribution of the 20’256 human sequences to the co-evolution predictions, we led two additional experiments: one where the co-evolving scores based on DCA are evaluated using solely the 20’256 human sequences (Fig 2C) and another where the co-evolving scores are evaluated by excluding the human sequences from the analysis (Fig 2D). These experiments revealed that the top predictions of DCA applied to human sequences did not highlight pairs of residues close in protein structures (Fig 2C). Reversely, when excluding all human sequences DCA predictions of co-evolving sites remained almost unchanged and still pinpointed mainly pairs of sites in the structural proximity (Fig 2D). Similar results are obtained using a threshold of 5Å to define the contact map (S2 Fig). Moreover when removing the sequences from species with more than 500 MHC-I sequences (Homo sapiens (Human); Macaca mulatta (Rhesus macaque); Macaca fascicularis (Crab-eating macaque) (Cynomolgus monkey); Acrocephalus schoenobaenus (sedge warbler); Parus major (Great tit); Macaca nemestrina (Pig-tailed macaque); Bos taurus (Bovine); Sus scrofa (Pig), we still observed that many of the top co-evolving sites are in structural proximity (S3 Fig). Altogether these experiments suggest that the co-evolution signal captured by DCA reflects molecular constraints in the course of vertebrate evolution, and not constraints on polymorphic sites within one species. This is in line with the low weight on human sequences due to their high homology in DCA within the full alignment (see Fig 1). Nevertheless, the lack of structurally meaningful correlations when considering only human sequences suggest that little co-evolution is observed among them, although polymorphic sites are contacting each other in the MHC-I binding site, and therefore could potentially display some level of correlation reflecting structural constraints.
To further investigate the relationship between polymorphism and co-evolving sites, we measured conservation in human using information content (see Materials and Methods) to derive a polymorphism score for each site. A position with a minimal score is rarely mutated in human MHC-I alleles whereas a position with a high score is highly mutated. We then used Enrichment Analysis (see Materials and Methods) to determine the overlap (or absence thereof) between sites displaying strong co-evolutionary constraints across vertebrates as measured by DCA and polymorphic sites in human population. DCA scores were established for each site based on the highest DCA values with any other site more than 4 amino apart in the sequence, and sites where ranked based on these scores (x-axis in Fig 3A, lower panel) to compute the enrichment (or absence thereof) in polymorphic sites among sites with highest DCA scores. Using a threshold of 0.01 on the information content to define polymorphic sites, our analysis showed that pairs of sites with the highest DCA score mainly comprise sites that are non-polymorphic in human (Fig 3A, P = 0.008). This observation holds for threshold values of 0.02 and 0.03 (S4 Fig), or when defining polymorphic sites based on the most frequent MHC-I alleles in Caucasian population (see Materials and Methods and S5 Fig). Similar results would be obtain by taking a threshold of 0.1 on the DCA score and using Fisher’s exact test to probe the depletion of points in the upper left part of Fig 3A (P = 0.003). The advantage of the enrichment approach is that is does not require fixing a threshold on the DCA scores. We further note that the cloud of points for DCA values lower than 0.08 in Fig 3A was expected since the majority of DCA values obtained from any alignment are significantly bigger than zero. However, as observed in previous studies, only the top ranking pairs give meaningful information about structural contacts. This is the reason why we used enrichment analysis in this work, as opposed to correlation coefficient whose value would be dominated by the low DCA scores, which cannot be interpreted in terms of biologically meaningful co-evolutionary constraints.
We then investigated the relationship between polymorphism and predicted importance for structural stability. Stability score of each site was evaluated using FOLD-X AlaScan software [10,27] using the X-ray structure of HLA-A02:01 in complex with a 9-mer ligand (PDB: 2BNR). Sites with different stability values were then used in the same enrichment analysis as before to compare with polymorphic sites. Here as well, we observed that polymorphic sites tend to be distinct from sites predicted to play a role in protein stability (Fig 3B, P = 0.04). This observation holds when considering other alleles and their corresponding pdb structures to evaluate stability score of each residue (Table 1). We further investigated the relationship between polymorphism and the number of structural contacts made by each residue (Materials and Methods). As expected from the stability analysis (Fig 3B), residues making many contacts tend on average to be enriched in non-polymorphic sites (Fig 3C), although the enrichment did not pass the 0.05 threshold for significance. In general, the fact that polymorphic sites that do not lead to dysfunctional proteins, such as those in MHC proteins, are less implicated in protein stability has been documented in many previous studies [28–32]. However, to our knowledge, our work is the first to perform such analysis specifically on MHC proteins.
To assess whether co-evolving pairs of residues may simply reflect sites involved in protein stability, we investigated the relationship between DCA scores and either stability or number of contacts. We observed a very poor correlation between DCA scores and stability scores (S6A Fig) or number of contacts (S6B Fig). As expected, we observed a higher correlation between stability scores and number of contacts (S6C Fig). These results show that amino acid correlation patterns are not simply recapitulating the importance of residues for protein stability and could highlight distinct constraints that cannot be captured by stability predictions or number of structural contacts.
MHC-I molecules are known to interact with many peptides and the presence of a peptide is required for MHC-I folding. To explore the effect of the presence of peptide ligands on DCA predictions, we built an expanded version of DCA, called DCApeptides, that can take as input several peptide ligands for each protein sequence. The set of peptides interacting with a given protein are used to compute the single and paired frequencies used in DCA, as described in Materials and Methods. Although major efforts have been invested in the field to experimentally characterize the MHC-I binding specificity repertoire in human and mice [33–36], the vast majority of MHC-I molecules do not have experimental ligands. To fill this gap, we selected 100’000 random 9-mer peptides from several organisms and evaluated the predicted binding affinity of MHC-I sequences to each of these peptides using NetMHCpan3.0 [37] (see Materials and Methods). For each MHC-I sequence we then selected the top 2% of the peptides, following the cut-off currently suggested by the authors of NetMHCpan [37]. These predicted ligands were included in the co-evolution calculations using the DCApeptides algorithm. Overall, results did not change much and we still observed the decoupling between co-evolving and polymorphic sites (Fig 4). However, it should be noted that these are predicted ligands and the signal captured by DCApeptides reflects at best what is implicitly modelled in the predictor and not necessarily the real inter-molecular constraints.
To further explore the DCApeptides algorithm in the case of experimental ligands, we restricted the study to human MHC-I alleles having experimental ligands in IEDB [36] (see Materials and Methods). The number of such alleles is much smaller (156) and, as expected, we did not observe good structural contact predictions (Fig 5A). However, when restricting the analysis to inter-molecular pairs, we observed that the top 4 inter-molecular DCA pairs mapped accurately to existing structural contacts (Fig 5B). Moreover, these 4 pairs of sites involved residues P2 and P9 in the MHC-I ligands, which are known to be the main specificity determining residues (so-called anchor residues). Overall, our results indicate that DCApeptides predictions are stronger among MHC-I residues then between MHC-I residues and their ligands. However, DCA predictions among MHC-I residues do not pinpoint structural contacts (as in Fig 2C), while DCA predictions between MHC-I residues and their ligands revealed known interactions.
We further extended our benchmarking of the DCApeptides algorithm to the human PDZ protein domains, which are also known to interact with several ligands (in our dataset, these ligands came from a phage display experiment [38], see Materials and Methods). Here as well, we observed stronger correlation among the PDZ domain residues (S7A Fig). Some of the DCA predictions mapped to known structural contacts (15/27). More interestingly, when focusing only on correlations between PDZ residues and their ligands, we saw that DCApeptides could accurately predict some of the contacting pairs of residues. In particular, the top 2 predictions involved both position -2 in the PDZ ligands (S7B Fig), which is known to be the main specificity determining position for PDZ ligands [39]. Altogether, our results suggest that, when focusing on domains with available ligands from one species, intra-molecular DCApeptides predictions are not able to identify residues in structural proximity (likely because of the much lower number of sequences imposed by the constraint of having experimental ligands available), but inter-molecular DCApeptides predictions can accurately pinpoint structural contacts.
Co-evolution analyses have been widely used in biological studies, focusing mainly on co-evolution across species [14,40]. To our knowledge, our work is the first co-evolution analysis of a protein family that displays at the same time high variability between species and high polymorphism within species. As MHC-I polymorphism is known to be functionally important to entail different alleles with a wide range of binding specificities, our observation that polymorphic sites tend on average to show less co-evolutionary constraints may reflect the importance of preserving high mutability of these sites. It is also interesting to note that the de-coupling between polymorphic sites and co-evolving sites was even stronger than between polymorphic sites and sites involved in protein stability (Fig 3), suggesting that co-evolution constraints captured by DCA may be especially detrimental for polymorphic sites.
To predict co-evolving sites within MHC-I molecules, we used the DCA model introduced in [12,23], [22]. DCA demonstrated its statistical power on protein domains for which many homolog sequences are available (typically >10’000 sequences, ideally spanning both eukaryotes and prokaryotes) [22]. This study demonstrates that DCA predictions are highly enriched in structural contacts in MHC-I protein family, although the number of species is restricted to 445 (Fig 1). As in all DCA analyses, we focused here on sites that are distant in the sequence (i.e., more than 4 amino acids apart), which ensures that predictions of structural contacts are not simply resulting from sequence proximity. As such our work suggests that polymorphic sites tend to show less co-evolutionary constraints with sites distant in the primary sequence. Importantly, polymorphic sites have similar numbers of structural contacts with residues distant in the sequence (S8 Fig) as other residues, and therefore the observations made in this study could not simply be explained by the absence of such contacts.
The co-evolution signal detected in our analysis likely comes from the presence of divergent vertebrate species in the dataset, since very similar predictions were obtained by excluding the 20’256 human sequences in the datasets (Fig 2C and 2D), or by excluding species with more than 500 sequences in the dataset (S3 Fig). We anticipate that the fast evolutionary dynamic of MHC-I proteins may contribute to generating a stronger co-evolutionary pattern compared to other protein families, which could explain why we were able to detect it, although the MHC-I family is restricted to vertebrates.
DCA does not consider the actual phylogeny and takes only the alignment of sequences as input [14,18]. However, MHC-I evolution is difficult to characterize especially because it was subject to several duplication events along vertebrate evolution. Moreover the rate of evolution and the role of MHC-I in the immune system differ from one vertebrate species to another [41–43]) making it even more challenging to use available phylogenetic-dependent methods to predict co-evolving constrained sites since these models assume a homogeneous rate of substitutions across species evolution.
Ligands binding to MHC-I molecules play a role in MHC-I binding stability, which is why we included the ligands in stability predictions based on HLA-A02:01 structure. In vivo, MHC-I molecules are known to interact with tens of thousands of different peptides [33,44] and their specificity cannot be summarized with one single peptide. This is the reason why we extended the DCA framework to consider multiple ligands per protein in the alignment (Fig 4). Unfortunately, due to the scarcity of experimentally determined MHC-I ligands in most species except for human and mouse, the co-evolution analysis could not be carried out only with experimental ligands for all alleles included in our dataset. We therefore used for each allele 2’000 predicted ligands corresponding to the top 2% of a set of 100’000 peptides randomly selected from different proteomes [37]. As such, it is likely that the inter-molecular co-evolutionary signal observed in Fig 4 only captures the signal that is present in the NetMHCpan predictor, and may therefore not capture signals coming from more distant species that are not included in the training set of this algorithm. Nevertheless, he fact that the decoupling between polymorphic and co-evolving sites was observed both without and with ligands suggests that our results do not depend significantly on the presence of ligands in our analyses.
Our extension of the DCA algorithm to consider multiple ligands of the same protein further enabled us to analyse inter-molecular co-evolution for both MHC-I and PDZ proteins with experimentally determined ligands. Remarkably, in both cases, the inter-molecular predictions pinpointed structural contacts, whereas the intra-molecular predictions did not (for the majority of them, at least). Similar results were recently reported in a study of Antibody-antigen interactions [45], where maximum-entropy models such as DCA could help predicting affinity between antigens and antibodies, but not structural contacts within antibodies. We anticipate that our extension of DCA (available at: https://github.com/GfellerLab/DCApeptides) will contribute to future analyses of the differences between inter- and intra-molecular amino acid co-evolution patterns.
MHC-I molecules have emerged recently in life history and are mainly restricted to vertebrate species. Despite the limited number of species that contain MHC-I genes, we observed that co-evolution constraints identified by statistical methods such as DCA accurately predicted several structural contacts. Moreover, we found that the co-evolution signal was dominated by inter-species amino acid changes and was not due to the variations between alleles within the same species (e.g., human). To our knowledge, this work is the first co-evolution analysis of a protein family that displays at the same time high variability between species and high polymorphism within species. Finally, our results suggest that MHC-I polymorphic sites, in addition to providing distinct binding specificities, preferentially avoid residues that show either high amino acid co-evolution patterns or play an important role in protein stability.
In this study, we analysed the PFAM domain family named Histocompatibility antigen, domains alpha 1 and 2 of class I with the identifier PF00129. In PFAM v30 the domain family was composed of a total of 40’856 protein sequences [2]. We removed 117 bacterial and viral sequences from the dataset and kept only vertebrate MHC-I for a total of 40’739 sequences. The human sequences constitute 49.7% of the family followed by the Rhesus macaque sequences that represent 4.9% of the family (Fig 1). We filtered highly gapped columns (>70%), and the final alignment corresponds to positions 2 to 179 in HLA-A02:01 allele (residue following the numbering in the crystal structures such as PDB:2BNR chain A).
We further collected the most frequent human alleles in the allele frequency database [46] for USA NMDP European Caucasian population (comprising a total of 1,242,890 individuals). 331 alleles had a frequency exceeding 0.00001 (97 HLA-A, 181 HLA-B and 55 HLA-C alleles).
We used Direct Coupling Analysis (DCA) model [12] for the intra-molecular analysis of co-evolving sites within MHC-I domain family alignment. DCA uses as input the frequency fi(A) of amino acid A in column i, the frequency fj(B) of amino acid B in column j, and the joint frequency count fij(A,B) for pairs of amino acid A and B in columns i and j within a protein alignment, for all pairs of position i and j. These frequencies are computed including reweighting of sequences with >80% sequence identity and pseudo counts equal to the effective number of sequences after reweighting, as described in [12]. The sum of weights displayed in Fig 1 for each clade corresponds to the sum of ‘ma’ values, where ma represents to the weight of sequence a (see Morcos et al. [12]), and can be interpreted as the effective number of sequences in this clade. Julia’s version of PlmDCA [26][25] was run on the same alignment with default parameters. The algorithm starts by removing the duplicate sequences. Once these sequences were removed PlmDCA analysed 22954 sequences, with an effective number of sequences Meff equal to 173.44.
As a reference structure for MHC-I domain, we used the structure of HLA-A02:01 in complex with a canonical 9-mer peptide (PDB: 2BNR; [47]). We consider that two sites are close in the structure if the distance between any of the heavy atoms is smaller or equal to 8Å, as suggested by the authors of the original DCA study [12], and built the contact map (grey dots in Fig 2). Similar contact maps were built using cut-off of 5Å in S2 Fig. To analyse the predictions of DCA with respect to structural contacts, we only considered pairs distant in the sequence (over 4 amino acids apart) and displayed in the contact maps of Fig 2 the top 44 predictions (25% of the MHC-I domain length). The performance plot in the insets were computed as follows:
DCA provides a score for every pair of sites. To reflect whether a site is under a co-evolutionary constraint we first ranked the scores in a decreasing order. We iteratively attributed individual score for each site as follow:
For human sequences in the PFAM alignment, we used one minus the Shannon entropy (i.e., 1+∑A=120fi(A)log{fi(A)}/log{20}, where fi(A) stands for the frequency of amino acid A at position i) to measure the polymorphism score at each position [48]. This score has a minimal value of zero when all amino acid frequencies in a site are equal and a maximal score of one when only one perfectly conserved amino acid is found at a given position. We omitted the gaps from the entropy measure. The polymorphism analysis was also performed using only the most frequent human MHC-I sequences (331 alleles, see before). To this end the human alleles were aligned with MUSCLE [49] and amino-acid to compute the Shannon entropy were weighted by the allele frequency in the USA NMDP European Caucasian population.
To evaluate the structural stability impact of each residue, the AlaScan function of the FOLD-X software [10,27] was used to calculate the energy contribution of each residue. The structures were first repaired using RepairPDB function. The stability score of each site was measured using a reformatted pdb structure of 2BNR [47] where MHC-I residues from position 1 to 179 and the ligand were merged on chain A.
The number of contacts of each site was measured using the pdb structure 2BNR (HLA-A02:01 allele in chain A and the ligand). For a given site, the number of contacts is the number of residues that are maximum 5Å distant from this site in the crystallized structure.
Enrichment Analysis was used to investigate the relationship between polymorphic sites and sites displaying strong co-evolution constraints as estimated by DCA. A site was considered to be non-polymorphic in human alleles when its polymorphism score was lower than a threshold of 0.01 (see S4 Fig for results with other thresholds). To compute enrichment curves, sites were ranked based on their DCA score (x-axis in lower panels of Fig 3). Whenever a non-polymorphic site is encountered along the ranking (yellow bars), the enrichment curve goes up. Whenever a polymorphic sites is found the enrichment curve goes down. The same enrichment analysis was also applied to investigate the relationship between polymorphic sites involved in structural stability or sites displaying many contacts in the crystal structure of HLA-A02:01. For the enrichment analysis and p-value calculations, we use a weighted version of the Kolmogorov-Smirnov statistic with exponent measure equal to 1, as in all standard enrichment analyses [50].
To model the existence of multiple (predicted) ligands for each MHC-I protein, the amino acid frequencies fi and fj for all sites and joint frequencies fij for all pairs of sites (i.e. including both sites in the MHC and sites in the ligands) were computed. Following the nomenclature used in [12] the point frequency for position i in ligand is computed as:
fi(A)=1Meff+λ(λq+∑a=1M1ma∑n=1Na1NaδA,Li,na)
where Li,na stands for the ith amino acid in the nth ligand of protein a, and Na stands for the number of ligands of a and M stands for the number of MHC-I sequences. The joint frequency between position i in the protein and position j in the ligand is computed as:
fij(A,B)=1Meff+λ(λq2+∑a=1M1maδA,Aia∑n=1Na1NaδB,Lj,na)
Where Aia stands for the ith amino acid in protein a. Finally, the joint frequency between two ligand positions (i and j) is computed as:
fij(A,B)=1Meff+λ(λq2+∑a=1M1ma∑n=1Na1NaδA,Li,naδB,Lj,na)
The sequence reweighting (ma) corresponds to the number of sequences with more than 80% sequence identity to protein a, and was computed considering only the MHC-I sequence identity. This implies that each ligand has a weight equal to the weight of its protein (1ma) divided by the number of ligands of this protein (Na), in order to ensure proper normalization. The same pseudo-count λ=Meff=∑a=1M1ma was applied as in the standard DCA. In the case of 9-mer MHC-I ligands, this resulted in a total alignment of 178+9 = 187 positions, where the first 178 positions are characterized by a single amino acid at each position, while the last 9 positions are characterized by a distribution of amino acids for each MHC-I and each position in the ligands. All the rest of the DCA algorithm remains the same (inversion of the (187*20) x (187*20) covariance matrix and estimation of the Direct Information scores). The script to run these calculations can be accessed at: https://github.com/GfellerLab/DCApeptides.
To explore the impact of MHC-I ligands on the enrichment analysis of Fig 3A, we attempted to run DCApeptides on the full alignment, including multiple peptide ligands for each MHC-I protein. Since the MHC-I ligand repertoire for the vast majority of MHC-I alleles in different species is still not experimentally available, we generated 100’000 random 9-mer peptides from 7 proteomes (Anguilla anguilla, Bos taurus (Bovine); Gallus gallus; Homo sapiens (Human); Larimichthys crocea; Mus musculus (mouse); Tinamus Guttatus) and predicted the binding affinity of MHC-I alleles to each of these peptides using NetMHCpan3.0 [37]. We then selected the top 2% predictions for each MHC-I allele in our alignment and computed the co-evolution patterns including these ligands based on DCApeptides (see above). Only MHC-I sequences without gaps at binding site positions used in NetMHCpan3.0 were considered (27,373 MHC-I sequences in total).
Experimental MHC-I ligands were retrieved from IEDB [36]. In total 156 human MHC-I alleles had experimental ligands (annotated as “Positive-High”, “Positive-Intermediate”, “Positive-Low” or “Positive”). Only 9-mers were considered and these ligands were used with DCApeptides. X-ray structure of HLA-A02:01 (PDB:2BNR) in complex with a 9-mer peptide was used to compute the contact maps of Fig 5.
Experimental PDZ ligands were retrieved from a large phage display screen performed for 54 human PDZ domains [38]. All ligands were aligned at their C-terminus. The contact map in S7 Fig was computed based on the X-ray structure of DLG2 (PDB: 2HE2) [51].
|
10.1371/journal.pntd.0001203 | The Effect of Egg Embryonation on Field-Use of a Hookworm Benzimidazole-Sensitivity Egg Hatch Assay in Yunnan Province, People's Republic of China | Current efforts to control human soil-transmitted helminths (STHs) involve the periodic mass administration of benzimidazole drugs to school aged children and other at- risk groups. Given that high levels of resistance to these drugs have developed in roundworms of livestock, there is a need to monitor drug efficacy in human STHs. The current study aimed to evaluate an in vitro egg hatch assay for measuring the sensitivity of human hookworms to benzimidazole drugs in an isolated field setting in southern Yunnan province, People's Republic of China.
Egg hatch assays were performed with hookworm (Necator americanus) eggs extracted from 37 stool samples received from local school-aged children. The mean IC50 was 0.10 ug/ml thiabendazole (95% CIs: 0.09–0.12 ug/ml). Observation of the eggs immediately prior to assay set-up revealed that a small percentage had embryonated in some samples. Scoring of % embryonation of eggs prior to the assay allowed for corrections to be made to IC50, IC95 and IC99 values. Examination of the data with and without this correction revealed that the embryonation of a small number of eggs did not affect IC50 values, but did increase IC95 and IC99 values for some samples.
This study has highlighted the impact of egg embryonation on the use of benzimidazole drug sensitivity assays for human hookworms in field settings. Given the greater flexibility required in human stool collection procedures compared to livestock studies, we suggest that embryonation of some eggs may be an unavoidable issue in some human studies. Hence, it needs to be measured and accounted for when analysing dose response data, particularly for generation of IC95 and IC99 values. The protocols used in this study and our suggested measures for accounting for egg embryonation should have widespread application in monitoring benzimidazole sensitivity at field sites worldwide.
| With the implementation of mass drug administration programmes for the control of human soil transmitted helminths there is a need to develop drug sensitivity monitoring tools to detect the emergence of resistance. The present study aimed to use an egg hatch assay to measure benzimidazole sensitivity in human hookworms in a field setting in Yunnan province, People's Republic of China, in order to assess whether the assay offered a practical means of monitoring drug sensitivity in human hookworms in such a location. The assay proved able to generate dose response data, which allowed for the drug sensitivity of the hookworms in the local children to be described; the mean IC50 was 0.10 ug/ml thiabendazole. The study also found that practical issues associated with stool collection procedures, specifically the embryonation of some eggs during the time elapsing between stool deposition and egg recovery, can have an impact on the drug sensitivity data. We suggest means for data analysis that overcome the impact of egg embryonation on drug dose response data, which should allow for the use of such assays at different field sites worldwide.
| Periodic mass administration of the benzimidazole drugs albendazole or mebendazole to school-aged children and other at-risk groups is the mainstay of all current programmes to control soil transmitted helminths (STHs) in humans [1]. The massive scale and increasing frequency of anthelmintic treatment mean that it is essential to monitor drug-exposed worm populations to ensure that any drug resistance is detected should it emerge. Early detection is a prerequisite for the implementation of mitigation strategies such as drug rotation to ensure that the effectiveness of the few existing anthelmintic drugs is preserved for as long as possible.
The detection of anthelmintic drug resistance has been the subject of much attention in the livestock area [2], [3]. The faecal egg count reduction test (FECRT), in which faecal egg counts are conducted before and after drug treatment to detect any reduced drug efficacy indicative of drug resistance, is the most readily available test that can be adapted for the human field. It is currently being assessed by the WHO as a drug resistance monitoring tool. However, this test suffers from a lack of sensitivity, and its performance depends on pre-treatment egg counts (infection intensity) and possibly density dependent fecundity [4], [5].
In vitro phenotypic assays and molecular biology-based tests have been studied extensively in livestock nematodes. An egg hatch assay has been described for measuring resistance to benzimidazole drugs [2], [3], and may therefore be applicable for the detection of benzimidazole resistance in human hookworms, the only common STH to hatch outside the human body. Molecular tests monitoring changes in beta tubulin genotypes have also been proposed [3], [6]. However, even in certain well–studied livestock parasite species the relationship between benzimidazole drug efficacy and genotype is not fully understood [3], and the importance of beta tubulin SNPs in benzimidazole sensitivity in human STHs has not yet been demonstrated. Hence, there is a need to develop and utilise phenotypic tests which examine the direct effects of a drug on the free living stages of the human STH in vitro until sensitive molecular tests become available.
The use of egg hatch assays for measuring benzimidazole sensitivity in human STHs in field sites has been reported by several groups [7], [8]. Recently, Kotze et al. [9] described a standardised egg hatch assay for human hookworms in a 96-well plate format using a concentration gradient of thiabendazole embedded in agar. The present study aimed to test this assay format at a field site in the Peoples Republic of China (P.R. China), and to examine logistical and technical issues such as sample collection, handling and storage, egg isolation and test evaluation which may have an impact on the use of the assay in the field.
The present study is an integral part of an on-going project for helminth infection surveillance among schoolchildren implemented by the Yunnan Institute of Parasitic Diseases in collaboration with educational authorities. It focuses on the epidemiology and control of intestinal helminth infections in this province in southwest P.R. China. The project has been approved by the Academic Board (Ethics Committee) of the National Institute of Parasitic Diseases, Chinese Center for Disease Control and Prevention in Shanghai, P.R. China. Informed consent to conduct the present study was given by local school authorities who informed the eligible students in the presence of the responsible personnel from the Yunnan Institute of Parasitic Disease. As the children board at the school during the week, and are under the responsibility of the school, this information session was not done in the presence of the parents/guardians. The students were informed about the study aims, procedures and potential risks and benefits, and were alerted to the possibility to withdraw from the study at any time without further obligation. Hence, assent to participate in the study was indicated by subsequent submission of a stool sample. By withdrawing, children did not forfeit their right to anthelminthic treatment at study completion. This procedure, with consent provided by school authorities and by the study subjects through choosing to participate, with no requirement for individual written informed consent, is in line with national and local standards for such studies which are based on diagnosis with no invasive procedures, and was approved by the ethics committee. A single standard dose of albendazole (400 mg) was provided free of charge to all study participants and their classmates after collection of stool samples.
Stool samples were collected among students of a primary school located in a suburb of Pu'er city, in southern Yunnan province, P.R. China (Figure 1). Labeled (identification number and name) stool collection containers were handed out to students in the afternoon, and students were advised to keep filled containers shaded and at room temperature. Filled containers were collected the next morning, and transferred to the local branch of the Yunnan Institute of Parasitic Diseases (YIPD), where all subsequent manipulations were performed, within 1 h. In the laboratory and during analyses, only the ID number was used to identify samples; a key linking ID numbers to names was retained by the YIPD.
Hookworm eggs were recovered from individual samples rather than pooling faecal samples in order to determine the range of responses in samples from within the same population. The presence of hookworm eggs in the collected samples was determined using the Kato-Katz method [10]. Whenever stool samples could not be processed for egg recovery on the day of collection, about 20 g of stool was suspended in ca. 40 ml water to arrest the development of hookworm eggs (anaerobic storage) [11], and the solution stored at room temperature.
The protocol for egg recovery was adapted from Kotze et al. [8]. About 20 g of the hookworm-positive stools, or whatever was available in case the sample was smaller, was mixed with water 1∶2, homogenized using a spatula and poured through a tea strainer. The runoff was transferred to 45 ml centrifuge tubes. The tubes were topped up with tap water, shaken carefully, and centrifuged in a bench top centrifuge for 2 minutes 30 seconds at 300 g. The supernatant was then poured off, the tube filled with saturated saline solution, and the sediment loosened using a wooden stick. Subsequently, the tube was centrifuged for 2 minutes 30 seconds at 130 g, and then left to stand for 5 minutes. The top 2 cm of the saline solution column were then transferred to a 15 ml tube using a disposable pipette. The 15 ml tube was filled with tap water and centrifuged for 2 minutes 30 seconds at 300 g, and the supernatant carefully removed using a disposable pipette. The tube was filled again with tap water, shaken and centrifuged as before. The supernatant was then removed leaving about 3 ml liquid above the sediment, and the sediment re-suspended.
Aliquots of 20 µl of the egg suspension were then placed on a slide and the number of hookworm eggs counted. Based on these counts, the egg concentration was adjusted to approximately 2 eggs per µl egg solution, or the highest possible concentration in cases where fewer eggs than required to achieve the target concentration had been recovered.
It was noted after the study had commenced that unusually high numbers of larvae were present in some drug assay plate wells with high drug concentrations. Subsequent observation of egg samples immediately following extraction from faeces revealed that some eggs had embryonated (embryos visible within the egg shell). Hence, from that time onwards the percentage of eggs that had embryonated prior to assay set-up was measured. A subsample of eggs was taken, and each egg was scored as being either embryonated (that is, with a larval shape visible within the egg) or not (that is, in a multicell stage). The mean number of eggs in the samples examined for embryonation (± SE) was 70±9. The number of samples for which embryonation was measured was 36. Only the egg hatch dose response data from these samples was analysed for the present study.
A PCR test designed to discriminate between Necator americanus and Ancylostoma duodenale was performed on a sub-sample of the recovered hookworm eggs conserved in ethanol and analysed according to the protocol published by Zhan et al. [12]. For DNA extraction, the DNEasy blood and tissue kit produced by Qiagen (Hilden, Germany) was used according to the manufacturer's protocol.
Assay plates were prepared as described by Kotze et al [9]. A stock solution of thiabendazole (10 mg/ml) was prepared in DMSO, and diluted 17.2-fold in DMSO to give a solution of 0.58 mg/ml, which was then serially-diluted 2-fold to produce a further 9 drug concentrations. Aliquots (2 µl) from this series of dilutions (starting at 0.58 mg/ml) were added to 96-well microtitre plates, such that each row of the plate comprised a gradient of ten dilutions. The first two wells of each row were utilized as control wells (i.e. received 2 µl of DMSO only), the 3rd well contained the lowest drug concentration, and the 12th well the highest. Each drug concentration was present in all 8 wells within each column of the plate. 200 µl of 2% agar (Davis Gelatine Co., powdered agar Grade J) was dispensed into each well of the plate and allowed to set. Thus, the concentration of thiabendazole across the plate ranged from 0.01 to 5 µg/ml (after addition of 30 µl of egg solution as described below). A piece of absorbent cloth soaked in an amphotericin B solution of 2.5 µg/ml was placed on top of the plate lids, and each plate was placed into a plastic press-seal bag and stored at 4°C. A box of plates was shipped at room temperature to the field site. The plates were refrigerated on arrival.
Aliquots of each egg solution (30 µl) were added to all wells in duplicate rows, or if insufficient egg solution was available, only every second well was loaded. The assay plates were incubated at 28°C for 48 hours before all larvae were killed by addition of 10 µl Lugol's iodine to each well. The number of larvae present in each well was then counted either by direct observation of the well under a dissecting microscope, or if too much debris or fungal/bacterial growth was present, the contents of the well was collected using a disposable pipette and transferred onto a microscope slide. In the latter case, both the empty well and the microscope slide were examined, and all larvae counted.
As the use of IC95 and IC99 values has been advocated for livestock nematodes as more sensitive measures of changes in drug response in worm populations than IC50 values (for example, Coles et al. [3]), we examined the data in terms of IC50, IC95 and IC99 values. For each sample, the mean number of larvae present in the 4 control (no drug) wells was calculated. The number of larvae present in each drug well was then expressed as a percentage of the control mean. Data was analysed by non-linear regression using GraphPad Prism software in order to generate IC50, IC95 and IC99 values, representing the drug concentrations which inhibited egg hatch (reduced numbers of larvae present in wells) by 50%, 95% or 99% relative to control wells.
Dose response data were analysed before and after the application of a correction for the % embryonation which had been measured prior to assay set-up for each egg sample. This correction was done for each sample of eggs based on the % embryonation measured for that sample. Firstly, the number of eggs embryonated in each sample was expressed as a percentage. The mean number of eggs in the dose response assay control (no drug) wells was then corrected by removing the number of larvae that would be expected to have been derived from embryonated eggs added to those assay wells when the assay was established; for example if a mean of 80 eggs were present in control wells for a sample which had been shown to be 15% embryonated prior to assay set-up, then the control mean would become 80−(15/100×80) = 68 eggs. Similarly, the numbers of larvae present in each drug assay well was also corrected by subtraction of the number of larvae expected to have been derived from embryonated eggs in that specific egg sample ( = 12 in this example). The corrected drug assay data point was then expressed as a percentage of the corrected control mean.
Corrected and uncorrected dose response data were analysed in two ways: firstly, by pooling the % egg hatch values for all 36 samples at each drug concentration to generate a single dose response for the uncorrected and corrected data sets (and, hence single IC50, IC95 and IC99 values for each data set); and, secondly by generating separate dose responses (and hence, separate IC50 and IC95 values) for the data derived from each study subject's egg sample.
The PCRs designed to discriminate between N. americanus and A. duodenale indicated the presence of only the former species in the egg samples. This is in agreement with the known predominance of N. americanus in the study population (Steinman, unpublished data). Eggs extracted from 37 faecal samples were examined in egg hatch assays. One of these assays was not included in the subsequent analysis as less than 20 larvae were present in the control (no drug) wells of the assay plates at the end of the incubation period. Hence, a total of 36 separate assays were analysed by non-linear regression.
The % embryonation measured in all samples prior to assay set-up is illustrated in Figure 2. The mean % embryonation was 6.3%, ranging up to 26% in one sample. The number of samples in which at least one egg had embryonated was 23 out of the 36.
Initially, we pooled the assay data to examine the overall population dose responses using data either corrected for embryonation or uncorrected (Figure 3, Table 1). Sigmoidal dose response relationships were apparent (Figure 3A), with the responses for both data sets showing very little difference at either the IC50 or IC95 levels, with overlapping 95% CIs (Table 1). Both data sets showed a plateauing of the response at the highest drug concentrations (Figure 3B). As a consequence, the uncorrected data set did not decrease to an IC99 level, while the embryonation-corrected data set did decrease to a level which allowed for the calculation of an IC99 (Figure 3B, Table 1). However, the egg hatch in this latter data set did not reach zero even at the highest thiabendazole concentration of 5 ug/ml.
The relationship between egg hatch and drug concentration in egg samples recovered from individual subjects is illustrated with 4 examples in Figure 4. In each case a plateau was present in the uncorrected data sets, with the % egg hatch at levels of 5–25% at the highest drug concentrations. After correction for embryonation, two effects were apparent: in A and B, the level of the plateau was reduced, but a plateau was still present; in C and D, the plateau was removed, and egg hatch was reduced to zero at the highest concentrations. In all cases the curves showed little change with regard to the IC50 point following correction for embryonation.
The variation in IC50 and IC95 values among the different individual faecal samples is illustrated in Figure 5. IC50 values showed a similar range (approximately 5-fold) across both data sets, and mean IC50 values were not significantly different before and after embryonation correction (paired t-test, P = 0.33). A comparison of the IC95 values for the uncorrected and corrected subsets in Figure 5B illustrates the effect of applying the egg embryonation correction in reducing the number of samples for which an IC95 could not be calculated (that is, IC95>5 ug/ml). However, even after the correction an IC95 could not be calculated for two samples. These two outlying IC95 values were derived from samples showing embryonation rates of 14 and 8%.
The present study has generated baseline data for drug sensitivity of human hookworms (overwhelmingly N. americanus) in a field setting in southwest P.R. China, and raised a number of important issues if such assays are to be standardized for widespread use. The IC50 for the pooled data (0.10 ug/ml, 95% CIs 0.09–0.12) was similar to that reported previously for N. americanus in Papua New Guinea by Kotze et al [8] (0.076 ug/ml) and in Pemba Island, Zanzibar, by Albonico et al. [7] (0.079 ug/ml), indicating a degree of consistency in IC50 determinations by such assays in quite different field settings.
The embryonation of some eggs is an important issue that has not previously been reported. Egg hatch assays in the livestock sphere are performed on eggs isolated from fresh samples (<3 hours old at the commencement of the egg extraction procedures), or, if this is not possible, the faeces is mixed with water and sealed in tubes to generate anaerobic conditions until egg extraction can commence [2], [11]. Le Jambre [13] compared egg hatch assays using embryonated and unembryonated eggs of Haemonchus contortus. The IC50 was approximately 10–20 fold higher if embryonated eggs had been added to assays than if unembryonated eggs had been used. In the present study, a degree of embryonation was observed and measured in some samples. Significantly though, this did not affect the pooled data IC50 and IC95 values (from Figure 3 and Table 1), or the mean IC50 values derived from corrected and uncorrected data set from individual study subjects (Figure 5).
The embryonation of some eggs did, however, have a significant effect on some individual dose responses, in particular the IC95 and IC99 values for these samples (from Figures 4 and 5), as well as the IC99 for the pooled data (Figure 3B and Table 1). The presence of a small proportion of embryonated eggs manifested itself as a plateau in the dose response curves. This was most likely due to the shorter period of drug exposure prior to hatching for these eggs compared to the majority, hence allowing them to hatch at drug concentrations that would otherwise have been lethal. Although the levels of embryonation observed here (mean of 6.3%) did not affect the IC50 and IC95 for the pooled data (as described above), the illustration of the impact of the egg embryonation on individual cases in Figure 5, and its effect on IC99 (from Table 1), and the data described above from Le Jambre [13], indicate that it could potentially have significant effects on IC50, IC95 and IC99 values if it occurred at higher rates and / or in a greater proportion of individual samples than seen in the present study.
Hatching of eggs at high drug concentrations could either be indicative of the presence of a small proportion of worms able to resist the effects of the drug, or simply the presence of a degree of egg embryonation in the original sample. Hence, if embryonation was solely responsible for the dose response plateau then the correction we applied to the data may be expected to remove it. While this occurred in some cases (eg. Figure 4C and D), it was not achieved in others (Figure 4A and B), and also did not occur for the pooled data (from Figure 3B). This is likely due to experimental error rather than the presence of a real dose response plateau unrelated to embryonation. Our scoring of samples to obtain a % embryonation figure was based on a single sample of eggs for each faecal preparation (mean sample size ± SE = 70±9). Hence, this represents at best an estimate of the % embryonation in the samples. If this estimate was too low in a particular sample, its application to the data would not have removed the plateau in the dose response (as most likely occurred in Figure 4A and B). Hence, given the lessons learnt here, we suggest that a greater number of eggs are scored for embryonation at the time of assay set-up than was done for the present study. A sample size of at least 100 eggs may be suitable.
It would be clearly desirable to prevent embryonation prior to assay set-up when using egg hatch assays in field studies. However, this is not as easily achieved in human studies as in livestock surveys due to sampling constraints in the former. In the present study, faecal containers were handed out to children in the afternoon, and then collected the next morning. The faeces was then immediately analysed with the Kato-Katz method and, if found hookworm positive, covered with water and mixed well in order to prevent further egg development. Egg extraction then took place from approximately noon till mid-late evening. At this field site it would be difficult to reduce the time between stooling and the mixing of the samples with water in the laboratory since the local children do not stool very often, possibly associated with their low-fibre diet. Hence, it is not possible to only collect stools that had been deposited in the morning if a representative population sample is required. In other field settings, for example Kyrgyzstan [14], it would be quite easy to ensure the freshness of samples as containers can be given out in schools in the morning and then collected for processing 2–3 hours later. Hence, given this difference in stooling habits in different field settings, it may be difficult to standardise this aspect of the assay across all sites worldwide.
We advise that rather than try to standardise on a less desirable but more universal method that could be applied in all sites (that is, an overnight stooling period along with an acceptance of a degree of embryonation in some cases), every effort is made to ensure that samples are as fresh as possible, even if this means that different methods are applied at different field sites. That is, in areas where it is possible, containers should be distributed in the morning and collected again after 2–3 hours for processing. In other cases, an overnight stooling period would need to be accepted. Wherever freshness cannot be guaranteed, the % embryonation should be scored based on a sample of at least 100 eggs at the time of assay set-up, and all egg hatch scores corrected. In areas where fresh stool samples can be accessed, embryonation could be checked in some samples, but not necessarily in all. When stool freshness is assured, confidence can be placed in IC50, IC95 and IC99 values. Where stool freshness cannot be assured, and % embryonation corrections are required, IC50 values can be regarded with confidence, however, IC95 and IC99 values need to be judged carefully. Where high IC95 and IC99 values are observed, repeat samples could be collected.
There was a degree of variability between IC50 and IC95 values over the separate assays. The variability in IC50s from embryonation-corrected assays using different samples of worm eggs amounted to a 5.3-fold range. Kotze et al [8] found a 4.1-fold range in IC50 values among samples from individual subjects in a Papua New Guinean village. The range for the data of Albonico et al [7] was not reported, but the 95% CIs from that study were similar to those for the pooled data dose response in the present study.
The extraction of eggs from stools is a laborious process. In the present study, approximately 15 samples could be processed by 2 people working full-time in the laboratory every day, aided by another 2 technicians who performed the Kato-Katz test. The study of individual faecal samples allowed us to look at the variability across separate assays. However, such a procedure would not be necessary for a more general sensitivity-monitoring exercise. Hence, for such studies, a degree of pooling of faecal samples is recommended. We suggest that an approximately equal weight (or volume) of faeces from 5–10 individuals (depending on study size) could be pooled. Such a strategy has been applied previously by Albonico et al [7] who pooled faecal samples from groups of 10 children.
In conclusion, this study has once again shown that it is possible to measure drug sensitivity using an in vitro assay in a field site with limited laboratory equipment. Difficulties associated with the recovery of hookworm eggs (most significantly the freshness of the stools) are not prohibitive, but need to be accounted for during test performance and as part of the data analysis. The protocols used in this study, and our recommendations concerning the pooling of samples and the accounting for egg embryonation, should have widespread application. Although the ability of the egg hatch assay to detect benzimidazole resistance in human hookworms has not been proven (in the absence of known resistant populations), its utility for the detection of benzimidazole resistance in livestock parasites [2], [3] should generate a degree of confidence that it will also be applicable in the human sphere. There is therefore a need to apply these assays widely in order to obtain baseline data for drug sensitivity in different human hookworm populations so that changes associated with the emergence of drug resistance may be detected, and to compare drug-naïve populations with those already exposed to repeated drug treatments.
|
10.1371/journal.pgen.1000984 | Inactivation of hnRNP K by Expanded Intronic AUUCU Repeat Induces Apoptosis Via Translocation of PKCδ to Mitochondria in Spinocerebellar Ataxia 10 | We have identified a large expansion of an ATTCT repeat within intron 9 of ATXN10 on chromosome 22q13.31 as the genetic mutation of spinocerebellar ataxia type 10 (SCA10). Our subsequent studies indicated that neither a gain nor a loss of function of ataxin 10 is likely the major pathogenic mechanism of SCA10. Here, using SCA10 cells, and transfected cells and transgenic mouse brain expressing expanded intronic AUUCU repeats as disease models, we show evidence for a key pathogenic molecular mechanism of SCA10. First, we studied the fate of the mutant repeat RNA by in situ hybridization. A Cy3-(AGAAU)10 riboprobe detected expanded AUUCU repeats aggregated in foci in SCA10 cells. Pull-down and co-immunoprecipitation data suggested that expanded AUUCU repeats within the spliced intronic sequence strongly bind to hnRNP K. Co-localization of hnRNP K and the AUUCU repeat aggregates in the transgenic mouse brain and transfected cells confirmed this interaction. To examine the impact of this interaction on hnRNP K function, we performed RT–PCR analysis of a splicing-regulatory target of hnRNP K, and found diminished hnRNP K activity in SCA10 cells. Cells expressing expanded AUUCU repeats underwent apoptosis, which accompanied massive translocation of PKCδ to mitochondria and activation of caspase 3. Importantly, siRNA–mediated hnRNP K deficiency also caused the same apoptotic event in otherwise normal cells, and over-expression of hnRNP K rescued cells expressing expanded AUUCU repeats from apoptosis, suggesting that the loss of function of hnRNP K plays a key role in cell death of SCA10. These results suggest that the expanded AUUCU–repeat in the intronic RNA undergoes normal transcription and splicing, but causes apoptosis via an activation cascade involving a loss of hnRNP K activities, massive translocation of PKCδ to mitochondria, and caspase 3 activation.
| In an earlier study, we showed that the mutation of spinocerebellar ataxia 10 (SCA10) is an enormous expansion of a gene segment, which contains a tandemly repeated 5-base (ATTCT) unit. Since SCA10 is the only known human disease that is proven to be caused by 5-base repeat expansion, it is important to learn how this novel class of mutation causes the disease. We found that the mutation produces an expanded RNA repeat, which aberrantly accumulates in SCA10 cells and interacts with a major RNA–binding protein. When we expressed expanded RNA repeats or decreased the RNA–binding protein level in cultured cells, either of these manipulations produced a specific type of cell death that is associated with a massive transfer of a key enzyme called protein kinase C delta to mitochondria. We also showed that either blocking the expanded AUUCU repeat or replenishing hnRNP K rescues cells from the cell death induced by the SCA10 mutation. Together, we conclude that the mutant RNA inactivates hnRNP K and kills cells by triggering the specific cell-death mechanism. Our data provide important clues for therapeutic intervention in SCA10.
| Spinocerebellar ataxia type 10 (SCA10) is an autosomal dominant neurodegenerative disease presented with progressive pancerebellar ataxia, leading to total disability [1]–[4]. Approximately 60% of the SCA10 patients also suffer from epilepsy with complex partial seizures and generalized tonic-clonic seizures, which become life-threatening due to development of status epilepticus [3]–[5]. The disease-causing genetic mutation is a large (up to 22.5 kb) expansion of a pentanucleotide, ATTCT, repeat present within the ninth intron of the ATXN10 gene on chromosome 22q13.31 [6].
In the last two decades, investigators identified a group of diseases caused by expansions of short tandem repeats, also known as microsatellite repeats. Most of these mutations involve unstable trinucleotide repeats located in different regions of respective genes. The roles of repeat expansion mutations in the pathogenic mechanism of these diseases are diverse and complex [7], [8]. However, in a simplistic view an expanded repeat in the coding region produces an elongated tract of repetitive amino acid residues with a gain of toxic function at the protein level, whereas a triplet repeat expansion in 5′- and 3′-untranslated regions (UTR) may result in an altered transcription level of the gene or a production of toxic RNA transcript containing expanded ribonucleotide triplets. Friedreich's ataxia (FRDA) is the only known disease caused by an expansion of an intronic trinucleotide repeat. Typical FRDA mutations are large GAA repeats located in intron 1 of the FXN gene, which severely hinders the transcription of the FXN gene, leading to the autosomal recessive phenotype [9].
SCA10 and myotonic dystrophy type 2 (DM2) are only human diseases caused by non-trinucleotide microsatellite expansion mutations although an insertion of a large pentanucleotide repeat has recently been reported to be associated with SCA31 [10]. In DM2 the mutation is a large (up to 44 kb) expansion of CCTG tetranucleotide repeat in intron 1 of the ZNF9 gene. Thus, it is an interesting coincidence that non-trinucleotide mutations in DM2 and SCA10 are both large expansions located in an intron and causing autosomal dominant phenotypes. In DM2, expanded CCUG tetranucleotide repeat transcripts accumulate mostly in nuclear foci, and sequestrate the muscleblind like 1 (MBNL1) protein into the RNA foci [11], [12]. The resultant depletion of MBNL1 causes splicing dysregulation of a variety of RNA transcritpts similar to DM1. Splicing misregulation is thought to be the primary pathogenic mechanism in DM1 and DM2.
In SCA10 the number of ATTCT repeats ranges from 10 to 29 in normal individuals, and increases up to 4,500 in patients [13], [14]. The ATXN10 gene consists of 12 exons spanning 172.8 kb, and encodes ataxin 10, which contains two armadillo repeats known to mediate protein-protein interaction. Knock-down of ATXN10 by RNAi induces cell death in primary cerebellar neurons [15], whereas over-expression of ATXN10 activates the mitogen-activated protein kinase cascade and promotes neurite extension in PC12 cells [16]. While ATXN10 is expressed in a wide variety of tissues, expression is especially strong in brain, heart and muscle. Although these data suggest that ataxin 10 plays a role in neuronal survival and differentiation, the exact function of ataxin 10 remains unknown. Thus, it is plausible that a large expansion of the ATTCT repeat may interfere with the transcription, like the GAA repeat expansion does in FRDA, leading to a loss of function of ataxin 10. However, we recently demonstrated that neither a gain nor a loss of the function of ATXN10 is the primary pathogenic mechanism of SCA10 [17]. Analyses of SCA10 fibroblasts showed that the ATXN10 mRNA levels remain unaltered in spite of the repeat expansion [6], [17]. In addition, transcription of the mutant alleles and post-transcriptional splicing of the mutant ATXN10 transcript remain largely unaltered in SCA10 patients [17]. Furthermore, homozygous Atxn10 knockout (Atxn10−/−) mice showed embryonic lethality while heterozygous (Atxn10+/−) mice showed no phenotype [17]. Finally, a recent report described patients with balanced translocation t(2;22)(p25.3:q13.31), in which the breakpoint of chromosome 22q13.31 disrupted intron 2 of ATXN10 [18]. These patients were totally asymptomatic, suggesting that haploinsufficiency of ATXN10 does not cause SCA10.
In the present study, we examine whether the expanded AUUCU RNA repeat in the mutant ATXN10 transcript is the principal pathogenic molecule capable of triggering neuronal death in SCA10. We demonstrate that the expanded AUUCU repeat within the spliced intron interacts with hnRNP K, and this RNA-protein interaction results in loss of hnRNP K function, translocation of Protein Kinase C δ (PKCδ) to mitochondria and activation of apoptosis in SCA10 cells. Furthermore, we observe that targeted inactivation of the mutant ATXN10 transcripts in SCA10 cells significantly reduces mitochondrial translocation of PKCδ. Together, these results define a key pathogenic mechanism of SCA10 and provide clues for potential therapeutic strategies.
We propose that the mutant ATXN10 transcripts containing expanded AUUCU repeats contribute towards the SCA10 phenotype. To investigate whether the sub-cellular distribution and fate of the mutant ATXN10 transcripts are altered, RNA FISH analysis with a Cy3-(AGAAU)10 riboprobe was performed on SCA10 fibroblasts containing ∼2000 or ∼1,000 ATTCT repeats and on normal fibroblasts expressing the wild type ATXN10 transcripts containing 12 AUUCU repeats. The (AGAAU)10 riboprobe detected the presence of nuclear and cytoplasmic aggregates in SCA10 fibroblasts (Figure 1A; arrows, also Figure S1A; arrow), but not in normal fibroblasts (Figure 1B). These aggregates observed in this and other Figures were resistant to DNAse and disappear after RNAse treatment. Since our previous study showed that the 9th intron of the ATXN10 gene (66,421 bp) encoding the expanded AUUCU repeats is spliced normally [17], our present results imply that the intron 9 sequences are spliced and partly translocated to the cytoplasm in SCA10 fibroblasts. FISH with an anti-sense probe specific for exon 9 of the ATXN10 gene showed no significant binding in the same SCA10 fibroblasts (data not shown), confirming that the aggregated AUUCU repeat sequences are spliced from the mutant ATXN10 transcripts. These findings suggest that intron 9 containing the expanded AUUCU repeat is spliced out of the mutant ATXN10 transcripts, but expanded AUUCU repeats within the spliced intron 9 are resistant to degradation, and deposited as aggregates in nuclei and in cytoplasm in SCA10 cells.
We determined whether expanded AUUCU repeats alone are sufficient to form aggregates. Untranslated ∼500 AUUCU repeats from a transgene (Figure 1C) were expressed in human neuroblastoma Sy5y cells. The transgene is designed to express an expanded ATTCT repeat within the rabbit β-globin intron downstream of the human α-enolase promoter and upstream of the LacZ reporter. Using RT-PCR analysis, we confirmed that the AUUCU-repeat-containing the rabbit β-globin intron is spliced from the transcript when the transgene is expressed in Sy5y cells (data not shown). FISH analysis of the Sy5y cells expressing ∼500 AUUCU repeats showed SCA10-like nuclear and cytoplasmic aggregates (Figure 1D). However, under identical conditions, aggregates were not detected in Sy5y cells expressing the lacZ transcripts encoding shorter repeats (12 or 25 repeats) (data not shown). FISH analysis of transfected normal human fibroblasts expressing ∼500 AUUCU repeats also showed SCA10-like nuclear and cytoplasmic aggregates (Figure S1B; arrow). These data indicate that even when the expanded AUUCU repeat is ectopically expressed, the intronic sequence is spliced from the transcript of the transgene, becomes resistant to degradation, and aggregates in nuclear and cytoplasmic foci in Sy5y cells.
We also determined whether transcripts with expanded AUUCU sequences form similar aggregates in mouse brain. Transgenic mouse lines using the construct described in Figure 1C were generated. Repeat-primed PCR (RP-PCR) analyses of genomic DNA from these mice showed the presence of expanded ATTCT repeats (Figure 1E), and Southern analyses confirmed the presence and integrity of ∼500 ATTCT repeats (Figure 1F). The (AGAAU)10 riboprobe detected distinct intracellular aggregates in brains from 6-month-old (Figure 1G) and 3-month-old (Figure S1C) transgenic mice, but not in control mouse brains. Importantly, similar to the SCA10 cells, we observed a large number of aggregates not only in the nucleus but also in the cytoplasm, and they were more abundant in 6-month-old than 3-month-old mice (Figure 1G and Figure S1C). The formation of SCA10-like aggregates in these cells and transgenic mouse brains confirms that the expanded AUUCU repeats are necessary and sufficient to form nuclear and cytoplasmic aggregates. Moreover, these large foci suggest that the expanded AUUCU-RNA repeats may aggregate as insoluble RNA-protein complexes, as described in other repeat expansion disorders [7], [8].
Light-microscopic analysis of the Sy5y cells expressing the ∼500 AUUCU repeats showed a dramatic increase in cell death (Figure 2A), whereas cells expressing normal-size repeats showed virtually no cell death (Figure 2B). A TUNEL assay revealed that more than 70% of cells expressing the ∼500 AUUCU repeats underwent apoptosis 48 hours after transfection (Figure 2A and 2C), while cells expressing 12 AUUCU repeats did not undergo apoptosis (Figure 2B and 2C) (p<0.0001). Furthermore, caspase-3 activity was significantly higher in cells expressing ∼500 AUUCU repeats than in control cells (p<0.0001) (Figure 2D), suggesting that the expanded AUUCU repeats activate caspase-3-mediated apoptosis. We also observed that expression of ∼500 AUUCU repeats cause apoptosis in PC12 cells (Figure S2).
The distinct aggregates in SCA10 cells and transgenic mouse brains led us to hypothesize that the expanded AUUCU repeat RNA may interact with proteins, and that such interactions may have pathogenic significance. We pulled down proteins from mouse brain extracts using biotin-labeled expanded AUUCU RNA repeats and analyzed them by SDS-PAGE (Figure 3A). The unique protein that was reproducibly and repeatedly pulled down (n = 6) was identified by mass spectrometry as hnRNP K (Figure 3A; arrow). hnRNP K contains three K-homology (KH) domains that mediate its interactions with RNA and a K interactive (KI) region with proline-rich docking sites important for src homology domain binding. To establish the specificity of the interaction of hnRNP K with AUUCU RNA, purified hnRNP K was incubated with single-stranded (AUUCU)15 RNA and extracted with buffers containing increasing salt concentrations. The data show a significant affinity of hnRNP K with (AUUCU)15 RNA even in the presence of 250 mM NaCl; in contrast hnRNP K is completely dissociated from the control RNA at significantly lower (≤100 mM) salt concentrations (Figure 3B). Thus, hnRNP K can bind tightly to AUUCU repeats, in addition to the consensus sequence, U(C3-4)U/A [19]. We next immuno-precipitated (IP) hnRNP K from SCA10 and normal fibroblasts and determined the presence of intron 9 of the ATNX10 transcript in the IP pellets. RT-PCR analysis of the IP pellets showed the presence of the intron 9 sequence of the ATXN10 transcript when hnRNP K was precipitated from SCA10 fibroblasts but not from normal fibroblasts (Figure 3C). We did not detect the presence of exon 9 or exon 10 in the pellets, further corroborating the idea that intron 9 containing the expanded AUUCU repeat is spliced from the ATXN10 transcript. These data indicate that hnRNP K is tightly associated with the expanded AUUCU repeat within the intron 9 sequence in SCA10 cells.
To demonstrate the in vivo interaction of the expanded repeat with hnRNP K, we investigated the co-localization of hnRNP K with AUUCU RNA in transgenic mouse brain. The expanded AUUCU repeat aggregates were visualized by FISH, and hnRNP K was detected with anti-hnRNP K antibody by immunofluorescence. Sagittal sections of hippocampus CA1 (Figure 3E) and cerebral cortex (Figure S3) from the 6-month-old transgenic mice showed distinct co-localization of ∼500 AUUCU aggregates with endogenous hnRNP K. In contrast, control mouse brains showed no foci (Figure 3D). We next transfected Sy5y cells with two plasmids: one to express the ∼500 AUUCU repeat (Figure 1C) and the other to express GFP-tagged hnRNP K. FISH analysis of the double-transfected cells revealed significant co-localization of the red fluorescence from the AUUCU RNA repeat and the green fluorescence from the GFP-hnRNP K (Figure 3F; arrow), indicating that hnRNP K exists as a RNA-protein complex with AUUCU RNA in vivo.
We assessed whether binding of hnRNP K with the expanded AUUCU RNA interferes with hnRNP K activity by studying hnRNP K-regulated alternative splicing of transcripts. hnRNP K is known to regulate alternative splicing of exon 6A and 6B, the mutually exclusive exons of the β-tropomyosin gene in vertebrates, and decreased hnRNP K activity has been shown to increase the inclusion of exon 6A and the exclusion of exon 6B [20], [21]. RT-PCR analysis showed that exon 6A is predominantly included in the mature β-tropomyosin transcripts in SCA10 cells compared to normal control cells (Figure 3G), suggesting that the hnRNP K activity is decreased in SCA10 cells. Consistent with these results, splicing of β-tropomyosin was also markedly altered in normal fibroblasts ectopically expressing expanded AUUCU repeats (data not shown).
To understand the possible pathogenicity of a loss of hnRNP K function in SCA10, we treated Sy5y cells with four different hnRNP K siRNA duplexes. The sequence that most significantly and reproducibly decreased hnRNP K protein level was used at titrating concentrations to knockdown hnRNP K in Sy5y cells. Western blot analysis showed that cells treated with 100 pM hnRNP K siRNA had >50% reduction in hnRNP K protein level, compared to that in cells treated with control siRNA (Figure 4A). We detected no significant cell death up to 48 hours after siRNA treatment, in accordance with previous studies [22], [23]. However, we observed a large number of dying cells 72 hours after transfection with the hnRNP K siRNA; in contrast, cells treated with control siRNA did not show significant cell death. Activation of cell death pathways in Sy5y cells transfected with hnRNP K siRNA was verified by significant caspase-3 activity (n = 3, p<0.001) (Figure 4B), and increased TUNEL-positive cells (n = 6, p<0.0001) (Figure 4C), 72 hours post-transfection. The concentration of hnRNP K siRNA sufficient to activate caspase-3-mediated apoptosis at 72 hours was 100 pM (n = 3, p = 0.0001) (Figure 4D). Thus, down-regulation of hnRNP K activates caspase-3-mediated apoptosis similar to that observed in cells expressing expanded AUUCU repeats.
Based on our data we postulated that an interaction of hnRNP K with the AUUCU repeat results in a loss of function of hnRNP K, leading to apoptotic cell death. We studied whether hnRNP K over-expression rescues cells from apoptosis induced by expanded AUUCU repeats. We established stably transfected Sy5y cell lines over-expressing hnRNP K, and then transiently transfected them with the plasmid shown in Figure 1C. Sy5y cells stably expressing a control plasmid in lieu of hnRNP K underwent massive cell death when expanded AUUCU repeats were expressed. In contrast, ∼50% over-expression of hnRNP K (Figure 4E) decreased the expanded AUUCU repeat-induced apoptosis by ∼30%, and this decrease in apoptosis is accompanied by reduced caspase-3 activity (n = 3, p<0.05) (Figure 4E). These data confirm our hypothesis that expanded AUUCU repeats activate apoptosis by suppressing hnRNP K function.
In vivo studies have shown that hnRNP K and PKCδ remain constitutively bound together within the cell [24]–[27]. Studies also showed that hnRNP K, when bound to nucleic acids, cannot be phosphorylated and cannot interact with PKCδ [25]. PKCδ has been implicated as an activator of apoptosis in many cell types, including neurons [28], [29]. Over-expression of PKCδ has been shown to activate apoptosis through a positive regulatory loop, in which caspase-3 activates PKCδ and activated PKCδ cleaves caspase-3 [30]. PKCδ over-expression results in its translocation to mitochondria, release of cytochrome c, and activation of caspase-3 [30]–[32]. Since binding of hnRNP K to the expanded AUUCU repeat is expected to reduce the formation of the hetero-dimeric complexes between hnRNP K and PKCδ, and mimic PKCδ over-expression, we investigated the ramifications of hnRNP K inactivation on sub-cellular localization of PKCδ in SCA10 fibroblasts and transgenic mouse expressing ∼500 AUUCU repeats in brain.
We first investigated whether cellular localization of PKCδ is altered in SCA10 cells. PKCδ was immunostained with green fluorescence and mitochondria were identified using mitotracker deep red 633. Immunostaining of the normal fibroblasts for PKCδ showed that PKCδ is present in the cytoplasm and the nucleus, but no significant PKCδ localization in mitochondria (Figure 5A). In contrast, PKCδ significantly overlaps with mitochondria in SCA10 fibroblasts as punctate staining around the nucleus, suggesting that a significant portion of PKCδ is translocated into the mitochondria (Figure 5B). To further verify that the interaction between hnRNP K and PKCδ is diminished in SCA10 fibroblasts, we immunoprecipitated hnRNP K from normal and SCA10 fibroblasts and analyzed the relative abundance of hnRNP K and PKCδ in the IP by Western blot analysis. The Western blot data show a significantly lesser amount of PKCδ in the IP from the SCA10 fibroblasts compared to normal fibroblast (Figure S4A). These data support our hypothesis that expanded AUUCU RNA interacts with hnRNP K and this binding results in the release of PKCδ, facilitating translocation of PKCδ to mitochondria in SCA10. To test that PKCδ is translocated to the mitochondria in SCA10 cells, we analyzed the mitochondrial protein fractions from SCA10 and control fibroblasts by Western blotting. Consistent with the immuno-histochemical data, the Western blot data showed elevated PKCδ level in SCA10 mitochondria (4B). Moreover, sagittal sections of transgenic mouse brain showed similar mitochondrial localization of PKCδ while negligible mitochondrial localization of PKCδ was seen in age-matched wild-type mice (Figure 5C and 5D). We also analyzed fibroblasts derived from patients with ataxia telangiectasia, and unlike SCA10 fibroblasts, these fibroblasts did not show presence of any detectable level of PKCδ in mitochondria (data not shown), suggesting the disease specificity of this mechanism in SCA10. These results suggest that PKCδ is translocated into the mitochondria of SCA10 cells.
To test the hypothesis that the interaction of AUUCU RNA with hnRNP K leads to a loss function of hnRNP K, which then results in translocation of PKCδ into mitochondria, we transfected normal fibroblasts with hnRNP K siRNA and studied the cellular localization of endogenous PKCδ. When hnRNP K is downregulated, a majority of PKCδ was translocated to mitochondria and a negligible amount of PKCδ was detected outside mitochondria (Figure 6A). In normal fibroblasts, or in fibroblasts treated with control siRNA, most of the PKCδ was detected within cytoplasm and nuclei, with no detectable translocation to mitochondria (Figure 6B). Importantly, downregulation of hnRNP K in normal fibroblasts did not alter the steady state level of PKCδ (Figure S4C).
We also expressed ∼500 AUUCU repeats in primary human fibroblasts and studied the expression and cellular localization of PKCδ in these cells to investigate whether AUUCU RNA interferes with the expression and/or subcellular localization of PKCδ. We found that the expression of PKCδ remained unaltered in cells expressing expanded AUUCU repeats and the red fluorescence from mitochondria significantly overlaps with the green fluorescence from PKCδ in fibroblasts expressing ∼500 AUUCU repeats (Figure 6C and Figure S4D) or in those cells expressing ∼200 AUUCU repeats (Figure S5), suggesting that PKCδ translocates to mitochondria in response to the expression of expanded AUUCU repeats. In contrast, a negligible translocation of PKCδ was observed when 12 AUUCU repeats were expressed in fibroblasts (Figure 6D). Together, these data corroborate our hypothesis that the expanded AUUCU repeat interacts with hnRNP K, suppresses its function, resulting in mitochondrial translocation of PKCδ and activation of apoptosis.
To test whether mitochondrial localization of PKCδ can be decreased by reducing the mutant ATXN10 transcript, we targeted the ATXN10 transcript in SCA10 fibroblasts with two different ATXN10 siRNA and studied the cellular localization of PKCδ. This resulted in significant decrease in the number of both nuclear and cytoplasmic AUUCU RNA aggregates (Figure 7A; left panel), whereas control siRNA did not reduce the number of AUUCU RNA foci in SCA10 fibroblasts (Figure 7A; right panel). Treatment of SCA10 fibroblasts with ATXN10 siRNA substantially restored normal PKCδ subcellular localization, with decreased amount in mitochondria (Figure 7B; top panel). As expected, Control siRNA had no significant effects on the distribution or amount of PKCδ in SCA10 cells (Figure 7B; center panel). Treatment of normal fibroblasts with ATXN10 siRNA did not have any effect on PKCδ cellular localization (Figure 7B; bottom panel). We conclude that by disrupting the hnRNP K-AUUCU complexes, hnRNP K can re-establish its normal function within the cell, alleviating the pathogenic mechanisms leading to apoptosis. These findings support our hypothesis that expanded AUUCU repeats are toxic and are sufficient to trigger PKCδ translocation to mitochondria and apoptosis.
Multiple inherited human neurological disorders are now attributed to expansion of short tandem repeats either in coding or non-coding regions of genes [2], [7], [8]. Genetic and molecular analysis of these disorders have revealed that the repeat expansion can result in either a loss of function of the gene (Fragile-X syndrome and Friedreich's ataxia) or a gain of function of the encoded protein (SCA1, SCA2, SCA3, SCA6, SCA7, SCA17, Huntington's disease, DRPLA, and oculopharyngeal muscular dystrophy) [7], [8]. RNA-mediated pathogenesis is believed to play a critical role in several other repeat expansion disorders, including Myotonic Dystrophy Type 1 (DM1) and Type 2 (DM2), SCA8, SCA12, Huntington's disease like 2 (HDL2), and fragile X tremor ataxias syndrome (FXTAS) [7], [8]. However, the pathogenic mechanism of DM1, SCA8, SCA12, HDL2 and FXTAS, which are caused by trinucleotide repeat expansions, may also involve qualitative or quantitative alterations of the protein products of the respective genes or genes on the opposite strand [33]–[35]. In contrast, SCA10 is the only human disorder proven to be caused by an expansion of a pentanucleotide repeat. Like the DM2 CCTG tetranucleotide repeat, the SCA10 ATTCT repeat shows repeat-number polymorphism, which makes these non-trinucleotide repeat highly unlikely to encode protein sequences from either strand. Furthermore, we have shown that the intronic repeat expansion does not alter ATXN10 transcripts [17]. Thus, SCA10 is likely to be a disorder solely caused by RNA-based mechanism, unlike most disorders that are caused by trinucleotide repeat expansions.
In the present study we provide evidence that SCA10 pathogenesis results from a trans-dominant gain-of-function of AUUCU repeats. First, transcription of the mutant allele produces transcripts that form aggregates in the nucleus and cytoplasm of the SCA10 cells and in transgenic mouse brain. Second: the expanded AUUCU repeat complexes with hnRNP K, leading to the loss of function of hnRNP K. Third, expression of expanded AUUCU repeat results in the accumulation of PKCδ in the mitochondria and caspase-3 mediated activation of apoptosis. Fourth, diminished hnRNP K activity recapitulates these events caused by expanded AUUCU repeats. And finally, over-expression of hnRNP K, as well as down-regulation of transcripts of expanded ATTCT repeat, rescues cells from apoptosis caused by expanded AUUCU repeats. Based on these findings, we conclude that the AUUCU RNA binds to and inactivates hnRNP K, triggering caspase-3-mediated apoptosis via translocation of PKCδ to mitochondria. Previous reports suggest that the presence of PKCδ in the mitochondria results in decreased membrane potential, release of cytochrome C, and activation of caspase-3 [30]–[32], further supporting our conclusion. Moreover, caspase-3 activates PKCδ and activated PKCδ further activates caspase-3 [30], and proteolytically activated PKCδ down-regulates hnRNP K protein in a proteasome-dependent manner [36]. Hence, positive feedback loops involving hnRNP K, PKCδ and caspase-3 may enhance this pathogenic pathway in SCA10. Since apoptosis is considered to be a major mechanism of cell death in a variety of human neurodegenerative disorders [37], the novel pathway of apoptosis induced by the mutant ATXN10 RNA is relevant to the neurodegenerative phenotype of SCA10. Our results provide strong evidence that this novel mechanism of trans-dominant RNA gain of function contributes to the pathogenic mechanism in SCA10.
The formation of aggregates may not necessarily be a required event for the mutant RNA to exert its toxicity. Binding of the soluble form of the mutant RNA to hnRNP K may be sufficient to cause the loss of function of hnRNP K with a release of PKCδ, and the aggregate formation could be a secondary phenomenon. We hypothesize that expanded AUUCU RNA pathologically binds to hnRNP K and prevents PKCδ from binding to the hnRNP K, mimicking over-expression of PKCδ within the cell. Previous studies have shown that hnRNP K is constitutively bound to PKCδ, but upon binding to nucleic acids, hnRNP K can no longer interact with PKCδ [25], [30]. Translocation of PKCδ to mitochondria in SCA10 cells, fibroblasts expressing expanded AUUCU repeat, and fibroblasts treated with hnRNP K siRNA argues for this mechanism. Studies have shown multiple apoptotic activators, including oxidative stress and over-expression of PKCδ, induce PKCδ translocation to the mitochondria, [32]. The mitochondrial translocation of PKCδ has been shown to cause an alteration in calcium signaling events and mediates the H2O2-mediated loss of membrane potential, release of cytochrome c, and activation of caspase-3 [38]. While it is possible that the expanded AUUCU repeat causes PKCδ translocation via other mechanisms, our data showing that over-expression of hnRNP K rescues AUUCU-mediated apoptosis argue for the mechanism mediated by a loss of function of hnRNP K.
Our present data do not rule out the possibility that additional proteins interact with the mutant ATXN10 transcripts. Also, expression of hnRNP K is ubiquitous within the cell, and diminished hnRNP K could lead to altered regulation of transcription, splicing and cell signaling, which may account for the phenotypic variability in SCA10 as illustrated in Figure 7C. We are investigating these mechanisms. However, our current data convincingly show that the hnRNP K inactivation and PKCδ mitochondrial translocation are a key pathogenic pathway mediating the RNA gain of toxic function in SCA10.
SCA10 fibroblasts were isolated from skin biopsy from a Mexican-American SCA10 patient with ∼2000 repeats and a Brazilian SCA10 patient with ∼1000 repeats under signed informed consent approved by IRB at UTMB and the Ethics Committee at Federal University of Parana. These human fibroblasts Cells were cultured in MEM with Eagle-Earle salt and 2 mM L-glutamine containing 15% fetal bovine serum and antibiotic in 5% CO2 at 37°C in 75 cm2 flasks. Human neuroblastoma Sy5y cells were cultured at Ham's F12K medium with 2 mM L-glutamine adjusted to contain 1.5 g/L sodium bicarbonate, 15% horse serum, 2.5% fetal bovine serum in 5% CO2 at 37°C in 75 cm2 flasks.
The cytomegalo virus (CMV) promoter sequences in plasmid pCDNA3.1-hygro-lacZ (Invitrogen) were replaced with the MfeI/BamHI fragment of the human α-enolase promoter sequences (∼5.0 kb). The 2nd intron of rabbit β-globin intron was cloned downstream of the enolase promoter and upstream of the lacZ. Expanded ATTCT repeats from the SCA10 hybrid cells [17] were PCR amplified with forward primer 5′-CCAAGGATGCAGGTGCCACAGCATCTC-3′ and reverse primer: 5′-ATATGCATCCAGCTTCTGATTACATGGACT-3′. A polylinker containing SwaI site was cloned into the MfeI site within the β-globin intron. Subsequently, the DNA fragment containing the ATTCT duplex was cloned into the SwaI site within the intron. Presence of the expanded ATTCT sequences in the transgenic plasmid was confirmed by digesting the plasmid DNA with NheI and HindIII sites that flank the SwaI site, and by sequencing. Plasmids encoding the expanded ATTCT repeats were grown in E. coli SURE bacteria at 16°C to minimize the deletion of the repeat sequences [39]. The transgenic plasmid DNA containing the LacZ and ∼500 ATTCT repeats was digested with MfeI and NaeI, and digested DNA was electrophoresed on agarose gel. The ∼10 kb DNA fragment containing the transgene was purified from agarose gel using gel extraction kit (Qiagen). The cloned ATTCT repeats under enolase promoter contain 650 bp of upstream and 500 bp of downstream ATXN10 sequence in addition to the ATTCT repeats. The control plasmids containing the same ATXN10 flanking regions and 12 ATTCT repeats were PCR amplified from a normal subject. The cDNA clones of human hnRNP K was purchased from Open Biosystems, USA and cloned in-frame into pCGFP-C1 (BD Biosciences, USA). For construction of plasmid for stable expression of hnRNP K, the ORF of the human hnRNP K was PCR-amplified from the pool of human cDNA (Clonetech) with the following primer sets: forward: 5′-CTGATTGGTGTGCCCGTTTAATAA-3′ and reverse: 5′-CTCCTTCAGTTCTTCACTAGTC-3′. The 1507 bp PCR product was purified from agarose gel using gel extraction kit (Qiagen) and the blunt-ended PCR product was cloned into the EcoRV site in the mammalian expression vector pcDNA3.1-Hygro(+) (Invitrogen) to generate recombinant plasmid pcDNA-hnRNP K. The coding sequence of hnRNP K in recombinant plasmid pcDNA-hnRNP K was sequenced to verify the proper orientation and sequence integrity of hnRNP K.
The transgene containing 500 ATTCT repeats within an intron (Illustrated in Figure 1C) was microinjected into the fertilized eggs and transplanted into the uterus of pseudo-pregnant surrogate mothers to obtain founder transgenic mice using standard procedure at the UTMB transgenic core facility. Presence of the transgene and the repeat in the transgenic founder mice were confirmed by both Southern blot as well as repeat primed PCR analyses. Animal experiments were performed under a protocol approved by IACUC at UTMB.
Mouse genomic DNA was collected from tail samples, and repeat-primed PCR and Southern blot analyses were performed as previously described [40].
Plasmid pcDNA3.1 control as well as pcDNA-(ATTCT)n (n = 500) were first linearized with Bam HI and the linear plasmid was in vitro transcribed with T7 RNA polymerase (Promega). Biotinylated rCTP was mixed with other rNTPs during transcription to incorporate the biotin-labeled rCTP into (AUUCU)n RNA. Nuclear extracts from brain were made from a 2 month old B6 mouse using NE-PER Nuclear and Cytoplasmic Extract Reagent (Pierce) according to vendors specification. The total protein mixture was incubated with (AUUCU)n RNA at 4°C overnight, and the unbound proteins were washed by using RNA washing buffer (0.5% NP- 40, 100 mM NaCl, 50 mM Tris-HCl) four times. The proteins that remain bound to the magnetic beads after extensive washing were extracted by boiling the magnetic beads in 1X SDS-PAGE loading buffer. The extracted proteins were electrophoresed on 5–12% PAGE and proteins that appear as unique bands were excised, digested with trypsin and then analyzed by MALDITOF assay at Biomolecular Resource Facility Core at UTMB and the sequence was identified by searching the rodent protein database.
The SCA10 and control normal fibroblasts were harvested and washed with 1X PBS, lysed in 20 mM HEPES pH 7.4, 1 mM EDTA, 100 mM NaCl, 1% NP-40, Leupeptin (10 µg/ml), aprotinin (10 µg/ml), 20 mM β-glycero-phosphate, 20 mM NaF and 1 mM Na Vanadate. hnRNP K was immuno-precipitated using anti-hnRNP K antibody and the IP pellet was washed three times in the lysis buffer, and bound proteins were eluted in SDS-containing sample buffer and separated on SDS-PAGE. The membrane was first immunoblotted with anti-hnRNP K antibody and then with PKCδ antibody. To detect the intron 9 sequences of ATXN10 in the IP pellet, total RNA was extracted from the IP pellets by phenol-chloroform extraction, and DNA contamination was removed with TURBO DNAse Kit (Ambion). The cDNA synthesis was carried out using 1 µg of total RNA using a RT-PCR kit (BD Biosciences). The cDNA aliquots were quantified and cDNA were used to detect the presence of intron 9 sequences of ATXN10 transcripts by PCR, using the forward primer 5′-AAGGATCAGAATCCCTGGAA-3 and the reverse primer 5′-TCATTCTGCCATCTGTTTTC-3′.
Splice isoforms of β-tropomyosin mRNA were analyzed using RT-PCR as previously described [20], [21] with a set of three primers; E5: 5′-GCCATGAAGGATGAGGAGAA-3′ (forward primer), E6a: 5′-CTGAGGTGGCCGAGAGGTAA-3′ (reverse primer to detect exon E6a) and E6b: 5′-TAAATGTGGGGACCTAGAGG-3′ (reverse primer to detect exon E6b).
1 mg of the Streptavidin-conjugated Dynabeads (Invitrogen) was washed once with Solution A (100 mM NaOH, and 50 mM NaCl) and twice with Solution B (100 mM NaCl). The magnetic beads were next incubated in 50 µl of 2X incubation buffer (10 mM Tris, pH 7.5; 1 mM EDTA; 2 M NaCl,) for 15 minutes. 1000 pmoles (50 µl) of biotinylated RNA [either (AUUCU)15 or control RNA: (UUUCC)3(CCCUU)3(UUUUC)3.] were added to the magnetic beads, incubated at room temperature for 1 hour, and washed twice with 1X incubation buffer. Five mg of purified hnRNP K were added to the magnetic bead-RNA mixture and incubated in a binding buffer (20 mM HEPES, pH 7.5; 10% glycerol; 1 mM DTT; 0.1 mM EDTA) at 4°C for overnight. The supernatant was discarded and the magnetic beads were resuspended in binding buffer (25 mM Tris, pH 7.5; 0.1 mM EDTA; 10 mM NaCl). RNA-bound hnRNP K was next sequentially eluted with buffers containing increasing NaCl concentrations. The eluted hnRNP K protein fractions were analyzed on PAGE by Coomasie staining and Western blotting.
Mouse monoclonal anti-hnRNP K was obtained from Acris Antibodies GmbH. Monoclonal anti-PKCδ (G-9) and Cytochrome C Oxidase (COXII) antibodies were purchased from Santa-Cruz Biotechnology. The Western blotting experiments were done according to standard procedure and the target proteins were detected using ECL Western kit (Amersham). Expression levels of β-actin (Abcam) were used as controls for protein loading.
Cells were grown in chamber slides overnight prior to TUNEL assay. TUNEL assay was performed according to vendor instructions (Roche). Student's t-test was used to calculate statistical significance.
Caspase-3 assay was performed according to instructions supplied by vendor (Calbiochem). P values were calculated using student's t-test.
Sy5y cells were transfected with plasmids or siRNA by Lipofectamine 2000 reagent (Invitrogen). For the targeted inactivation of hnRNP K, the On-target-Plus siRNA duplexes were purchased from Dharmacon. The control non-targeting siRNA from Dharmacon was used as a negative control for all siRNA experiments. Approximately 100 and 200 nmoles of the siRNA pools were used for the targeted down-regulation of hnRNP K, and different assays were applied 72 hours after the transfection. Fibroblasts were transfected using the human dermal fibroblast nucleofector kit for electroporation (Amaxa Corporation). Plasmids expressing either 12 or 500 ATTCT repeats were transfected at 3 µg according to kit instructions, and siRNA (both hnRNP K and ataxin 10) was transfected according to kit instructions. For stable over-expression of hnRNP K in Sy5y cells, pcDNA-hnRNP K was digested with MfeI and the linear plasmid DNA was tranefected into Sy5y cells using Lipofectamin 2000 reagent (Invitrogen), and the stable clones were selected with hygromycin (300 mg/ml). Total protein was isolated from the stably transfected cells and the hnRNP K expression level was analyzed by Western blotting with the anti-hnRNP K antibody.
RNA foci were detected using a Cy3-labeled (AGAAU)10 RNA riboprobe. Slides were pre-hybridized at 65°C in RNA hybridization buffer for 1.5 hours. Slides were then hybridized overnight in 250 ng (AGAAU)10/1 ml hybridization solution at 45°C. Slides were rinsed with PBS three times and then extensively washed 4 times 5 minutes each to remove all non-specific binding probes. Slides were then mounted with DAPI mounting medium.
Transgenic mice anesthetized with Avertin were perfused through the aorta, first rinsing for 15 minutes with PBS and then 60 ml of fresh 4% Paraformaldehyde (PFA) in DEPC water. The brain was carefully removed and stored in 4% PFA at 4°C with gentle agitation overnight. Brain tissue was then placed in 30% sucrose overnight. Mouse brains were fixed in paraffin and sectioned sagittally. RNA foci were stained using a Cy3-labeled (AGAAU)10 riboprobe. First, paraffin was removed from the brain sections and slides were dehydrated with 70%, 95% and 100% Ethanol in DEPC water, and washed using DEPC PBS. Following FISH, hnRNP K was immunodetected. Sections were blocked with DAKO antibody blocking solution (serum-free) and later double stained with anti-hnRNP K 1∶1000 in DAKO antibody diluent. Goat anti-mouse 488 was used to identify hnRNP K and slides were visualized using a Hamamatsu Camera Controller using DP controller software in histopathology lab at UTMB.
Fibroblasts were transfected with plasmids pcDNA-(ATTCT)12, pcDNA-(ATTCT)500, hnRNP K siRNA or control siRNA through electroporation. Transfections were conducted in chamber slides. Thirty-six hours after repeat transfection and 72 hours after RNAi transfection, the cells were treated with mitotracker deep red 633 (Invitrogen) at a concentration of 250 nM in cell culture medium. Cells were incubated at 37°C for 30 minutes. After washing the cells three times with PBS, cells were then fixed with 4% PFA for 30 minutes at room temperature. Cells were washed 3 times with PBS and stored in 70% Ethanol for up to 24 hours. Cells were blocked with DAKO antibody blocking solution (serum-free) and later double stained with anti-PKCδ 1∶500 in DAKO antibody diluent. Goat anti-mouse 488 was used to identify PKCδ. Fluorescent photomicrographs were taken using a Hamamatsu Camera Controller using DP controller software in the histopathology core lab at UTMB.
The sagittal section of the transgenic brain was processed according to the procedure described above and immuno-stained with anti-PKCδ and Cox II antibodies to detect mitochnodria and PKCδ respectively. The cytoplasmic, nuclear and mitochondrial protein fractions from normal and SCA10 cells were isolated using the mitochondria isolation kit and Sub-cellular Protein Fractionation kits (Thermo Scientific-Pierce). The isolated proteins were analyzed by Western blotting using anti-PKCδ antibody.
|
10.1371/journal.pcbi.1004467 | Genome-Wide Detection and Analysis of Multifunctional Genes | Many genes can play a role in multiple biological processes or molecular functions. Identifying multifunctional genes at the genome-wide level and studying their properties can shed light upon the complexity of molecular events that underpin cellular functioning, thereby leading to a better understanding of the functional landscape of the cell. However, to date, genome-wide analysis of multifunctional genes (and the proteins they encode) has been limited. Here we introduce a computational approach that uses known functional annotations to extract genes playing a role in at least two distinct biological processes. We leverage functional genomics data sets for three organisms—H. sapiens, D. melanogaster, and S. cerevisiae—and show that, as compared to other annotated genes, genes involved in multiple biological processes possess distinct physicochemical properties, are more broadly expressed, tend to be more central in protein interaction networks, tend to be more evolutionarily conserved, and are more likely to be essential. We also find that multifunctional genes are significantly more likely to be involved in human disorders. These same features also hold when multifunctionality is defined with respect to molecular functions instead of biological processes. Our analysis uncovers key features about multifunctional genes, and is a step towards a better genome-wide understanding of gene multifunctionality.
| Almost every aspect of cellular function depends on protein activity. In spite of being fine-tuned to carry out highly specific functions, proteins can also multitask. Experimental studies have identified genes and proteins endowed with more than one molecular function, or participating in very different biological processes. These studies suggest that the degree of functional plasticity exhibited by proteins might go well beyond a simple “one protein—one function” relationship. However, systematic studies of the properties of multifunctional genes (and their encoded proteins) have been limited. Here we present a computational framework to identify putative multifunctional genes, and compare their properties with those of other genes. We find that multifunctional genes are significantly different from other genes with respect to their physicochemical properties, expression profiles, and interaction properties. We also observe that multifunctional genes tend to be more conserved, and that a greater fraction of them are associated with human disorders. Taken together, these results represent a step towards a more complete understanding of the role multifunctional genes play in the functional organization of the cell.
| Multifunctionality can be defined as the involvement of a gene in multiple cellular processes [1]. This can come about either because a protein coded by a gene is capable of performing distinct molecular functions [2–6], or as a result of a single molecular function being performed in different contexts [7, 8]. For example, pioneering experimental work led to the surprising finding that crystallins—the proteins responsible for the optical properties of the eye lens—can also play non-refractive roles and have enzymatic activity in other tissues [2]. This evolutionary strategy was named “gene sharing” [9]. Further examples of proteins performing multiple molecular functions were subsequently described: a uracil-DNA glycosylase that can also function as a glyceraldehyde-3-phosphate dehydrogenase, or the enzyme thrombin that can moonlight as a ligand for surface receptors [3]. More recently, a large-scale screening of mutants in yeast was performed to measure the pleiotropic effects of genes under different conditions [10]. In the case of pleiotropy, a gene may perform only one molecular function, but it can be involved in multiple biological processes, and its perturbation can therefore have pleiotropic consequences.
Though multifunctionality has been characterized in detail only for a few case studies, it is likely to be a common phenomenon. Nevertheless, multifunctionality remains poorly understood. Fortunately, the current state of known gene functional annotations for several organisms gives us an opportunity to systematically identify multifunctional genes and analyze their properties. Earlier computational studies have attempted to identify multifunctional genes from functional annotations available for genes in different organisms. Several previous works measured multifunctionality by simply counting the number of distinct Gene Ontology (GO) biological process terms annotating a gene product [11–14]. While intuitive and straightforward, these approaches do not always guarantee that a gene annotated with more than one GO term is indeed involved in two distinct biological processes. In particular, this assumption is incorrect when one term is a descendant of another term in the GO hierarchy. To better handle the hierarchical organization of GO, an alternate approach considered the total number of distinct GO “leaf” terms annotating a gene [15], and a recent analysis used semantic similarity between GO terms to identify moonlighting proteins [16]. However, problems may also arise even when two terms are in completely different branches of the ontology, as idiosyncrasies in GO may lead to similar processes being categorized in distinct places in the ontology. Methods to overcome this redundancy by focusing on a manually curated subset of terms (e.g., GO Slim or other gold standards [17–19]), even though suitable for tasks such as function prediction, can introduce a bias from manual curation to the analysis of gene multifunctionality, and also may not be generalizable as more annotations become available. Other approaches have used protein-protein interaction data and defined proteins as multifunctional if they are located at the intersection of overlapping clusters [20]. However, computationally derived clusters can differ substantially depending on the algorithm used [21], thereby leading to imprecise views of multifunctionality. Further, using interaction data to define multifunctional genes has the obvious drawback of preventing an unbiased analysis of these genes’ network properties.
In our work, we develop a computational approach to identify multifunctional genes that leverages GO functional annotations in a systematic and robust manner. To handle similar terms that appear in distant places in GO, we explicitly select sets of terms that do not co-annotate an enriched number of genes; these terms are then used to identify multifunctional genes. We apply our procedure to detect multifunctional genes to three organisms—human, fly and yeast—and then compare in each organism the properties of multifunctional genes (and the proteins they encode) with those of other annotated genes. Our results across these species consistently show that, as compared to other genes, multifunctional genes possess distinct physicochemical properties, are more broadly expressed across cell types and tissues, tend to be more evolutionarily conserved, are more likely to be essential, and are topologically distinct in protein-protein interaction networks, in regulatory transcription factor–gene networks and in genetic interaction networks. We also find that multifunctional genes are significantly more likely to be involved in human disorders than other genes. Overall, our analysis leads to a more complete understanding of the role multifunctional genes play in the functional organization of the cell.
We use functional annotations of genes in three organisms, H. sapiens, D. melanogaster, and S. cerevisiae, to identify multifunctional genes in each of them at the genome-wide level. To accomplish this, we use Biological Process GO annotations [22], though in subsequent analyses we also consider multifunctionality with respect to Molecular Function. In the remaining text, when we refer to GO annotations, we refer to Biological Process terms unless otherwise specified. Our method for detecting multifunctional genes is shown schematically in Fig 1 and is briefly described below (see Materials and Methods for details).
The Biological Process GO is a hierarchy of terms representing different aspects of biological processes, where the terms range from very general to very specific and a relationship between terms indicates if one term implies another. We therefore start by selecting a subset of comparable terms that do not have ancestor or descendant relationships amongst themselves. This set of terms can be chosen at different specificity levels, represented by a parameter N corresponding to the number of genes annotated by a term. Lower values of this parameter produce larger numbers of more specific terms, and higher values result in smaller numbers of more general terms (S1 Fig). We consider several distinct levels of specificity and call multifunctional all genes for which we find evidence of multifunctionality at any specificity level.
Once the terms have been selected at a particular specificity level, we extract all genes annotated with at least two such terms. In order to select only pairs of distinct terms and make sure a gene annotated by both terms is truly multifunctional, we apply several filters to pairs of terms. From the collection of all pairs of terms at a particular specificity level, we filter out those that either share a common ancestor (other than the root) or have a common descendant term in the GO graph, as these events indicate that the terms are semantically related. However, this is not sufficient to claim that the remaining pairs of terms are distinct. For example, the terms aerobic respiration and mitochondrial translation do not have any ancestral or descendant term in common in the GO hierarchy graph besides the most general biological process term, but often co-annotate mitochondrial ribosomal proteins and capture semantically distinct aspects of the same function. Therefore, we further remove all pairs of terms that co-annotate more genes than expected by chance (as detected by the hypergeometric test). All genes co-annotated by some pair of chosen terms passing these two filters, for any set of chosen terms at each specificity level N considered, are called multifunctional. We note that, depending upon the application, our filters can be relaxed to consider more genes as multifunctional; for example, two biological processes may be considered distinct if they share a common ancestor that is sufficiently general. However, here we aim to identify genes that have the strongest evidence of multifunctionality.
In what follows, we compare multifunctional genes detected in fly, human, and yeast with all other annotated genes in these organisms in order to uncover whether there are significant differences between the two groups with respect to various biological properties. The number of multifunctional genes and the total number of annotated genes for each organism is given in Table 1, and the actual lists of identified multifunctional genes are provided as S1 File for fly, S2 File for human, and S3 File for yeast. We note that a small number of experimentally verified human, fly and yeast genes with multiple functions are known [23, 24], and our method is able to successfully detect a significant fraction of these genes (see Section 1.1 in S1 Text).
We start the analysis by studying some basic physicochemical properties of proteins. First, we hypothesized that multifunctional proteins may be longer than other proteins in order to accommodate more functional domains. To test this hypothesis, we compare the lengths of proteins encoded by multifunctional and other annotated genes in D. melanogaster, H. sapiens, and S. cerevisiae, and indeed find that multifunctional genes are significantly longer than other genes (p-values 1e-39, 1e-9, and 8e-12, respectively, Mann–Whitney U test), on average by 39%, 16%, and 15%, respectively (Fig 2). We also observe that proteins encoded by multifunctional genes have significantly higher numbers of distinct domains per protein (p-values 2e-7, 1e-10, and 2e-4, respectively), on average by 17%, 13%, and 8%, respectively (Fig 2); this is consistent with the earlier finding of a small but statistically significant positive correlation between the number of GO biological process leaf terms a gene has and its number of Pfam domains [15]. However, we also note that longer proteins have more domains, so the difference in length between multifunctional and other genes can potentially explain the observed difference in the number of domains (see Section 1.2 in S1 Text).
Another mechanism that has been proposed to play a role in protein multifunctionality is the presence of intrinsically unstructured regions, which are thought to increase the structural adaptability of interaction surfaces of proteins to allow them to bind to the same or distinct partners with different effects [25]. To determine whether multifunctional proteins tend to be more disordered, we predict the fraction of disordered residues using the IUPred program [26, 27], and find that multifunctional genes in D. melanogaster, H. sapiens, and S. cerevisiae have a significantly higher fraction of predicted disordered residues (p-values 6e-21, 7e-4, and 3e-14, respectively), on average by 26%, 5%, and 31%, respectively (Fig 2). These results are in agreement with recent analyses of disordered regions in experimentally verified moonlighting proteins and a small set of computationally inferred moonlighting proteins in E. coli [16].
Overall, we find that proteins encoded by multifunctional genes are longer, have more domains and are more disordered than proteins encoded by other annotated genes.
Differential gene expression is evident across tissues and cell types. A gene expressed in different contexts may have different functions depending upon how and when it is expressed. Therefore, we hypothesized that a gene associated with multiple distinct functions may be expressed in a larger number of contexts. In order to assess the relationship between gene expression and gene multifunctionality, we use genome-wide mRNA expression data and count in how many conditions, tissues or cell types each gene is expressed. For fly, we use two datasets: (1) FlyAtlas [28], the Drosophila microarray gene expression atlas across different tissues in larva and adult, and (2) RNA-seq data from modENCODE across many different tissues and development time points, as aggregated by FlyBase [29, 30]. For human, we use information about organism parts in which genes are expressed, obtained from Ensembl BioMart [31]. We observe that in both human and fly, multifunctional genes are expressed more broadly than other annotated genes; that is, they are expressed in a significantly larger number of tissues or organism parts (p-values from 7e-38 to 2e-4, Mann–Whitney U test; Fig 3A and 3B).
A potential mechanism for gene multifunctionality is the production via alternative splicing of multiple protein isoforms with different functions. Indeed, we observe that multifunctional genes have a significantly larger number of known isoforms in fly and human (S2 Fig). If different isoforms of a gene have different expression patterns, this gene may be detected as broadly expressed in genome-wide assays, which currently report expression only at the gene level, merging information about the expression of different isoforms. Indeed, we observe a significant positive correlation between the number of isoforms per gene and the number of tissues or organism parts in which it is expressed (S1 Table). However, when comparing genes with an equal number of known isoforms, we still observe that multifunctional genes are expressed in larger numbers of tissues or organism parts (although most p-values for human are above our significance threshold of 5%; S2 Fig). This indicates that multifunctional genes are more broadly expressed regardless of the number of isoforms.
Acquiring multiple functions may constitute a special evolutionary strategy and limit gene evolutionary rates [9]. In order to study the evolutionary dynamics of gene multifunctionality at the genome-wide level and in an unbiased manner, we use evolutionary conservation scores from phastCons [32]. Scores in phastCons are computed using phylogenetic hidden Markov models of multiple sequence alignments of D. melanogaster with 14 other insect genomes, of H. sapiens with 99 other vertebrate genomes, and of S. cerevisiae with 6 other yeast species. For each nucleotide of the genome, phastCons produces a score between 0 and 1, where higher values indicate stronger evolutionary conservation. For each gene, we average the scores of all nucleotides of each isoform of the gene, and then average over all isoforms of the gene to obtain a single value for each gene as an estimate of how evolutionarily conserved the gene is. Previously, a positive correlation between the number of biological process GO terms a protein is annotated with and its evolutionary conservation was observed for yeast [7, 11, 33]. In agreement with this, we find that in fly, human, and yeast, multifunctional genes are significantly more evolutionarily conserved than other annotated genes (p-values 5e-13, 6e-10 and 0.02, respectively, Mann–Whitney U test; Fig 4).
Having shown that multifunctional genes tend to evolve more slowly, we next hypothesized that multifunctional genes independently detected in different organisms may be orthologous to each other. In order to test this, we compare the property of multifunctionality for orthologous proteins from different organisms. We use information about protein orthology from P-POD [34] and count how many orthologous pairs are observed where both corresponding genes are identified as multifunctional. Between fly and human, we observe 1725 orthologous pairs of genes where one gene is classified as multifunctional in fly and the other gene is classified as multifunctional in human. To assess significance, we compute the same number when randomly reshuffling multifunctional and other annotated genes from orthologous pairs in each organism, and observe on average only 845.1 ± 90.0 orthologous pairs where both genes are classified as multifunctional; thus, the actual value is 2.0 times higher (empirical p-value < 1e-3). For fly and yeast, we find 388 orthologous pairs between multifunctional genes (2.1 times higher than 184.7 ± 20.2 expected by chance, p < 1e-3). For human and yeast, we find 576 orthologous pairs between multifunctional genes (2.2 times higher than 267.2 ± 32.6 expected by chance, p < 1e-3). We conclude that the property of multifunctionality is conserved across orthologous genes of different organisms. This observation also supports the validity of our method for detecting multifunctional genes.
Functional annotations of genes are in part determined by transferring information between organisms via sequence similarity, and this could potentially confound our evolutionary analysis of multifunctionality. To address this, we repeat the analysis excluding GO annotations based on sequence or structural similarity and observe the same trends (see Section S1.3 in S1 Text and S14 Fig).
Genes responsible for multiple functions may require more complex regulatory programs to differentiate functions across multiple tissues or conditions. In order to study how regulated multifunctional genes are, we use regulatory interactions from high-throughput ChIP experiments [35–38]. For each gene, we count the number of transcription factor–target interactions this gene participates in as a target. In all three organisms, we observe that multifunctional genes are regulated by a significantly larger number of transcription factors than are other annotated genes (p-values from 3e-54 to 7e-4, Mann–Whitney U test; Fig 5).
In addition to requiring more complex regulatory programs, multifunctional genes may also be associated with more complex phenotypes that require involvement with many other genes; this would be reflected in a gene’s genetic interactions. In order to compare the distribution of genetic interactions between multifunctional and other annotated genes, we use a collection of genetic interactions curated by FlyBase [30] for fly and by BioGRID [39] for yeast. Previously, a positive correlation between the number of biological process GO annotations a gene has and its number of genetic interactions was observed for yeast [19]. In agreement with this, we observe that in fly and yeast, the number of genetic interactions is significantly higher for multifunctional genes than for all other annotated genes (p-values 5e-25 and 2e-55, respectively; Fig 5). Moreover, in a more refined comparison for yeast, we observe that both the number of positive and the number of negative genetic interactions are significantly larger for multifunctional genes than for other annotated genes (p-values 9e-30 and 1e-40, respectively; Fig 5).
A gene associated with multiple functions may be more important for the normal functioning of the cell and therefore may potentially be more critical for survival than a gene associated with a single function. In order to test this hypothesis, we consider the relationship between gene essentiality and multifunctionality.
For fly, we call essential all genes with a lethal phenotype (as curated by FlyBase [30]) and observe that 74% of multifunctional genes are essential, whereas only 44% of other annotated genes are essential (p < 2e-86, hypergeometric test; Fig 6A). In addition, we use data from genome-wide RNAi screens in cell lines [40] and observe that, even though only a small fraction of genes in the study overall are detected as essential, multifunctional genes have a significantly higher fraction of essential genes than other annotated genes do (3.8% and 2.9%, respectively, p < 0.046; Fig 6B).
For human, we call essential all genes that have a mouse ortholog with a lethal phenotype (according to MGI [41]). We find that 53% of multifunctional genes are essential, whereas only 42% of other genes are (p < 7e-16; Fig 6C). Using data from a genome-wide RNAi screen in human mammary cells [42], we also observe that multifunctional genes are essential significantly more often (p < 1e-34; Fig 6D). In a more detailed analysis using quantitative data about essentiality in 72 human cancer cell lines [43, 44], we confirm that in all 72 cell lines, multifunctional genes tend to be more essential (S3 Fig).
Gene essentiality has been found to correlate with evolutionary rate [45, 46], and we observe that multifunctional genes tend to be more evolutionarily conserved; thus, the increased evolutionary conservation of multifunctional genes could potentially explain their preferential essentiality. We confirm that whether a gene is essential is correlated with its evolutionary conservation, but observe that multifunctional genes are still significantly more essential when controlling for evolutionary conservation (see Section 1.4 in S1 Text). We note, however, that the relationship between multifunctionality and evolutionary conservation becomes much weaker when controlling for essentiality, and thus the tendency of essential genes to be more evolutionarily conserved may indeed explain the tendency of multifunctional genes to be more evolutionarily conserved (see Section 1.4 in S1 Text).
In contrast to fly and human, for yeast, when using information about essentiality for growth in rich medium, we do not observe a significant difference in essentiality: 24% of multifunctional genes and 26% of other annotated genes are essential (p = 0.11; Fig 6E). However, in a genome-wide screen of yeast homozygous and heterozygous deletion strains across a variety of conditions, up to 97% of yeast genes are reported as essential in at least one condition [47]. Using these data, we count the number of conditions in which each gene is detected as essential, and find that multifunctional genes are essential in a significantly larger number of conditions than are other annotated genes (p-values 2e-04 and 3e-03 for homozygous and heterozygous screens, respectively; Fig 6F).
As multifunctional genes are more critical than other genes for the survival and normal functioning of the cell, they may potentially also be more likely to be associated with diseases. To address the relationship between gene multifunctionality and involvement in human disorders, we use the gene-disease “morbid map” from the Online Mendelian Inheritance in Man (OMIM) catalog [48], and calculate the fraction of genes with an OMIM annotation among multifunctional genes found for human. We find that 32% of all multifunctional genes are involved in at least one Mendelian disorder, whereas the fraction of other annotated genes involved in at least one Mendelian disorder is 21% (p < 8e-30, hypergeometric test; Fig 7A).
To further investigate the relationship between multifunctional genes and their involvement in human disorders, we look at genes involved in multiple distinct disorders. We map OMIM terms onto the Disease Ontology [49] and identify genes with at least one pair of disjoint OMIM terms (i.e., diseases that fall into separate branches of the Disease Ontology). We consider these genes to be involved in two or more distinct diseases. When considering genes involved in at least one disease from the Disease Ontology, we find that 18% of multifunctional genes are involved in at least two diseases, while only 8% of other such genes are involved in at least two diseases (p < 4e-8; Fig 7B).
One might expect that genes involved in more disorders, as well as multifunctional genes, are more actively studied by the research community, and that this could potentially introduce a study bias affecting our observations [12]. Using the number of PubMed publications associated with a gene as a proxy for how well studied it is, we indeed confirm that multifunctional genes are more actively studied (S4 Fig); however, even when only comparing gene sets with the same number of associated publications, we observe that the fraction of genes associated with disease is higher for multifunctional genes than for other genes (S5 Fig).
Overall, we observe that multifunctional genes are associated with diseases significantly more often than are other annotated genes.
Genes associated with multiple functions may potentially play a more central role in the global functional organization of the cell. Large-scale networks of physical protein-protein interactions provide a comprehensive view of the cellular functional landscape. In order to study how multifunctional genes are positioned in protein interaction networks, we use interaction data curated by BioGRID [39]. We use three measures of centrality: degree, betweenness centrality, and participation coefficient. Degree is the number of interactions in which a protein is involved. Betweenness centrality is the number of shortest paths passing through a node in the network, and nodes with higher betweenness are more globally central in the network. Participation coefficient shows how well a protein’s interacting partners are distributed among clusters in the network, so that proteins with low participation are mostly interacting with proteins from the same cluster, whereas proteins with high participation have their interactions spread across many clusters.
We observe that with respect to all three considered measures, multifunctional genes are significantly more central than other genes (p-values from 2e-13 to 3e-50, Mann–Whitney U; Fig 8). However, not surprisingly, degree is correlated with betweenness and participation (S6 Fig), and thus the correlation between multifunctionality and degree could potentially explain the correlation with the other two more complex measures. In order to test for this, we compare multifunctional and other annotated genes with respect to their betweenness and participation when controlling for degree distribution, and still observe that multifunctional genes have significantly larger betweenness and participation (S6 Fig and S2 Table).
In order to show that our observations are not affected by potential study biases, we repeat the comparisons of degree, betweenness, and participation between multifunctional and other annotated genes in networks containing only interactions from high-throughput experiments (as reported in BioGRID [39] and HINT [50]) and observe similar results (S7 Fig). Furthermore, in order to show that potential bias in the selection of baits in these high-throughput experiments does not affect our conclusions, we compare only the number of bait-to-prey interactions reported in these high-throughput experiments. In particular, we only compare multifunctional and other genes that are baits in these experiments, and observe the same trends (S7 Fig). Overall, we conclude that multifunctional genes are more centrally positioned in protein interaction networks, and this suggests that they may play an intermodular role within interactomes.
The main focus of our analysis thus far has been on multifunctional genes detected using the Biological Process ontology. However, the same procedure for detecting multifunctional genes can be applied to the Molecular Function ontology instead, thereby providing an orthogonal view of gene multifunctionality. For clarity, in this section we call the genes detected as multifunctional using the Biological Process ontology as BP-multifunctional and those detected as multifunctional using the Molecular Function ontology as MF-multifunctional.
We identify sets of MF-multifunctional genes for each organism and observe that MF-multifunctional genes have the same distinct biological properties when compared with other annotated genes as has been reported in the previous sections for BP-multifunctional genes (although some p-values for yeast are above our significance threshold of 5%; see S8, S9 and S10 Figs).
In order to see if the involvement of a gene in multiple biological processes (BP-multifunctional) can be explained by multiple functions of the gene at the molecular level (MF-multifunctional), we directly compare the two sets of multifunctional genes derived from the two ontologies. We observe that 12% to 35% of BP-multifunctional genes are also MF-multifunctional, which constitutes a significant overlap (p < 6e-18; S3 Table), while the remainder may potentially be explained by other modes of gene multifunctionality. In contrast, a gene involved in multiple molecular functions might be expected to have these molecular functions while performing different biological processes, and indeed most MF-multifunctional genes are also BP-multifunctional (56% to 78%; S4 Table). These results are consistent with previous observations made using a simpler multifunctionality definition counting leaf GO terms associated with each protein [15]. Note, however, that the total number of MF annotations is lower than the total number of BP annotations (S3 and S4 Tables), and thus the total number of genes identified as MF-multifunctional is lower than the total number of genes identified as BP-multifunctional (S4 Table).
Most proteins are—at least to some extent—multifunctional. Even within this context, previous experimental studies have identified proteins that perform remarkably different molecular functions [2–6], or that affect several distinct biological processes [7, 8, 10]. These findings suggest the existence of a subset of genes that are endowed with a particularly high degree of functional plasticity. There is increasing evidence that the phenomenon of gene multifunctionality is actually very common; thus, studying multifunctionality at a systems level can help elucidate the functional organization of the cell. In this paper, we introduce a computational approach to systematically identify multifunctional genes using existing functional annotations, and show that multifunctional genes are characterized by distinct properties as compared to other genes. To the best of our knowledge, our work represents the largest-scale characterization of gene multifunctionality to date, with whole-genome analysis across several organisms.
As compared to other studies, our approach specifically addresses some previous weaknesses in handling GO functional annotations. In several previous publications, a simple count of the number of GO terms annotating a gene was used as a proxy for gene multifunctionality [11, 12, 19]. However, idiosyncrasies of GO may result in similar functions or processes being categorized in distinct places in the ontology. In our approach to identify multifunctional genes, we explicitly select semantically distinct terms that co-occur less frequently than expected by chance. Special care is also taken in gauging the effects of study bias, particularly in the case of interaction network properties and disease genes; that is, multifunctional genes may appear more often in the results of various experiments and thus be more actively studied by researchers, and this could potentially introduce a study bias in our analysis. In order to avoid this, we mostly analyze high-throughput and whole-genome data sets. When looking at associations of multifunctional genes with manually curated data (e.g., association with diseases), which could potentially suffer from study bias, we directly correct for this bias. Further, we carry out inter-species comparisons, and observe similar trends across three different organisms, thereby minimizing the effects of organism-specific annotation biases.
A remaining challenge in characterizing multifunctional genes at the genome-wide level is that current knowledge about gene function is far from complete; thus, new experimental information about the function of some genes could result in their reclassification as multifunctional. In the limit, we may expect that nearly all genes are, to varying degrees, multifunctional. Nevertheless, the robustness of our results—both across a diverse set of organisms with distinct functional annotations and biases as well as within a single organism when explicitly controlling for study bias—suggests that we have identified specific biological features that are associated with the degree of functional plasticity of a gene.
We find that gene multifunctionality is associated with several distinct properties that have important functional consequences. In the protein interactome, multifunctional proteins have a tendency to occupy more central and intermodular regions, even after controlling for potential study bias; this suggests that multifunctional proteins connect distinct and more specialized parts of the interactome, and are critical for information flow within the cell. Consistent with their important role within the cell, we also observe that multifunctional genes are more likely to be essential and are more often found to be associated with diseases. At the expression level, multifunctional genes are more broadly expressed across different conditions or cell types than are other genes. It is therefore possible that only subsets of functions are performed by multifunctional proteins under specific conditions or in particular cell types. We also observe that the expression of multifunctional genes appears to be finely regulated, as it involves a larger number of transcription factors than expected. At the molecular level, we find that multifunctional proteins have a larger number of unique domains as compared to other proteins; this is consistent with the wider spectrum of functions that they carry out. However, consistent with previous reports [25], we also find that multifunctional proteins have a higher degree of structural disorder. Determining which of these properties or combinations of properties represent the main mechanism underlying the functional plasticity of a gene is of great interest. It is also possible to speculate that multifunctionality may be achieved via class-specific mechanisms where certain mechanisms may be at play only for a given class of genes.
As part of our analysis, we perform a cross-genomic analysis of gene multifunctionality. We find that multifunctional genes are more evolutionarily conserved than other genes; this may be due to their being under stronger evolutionary pressure as they perform multiple functions, with different functions potentially performed in different conditions. Further, orthologous genes tend to share their propensity for multifunctionality; this suggests that the multifunctionality of many genes may have an early evolutionary origin. It also further supports the validity of our method to detect multifunctional genes, as they are uncovered in each organism independently.
Our method to detect genes annotated with distinct functional terms can be applied to any of the vocabularies in GO, and this allows us to look at the phenomenon of gene multifunctionality from different perspectives. We observe, not surprisingly, that most genes identified as being involved in multiple molecular functions are also identified as participating in multiple biological processes. However, we detect many genes involved in multiple biological processes for which there is no evidence of association with multiple molecular functions. While this may be partly due to the fewer number of molecular function annotations, it also suggests that these genes may perform the same molecular function while carrying out different biological processes, depending upon a spatio-temporal context. Being able to tease apart the conditions under which a specific function is performed by a gene is an important avenue for future research in functional genomics, and could even lead to the development of a context-specific GO vocabulary. In this ontology, the terms used to annotate genes could be qualified with other terms specifying the cell type, the developmental stage, or the stage in the cell-cycle in which a given function is most likely to be carried out by a gene.
In conclusion, a comprehensive understanding of gene and protein function has been a major goal of computational biology since the emergence of the field. In this work, we develop a computational method for genome-wide detection of multifunctional genes using existing functional annotations. We make a number of novel observations about gene multifunctionality across several organisms, as well as confirm some previous findings (including many cases where only anecdotal evidence existed). Overall, our work contributes to a better systematic understanding of the functional landscape of the proteome, and can be the basis for future work in this direction as more specific and detailed functional genomics data become available.
Gene Ontology (GO) [22] terms and gene association data for each organism were downloaded from http://www.geneontology.org/ on July 12, 2013. For the main analysis reported in the paper, we include all functional associations with evidence codes EXP (“Inferred from Experiment”), IDA (“Inferred from Direct Assay”), IMP (“Inferred from Mutant Phenotype”), IGI (“Inferred from Genetic Interaction”), IEP (“Inferred from Expression Pattern”), ISS (“Inferred from Sequence or structural Similarity”), ISO (“Inferred from Sequence Orthology”), ISA (“Inferred from Sequence Alignment”), ISM (“Inferred from Sequence Model”), IGC (“Inferred from Genomic Context”), IBA (“Inferred from Biological aspect of Ancestor”), IC (“Inferred by Curator”), TAS (“Traceable Author Statement”), and NAS (“Non-traceable Author Statement”). We exclude all annotations with the qualifier NOT. We also perform additional analyses restricting ourselves to GO annotations with evidence codes EXP, IDA, IMP, IEP, IC, and TAS; these results are consistent with those reported in the main body of the paper (see S11, S12, S13 Figs and S6 File). For all GO analysis, we use code from the project goatools (https://github.com/tanghaibao/goatools).
We call multifunctional every gene that is annotated with at least “two sufficiently distinct functional terms of comparable specificity,” as explained next. First, to define terms of about equal specificity, we start with the notion of informative terms used previously in the literature [51–54], which selects for a given N all terms that annotate ≥ N genes, but whose descendant terms annotate < N genes. However, we observe that a very general term annotating many genes may have all descendant terms annotating only small numbers of genes, even if it annotates many more than N genes. For example, a fly term imaginal disc-derived wing morphogenesis (GO:0007476) annotates 508 genes, but its descendant terms annotate no more than 82 genes each (248 genes in total), and it may be undesirable to call this term informative for N ≈ 100, as it is actually a much more general term than terms that annotate approximately 100 genes. To overcome this problem, we select the set TN of all terms which annotate ≥ N genes, but < 2N genes, and whose every descendant term annotates < N genes (this includes terms with no descendants). Next, from all genes annotated by terms from TN we extract the genes annotated with at least two such terms. In order to consider annotations by distinct terms only, from the collection of all pairs of selected terms {(t1, t2): t1, t2 ∈ TN}, we further select pairs of terms that are sufficiently distinct. First, we filter out pairs of terms that have a common descendant term, as this may be an indication of similarity between the terms. We also remove all pairs of terms that have pairwise semantic similarity larger than zero [55]; though alternate thresholds of semantic similarity could be used, here we select only pairs of terms whose least common ancestor is the root of the ontology. Finally, terms annotating similar sets of genes may correspond to similar functions, so we filter out all pairs of terms that annotate significantly overlapping sets of genes (hypergeometric test, p < 0.1). A gene co-annotated by some pair of selected terms from TN passing these filters is called multifunctional. In order to focus on more specific biological process terms and avoid considering less informative (i.e., more general) terms annotating a lot of genes, we require that N is not greater than a certain threshold M; we choose M = 120 for the analysis in the main text. The final set of multifunctional genes is given by the union of all sets obtained for different N, where N ranges from 10 up to M, with an increment of 10. We compare multifunctional genes with all other genes that are annotated with any selected term from TN for N between 10 and M (with an increment of 10). We show that, for all our results, the same trends are observed when varying the parameter M (S4 File), the p-value threshold in co-annotation filter (S5 File), and when restricting the analysis to a subset of the most reliable GO annotations (S11, S12, S13 Figs and S6 File).
Protein ortholog information was obtained from version 4 of the Princeton Protein Orthology Database (P-POD) [34, 68]. Two proteins from different organisms are considered orthologous if they belong to the same family, as detected by P-POD using either OrthoMCL or MultiParanoid. For each pair of organisms, we compute how many orthologous pairs of multifunctional genes are found where one gene in a pair is from one organism and the other gene in the pair is from the other organism. To assess significance, we repeat the computation 1000 times with randomization. In each random trial, we permute the labels of multifunctional and other annotated genes within each organism, while considering only genes involved in orthologous relationships. The orthology relationship between genes of different organisms is preserved. Then we compute the average and standard deviation of the counts in random trials along with an empirical p-value of the real count with respect to the randomized counts.
The degree of a vertex is the number of interactions that the corresponding protein has in the network. The betweenness centrality of a vertex v is the number of shortest paths between all pairs of vertices in the network that pass through v, with the shortest paths between two vertices s and t weighed inversely to the total number of distinct shortest paths between them. The participation coefficient [69, 70] of a vertex with respect to a set of clusters in a network is defined as P = 1 − ∑ i( k i k ) 2, where the summation is over all clusters, k is the vertex degree, and ki is the number of edges going from the vertex to vertices from the cluster i. The rationale is to have P = 0 if all edges from the vertex go to a single cluster, and to have p closer to 1 if edges from the vertex are more uniformly distributed over clusters. To find clusters in the network, we use the SPICi clustering algorithm [71] with parameters optimized with a simple exhaustive search procedure to approximately maximize Newman’s modularity [72], as described earlier [73]. For network analysis, we use the python interface to the igraph library, version 0.6.5 (http://igraph.sourceforge.net/).
|
10.1371/journal.pntd.0004332 | Whole Genome Sequencing of Mycobacterium africanum Strains from Mali Provides Insights into the Mechanisms of Geographic Restriction | Mycobacterium africanum, made up of lineages 5 and 6 within the Mycobacterium tuberculosis complex (MTC), causes up to half of all tuberculosis cases in West Africa, but is rarely found outside of this region. The reasons for this geographical restriction remain unknown. Possible reasons include a geographically restricted animal reservoir, a unique preference for hosts of West African ethnicity, and an inability to compete with other lineages outside of West Africa. These latter two hypotheses could be caused by loss of fitness or altered interactions with the host immune system.
We sequenced 92 MTC clinical isolates from Mali, including two lineage 5 and 24 lineage 6 strains. Our genome sequencing assembly, alignment, phylogeny and average nucleotide identity analyses enabled us to identify features that typify lineages 5 and 6 and made clear that these lineages do not constitute a distinct species within the MTC. We found that in Mali, lineage 6 and lineage 4 strains have similar levels of diversity and evolve drug resistance through similar mechanisms. In the process, we identified a putative novel streptomycin resistance mutation. In addition, we found evidence of person-to-person transmission of lineage 6 isolates and showed that lineage 6 is not enriched for mutations in virulence-associated genes.
This is the largest collection of lineage 5 and 6 whole genome sequences to date, and our assembly and alignment data provide valuable insights into what distinguishes these lineages from other MTC lineages. Lineages 5 and 6 do not appear to be geographically restricted due to an inability to transmit between West African hosts or to an elevated number of mutations in virulence-associated genes. However, lineage-specific mutations, such as mutations in cell wall structure, secretion systems and cofactor biosynthesis, provide alternative mechanisms that may lead to host specificity.
| Mycobacterium africanum consists of two lineages, lineages 5 and 6, within the Mycobacterium tuberculosis complex (MTC) that cause human tuberculosis in West Africa, but are found rarely outside of this region. Our analysis of the whole genome sequences of 26 lineage 5 and 6 isolates, and 66 isolates from other lineages within the MTC, reveal that M. africanum does not meet modern criteria to be considered an independent species. We analyzed drug resistance-associated genes and found that drug resistance evolves within these lineages through similar mechanisms as observed for the rest of the MTC in Mali. Though we did not see an elevated number of mutations in virulence-associated genes in these two lineages, we identified a number of lineage-specific mutations, pseudogenes and changes in gene content that may impact virulence and host specificity, and improve, overall, our understanding of what make these lineages unique.
| Mycobacterium africanum is a member of the Mycobacterium tuberculosis complex (MTC) that causes up to half of all tuberculosis cases in West Africa [1]. First identified by Castets in 1968, it was originally characterized as having biochemical characteristics intermediate between Mycobacterium tuberculosis, which consists of lineages 1, 2, 3, 4, and 7 and is the main cause of human tuberculosis, and Mycobacterium bovis, an animal-adapted lineage that causes bovine tuberculosis [2]. Later work divided M. africanum into two lineages, M. africanum West African type I and M. africanum West African type II, which became known as lineages 5 and 6, respectively, within the MTC [3, 4].
Lineages 5 and 6 cause a disease similar to classically defined M. tuberculosis, although it has been suggested that human disease caused by these lineages may differ compared to that caused by lineages 1–4. For example, patients with lineage 6 disease have been reported to show attenuated ESAT-6 responses compared to patients with classical M. tuberculosis lineage disease [5, 6]. In addition, in liquid culture systems it has been reported that M. africanum has a slower growth rate with a larger bacillary size than M. tuberculosis [7, 8]. While some studies have found that M. africanum is less virulent than M. tuberculosis, both in animal models and human patients [7, 9–11], others show that there is no difference [12]. Though these contradicting results may be due to differences in the study populations, they underscore how little is known about lineages 5 and 6.
Contributing to this lack of knowledge, while lineages 1–4 are widely distributed around the globe, lineages 5–7 are limited to certain regions of Africa [13]. Lineage 7 has only been found in Ethiopia [14], and lineages 5 and 6 are found almost exclusively in patients living in West Africa, with very few cases occurring outside of this region, mostly involving recent immigrants from West Africa [1]. The reason for the apparent geographic restriction of lineages 5 and 6 is unknown. One hypothesis is the presence of an undiscovered animal reservoir endemic to West Africa, which is supported by the close relationship between lineages 5 and 6 and the animal-adapted lineages of the MTC [15, 16]. Another hypothesis is that lineages 5 and 6 have a unique predilection for humans with genetic backgrounds common in West Africa. In fact, using a retrospective epidemiological study of the MTC in Ghana, Asante-Poku et al. showed that lineage 5 is associated with the Ewe ethnic group [17]. A third hypothesis is that lineages 5 and 6 are unable to compete with other lineages outside of West Africa, either due to loss of fitness or decreased transmissibility, thus explaining their limited global distribution [7].
Historically, mycobacterial subspecies were defined by biochemical assays, but, as genetic tools became more readily available, it is now possible to identify genomic regions that define MTC lineages [18]. The publication of the whole genome sequence of M. africanum GM041182, a single lineage 6 strain, provided valuable insights into the genetics of this lineage [19]. For instance, the authors identified lineage 6-specific pseudogenes, a novel region not present in M. tuberculosis, and single nucleotide polymorphisms (SNPs) in key genes, all of which may play a role in the geographic restriction of lineage 6. A later study sequenced four additional lineage 6 isolates and was able to confirm many of these findings, but also showed that not all mutations identified in M. africanum GM041182 are shared by other members of this lineage [8]. To our knowledge, no study has closely analyzed the genetics of lineage 5.
From these studies, it is clear that more sequenced isolates are needed to fully characterize the genetics of lineages 5 and 6 and to illuminate mechanisms that may explain its geographic isolation. Toward this end, we sequenced 92 clinical MTC isolates from Mali, a country in West Africa in which 26.2% and 1.6% of tuberculosis cases are caused by lineage 6 and lineage 5, respectively [20] [1]. Using these and previously published data, we performed both alignment- and assembly-based comparative analyses to further refine our understanding of lineage-specific genomic features that might explain the geographic distribution of lineages 5 and 6. To our knowledge, this is the largest collection of lineage 6 strains sequenced to date, and the first in depth whole genomic characterization of lineage 5.
101 strains were selected from clinical isolates collected in Bamako, Mali [20], and included all strains identified by spoligotyping as M. africanum, M. tuberculosis T1, or M. bovis. Of these strains, 92 were still viable and were submitted for whole genome sequencing. These 92 strains will be referred to as the “Mali Collection” (S1A Table). In addition, to improve MTC lineage representation, we selected additional whole genome assemblies that matched the quality of our assemblies. These included four finished M. bovis genomes available from GenBank (M. bovis AF2122/97 [21], M. bovis BCG Mexico [22], M. bovis BCG Pasteur 1173P2 [23], and M. bovis BCG Tokyo 172 [24]), a set of 40 M. tuberculosis strains (9 lineage 1 strains, 12 lineage 2 strains, 7 lineage 3 strains, and 12 lineage 4 strains) from South Africa [25], the finished M. africanum genome from Genbank (M. africanum GM041182) [19], and our outgroup, M. canettii CIPT 140010059 [26]. Combined with the Mali Collection, these 137 strains will be referred to as the “Assembly Collection” (S1B Table). Finally, all 161 strains (122 lineage 2, two lineage 3, and 37 lineage 4) from a study in China were included in the variant analysis to improve geographical and lineage representation [27]. The samples from the China study (S1C Table) combined with the samples from South Africa and Mali (for a total of 289 strains) will be referred to as the “Alignment Collection”.
The study protocols for the Mali samples were approved by the Ethics Committee of the University of Bamako and the Institutional Review Board of the National Institute of Allergy and Infectious Diseases, National Institutes of Health (NIAID/NIH), Bethesda, MD, USA. For all samples, written informed consent was obtained from study participants prior to cohort enrollment [20]. For the South African samples, Biomedical Research Ethics Council (BREC) approval from the University of KwaZulu-Natal was granted for collection of sputum specimens from study participants and for whole genome sequencing of clinical strains. Written informed consent was obtained from study participants prior to cohort enrollment, or waived by BREC [25].
Drug resistance to isoniazid, rifampicin, ethambutol and streptomycin was tested for all Mali strains as previously described [20]. We confirmed those results by submitting 17 strains to National Jewish Health in Colorado for agar proportion testing of isoniazid, rifampicin, ethambutol, ofloxacin, niacin, kanamycin, ethionamide, capreomycin, amikacin, cycloserine and para-aminosalicylic acid, as well as radiometric testing of ciprofloxacin and pyrazinamide. The agar proportion results confirmed the mycobacterial growth indicator tube (MGIT) tests performed in Mali. Genotypic drug resistance was determined for rifampicin, isoniazid, ethambutol, streptomycin, ofloxacin, kanamycin and ethionamide using genetic markers from line-probe assays (S2 Table).
Extraction of genomic DNA was performed on 10 mL cultures grown in 7H9 broth using the CTAB-lysozyme method as previously described [28]. Library preparation and whole genome sequencing (WGS) were performed as previously described [29–31]. GenBank accessions for all strains used in this analysis can be found in S1B Table, along with assembly statistics for the new sequences generated at the Broad Institute (92 sequences from Mali generated for this study, and 40 sequences from South Africa).
All genomes in our Assembly Collection were uniformly annotated by transferring annotations from M. tuberculosis H37Rv. The reference M. tuberculosis H37Rv genome (accession CP003248.2) was aligned to draft assemblies using Nucmer [32]. This alignment was used to map reference genes over to the target genomes. Using this methodology, annotations were successfully transferred onto all 137 strains for 3466 of the M. tuberculosis H37Rv genes; the rest of the M. tuberculosis H37Rv genes transferred to a subset of the genomes.
For those genes not cleanly mapping to M. tuberculosis H37Rv, the protein-coding genes were predicted with the software tool Prodigal [33]. tRNAs were identified by tRNAscan-SE [34] and rRNA genes were predicted using RNAmmer [35]. Gene product names were assigned based on top blast hits against the SwissProt protein database (> = 70% identity and > = 70% query coverage), and protein family profile search against the TIGRfam hmmer equivalogs. Additional annotation analyses performed include Pfam [36], TIGRfam [37], Kyoto Encyclopedia of Genes and Genomes (KEGG) [38], clusters of orthologous groups (COG) [39], Gene Ontology (GO) [40], enzyme commission (EC) [41], SignalP [42], and Transmembrane Helices; Hidden Markov Model (TMHMM) [43].
Reads from each isolate were aligned against the 43 spacer sequences traditionally used in wet lab spoligtyping [28, 44]. From these alignments, the number of matching reads was used to determine if the spacer was present. The spacer was considered absent if the read count total was in the lowest quartile of counts. Spacers were defined as present by using a Bonferroni corrected p-value based on an exponential distribution of the average absent spacer counts. If the p-value was <0.01 the spacer was considered to be present. The spacer pattern was matched to the SITVITWEB database to generate a named spoligytpe for each isolate and to determine the spoligotype international type (SIT) [45].
SYNERGY2 [46–48], available at http://sourceforge.net/projects/synergytwo/, was used to identify cluster-based orthogroups across our Assembly Collection of 137 genomes, which we will refer to as “SYNERGY orthogroups”. In addition, for each M. tuberculosis H37Rv gene, we defined a second set of annotation transfer-based ortholog groups as the set of genes for which annotations were transferred from this M. tuberculosis H37Rv gene in our annotation protocol, which we will refer to as “M. tuberculosis H37Rv-based orthologs”. Genes without M. tuberculosis H37Rv orthologs were manually examined in the context of their SYNERGY orthogroups to identify lineage-specific novel genes.
Phylogenetic trees were generated by applying RAxML [49] to a concatenated alignment of 3343 single-copy core SYNERGY orthogroups (excluding orthogroups with paralogs) across all 137 organisms. Bootstrapping was performed using RAxML’s rapid bootstrapping algorithm (1000 iterations).
Calculations of ANI were done as previously described [50, 51] using the SYNERGY orthogroups calculated from the Assembly Collection.
PAUP [52] was used to reconstruct gain and loss of M. tuberculosis H37Rv-based orthologs at ancestral nodes of the Assembly Collection phylogenetic tree using parsimony. In order to analyze changes in gene content, we used a cost matrix with values of 10 for a gene gain, 5 for a gene loss, and 0.2 for an increase or decrease in copy number. We looked for orthologs found within all members of one clade, and absent in other clades. As a further filter, we also required that orthogroups be found in >80% of the clade of interest, and <20% of other strains. We performed this analysis for four key clades: lineage 5, lineage 6, the clade including M. bovis and lineage 6, and the clade including lineages 5, 6 and M. bovis.
In addition, we selected the Pfam gene categories most expanded or reduced in each clade of interest. We determined significance using Fisher’s test (Q<0.05). For each of the clades described above, we compared the strains below this node versus all other strains in our analysis.
For our Alignment Collection, reads were mapped onto a reference strain of M. tuberculosis H37Rv (GenBank accession number CP003248.2) using BWA version 0.5.9.9 [53]. In cases where read coverage of the reference was greater than 200x, reads were down-sampled using Picard [54] prior to mapping. Variants, including both single nucleotide polymorphisms (SNPs) and multi-nucleotide polymorphisms, were identified using Pilon version 1.5 as described [29] and were used to construct phylogenetic trees using FastTree [55].
We defined lineage-specific variants for lineage X, as those occurring in at least 95% of the strains of lineage X (true positive rate >95%), missing in less than 5% of strains of lineage X (positive predictive value >95%), occurring in less than 5% of the strains that do not belong to lineage X (true negative rate >95%) and not occurring in at least 95% of strains not belonging to lineage X (negative predictive value >95%). The absolute number of true positives must exceed seven. Formulas are schematically presented in S3 Table.
Mutations were considered M. africanum-specific (lineage 5 and 6-specific, identified as LIN-Maf in S4 and S5 Tables) if they met these cutoffs for lineage 5 and 6 combined, and were present in both lineage 5 strains. Similarly, mutations were considered M. tuberculosis-specific if they met these cutoffs for lineages 1–4 combined. No M. tuberculosis-specific mutations were identified. Due to inclusion of only two lineage 5 strains in our dataset, no lineage-specific variants were identified in lineage 5. Thus, for this lineage only, we used a less stringent requirement to define lineage-specific variants: we required that variants be present in both lineage 5 strains and in <5% of the strains in each other lineage. We classified each gene containing a lineage-specific variant into functional group categories, including GO [40], KEGG [38], Pfam [36], and COG [39]. We then evaluated enrichment using Fisher's Exact test and corrected for multiple comparisons using the Storey method for functional group categories [56].
A pseudogene was defined as any gene that had a loss of function mutation anywhere within the coding sequence. Loss of function mutations were defined as nonsense mutations, or insertions or deletions with lengths that were not multiples of 3 base pairs or were greater than 30 base pairs. Lineage-specific pseudogenes were determined using the same definitions as for variants on a per gene basis (positive predictive value > 95%, negative predictive value >95%, true positive rate >95%, true negative rate >95%, number of true positives >7, with the exception of lineage 5, which used the SNP cutoffs of pseudogene in both lineage 5 strains and in <5% in each other lineage).
The effect of select non-synonymous mutations on protein function was assessed using the online version of SIFT at default settings [57], unless there was low confidence in the prediction, in which case SIFT was run for each of the four available databases (UniRef90 from April 2011 [default], UniProt-SwissProt 57.15 from April 2011, UniProt-TrEMBL from March 2009 and NCBI nonredundant from March 2011). Peptide binding was predicted using the NetMHCII online tool with default settings [58].
Our collection of 92 clinical MTC strains was isolated from patients presenting with pulmonary tuberculosis at Point G, Bamako, Mali between 2006 and 2010 as part of a cross-sectional study to analyze the diversity of the MTC in Mali [20]. All patients were Mali natives, with the exception of one patient born in central Africa (S1A Table). We sequenced this collection using the Illumina sequencing platform, and the resulting reads were both assembled into contigs and aligned against the M. tuberculosis H37Rv reference genome. Based on our phylogenetic reconstructions, our collection included one lineage 1, two lineage 2, zero lineage 3, 63 lineage 4, two lineage 5 and twenty-four lineage 6 strains (Fig 1). The spoligotype distribution of our collection is representative of what has previously been observed in West Africa, except that we had a higher proportion of SIT53 (T1) strains and a lower proportion of SIT181 (AFRI_1) strains (Figs 1 and S1) [45]. In order to perform statistical comparisons of the M. tuberculosis, M. africanum and M. bovis lineages, our newly sequenced dataset (the “Mali Collection”) was combined with data from additional strains from GenBank and South Africa (“Assembly Collection”, Fig 2), as well as data from China (“Alignment Collection”; see Materials and Methods and S1 Table). These additional comparator genomes enabled us to examine in detail the distinguishing characteristics of lineages 5 and 6 that might explain their geographic restriction.
Since this represents the largest collection of whole genome sequences of lineage 5 and 6 strains to date, we used our Assembly Collection to conduct a detailed examination of their phylogeny and characteristics in relation to other members of the MTC, including M. bovis and M. tuberculosis. M. bovis is considered an animal strain that mainly infects cattle and rarely humans, while M. tuberculosis is human adapted, and lineages 5 and 6 are thought to be intermediate between the two [1, 15]. Using our Assembly Collection, we constructed a high-resolution phylogenetic tree using 3,343 single-copy core orthogroups (sets of orthologs) conserved across all 137 strains (Materials and Methods). This tree was rooted using the outgroup M. canettii and agreed with phylogenies observed in other studies, including the fact that each of the lineages was clearly separated from the other, with lineage 5 being more closely related to human-adapted strains and lineage 6 being more closely related to M. bovis (Fig 2) [13, 15].
It has been previously shown, using average nucleotide identity (ANI) analysis, that separate bacterial species share <65–90% of genes and have no more than 94–95% ANI among shared genes [50, 51]. Using gene content and nucleotide variation among shared genes, we examined the genetic distances between strains within the Assembly Collection to understand how mycobacterial species fit within this framework. In agreement with previous studies showing the close relationship between MTC subspecies, including M. africanum, we observed that there was little diversity between the lineages analyzed [61]. Strikingly, values from inter-lineage comparisons of M. tuberculosis, M. bovis, and M. africanum strains overlapped those from intra-lineage comparisons, showing very little separation, with >99% ANI and >94% fraction of shared genes (Fig 3). These results are in agreement with previous observations that these different organisms should not, in fact, be named different species [61].
In contrast, MTC pairwise comparisons with M. canettii revealed a clear separation between the two groups, suggesting that they occupy distinct niches (Fig 3). M. canettii is a smooth tubercle bacilli that causes human tuberculosis in East Africa and is considered an emerging pathogen in some parts of the world, but its natural host(s) and reservoirs remain unknown [62]. Thus, it might be argued, based on these data and the traditional cutoffs set by ANI analysis, that all MTC members should be named the same species, and that even M. canettii should be included since pairwise identities with MTC exceeded these thresholds (Fig 3). However, as Smith et al. have previously discussed [61] changes in nomenclature can cause confusion in the literature, and so we will continue to refer to M. africanum-associated lineages as either lineage 5 or 6 within the MTC.
Despite the fact that lineages 5 and 6 are so closely related to lineages 1–4, as demonstrated by Fig 3, they are still unique in being geographically restricted compared to these other lineages. One hypothesis for this restriction is that they are less fit, unable to compete with other lineages within the MTC. To examine this possibility, we looked within the Mali Collection for clues that lineage 6 strains were not as diverse as strains from lineage 4, the other predominant lineage within the region. We analyzed the breadth of pairwise diversity within lineage 6 using the ANI output and compared this diversity to that of lineage 4 strains isolated within Mali. ANI diversity was not statistically different when comparing these two groups of strains (Fig 4). Although this result does not eliminate the possibility of differing ecologies, such as an animal reservoir for lineage 6, as has previously been hypothesized [16], it does suggest that lineage 6 has not undergone a recent selective sweep or population bottleneck that would make lineage 6 populations circulating within Mali less diverse than lineage 4 populations [63].
In addition to being diverse, we also observed highly similar lineage 6 strains among this collection. Three pairs of lineage 6 strains were separated by less than 10 SNPs relative to M. tuberculosis H37Rv (Fig 1B; see Materials and Methods), including isolates from both HIV-positive and immunocompetent patients. There were also six such clusters within lineage 4. A cutoff of 12 SNPs has previously been used to determine recent transmission [64]. Thus, strains separated by less than 10 SNPs provide evidence of transmission, suggesting that 6 of 24 (25%) of our lineage 6 strains and 13 of 63 (21%) of our lineage 4 strains were involved in recent transmission events, confirming previous observations based on alternative genotyping approaches that there is robust ongoing transmission of lineage 6 within this region [9].
Given the reports of lineages 5 and 6 strains having decreased virulence [7, 9–11], we hypothesized that altered virulence may contribute to geographical restriction, either due to changes in host requirements or to a reduction in fitness. To test this hypothesis, we examined lineage-specific pseudogenes (truncated genes) and non-synonymous SNPs in known essential genes, slow growth genes, and genes required for virulence in mice and growth in macrophages to determine whether lineages 5 and 6 had an enrichment of defects in these genes that might contribute to overall altered virulence [65–67]. Although both lineages 5 and 6 had lineage-specific mutations in these gene categories, so did other lineages (S4A and S5 Tables), and the proportion of mutated genes in lineage 6 was not significantly different from that of the other MTC lineages [8] (Fig 5). Lineage 4 was not included on this graph because it only had one lineage-specific mutation in an intergenic region when aligned to M. tuberculosis H37Rv, which is a member of lineage 4, and lineage 5 was excluded due to low sample size. We performed a similar analysis on the full length of genes encoding known T cell antigens as defined by Comas et al. [4] to explore whether alterations in these genes might be restricting host specificity, but again we observed no significant difference in the proportion of lineage 6-specific mutations that fell within these genes as compared to lineages 1, 2 and 3 (Fig 5). Similarly, we looked for enrichment of lineage-specific mutations in COG, GO, KEGG, Pfam and TIGRfam gene categories, but found no enrichment in any of these categories, either for pseudogenes or non-synonymous SNPs (Q > 0.05). These results corroborate our observations from ANI that the lineages of the MTC are very similar in their overall genetic composition and suggest that lineage 6 is not enriched for mutations in virulence genes relative to other lineages. However, while the overall number of mutations in virulence genes was not enriched, we identified mutations in these genes that might have an impact on virulence that will be discussed below.
Studies have shown that lineages 5 and 6 evolve drug resistance less often compared to other MTC lineages, including the study from which these sequenced strains were obtained [20, 68]. Thus, one hypothesis for the limited geographic range of lineages 5 and 6 could be decreased fitness relative to strains better able to evolve antibiotic resistance. In this case, we might expect that mutations driving drug resistance in these two lineages would be different from those evolving in more successful lineages. Thus, we analyzed our newly sequenced strains from Mali for the presence of mutations known to confer drug resistance and used in common nucleic acid-based commercial tests [59, 60] for the detection of drug resistance [69–75] (S2A Table). Forty (60%) strains in lineages 1–4 and only four (15%) of the lineage 5 and 6 strains were phenotypically resistant to at least one of the four tested drugs. We observed that mutations used in commercial tests were sensitive in detecting phenotypic resistance to rifampicin, isoniazid and ethambutol (S2B Table and Figs 1 and S1).
Streptomycin resistance mutations were not included among our list of known resistance mutations (S2A Table). Therefore, we searched for potential resistance mutations in a set of genes previously known to affect streptomycin resistance, including rrs, rpsL, and gidB [76–78]. We identified a point mutation in gidB that caused a non-synonymous change (leucine to serine at residue 79) that is predicted to affect protein function [57] (S1D Fig; see S1 Text for more details). This mutation was found in 23 streptomycin resistant strains and no streptomycin susceptible strains in our dataset and likely represents a previously uncharacterized mutation that confers resistance to this drug. Previous studies have identified loss of function mutations in gidB affecting streptomycin resistance [77], as well as point mutations in the region of gidB close to residue 79, including at residues 75 and 82 [78].
In addition, we identified known mutations in genes associated with resistance to drugs that were not phenotypically assessed, including ofloxacin, kanamycin, and ethionamide. Using the list of mutations in S2A Table, we found that 25 (38%) of the Mali strains belonging to lineages 1, 2 or 4 could be classified as MDR (multi-drug resistant; resistant to isoniazid and rifampicin), and two (3%) could be classified as pre-XDR (pre-extensively drug resistant; resistant to isoniazid, rifampicin, plus either ofloxacin or kanamycin). In contrast, three (11%) of the lineage 5 and 6 strains could be classified as MDR, and one (4%) could be classified as pre-XDR. The presence of these pre-XDR strains is of particular concern, as XDR has not been reported in Mali, and testing is not currently performed routinely for second line antibiotics [79, 80].
Similar resistance-conferring mutations were found among the lineages (S2 Fig). Although we cannot eliminate the possibility of cross resistance and other alternate genetic mechanisms of lineage 5 and 6 drug resistance, or of differences in drug tolerance or rates of persister cells, it appears that the mechanism of genetic drug resistance was similar between lineages 2, 4, 5 and 6. Thus, although the sample size was small, our results suggest that drug resistance, while less frequent in lineage 6, evolves through acquisition of similar mutations as observed in lineages 2 and 4 in Mali, including combinations of mutations leading to pre-XDR, and that this resistance could be detected using current molecular diagnostic approaches.
Previous analyses pinpointing lineage 6-specific genomic features have compared limited numbers of strains, which might have caused these studies to miss important features or to identify features that are not actually found in a broader set of lineage 6 strains [8, 19]. Also, these studies have not examined genomes of lineage 5 in detail. Using both our Alignment and Assembly Collections, containing representatives from lineages 1 through 6 and M. bovis, we sought to robustly identify distinguishing features of lineages 5 and 6, focusing on traits that could have caused geographic restriction. Using our Assembly Collection, at each node labeled A-D in Fig 2 (representing genetic diversification events that may correlate with ecological specialization), we identified gene gains and losses (Table 1; Materials and Methods). Many of our findings agreed with previous observations describing regions of difference, as determined through genomic hybridizations [3, 18]. However, we also identified a small number of genes that were not previously identified as being part of these known regions of difference (see S1 Text for full details), including a gain of genes encoding a PE-PGRS and hypothetical protein at the last common ancestor of lineage 6 and M. bovis, the loss of Rv1523 (a methyltransferase) and Rv3514 (PE-PGRS57) in lineage 5, and the loss of a gene encoding a TetR family regulator and the gain of one PPE protein-encoding gene at the last common ancestor of lineages 5, 6 and M. bovis.
In addition, using our alignments to M. tuberculosis H37Rv, we identified a number of lineage-specific mutations, including pseudogenes that affect protein function (Tables 2, S4, S5 and S6). From these data, we identified 681 lineage 6-specific mutations shared across all lineage 6 strains, including eight truncated pseudogenes. These data also provided the first in-depth analysis of lineage 5 assemblies, which revealed 952 lineage-specific mutations and 43 pseudogenes as shared by our two lineage 5 strains (see S1 Text). The larger number for lineage 5 compared to other lineages likely results from our small sample size. Key categories of lineage-specific mutations and pseudogenes that might contribute to the geographic restriction of lineages 5 and 6 are discussed below, and in more detail in the S1 Text.
One distinguishing clinical characteristic of lineage 6 is an attenuated T cell response to ESAT-6, one of the proteins secreted through the ESX secretion system, as compared to patients infected with lineages 1–4 [5]. This altered immune response supports the hypothesis that lineage 5 and 6 have specificity for a particular host immunogenic background. Although our data cannot address whether ESAT-6 production has been affected, we observed non-synonymous polymorphisms, including indels, in genes encoding ESX secretion systems that could contribute to the different immune responses of lineage 6-infected patients (Table 3). Furthermore, we observed lineage-specific mutations in ESX-encoding genes in all lineages, suggesting that each lineage may have unique interactions with the host (Table 3; S1 Text).
Lineages 5 and 6 had lineage-specific mutations, including pseudogenes, in genes encoding multiple components of cofactor biosynthetic pathways, including molybdenum, vitamin B12, and vitamin B3 (S1 Text and Tables 3, S4 and S5). Molybdenum cofactors are key catalysts for redox reactions, and are hypothesized to have played an important role in the evolution of pathogenic mycobacteria [81]. In addition, mycobacteria are one of the few bacterial pathogens with the ability to synthesize vitamin B12 [82]. Thus, both of these cofactors have specifically evolved in mycobacteria and loss of these cofactor biosynthetic pathways could affect the function of proteins that use these cofactors, which include proteins that are important for many cellular functions. These mutations may affect the host range of lineages 5 and 6, supporting the hypothesis of a unique host preference.
It has been shown previously that M. africanum GM041182 has a distinct physiology as compared to that of M. tuberculosis H37Rv, including a larger cell size and slower growth rate [7]. Possibly explaining these differences, we identified lineage 6-specific non-synonymous SNPs in genes encoding the L,D transpeptidases, ldtA and ldtB (Rv0166c and Rv2518c), previously shown to form cross-linkages within peptidoglycan (Tables 3 and S4A) [83] and to be key drivers of cell shape, size, surface morphology, growth and virulence [84]. Lineage 5 also contained a non-synonymous SNP predicted to affect LdtA protein function (Tables 3 and S4A). No other lineages had a lineage-specific mutation in an L,D-transpeptidase.
We observed that lineages 5 and 6 had lineage-specific mutations in genes encoding adenylate cyclases, the enzymes that synthesize cyclic AMP (cAMP), an important cell signaling molecule. Although the affected genes were different between the two lineages, no other lineage had lineage-specific mutations predicted to affect adenylate cyclase function. Deletion of one of the 17 adenylate cyclases in M. tuberculosis, Rv0386, has been shown to reduce virulence and alter the immune response [85]. Bentley et al. also previously found that this gene was a pseudogene in M. africanum GM041182, although here we find that pseudogenization of Rv0386 was not lineage specific (S6A Table). Nevertheless, given the number of affected adenylate cyclases, there may be differences in cAMP signaling within lineages 5 and 6, leading to altered pathogenicity.
In order to shed light on the reported lower rates of drug resistance in lineages 5 and 6, we screened our lineage-specific mutations to investigate if there were any changes in known drug resistance genes that were not on the list of mutations used before and that might affect the development of antibiotic resistance [20, 68]. In lineage 6, we observed two lineage-specific non-synonymous mutations in rpoB, and one lineage-specific non-synonymous mutation in embC (S1A and S1C Fig and Tables 3 and S4A) not previously implicated in antibiotic resistance. Lineage 5 strains had non-synonymous mutations in genes encoding AtpH (Rv1307) and AtpG (Rv1309), both of which are subunits of ATP synthase [86] (S4A Table), and a target of bedaquiline, a new antibiotic reserved for the treatment of drug resistant tuberculosis [87]. Both of these mutations were predicted to affect protein function by SIFT [57], and may affect bedaquiline efficacy in countries with a high proportion of patients infected with lineage 5. Thus, both of these lineages have non-resistance conferring mutations in genes associated with drug resistance that might influence the frequency at which drug resistance develops in these lineages.
M. tuberculosis H37Rv contains four mammalian cell entry (MCE) operons, which play an important role in mycobacterial virulence [88]. In addition to confirming earlier reports that lineage 6 strains lacked one of these four operons (operon 3; Table 1) [18], we observed lineage-specific mutations in several of the other MCE operons (lineage 6 had mutations in operons 1 and 2; lineage 5 had mutations in operons 1 and 3). We also observed a non-synonymous mutation in mce1B that was shared by lineages 5 and 6 strains and was predicted by SIFT to affect Mce1B protein function [57]. In comparison, the other lineages had nearly identical MCE operons as compared to M. tuberculosis H37Rv (Tables 3 and S4A).
Our study describes the largest collection of sequenced lineage 6 isolates to date, and, to our knowledge, the first in-depth analysis of the genetics of lineage 5. Through our work, we have characterized the genetic basis of antibiotic resistance in lineage 6 strains from Mali, shown that M. africanum and M. tuberculosis are part of the same species, and better defined the mutations and changes in gene content that typify these lineages. Collectively, this work provides insights into these understudied lineages and provides testable hypotheses as to why they are geographically restricted.
We evaluated 92 Mali MTC isolates using both assembly and alignment-based approaches. Our assemblies revealed several new regions of difference and our alignments identified smaller lineage-specific changes. In addition to our conclusion that M. africanum is not a separate species, we observed that some M. africanum-M. tuberculosis pairs of strains have greater average nucleotide identity than some pairs of strains from within the same lineage. Furthermore, our ANI data demonstrated that there is comparable diversity in lineages 4 and 6, suggesting that lineage 6 has not undergone a recent population bottleneck. This emphasizes the extremely close relationship between all MTC lineages, highlighting the role that small changes within the MTC have played in geographical restriction and altering host preferences. Since our assemblies were of very high quality, we were able to observe changes in genes that previous studies could not, thus providing a prioritized list of genes for investigating lineage 5 and 6 characteristics.
One hypothesis for the geographical restriction of lineages 5 and 6 is the presence of an unknown non-human reservoir. M. africanum has been found in animals, including monkeys, cows, pigs and hyrax [89–94]. Unfortunately, given genomic data from human clinical isolates alone, we cannot address this hypothesis directly. However, given the similar level of diversity between lineage 4 and 6 in Mali and the evidence of person-to-person transmission, a non-human reservoir seems unlikely to explain the geographic restriction, as lineage 6 appears well adapted to spread in humans living in this geographic setting, unlike M. bovis in this and other settings [95–97].
Another hypothesis for why lineages 5 and 6 occur almost exclusively in West Africa is a preference for hosts of West African ethnicity, supported by previous evidence, including a study linking lineage 5 to the Ewe ethnic group [17]. We identified lineage-specific mutations in ESX genes in every lineage, indicating that each lineage may interact uniquely with the host immune system. Mycobacteria have five ESX secretion systems, also known as type VII secretion systems, which secrete small proteins across the bacterial cell envelope and are important to mycobacterial virulence [98, 99]. For example, ESX-1 secretion is lost as part of RD1 in M. bovis BCG vaccine strains, resulting in loss of ESAT-6 and CFP-10 secretion, and thus attenuation of the bacterium [100, 101]. The lineage-specific mutations in ESX genes could lead to alterations in the pathogen-host immune interaction, resulting in a requirement in lineages 5 and 6 for the West African immune system. In fact, an altered response to ESAT-6 in patients infected by lineage 6 has previously been reported [5]. Thus, the specific ESX mutations in lineages 5 and 6 could represent adaptations to the niche of the West African host.
Lineage-specific mutations in cobalamin biosynthesis could also contribute to adaptation of these lineages to the specific ecological niche of the West African host. The hypothesis of adaptation to a different host cofactor environment for lineages 5 and 6 is supported by several studies that have found increased levels of vitamin B12 plasma concentrations in West Africans compared to Europeans and Mexicans [102, 103]. One unique characteristic of mycobacteria compared to many other bacterial genera is that they are capable of synthesizing vitamin B12. Furthermore, vitamin B12 may play a crucial role in M. tuberculosis infection [82, 104, 105]. Thus, the lineage-specific mutations in cobalamin pathways in lineages 5 and 6 may alter these strains’ ability to synthesize vitamin B12, which may be tolerated in West African hosts with higher levels of plasma B12. Adaptation to this B12-rich West African niche might prevent these lineages from infecting other ethnic groups with lower B12 bioavailability; however, further studies would be required to confirm this hypothesis.
A third hypothesis for the geographic restriction of lineages 5 and 6 is that they are less fit, either for transmission or in-host virulence, resulting in a decreased ability to survive outside of West Africa. Several papers have shown no difference in transmission rates between M. tuberculosis-associated strains and M. africanum-associated strains [5, 9, 106, 107]. In agreement with previous findings, our Mali Collection revealed three pairs of lineage 6 strains separated by 10 or fewer SNPs when aligned to M. tuberculosis H37Rv, suggesting recent transmission of strains between patients within the ethnic backgrounds prevalent in Mali [64]. That these transmission events were not exclusive to HIV positive patients suggests that a compromised immune system is not required for a transmission event. These results indicate that lineages 5 and 6 do not have a reduced ability to transmit.
A decrease in fitness could also be reflected in a decrease in virulence. It has been hypothesized that M. africanum is less virulent within humans, mice and guinea pigs than is M. tuberculosis [7, 9–11]. However, lineage 6 was not enriched for mutations in virulence and growth-related genes compared to lineages 1, 2 and 3, suggesting that lineage 6 does not contain an overall numerical loss of virulence or growth-associated genes. Despite this, individual mutations can still greatly affect disease outcome, and analysis of lineage-specific mutations identified several potential mechanisms that could lead to changes in how lineages 5 and 6 proliferate and cause disease. The lineage-specific mutations discussed above that could relate to a niche adaptation in hosts of West African ethnicity, including the lineage-specific mutations in ESX genes and cofactor biosynthesis genes, are also involved in virulence.
Another key set of virulence genes with lineage 5 and 6-specific mutations are the MCE operons. The MCE operons play an important role in the virulence of M. tuberculosis, particularly in mycobacterial growth in macrophages [67]. Antibodies to MCE1 proteins have been identified in patients [108], and operons 1–3 are required for virulence in mice [88]. Despite this apparent role in virulence, lineage 6 contains mutations that affect protein function in operons 1–3, while lineages 1–3 have nearly identical MCE operons to M. tuberculosis H37Rv, suggesting one potential mechanism of decreased virulence.
Another set of virulence-related genes with lineage-specific mutations are adenylate cyclases, which synthesize cAMP, an important second messenger [109]. M. tuberculosis encodes 17 adenylate cyclases, and deletion of one of them (Rv0386) has been shown to affect virulence and host response [85], highlighting the importance of this set of genes to pathogenicity. Both lineages 5 and 6 contained lineage-specific mutations predicted to affect the protein function of several adenylate cyclases, suggesting altered cAMP signaling in these strains, and a potential effect on the virulence of lineages 5 and 6.
Another pathway that affects bacterial growth and host response is the synthesis of the cell wall. Both lineage 5 and 6 contained lineage-specific mutations in L,D-transpeptidases. L,D-transpeptidases are critical to the structure of mycobacterial peptidoglycan and are involved in bacterial structure and growth [84], providing a possible explanation for the reported changes in cell size and doubling time in M. africanum GM041182 compared to M. tuberculosis H37Rv [7]. An altered cell wall could support either the hypothesis of decreased virulence, or suggest the need for a specific host immune system.
In addition, we saw high variability in PE, PPE and PE-PGRS genes, including changes in gene content. These repetitive regions are difficult to sequence and are often ignored, but may play a crucial role in antigenicity and the host-pathogen interaction [110, 111]. Using our high quality assemblies and alignments, we were able to identify lineage-specific mutations in these genes, as well as altered gene content. These mutations highlight the possibility of a critical role for these proteins in host-pathogen interactions and emphasize the need for a more detailed analysis of these regions. Furthermore, there were also a number of mutated hypothetical proteins and proteins of unknown function, all of which may play a critical as yet undiscovered role.
In addition to exploring mechanisms of geographic restriction, we also identified mutations that may have clinical implications for the region. We found that in Mali, M. africanum-associated and M. tuberculosis-associated strains evolved antibiotic resistance through similar mutations, and thus standard line-probe assays can still be utilized in West Africa. However, we also found a gidB polymorphism not previously described which might account for much of the streptomycin resistance in Mali. Also of concern, we identified several cases of pre-XDR in Mali, suggesting that Mali may need to begin testing for XDR cases. Furthermore, we identified lineage 5 or 6-specific mutations that may affect the evolution of drug resistance, particularly bedaquiline. Thus, whole genome sequencing surveys like this one are useful in revealing new mechanisms for drug resistance, informing development of molecular diagnostics.
One weakness of our study was that we were limited in our sample size for lineage 5 and M. bovis strains. Our collection was not representative of M. bovis genomic diversity, as three of the four M. bovis strains in our analysis were attenuated M. bovis BCG vaccine strains. However, we only used the M. bovis strains in our ANI and gene content analysis, and required that any observations be consistent with wild-type M. bovis sequence, AF2122/97, and our results corroborated all previous findings of M. bovis regions of difference. Another weakness of our study was that our observations may be specific to Mali, since all lineage 5 and 6 isolates sequenced for our study were isolated in Mali, although these lineages are found throughout West Africa. However, our lineage 6 isolates were genetically diverse, and represented multiple spoligotypes, and our isolates from other lineages did not cluster separately on the phylogenetic tree from strains isolated from South African patients. Thus, our collection reflected substantial diversity and did not originate from a clonal outbreak. In fact, the study from which we selected our samples found a wide diversity of strains in Mali, which covered 55% of all known spoligotyped strains [20]. Furthermore, based on spoligotyping, many similar strains can be found in neighboring countries [54, 68, 112–115]. And, finally, studies that employ genomic data alone are insufficient to address causality. However, we believe that this in-depth genomics analysis of the neglected pathogen, “M. africanum”, provides a strong foundation from which causal relationships between lineage-specific variation and geographic restriction can be made.
This collection provides valuable insights into the distinguishing genomic features of M. africanum. Here, we have analyzed in detail the genomes of lineage 5 and 6 isolates from Mali and identified several potential genetic reasons for the geographical restriction of lineages 5 and 6, such as alterations in vitamin B12 pathways and genes associated with virulence, which provide a guide to future studies focusing on the effects of specific genes. Although we cannot specifically point to a single reason why these lineages are geographically restricted, we have found mutations that support the hypothesis of a unique requirement for a host of West African ethnicity and for the hypothesis of loss of bacterial fitness. These hypotheses are not mutually exclusive, and we anticipate that these observations will be able to inform and fast-track experiments on mycobacterial pathogenicity and virulence, particularly with regard to this unique member of the MTC.
|
10.1371/journal.pntd.0007681 | Identification of anti-flaviviral drugs with mosquitocidal and anti-Zika virus activity in Aedes aegypti | Zika virus (ZIKV), an emerging arbovirus belonging to the genus Flavivirus, is transmitted by Aedes mosquitoes. ZIKV infection can cause microcephaly of newborn babies and Guillain–Barré syndrome in adults. Because no licensed vaccine or specific antiviral treatment is available for ZIKV infection, the most commonly used approach to control the spread of ZIKV is suppression of the mosquito vector population. A novel proposed strategy to block arthropod virus (arbovirus) transmission is based on the chemical inhibition of virus infection in mosquitoes. However, only a few drugs and compounds have been tested with such properties. Here we present a comprehensive screen of 55 FDA-approved anti-flaviviral drugs for potential anti-ZIKV and mosquitocidal activity. Four drugs (auranofin, actinomycin D (Act-D), bortezomib and gemcitabine) were toxic to C6/36 cells, and two drugs (5-fluorouracil and mycophenolic acid (MPA)) significantly reduced ZIKV production in C6/36 cells at 2 μM and 0.5 μM, respectively. Three drugs (Act-D, cyclosporin A, ivermectin) exhibited a strong adulticidal activity, and six drugs (U18666A, retinoic acid p-hydroxyanilide (4-HPR), clotrimazole, bortezomib, MPA, imatinib mesylate) significantly suppressed ZIKV infection in mosquito midguts. Some of these FDA-approved drugs may have potential for use for the development of ZIKV transmission-blocking strategies.
| Zika virus (ZIKV) is a human threat with a global health burden. As many as 86 countries and territories have reported evidence of mosquito-transmitted Zika infection, and there is no effective means of control. Recently, several studies have identified FDA-approved drugs exerting anti-ZIKV activity in mammalian cells. Here, we have screened such drugs for the ability to reduce mosquito viability or suppress ZIKV infection of mosquito cells. We identified three drugs that significantly increased mosquito mortality and six that significantly suppressed ZIKV infection in mosquito midguts. Altogether, our study provides a list of candidate agents for potential use to block ZIKV transmission in mosquitoes by chemical inhibition.
| Zika virus (ZIKV) belongs to the Flavivirus genus, which also includes dengue virus (DENV), West Nile (WNV), yellow fever virus (YFV) and Japanese encephalitis viruses (JEV) and is mainly transmitted by Aedes mosquitoes, including Ae. aegypti and Ae. albopictus, in an urban cycle [1]. The typical symptoms of Zika are very similar to dengue fever, including conjunctivitis (red eyes), muscle and joint pain, mild fever, rash, and severe headache [1]. ZIKV can also be directly transmitted between humans through sexual contact, or vertically from mother to child through during pregnancy [2–4]. ZIKV was first isolated from a rhesus monkey in the Zika forest of Uganda in 1947 and was then believed to be mainly circulated in a sylvatic cycle between non-human primate hosts and mosquitoes, until an outbreak occurred in Libreville, the capital of Gabon in 2007 [5, 6]. The most recent ZIKV outbreak in the Americas gained significant public attention, since Zika infection can result in microcephaly of newborn babies and Guillain–Barré syndrome in adults [7, 8].
No approved vaccines or specific therapies to prevent or treat ZIKV infection exist. Therefore, vector population suppression remains the most effective approach to control the spread of ZIKV. Control efforts to limit the population of Aedes mosquitoes and prevent mosquito biting rely mainly on insecticide application, removal of mosquito breeding sites, and the use of repellents and door/window curtains. However, these control methods are plagued by limitations such as insecticide resistance and logistics that hamper disease control, and novel control strategies are therefore needed.
Following ingestion of a viremic bloodmeal from a vertebrate host, an arthropod virus (arbovirus) needs to infect and replicate in several tissues and escape the vector’s immune defenses to be transmitted to another vertebrate host during blood feeding [9–12]. Effective blocking of arbovirus infection in the mosquito vector will result in transmission blocking. Numerous mosquito-encoded virus host factors (agonists) and restriction factors (antagonists) have been identified and shown to play essential roles in influencing arbovirus infection [13]. The rather small genomes of flaviviruses, comprising about 10 genes, do not allow for much functional diversity with regard to the viruses’ interaction with the vertebrate host’s and insect vector’s cellular machineries; hence, some virus host factors are conserved between the two hosts. Accordingly, a recently explored transmission blocking strategy that is based on chemical inhibition of host factors has shown significant reduction in DENV in midgut and salivary gland after injection of, or feeding on, chemical compounds that had previously been shown to inhibit infection in vertebrate cells [14].
An ideal transmission-blocking compound should either be safe for human consumption, or be environmentally safe through delivery in attractive toxic sugar bait (ATSB) systems. In both cases the compound should either inhibit virus infection of the vector or kill the vector [15, 16]. Several studies have reported successful identification of Food and Drug Administration (FDA)-approved drugs exerting ZIKV-inhibition in mammalian cells [17–19], involving diverse mechanisms of action such as the inhibition of DNA/RNA synthesis, DNA replication, proteolysis, or purine synthesis. Such compounds could qualify for the development of the above-mentioned control strategies. Here, we report the screening of such FDA-approved drugs to evaluate their anti-ZIKV activity in mosquito cells and adult mosquitoes and their ability to influence mosquito viability.
We searched the literature for drugs and compounds that have been reported to exert anti-ZIKV and/or anti-DENV activity in mammalian or mosquito cells, using the following criteria: 1) significant inhibition of flavivirus (ZIKV or DENV) infection in mammalian or mosquito cells; 2) no obvious toxicity to the tested cells; 3) approval already granted for human use or clinical trials. A total of 55 drugs were selected on the basis of these criteria (S1 Table). It should be noted that the anti-flaviviral activity of some drugs had been reported in multiple studies. For example, the cholesterol-lowering drug Lovastatin has been shown to inhibit ZIKV replication in the Huh7 human hepatoma cell [17, 18] as well as to have an impact on DENV RNA replication and virion secretion in primary human monocytes and cultured cell lines [20, 21].
We first used a mosquito cell line-based assay to screen the 55 selected drugs for inhibition of ZIKV infection. C6/36 cells grown overnight to 80% confluence were pre-treated for 1 h with a 20 μM concentration of each drug or the vehicle control (DMSO), and ZIKV was then inoculated into the cells at a multiplicity of infection (MOI) of 0.5. The cell mortality and virus titer in the medium were determined by microscopy and plaque assay, respectively at 3 days post-infection (dpi). We found that 15 drugs induced moderate cell mortality based on a >50% cell death as indicated by floating cells or highly granulated deformed cells, when compared to the DMSO-treated control (Fig 1A and S2 Table), and 3 drugs (5-fluorouracil, deferasirox, and mycophenolic acid [MPA]) mediated a significant reduction in virus titer when compared to DMSO treatment. Because of cell toxicity at the initially tested concentration of some of the drugs, we repeated the assay with two lower concentrations, 10 μM and 2 μM, for 15 of the drugs. At 10 μM, 8 of 15 drugs demonstrated cell toxicity at 3 days post-treatment, and 3 drugs inhibited virus infection without noticeable cell toxicity. At the 2 μM concentration, 4 of the 15 drugs induced cell toxicity, and 2 drugs reduced the virus titer, producing no observable toxicity (Fig 1B and S3 Table). In summary, we identified four drugs (auranofin, Act-D, bortezomib, and gemcitabine) that induced strong mosquito cell toxicity, and two drugs (5-fluorouracil and MPA) that mediated significant inhibition of ZIKV infection of C6/36 cells without causing cell toxicity (Fig 1C). Next, the cytotoxicity of these 4 drugs to C6/36 cells was determined, and the IC50 of auranofin, Act-D, bortezomib, and gemcitabine was 254.0 ng, 405.8 ng, 4.5 ng and 50.2 ng, respectively (Fig 1D). The anti-ZIKV effect of 5-fluorouracil and MPA was also evaluated in C6/36 cells, and 5-fluorouracil significantly inhibited ZIKV infection at 2 μM and MPA inhibited ZIKV infection at 0.5 μM (Fig 1E).
Since some drugs showed a strong toxicity to mosquito cells, we tested whether the 55 selected drugs influenced the viability of adult mosquitoes. Using an artificial membrane feeder, we orally fed each drug in combination with human blood, and monitored mosquito mortality daily up to 10 days. When adult female Ae. aegypti were fed 100 μM of each test drug in human blood, nine drugs induced mosquito mortality at 4 days and killed at least twice as many adults as the DMSO control at 10 days after exposure (Fig 2A and S4 Table). Three drugs showed a strong adulticidal activity, killing almost all the adult mosquitoes within 2 days (Fig 2A and S4 Table). When we serially 10-fold diluted these three drugs and tested them in adult females, we found that Act-D and cyclosporin A (CsA) exhibited a high mortality at 100 nM and 1 μM, respectively, while ivermectin induced the strongest mortality at 100 nM (Fig 2B–2D). Ivermectin is a known mosquitocidal for both Aedes and Anopheles species, whereas Act-D and CsA have not previously been reported to exert adulticidal activity.
Based on our C6/36 cell-based screening, 18 drugs that demonstrated cell toxicity or ZIKV inhibition were selected for testing in adult female Ae. aegypti. In a first assay, we fed one-week-old females with each drug at a final concentration of 100 μM in virus-containing blood, then measured midgut virus titers at 7 dpi by plaque assay. Because of the high toxicity of drugs #26 and #35, a final concentration of only 10 nM for #26 and 1 μM for #35 was used in the blood-virus mixture. The virus midgut infection intensity and prevalence were compared between each drug and the DMSO control. Three drugs (retinoic acid p-hydroxyanilide (4-HPR), bortezomib, and MPA) significantly reduced virus intensity, and five drugs (4-HPR, daptomycin, deferasirox, clotrimazole, and mefloquine-HCl) significantly reduced prevalence, whereas two drugs (#51: gemcitabine HCl and #52: clofazimine) actually significantly increased virus infection intensity (Fig 3A). It was particular noteworthy that 4-HPR dramatically reduced both virus infection intensity and prevalence. However, in the case of three drugs that reduced virus infection intensity in this first round of experiments (daptomycin, deferasirox, and mefloquine HCl), their ability to inhibit ZIKV could not be validated in a second round of virus challenge experiments with a higher ZIKV titer (Fig 3B); the higher viral load likely overcame the inhibitory action of these drugs. In addition, two drugs (4-HPR and MPA) also significantly reduced DENV2 infection intensity or prevalence in midguts in this second round (Fig 3C).
Notably, two drugs (5-fluorouracil and clofazimine) inhibited ZIKV infection in mosquito cell lines but did not suppress ZIKV infection in mosquitoes, indicating differences in the cellular physiology and interactions with the virus in vitro and in vivo situations. Therefore, we tested all 55 drugs for anti-ZIKV activity in Ae. aegypti. In a pre-screen, nine females from each drug group were tested. One drug (imatinib mesylate) significantly decreased midgut infection intensity, and two drugs (doxazosin and U18666A decreased prevalence, whereas two drugs (#12: MG-132 and #27: pyrimethamine) significantly increased infection intensity (Fig 4A). However, we could not verify doxazosin as being able to reduce infection prevalence in the second virus challenge experiment using a higher ZIKV titer in a blood meal (Fig 4B). In summary, we identified six drugs (U18666A, 4-HPR, clotrimazole, bortezomib, MPA, and imatinib mesylate) (Fig 5) showing significant inhibition of ZIKV infection in Ae. aegypti.
Taken together, four and two drugs displayed a strong mosquito cell toxicity or significant inhibition of ZIKV replication, respectively, on the basis of C6/36 cells mortality and infection assays, three and six drugs possessed mosquitocidal or transmission blocking activity, respectively, as assayed by mosquito mortality and infection assays (Fig 5 and S5 Table). Additionally, MPA inhibits ZIKV infection in both C6/36 cells and adult female mosquitoes; Actinomycin D exerts mortality to both C6/36 cells and female adults, while Bortezomib exerts mortality to C6/36 cells and inhibits ZIKV infection in mosquito midguts.
Numerous small compounds have been described as exhibiting flavivirus-inhibiting activity in mammalian cells [17–19, 21]. However, only a few have been tested for anti-flaviviral activity in mosquito cells, and chemical inhibition of flavivirus infection in adult mosquitoes has only been reported by our recent study [14]. Here, we tested 55 FDA-approved drugs with known anti-flaviviral activity in mammalian cells for potential mosquitocidal and anti-ZIKV activity in mosquitoes and cells. Several drugs were found to exert adulticidal activity or suppress ZIKV infection in the midgut tissue, indicating that these drugs could be considered for further exploration as arboviral transmission-blocking agents. It is important to point out that some drugs (lovastatin[17, 18, 21], palonosetron-HCL [17, 18, 21] and 6-azauridine[18, 22]) have been shown to exhibit moderate anti-flaviviral activity in mammalian cells but did not affect ZIKV infection in either mosquito cells or adult females in our experiments, pointing to the existence of differences in host-virus interaction or drug-host interactions. This finding challenges the belief that the virus’ interaction with the vector and host entail mostly conserved molecular mechanisms and factors because of the small size of the viral genome.
Act-D was the only drug displaying toxicity to both mosquito cells and adult females. However, only a relatively high dosage of Act-D (10 μM in blood) exerted mosquitocidal activity within 2 days after feeding, suggesting this drug may not act as a highly potent mosquitocide. Act-D is a polypeptide antibiotic with activity against Gram-negative bacteria that has also been clinically used as an anticancer drug. The anticancer mechanism of Act-D involves its ability to inhibit DNA-dependent RNA polymerases [23]. Interestingly, Act-D also induces apoptosis in the lepidopteran insect cell line Sf21 (established from Spodoptera frugiperda) by inhibiting cellular RNA synthesis [24, 25]. It is likely that Act-D induces midgut cell apoptosis, thereby killing the insect.
Two drugs, ivermectin and CsA, also exhibited adulticidal activity. Ivermectin is a broad-spectrum antiparasitic agent, and previous studies have established it as a general mosquitocide (including Aedes and Anopheles) [26, 27]. A recent study has shown that the LC50 for ivermectin is about 2.5ng/ml in adult Ae. aegypti [26], which is similar to the lowest dosage used in our experiments. CsA is an immunosuppressive agent used in patients receiving organ transplants. CsA was first reported to exert mosquito larvicidal activity in 1988, killing Culex pipiens autogenicus L4 larvae upon exposure to CsA with a LC50 = ~0.6 μg/ml at 48 h [28]. Another study has shown that CsA is a metabolite forming the spore surface layer of Tolypocladium tundrense and T. terricola, which kill Ae. aegypti larvae by vacuolization and subsequent destruction of their midgut cell mitochondria [29]. In addition to its toxicity in mosquito larvae, CsA has also been reported to suppress the humoral immune response of the wax moth Galleria mellonella [30]. However, the adulticidal activity of CsA appeared to be weaker than the larvicidal activity, requiring 1 μM in the blood meal.
We report here that three drugs (5-fluorouracil and MPA) inhibited ZIKV infection of mosquito cells, and six drugs significantly suppressed ZIKV infection in mosquitoes, but only MPA showed a significant anti-flaviviral activity in both mosquito cells and adults. MPA is an inhibitor of inosine monophosphate dehydrogenase, which blocks the synthesis of xanthosine monophosphate. MPA inhibits flavivirus infection in mammalian cells by preventing the synthesis and accumulation of viral RNA [31, 32]. MPA also suppresses DENV2 infection in mosquito midguts and dissemination to salivary glands when exposed through a blood or sugar meal [14]. Our study revealed that MPA can also inhibit ZIKV infection in both mosquito cells and adults.
We also show that 4-HPR potently inhibits both ZIKV and DENV infection in mosquito midguts. 4-HPR is a retinoic acid (RA) derivative and a potential cancer-preventive agent that acts by inducing apoptosis in cancer cells [33], and has been extensively tested in humans [34, 35]. Recently, 4-HPR was also shown to be antiviral, since it inhibits DENV by limiting the accumulation of viral RNA in mammalian cells at the late stage of infection [36], as well as several other flaviviruses in mammalian cells [37]. It has also been shown that 4-HPR potently inhibits ZIKV infection in multiple mammalian and mosquito cell lines, and prophylactic delivery of the drug results in significant reductions in both serum viremia and brain viral burden in a murine ZIKV infection model [38, 39]. A significantly lower antiviral activity of 4-HPR has also been observed in C6/36 cells [39]. Accordingly, we did not detect a significant reduction in ZIKV titer in C6/36 cells treated with 4-HPR. However, 4-HPR significantly suppressed ZIKV infection and reduced DENV infection prevalence in mosquito midguts, suggesting that the inhibition likely occurs through viral host factors in mosquitoes. Further investigation of which host factors are involved in 4-HPR inhibition of ZIKV infection in mosquitoes is necessary to uncover the underlying mechanisms.
The FDA-approved 20S proteasome inhibitor bortezomib has been shown to inhibit infection of all four serotypes of DENV in primary monocytes, and bortezomib treatment of DENV-infected mice reduces viral load and signs of dengue pathology [40]. A recent study has also shown that bortezomib can suppress ZIKV infection in C6/36 cells in a dose-dependent manner, as well as reduce the viral load and signs of ZIKV pathology in treated mice [41]. Our C6/36 cell-based assay showed that ZIKV virus load was significantly reduced when cells were treated with 2 mM bortezomib, but treatment with the drug also affected cell viability, consistent with a study by Xin et al. that showed a concentration-dependent cytotoxicity in C6/36 cells (50% cytotoxic concentration (CC50) of >160 nM) [41]. Combined with our finding that bortezomib inhibited ZIKV infection in both mosquito cells and adult females, these results suggest that bortezomib is a good candidate to explore for blocking ZIKV transmission in mosquitoes and treatment of ZIKV infection in patients.
In addition to MPA, 4-HPR, and bortezomib, we also identified other three drugs (U18666A, clotrimazole, and imatinib mesylate) with activity against ZIKV infection in Ae. aegypti. U18666A is an amphipathic steroid that is widely used to block the intracellular trafficking of cholesterol by inhibiting oxidosqualene cyclase and desmosterol reductase [42]. U18666A has been shown to inhibit DENV entry and replication in mammalian cells by suppressing de novo sterol biosynthesis and retarding viral trafficking in the cholesterol-loaded late endosomes/lysosomes of host cells [43]. U18666A can also suppress infection and replication of the alphavirus Chikungunya virus (CHIKV) in human skin fibroblasts without any cytotoxic effects [44]. U18666A has not been tested to assess its effect on arbovirus infection in mosquitoes or insect cells. We found that U18666A significantly reduced ZIKV infection prevalence in mosquito midguts. Clotrimazole is a synthetic imidazole that acts against fungi by inhibiting the biosynthesis of sterols (ergosterol) of the fungal cell membrane, and it is widely used to treat fungal infections. Clotrimazole has been identified as suppressing ZIKV infection in mammalian cells [18]. Our study showed that clotrimazole also inhibits ZIKV infection in mosquito midguts. Its mechanism of inhibition of ZIKV infection has not been studied in mammalian cells or mosquitoes. Imatinib mesylate is an FDA-approved tyrosine kinase inhibitor that is used to treat several types of cancers. In addition to inhibiting ZIKV infection, imatinib also inhibits alphavirus (Sindbis virus) replication in mammalian cells [45].
The anti-flaviviral and mosquitocidal effects of the different compounds are most likely mediated by diverse mechanisms since their functions as therapeutic agents in humans are diverse. For example, MPA and CsA are used as immunosuppressants in kidney and liver transplant recipients [46, 47]; bortezomib, a proteinase inhibitor, is used to treat relapsed multiple myeloma and mantle cell lymphoma [48]; imatinib mesylate, a tyrosine kinase inhibitor, is used to treat chronic myelogenous leukemia (CML), gastrointestinal stromal tumors (GISTs) and a number of other malignancies [49]. The therapeutic dosage of drugs differ as does the drug concentration in patient’s blood or plasma. For example, serum concentration of imatinib mesylate ranges between 0.138 mg/ml and 2.816 mg/ml (median = 1.344 mg/ml) [50], which is higher than what we used for mosquito feeding (58.97 μg/ml); the patient serum level of MPA ranges from 0.45 to 6.5 mg/l (median = 2.1 mg/l) [51, 52], which is lower than that we fed mosquitoes on (32.04 μg/ml); the level of CsA in patient blood ranges from 212 to 1358 ng/ml [53], which is higher than that in blood meal can kill mosquitoes (120.26 ng/ml); the median concentration of Actinomycin D in patient plasma ranges from 24.4 to 128 μg/l between 5 and 15 min post administration [54], which is similar to that in blood meal can kill mosquitoes (125.5 ng/l). It is standard practice to use uniform concentrations when screening multiple compounds for bioactivity such as anti-viral action [17, 55]. Since we did not observe complete virus-blocking (infection intensity of zero) for any of the drugs, even at concentrations higher than patient serum levels, we did not pursue testing of lower concentrations. Nevertheless, the mosquitocidal or virus-blocking concentration is irrelevant when considering the design of a control strategy based on artificial toxic sugar bait (ATSB) delivery of compounds to mosquitoes[16, 56].
In conclusion, we have screened 55 FDA-approved drugs with known anti-flaviviral activity in mammalian and/or mosquito cells. Based on C6/36 cell assays, four drugs (auranofin, Act-D, bortezomib, and gemcitabine) were identified as exerting moderate toxicity to mosquito cells, and two drugs (5-fluorouracil and MPA) were identified as significantly reducing ZIKV infection. Our mosquito mortality assays revealed that nine drugs had mosquitocidal activity, and three of these (Act-D, CsA, and ivermectin) exhibited a moderate adulticidal activity and should therefore be further explored for utility in mosquito control. Finally, six drugs (U18666A, 4-HPR, clotrimazole, bortezomib, MPA, and imatinib mesylate) demonstrated anti-ZIKV activity in mosquito midguts, and three of them (MPA, 4-HPR, and bortezomib) showed particularly potent ZIKV-blocking ability, rendering them interesting candidates for the development of transmission-blocking strategies.
This study was carried out in accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health, the Animal Care and Use Committee (ACUC) of the Johns Hopkins University, and the institutional Ethics Committee (permit number: M006H300). The IACUC committee approved the protocol. Mice were only used for mosquito rearing. Commercial, anonymous human blood was used for virus infection assays in mosquitoes, and informed consent was therefore not applicable.
The chemicals used in this study were purchased from Sigma or MedChemExpress (MCE). ZIKV (Cambodia, FSS13025) and DENV serotype 2 (New Guinea C strain, DENV2) were used.
Ae. aegypti (Rockefeller) larvae were reared with fish food, and adults were maintained on a 10% sucrose solution at 27°C and 85% humidity with a 12-h light/dark cycle. The mice were used for blood feeding and colony maintenance. The C6/36 (Ae. albopictus) cell line that was used for virus propagation was grown in minimal essential medium (MEM, Gibco, Carlsbad, CA, USA) with 10% heat inactivated FBS, 1% L-glutamine, 1% penicillin-streptomycin, and 1% non-essential amino acids at 32°C with 5% CO2. The baby hamster kidney BHK-21 and Vero cell lines that were used for plaque assays were maintained on Dulbecco's modified Eagle's medium (DMEM, Gibco, Carlsbad, CA, USA) supplemented with 10% FBS, 1% L-glutamine, 1% penicillin-streptomycin, and 5 μg/ml plasmocin (Invitrogen, Carlsbad, CA) at 37°C with 5% CO2.
Day 1: C6/36 cells were cultured in 24-well plates, then used for ZIKV infection when the cells were 80% confluent. Day 2: The cells were pretreated with drugs or controls. Drugs were diluted to a final concentration of 20 μM (or 10 μM or 2 μM) in complete MEM medium and added to the cells, which were incubated at room temperature in the rocking machine for 15 min, then incubated for 1 h in an incubator at 32°C with 5% CO2. Approximately 1 h after the addition of a test compound, ZIKV was added to each well at an MOI of 0.5. Infected cells were incubated at 32°C with 5% CO2. Each drug treatment had at least three replicates. Day 3–5: The cells were observed while growing, and the floating and highly granulated deformed cells were noted under the microscope and recorded each day. Day 5: If the mock-infected cells showed signs of infection, 200 μl of conditioned cell culture medium was harvested from each well and kept at -80°C for virus tittering. To measure IC50 of the selected drugs, C6/36 cells were plated at densities of 1 × 104 cells/well in 96-well plates. The cells were incubated overnight and then treated with the compounds at a final concentration of 10 μM, 2 μM, 0.5 μM, 0.1 μM, 20 nM, 5 nM, 1 nM, 0.2 nM and 0.05 nM, respectively. C6/36 cells were incubated for another 96 h. Cell viability was measured using the CellTiter 96 Aqueous One Solution Reagent (Promega, Madison, WI, USA) and a Spectramax M2 plate reader (Molecular Devices, Sunnyvale, CA, USA) following manufacturer's protocol. Results from drugs treated cells were compared to results from DMSO controls. The experiments were performed in triplicate. IC50 of each drug was determined using nonlinear regression with GraphPad Prism.
ZIKV or DENV2 was propagated in C6/36 cells in T25 flasks and infected with ZIKV at an MOI of 0.5 or DENV2 at an MOI of 0.1 using MEM with 10% FBS, 1% L-glutamine, 1% penicillin-streptomycin, and 1% non-essential amino acids at 32°C with 5% CO2. After 5–6 days of incubation at 28°C, infected cell culture medium was mixed with an equal volume of commercial human blood supplemented with 10% human serum containing 10 mM ATP and a final concentration of 100 μM of the test drug. One week-old females were fed for 30 min with the mixture at 37°C using a single glass feeder per cup. Fully engorged females were selected and maintained on 10% sucrose at 27°C and 85% humidity with a 12-h light/dark cycle. Midguts were dissected from individual females at 7 dpi and kept it at -80°C.
Samples harvested from cell culture were diluted 1:1000 with DMEM, and 10 μl were used for a plaque assay. Infected mosquito midguts were homogenized in DMEM with a Bullet Blender (Next Advance Inc., Averill Park, NY, USA) with glass beads, centrifuged at 10,000g for 5 min at 4°C, and the supernatants were harvested for plaque assay. DENV2 was tittered with BHK 21 cells, and ZIKV for C36/36 cell-screening was tittered with Vero cells and for mosquito infection assay with BHK cells. In brief, infected medium or midgut samples were serially 10-fold diluted and inoculated into cells seeded and grown to 80% confluence in 24-well plates. Plates were rocked for 15 min at RT, followed by incubation at 37°C with 5% CO2 for 45 min. Subsequently, 1 ml of DMEM containing 2% FBS, 0.8% methylcellulose and 1% penicillin-streptomycin were added to each well, and the plates were incubated for 5 days at 37°C with 5% CO2. The plates were then fixed and visualized with methanol/acetone mixture (1:1 volume) and 1% crystal violet mixture for 30 min at RT, and the plaque-forming units were counted.
Thirty one-week-old female mosquitoes were transferred to a paper cup the day before blood feeding. Mosquitoes were fed with human blood containing the tested drug on a glass feeder. Twenty fully engorged mosquitoes per cup were retained for a survival study. For the initial screening, a final concentration of 100 μM was tested for each drug. For those chemicals showing high mortality, a 10-fold serial dilution was used for the next round of screening. Dead mosquitoes were recorded and removed every day for 10 days or 14 days. Each chemical feeding had at least three replicates. Survival curves were generated using GraphPad Prism.
Statistical analysis was performed using GraphPad Prism. Differences in virus intensity and prevalence of infection in midguts between chemical treatment and the control were compared using the non-parametric Mann-Whitney U-test or Fisher's exact test, respectively. In the experiments screening C6/36 cells, data from each sample were analyzed using Student’s t-test. All tests were considered significant at P < 0.05.
|
10.1371/journal.pbio.3000189 | Ablating astrocyte insulin receptors leads to delayed puberty and hypogonadism in mice | Insulin resistance and obesity are associated with reduced gonadotropin-releasing hormone (GnRH) release and infertility. Mice that lack insulin receptors (IRs) throughout development in both neuronal and non-neuronal brain cells are known to exhibit subfertility due to hypogonadotropic hypogonadism. However, attempts to recapitulate this phenotype by targeting specific neurons have failed. To determine whether astrocytic insulin sensing plays a role in the regulation of fertility, we generated mice lacking IRs in astrocytes (astrocyte-specific insulin receptor deletion [IRKOGFAP] mice). IRKOGFAP males and females showed a delay in balanopreputial separation or vaginal opening and first estrous, respectively. In adulthood, IRKOGFAP female mice also exhibited longer, irregular estrus cycles, decreased pregnancy rates, and reduced litter sizes. IRKOGFAP mice show normal sexual behavior but hypothalamic-pituitary-gonadotropin (HPG) axis dysregulation, likely explaining their low fecundity. Histological examination of testes and ovaries showed impaired spermatogenesis and ovarian follicle maturation. Finally, reduced prostaglandin E synthase 2 (PGES2) levels were found in astrocytes isolated from these mice, suggesting a mechanism for low GnRH/luteinizing hormone (LH) secretion. These findings demonstrate that insulin sensing by astrocytes is indispensable for the function of the reproductive axis. Additional work is needed to elucidate the role of astrocytes in the maturation of hypothalamic reproductive circuits.
| Astrocytes are a major cell type in the central nervous system, yet their impact on the neuroendocrine circuits that control fertility is under appreciated. Here, we show in mice that ablation of insulin signaling in astrocytes leads to delayed puberty, hypothalamic-pituitary-gonadotropin (HPG) axis dysfunction, and reduced fertility. These findings are the first demonstration that astrocytes and a metabolic signal collaborate to permit the maturation of the reproductive axis and adult fertility.
| Reproduction is essential for species survival. Because energy is required to locate a mate, maintain a pregnancy, and rear young, fertility is modulated by the status of energy stores [1–3]. Excessive energy expenditure or insufficient caloric intake in humans and rodents delays the pubertal transition and reduces fertility [4, 5]. Moreover, diseases that cause metabolic disturbances, such as thyroid disease, chronic inflammatory states, and malnutrition, are associated with a disruption of the normal timing of puberty [6].
The pancreatic hormone insulin serves as one metabolic signal linking hypothalamic function with metabolic state [7–9]. Postnatal deletion of insulin receptors (IRs) in glial fibrillary acidic protein (GFAP)-expressing cells decreased the activation of pro-opiomelanocortin (POMC) neurons by glucose [10]. Additionally, mice with IR ablated from astrocytes in the mediobasal hypothalamus became insulin and glucose intolerant [10]. These findings suggest that IRs on hypothalamic astrocytes play a role in regulating glucose metabolism.
Insulin is a key regulator of the gonadotropin-releasing hormone (GnRH) network that controls fertility [8, 11–14]. Insulin increases GnRH-dependent luteinizing hormone (LH) secretion in adult male mice [2, 15]. Similarly, hyperinsulinemic clamps in women significantly increase LH pulsatility [2, 16, 17]. Insulin signaling in the brain may also provide a prerequisite signal for the initiation of puberty [18, 19]. Insulin increases in children around the time of adrenarche in association with increasing circulating insulin-like growth factor 1 (IGF1) levels [2]. Administering metformin to girls with precocious pubarche to reduce their insulin levels results in a delay in the onset of puberty [20, 21]. However, the specific mechanisms underlying insulin modulation of pubertal timing are largely unknown.
A seminal paper by Brüning and colleagues [8] showed that 50% of mice lacking the IR in cells expressing nestin (NIRKO mice) displayed hypogonadotropic hypogonadism in adulthood. Targeted deletion of IRs in specific neuronal populations, however, has failed to induce the subfertile phenotype and GnRH network dysregulation of NIRKO mice [2, 3, 6, 22, 23]. For instance, Divall and colleagues found that mice with IR deletion in GnRH neurons experienced normal pubertal timing and fertility [6]. Mice with IR deletion in kisspeptin neurons displayed a 4–5 day delay in pubertal onset but normal fertility and gonadal hormonal levels in adulthood [2]. In another example, mice with IR deletion in gamma-amino butyric acid (GABA)-ergic or glutamatergic cells showed normal pubertal progression, estrous cyclicity, and fertility [23]. More widespread deletion of IR in Ca2+/calmodulin-dependent protein kinase-expressing neurons, located in the dentate gyrus, cortex, olfactory bulb, amygdala, striatum, thalamus, and hypothalamus [24], also produced mice with normal reproductive maturation and fertility [3]. These numerous negative results suggest that insulin action in neurons does not play an essential role in hypothalamic-pituitary-gonadal (HPG) axis function.
Alternatively, it has been suggested [3] that the hypothalamic hypogonadism observed in NIRKO mice results from the chronic absence of insulin signaling in glia rather than neurons. Indeed, the nestin-cre line drives deletion of IR in both neuronal and non-neuronal cells [8, 25–27]. Glial cells, which include astrocytes and tanycytes, are known to play an important role in the puberty onset, estrus cyclicity, and fecundity [28, 29]. Therefore, we hypothesized that astrocytic insulin sensitivity is required for normal GnRH release during the pubertal period and in adulthood. We tested this hypothesis by using the cre-lox system to examine the effect of chronic astrocyte IR deletion on fertility.
To generate mice with IR deletion in astrocytes, we crossed IRloxp and GFAP-cre mouse lines. To assess whether Cre expression was restricted to astrocytes in the resulting mice, we crossed experimental mice with tdTomato-loxP reporter mice, which express red fluorescent protein (RFP) in a cre-dependent manner. RFP was found in IRKOGFAP brains but in not those of control mice that carried only the IRloxp allele (Fig 1A). Our data confirm the specificity and selectivity of IR gene and transcript deletion to the brain and not other tissues, including the gonads (S1 Fig). Double immuno-staining labeling of GFAP and tdTomato showed sufficient cre activity to drive tdTomato expression in 94% of GFAP positive cells. When neurons were labeled with the neuronal nuclear antigen NeuN, there was no colocalization with cre-driven tdTomato expression (Fig 1B). We performed immuno-staining colocalization studies in various regions of the brain, including the arcuate nucleus (ARC), anteroventral periventricular nucleus (AVPV), and the cortex to further confirm the wide-spread deletion of IR in astrocytes (S2 Fig).
Fluorescence-activated cell sorting (FACS) was performed on isolated brain cells using tdTomato as a marker of cre expression. The data show that 46.0% of isolated brain cells were positive for astrocyte cell surface antigen-1 (ACSA-1) and tdTomato, whereas 11.2% of cells were positive for ACSA-1 yet negative for tdTomato in the IRKOGFAP mice. In addition, very few cells (0.7%) were positive for tdTomato and negative for ACSA-1 in brain cells isolated from IRKOGFAP mice (Fig 1C). Astrocytes isolated by this method (tdTomato+ allopycocyanin+ [APC]) showed a substantial reduction in IR mRNA levels in IRKOGFAP mice when compared to IRloxp (tdTomato− APC+) (Fig 1C). Meanwhile, the expression levels of IR mRNA in the isolated nonastrocyte cells (tdTomato− APC−) from IRKOGFAP mice were comparable to the IRloxp group, confirming the specificity of the deletion (Fig 1C). Previous studies have suggested that tanycytes near the third ventricle express GFAP [30]. Therefore, to further verify the purity of astrocytic FACS isolation, we measured gene expression of different markers of neuronal, tanycytic, microglia, and endothelial markers and confirmed the specific isolation of astrocytes via FACS (S3 Fig). In addition, western blotting of brain tissues confirmed decreased levels of IR protein in IRKOGFAP mice when compared to the IRloxp group (Fig 1D) (S4 Fig). Because it is still unclear if astrocytes are derived from erythromyeloid progenitors, the same lineage that produces macrophages in the periphery [31, 32], we tested whether macrophages, which originate as monocytes produced in bone marrow, exhibited loss of IRs. Expression of IRs and GFAP was not different in macrophages from IRloxp and IRKOGFAP mice (S5 Fig).
Balanopreputial separation serves as an indicator of the initiation of puberty in males. IRKOGFAP male mice showed a significant delay in the postnatal day (PND) of balanopreputial separation (PND 33.36 ± 0.67) when compared to IRloxp control mice (PND 28.44 ± 0.36) (Fig 2A). In contrast, we found that the GFAP-cre mouse line alone has no phenotype in comparison to IRloxp mice (S6 Fig).
To assess the progression of puberty in female mice, vaginal opening and timing of the onset of estrus cycling were measured. IRKOGFAP mice exhibited a delay in vaginal opening of approximately 4 days (PND 34.08 ± 0.69) when compared to IRloxp mice (PND 29.44 ± 1.05) (Fig 2B). IRKOGFAP mice showed a significant delay in the age of first estrus by approximately 5 days (PND 42.55 ± 0.45) when compared to IRloxp mice (PND 36.00 ± 1.01) (Fig 2C). In addition, no differences were seen in body weight or body growth at 3 weeks of age between IRKOGFAP and IRloxp mice (S7 Fig).
IRKOGFAP females exhibited irregular cyclicity and longer estrous cycles. The estrus cycle length was approximately 2 days longer in IRKOGFAP females (PND 6.25 ± 0.21) when compared to IRloxp mice (PND 4.80 ± 0.13) (Fig 2D). IRKOGFAP mice spent significantly less time in estrus and a longer time in diestrus when compared to IRloxp females (Fig 2E) (S8 Fig).
To assess fertility in IRKOGFAP mice, pregnancy rate, litter size, and mating success were measured. IRKOGFAP males produced fewer pregnancies when paired with fertile wild-type (WT) females (54% induced pregnancies), while IRloxp males were 90% successful in producing pregnancies (Fig 2F). IRKOGFAP females, when paired with fertile WT males, exhibited a significantly reduced pregnancy rate of 45%, compared to 89% for IRloxp females (Fig 2L). The interval from mating to birth did not differ between groups (Fig 2H and 2N). However, IRKOGFAP male and female mice exhibited a significant decrease in litter size when compared to IRloxp mice (litter size for IRloxp 7.44 ± 0.97 versus IRKOGFAP 2.55 ± 1.02) (Fig 2G and 2M).
We next assessed the function of the HPG axis in adult male and randomly cycling female mice by measuring LH, follicle-stimulating hormone (FSH), and sex steroid levels between 8 and 10 AM. IRKOGFAP males showed a significant decrease in LH and testosterone levels (Fig 2I and 2K) but no change in FSH when compared to IRloxp mice (Fig 2J). LH, FSH, and estradiol levels were significantly decreased in IRKOGFAP females when compared to IRloxp mice (Fig 2O–2Q). LH pulse amplitude and frequency have been reported to be reduced on estrus, although basal levels of LH are similar on all days of the cycle [33]. Since IRKOGFAP female mice spent less time in estrus yet had lower LH levels, mouse cycle stage is unlikely to explain these findings.
Gonadal morphology was examined in both sexes. There was a reduction in the sperm count per seminiferous tubule cross-section in all stages (Fig 3A and 3B). Spermatogonia, spermatocytes, spermatid, and spermatozoa counts were significantly reduced in the seminiferous tubules of IRKOGFAP males (128.3 ± 16.53, 128.0 ± 7.16, 209.0 ± 15.76, and 138.3 ± 12.61) when compared to IRloxp mice (212.0 ± 13.72, 229.0 ± 14.01, 361.0 ± 48.30, and 278.0 ± 31.10) (Fig 3C–3F). IRKOGFAP female mice exhibited altered ovarian morphology when compared to IRloxp mice (Fig 3G and 3H). Similarly, the number of primary follicles, preovulatory follicles, and corpora lutea per ovary cross-section were significantly lower (3.00 ± 0.57, 1.67 ± 0.33, and 3.33 ± 0.33) when compared to IRloxp mice (7.50 ± 0.64, 4.00 ± 0.57, and 7.00 ± 1.00) (Fig 3I–3M). Primordial and secondary follicle numbers were not different between groups.
Because astrocytic insulin signaling has been linked to depressive-like behavior [69], we examined sexual behavior in these mice to determine whether reduced fertility in IRKOGFAP mice could be partially attributed to reduced sexual motivation or performance. IRKOGFAP and IRloxp females were paired with WT gonadectomized males, and multiple parameters were measured, including lordosis, mounting attempts, lordosis quotient, and latency to first lordosis. IRKOGFAP and IRloxp female mice showed no differences in any of these parameters (S9 Fig). Likewise, IRKOGFAP and IRloxp male mice showed no differences in mounting attempts, latency to first mount, and latency to first intromission when paired with control females (S9 Fig).
Astrocytes release specific growth factors that stimulate the secretion of GnRH. In particular, prostaglandin E2 (PGE2) release stimulates the secretion of GnRH; Clasadonte and coworkers investigated the firing activity of GnRH neurons in mice with deficient PGE2 synthesis in astrocytes and found the excitability of these neurons significantly decreased [34]. We therefore measured protein levels of prostaglandin E synthase 2 (PGES2), which catalyzes the conversion of prostaglandin H2 to prostaglandin E2, in isolated astrocytes from IRKOGFAP and control mice. IRKOGFAP astrocytes exhibited a significant reduction in PGES2 levels when compared to IRloxp astrocytes (Fig 4A–4C).
Astrocytes assist neurons through nutritional and structural support and by promoting neurotransmitter release and recycling. They also appear to contribute to information processing by the brain [35, 36]. Astrocytes possess a dense network of fine processes whose membranes contain potassium channels [37, 38], aquaporins [39], glutamate transporters [40], and lactate transporters [41]. These processes enwrap neuronal synapses and ensure effective synaptic transmission. Astrocytes also display increased intracellular calcium (but not electrical excitability) in response to chemical and neuronal cues [42], which is believed to lead to the release of gliotransmitters, such as adenosine, polyphosphate, D-serine, glutamate, GABA, and lactate, that can alter neuronal activity [43–48]. As one critical element of the blood–brain barrier, astrocytes are readily able to sense circulating metabolic and endocrine signals [49, 50]. Notably, insulin acts on IRs in primary human astrocytes, promoting glycogen synthesis [51]. Astrocytes are also able to release vasoactive molecules to regulate cerebral blood flow and to ensure a sufficient supply of oxygen and glucose to active neurons [52]. Astrocytes are therefore believed to play a critical role as central nervous system (CNS) metabolic sensors [53].
The current study demonstrates that insulin is a critical metabolic signal acting through astrocytes to permit reproductive competency via the GnRH network; astrocyte insulin signaling prevented hypogonadism and allowed normal fertility in adulthood. Similar to NIRKO mice [54], IRKOGFAP mice exhibited impaired spermatogenesis, folliculogenesis, and ovulation, resulting in an almost 50% decrease in pregnancy rate and a nearly 69% reduction in litter size. IRKOGFAP mice also showed a significant decrease in LH and testosterone levels in males and LH, FSH, and estradiol levels in females. These findings indicate that disruption of astrocytic insulin signaling leads to hypogonadotropic hypogonadism [55, 56]. Given that IRKOGFAP mice exhibit a delay in vaginal opening and first estrous in females and balanopreputial separation in males, disruption of astrocytic insulin action also serves as a critical role in the maturation of the HPG axis.
Astrocytes have the potential to control GnRH release in several ways. GFAP-immunoreactive astrocyte processes have been shown to ensheath GnRH cell bodies in the rostral preoptic area of the rat [57] and GnRH cell bodies in the medial basal hypothalamus of monkeys [58, 59]. In addition, GnRH processes in the median eminence are apposed largely by astrocytes, with the support of tanycytes [60]. The structural relationships at both sites are dynamic and regulated by gonadal steroids in rodents and rhesus monkeys [57, 58, 61, 62]. GnRH neurons adhere to astrocytes using heterophilic (contactin/RPTPβ) and homophilic synaptic cell adhesion molecule (SynCAM) interactions; these molecules have signaling capabilities, suggesting they can activate intracellular signaling cascades in astrocyte and GnRH neurons [63]. Indeed, transgenic mice that express a dominant negative SynCAM1 under the control of a human GFAP promoter had a delayed onset of puberty, disrupted estrous cyclicity, and reduced fecundity associated with low GnRH release [29].
Astrocytes also synthesize and release factors that regulate GnRH secretion [28]. Astrocytes are believed to produce growth factors such as basic fibroblast growth factor IGF1 and transforming growth factor (TGF)-β1 that act directly on GnRH neurons to stimulate production of GnRH. In addition, in vitro evidence suggests that their production of growth factors of the epidermal growth factor family (TGFα and neuroregulins) causes glial release of mediators like PGE2 that stimulate GnRH release [64]. Mice expressing a dominant-negative Erbb2 receptor tyrosine kinase 4 receptor, which responds to EGFs, under the control of the GFAP promoter exhibit delayed sexual maturation and a diminished reproductive capacity in early adulthood due to impaired release of GnRH [65]. Interestingly, human hypothalamic hamartomas associated with sexual precocity in humans contain numerous astrocytes expressing TGFα and erbB1 receptors [66].
Astrocytes also release substances, like calcium, glutamate, and ATP, capable of stimulating GnRH release [67, 68]. Cai and coworkers (2018) recently found that insulin signaling can target astrocyte-specific soluble NSF attachment protein receptors to regulate exocytosis of ATP [69]. Thus, IR deletion in IRKOGFAP mice may lead to impaired tyrosine phosphorylation of mammalian uncoordinated-18, leading to decreased astrocytic ATP exocytosis [69]. Finally, neurons require glial-provided precursors such as glutamine to synthesize glutamate and GABA. This mechanism allows astrocytes to influence neuronal glutamate production and availability at the synaptic cleft by expressing glutamine synthase [70, 71]. This regulation is responsive to estradiol levels and pubertal progression [72, 73]. Overall, these studies demonstrate that astrocytes can influence GnRH release through multiple pathways.
Studies have shown that hypothalamic astrocytes release PGE2 in response to cell–cell signaling. PGE2 release stimulates the secretion of GnRH to regulate the pituitary release of LH and FSH [34]. Our work shows decreased levels of astrocytic PGES2 protein levels in knockout mice when compared to controls, suggesting reduced production and release of PGE2. Interestingly, PGE2 release is mediated by exocytosis. Shimada and colleagues have shown that solute carrier organic anion transporter family member 2A1, a PGE2 transporter, is responsible for loading intracellular PGE2 into lysosomes in macrophages; PGE2 is then released via exocytosis induced by Ca2+ influx [74]. Future studies should therefore investigate whether impaired insulin-dependent exocytosis could also affect PGE2 release from astrocytes. Another important consideration for future study is the role of astrocyte insulin action during development versus its actions in the adult animal. Indeed, insulin and IGFs may directly influence brain development and neuronal survival [75–77]. While the contribution of astrocyte insulin signaling to the establishment of neuroendocrine function is unknown, it may play a role during the organization of reproductive circuitry.
In summary, our findings suggest that impaired insulin sensing in astrocytes delays the initiation of puberty and dramatically reduces adult reproductive success. These effects are due to dysfunction of the HPG axis, leading to hypogonadotropic hypogonadism, and are associated with decreased PGES2 levels in astrocytes. This model is the first to recapitulate the effects of brain IR deletion on fertility. Our findings emphasize the importance of astrocytic signaling in the regulation of reproduction and lay the foundation for future studies addressing this communication at different stages of development. Additional studies are warranted to investigate the mechanism of how insulin action on astrocytes modulates the GnRH network.
All procedures were approved by the Institutional Animal Care and Use Committee (IACUC) of the University of Toledo College of Medicine and Life Sciences in Toledo, Ohio. All experiments were performed in accordance with the relevant guidelines and regulations described in the IACUC-approved protocol number 106448.
To create an astrocyte-specific deletion of IR (IRKOGFAP mice), GFAP-Cre mice (C57Bl/J6) (Frederick National Laboratory for Cancer Research, Frederick, Maryland, United States) were crossed with IRloxp mice (C57Bl/J6) in which exon 4 of the IR gene was flanked by loxP sites [22]. GFAP is the main intermediate filament protein in mature astrocytes and an important component of the cytoskeleton in astrocytes during development [78, 79]. After the first generation of the breeding, GFAP-Cre, IRloxp mice were crossed with homozygous IRloxp mice to generate the experimental mice. IRloxp mice littermates lacking Cre expression were used as controls; comparisons between IRloxp mice and GFAP-Cre mice were also performed where specified. Where noted, the mice also carried the tdTomato gene inserted into the Gt(ROSA)26Sor locus to serve as a reporter under the control of Cre recombinase expression. Mice were housed in the University of Toledo College of Medicine animal facility at 22°C–24°C on a 12-hour light/dark cycle and were fed standard rodent chow. Mice were weaned on postnatal day (PND) 21. Genotyping was performed by Transnetyx, Inc. (Cordova, Tennessee, US) using a real-time RTPCR–based approach.
Mice were sacrificed via ketamine/xylazine injections, and the brain and other tissues were removed. Total RNA was extracted using an RNeasy Lipid Tissue Mini Kit (Qiagen, Valencia, California, US). Single-strand cDNA was synthesized by a high-capacity cDNA Reverse Transcription Kit (Applied Biosystems). Bone marrow–derived macrophages were obtained, as previously described [80]. Specifically, femurs and tibias were collected and flushed with medium containing sterile RPMI, 1% penicillin/streptomycin, and L929‐conditioned medium to isolate bone marrow cells. These cells were then allowed to differentiate for 7 days (37°C, 5% CO2 atmosphere) with a change of media on day 4. Then, RTPCR was performed [81]. Briefly, total RNA was prepared from BMDMs using Perfect Pure RNA Tissue kit (5Prime kit) according to manufacturer's instructions. cDNA was synthesized with random primers and reverse transcriptase (Applied Biosystems) using 1 μg of total RNA. cDNA was evaluated with quantitative RTPCR using True Amp SYBR green qPCR Supermix (Applied Biosystems). The relative amount of mRNA was calculated by comparison to the corresponding controls and normalized relative to Glyceraldehyde 3-phosphate dehydrogenase (GAPDH). RQ is expressed as means ± SE relative to IRloxp. Sequences of primers used are as follows: IR: Forward—CCCCAACGTCTCCTCTACCA, Reverse—TGTTCACCACTTTCTCAAATG; GFAP: Forward—ACATCGAGATCGCCACCTAC, Reverse—ATGGTGATGCGGTTTTCTTC; CD68: Forward—TCCAAGCCCAAATTCAAATC, Reverse—ATATGCCCCAAGCCTTTCTT; MAP-1: Forward—AGTGAGAAGAAAGTTGCCATCATC, Reverse—TTAATAAGCCGAAGCTGCTTAGG; CD11b: Forward—TGCCAAGACGATCTCAGCAT, Reverse—GCCTCCCACCACCAAAGT; Hes-1: Forward—CAACACGACACCGGACAAAC, Reverse—GTGGGCTAGGGACTTTACGG; Hes-5: Forward—GGTACAGTTCCTGACCCTGC, Reverse—AGAGGGTGGGCCCTGATTAT; vWF: Forward—CTACCTAGAACGCGAGGCTG, Reverse -CATCGATTCTGGCCGCAAAG; GAPDH: Forward—CCAGGTTGTCTCCTGCGACT, Reverse—ATACCAGGAAATGAGCTTGACAAAGT.
Mice were sacrificed via ketamine/xylazine injections, and brains were collected. The hypothalami were then excised and minced with a razor blade on an ice-cold glass plate and placed in a microfuge tube with 1 ml of hibernate A (HA-LF; Brian Bits, Springfield, Illinois, US). Hibernate A was then replaced with 1 ml Accutase (SCR005, Millipore, Temecula, California, US), and tubes were rotated for 30 minutes at 4°C. Samples were centrifuged at 425 x g for 2 minutes and each pellet was resuspended in 250 μl of ice-cold Hibernate A [82]. For cell dissociation, samples were triturated 10 times with a large Pasteur pipet and then placed on ice. Large pieces were allowed to settle, and 600 μl of supernatant was transferred to a 15-ml Falcon tube on ice. 600 μl of Hibernate A was added to the original tube, and the same procedure was repeated with medium and small Pasteur pipets. The collected supernatants were transferred to a 15-ml Falcon tube. Lastly, 750 μl of Hibernate A was added to the original tube, and 800 μl of supernatant was added to the 15-ml Falcon tube. Large debris was removed from the cell suspension by serial filtration through 100-μm and 40-μm cell strainers into 50-ml Falcon tubes, respectively (Falcon 352360; Falcon 352340; BD Biosciences, San Jose, California) [82]. The cell suspension was then centrifuged at 300 x g for 10 minutes and supernatant was aspirated completely. 100 μl of buffer (PBS +5% FBS) per 106 nucleated cells was added to the pellet. Then, 10 μl of ACSA-1 antibody (MACS Cat. #130-095-814) was added, mixed well, and incubated for 10 minutes in the dark. Cells were washed by adding 1 ml of buffer and centrifuged at 300 x g for 10 minutes. The supernatant was then aspirated completely. Lastly, the cell pellet was resuspended in 500 μl of buffer. Cells were sorted in FACSAria (BD Biosciencs, San Jose, California) using tdTomato and ACSA-1-APC appropriate wavelengths (581 nm and 660 nm, respectively) [83]. Astrocytes were isolated from IRloxp (tdTomato− APC+), and IRKOGFAP (tdTomato+ APC+). In addition, nonastrocyte cells were isolated from IRKOGFAP (tdTomato− APC−/ tdTomato+ APC−/ tdTomato− APC+). RNA from these cells were purified to determine IR gene expression [84].
Mice were sacrificed via ketamine/xylazine injections, and brains were collected, then excised and minced with a razor blade on an ice-cold glass plate and placed in a microfuge tube with 1 ml of hibernate A (HA-LF; Brian Bits, Springfield, Illinois). A similar procedure was followed to isolate brain cells, as previously described in the FACS method section. Then, astrocytes expressing NA+-dependent glutamate transporter (GLT-1) were positively selected using rabbit anti GLT-1 antibody (Cat. #OSE0004W, ThermoFisher Sci) and goat antirabbit IgG magnetic beads (Cat. #S1432S, Biolabs). Full details of the procedure were described previously [85]. For protein expression, isolated astrocytes were lysed in RIPA buffer (Cat. #SC-24948, Santa Cruz Biotech). Lysate was centrifuged, followed by BCA assay to determine protein concentration. The primary antibodies used were as follows: IRβ (Cat. #3025S, Cell signaling); PGES2 (Cat. #bs-2639R, Bioss) [86, 87]; and GADPH (Cat.# SC-32233, Santa Cruz Biotechnology). Secondary antibodies used were as follows: goat antirabbit-800 (LI-COR, P/N 925–32211) and donkey antimouse-680 (LI-COR, P/N 925–68075). Images were captured using the LI-COR odyssey infrared imaging system, and only the contrast and brightness were adjusted for this purpose.
Adult males and females (in diestrus) were perfused at the age of 7–8 months. Brains of the mice were collected and postfixed with 10% formalin at 4°C overnight, followed by immersion in 10%, 20%, and 30% sucrose for 24 hours each. A sliding microtome was used to cut sections (35–40 μm) of the brain into five series [2, 88]. For immunofluorescence, these sections were permeablized in 1 x PBS / 0.4% Triton x 100 for 1 hour at room temperature. Then, they were blocked in 1% BSA/5% normal donkey serum in 1 x PBS/Triton 0.4% at room temperature for 1 hour. After that, tissues were incubated with primary antibodies in blocking buffer at 4°C overnight, followed by five washes in PBST, with each wash lasting 10 minutes. Then, the tissues were incubated with secondary antibodies in blocking buffer for 2 hours at room temperature, followed by five washes in PBST. Sections were mounted on slides, air-dried overnight, and coverslipped with fluorescence mounting medium containing DAPI (Vectasheild, Vector laboratories, Inc. Burlingame, California). Brain sections were visualized for the expression of tdTomato, GFAP, and NeuN fluorescence in IRKOGFAP mice using Total Internal Reflection Microscopy (B&B microscopy limited Olympus IX-81) and Confocal Microscopy (Leica) and captured via Metaphore for Olympus Premier software. The primary antibodies used are as follows: anti-dsred 1° antibody ([1:50] Clone Tech, Cat. #632496), rabbit anti-GFAP polyclonal antibody-FITC conjugated (Bioss, Cat# bs-01994-FITC), and rabbit anti-NeuN ([1:100] abcam, Cat. #ab177487). The secondary antibodies used are as follows: Alexa Fluor 594 (1:1,000, Life Tech, Lot #1256153) and Alexa Flour 488 (1:1,000, Thermofisher Scientific, Cat. #A-21206). Only the contrast and brightness were adjusted during imaging.
Males and females were checked for onset of puberty daily starting after weaning at 3 weeks of age. Balanopreputial separation in males was checked by attempting to manually retract the prepuce with gentle pressure. For females, vaginal opening was checked daily [89]. Thereafter, vaginal lavages were collected from experimental mice for at least 3–4 weeks. Cytology of collected cells was examined to assess estrus stages. Predominance of leukocyte cells was taken to indicate a diestrous stage, predominance of nucleated cells a proestrous stage, and predominance of cornified epithelial cells an estrous stage [90, 91]. First estrous was defined as the first day of predominant cornified epithelial cells after the completion of one initial estrous cycle. For fertility studies, adult control IRloxp and IRKOGFAP females 3–4 months old were placed with WT males. Length of time until birth of the first litter and litter size were then determined [2]. The mice were paired for 8 days, and copulatory plugs were observed for evidence of successful mating. After that, mice were separated, and the delivery date was recorded. Similar procedures were used for IRloxp and IRKOGFAP male mice paired with WT females.
IRloxp and IRKOGFAP male mice were paired with WT females on the day the female was in proestrus. IRloxp and IRKOGFAP females were paired with experienced vasectomized males. Mating behavior was captured using infrared cameras (Swann) placed beside individual cages. Mice were placed in the procedure room at 1 PM to acclimate to the new environment and then the lights were turned off at 6 PM to begin the dark phase. After 2 hours in the dark (8 PM), a female in proestrus was introduced into each cage with a single male. Filming began at 8 PM and continued until 2 AM. The following morning, the female mice were checked for copulatory plugs, as previously described [92]. The video files were collected and analyzed for specific hallmarks of female sexual behavior, such as lordosis events and latency to first lordosis, as well as indicators of male sexual interest, such as latency to first mount and number of mounting attempts. A single-blinded rater completed the analysis to ensure consistency and reliability.
Submandibular blood was collected from IRloxp and IRKOGFAP diestrus female and male mice between 8–10 AM in randomly cycling mice to avoid the rise in LH that occurs on proestrus afternoon. LH and FSH levels were measured using multiplex testing performed by the University of Virginia Center for Research in Reproduction (Charlottesville, Virginia). Multiplex LH and FSH levels were measured with intra-assay CV < 20% and reportable range of 0.24–30 ng/ml for LH and 2.4–300 ng/ml for FSH. Female serum estradiol was measured using ELISA (Calbiotech. Spring Valley, California) with sensitivity of 3 pg/ml and intra-assay CV < 10.5%. Male serum testosterone levels were measured by ELISA (Calbiotech. Spring Valley, California) with sensitivity of 0.1 ng/ml and intra-assay CV of 3.17% [93].
At 6–7 months of age, adult males and diestrous females were perfused with 10% formalin and organ tissues including the testis or ovary were collected and postfixed immediately in 10% formalin overnight. Next, the tissues were kept in 70% ethanol overnight. Then, tissues were embedded in paraffin, cut into sections, and stained by hematoxylin and eosin [2]. Histological section were visualized via Olympus BX61US microscope (X-cite 120 LED boost EXCELITAS technology) and captured via OlyVia 2.9 software. Ovary sections (4 per mouse) were analyzed by evaluating follicle maturation, including counting the number of primordial, primary, secondary, and preovulatory follicles and corpora lutea. Testes sections were analyzed by evaluating sperm stages, including counting the number of spermatogonium, spermatocytes, spermatid, and spermatozoa. Sperm and follicle counts are reported per seminiferous tubule/ovary cross-section. Only the contrast and brightness were adjusted during imaging.
Data are presented as the mean ± SEM. Two-tailed, unpaired t testing was used for comparisons of two groups. One-way ANOVA was used to compare three groups, followed by Bonferroni multiple comparison test. Chi-squared test was used to analyze statistical differences in fertility studies. Data were analyzed using Prism 6 software (GraphPad). P < 0.05 was considered statistically significant. The numerical data used in all figures are included in S1 Data.
|
10.1371/journal.pntd.0006471 | Interruption of onchocerciasis transmission in Bioko Island: Accelerating the movement from control to elimination in Equatorial Guinea | Onchocerciasis, also known as river blindness, is a parasitic disease. More than 99 percent of all cases occur in Africa. Bioko Island (Equatorial Guinea) is the only island endemic for onchocerciasis in the world. Since 2005, when vector Simulium yahense was eliminated, there have not been any reported cases of infection. This study aimed to demonstrate that updated WHO criteria for stopping mass drug administration (MDA) have been met.
A cross-sectional study was conducted from September 2016 to January 2017. Participants were 5- to 9-year-old school children. Onchocerciasis/lymphatic Filariasis (LF, only in endemic districts) rapid diagnostic tests (RDTs) were performed. Blood spots were collected from RDT positive children and 10 percent of the RDT negatives to determine Ov16 and Wb123 IgG4 antibodies through enzyme-linked immunosorbent assay (ELISA). Skin snips were collected from RDT positives. Filarial detection was performed by PCR in positives and indeterminate sera. Black fly collection was carried out in traditional breeding sites. A total of 7,052 children, ranging from 5 to 9 years of age, were included in the study. Four children (0.06%) were Ov16 IgG4 RDT positives, but negative by ELISA Ov16, while 6 RDT negative children tested positive by ELISA. A total of 1,230 children from the Riaba and Baney districts were tested for LF. One child was Wb123 RDT positive (0.08%), but ELISA negative, while 3 RDT negative children were positive by Wb123 ELISA. All positive samples were negative by PCR for onchocerciasis and LF (in blood spot and skin snip). All fly collections and larval prospections in the traditional catching and prospection sites were negative.
WHO criteria have been met, therefore MDA in Bioko Island can be stopped. Three years of post-treatment surveillance should be implemented to identify any new occurrences of exposure or infection.
| Onchocerciasis, commonly called river blindness, is a chronic parasitic disease particularly prevalent in Africa. It is transmitted through the bites of infected Simulium blackflies. Onchocerciasis is endemic in Equatorial Guinea. Huge achievements have been made in human and vector control during the last two decades, especially on Bioko Island. Eliminating onchocerciasis transmission on Bioko is feasible given its isolation from other landmasses, which also reduces the risk of reinvasion by the disease vector. Recently updated WHO guidelines for stopping mass drug administration (MDA) and verifying elimination of human onchocerciasis (2016) established a new critical threshold to verify elimination of onchocerciasis transmission based on novel serological tests. We applied these techniques in a representative sample of 5- to 9-year-old school children. An entomological assessment was also carried out. We found no evidence of current infection or recent transmission. There was no evidence of onchocerciasis vectors, and our results from the sample population meet the current WHO serologic criteria for stopping MDA. Based on these results, we recommended to the Ministry of Health and Social Welfare of Equatorial Guinea that MDA on Bioko Island be stopped and that 3 years of post-treatment surveillance should be undertaken to identify any new occurrences of exposure or infection.
| Onchocerciasis is a parasitic disease caused by the filarial worm Onchocerca volvulus. It is transmitted through the bites of infected Simulium blackflies, which breed in fast-flowing streams and rivers. Symptoms include rashes, severe itching and various skin lesions, and blindness. The disease is endemic in 31 countries in sub-Saharan Africa, two countries in Latin America, and in Yemen. An estimated 18 million people are infected with the disease and have dermal microfilariae. 99% of the infected individuals live in Africa [1,2].
Human onchocerciasis is one of the two filarial helminth “neglected tropical diseases” targeted for geographically local elimination [3]. In the Americas, onchocerciasis elimination has traditionally been considered feasible as most onchocerciasis foci were confined and usually small. Since 2013, the World Health Organization (WHO) has certified four countries in Latin America as free of human onchocerciasis [4]. In Africa, where onchocerciasis has been endemic over vast areas, with highly efficient vectors and much higher endemicity levels, elimination was not initially considered to be feasible [5].
The Onchocerciasis Control Programme in West Africa (OCP) was launched in 1974 by the World Health Organization (WHO), followed by the African Programme for Onchocerciasis Control (APOC), initiated in 1995. Both programs established mass treatment with ivermectin combined with aerial spraying of breeding sites with selected insecticides in fast-flowing rivers as principal methods for controlling onchocerciasis [3]. Great progress has been made towards elimination. In most OCP/APOC countries, nationwide onchocerciasis elimination now seems to be an obtainable objective [6]. This paradigm shift from control to elimination occurred in 2012 due to success in Latin America, the shift to integrated NTD control, and the successful elimination of onchocerciasis in some parts of Africa [7].
The island of Bioko is part of the Republic of Equatorial Guinea and is the only island in the world where onchocerciasis is endemic [8]. Initially, the island was considered free of loiasis. The presence of intermediate hosts [9], and the recent reporting of an imported case of loiasis in a US traveler returning from the island [10] call for attention and further research. Two out of four districts have reported cases of Lymphatic Filariasis (LF). In 1990, several control activities were launched by the OCP, including long-term ivermectin mass treatment in all 52 island communities [11,12]. Afterwards, APOC became the sponsoring agency and introduced community-directed treatment with ivermectin (CDTI) throughout the island in 2000 [8]. In addition, a vector elimination project started when APOC was established in 1995. A feasibility study was carried out in 1996, confirming the high vectorial efficiency of the endemic Bioko form of Simulium yahense [13] and the distribution of the vector breeding sites [14]. From 2001 to 2005, a large-scale larviciding trial using ground-based applications was undertaken using helicopter and ground-based applications of temephos [8,15]. In 2005, the endemic Bioko form of S. yahense was finally eliminated from the island. Since then, there has not been any reported transmission or any serious epidemiological situation in Malabo City or elsewhere on the island [8]. According to personal communication from the Ministry of Health and Social Welfare of Equatorial Guinea (MINSABS in Spanish), the last MDA with ivermectin was administered in 2012 in urban Malabo and in 2016 elsewhere on the island.
A 2014 cross sectional study found no positive MF skin snip assessment in 544 study participants ages 5 years and older [16,17]. In the recently updated WHO guidelines for stopping mass drug administration and verifying elimination of human onchocerciasis (2016), new tools were proposed to verify the transmission interruption, stop CDTI and begin post-treatment surveillance. Following the WHO updated recommendations, the objectives of this study were: a) to verify onchocerciasis transmission interruption in Bioko Island, Equatorial Guinea; b) to validate a methodology to assess Ov16 prevalence in children younger than 10 years of age and; c) to develop a protocol to verify onchocerciasis elimination that can be applied in other African countries with hypoendemic intensity of transmission or where mass drug administration (MDA) has been conducted for a number of years. Data on LF were also collected as part of the study.
The Island of Bioko is part of the Republic of Equatorial Guinea, which also includes Rio Muni on the mainland and the island of Annobon. It is located in the Bay of Guinea in Central Africa, about 40 km southwest of the Cameroon coast. Bioko Island covers an area of approximately 2,017 km2 (779 square miles). It is 72 km (44.7 miles) long and is divided into four districts (Malabo, Luba, Riaba and Baney). Bioko Island has 334,463 inhabitants. Most of the population is concentrated on the northern part of the island in the Malabo district, where Malabo, the capital of Equatorial Guinea is located [18]. Tropical rainforest covers much of the interior of the island and the topography is characterized by steeply sloping volcanoes and calderas. Bioko Island has a humid tropical climate with an average annual temperature of 25°C and two distinct seasons: a dry season from November to March and a rainy season from April to October.
A cross-sectional study was conducted from September 2016 to January 2017. The eligible participants were 5- to 9-year old school children who had lived in Bioko Island for the past three years. According to the most recent WHO guidelines, a sample size of 1,100 to 2,000 children younger than 10 years of age per administrative unit is required for Ov-16 serology testing in order to detect a prevalence of less than 0.1% at the upper bound of the 95 percent confidence interval. When the eligible population of children is less than 1,100, all eligible children are to be tested [19].
First, we obtained an updated census report on school-aged children from the Ministry of Education. According to this data, sampling was not required in two districts (Riaba and Luba), since the population was below 1,100. All the children from the Baney district and rural Malabo who met the inclusion criteria were included in the study because the numbers were lower than first estimated. In urban Malabo, a random sampling method was implemented.
A second visit took place in May 2017 to obtain a second RDT and blood spot for ELISA/PCR in patients whose results were RDT/ELISA positives and/or were in the detection threshold limit.
Entomological surveys were carried out in the rivers with potential breeding sites.
Two teams, consisting of three local technicians, one supervisor and two coordinators were assembled and trained before the field work began.
Prior to starting the study, a comprehensive field training program was provided along with training on the proper use of diagnostic tools. The participation questionnaire was pre-tested one week before the beginning of the project. Questionnaires were provided to each school administrator, who distributed them to the children before the science team came to the school to conduct the testing. Children were instructed to take the forms home and have their parents or guardians complete them before they were returned. At least two reminders were given before the testing date. The form required the participant’s sex, birthdate, age, and asked whether the child had lived in Bioko Island for the past three years and if the child had ever take ivermectin. Test results were later recorded in the same document.
The geographic coordinates of each school (latitude and longitude) were collected using the offline maps application MAPS.ME.
This study combined the use of serological tests based on recombinant antigens with the later confirmation of diagnostic results by molecular diagnosis. The SD BIOLINE Onchocerciasis IgG4 rapid diagnostic test, manufactured and distributed by Standard Diagnostics, Inc. (SD), was performed following the product instructions. Further information on this technique has been described elsewhere [20]. Sterile techniques were used to obtain finger-prick blood samples. Fingers were cleaned using an alcohol swab and sterile cotton balls. Technicians wore disposable gloves. All materials were safely disposed of. In those districts considered co-endemic to lymphatic Filariasis and onchocerciasis (Riaba and Baney), SD BIOLINE LF IgG4 RDT was also performed, together with SD BIOLINE onchocerciasis/LF IgG4 biplex.
Blood spots were collected from all the children with either a positive Ov RDT or a positive LF RDT and from a random sample of 10 percent of the negative RDT children for determination of Ov16 and Wb123 IgG4 antibodies through enzyme-linked immunosorbent assay (ELISA). This subsample (10% of negative RDT) was randomly selected by equal probability systematic sampling. Every tenth RDT negative child was selected for further blood spots collection.
Sterile techniques were used to place finger-prick blood spots onto circles on 5 x 5 cm Whatman 903 protein saver cards. Skin samples were obtained near the iliac crest of each individual through Walser matrix forceps. Samples, stored and transported at 4°C, were delivered to the National Center of Microbiology (NCM) laboratory at the Institute of Health Carlos III in Madrid for further analysis.
An ELISA protocol was run to detect anti-Ov16 IgG4 antibodies in the eluted blood from the Whatman filter paper (S1 ELISA Protocol). Plates were sensitized with 0.5 μg/ml of Ov16 recombinant protein, obtained and purified as described in Hernández-Gónzalez et al., 2016 [16]. A poly-His tail was added to the carboxy-terminus of the Ov-16 sequence amplified from a vector (kindly donated by Professor JE Bradley, School of Life Sciences and University of Nottingham, UK). Then, the construction was subcloned into pGEX-6P-1 plasmid (GE Healthcare, Little Chalfont, UK). Further expression and purification steps were undertaken as explained in the indicated paper. Two positive controls were included: (i) an anti-Ov16 human recombinant monoclonal antibody (hIgG4, Bio-Rad) and (ii) a pool of positive sera from patients with onchocerciasis (clinical and parasitological confirmation). A standard curve from each positive control was used to identify positive samples on each plate allowing comparisons among plates and days. The anti-Ov16 IgG4 mAb positive control was diluted as follows: 12, 6, 3, 2, 1.5 and 1 ng/ml and in the case of the positive sera, pool dilutions ranked from 1/400 to 1/3.200. The cut- off for the recombinant positive control (anti-OV16 mAb; stock: 2.9 mg / ml), was set at 2.01 ng / ml following the indications of Golden et al., 2017 [21]. The dilution 1/800 from the pool of positive sera was chosen as the cut-off, with optical densities similar to those obtained with the anti-Ov16 mAb at 2.01 ng/ml.
A similar ELISA protocol was run to detect anti-Wb123 IgG4 antibodies. The only difference in methodology was that wells were sensitized with Wb123 recombinant protein. The Wb123- pUC57 construction, corresponding to the GenBank HQ438580 sequence, was obtained from Genscript (Piscataway, NJ, USA) and further directionally subcloned in pQE-30 expression vector. Isopropyl β-D-1-thiogalactopyranoside (IPTG) 0.01 mM was used to induce protein expression with ON incubation at 16°C. The Wb123 recombinant protein was purified by Ni2+- Sepharose 4B affinity chromatography (GE Healthcare) in native conditions and according to the manufacturer’s recommendations. Finally, the Wb123 recombinant antigen was dialyzed against PBS and quantified using the Pierce BCA Protein Assay Kit (Thermo Scientific Rockford, IL, USA). A standard curve was developed; in this case, with serial dilutions (1/400 to 1/3,200) of a Wuchereria bancrofti positive sera pool (WHO collection provided by Dr. N. Weiss, Swiss Tropical Institute, Basel). The chosen cut-off was the OD corresponding to the 1/1600 control sera dilution.
In both ELISAs (Ov16 and Wb123), the sera were classified as positive, indeterminate or negative, following the criteria described below:
Molecular protocols (PCR) were undertaken to confirm serological positive samples: blood spots for FL and onchocerciasis positives and skin snip for Onchocerca positives. DNA extraction from skin snips specimens and the blood spots were performed with the QIAamp DNA mini kit (QIAGEN, IZASA, Madrid). Filarial detection was carried out by two independent Polymerase chain reaction (PCR) methods. A real-time PCR, modified by Tang et al. (2010) [22], targets the internal transcribed spacer in one region of the ribosomal gene (ITS-1), which enables the identification of positive samples by post reaction analysis by melting temperature (Tm) curve of the amplified fragments (Tm = 77.50°C ± 1.0°C) and by the size of the amplified fragment (344bp for O. volvulus, 312bp for Mansonella spp., 301bp for W. bancrofti and 286bp for Loa loa) and a conventional PCR targeting of the mitochondrial COI gene [23] (see further details in S1 Table). All samples were made in duplicate.
PCR products were analyzed by automatic electrophoresis (QIAxcel, QIAGEN, IZASA, Madrid) or conventional electrophoresis in 2 percent agarose gels stained with Pronosafe (Pronadisa, Madrid). The confirmation of the filarial species was performed by sequencing the amplified fragment using the Big Dye Terminator v3.1 Cycle Sequencing Kit (Applied Biosystems, Massachusetts, EEUU) on an ABI PRISM 3700 DNA analyzer (Applied Biosystems, Massachusetts, USA). Previously, PCR products were purified using the Illustra DNA and Gel Band Purification Kit (General Electric Healthcare, Little Chalfont, UK). All amplified products were sequenced twice in both directions.
All sites accomplishing APOC experts’ criteria for harboring larvae of the simulids were visited. Black fly collection was carried out in representative catching sites by one team in each community, consisting of two fly collectors and a human attractant (human bait). The human bait method involved having the individual sit in one place barefoot for up to one hour. During this worktime, the fly collector exposed his legs and caught all landing Simulium flies with the help of an exhaustor (“pooter”). Collections began, local time, at 0700 and ended at 1700. Collectors received ivermectin 1 week before beginning the collection process.
Individual data, RDT and laboratory results were analyzed to obtain frequencies of each variable. Prevalence (with 95 percent confidence intervals) of onchocerciasis and LF were calculated from RDT results. In three out of the four districts (and rural Malabo), all the eligible children attending school the day of the visit were included. It was not confirmed whether any children were absent on the day of the survey or whether any local children were not attending school at all. Prevalence and confidence limits were computed following the recommendations of Brown et al. for interval estimation for a binomial proportion [24]. The Wilson interval for small n (when n<100), and the interval suggested in Agresti and Coull for larger n (when n≥100), were used.
A univariate analysis was performed to explore if any variable showed a significant relationship with positive onchocerciasis RDT cases. Analyses were performed using the statistical package Stata 14.0. Schools were mapped using the free software QGIS version 2.18.7.
The study was approved by the MINSABS on Bioko Island and the research ethics and animal welfare committee at the Health Institute Carlos III (ISCIII in Spanish) in Spain, under the number CEI PI 22_2016-v3. The school headmasters and children´s parents/guardians were informed of the day of the visit and the scope of the study by an official letter from the MINSABS. Written informed consent was obtained from all parents or guardians prior to study inclusion. Data were analyzed anonymously.
A total of 7,052 school children, ages 5 to 9 years old and living in Bioko Island for the last three years or more, participated in the epidemiological assessment. The study sample was recruited from 147 schools, distributed all around the Island (Fig 1). In the Riaba, Luba and Baney districts and rural Malabo, all of the eligible 5- to 9-year-old children agreed to participate and were thus included. In urban Malabo, several parents did not sign the study agreement form (around 25%), and replacements for these individuals were found to obtain a sample of 4,342 randomly selected children.
The majority of the 5- to 9-year-old children lived in urban Malabo. Overall, ages were evenly distributed. The sample was 52.4% female. 96 percent of the respondents reported that their children had never received ivermectin (Table 1).
Finger-prick blood samples for Ov16 IgG4 RDT were obtained from all 7,052 school children. From the overall sample, four children (0.06 percent) were found to be positive for Ov16 IgG4 antibodies by RDT (95 percent CI upper limit = 0.11). One was from the Riaba district and three lived in rural Malabo (Fig 2). No significant association with onchocerciasis RDT positive cases was found for any variable, except for district. Rural Malabo presented the highest frequency of onchocerciasis RDT positive cases (n = 3; p<0.001).
Blood spots were collected from 720 school children. Five of them were obtained from positive individuals and 715 (10 percent) were collected from a randomized selection of RDT negative children. Skin snips were obtained from the RDT positive children (n = 4). The 4 onchocerciasis RDT positive children were negative by ELISA Ov16, while 6 RDT negative children from urban Malabo tested positive to onchocerciasis by ELISA (prevalence = 0.83 percent; 95 percent CI = 0.34–1.85). Three samples were considered as indeterminate sera, because they were close to the ELISA cutoff threshold. PCR analysis was negative for all positive/indeterminate blood and skin samples (n = 13; Table 2).
A total of 1,230 children from Riaba and Baney were tested for lymphatic Filariasis. In one Baney district school (with n = 19), the LF IgG4 RDT was not performed due to logistical problems. One 9-year-old girl (0.08%) from Baney was found positive for Wb123 by LF IgG4 RDT (Table 3), while no child tested positive for Wb123, according to onchocerciasis/LF IgG4 biplex results.
Samples were processed by ELISA with recombinant Wb123. The LF RDT positive child was ELISA negative, while 3 children from Baney and 2 from Riaba who tested negative by RDT tested positive when sera was processed by Wb123 ELISA (Fig 2). One onchocerciasis RDT positive child was also positive to LF by Wb123 ELISA. All LF serologically positive children (n = 6) were negative by PCR (Table 3).
In May 2007, a second serum sample was obtained from 16 children who were either RDT positive (Onchocerciasis or LF) or ELISA positive/indeterminate for confirmatory test (n = 18). Two school children were unavailable during the second visit.
Out of the above population, 2 children were onchocerciasis RDT positive (prevalence = 0.03%; 95% CI = 0.00–0.11), 1 by Ov16 IgG4 RDT and the other by Oncho/LF biplex IgG4 RDT. Both were negative by Ov16 ELISA and blood spot/skin snip PCR. The child, who was previously found positive to LF, was negative by biplex RDT in this second round. The second Ov16-IgG4 ELISA analysis was negative for all the samples (n = 16) except one. Regarding those samples that were positive by ELISA with recombinant Wb123 in the preliminary analysis (n = 5), 3 remained positive, 1 became negative and the fifth one had been collected from one of the two unavailable children. Skin snips were obtained from the confirmed biplex RDT (n = 2) and Ov16 ELISA (n = 1) positive children. PCR analysis was negative for all of them (Table 4; S2 Table).
Updated WHO guidelines established that serologically positive children found negative by PCR testing of skin snips are considered negative for patent infection with O. volvulus and are accepted as not contributory to the 0.1% threshold calculation [19]. According to this criterion, no child was positive for onchocerciasis (Table 2). The 2 RDT and the remaining ELISA positive child (according to the sera obtained on the second visit) were considered as O. volvulus “exposed.” The MINSABS will re-examine them 1 to 1.5 years after the first visit to determine if they have developed patent infection. If so, they will be treated accordingly, following WHO recommendations [19].
Entomological assessment was performed from the 29th of August to the 23rd of September, 2016. Due to fieldwork difficulties and the need for extra personnel, the Ureka area (breeding sites placed in rivers Mohaba, Ehola and Osha) was visited in February 2017 (Fig 3). All fly collection and breeding site prospection were negative.
We conducted an evaluation of O. volvulus transmission in Bioko Island, Equatorial Guinea, guided by updated WHO guidelines [19]. We found no evidence of current infection or recent transmission: there was no evidence of onchocerciasis vectors, and our results in 5- to 9-year-old children meet the current WHO 0.1 percent serologic criteria for stopping MDA. Moreover, none of the children tested positive when RDT was combined with ELISA. Based on these results, we recommended to the Ministry of Health and Social Welfare in Equatorial Guinea (MINSABS) that MDA in Bioko Island be stopped and that 3 years of post-treatment surveillance should begin, to identify any new occurrences of exposure or infection.WHO defines disease elimination as the reduction to zero of the incidence of infection caused by a specific pathogen in a defined geographical area, with minimal risk of reintroduction, as a result of deliberate efforts [25]. To measure the progress towards elimination, three phases have been defined for onchocerciasis control programs. The phase 1 or intervention phase is characterized by regular ivermectin treatment with a minimum requirement of 80 percent therapeutic coverage. This phase typically lasts at least 12 to 15 years, which corresponds to the reproductive lifespan of the adult worm when exposed to drug pressure [19]. Phase 2, also called “post-treatment surveillance” follows the intervention phase and typically lasts 3 to 5 years. The final phase (phase 3) starts at the end of the 3 to 5 years of post-treatment surveillance and is also known as “post-elimination surveillance.” It follows the confirmation of the initial assessments at the end of phase 2, thereby providing strong evidence that transmission has been permanently eliminated in a country. According to our results, Bioko Island is currently in phase 2. On Bioko Island, initial onchocerciasis control activities began in 1990, using ivermectin distributed by mobile teams [12,26]. After 8 years, a significant reduction in onchocerciasis prevalence (from 74.5 to 38.4 percent), and community MF load (from 28.3 to 2.3) [11] was achieved. Afterwards, APOC became the sponsoring agency and introduced CDTI throughout the island in 2000 [8]. Since then, therapeutic coverage has ranged from 51 [11] to 75 percent [27]. Although the therapeutic coverages have been relatively low, treatment intervention has been steady for more than 20 years. On the other hand, Equatorial Guinea (Bioko island in particular) is one of the three African countries (Uganda and the United Republic of Tanzania being the other two) that supplemented MDA together with vector control [19]. Furthermore, given that Bioko Island has been free of onchocerciasis vectors for the past 12 years [3,8] and the insular isolation of Bioko (32 km from the Cameroon coast and 400 km from the Guinean mainland coast), WHO might reconsider if the post treatment surveillance is truly needed in this case.
This study also reports the first use of anti-Ov16 IgG4 RDT in a primary field screening followed by confirmatory ELISA/PCR in the laboratory to certify onchocerciasis elimination in an African endemic area. Several markers for infection have been used for measuring onchocerciasis disease burden; however, most of them seem to be insensitive to certifying elimination [28]. Commonly used methods for diagnosing O. volvulus infections (microscopic detection of microfilariae (MF) in skin snips and nodule palpation) are insensitive, especially when MF skin densities are low. PCR of the skin snips may provide greater sensitivity but still require sampling skin snips. Moreover, the cost of performing PCR on entire populations, or even in sentinel groups, is high [29]. An alternative approach is applying antibody detection to O. volvulus specific antigens that are expressed by the larval stages of the parasite. This method seems to be particularly useful to identify incident infections in communities having already undergone MDA [30]. Several serological markers for exposure to onchocerciasis have been assessed in the past [31,32]. The most widely used and the one adopted as a tool for monitoring onchocerciasis control and elimination in the Americas is the Ov-16 antigen [33]. This test has been mainly performed by enzyme immunoassay (EIA or ELISA), that detects IgG4 to this antigen. IgG4 detection results are more reliable than IgG detection, leading to fewer false positive results, which is critical for its use in a low prevalence, elimination scenario [30]. In Sub-Saharan Africa, the Ov16 ELISA was used to prove that onchocerciasis transmission was interrupted in the Wadelai focus of Northwestern Uganda [34] and the Abu Hamed focus, in Sudan [35].
In this new paradigmatic scenario, where elimination seems obtainable, a standardized affordable, simple and rapid alternative for testing in the field is necessary. To fill this gap, a new RDT was recently developed by PATH [20,29]. To date, this Ov16 RDT accuracy has been mainly assessed in surveillance studies and/or African focus with high onchocerciasis prevalence. Its usefulness in defining when to stop MDA is still limited [29,36,37]. Our study is the first one to provide evidence on its accuracy and usefulness in an African focus with low-prevalence conditions prevailing at the end-stage of control programs.
Combining this method with other diagnostic strategies (ELISA in 10% of RDT negatives and RDT positives, and PCR in all seropositives), we also contribute to a better understanding of the dynamics of Ov16 antibody responses for accurate interpretation of seroprevalence data. In our study, four children (0.06%) were positive by RDT, but negative by ELISA and PCR. This prevalence was even lower after verification of results during a second visit (0.03 percent). Regarding the Onchocerciasis and Lymphatic Filariasis IgG4 biplex RDT, it also performed well in the present study. One positive child out of 1,230 (0.08 percent) was deemed negative by ELISA and PCR. The integration of two filarial antigens in one device is of added value in co-endemic areas where MDA has been ongoing for several years. It saves time, money, and human resources in regions where co-endemicity is suspected. Nevertheless, it may not provide accurate results in areas that are highly endemic for loiasis [38], which was not the case on Bioko Island. Therefore, the high sensitivity and specificity shown by both RDTs support their usefulness to determine the interruption of onchocerciasis transmission.
In summary, we have developed and tested a protocol for stopping MDA and evaluating onchocerciasis elimination following the updated WHO guidelines [19]. Its implementation on Bioko Island has been a success, but some considerations should be taken into account before applying it to other African countries. Previous research has proven that onchocerciasis elimination with ivermectin treatment might be feasible in some endemic foci in Africa [39,40]. In Bioko island, MDA was quite successful, but interruption of the disease transmission might have been partially or completely due to the disappearance of the Simulium vectors, as was the case in Uganda [34]. In the case of this study, the elimination of the vector, combined with the isolated nature of Bioko and the long history of MDA, were responsible for a particularly low baseline level. This should be taken into account before using this protocol in other contexts.
Our study has several limitations. First, the lower participation rate in urban Malabo may affect the interpretation of the results in this particular district. Moreover, the school attendance rates for 5- to 9-year-old children were not provided by the Ministry of Education (probably this information is not systematically recorded). Nevertheless, the sample size was larger than would have been required in this administrative unit according to the WHO guidelines. Second, the recombinant positive control antibody AbD19432_hIgG4 was complex to standardize, due to the reagent instability when it was diluted to the final concentration to be used in ELISAs. Secondary extractions from indeterminate serological samples (within the detection limit threshold) were obtained in May 2017 to verify the results. Only one of the conflictive samples requested yielded a positive absorbance value by both positive controls (recombinant control and serum control). Third, pre-analytical bias due to the complex field logistics might exist. To reduce these potential biases, detailed guideline and standard operational procedures (SOPs) were prepared and piloted prior to the field work. All the SOPs were also tested by internal control before and during the field work.
In conclusion, WHO criteria have been met in Bioko Island. Consequently, MDA can be stopped and 3 years of post-treatment surveillance should begin to identify any new occurrences of exposure or infection. Successful elimination of onchocerciasis infection throughout Equatorial Guinea may be a feasible goal for the relatively near future. To that end, a survey extending to the continental area is needed before talks about countrywide elimination begin.
|
10.1371/journal.pbio.1000085 | Coordinated Movement of Cytoplasmic and Transmembrane Domains of RyR1 upon Gating | Ryanodine receptor type 1 (RyR1) produces spatially and temporally defined Ca2+ signals in several cell types. How signals received in the cytoplasmic domain are transmitted to the ion gate and how the channel gates are unknown. We used EGTA or neuroactive PCB 95 to stabilize the full closed or open states of RyR1. Single-channel measurements in the presence of FKBP12 indicate that PCB 95 inverts the thermodynamic stability of RyR1 and locks it in a long-lived open state whose unitary current is indistinguishable from the native open state. We analyzed two datasets of 15,625 and 18,527 frozen-hydrated RyR1-FKBP12 particles in the closed and open conformations, respectively, by cryo-electron microscopy. Their corresponding three-dimensional structures at 10.2 Å resolution refine the structure surrounding the ion pathway previously identified in the closed conformation: two right-handed bundles emerging from the putative ion gate (the cytoplasmic “inner branches” and the transmembrane “inner helices”). Furthermore, six of the identifiable transmembrane segments of RyR1 have similar organization to those of the mammalian Kv1.2 potassium channel. Upon gating, the distal cytoplasmic domains move towards the transmembrane domain while the central cytoplasmic domains move away from it, and also away from the 4-fold axis. Along the ion pathway, precise relocation of the inner helices and inner branches results in an approximately 4 Å diameter increase of the ion gate. Whereas the inner helices of the K+ channels and of the RyR1 channel cross-correlate best with their corresponding open/closed states, the cytoplasmic inner branches, which are not observed in the K+ channels, appear to have at least as important a role as the inner helices for RyR1 gating. We propose a theoretical model whereby the inner helices, the inner branches, and the h1 densities together create an efficient novel gating mechanism for channel opening by relaxing two right-handed bundle structures along a common 4-fold axis.
| Maintaining a precise intracellular calcium concentration is key for cell survival. In skeletal muscle, ryanodine receptor type 1 (RyR1) is an intracellular calcium-release channel that is critical for contraction. Here, we used single-channel techniques to demonstrate the presence of functionally homogenous populations of RyR1 in either the closed or open state and then applied cryo-electron microscopy and image processing to determine the 3D structure of each state. The 3D structures show that RyR1′s ion pathway is formed by two sets of bundles, each containing four rods along a common axis. One set (inner helices) stretches from the lumen to the ion gate, whereas the second (inner branches) stretches from the ion gate to the peripheral cytoplasmic domains. The configuration of the two bundles is clearly different in the two physiological states, allowing a 4 Å increase in diameter of the ion gate upon opening. This diameter increase is sufficient to ensure flow of calcium ions. Upon gating, the cytoplasmic domains undergo a conformational change that converges on the inner branches, revealing a long-range allosteric mechanism that directly connects effectors acting on the cytoplasmic moiety with the ion gate.
| Maintaining a precise intracellular Ca2+ concentration that is 10,000-fold lower than the surrounding environment of the cell, and the ability to dramatically increase intracellular calcium to trigger downstream events in response to specific stimulus are key for cell survival [1]. Ryanodine receptors (RyRs) are high-conductance intracellular Ca2+ channels regulated by both exogenous and intracellular mediators, which release Ca2+ stored in the endoplasmic reticulum. RyRs are the largest ion channels known, with an average molecular weight of 2.26 MDa, with most of its mass (∼4/5) forming the cytoplasmic domain. The skeletal muscle isoform, RyR1, has a bidirectional interaction with the slow voltage-gated calcium channel in the cell membrane, or dihydropyridine receptor (DHPR), which acts as RyR1′s voltage sensor for cell membrane depolarization [2]. Two key questions to understand RyR1′s function are how are signals transmitted from peripheral cytoplasmic domains to the ion gate, and what is the gating mechanism itself.
Cryo-electron microscopy (cryoEM) and single-particle image analysis of frozen-hydrated RyR1 revealed the 3D structure of RyR1 at approximately 25–30 Å resolution [3–5]. Its cytoplasmic domain is shaped like a flat square prism of 290 Å side and 120 Å high, with at least 12 reproducible domains that have been assigned numerals 1–12 [6,7]. Using the same technique, it has also been possible to map the binding sites for several ligands: the FK506-binding protein 12 kDa (FKBP12), calmodulin (Ca2+-CaM, apoCaM), and imperatoxin A (IpTxA) [8–10]. All these interactions, which are known to modulate RyR1 gating, take place at distances at least 130 Å away from RyR1′s putative ion gate and suggest that RyR1 makes use of long-range allosteric pathways between the cytoplasmic sensing domains and the ion gate. The 3D reconstructions of RyR1 in the open conformation indicated several conformational changes involving both the cytoplasmic and transmembrane domains with respect to the closed conformation [11,12]; however, the resolution of these reconstructions (∼30 Å) is insufficient to understand the connection between the two domains or to distinguish the substructure within the transmembrane domain itself.
The wealth of atomic structures of K+ channels solved by X-ray crystallography obtained in the last decade has allowed extensive study of the structural rearrangements underlying ion gating for this channel family. In the prevalent model for the ion gating of the K+ channel, the inner helices bend outwards around their midpoint (through a Gly or a Pro-X-Pro hinge) to increase the diameter of the ion gate so that it becomes permeable to ion flow. These inner helices are connected to their sensing domains using a plethora of structural arrangements to respond to a variety of effectors (voltage, ion concentration, pH, redox state, small molecules, and ligands). However, with one exception [13], models of K+ channel gating have been deduced from comparison of unrelated K+ channels. The only other case in which structural data at near-atomic detail is available for both the open and closed states in the same channel is for the nicotinic acetylcholine receptor (nAChR), as determined by electron crystallography [14]. Unlike other K+ channels, the nAChR is a pentamer with its ion gate formed by a hydrophobic girdle in the middle of the membrane. Binding of acetylcholine induces a rotation in protein chains that communicates to the inner helices of the pore, resulting in modulation of the ion gate diameter. To date, nothing is known about the ion gating mechanism of RyR1.
Using cryoEM, we previously defined the architecture of RyR1′s transmembrane domain in the closed state at higher detail [7]. RyR1′s closed ion pore is defined by an axial structure formed by two sets of four rods each forming a right-handed bundle, which we defined as the inner helices, and the inner branches. The inner helices shape the core of the transmembrane assembly. The inner branches are in the center of the cytoplasmic assembly and are directly connected with the peripheral cytoplasmic domains. The two bundles converge into a ring of high density, which we presumed to be the ion gate. A second constriction, which would correspond to the selectivity filter, is on the sarcoplasmic reticulum (SR) luminal side of the transmembrane assembly. These two constrictions define a central cavity. The structure formed by RyR1′s inner helices in the closed state appears to be parallel to the canonic structure of the inner helices of closed K+ channels [15,16]. A second group of investigators has also reported the presence of the inner helices in the core of the transmembrane assembly of RyR1 in the closed state, achieving similar resolution using the same method and almost identical biochemical conditions [17]. Intriguingly, the conformation that they reported for the inner helices corresponded best to that of an open K+ channel, and suggested that the ion gating mechanism used by RyR1 must be radically different than that used by K+ channels.
To better understand the basis for RyR1′s gating and to solve the controversy on the conformation of the inner helices in the closed state, we sought to obtain the open and closed conformations of RyR1 in their (frozen) hydrated state using single-particle cryoEM. Furthermore, we used single-channel biophysical characterization of the two states in bilayer lipid membranes (BLMs) using identical samples and conditions, to have a more direct correspondence between conformation and biophysical state of the channel. Here, we present the first demonstration, to our knowledge, of a midlevel resolution 3D model for the open state of RyR1 bound to its accessory protein FKBP12. A 3D reconstruction of RyR1-FKBP12 in the closed state was obtained in parallel for comparison. Thus, in this study, we are able to directly compare both conformations of the same protein, rather than comparing related proteins. Furthermore, both structures correspond to the protein in its fully hydrated state, and both are correlated directly to a functionally characterized biophysical state. We found that upon opening, the cytoplasmic domain undergoes an overall conformational change that involves the connections with the transmembrane domain. In the transmembrane assembly, we find that the inner helices corresponding to the open and closed states of RyR1 have a high cross-correlation with parallel structures of K+ channels in the corresponding state. Nevertheless, the ion pathway of RyR1 has features not present in K+ channels, which has allowed us to create a novel heuristic model for RyR1′s ion gating.
To obtain the resolution necessary for the visualization of secondary structure (∼9 Å), it is critical to obtain a highly homogeneous dataset. Obtaining a homogeneous population of RyR1 in the closed state is relatively easy. By contrast, the typical flickering behavior of RyR1 under physiologic activating conditions represents a significant limitation, since it produces a mixed population of open and closed states, e.g., under maximum Ca2+ activating conditions (50 μM Ca2+ on the cytoplasmic side), the channel open probability (Po) of reconstituted purified RyR1-FKBP12 channels is less than 30% (unpublished data). Our previous studies using vesicles demonstrated that the neuroactive noncoplanar polychlorinated biphenyl 2,2′,3,5′,6-pentachlorobiphenyl (PCB 95) had an unprecedented activating effect on RyR1 [18,19], suggesting that it would be a candidate small molecule to stabilize RyR1′s open state. The BLM studies of reconstituted purified RyR1-FKBP12 channels indicate that PCB 95 stabilizes the full open (conducting) state in ten out of ten reconstituted channels, resulting in extremely long-lived openings interspersed with rare short-lived transitions to the closed state. This results in a mean Po of 0.96 and thus produces a highly homogeneous dataset (Figure 1C–1E). By contrast, addition of 2 mM EGTA to the cis chamber (pCa2+ < 108) after fusion of an actively gating channel completely stabilized the fully closed state of the channel with no gating transitions observed for the entire recording period (Po = 0) (Figure 1A, 1B, and 1E). High-affinity [3H]ryanodine binding experiments query the conformational state of a large number of RyR1s [20]. The presence of EGTA (2 mM) in the assay buffer negates specific binding of [3H]ryanodine because the channels are in a closed conformation. By contrast, the presence of PCB 95 and optimal Ca2+ produced nearly 10 pmol of binding sites per milligram of SR protein (∼35,000 disintegrations per minute [dpm]/25 μM of SR protein) at steady state (Figure 1F), indicative of the fact that the channels are stabilized in the open state. These biophysical and biochemical data provide two independent measures of the ability of PCB 95 to stabilize the open state of the RyR1 channel having a unitary current level indistinguishable from a native channel in the full open state. The unitary current is a fundamental parameter for any given channel [21], thus it is safe to assume that the PCB 95–stabilized RyR1 has a pore structure representative of the native open state (in which only the kinetic/thermodynamic parameters have been altered). To exert its effect, PCB 95 requires that RyR1′s FKBP12 accessory subunit be bound [22]. In vivo, FKBP12 is constitutively bound to RyR1 and is known to stabilize its fully closed state and minimize subconductance states [23,24]. Both the position and orientation of FKBP12′s atomic coordinates with respect to RyR1 have been mapped and have been shown not to alter RyR1′s closed-state conformation at 16 Å resolution [9].
Our RyR1 purification method [7] produced a single band on PAGE (Figure 2A) indicative of a pure RyR1 preparation, and RyR1s with well-preserved structure when viewed with cryoEM (Figure 2B). The relatively high concentration of RyR1, approximately 2 mg/ml, enabled the successful cryo-preparation of RyR1 suspended over holes instead of lying on a carbon support, a method that allows increased resolution of the 3D reconstruction because it considerably increases the randomness of orientations [7]. CryoEM and image processing of two frozen-hydrated RyR1-FKBP12 datasets corresponding to the open and closed states, with approximately 17,000 particles each, yielded two 3D reconstructions. The homogeneous angular distribution for both datasets (Figure 2C) indicates that all orientations are equally represented in both datasets; thus the two 3D reconstructions have isotropic resolution and are free of the missing-cone artifact [25]. The nominal resolution of the two reconstructions, 10.2 Å, was determined by Fourier shell correlation (FSC) using a cutoff criterion of 0.143 [26] (Figure 2D), which in this study was a conservative value relative to the five times noise-correlation cutoff. The resolution value of 10.2 Å appears reasonable, taking into account the fact that in general, positive identification of secondary structure is indicative of 9 Å or better resolution. We have focused our analysis on only those structures readily visible in the cryoEM density map without any further manipulation. Specifically, we have centered our study on structures with densities at least 2.8 σ levels above the mean value.
When comparing the 3D reconstructions corresponding to closed and open RyR1, they look rather similar (Figure 3). However, careful analysis reveals that they are different conformomers of the same molecule. The coarse conformational changes may be better appreciated when the two 3D reconstructions are filtered to lower resolution and directly superimposed (Figure 4A), or when the two 3D structures alternate between the closed and open states (Video S1). Whereas most of the domains appear to move, the largest conformational changes take place in the distal regions of the cytoplasmic domains. The larger conformational changes are also evident in the 3D difference maps (Figure 4B). The difference was performed in both directions (closed minus open, and open minus closed), providing the regions of mass that were exclusive for the closed and open states, respectively.
Because the open- and closed-state datasets were processed in parallel, starting from a common low-resolution structure, and result in clearly different conformations, we believe that these are genuine representations of the two physiological states. Furthermore, given the large dimensions of the RyR1 (e.g., 30× larger than the K+ channel KcsA), domains separated by more than 100 Å may be regarded as resolved independently from each other. Yet, these domains follow the same direction of movement when they are connected by intervening density. Finally, for each domain that moved, there is a pair of complementary differences (see Figure 4B), which is also indicative of high data quality and actual movement.
The largest-magnitude conformational change occurs in the cytoplasmic domain, whereby each of the quadrants swivels outwards. The corners or clamp domains (domains 9 and 10) together with the structure formed by domains 7, 8, and 8a move away from the T-tubule and towards the SR membrane by approximately 8 Å. Concomitantly, domain 2, more central and facing the T-tubule, moves approximately 4 Å towards the T-tubule, and outwards away from the 4-fold axis (Figure 4A). We do not see an opening of the clamp domains in the open state as was suggested previously (see Discussion). Domain 6, protruding towards the T-tubule, moves approximately 5 Å outwards when the channel is in the open state, and a similar magnitude of outward movement takes place at domain 11, facing the SR membrane. The main effect of this swiveling movement is that the mass moves from the center to the outside, making the 4-fold axis less crowded. This movement in the cytoplasmic regions is clearly conveyed to the inner branches (Figure 3A and 3C, Video S1).
In the closed state, the overall structure of the inner helices and inner branches of RyR1-FKBP12 display a structure almost identical to the structure of the closed state of RyR1 that we determined previously in the absence of FKBP12 [7] (compare Figures 3B, 3D, and 5A). As in our earlier report, the inner helices have a tepee-like arrangement that overlaps directly with the tepee structure described for the ion pathway in the atomic models of K+ channels [15,16,27–29] (e.g., see Figure 6). Although a resolution of 9 Å or better is needed to visualize α helices [30,31], it has been described that resolution of 10 Å or even less may suffice to identify α helices, if they are separated from surrounding structures [32]. Another report of the closed state of RyR1 at 10 Å [17] also indicated four inner helices in the same location—although in a different configuration—supporting this finding (Figure 5B). When compared to our closed-state reconstruction, the inner branches in the open state are clearly recognizable but in a different conformation, and the central passage has significantly lower density than in the closed state (stereo pairs shown in Figures 3B and 5A). The inner branches and inner helices define three main constrictions along the 4-fold axis, represented in Figure 7. The upper, or cytosolic, constriction is defined by the distal enlargement of the inner branches (Figure 7A). The meeting point between the inner branches and the inner helices defines the ion gate (Figure 7B). The lowest constriction defines the opening to the SR lumen (Figure 7C), and is formed by the pore helices in a region that corresponds to the selectivity filter of the K+ channels. The inner helices, the ion gate, and the putative selectivity filter surround the central cavity (see Figures 3, 8B-c, and 8D-c).
In agreement with our previous 3D structure of RyR1 [7], the transmembrane assembly of both new 3D reconstructions reveals at least six distinct regions of high density per subunit that can be attributed to α helices (Figure 8). These rod-like structures have a density >3 σ above the mean and are clearly differentiated from their surroundings; they are identified as red contoured regions in Figure 8. There is a remarkable similarity between the arrangement of all six α helices of the mammalian voltage-dependent shaker channel Kv1.2 in the open state and the putative RyR1 transmembrane α helices in the same condition (see Results and Figure 9). For the purpose of comparison, we designate the putative α helices of RyR1 as R1–R6, where R6 is the inner helix. These are named according to the comparably positioned α helices of the K+ channel (S1–S6).
We compared the diameter of the ion gate of our open/closed RyR1-FKBP12 with that of previous 3D determinations of RyR1 in the closed state [7,17]. Furthermore, taking into account that the diameter of the K+ and Ca2+ ions is very similar, around 4 Å, we compared the diameter of the ion gate of RyR1 with that of the K+ channels and nAChR in open/closed conformations that have been determined at atomic resolution. To accomplish this for RyR1, we measured the diameter of the ion gate at a density threshold corresponding to the secondary structure (Figure 11). The diameter of the ion gate of our closed RyR1-FKBP12 and closed RyR1 [7] 3D reconstructions is 8 Å, whereas the diameter of the ion gate of the closed RyR1 obtained at similar resolution in the same conditions by another group is 15 Å [17]. The diameter we find here for the RyR1-FKBP12 ion gate in the open state is 12 Å. From the known atomic structures of K+ channels and nAChR, we took the equivalent measurement, defined by the inner edge of the inner helices, and find that in the closed state, their ion gate diameters range between 7–8 Å (closed K+ channels, which are tetramers) and 10 Å (closed nAChR, which is a pentamer). For all of the open channels, the diameter is 12–13 Å [29,33–35].
Taken into this general context, our measurement of 8 Å for RyR1′s closed ion gate falls within the values found for the closed conformations, and the measurement of 12 Å for RyR1′s open ion gate corresponds to that found for the open conformations of the known K+ channels (Figure 11). When the side chains of the K+ channel's atomic model are taken into consideration, the actual diameter of the closed pore is 4 Å [15]. Thus, it is likely that when atomic resolution of RyR1′s structure is obtained, our 8 Å diameter will result in similar pore dimensions, which is an appropriate conformation for a closed Ca2+ channel. Likewise, the observed increase to 12 Å diameter in the open state should be sufficient to enable Ca2+ flow.
When the atomic structure of open Kv1.2 [35] is superimposed on the open RyR1 density map, the positions of the α helices of Kv1.2, S1–S6, correlate well with high-density regions of RyR1 (Figure 9). Starting from the 4-fold axis, we assign S6, the four inner α helices of the K+ channel, to the four central rod-like structures (inner helices) in RyR1 (R6, see Figures 8D-b through 8D-d and 9B–9D). The tips of the pore helices in Kv1.2 also overlay those of RyR1 (p in Figures 8D-d and 9D). Four rod-like structures that are in the same position as the outer helices of the K+ channel (S5) can be identified in the region of RyR1′s transmembrane domain proximal to the lumen (R5) (see Figures 8B-d, 8D-d, and 9D). We suggest that they are the putative outer helices, or R5. The S1–S4 helices form the voltage sensor of Kv1.2. Although RyR1 does not have known voltage-sensing activity, we observe that S1–S4, which form the voltage sensor in Kv1.2, overlap with the corners of the transmembrane assembly of RyR1. Two densities in RyR1, R1 and R3, are in a similar configuration to S1 and S3, although slightly farther away from the 4-fold axis (Figures 8D-c, 8D-d, 9B, and 9D). R2, a weaker density, matches with S2, and the intervening density between R3 and R5, indicated as R4 in Figure 8D-d, could correspond to S4. At the level of the ion gate, R6 continues to overlap with S6, and the horizontal rod-like density 1 (h1) of RyR1 overlaps with the S4–S5 linker structure (Figures 7B, 8D-b, and 9B). The h2 structure coincides with an α helical structure of unknown sequence in the Kv1.2 atomic model [35] (see u in Figure 9A superimposed on the open RyR1 density map). Despite the structural similarity, we could not find sequence homology between the transmembrane segments of Kv1.2 and the aliphatic segments of RyR1. In contrast with Kv1.2, two other atomic models of K+ channels with six transmembrane α helices per subunit [29,34] do not match well with our cryoEM density map. The region of discordance in these atomic models is the S1–S4 formation; however, this could well be the result of the presence of the Fab/Fv fragments against the voltage sensors that were needed for crystallization.
A previous 30 Å resolution reconstruction of RyR1 was prepared in conditions designed to represent the open conformation (100 μM Ca2+, 100 nM ryanodine) [11]. This reconstruction indicated that in going from the closed to the open state, the protein undergoes several conformational changes: a counterclockwise rotation of the transmembrane domain with respect to the cytoplasmic assembly, an elongation of approximately 10 Å of the overall structure in the 4-fold axis direction, an opening of the clamp domains between domains 9 and 10, and an increase in pore diameter from 0 to approximately 18 Å. Two other 3D reconstructions of RyR1 at similar resolution were prepared to represent fully and transiently open states (100 μM Ca2+, and 100 μM Ca2+ plus 1 mM AMP-PCP, respectively) [12] also indicated an opening of the clamp domains between domains 9 and 10. In these cases, no elongation along the 4-fold axis was observed. Due to the limited resolution, the pore diameter was highly threshold-dependent and in a range between 0–7 Å diameter.
Many of these features are not compatible with our observations, and because of the low resolution of these reconstructions, some genuine structural differences were likely to have been confounded by effects resulting from low or anisotropic resolution. In our current reconstructions and a previous closed-state reconstruction [7] of RyR1, all at around 10 Å resolution, the only connection from domain 10 to the rest of the structure is domain 9, making it impossible for the clamp domains to “open” during gating by separating domains 10 and 9 [11,12]. Their observed gap is likely to be a consequence of the lower resolution and the threshold nonequivalence between the open state and closed states as it is known that the choice of threshold in low-resolution reconstructions dramatically affects the surface representations. Second, the elongation in the z direction that they observed, but was not observed in our reconstruction, is likely due to averaging of the central domains moving away from the transmembrane assembly and the peripheral domains moving toward it. In addition, the missing-cone artifact [25], whereby a large proportion of 4-fold views with respect to side views, could provoke an artifactual elongation along the 4-fold axis. Last, the diversity of dimensions for the open pore and the much lower resolution in the previous open-state 3D reconstructions [11,12] do not warrant a comparison of pore dimensions.
In trying to elucidate the molecular mechanism for ion gating using cryoEM, we have found that upon channel opening, structural changes in the cytoplasmic domains are coordinated with structural changes in the ion gate. All domains appear to move in an orchestrated manner, resulting in a significant lowering of density along the 4-fold axis of the protein and an increase of the ion gate diameter. The most obvious connection that we can see between changes at the cytoplasmic domains and ion gate opening is how the upward and outwards movement of the cytoplasmic domains pulls the inner branches in that same direction. Because the inner branches are directly connected to the ion gate, it is straightforward to see how their being pulled apart increases the diameter of the ion gate.
RyR1′s large cytoplasmic domain interacts with several proteins such as the voltage sensor (DHPR), FKBP12, CaM, and IpTxa, and all four affect RyR1′s gating. In intact skeletal muscle, RyR1 appears to open exclusively under the control of the DHPR. Removal of FKBP12 or addition of IpTxa is known to induce subconductance states, whereas CaM modulates the Ca2+ dependence of RyR1′s probability to open. The binding sites for FKBP12, CaM, and IpTxa have been mapped by cryoEM and 3D difference mapping [8–10,49,50], and in all cases, they bind at least 130 Å away from the ion gate (positions of FKBP12 and Ca2+-CaM binding sites are indicated in Figure 4A). We suggest that the conformational changes associated with gating that we have found here are very likely to be the same as the long-range allosteric pathways that convert remote signals sensed through protein/peptide/small molecule–protein interaction in RyR1′s cytoplasmic domains into the appropriate response (e.g., the probability of RyR1′s ion gate to open).
By superimposing the open/closed 3D reconstructions, one can observe regions of density displacement near regions that remain almost stationary. This indicates the presence of structural hinges, i.e., boundaries between regions of RyR1 that move with different breadths. The two more noticeable regions where this takes place are the crevice near domain 4, and the one between domains 5, 9, and 3 (Figure 4A). Interestingly, these hinges correspond to previously mapped binding sites. The crevice near domain 4 is the target for IpTxa and Ca2+-CaM [10,50]. Likewise, the intersection between domains 5, 9, and 3, constitutes the FKBP12 binding site [9,10]. Thus, it appears that the hinges may constitute regulatory sites where binding of a relative small effector could produce optimal effect.
It has been previously reported that the dimension of the closed ion gate in a 9.6 Å reconstruction of RyR1 is 15 Å [17]. This is surprising because it is almost twice the size of the pore we have found in our 10.2 and 10.3 Å resolution reconstructions of RyR1 in the closed state and 20% larger than the open ion gate reported here (Figure 11). A pore of 15 Å would leave a large gap that, based on the dimensions of open ion gates for other known cation channels, should not be impermeable to Ca2+ ions. The inner branches were not observed in this 9.6 Å 3D reconstruction, and the density in several portions of their putative inner helices is discontinuous (Figure 5B), which raises the possibility that this reconstruction was obtained from a preparation that contained a mixture of open and closed conformations. Such heterogeneity would give a low signal-to-noise ratio in those parts of the structure that change conformation during ion gating. The presence of a low signal-to-noise ratio in their reconstruction required the assistance of helix hunter [51] to identify the putative helices rather than being able to see them directly by increasing the threshold as was done here. There are several portions of their putative inner helices that do not overlap with either our closed- or open-state 3D maps (compare Figure 5B with Figures 3B, 3D, and 5A). Ludtke et al. interpreted their results as meaning that the inner helices of RyR1 in the closed state are more similar to an open than a closed K+ channel conformation. This contradicts our report here in which cross-correlation measurements between our open and closed states and all K+ channels indicated a direct equivalence of physiological state and inner helix conformation (Figure 6). Finally, the fact that we provide three independent 3D reconstructions supports further the ion gate dimensions and inner helix conformation of open/closed RyR1, and that they are in a similar range of these reported for the K+ channels.
Based on our results, we propose that three structures, the inner branches, the inner helices, and h1 densities, by forming a mobile axial structure, are the three main gating effectors. In the closed state, the two right-handed bundles (inner helices, inner branches) form the high-density constriction (ion gate) at their meeting point. In going from the closed to the open state, both sets of bundles relax and appear to contribute equally to lowering the density of the ion gate (see arrows in Figure 3B). The h1 densities also contribute to the constricting effect in the closed state and move outwards as the gate opens. The resulting profile of the pore along the ion pathway looks dramatically different as it transitions from the closed to the open state. Such a three-way mechanism appears to constitute a very efficient mechanism to open and close the ion gate and is compatible with the complex regulation of RyR1 through its interaction with the DHPR and other exogenous or intracellular modulators [52].
In summary, we have obtained the 3D reconstructions of the hydrated RyR1-FKBP12 complex both in open and closed conformations. The use of neuroactive PCB 95 [18,53] to favor the stability of the full open conformation of the RyR1 channel enabled 3D reconstructions of the ion pathway with high detail. The conformational change of the peripheral cytoplasmic domains is directly related to conformational changes in the transmembrane domain. The architecture of the RyR1 appears to be designed to support precise long-range allosteric pathways such as these involved in efficient coupling with the voltage sensor and in the modulation by ligands such as FKBP12 and CaM. Finally, we have shown that there is a striking similarity between the architectural organization of the transmembrane α helices of the K+ channel family and those of RyR1. Beyond this similarity, we find that the inner branches, a structure that connects the cytoplasmic domains of RyR1 to the ion gate, appear to play a direct role in ion gating.
[3H]Ryanodine ([3H]Ry; 60–90 Ci/mmol; >99% pure) was purchased from Perkin-Elmer New England Nuclear. PCB 95 (>99% pure) was purchased from Ultra Scientific. All other reagents were of the highest purity commercially available.
RyR1 was purified from rabbit skeletal muscle to concentrations of 2 mg/ml as previously described [7]. Prior to freezing, all RyR1s were incubated with FKBP12 (Sigma) at a molar ratio of 8× for 20–40 min. Final buffer conditions to lock the RyR1 into the closed state were 20 mM Na-MOPS (pH 7.4), 0.9 M NaCl, 0.5% CHAPS, 2 mM DTT, 2 mM EGTA, 5 μg/ml aprotinin, 5 μg/ml leupeptin, and 2.5 μg/ml Pefabloc. To set RyR1 in the open state, the same buffer was used except that 10 μM PCB 95 and 50 μM Ca2+ were added and EGTA was excluded.
Bilayers were made of phosphatidylethanolamine: phosphatidylserine: phosphatidylcholine (5:3:2 w/w, Avanti Polar Lipids) dissolved in decane at a final concentration of 30 mg/ml. The bilayer partitioned two chambers (cis and trans) containing buffer solution (in mM) 500 CsCl, 50 μM Ca2+, and 20 Hepes-Tris (pH 7.4) on cis and 50 CsCl, 7 μM Ca2+, and 20 Hepes-Tris (pH 7.4) on trans. The addition of protein was made to the cis solution that was held at the virtual ground, whereas the trans solution was connected to the head stage input of an amplifier (Bilayer Clamp BC 525C; Warner Instruments). Purified RyR1-FKBP12 complexes preincubated for 20–40 min were introduced to cis solution. Upon the incorporation of a single RyR1 channel into BLM, the cis chamber was perfused with cis solution to prevent additional channel incorporation. Single-channel gating was monitored and recorded at a holding potential of −40 mV (applied to the trans side). The sidedness (cytosolic) of the channel was verified by the positive response to addition of micromolar Ca2+ and response to 2 μM ryanodine and 5 μM Ruthenium Red (at the end of the experiment). The amplified current signals, filtered at 1 kHz (Low-Pass Bessel Filter 8 Pole; Warner Instrument,) were digitized and acquired at a sampling rate of 10 kHz (Digidata 1320A; Axon-Molecular Devices). All the recordings were made for a duration between 12 s and 6 min under each experimental condition. The channel open probability (Po) was calculated using Clampfit, pClamp software 9.0 (Axon-Molecular Devices) without further filtration.
Equilibrium measurements of specific high-affinity [3H]Ry binding were determined as previously indicated [20,54]. Junctional SR vesicles of rabbit skeletal muscle (50 μg protein/ml) were incubated with or without PCB 95 in buffer containing 20 mM HEPES (pH 7.4), 250 mM KCl, 15 mM NaCl, defined concentration of CaCl2, and 2 nM [3H]Ry for 3 h at 37 °C. The reactions were quenched by filtration through GF/B glass fiber filters and washed twice with ice-cold harvest buffer (20 mM Tris-HCl, or 20 mM Hepes, 250 mM KCl, 15 mM NaCl, 50 μM CaCl2 [pH7.4]). Nonspecific binding was determined by incubating JSR vesicles with 1,000-fold excess unlabeled ryanodine. Each of the conditions was replicated four times in two separate junctional SR preparations, and each of the readings was performed in triplicate or quadruplicate.
A 5-μl aliquot of the 2–4 mg/ml RyR1-FKBP12 complex incubation mixture was adsorbed onto a glow-discharged quantifoil holey grid and the excess blotted off with Whatman 540 filter paper. The sample was vitrified by plunging the grid into liquid ethane. CryoEM was performed on a FEI Tecnai F20 FEG microscope operated at 200 kV. Untilted images with defoci between 2.5 and 4.0 μm were recorded on Kodak SO-163 film under standard low-dose conditions (dose <10 e−/Å2) at a nominal magnification of 50,000×.
A total of 257 and 233 micrographs for the closed and open states, respectively, were digitized on a Zeiss SCAI scanner at a step size of 7 μm, and subsequently binned down to a final pixel size of 2.8 Å. A total of 15,625 and 18,527 particles for the closed and open states, respectively, were selected interactively using the program WEB. The defocus parameters were determined for every particle using CTFTILT [55]. Individual particles were subjected to a reference-based algorithm starting from an initial 3D model of RyR1 [7] filtered to 40 Å resolution where no substructure is detectable, thus avoiding model bias. Fifty percent of the particles from each dataset with the lowest cross-correlation with the 3D model were discarded. This was followed by several iterations of refinement until the shifts and rotations stabilized. The final number of particles was 9,331 and 8,133 particles for the closed and open states, respectively. Reference alignment and 3D reconstruction enforcing 4-fold symmetry were performed using the program FREALIGN [56], which takes account of phase and amplitude contrast transfer function (CTF) correction for every particle. This program has implemented a weighting scheme to correct for noise bias, an artifact that could result in an artificial overestimation of the resolution [57]. Resolution values were calculated according to the Fourier shell correlation (FSC) curve between two half datasets. The 0.143 cutoff [26] was chosen because it was optimistic with respect to the 5 σ noise correction calculated taking into account the 4-fold symmetry (and thus data redundancy) of the RyR1. The final 3D structure of RyR-FKBP12 was normalized and filtered to a resolution of 10.2 Å using a B factor of −300 Å3. The mean and standard deviation values of the volume were calculated within a spherical mask of the same diameter as that used in the iterative alignments. For 3D difference mapping, both 3D reconstructions were filtered to 18 Å resolution and normalized by adjusting the average and standard deviation of densities in both reconstructions to the same level as previously done [9]. Then the open-state RyR1 3D reconstruction was directly subtracted from the closed-state RyR1 3D reconstruction and vice versa. No further data manipulation such as postsubtraction filtering or masking was performed. SPIDER software [58] was used for preparation of the initial volumes, normalization, 3D difference mapping, filtration of the Protein Data Bank (pdb) files for comparison with the cryoEM density maps, and calculation of cross-correlation values. Image rendering, docking of atomic structures, and alignment of the other RyR1 3D reconstruction from the database were performed in Chimera [59] (http://www.ebi.ac.uk/pdbe/emdb/). Both closed-state RyR1 3D reconstructions that have been previously published [7,17] are available in the Electron Microscopy Database (http://www.ebi.ac.uk/msd-srv/docs/emdb/).
Hydropathicity, transmembrane propensity, and α helical prediction analyses were performed using several packages available on public servers. The different packages for α helical prediction provided reasonable overlapping sequence segments. The proposed secondary structure is based on the PSIPRED prediction method [60] (http://bioinf.cs.ucl.ac.uk/psipred/).
Electron Microscopy Data Bank (http://www.ebi.ac.uk/pdbe/emdb/) accession numbers for the structures of the RyR1 in closed and open conformations are 1606 and 1607 respectively.
|
10.1371/journal.ppat.1003184 | Increased Functional Stability and Homogeneity of Viral Envelope Spikes through Directed Evolution | The functional HIV-1 envelope glycoprotein (Env) trimer, the target of anti-HIV-1 neutralizing antibodies (Abs), is innately labile and coexists with non-native forms of Env. This lability and heterogeneity in Env has been associated with its tendency to elicit non-neutralizing Abs. Here, we use directed evolution to overcome instability and heterogeneity of a primary Env spike. HIV-1 virions were subjected to iterative cycles of destabilization followed by replication to select for Envs with enhanced stability. Two separate pools of stable Env variants with distinct sequence changes were selected using this method. Clones isolated from these viral pools could withstand heat, denaturants and other destabilizing conditions. Seven mutations in Env were associated with increased trimer stability, primarily in the heptad repeat regions of gp41, but also in V1 of gp120. Combining the seven mutations generated a variant Env with superior homogeneity and stability. This variant spike moreover showed resistance to proteolysis and to dissociation by detergent. Heterogeneity within the functional population of hyper-stable Envs was also reduced, as evidenced by a relative decrease in a proportion of virus that is resistant to the neutralizing Ab, PG9. The latter result may reflect a change in glycans on the stabilized Envs. The stabilizing mutations also increased the proportion of secreted gp140 existing in a trimeric conformation. Finally, several Env-stabilizing substitutions could stabilize Env spikes from HIV-1 clades A, B and C. Spike stabilizing mutations may be useful in the development of Env immunogens that stably retain native, trimeric structure.
| A vaccine is needed to prevent HIV/AIDS but eliciting potent neutralizing antibodies (Abs) against primary isolates has been a major stumbling block. The target of HIV-1 neutralizing antibodies is the native envelope glycoprotein (Env) trimer that is displayed on the surface of the virus. Virion associated Env typically elicits antibodies that cannot neutralize primary viruses. However, because native Env trimers can dissociate and coexist with non-fusogenic forms of Env interpreting these results are difficult. Here, we used directed evolution to select for virions that display native Env with increased stability and homogeneity. HIV-1 virions were subjected to increasingly harsh treatments that destabilize Env trimers, and the variants that survived each treatment were expanded. We could identify seven different mutations in Env that increased its stability of function in the face of multiple destabilizing treatments. When these mutations were combined, the resulting mutant Env trimers were far more stable than the original Env protein. Incorporating trimer-stabilizing mutations into Env-based immunogens should facilitate vaccine research by mitigating the confounding effects of non-native byproducts of Env decay. A similar approach may be used on other pathogens with potential vaccine targets that are difficult to isolate and maintain in a native form.
| For an HIV/AIDS vaccine to be effective, it is widely thought that it should elicit high titers of broadly neutralizing antibody (Ab) [1], [2]. HIV-1 neutralizing Abs target the envelope glycoprotein (Env) spike, which is a trimer containing three copies each of the surface subunit, gp120, and the transmembrane subunit, gp41 [3]. A major confounding issue in the rational development of Env as a vaccine is that fusion-competent Env trimers are often labile and heterogeneous, so distinguishing fusogenic from other forms of Env can be challenging [4]–[8]. Non-native forms of Env include dissociated gp120 monomers and dimers, gp41 stumps, monomers and oligomers of unprocessed gp160, as well as Env with aberrant disulfides and heterogeneous glycosylation [6], [7], [9]–[11]. In particular, non-native forms of Env may serve as immune decoys and elicit non-neutralizing Abs [6], [12]–[14]. Envs that are truncated prior to the gp41 transmembrane (TM) domain have in some cases been engineered as trimers, but these are not in a native conformation as, unlike native Env, they are typically recognized by non-neutralizing Abs and also elicit non-neutralizing Abs after immunization [15]–[20]. Thus, limiting exposure to the immune system of non-fusogenic forms of Env through stabilization of the native structure may facilitate HIV-1 vaccine design.
HIV-1 Env spikes are held together by non-covalent interactions among its subunits. Mutations that accelerate spontaneous or CD4 receptor-induced dissociation of gp120 from the HIV-1 Env complex can be found in various regions including the N-heptad repeat (NHR) [21], the disulfide loop (DSL) [22] and C-heptad repeat (CHR) regions [21], [23] of gp41, as well as in the C1 [24], V3 [25], β3–β5 loop of C2 [26], and C5 [27] regions of gp120. This may be expected on chance, as random mutations are much more likely to disrupt than stabilize the structure-function of a protein. Indeed, mutations that would stabilize Env trimers in the active membrane-anchored form have not been forthcoming or even reportedly sought after. One potential solution has been the introduction of a disulfide-bond between gp120 C5 and the DSL of gp41 (e.g. 501C and 605C; known as “SOS”), which, when exposed to a reducing agent, breaks and allows for productive entry of SOS-modified HIV-1 into target cells [28], [29]. However, the disulfide bond is subject to exchange and can also leave many non-neutralizing epitopes exposed, at least in soluble forms of the SOS molecule [30], [31]. Thus, we envisioned an alternative strategy that allows the virus to select for mutations that stabilize the Env trimer naturally, without compromising native structure or antigenicity.
We have shown previously that, depending on the viral isolate, virion associated Env can have different levels of heterogeneity and can have a range in stabilities to conditions such as elevated temperature, prolonged incubation at 37°C or exposure to denaturants [6], [7]. Env from the clade B isolate ADA is both labile and heterogeneous [7]. An ADA variant, AD8, has recently been associated with rapid elicitation of broadly neutralizing Abs in a macaque model and so is relevant to vaccine research [32]. Heterogeneous and labile Envs can be problematic to study because non-fusogenic forms of Env can accumulate and confound measurements [7]. Diversity in glycosylation may account for some of the observed heterogeneity, even in functional Env, as certain cloned isolates of HIV-1 are incompletely neutralized by glycan-sensitive Abs such as PG9 and PG16 [10].
Directed evolution is often used in virology to study cellular tropism or resistance to neutralizing antibodies or antiviral drugs. HIV-1 has been selected to resist spontaneous inactivation of Env at cold temperatures or to overcome functional defects associated with truncated Envs [33], [34]. Increasing the stability and homogeneity of native Env might also improve Env-based vaccines by limiting potentially distracting non-neutralizing and immunogenic surfaces of Env, improve correlations between observed Env structures and their associated functions, as well as inform the design of more molecularly defined immunogens [4]–[6]. We therefore devised a strategy in which Env is selected specifically for increased stability in its unliganded, functional state. Thus, HIV-1 is subjected to iterative cycles of harsh destabilizing conditions and subsequent viral expansion. We show that this approach can overcome as well as help understand the molecular heterogeneity and lability of viral spikes, which may have implications for the design of immunogens based on functional, unliganded and membrane-anchored Env of multiple clades of HIV-1.
We previously showed that Env from the clade B, R5 isolate ADA was relatively heterogeneous as well as labile to heat, guanidinium hydrochloride (GuHCl) and to spontaneous inactivation at physiological temperature [7]. We generated mutant pools of HIV-1 ADA using site-directed mutagenesis targeted to regions of Env shown previously to affect Env trimer stability (Figure S1A). However, random mutant clones were found to be mostly non-infectious and when library DNAs were transfected into 293T cells, the virus produced was of very low titer, most likely because of the high number of mutations targeted to conserved regions of Env (Figure S1B and C). Nevertheless, extensive passaging in MT2-CCR5ΔCT cells yielded high titer virus with each pool (Figure S1D).
To select for stable ADA variants, the pools of virions were treated separately with incremental concentrations of GuHCl, urea, hyper-physiological temperatures or prolonged incubations at physiological temperature that inactivates wild-type ADA [7]. After destabilizing treatment, the surviving infectious viruses were rescued on MT2-CCR5ΔCT cells. Following three rounds of selection, some virion pools were more stable than ADA wild-type to GuHCl and heat treatment (e.g. B21 and B22), one was resistant to heat (e.g. C11), and two pools were resistant to 37°C decay (e.g. C12 and M2; Figure 1). None of the selected library pools were found to have increased stability to urea.
To identify individual Env mutants of increased stability, viral RNA was purified from the stability-enriched pools and env was amplified using RT-PCR. Only the ectodomain portion of Env was subcloned back into the pLAI display vector in order to rule out mutations in the gp41 TM and cytoplasmic tail (CT) domains that might affect interactions below the viral membrane, such as between the gp41 CT and Gag [35], and would add further complexity to the analysis. Individual Env clones were picked from each pool and the corresponding virions were assayed for resistance to each of the selection conditions (Figure 2). A variety of stability phenotypes were observed. Notably, some clones from the GuHCl and heat-treated pools fully recapitulated the stability phenotype of the originating viral populations (Figure 2A, B, and C).
For virion pools that showed resistance to decay at 37°C (i.e. PM and PC1), none of the cloned Envs approached the stability of the corresponding pool (Figure 2D and data not shown). We therefore took an alternative limiting dilution approach to identify stable mutants. Using MT2-CCR5ΔCT cells as target cells, six limiting dilution wells for each pool were found to be equal in stability to the parental pools (Figure 2E and F). In summary, we successfully identified HIV-1 mutant clones that were stable to the same conditions as the pools from which they were derived.
To understand the basis of the hyperstable phenotypes of the rescued mutant clones, we determined the primary Env sequences. Mutations were identified, remarkably, only in regions of ADA Env not targeted by our mutagenesis procedure, most likely because the targeted mutations negatively impacted infectivity and any mutants were out-competed by the small number of virions incorporating mostly wild-type Env sequence. From the heat-stable library pool HC11, all of the clones contained the substitution S649A in the CHR region of gp41, whereas some but not all clones contained two substitutions in gp120 C5 (S462G and D474N) and an alteration to gp120 V1 (i.e. deletion of N139/I140 plus an N142S substitution that will hereby be referred to as “V1alt”). These mutations in the HC11 clones appeared to arise de novo. (Figure 3A).
Sequencing of Env from the GuHCl-stable and heat-stable library pools, G and HB2, surprisingly revealed mutations to residues in common with the LAI strain that was used in engineering the display vector, pLAI (Figure 3B and 3C). In these clones, recombination events involving a very small amount of LAI DNA appear to have occurred during the PCR used in library production. Importantly, all of the selected clones were more stable than either the parental strain, ADA, or the LAI strain (Figure 3C). Multiple recombination events appear to have taken place during the selection process resulting in virions containing different amounts of LAI-derived gp120 sequence, and among the most stable clones, LAI-derived sequence from gp41, but not from gp120, was associated with improved Env stability. Thus, the presence of LAI gp120 was significantly inversely correlated with Env stability (Figure 4) and the most stable clone (GB21-6) contained full LAI gp41 and ADA gp120. In total, 9 amino acid residues in the gp41 ectodomain differ between ADA and LAI (Figure 3). In addition, a tenth mutation, K574R, arose de novo and was conserved amongst the stable clones.
The stability of the selected Env clones can either be specific to the selection conditions, or can impart a broader resistance to multiple destabilizing treatments. To investigate, the most stable clones from the GuHCl and heat-selected pools, GB21-6 and HC11-1, were subjected in parallel to GuHCl, heat and prolonged incubation at physiological temperature. Both GB21-6 and HC11-1 maintained infectivity with increased resistance to each of these conditions relative to wild-type ADA (Figure 5). Clone GB21-6 was consistently the most stable in each case. In keeping with our prior observation of a negative correlation between the number of LAI gp120 residues and Env stability (Figure 4), LAI was less stable than GB21-6 when treated with heat or GuHCl, but notably the two viruses displayed similar rates of decay at physiological temperature (Figure 5).
To determine the relationship between functional stability and oligomeric stability of the unliganded mutant Env trimers, we turned to BN-PAGE. Two GuHCl resistant and two heat resistant clones were chosen for the analysis: GB21-6, GB22-1, HB22-5 and HC11-4, the latter of which shares similar stability and most of the same mutations as HC11-1 (Figure 3). The clones were subjected to increasing temperature or GuHCl concentrations, and samples were analyzed for infectivity and by BN-PAGE. In each case, dissociation of the Env trimer on BN-PAGE closely correlated with the loss of infectivity, with stability selected Envs clearly maintaining trimeric association under conditions that caused dissociation of wild-type ADA trimers (Figure 6). HC11-1, used above and in subsequent analyses, was also found to be more stable (data not shown). Importantly, the stable Env variants also appear more homogeneous than that of wild-type ADA as non-trimeric species in the former are much less apparent than in the latter.
Certain neutralizing monoclonal Abs (mAbs) and other inhibitory ligands, such as soluble CD4 (sCD4), can destabilize and irreversibly inactivate the Env trimer upon binding [36]–[38]. We used two approaches to determine whether the stabilized clones of HIV-1 Env would resist ligand destabilization. In the first approach, we employed a modified virus capture assay (VCA) that was designed to crudely mimic Env destabilization through host receptor engagement. Thus, GB21-6 or HC11-1 virions were incubated in solution with increasing concentrations of sCD4 for 15 min, and then overlaid on microwells coated with either mAb hNM01 (anti-V3) or X5 (anti-coreceptor binding site (CoRbs)). Unbound virus was then washed away and TZM-bl target cells were overlaid to measure remaining infectivity. Binding of sCD4 to the Env trimer initially exposes the V3 loop and CoRbs in a fusion-active state [39], but after a short period the sCD4-bound trimer decays to an inactive state [36]. The VCA design provides an aggregate measure of induction of conformational changes by sCD4 and the functional stability of the sCD4-activated state. As expected, capture of infectious wild-type ADA was increased by low concentrations of sCD4 that promotes exposure of the epitopes of the capture mAbs, but decreased at higher concentrations as virions were inactivated (Figure 7). By contrast, when using the same initial concentrations of sCD4 capture efficiency of infectious virus was increased for both GB21-6 and HC11-1, and, at high concentrations of sCD4 that inactivated wild-type ADA, the infectivities of GB21-6 and HC11-1 were still intact. We note that in the absence of sCD4 mAbs X5 and hNM01 captured lower levels of GB21-6 and HC11-1 relative to wild-type ADA (Figure 7). This may be related to the BN-PAGE results showing that these viruses display more homogeneous trimers (Figure 6), as mAb X5 does not appear to bind the unliganded trimers of most primary isolates and likely captures virions via non-native Env [6].
If an inhibitor binds and inactivates Env, the IC50 of that inhibitor is expected to decrease over time [37]. In a second approach to assay ligand-induced Env destabilization, GB21-6 and HC11-1 viruses were pre-incubated with mAbs or inhibitors for two different periods of time prior to measuring viral infectivity on target cells. Of the inhibitors we tested, mAbs b12 (anti-CD4 binding site; CD4bs), 4E10 and 2F5 (anti-membrane proximal external region; MPER), as well as sCD4 have all been shown to destabilize Env trimers [36], [37]; mAb VRC01 (anti-CD4bs) has a much weaker destabilizing effect [40]; mAb PG9 (anti-V2/V3) has not been well studied in this context [41]; C34 (anti-NHR) should have no activity towards unliganded Env as it only binds Env post-CD4 engagement [42]; and PF-348089 is an analogue of BMS-378806 that binds to gp120 and prevents CD4-induced conformational changes [43], [44]. Consistent with the VCA results above, both GB21-6 and HC11-1 resisted inactivation by sCD4; similar resistance was also observed using b12, 4E10 and 2F5 (Figure 8A; Table S1). Thus, whereas the IC50s of these inhibitors against wild-type ADA decreased 9–15-fold from the first (1 h) to the second (20 h) pre-incubation time point, the IC50 decreases with GB21-6 and HC11-1 were only 3–6-fold and 4–8-fold, respectively. Notably, VRC01 affected all three viruses equivalently. PG9 inactivated both wild-type ADA and clone GB21-6 with IC50 decreases of ∼10-fold between pre-incubation times. In contrast, HC11-1 resisted PG9 inactivation, but this clone contains an alteration in V1 that may directly affect the PG9 epitope. Finally and as expected, C34 and PF-348089 did not inactivate any of the viruses over time.
The above inhibition data can also be expressed as a ratio of IC50s of the mutants to that of wild-type ADA using the standard 1 h pre-incubation time (Figure 8B and Table S1). Expressed in this way, GB21-6 is 12–30-fold more resistant than wild-type virus to sCD4, b12, 4E10 and 2F5, while HC11-1 is also resistant to these inhibitors, but to a lesser extent. The same two mutants are not generally more resistant to VRC01, PG9, and C34, with some exceptions that are most likely due to sequence changes that directly affect ligand binding (Figure 3). Hence, the stable clones tend to be more resistant to ligand-induced inactivation, except when the ligand is not of the destabilizing type (e.g. VRC01 and C34) or the inhibitor epitope is affected. Interestingly, both GB21-6 and HC11-1 were 5–10-fold more sensitive to PF-348089, an inhibitor that prevents CD4-induced conformational changes presumably by stabilizing the CD4-unbound state.
It has been shown that mAbs PG9 and PG16 can neutralize ∼92% of HIV-1 isolates in a large multi-clade viral panel but can occasionally give rise to inhibition curves that plateau below 100% neutralization for certain sensitive isolates [10], [41]. The latter phenomenon is thought to be caused by heterogeneity in glycosylation on the Env trimer [10]. We observed such a phenomenon with wild-type ADA in which PG9 and PG16 exhibited a maximal percent inhibition (MPI) of 74% and 90%, respectively (Figure 9). Interestingly, both GB21-6 and HC11-1 exhibited higher MPIs against both mAbs; PG9 and PG16 had 90% and 96% MPI, respectively. Thus, not only do these two mutants of ADA exhibit decreased levels of heterogeneity and “cleaner” trimer bands on BN-PAGE, but they also show less heterogeneity within the pool of functional trimers as probed by mAbs PG9/16.
Based on the stable mutant Env sequences we selected single amino acid residue changes to introduce into wild-type ADA and examined their effect on Env stability. H625N and T626M were introduced as a double substitution, as these residues were adjacent to one another and seemed to co-vary. Although none of the point mutants completely recapitulated the phenotype of the stable clones, stabilizing effects were clearly observed and could be narrowed down to a few residues in each case (Figure 10). Thus, from the B2 pools, I535M, L543Q, and K574R in the NHR and H625N/T626M in the CHR each partially stabilized ADA Env to both heat and GuHCl treatment. In the case of the HC11 clones, the CHR mutation S649A played the largest role in stabilization and the V1alt substitution provided a more limited increase in Env stability.
Among the gp41 amino acid changes identified in the GB2 and HB2 virion pools that increased functional trimer stability, K574R was the only mutation that was not of LAI origin. The conserved Lys at position 574 residue has previously been shown to be crucial for stability of the six-helix bindle (6HB) protein that is a mimetic of gp41 in a post-fusion form [45]. Other studies have shown that mutations to the NHR of gp41 can affect neutralization sensitivity of HIV-1 [46], [47], and as shown above Env trimer stability can also alter sensitivity to certain inhibitors. To further characterize the relationship between Env stability and neutralization sensitivity due to mutation in the NHR, we examined how non-conservative mutation at position 574 affects trimer stability. We first performed this analysis using the mutant K574A in the LAI strain, which was generated in a previous study [48]. The K574A mutation profoundly decreased the T90 (the temperature at which viral infectivity is diminished by 90% in one hour) of HIV-1 LAI by ∼4°C and also globally increased sensitivity to a number of Env-destabilizing ligands (e.g. b12, sCD4, 2F5, 4E10, and b6; Figure 11A and Table 1). The most profound effect was observed with the weakly neutralizing CD4bs mAb b6 which was 250-fold more potent against K574A than wild-type LAI. In contrast, the substitution K574A had a less pronounced effect on neutralization by the non-destabilizing mAb 2G12 [37], and was only 2-fold more sensitive to mAbs and inhibitors that target Env in a pre-fusion intermediate state.
We also introduced the K574A mutation into the relatively stable primary isolate, JR-FL. Again, K574A destabilized JR-FL (i.e. T90 decreased by ∼3°C) and also resulted in broader sensitivity to a variety of destabilizing inhibitors (Figure 11B; Table 1). In particular, K574A made JR-FL ∼50-fold more sensitive to b6. Overall, our results suggest that residue K574 also plays a crucial role in regulating stability of the receptor-naive Env trimer.
To see if the stabilizing substitutions we identified could stabilize other viral Envs, we first targeted the relatively stable and homogeneously trimeric Env, JR-FL. L543Q, K574R, S649A, and V1alt were introduced, and, since JR-FL already contained M535 and N625/M626, the reverse mutations M535I and NM625/6HT were introduced to see if they would destabilize the trimer. When the mutants were tested for stability in the heat gradient assay, L543Q, K574R, and S649A all increased the T90 of JR-FL by 0.7–1.5°C, while M535I and NM625/6HT both decreased the T90 of JR-FL by ∼2°C (Table 2). V1alt did not have any effect on JR-FL stability, as might be expected due to the extreme sequence variability in this region. Thus, substitutions to these amino acid residues in the heterologous isolate JR-FL have similar effects on Env stability as in ADA.
Next, we introduced stabilizing substitutions into isolates from multiple clades that have previously been shown to be labile, including Q769.b9 (clade A), RHPA4259 (clade B) and ZM109F (clade C) [7]. Substitutions I535M, K574R and S649A were introduced into all three strains; L543Q was only introduced into RHPA4259, as Q769.b9 and ZM109F already contained Q543; all three isolates already contained N625/M626. All mutations inserted into the clade B isolate RHPA4259 increased its T90 by 0.8–1.3°C (Table 2). Substitutions in the non-clade B isolates showed mixed effects. Thus, K574R increased the T90 of both isolates, S649A increased the T90 of Q769.b9 but not ZM109F, and I535M did not affect the stability of either strain. Thus, the stabilizing mutations identified in this study appear to have a similar effect on all clade B isolates tested, and some of the substitutions (i.e. K574R and S649A) impart a stabilizing phenotype on Envs across the three major clades of HIV-1.
In an attempt to reconstitute the phenotypes of the most stable variants, we combined the stabilizing mutations identified from the GB2 and HC11 viral pools to produce Gmut (I535M, L543Q, K574R and H625N/T626M) and Hmut (S649A and V1alt). We also generated comb-mut, which combines both sets of consensus mutations from the two unrelated stable clones. When challenged with GuHCl, both Gmut and Hmut clearly recapitulated the phenotype of the clones from which they were derived. Notably, comb-mut was even more stable than either variant, being resistant to GuHCl at 2 M (Figure 12A). Similar results were seen with heat treatment, although in this case comb-mut was only slightly more stable than the others. Similarly, following prolonged incubation at 37°C, all of the stabilized mutants showed much improved half-lives compared to wild-type ADA, but no significant differences could be seen between the mutants. Loss of infectivity due to heat or 37°C incubation involves multiple viral components and we have seen that functional Env stability beyond a T90 of ∼50°C or a half-life of ∼20 h at 37°C cannot be quantified under the conditions of the assay [7].
We further tested the functional stability of comb-mut Env using destabilizing ligands. ADA wild-type or comb-mut virus was pre-incubated with the same mAbs and inhibitors used in Figure 8 for two different time periods. The inhibition data was again plotted both as the ratio of IC50s after a one hour incubation to that of a 20 hour incubation, as well as the ratio of IC50s of comb-mut to that of wild-type ADA using just the one hour incubation time. Comb-mut resisted destabilization by sCD4, b12, 4E10, and 2F5, as seen previously with GB21-6, and likewise was more sensitive than wild-type ADA to PF-348089 (Figure 12B and C). However, comb-mut was even more resistant to inhibition by sCD4 than GB21-6, exhibiting an 80-fold increase in IC50 relative to wild-type virus. In order to verify that resistance to destabilizing mAbs is not due to changes in binding site accessibility or integrity, we measured binding of a panel of mAbs to virion-displayed Env in a simplified virus ELISA format [11]. From this panel, we found that all of the broadly neutralizing mAbs to gp120 bound at least somewhat more strongly to comb-mut virus than to ADA (Table S2). In particular, neutralizing mAbs that bind the outer surface of gp120 (e.g. PGT128, PG9, PG16, and 2G12) bound ∼10–50-fold more strongly to comb-mut than to ADA, while mAbs against the more recessed CD4bs bound ∼4-fold better. Binding of neutralizing mAbs to gp41 was equivalent between comb-mut and wild-type ADA. In contrast, all of the non-neutralizing mAbs tested in this assay exhibited somewhat reduced binding affinity to comb-mut virus relative to ADA. In particular, mAb 7B2 against the immunodominant disulfide loop region of gp41 showed 20-fold lower binding to comb-mut. These results show that functional comb-mut Env trimers are resistant to destabilization by various inhibitors and the binding sites of these inhibitors appear to be intact on virions, while the epitopes of non-neutralizing mAbs appear to be diminished on virions, although not eliminated.
We further examined the direct effect of heat on Env trimer dissociation by comparing wild-type ADA and comb-mut using BN-PAGE. As expected, the Env trimer of wild-type ADA dissociated on heat treatment at a temperature slightly above the T90 of ADA (Figure 13A and B) [7]. In contrast, the Env trimer of comb-mut was much more resistant to this treatment and did not significantly dissociate until an incubation temperature of 63.6°C. The observed increase in oligomeric stability of mutant Env trimers might be due at least in part to an increase in the level of uncleaved gp160 incorporated into the virus. We therefore analyzed the relative levels of cleaved gp120 and uncleaved gp160 associated with each virus by reducing SDS-PAGE. We observed no effect on cleavage as a result of the stabilizing mutations, as all Env variants appear to be ∼95% cleaved (Figure S2).
Immunization with virus particles typically does not elicit neutralizing Abs to the autologous virus, suggesting that native Env might be degraded rapidly in vivo. In addition to spontaneous dissociation, a possible cause of Env degradation is proteolysis. To explore their protease sensitivity, we treated ADA and comb-mut virions with a cocktail of trypsin, chymotrypsin, and proteinase K and then measured virus infectivity over a time course at physiological temperature. We note that with concentrated (500-fold) virus, the infectivities of ADA and comb-mut decreased much more rapidly than with unconcentrated virus, possibly due to an effect of lysosomal proteases or other cellular agents that might pellet with the virus. After normalizing for this effect, we found that ADA infectivity was reduced by ∼50% immediately upon treatment and the virus was almost completely inactivated after two hours (Figure 12D). In contrast, comb-mut was significantly more resistant to this treatment and viral infectivity did not drop below 60% of the untreated virus control over the same time period. When analyzed using BN-PAGE, the majority of protease-treated ADA Env trimer was already consumed at the earliest time point analyzed, while the effect on comb-mut Env trimers was significantly delayed and less complete (Figure 13C and D). Thus, the trimer-stabilizing mutations in comb-mut appear to make the Env complex less susceptible to degradation by a cocktail of different protease specificities.
While membrane-anchored Env is arguably most relevant for structural studies and vaccine development, truncated gp140 trimers are also of considerable interest. However, the two forms are likely to have stability requirements that are at least somewhat different since the TM and viral membrane play critical roles in stabilizing native spikes. In addition, techniques employed to artificially trimerize gp140s have typically altered its conformation, which poses a conundrum as to which truncated forms of Env trimer to use to evaluate mutations that were selected in the membrane-anchored context. Rather than investigate artificial trimerization motifs, disulfide bonds, or cleavage site knockout mutations, we decided to determine the oligomerization state of secreted gp140s without further genetic modification. We chose to produce gp140s by transient transfection of 293S (GnTI−/−) cells that result in relatively homogeneous glycosylation (i.e. only Man5, Man8, and/or Man9), as ADA gp140 produced in GnTI−/− cells has been shown to form trimers, at least in the uncleaved form [49]. We generated cleavage-competent ADA and comb-mut Envs that were truncated after amino acid position 664, since trimers with this truncation have been shown to be relatively well-behaved [50]. We observed that ADA and comb-mut gp140 produced in 293S cells was indeed at least partially trimeric as measured by BN-PAGE, whereas the trimeric fraction was negligible when produced for comparison in 293T cells (Figure S3A). The Env constructs have an intact cleavage site between gp120 and gp41, so we wished to determine the actual level of cleavage in the soluble Env preparations using SDS-PAGE. We observed that both ADA and comb-mut soluble gp140s were approximately 50% processed (Figure S3B).
When the oligomeric states of secreted comb-mut and ADA gp140s were compared using BN-PAGE, we observed a statistically significant increase over wild-type in the proportion of comb-mut Env that spontaneously formed trimers (51% trimer for comb-mut and 25% trimer for ADA, p = <0.0001, n = 10), which was accompanied by a corresponding decrease in bands corresponding to non-trimeric Env (Figure 14A, B, and C). To rule out the possibility that this apparent increase in the trimeric population could be an artifact of BN-PAGE/Western blots, we used the same Env preparations in ELISA to analyze binding to PG9 - that has a strong preference for trimeric Env [41] - along with several control mAbs. With the control mAbs, we observed strong binding by both neutralizing (e.g. 2G12 and b12) and non-neutralizing (e.g. b6 and 7B2) control mAbs, with no change in binding between the two Envs with these or other mAbs (Figure 14D and data not shown). However, much weaker but highly reproducible binding was observed using PG9 against both ADA and comb-mut soluble Env, and, consistent with the BN-PAGE data, there was a statistically significant two-fold increase in PG9 binding to comb-mut relative to ADA (Figure 14D). Thus, the secreted Env is comprised of Env trimers that either lack certain antigenic features of native Env, or those Env trimers with native antigenicity would have to be a relatively minor constituent of the total Env population. However, in addition to slightly enhancing trimerization of soluble gp140, the stabilizing mutations in comb-mut also cause a small but significant increase in the proportion of PG9-reactive molecules, which are both sought after features in Env immunogen design.
To determine the stability of soluble trimers of comb-mut and wild-type gp140s, we assayed heat induced trimer dissociation using BN-PAGE (Figure 14E). Interestingly, the lower molecular weight species of Env disappeared from the Western blot at intermediate temperatures and appeared to form larger complexes, while the trimer band disappeared at higher temperatures. Corresponding monomeric dissociation byproducts did not appear concomitantly with the disappearance of the trimer band and all staining became undetectable after treatment at higher temperatures, presumably due to product aggregation. Under these conditions the comb-mut trimer did appear more stable than ADA, as it consistently disappeared from the blot at a higher temperature (i.e. 50% trimer disappearance at 68.4°C for comb-mut vs. 62.3°C for ADA; Figure 14E and F). Importantly, both ADA and comb-mut soluble gp140 trimers were much more thermostable than their functional Env trimer counterparts on the virion, as the gp140s retained a significant amount of trimer following incubation at 68°C for one hour.
Because the gp140s are largely unprocessed (Figure S3B), and because uncleaved Env has been shown to be more stable than its cleaved counterpart on the membrane surface [6], we wished to determine the stabilities of uncleaved and cleaved gp160s of ADA and comb-mut. We compared replication-competent virions that display mostly cleaved Env, pseudotyped virus particles that display mostly uncleaved gp160, and Env produced by DNA transfection in the absence of viral backbone that is essentially uncleaved (Figure S4B and D). When subjected to heat and visualized by BN-PAGE, the uncleaved gp160 formed a less discrete oligomeric band at 57°C, which appeared to increase in size at higher temperatures that cause cleaved ADA Env trimers to dissociate (Figure S4A and C). These results are quite similar to what was observed when soluble gp140s were exposed to heat, suggesting that uncleaved gp140 trimers may share some stability features with their uncleaved gp160 counterparts.
Because the stability of native Env trimers is dependent on interactions with the membrane [7], we wished to investigate comb-mut Env stability in detergent. We previously showed that fully mature, virion-associated Env of a clade B primary isolate, JR-FL, dissociated at physiological temperature in under four hours following solubilization in the mild detergent, DDM [7]. We therefore used BN-PAGE to analyze the stability of ADA wild-type and comb-mut trimers in DDM over time at 37°C. ADA Env trimers quickly dissociated under these conditions and had almost completely decayed after one hour (Figure 13E and F). However, under identical conditions Env from comb-mut retained a substantial fraction of trimeric Env (30%) a full 24 hours following DDM treatment. Hence, the comb-mut Env spike is not only relatively resistant to heat, proteolysis and GuHCl treatment but also exhibits greatly increased stability after being detergent-solubilized.
HIV-1 has been experimentally subjected to various evolutionary selection pressures in order to study its fitness, tropisms, and various aspects of Env structure-function including mechanisms of escape from drugs and neutralizing antibodies [30], [33], [34], [51]. Here, we used directed evolution to identify amino acid changes in HIV-1 Env that increase the stability and homogeneity of the unliganded spike without grossly altering its function or its antigenic properties, and without the aid of structural data. Previous engineering approaches have sought to stabilize Env trimer-based immunogens using intermolecular disulfides, cleavage site knockouts, and artificial trimerization domains, but each approach has adversely affected the function and antigenic profile of the cognate native trimer [18], [28], [52], [53].
Selection strategies may be devised to identify HIV-1 Env trimers with even greater stabilities than we observed here. Env requires a degree of conformational flexibility in order to mediate fusion of the viral membrane with the target cell membrane [54]. Screening for stable Env trimers in the absence of an infectivity requirement may therefore identify a greater diversity of Env-stabilizing mutations. Nevertheless, such screens can also lead to non-native conformations of Env so specific counter-screens may also be necessary.
In our screen for Env stability, several trimer-stabilizing mutations were identified in the NHR of gp41. A prior study identified substitutions I535M and L543Q in the NHR that led to decreased levels of non-trimeric Env on pseudotyped virus, but native Env trimer stability was not explicitly measured [55]. In the current study we show that these substitutions in the NHR stabilize the functional form of HIV-1 Env. The K574R substitution we identified affects a highly conserved residue in the NHR, with >99.5% of group M isolates having a Lys at position 574 in the LANL HIV sequence database. Non-conservative substitutions with K574 tend to destabilize the post-fusion (6HB) conformation of gp41 [45], but their effect on unliganded trimer stability has not previously been studied. We show here that mutations to K574 can either stabilize (e.g. K574R) or destabilize (e.g. K574A) the unliganded, native trimer. Thus, position 574 appears to have a pivotal role in regulating multiple conformations of Env, which may explain the high degree of sequence conservation at this position. Which interactions this residue makes as it transitions between the unliganded and receptor primed forms of Env is unclear as the structural details of these states are currently lacking.
Post-CD4 engagement, the NHR region of gp41 forms a homotrimeric coiled-coil that is transiently accessible to peptide inhibitors and Abs [36], [56]. However, details of NHR structure in the unliganded (ground) state have not been described. The NHR appears to form a homo-trimer in receptor activated and post-fusion conformations of Env and has even been implicated in trimerization of the unliganded Env complex, though apparently not by equivalent mechanisms [55], [57]. The NHR mutations we identified here may enhance subunit-subunit interactions within Env, which could resist structural transitions out of the native state and into CD4-bound, antibody-bound, and other inactive states (see below). We speculate that while the K574R mutation maintains hydrophilicity and charge, the guanidino group may enhance electrostatic interactions or hydrogen bonds with adjacent elements on Env. The mutation L543Q, and to a lesser extent I535M, involve the substitution of a hydrophobic side-chain with a polar residue that is more likely to be found on the surface of the protein, suggesting that this portion of the NHR might be at least somewhat solvent-exposed and poised to interact with other hydrophilic elements. Cryo-electron microscopic (cryo-EM) structures of the Env trimer have shown the presence of a hole in the center of the trimer [58], [59], and a recent study suggests that the NHR helices may line this cavity [60].
We also identified Env stabilizing mutations in the CHR region of gp41: H625N, T626M and S649A. Residues N625 and M626 occur commonly among HIV-1 isolates. However, S649 is conserved in group M (95.3% of isolates), while A649 predominates in groups N and O and SIVcpz. A number of studies have implicated the CHR of gp41 in Env subunit-subunit interactions. Thus, mutations in the CHR can disrupt gp120-gp41 interactions and increase spontaneous shedding of gp120 [23]. In addition, a peptide corresponding to the DSL and CHR regions of gp41 can bind to monomeric gp120 through interactions with the C5 and C1 regions of gp120 [61]. Another study showed that peptides corresponding to gp120 C4 can interact with the peptide fusion inhibitor T-20, the latter of which is comprised mostly of CHR residues [62]. Residues in the CHR immediately N-terminal to position 646 have also been shown to contribute to gp41 trimerization [63]. Thus, the CHR in the unliganded, native trimer could conceivably interact with the inner domain and/or base of gp120 as well as with other gp41 protomers [58]. The presence of the dipeptide motif HT at positions 625/626 has been shown to increase virion infectivity in a CD4-independent manner and makes HIV-1 more sensitive to sCD4 and cold inactivation [64]. We note that, the substitution of NM for HT at positions 625/626 causes a putative N-glycosylation site (PNGS) to shift from N624 to the new N625. Studies have shown that this glycosylation site is occupied [50], [65], so its alteration might explain at least part of the stabilizing effect of this mutation. The S649A substitution involves a hydrophilic to hydrophobic residue change that could mean that this residue is in a more hydrophobic environment (i.e. buried) in the unliganded trimer.
The only stabilizing mutation identified in gp120 was the V1 alteration (N139/I140 deletion, N142S). In cryo-EM models, V1V2 is located at the apex of the trimer where it may interact with adjacent protomers by contacting other V1V2s [58], V3 [66] and/or other elements nearby on Env. V1 is heavily N-glycosylated and likely O-glycosylated as well [67]–[69]. The N142S mutation we identified eliminates a PNGS in V1, so glycosylation at this site and glycosylation proximal to this site may contribute to Env trimer stability as well.
The Env mutants that we identified in the B2 and HC11 library pools have distinct sequences and yet possess a stability phenotype that appears to be largely independent of the method of destabilization (i.e. GuHCl, heat, prolonged incubation at 37°C, destabilizing ligands, proteolysis, and detergent). Moreover, both Env mutants are hyper-sensitive to an entry inhibitor that opposes conformational changes in trimeric Env (i.e. PF-348089). Collectively, the results suggest that more than one element within Env may cooperate to resist trimer-destabilizing treatments, and that the different treatments may inactivate Env through a cooperative mechanism. We previously showed a correlation between heat stability and resistance of HIV-1 to 37°C decay, and, for the isolates JR-CSF and ADA, resistance to GuHCl also correlated with resistance to heat and spontaneous decay [7]. Recently, residues in gp41 including H625/T626 were found to increase CD4 independent infection, global neutralization sensitivity, and sensitivity of Env to cold inactivation [64]. Conformational changes in the Env trimer that lead either to infection or inactivation both involve an irreversible transition over an activation energy barrier [54], [70], [71]. The stability of native Env can either be increased by reducing the Gibbs free energy of the unliganded Env trimer or by increasing the free energy of the transition state that leads to a new state. It seems likely that the mutations we selected tighten interactions between subunits in the unliganded trimer. However, the mutations identified here may also destabilize the transition state that leads to inactive conformations, thus making it less likely that Env will decay. Elucidation of the mechanisms of Env stabilization might reveal structural distinctions between functional forms of HIV-1 Env trimers.
In addition to increasing the stability of Env, the mutations identified here also made the native Env trimer less sensitive to protease digestion. The mutations introduced into comb-mut are not expected to significantly impact the preferred cleavage sites of the enzymes tested. Most likely then the mutations in Env cause it to assume a more closed conformation in which protease cleavage sites are less accessible. It is possible that digestion of Env immunogens in vivo restricts elicitation of certain Abs [72], [73], so incorporation of comb-mut mutations into Env trimer-based immunogens might offer a level of protection against such degradation. We note that the specific enzymes used here would not be encountered at the site of vaccination, but comb-mut Env trimers are resistant to a cocktail of multiple proteases with different specificities so the effect may be more general. Binley and colleagues have shown that sequential glycosidase-protease digests can degrade non-functional Env species with greater efficiency than native Env trimers, but the process does reduce viral infectivity by ∼70% suggesting some effect on functional Env [11], [13]. By including stabilizing mutations in Env it may be possible to remove irrelevant Env without loss of native trimer.
As Env stability and Env homogeneity are not always correlated [6], [7], it is notable that the stable Env mutants selected in this study were also homogeneously trimeric on BN-PAGE. In support of the BN-PAGE results, the stabilized mutant Env virions were also captured with much lower efficiency than wild-type ADA using mAbs that bind poorly to unliganded native trimers (i.e. X5 and hNM01) [6], suggesting that less non-native Env exists on the stable ADA mutants. Reasons for the decrease in non-native Env may include less cellular biosynthesis of aberrantly folded or improperly glycosylated forms of Env prior to incorporation of Env onto the budding virion and/or slower decay of the folded Env trimer, both prior to budding from the cell and on the virus surface [7], [9], [13].
We found that mAbs PG9/16 neutralized the stable Env mutants of ADA more completely than the wild-type virus [10]. When certain PG9/16 resistant viruses are produced in GnTI−/− cells, PG9/16 can neutralize the resulting viruses more efficiently and completely, due at least in part to changes in glycans on the variable loops of gp120 that become enriched in Man5GlcNAc2 and oligomannose structures [10]. The stabilizing mutations therefore seem to reduce glycan heterogeneity within the functional population of Env. Mutations that stabilize the native trimer might increase the packing density of glycans and affect the ability of glycosylation enzymes to trim high mannose residues and add complex glycans [74], despite being distal from the actual glycosylation sites. It is notable that both GB21-6 and HC11-1 have completely different mutations, but both increase the proportion of PG9/16 sensitive Env.
Env spikes with high functional stability and high structural homogeneity might be useful for immunization studies [5]–[7], [13]. However, many factors besides trimer stability can influence immunogenicity of Env including mode and density of display, choice of adjuvant, ability to elicit T-cell help, and the capacity to stimulate the appropriate germline B cells and drive affinity maturation. Virus particles, while displaying native Env trimers, do so at low levels (∼10 copies/virion) and typically induce only weak neutralizing Ab titers [14], [75]. Env displayed at higher density (e.g. as soluble protein or on nanoparticles) may lower the affinity threshold of BCR activation by taking advantage of the avidity effect [76]. The Env-stabilizing mutations we identified increased both the trimerization and stability of secreted gp140s, although these effects were quite modest. However, we also show that this trimeric truncated form of Env is largely uncleaved and is more thermostable than that of native Env. Uncleaved Env trimers, whether membrane-anchored or soluble, can be relatively stable but have antigenic features that are not native-like (i.e. binding of non-neutralizing mAbs) [6]. Described soluble gp140 trimers that include artificial stabilizing alterations also differ antigenically from native Env [15]–[20], [49], [77]–[79]. One factor that might contribute to the non-native properties of secreted Env is that it may traffic in the cell differently from membrane-anchored protein, resulting in differences in folding, processing, and post-translational modification [74]. Other stabilizing modifications or re-routing of soluble gp140 through specific folding and processing pathways may be required to compensate for the gp41 TM/CT truncation.
As an alternative to membrane-anchored or soluble gp140s, detergent-solubilized Env spikes may be purified from virions and used in immunization or structural studies. A detergent solubilization strategy was recently used to prepare Env trimers for cryo-EM analysis, but uncleaved rather than cleaved Env trimers were used due to instability of the latter form of Env [80]. The mutations we identified here greatly increased the stability of fully cleaved virion spikes in detergent raising the possibility that cleaved Env trimers could also be purified. Future studies will be directed at how to incorporate native Env-stabilizing mutations into an immunogen that can elicit neutralizing Ab.
Virus was produced from 293T cells by transient transfection using the polyethylene imine (PEI) as previously described [6]. When virus was amplified in MT2-CCR5ΔCT cells, cells were infected at an m.o.i. of 0.01. Every 2–3 days, one half of the cells and virus-containing media was removed and replaced with media containing fresh cells. This procedure was continued for 10–12 days.
Oligonucleotide directed mutagenesis was used to generate HIV-1 ADA mutant pools by targeting four different regions of Env that have been shown to be involved in subunit-subunit interactions in Env. Mutagenesis was targeted to the C1 region of gp120 (pools C1 and C2), the β3–β5 loop of gp120 (pools B1 and B2), the disulfide loop region (DSL) of gp41 (pools D1, D2, and D3), and the membrane proximal external region (MPER) of gp41 (pool M) (Figure S1A). Mutagenesis was restricted to amino acid residues found to naturally occur in the Los Alamos National Laboratory (LANL) HIV Sequence Database. The libraries were created using two PCRs: one 3′ PCR using a primer containing degenerate codons in the region targeted and a 5′ PCR that would partially overlap with the 3′ PCR upstream of targeted region. These two PCR products were then joined by splicing-overlap-extension PCR and the mutant Envs were subcloned into the molecularly cloned HIV-1 Env display vector, pLAI-ADA. Randomly selected test clones were sequenced and each was found to be a unique variant containing between 1 and 8 mutations in the targeted region (Figure S1B and C). The bulk ligated DNA was used to transfect 293T cells and virus-containing cell culture supernatant was used to infect MT2-CCR5ΔCT cells at an m.o.i. of 0.01 to produce pools of replication-competent viruses.
Incremental concentrations of denaturant (0.25–2 M GuHCl or 0.5–4 M Urea), hyper-physiological temperatures (45.7–53.6°C), and incubation time periods at 37°C (4–6 days), designed to be in the range that inactivates wild-type ADA [7], were used separately to select the 8 virion pools in duplicate. In the case of denaturants, treated viruses were pelleted by centrifugation and the denaturant was washed away prior to the infection step. After the destabilizing treatment, an aliquot of each viral pool was analyzed for infectivity in TZM-bl cells to determine the proportion of virus inactivated, and the remaining virus was rescued on MT2-CCR5ΔCT cells. A total of three such rounds were performed for each virion pool.
Following 3 rounds of selection, individual clones were rescued from each stability-enhanced library pool. Whole RNA was isolated from virions in culture supernatant using the QIAamp Viral RNA kit (Qiagen). SMARTScribe Reverse Transcriptase (Clontech) was used to produce cDNA from the viral RNA using the primer NefOR (AGGCAAGCTTTATTGAGG; donated by D. Mosier, TSRI) which binds downstream from env. Next, env was amplified using the Expand High Fidelity PCR System (Roche) and Env-specific primers (i.e. pLAI5EnvF 5′-TAGGCATCTCCTATGGCAGGAAG-3′ and pLAI3EnvR 5′-GTCTCGAGATGCTGCTCCCACCC-3′). Amplified env was subcloned into pLAI-ADA using a BamH I and Bgl I restriction sites. Individual plasmid DNA, amplified in E. coli, was purified and full-length env was sequenced.
A serial 5-fold dilution was performed for each stability-enhanced virion pool and the virions were added to MT2-CCR5ΔCT cells. After 24 h, the media was replaced, and following a 7 day incubation, cell culture supernatants were harvested and tested for the presence of infectious virus in the TZM-bl assay. The media from the highest dilution to produce infectious virus was saved for stability tests.
Virions were exposed to incremental concentrations of GuHCl or urea for 1 h, increasing temperature for 1 h, or extended incubation at 37°C. Samples treated with denaturants were pelleted in a microcentrifuge (20,000×g at 4°C) and were washed with fresh media twice before being resuspended in an equal concentration of media. Virus was then added to TZM-bl cells and luciferase activity was determined 72 h later using the Bright-Glo System (Promega) and an Orion microplate luminometer (Berthold Instruments). Residual infectivity was determined, and results are expressed relative to untreated virus. All experiments were performed in triplicate.
Virus used for BN-PAGE was pelleted in an Optima ultracentrifuge (Beckman; 60,000×g at 4°C) and resuspended 100-fold concentrated in PBS. Virions were exposed to destabilizing conditions as above. BN-PAGE was performed as previously described [7]. Briefly, samples were treated with 1% DDM for 20 min on ice, and then electrophoresed on 4–16% NativePAGE Bis-Tris gels (Invitrogen). Proteins in the gel were then transferred to a PVDF membrane, membranes were blocked in 5% non-fat dry milk and blotted overnight at 4°C using a cocktail of mAbs to gp120 (b12, 2G12 and F425-B4e8, 2 µg/ml) or to gp41 (2F5, 4E10 and Z13e1 each at 1 µg/ml). After washing, membranes were probed for 30 min at room temperature with a goat anti-humanFc-HRP conjugated Ab (Jackson), and peroxidase activity was assayed using Super Signal West Pico Chemiluminescence (Pierce).
HIV-1 virions were concentrated 500-fold. The following proteases were added in Trypsin buffer (50 mM Tris-HCl, 20 mM CaCl2, pH 8.0): trypsin (50 µg/ml), chymotrypsin (50 µg/ml), and proteinase K (1 mg/ml; all NEB). Virions were incubated at 37°C for the indicated time periods and the digestion was stopped by addition of Complete Protease Inhibitor Cocktail (Roche) and stored at −80°C until analyzed. Samples were analyzed for infectivity and by BN-PAGE.
Env was solubilized from virus particles by addition of n-Dodecyl β-D-maltoside (DDM) to a final concentration of 1% at 37°C. Samples were removed at different time points and analyzed using BN-PAGE as described above.
VCAs were modified from a previously detailed protocol [6]. Microtiter wells were coated overnight at 4°C with capture mAb (5 µg/ml in 50 µl of PBS). Wells were blocked using 4% non-fat dry milk (NFDM) in PBS for 1 h at 37°C. Incremental concentrations of soluble CD4 were added to 50 µl of virus in cell culture supernatant and, after a 15 min incubation, virions were added to the blocked wells and incubated for 2 h at 37°C. Wells were washed 6 times with PBS, and TZM-bl target cells were overlaid (104 cells/well). Luciferase activity was determined after a 72 h incubation as described above.
HIV-1 infectivity and neutralization was determined as described previously [6]. Briefly, TZM-bl reporter cells were seeded in 96-well plates at 104 cells per well in 100 µl complete DMEM and incubated for 24 h at 37°C. Virus samples were incubated with mAbs or inhibitors for 1 h or 20 h at 37°C, and the mixture was added to cells in a total volume per well of 200 µl. Cells were harvested 72 h post-infection, luciferase activity in the cells was determined as above.
The virus ELISA was adapted from a previously described protocol [11]. Virions were immobilized directly on microtiter wells for 2 hours at 37°C (2 ng p24 equivalents per well). Plates were washed (all washes were performed using PBS without detergent) and wells were blocked using 4% NFDM in PBS for 1 h at 37°C. After washing, primary Abs were added in PBS containing 0.4% NFDM for 1 h at 37°C. Plates were washed again, goat anti-human-Fcγ-HRP secondary Ab (Jackson) was added, and the plates incubated for 45 min at 37°C. Following another wash, TMB substrate (Pierce) was added and absorbance read at 450 nm.
ADA wild-type and comb-mut gp140 Env expression vectors were generated by introducing mutations D664G and K665stop in pcDNA-ADA using Quikchange site-directed mutagenesis (Agilent). Cleavage-competent gp140 proteins were produced by transient transfection of 293T and 293S (GnTI−/−) cells as described above for virus production. The oligomeric state of soluble gp140 in cell culture supernatant was analyzed by BN-PAGE and Western blot using only the anti-gp120 mAb cocktail, because the anti-gp41 mAbs used for Western blot staining bind to the region of gp41 removed by the truncation after position 664. The identity of the Env trimer band was verified by comparison with KNH1144 SOSIP [104] and JRFL-foldon soluble trimers [17] (gifts from I. Wilson and R. Wyatt (TSRI), respectively). Relative density of the BN-PAGE bands was analyzed using ImageJ software (NIH) and compared by t-test using GraphPad Prism.
Microtiter wells were coated with Galanthus nivalis lectin (GNL; Sigma) at 5 µg/ml in PBS overnight at 4°C. Plates were then washed using PBS containing 0.05% Tween (PBST); all washes are with PBST. Plates were blocked with 4% non-fat dry milk (NFDM) in PBS for 1 h at 37°C. Next, plates were washed and gp140 cell culture supernatant was added for 2 h at 37°C. Following this incubation, plates were washed and mAb binding was assayed as with the Virus ELISA above, except that 0.05% Tween was included in all steps.
|
10.1371/journal.pntd.0003784 | Free-Roaming Dog Population Estimation and Status of the Dog Population Management and Rabies Control Program in Dhaka City, Bangladesh | Beginning January 2012, a humane method of dog population management using a Catch-Neuter-Vaccinate-Release (CNVR) program was implemented in Dhaka City, Bangladesh as part of the national rabies control program. To enable this program, the size and distribution of the free-roaming dog population needed to be estimated. We present the results of a dog population survey and a pilot assessment of the CNVR program coverage in Dhaka City. Free-roaming dog population surveys were undertaken in 18 wards of Dhaka City on consecutive days using mark-resight methods. Data was analyzed using Lincoln-Petersen index-Chapman correction methods. The CNVR program was assessed over the two years (2012–2013) whilst the coverage of the CNVR program was assessed by estimating the proportion of dogs that were ear-notched (processed dogs) via dog population surveys. The free-roaming dog population was estimated to be 1,242 (95 % CI: 1205–1278) in the 18 sampled wards and 18,585 dogs in Dhaka City (52 dogs/km2) with an estimated human-to-free-roaming dog ratio of 828:1. During the two year CNVR program, a total of 6,665 dogs (3,357 male and 3,308 female) were neutered and vaccinated against rabies in 29 of the 92 city wards. A pilot population survey indicated a mean CNVR coverage of 60.6% (range 19.2–79.3%) with only eight wards achieving > 70% coverage. Given that the coverage in many neighborhoods was below the WHO-recommended threshold level of 70% for rabies eradications and since the CNVR program takes considerable time to implement throughout the entire Dhaka City area, a mass dog vaccination program in the non-CNVR coverage area is recommended to create herd immunity. The findings from this study are expected to guide dog population management and the rabies control program in Dhaka City and elsewhere in Bangladesh.
| Rabies is a public health problem in Bangladesh. A CNVR program was commenced in Dhaka City in January 2012 as part of the Bangladesh national rabies control program. We describe the findings of a dog population survey that was conducted to estimate the free-roaming dog population using a mark-resight framework and the progress of the CNVR program. The free-roaming dog population in the 18 sampled wards was estimated to be 1,242 dogs, and 18,585 dogs in Dhaka City (52 dogs/km2). The estimated human-to-free-roaming dog ratio was 828:1. Between 2012 and 2013, 6,665 dogs (3,357 male, 3,308 female) were neutered and vaccinated against rabies in 29 of the 92 city wards. A pilot survey conducted in 18 wards indicated a CNVR coverage of 60.6% (range 19.2–79.5%). We recommend conducting an annual mass dog vaccination in the non-CNVR coverage area during the intervening period of the CNVR program to create herd immunity against rabies and break the cycle of rabies transmission.
| Rabies kills an estimated 2,000–2,500 people every year in Bangladesh, ranking it third globally after India and China in terms of human impact [1–4]. In Bangladesh an estimated 166,590 (95% CI: 163,350–170,550) cases of animal bites in humans are reported each year, contributing to an estimated annual incidence of 1.40 human rabies deaths per 100,000 population [3]. The Infectious Disease Hospital (IDH) located in Dhaka City is the main referral centre for rabies patients in Bangladesh; it provides free treatment to 350 to 450 dog bite victims daily [1, 5]. For example, between 2004 and 2012 the IDH in Dhaka City reported 1,152 human rabies deaths [1, 5]. At the district level, 65 rabies prevention and control centers provided free anti-rabies vaccine and treatment to dog bite victims [5]. In domestic animal populations, 3,425 rabies deaths (cattle: 2845; goats: 547; sheep: 13) were reported in a passive surveillance system (2010–2012) in Bangladesh [6]. However, rabies cases in dogs were not captured by this surveillance system and other reliable data are scarce. The mortality in both animals and humans may be several fold higher than reported since rabies is not a notifiable disease in Bangladesh [1].
In Bangladesh, domestic dogs act as the main source of rabies for both domestic animals and humans [1,3]. Until late 2011, mass dog culling was implemented in major towns in Bangladesh, in an unsuccessful attempt to control rabies [1]. For example between 2003 and 2008 there were 139,391 stray dogs culled in five major towns (Dhaka, Khulna, Rajshahi, Sylet and Tongi) in Bangladesh, of which 80% (n = 112,078) were culled in Dhaka City alone (average 22,415 dogs per year) [1], yet rabies infection remained endemic. Therefore, Obhoyaronno-Bangladesh Animal Welfare Foundation (OBAWF) carried out an advocacy campaign against mass culling of dogs and recommended a humane method of dog population and rabies control in Dhaka City. Subsequently, the government approved and provided support and commitment to the OBAWF to implement a long-term humane dog population management and rabies control program in Dhaka City through Catch-Neuter-Vaccinate-Release (CNVR). Since January 2012, mass culling of dogs was stopped in Dhaka City. On 1 April 2012, OBAWF signed a memorandum of understanding with both Dhaka North and South City Corporations to assume the responsibility of managing Dhaka City’s entire dog population humanely. OBAWF carried out CNVR as part of the National Rabies Control program in Dhaka City.
Knowledge of the size of the dog population is crucial for dog management and for assessing the effectiveness of dog population and rabies control strategies. However, no studies have been conducted to estimate the size of the free-roaming dog population in Dhaka City. In addition, since the start of the CNVR program in January 2012, no assessment has been made of the coverage of the CNVR program in Dhaka City.
The objectives of this study were to: (1) estimate the size of the free-roaming dog population using a mark-resight framework, (2) describe the status of CNVR program conducted during 2012–2013, and (3) estimate the proportion of free-roaming dogs that had been sterilized and vaccinated against rabies in the CNVR program in Dhaka City. The findings from this study are expected to guide the dog population management and rabies control program in Dhaka City and elsewhere in Bangladesh.
The People's Republic of Bangladesh is located in South Asia. It is divided into seven administrative divisions (Barisal, Chittagong, Dhaka, Khulna, Rajshahi, Sylhet, Rangpur) and 64 districts. Each district is further divided into Upazilla (sub-district) while the metropolitan areas are divided into wards (Fig 1, panel A-D).
Dhaka is the capital of Bangladesh and is also the principal city of Dhaka Division and Dhaka District. The city is administered by the Dhaka City Corporation (DCC) which is divided into two administrative parts—North and South Dhaka City Corporations—ensuring better civic facilities. The city is divided into 10 zones and further sub-divided into 92 wards (Fig 1, panel C and D). An estimated 15.391 million people live in Dhaka City and it is one of the most densely populated cities in the world [7].
In this paper, free-roaming dogs are defined as any dogs seen in public areas that are not confined to its owner's house or property and are not currently under direct human control. Therefore, this term encompasses owned, unowned and community owned dogs but not for those on leashes or under direct human control at the time of survey [8]. However, the proportion of owned dogs that are typically free-roaming is unknown in Dhaka City, but is likely that some owned dogs are free-roaming.
During January to March 2011, a mark-re-sight procedure was used to estimate the free-roaming dog population in Dhaka City. As part of the sampling process, a polygon map for each ward was coloured with one of five colors: brown, red, green, pink and blue according to World Society for the Protection of Animals (WSPA) guidelines (Fig 1, panel D) [8]. A lottery was then used to decide which of the denoted wards would be chosen for surveys. Red colour polygons were selected and these consisted of 19 wards. Then a detailed route map of each of the 19 selected wards were prepared for the survey based on the neighbourhood road networks. Some of the larger wards were split and sub-divided into smaller, more manageable blocks for the survey.
Three two-person survey teams were formed for the population survey. Each team was assigned wards and fixed routes to follow on a daily basis for the survey. On day 1, the teams travelled by motorcycle through their predetermined routes and sprayed water-soluble blue colour vegetable dye on to the dogs observed in the street, without capturing the dogs. A farmers' sprayer machine was used for marking the dogs. The harmless colour mark remains on the dog body for at least two weeks and can be easily resighted from a distance during secondary sampling. The team made three to five visits to mark the dogs within the designated neighbourhood. At the time of marking, the team recorded the total number of dogs marked in each ward. The number of dogs marked with colour on the first day represented the n1 within the mark-resight framework [9]. On the second day (7.00–12.00 h), each team returned to the same location/neighbourhoods where the dogs were marked on the previous day and carried out counting of free-roaming dogs. All survey teams walked through the street in one direction without overlapping the area to avoid double counting of dogs. During the counting process, the teams recorded the presence of a paint mark on the body of sighted dogs. The total number of dogs counted on day 2 forms the n2 while the number of colour marked dogs resighted forms m in the mark-resight framework (Table 1). The same process of marking and counting event was done in all wards sequentially.
The Lincoln—Petersen’s formula with Chapman’s correction was applied to estimate the free-roaming dog population [10].
N=[(n1+1)(n2+1)(m+1)]−1
(1)
where, N is the estimate of the total population size, n1 is the total number of animals sighted and marked on day 1, n2 is the total number of animals sighted on day 2, and m is the number of marked animals on day 1 that were sighted on day 2.
An approximately unbiased variance of N was estimated by using Seber's formula [9]:
var(N)=[(n1+1)(n2+1)(n1−m)(n2−m)(m+1)2(m+2)]
(2)
And the 95% confidence interval for N was estimated as:
N±1.965var(N)
(3)
The 18 surveyed wards and DCCs had an approximate area of 23.880 and 126.59 km2 respectively. The total number of free-roaming dogs within the areas of DCC was estimated by adjusting the total estimated free-roaming dog population in the studied wards with ward area sampling fraction (1242 ÷ 23.880/126.59 = 1242/0.1886) [7]. To estimate the free-roaming dogs in the Dhaka metropolitan area (DMA), a total density of dogs estimated per km2 in the surveyed wards was adjusted with the entire metropolitan area (357.42 km2) by assuming homogeneity of the area in terms of habitat and food resource availability for free-roaming dogs within the city. The analysis was conducted using Microsoft Excel (Microsoft Excel 2007, Redmond, USA).
The first CNVR program in Dhaka City commenced in January 2012 and was implemented by Obhoyarono-Bangladesh Animal Welfare Foundation. The CNVR program was conducted in 18 wards (Ward No 7 to 14; 39, 40, 43 to 50) during 2012 and in 29 wards (Ward No 2 to 14; 39 to 50; 54, 55, 58, 60) during 2013 of Dhaka North City Corporation (Table 2). The CNVR program is ongoing and will be continued until all 92 wards in Dhaka City have been covered, based on the memorandum of understanding signed between OBAWF and DCCs.
Prior to the start of the program, four veterinary surgeons and four para-veterinarians were trained for two months (September–October 2011) by the HSI in India on a range of dog population management aspects including humane dog catching, aseptic surgical sterilization methods (ovario-hysterectomy and castration), monitoring of post operative complications, dog population survey and data management.
During the program, free-roaming dogs sighted in the street were humanely captured by the trained dog catchers either by hand or using nets in the morning (6.30–10.00 h) and brought using a mobile van to the clinic set up at Boshila, Mohamadpur (Ward no 46), under Dhaka North City Corporation (Fig 1, panel D). However, pregnant and lactating dogs were not captured for sterilization and vaccination. In few instances, the program had to be suspended whenever there was political unrest and strikes in the city, public holidays and also due to logistic constraints including shortage of drugs.
Dogs were classified as either owned, community or ownerless dogs. Owned dogs were brought in to the clinic by the owners; ownerless dogs were captured in the street by the dog catchers. When communities claimed that the dogs lived within their community and that they fed the dogs, they were classified as community dogs. All dogs were handled and neutered—including anesthetic and drug administration—according to the standard protocols developed by HSI. Each dog was given a rabies vaccine (Rabisin, Merial, France) injection. The dogs were also given an ivermectin (Techno Drug Ltd, Narsingdi, Bangladesh) injection to control internal and external parasites. A V-shaped ear notch was performed on all neutered and vaccinated dogs that were classified as ownerless or community dogs while under anesthesia to permanently identify processed dogs. The surgical procedure was completed before 14.00 h to ensure that all operated dogs were fully recovered from anesthesia before being released back to the place of capture in the same afternoon. Recovery from anaesthesia typically takes about two to three hours post operation. Any dog that displayed discomfort, weakness after surgery or took longer to recover from anesthesia were retained at the clinic and treated before being released back to their territory.
Two years’ (January 2012 to December 2013) data on the CNVR program were retrieved from the database and descriptive analyses were performed to understand the sex, ownership, and space-time pattern of dogs caught and processed in the clinic.
During January and February 2014 a pilot assessment of CNVR coverage—via dog count surveys—was carried out in 18 of the 29 randomly selected wards of Dhaka North City Corporation (Ward No: 2, 3, 4, 5, 6, 8, 9, 13, 14, 41, 43, 46, 47, 48, 50, 54, 55 and 58) where the CNVR program was implemented during 2012–2013 (Table 2). Three teams each comprising of three persons walked predetermined routes through each selected ward and counted the number of dogs by direct observation (without capturing or handling dogs). For each dog counted, the teams recorded the details based on visual assessment: sex (male, female), presence of ear notch (yes, no), body (1—very thin, 2—thin, 3—normal, 4—stout, 5—overweight) and skin (normal—no observable skin lesions, mildly diseased—few skin lesions, moderately diseased—moderate number of skin lesions, severely diseased—severe skin lesion with apparent mange infestation) condition scores and the approximate age (pup—up to 6 months, juvenile—6 to 12 months and adults—more than 12 months) (Table 3). Counting was conducted between 6.30–12.00 h when dogs were most active and likely to be visible to the counting team.
Vaccination and sterilization coverage of the free-roaming dog population in the 18 sampled wards were estimated as the proportion of ear-notched dogs sighted and counted during the population survey. Descriptive analyses were performed to describe the male-to-female ratio and the health condition of the dogs.
Only 18 out of the 19 selected wards in Dhaka City could be surveyed: ward 90 could not be surveyed due to logistic constraints. A total of 816 dogs (n1) were marked with colour paint spray between January and March 2011 and 775 dogs (n2) were counted at the secondary count event; of these 775 dogs, 518 (m) were re-sighted with colour paint marks (Table 1). The re-sighting probability was 63%. Using the Lincoln-Petersen index-Chapman correction methods, the free-roaming dog population was estimated to be 1,242 (95% CI: 1205–1278) in the 18 sampled wards and 6,584 dogs within the DCC administrative areas (Table 1). The DMA area has an area of approximately 357.42 km2 and by using the dog density estimates from the sampled wards, the overall free-roaming dog population in DMA during early 2011 was estimated to be 18,585 dogs (52 dogs per km2). Using the human population of Dhaka City to be 15.391 million, the human-to-dog ratio was estimated to be 828:1.
A total of 6,713 (2,587 in 2012 and 4,126 in 2013) dogs were caught and presented to the CNVR clinic for vaccination and sterilization between January 2012 and December 2013. Of these, 6,665 (99.3%) dogs (2553 in 2012 and 4112 in 2013) were processed (vaccinated, sterilized and released). The remaining 48 dogs had various health conditions and were thus unfit for surgical intervention, but nonetheless were given an anti-rabies vaccine injection. Almost equal number of male (3,357; 50.4%) and female (3,308; 49.6%) dogs were processed during 2012 and 2013 (male-to-female ratio 1.01:1). The majority of the dogs processed were classified as ownerless (5,266; 78%) whereas 14% (942) were owned and 8% (505) were community dogs. Fig 2 shows the gender-specific monthly pattern of dogs processed at the clinic during the two year period. The highest number of dogs (n = 2,354) were processed in wards 46 and 48 (Table 2). The relationship between frequency of visits and CNVR coverage in each ward is shown in Table 2.
The survey team counted 6,341 dogs, of which 3,844 were found to be ear-notched, in 18 wards. The overall point estimate of vaccination and sterilization coverage was estimated to be 60.6% (95% CI: 59.4–61.8). Of the 18 wards, eight had an estimated coverage >70% (Table 2). There was a higher proportion of male (3,528) than female (2,759) dogs in the sampled wards (overall male-to-female ratio of 1.28:1). The CNVR coverage was 62.1% (2,191) in male and 58.9% (1,618) in female dogs.
The age distribution of ear-notched dogs was found to be skewed towards the adult groups (4,832 [78.3%]; 95% CI: 77.2–79.3), with fewer puppies (938 [15.1%]; 95% CI: 14.2–15.9) and juveniles (413 [6.6%]; 95% CI: 6.1–7.3). The estimated CNVR coverage in adults, puppies and juveniles was 63.4% (3,088), 50.7% (476) and 45.3% (187), respectively.
The majority of the free-roaming dogs sighted in this study had good body condition (score 3) and normal skin condition. Neutered dogs and male dogs had higher BCS and better skin condition that intact dogs and female dogs respectively (Table 3).
As far as we are aware, this was the first study conducted to estimate the size of the free-roaming dog population in Dhaka city, Bangladesh, using mark-re-sight methods. This information was needed to design the subsequent CNVR program, for example to estimate the doses of rabies vaccine and other drugs required and to determine how many dogs would need to be sterilized and vaccinated to control the dog population and to reduce the rabies transmission risk. However, this population survey was conducted in 2011 after a citywide mass dog culling operation conducted by the city corporations in which more than 22,400 dogs were eliminated every year (2003–2008) in an attempt to control rabies and reduce the dog population in Dhaka City [1]. The current study generated a point estimate of the dog population based on data collection during early 2011. It may therefore underestimate the existing population within the city because the mass dog culling campaign was replaced by a long-term sustainable dog population management and rabies control via CNVR in January 2012. Nevertheless, this baseline information can be used to understand the dog population size in Dhaka City and to compare future estimates.
The mark-resight method is considered to be a practical way of estimating the number and distribution of a free-roaming dog population, if the assumption of a closed population (no appreciable births, deaths, immigration and emigration of dogs) is fulfilled during the primary and secondary sampling intervals [9,11]. In this study, marking and subsequent counting events were completed within three days and thus the assumption of a closed population was likely valid because of the very short period between counting events. Only a few published studies have used mark-resight and sight-resight methods (by observing the natural body marks on dogs, counting permanently identifying features such as ear-notch status, collaring of dog or colour paint spray) to estimate the size of free-roaming dog populations [12–18]. The method chosen for population estimation will depend on the availability of resources and their practicality in the field. In the current study, we applied vegetable colour paint marks to dogs followed by secondary counting events and resight of the marked and unmarked dogs to estimate the free-roaming dog population. We have found previously that such colour paint marks remain on the dog's body for at least two weeks (even when it rains) and that the marked dogs can be easily sighted from a distance during the secondary resighting sampling process [17]. This method is quick and cheap and can be used for population surveys or to mark dogs during vaccination campaigns to assess the vaccination coverage when other methods are unavailable or impractical.
Within the DMA the estimated free-roaming dog population was 18,585 dogs and 52 dogs/km2, respectively. There are huge differences in estimated dog densities between wards (Table 1) which may be associated with variation in human density and the availability of food resources. The population of free-roaming dogs in Dhaka City cannot be assumed to be homogenous, an important finding for planning CNVR program and other disease response plans. Therefore, available resources could be used more effectively by focusing dog and rabies control measures in areas known to have higher dog density to achieve high vaccination coverage. The estimated human-to-free-roaming dog ratio of 828:1 is relatively high compared to international studies presumably due to high human density in Dhaka City. The dog density (dogs/km2) estimate in Dhaka City is moderate compared to other Asian cities which range from 5.78 in Timor Leste to 2,930 in Kathmandu, Nepal, with human-to-dog ratio ranging from 4.7:1 in Kathmandu to 23:1 in Timor Leste [15, 18, 19]. In contrast, Hossain et al., [4] estimated 14 dogs per km2 with human-to-dog ratio of 120:1 in rural Bangladesh. The density of dogs is expected to be lower in rural than urban areas. Also, estimates of the density of dogs may vary between different areas and countries due to socio-cultural differences and the type of dogs included in counts. In the current study, only free-roaming dogs (but not owned restrained dogs) were counted, although it is expected that some proportion of owned dogs might have been included in the survey.
During the two years (January 2012–December 2013) CNVR program in Dhaka City, more than 6000 dogs were processed with variable coverage between the neighbourhoods (Table 2). The teams, however, attempted to improve the coverage by making repeated visits to areas that had previously been targeted. For instance, during 2013 the CNVR program was re-focused on those neighbourhoods/wards which were already targeted during 2012 to increase the overall level of coverage (Table 2). However, some wards were visited only once (for example ward 42 and 60) during 2012 and 2013. This is because when the dog catching team assessed few free-roaming dogs within the neighborhood, the team moved to the next wards for capturing and in this way, the free-roaming dogs within the city wards were captured and taken to the clinic for processing. Nevertheless, there has been a consistent increase in the total number of dogs processed at the CNVR clinic from 2,553 dogs in 2012 to 4,112 in 2013.
Most dogs processed at the clinic were classified to be ownerless, rather than owned or community dogs. It is possible that owned dogs may not have been presented to the clinic for sterilization because the dog owners did not want their dogs to be sterilized, or the clinic was too far away and the owners did not have enough time to take their dogs. In some cases, owned dogs may have been presented to the government veterinary clinics or private clinics, although the service was provided free of charge at the CNVR clinic.
The point estimate for CNVR coverage among the free-roaming dogs in the sampled 18 wards was 60.6%. However, assuming that some owned dogs are vaccinated and neutered but not ear-notched and are free-roaming in the street, this CNVR coverage estimates may have been underestimated. Of these 18 wards, eight had an estimated coverage level >70%, exceeding the WHO recommended threshold for rabies eradication [20, 21]. Within these eight wards there were repeated visits by the dog catching teams and longer program duration. For example, ward numbers 43, 46 and 48 were visited 45, 83 and 84 times, respectively, by the dog catching teams and thus achieved higher coverage compared to other wards. Since city wards are demarcated mainly for administrative purposes and have no physical boundary, a certain amount of movement of dogs between the wards/neighborhoods is likely to have occurred during, as well as following, the CNVR program. For instance, a higher number of ear-notched dogs were observed in certain wards during the post-CNVR program assessment survey, compared to the initial number of dogs processed in those wards (Table 2), indicating movement of dogs between neighbourhoods.
Although almost equal number of male and female dogs were processed during CNVR program in 2012 and 2013, a higher number of male than female free-roaming dogs were found during a pilot assessment surveys conducted during early 2014. This may be due to higher survival of males than females as BCS of males were found to be better than females during a pilot study. The finding of more male than female dogs in this pilot assessment study was consistent with what has been reported elsewhere in the world where male dogs predominate [13, 17, 22, 23]. The information on sex-ratio of free-roaming dogs will be useful for planning the logistical arrangements for CNVR program in Dhaka City.
This pilot assessment study has helped to identify those wards/neighbourhoods during 2014 where the program coverage was low. Most of those wards with lower coverage and new wards have been covered during 2014. Such evaluation surveys are necessary to assess the coverage and then modify the program accordingly. In this survey, not all 29 wards could be included due to logistical reasons, which is why these results pertain to only 18 wards. Although there is no rule-of-thumb regarding what proportion of the dog population needs to be neutered to control and stabilize the dog population, WSPA recommend 70% coverage (similar to the target vaccination coverage for rabies elimination), but this will require long-term planning and sustained resources [13, 24]. Nevertheless, OBAWF have committed to cover the entire Dhaka City to manage the dog population. Since the CNVR program may take few more years to cover the entire Dhaka City dog population, mass dog vaccination against rabies is recommended in the non CNVR areas during the intervening period to create herd immunity and break the transmission chain of rabies virus. The CNVR program needs to cover more areas (city wards) and dogs within a short time period. This can be strategically achieved by setting up additional CNVR clinics in different wards to improve the level of coverage. This will be more efficient since collecting dogs from far off neighborhoods and transporting them to and from the existing clinic located at ward No 46 (Fig 1, panel D) takes considerable time and resources.
The majority of the free-roaming dogs sighted in the street had good BCS, with neutered dogs appearing healthier than intact dogs, and also males appearing healthier than female dogs (Table 3). The improvement of BCS and skin condition of the processed dogs could be due to the direct benefit of the CNVR program [25]. For instance, all dogs presented to the clinic were given one dose of ivermectin injection to control internal and external parasites and also treated against skin infections and other health ailments encountered during the processing. One of the objectives of the CNVR program is to improve the health and welfare of free-roaming dogs. The health benefits of sterilization in both males and females include preventing multiple pregnancies, reducing sexually transmissible diseases (such as transmissible venereal tumor), improving body weight and condition, and increasing life span [25–27]. In females repeated pregnancies can physically stress animals, while the absence of pregnancy can improve health and lifespan [25]. The most important benefit of sterilization is stabilization of the population by reducing the population turnover rate, but this requires a sustained program over a longer period of time. Because of a longer life span of neutered [25,26] and vaccinated dog, herd immunity against rabies is also maintained and not lost as rapidly between rounds of vaccination. It is expected that the health and welfare of free-roaming dogs may improve in the future as more dogs are covered by the CNVR program.
This study provides baseline information about the dog population size and the status of CNVR coverage in Dhaka City. We recommend that indicator counts are conducted every year to assess the dynamics (size and distribution) of the dog population and CNVR coverage. Also, recording the number of dog bites in humans and incidences of rabies in the city would provide information about the impact of the program. Because the CNVR program will take time to cover the entire Dhaka city area simultaneously, annual mass dog vaccination is recommended in the non-CNVR coverage area to create herd immunity among the dog population and to break the transmission chain of rabies virus. In addition, the setting up of additional CNVR clinics (e.g mobile clinics) could increase coverage and reduce the dog population and rabies incidence, as well as improve the health and welfare of dogs in Dhaka City. There is also a need for inter-agency collaborations and the One Health approach engaging human health and veterinary professionals to control rabies in Bangladesh.
|
10.1371/journal.pntd.0001890 | Filariasis Attenuates Anemia and Proinflammatory Responses Associated with Clinical Malaria: A Matched Prospective Study in Children and Young Adults | Wuchereria bancrofti (Wb) and Mansonella perstans (Mp) are blood-borne filarial parasites that are endemic in many countries of Africa, including Mali. The geographic distribution of Wb and Mp overlaps considerably with that of malaria, and coinfection is common. Although chronic filarial infection has been shown to alter immune responses to malaria parasites, its effect on clinical and immunologic responses in acute malaria is unknown.
To address this question, 31 filaria-positive (FIL+) and 31 filaria-negative (FIL−) children and young adults, matched for age, gender and hemoglobin type, were followed prospectively through a malaria transmission season. Filarial infection was defined by the presence of Wb or Mp microfilariae on calibrated thick smears performed between 10 pm and 2 am and/or by the presence of circulating filarial antigen in serum. Clinical malaria was defined as axillary temperature ≥37.5°C or another symptom or sign compatible with malaria infection plus the presence of asexual malaria parasites on a thick blood smear. Although the incidence of clinical malaria, time to first episode, clinical signs and symptoms, and malaria parasitemia were comparable between the two groups, geometric mean hemoglobin levels were significantly decreased in FIL− subjects at the height of the transmission season compared to FIL+ subjects (11.4 g/dL vs. 12.5 g/dL, p<0.01). Plasma levels of IL-1ra, IP-10 and IL-8 were significantly decreased in FIL+ subjects at the time of presentation with clinical malaria (99, 2145 and 49 pg/ml, respectively as compared to 474, 5522 and 247 pg/ml in FIL− subjects).
These data suggest that pre-existent filarial infection attenuates immune responses associated with severe malaria and protects against anemia, but has little effect on susceptibility to or severity of acute malaria infection. The apparent protective effect of filarial infection against anemia is intriguing and warrants further study in a larger cohort.
| In many regions of the world, including sub-Saharan Africa, concomitant infection with multiple parasites is common. In order to examine the effects of filariasis, a chronic helminth infection, on immune responses and clinical manifestations of acute malaria infection, the authors followed 31 filaria-infected (FIL+) and 31 filaria-uninfected (FIL–) individuals living in a malaria-endemic area of Mali through an entire malaria transmission season for the development of clinical malaria (fever or other symptoms of malaria in the setting of detectable blood parasites). Serum levels of inflammatory cytokines previously associated with severe malaria were decreased in FIL+ subjects at the time of acute clinical malaria. Although there were no differences between FIL+ and FIL– subjects with respect to the time of first episode of malaria or the number or severity of malaria episodes, filarial infection appeared to protect against the development of anemia during the malaria transmission season. These findings demonstrate that chronic filarial infection modulates the immune response to acute malaria. The apparent effect on anemia is intriguing and deserves further study.
| Filarial infections and malaria are coendemic in many areas of the world, including sub-Saharan Africa, where human coinfection with malaria and filarial parasites is common [1], [2], [3]. Chronic filarial (helminth) infection is associated with skewing of parasite-specific immune responses towards a Th2/Treg cytokine pattern [4]. Furthermore, studies have demonstrated extension of this skewing of the immune response in helminth-infected individuals to bystander antigens with clinically relevant effects on responses to immunization with tetanus toxoid [5], [6] and oral cholera vaccine [7]. In contrast to tissue invasive helminth infection, malaria is characterized by an acute inflammatory response with increases in serum proinflammatory cytokine and chemokine levels, including IL-1β, IL-6, IL-8, TNFα, IP-10, IL-1ra and IL-12p70 [8], [9], [10], [11], [12], [13], [14]. Although several of these mediators have been implicated in control of the infection [9], [10], they are also associated with increased severity of clinical disease [11], [12], [13], [14]. Consequently, with the implementation of worldwide programs to eliminate filariasis and other helminth infections, questions have arisen regarding the impact of antifilarial chemotherapy (and de-worming) on immune responses to acute malaria infection and the potential effect of this immune modulation on malaria morbidity and mortality [15].
A number of studies have demonstrated that chronic filarial infection can suppress human immune responses to malarial antigens in vitro, including a recent study that showed an association of patent filarial infection with decreased malaria antigen-specific IL-12p70/IFNγ production [4]. This effect was reversed by the addition of neutralizing antibodies to IL-10. Furthermore, filaria-infected individuals with asymptomatic malaria parasitemia have been shown to have lower frequencies of malaria-specific Th1 and Th17 cells [16]. Despite clear evidence for chronic modulation of malaria-specific immune responses in vitro, few studies have addressed the effects of chronic filarial infection on immune responses and clinical outcomes in the setting of acute malaria in human populations.
The current study was designed to determine if pre-existent filarial infection alters the frequency and/or severity of clinical malaria in children and young adults in a coendemic area in Mali where malaria transmission follows a seasonal pattern. Cytokine profiles at the time of presentation with acute malaria were also compared between subjects with and without concomitant filarial infection.
The study was designed as a matched prospective study of the effects of filarial infection on the incidence, severity and immune responses to malaria infection.
The study was conducted in the villages of Tieneguebougou and Bougoudiana in the district of Kolokani, approximately 105 km northwest of Bamako, Mali, between June 2007 and December 2007. Prior studies in these villages had demonstrated a high prevalence of Wuchereria bancrofti (Wb) and Mansonella perstans (Mp) microfilaremia in the population, but no evidence of onchocerciasis. A single round of mass drug administration (MDA) with ivermectin and albendazole was conducted by the National Program to Eliminate Lymphatic Filariasis prior to the study in May 2006, and study subjects participated in a second round of MDA during the study in August 2007. Malaria transmission in this region is seasonal (June–December) with a cumulative EIR of 19.2 infective bites/person in a neighboring village in 2000 [17].
The study flowchart and selection of study participants is shown in Figure 1. Non-pregnant volunteers (n = 539) of both genders, aged from 1 to 20 years, were screened with a brief medical history and physical examination. Subjects with evidence of severe or chronic illness, a history of allergy to anthelmintic or antimalarial agents or plans to relocate out of the village during the study were excluded, as were subjects who refused venipuncture or had difficult venous access. The remaining subjects underwent laboratory testing, including pregnancy testing (if appropriate), hemoglobin (Hgb) measurement using a portable analyzer (Hemocue, Lake Forest, CA), Hgb typing by HPLC (D-10 instrument; Bio-Rad, http://www.bio-rad.com), daytime thick smear for detection of malaria and/or Mp infection, and circulating antigen (CAg) ELISA for Wb (TropBio, Townsville, Australia). Based on these results, potential participants were divided into two groups: 1) FIL+: individuals with confirmed active filarial infection with Wb and/or Mp, as defined by the presence of CAg and/or microfilaremia (mf) and 2) FIL−: individuals without evidence of active filarial infection (negative for CAg and mf). Thirty-one FIL+ subjects were identified and matched to FIL− subjects on the basis of gender, age (within 4 years), and Hgb type. Glucose-6-phosphate dehydrogenase (G6PD) deficiency was assessed by restriction enzyme analysis [18] in all 62 subjects. Filarial infection status was confirmed at the time of enrollment in the prospective study, and reassessed after 3 and 6 months by Nuclepore™ filtration of 1 ml of blood drawn between 10 pm and 2 am and CAg ELISA (TropBio).
All participants were evaluated weekly throughout the malaria transmission season. At each visit, symptoms and signs of clinical malaria were assessed, and if present, blood was obtained from a finger stick for preparation of Giemsa-stained thick and thin blood smears for detection and speciation of malaria parasites and Hgb measurement. In addition, subjects were encouraged to present to the study physician for evaluation if they felt ill. Blood smears and Hgb determinations were also performed at the start of the study and monthly thereafter, regardless of the presence of clinical signs or symptoms. Reading of these monthly blood smears was deferred until the end of the study unless the subject was symptomatic at the time of the blood draw. To minimize the potential effects of concomitant helminth infection, all subjects were treated with single dose praziquantel (40 mg/kg) and albendazole (400 mg) at the start of the study. Albendazole treatment was repeated every two months during the study period.
Clinical malaria was defined as the presence of fever (axillary temperature ≥37.5°C) or another symptom or sign compatible with malaria infection plus the presence of asexual malaria parasites on a thick blood smear. Malaria parasitemia was determined by counting the number of asexual P. falciparum parasites on Giemsa stained thick blood films until 300 leukocytes were observed. All slides were read by two independent microscopists, and differences of >10% were resolved by an expert microscopist. Parasite densities were calculated by multiplying the mean value for the two readers by 25 to give parasites/microliter.
Cases of uncomplicated malaria were treated with standard recommended doses of artesunate-amodiaquine (4 mg/kg artesunate and 10 mg/kg amodiaquine per day for 3 days). Cases of therapeutic failure were treated with quinine (10 mg/kg/day for 3 days). On days 1, 2, 3, 7 and 14 following treatment, subjects were evaluated for clinical signs and symptoms, and malaria smears and Hgb measurements were performed. Plasma levels of interferon (IFN)γ, interleukin (IL)-1α, IL-1β, IL-1ra, IL-6, IL-8, IL-10, IL-12p70, IP-10, and tumor necrosis factor (TNF)α were measured in plasma samples obtained prior to treatment by suspension array in multiplex (Millipore Corp, St. Charles, MO), according to the manufacturer's instructions. The limits of detection of the assay are 0.5 pg/ml for IFNγ, 0.63 pg/ml for IL-1α, 0.19 pg/ml for IL-1β, 10.97 pg/ml for IL-1ra, 0.79 pg/ml for IL-6, 0.32 pg/ml for IL-8, 0.41 pg/ml for IL-10, 0.23 pg/ml for IL-12p70, 1.14 pg/ml for IP-10, and 0.22 pg/ml for TNFα.
The study (NCT00471666) was approved by the ethical review committees of the Faculty of Medicine, Pharmacy, and Dentistry at the University of Bamako (Mali) and of the National Institutes of Allergy and Infectious Diseases (Bethesda, Maryland). Community permission for the study was obtained from village elders, and individual oral or written informed consent (for subjects ≥18 years of age) or parental informed consent and assent (for subjects <18 years of age) was obtained from all participants in French or Bambara, the local language. Oral consent was obtained for illiterate participants and parents of participants as approved by the ethical review committees and was documented by the signatures of a member of the study team and a witness.
The ratio of the rates of clinical malaria events was calculated by Poisson regression adjusting for the time at risk [19]. Person-season was defined as the sum of the days at risk for all subjects during the first season divided by 168 days (24 weeks). The time to first episode of clinical malaria analysis used the Kaplan-Meier estimates and Cox proportional hazards [20]. Differences between the two groups were tested using the Mann-Whitney test (continuous responses) or Fisher's exact test (binary responses). Confidence intervals (CI) on the ratio of geometric means (GM) were calculated using a t-test on the log-transformed values and CIs on the difference in arithmetic means (for temperature) used t-tests on untransformed values. Wilcoxon signed rank test was used to test for changes over time and Spearman rank for assessment of correlation. Analyses were done in R [20] or GraphPad Prism (V5.0).
Baseline characteristics for the two groups are shown in Table 1. Overall, the subjects had a median age of 13 years, 65% were male, and the majority had normal Hgb (HgbAA). G6PD deficiency was present in 6/62 (14%) subjects and was equally common in the two groups. Baseline GM Hgb was similar between subjects with and without filariasis (12 g/dL in both groups). Malaria parasites were detected in a high proportion of subjects at baseline in both groups (58.1% and 64.5% in FIL+ and FIL−, respectively), but among those with detectable malaria parasites, parasitemia was low (GM 233 parasites/µl in FIL+ and 147/µl in FIL−, respectively; p = 0.06), with a level >1000 malaria parasites/µl in only one subject in the FIL− group. No intestinal helminth infections were detected in any of the subjects by stool examination, although intestinal protozoa (Giardia and Entamoeba) were found in 4 and 11 subjects, respectively.
Filarial status was assessed at baseline and at 3 and 6 months. Among the 31 FIL+ subjects, 11 (35.5%) were infected with Wb and 20 (64.5%) with Mp. Two subjects had detectable Wb microfilariae in the peripheral blood. During the course of the study, 3 FIL+ subjects, all of whom were Mp-infected, became FIL− (Table 2). As expected, the two subjects with Wb microfilaremia cleared their circulating mf at 3 months (after participating in the mass drug administration of ivermectin and albendazole in their village). Eleven FIL− subjects acquired filarial infection during the study: 7 became positive for CAg, 3 for Mp mf and 1 for both CAg and Mp mf.
Although not all subjects were followed for the full 24 weeks season, 51.3 person-seasons of data were available for analysis. During the malaria transmission season, 38 (61%) subjects developed clinical malaria. There was no significant difference in the rates of clinical malaria between the FIL+ and FIL− groups (Rate ratio [FIL+/FIL−] = 0.85; 95% CI 0.49, 1.46; p = 0.56; Figure 2). Similarly, the time to first episode of clinical malaria was comparable with a median time to first episode of 17 weeks in the FIL+ groups and 15 weeks in the FIL− group (Hazard Ratio = 0.77; 95% CI 0.41, 1.45; p = 0.42; Figure 3). GM parasitemia at the time of the first episode of clinical malaria was also comparable between FIL+ and FIL− subjects (584 parasites/µl and 1216 parasites/µl, respectively; p = 0.18; GM ratio = 0.48; 95% CI 0.16, 1.40; Figure 4).
Clinical signs and symptoms were similar between the two groups at the time of diagnosis of the first episode of clinical malaria (Table 3), and no subjects met WHO criteria for severe malaria during the study. Fever, as defined by temperature >37.8 degrees was documented in 7/18 FIL+ and 5/20 FIL− subjects during their first episode of clinical malaria (p = 0.49). Mean temperatures were also comparable between the two groups (37.7 vs. 37.4 degrees Celsius in FIL+ and FIL−, respectively; p = 0.15; difference in means = 0.34; 95% CI 0.13, 0.81).
Hgb levels were measured monthly throughout the malaria transmission season to provide an indirect measure of chronic malaria morbidity (Figure 5). Values obtained during monthly visits at which clinical malaria was diagnosed were excluded from the analysis. GM Hgb levels were similar between the FIL+ and FIL− subjects at baseline (11.5 g/dL vs. 11.3 g/dL, respectively) and rose significantly in both groups at month 1 (11.5 to 12.2 g/dL in FIL+ and 11.3 to 12.1 g/dL in FIL−; p<0.01, Wilcoxon signed rank test). GM Hgb levels continued to slowly rise over the course of the transmission season in the FIL+ group, but decreased in the FIL− subjects during months 3 and 4 at the height of the malaria transmission season (from 12.1 to 11.4 g/dL; Figure 5) and were significantly lower than the GM Hgb values for the FIL+ group. Of note, the prevalence of asymptomatic malaria parasitemia was similar between the two groups at baseline (Table 1) and remained comparable throughout the transmission season (data not shown).
Plasma levels of cytokines and chemokines previously reported to be associated with the severity of acute malaria were measured at the time of presentation with clinical malaria in FIL+ (n = 10) and FIL− (n = 14) subjects. Among the analytes measured, GM levels of IL-1ra, IP-10 and IL-8 were all significantly decreased in FIL+ subjects (99, 2145 and 49 pg/ml, respectively) as compared to FIL− subjects (474, 5522 and 247 pg/ml, respectively; Figure 6). In contrast, GM plasma levels of IL-10 were increased in FIL+ subjects (172 pg/ml) as compared to FIL− subjects (62 pg/ml), although this difference did not reach statistical significance (p = 0.08). Plasma levels of IFNγ, IL-1α, IL-1β, IL-6, IL-12p70 and TNFα were comparable between the two groups.
In keeping with prior reports of an association between plasma levels of IP-10 and malarial anemia, plasma IP-10 levels measured at the time of acute clinical malaria were inversely correlated with Hgb levels at the peak of malaria transmission (r = −0.54, 95% CI −0,79, −0.11; p = 0.01; Figure 7). A negative correlation was also observed with the next measured Hgb level after the episode of acute malaria (r = −0.62, 95% CI −0.86, −0.16; p = 0.01; data not shown).
Despite considerable data supporting an effect of chronic helminth infection on immune responses to malaria parasites, longitudinal studies examining the incidence and severity of clinical malaria in helminth-infected and –uninfected individuals have been few. In the present study, 62 children and young adults with (FIL+) and without (FIL−) chronic filarial infection were matched for age, gender and hemoglobin type and followed longitudinally through a malaria transmission season. No significant differences were detected between FIL+ and FIL− subjects with respect to the incidence of clinical malaria, time to first episode of clinical malaria, number of episodes of clinical malaria during the transmission season, parasitemia at first episode, or clinical signs and symptoms at first episode. A prospective study of similar design that examined the effects of coinfection with schistosomiasis on clinical malaria in 338 Malian children aged 4–14 years of age demonstrated a significant increase in the time to first episode of clinical malaria by 10–14 days in coinfected children (p = 0.04), but only in the 4–8 year old age group [21]. Coinfected children, aged 4–8 years, also experienced slightly fewer episodes of clinical malaria (1.55 vs. 1.81 episodes, p = 0.03) during the transmission season. No differences were observed in the incidence of clinical malaria, number of episodes during the transmission season or parasitemia during an episode. Taken together, these two studies suggest that the effect of chronic helminth infection on the incidence and severity of acute clinical malaria is likely small and limited to younger age groups.
Despite the lack of an observable effect on the frequency or severity of clinical malaria, chronic filarial infection appeared to protect against the development of anemia during the height of the malaria transmission season (Figure 5). This is the exact opposite of what has been reported in the setting of hookworm infection [22], where hemoglobin levels were found to be significantly decreased in coinfected subjects compared to those with either hookworm or malaria infection alone. In the latter study, a protective effect of helminth infection on hemoglobin levels may have been outweighed by gastrointestinal blood loss in the hookworm-infected subjects. Although the mechanism by which chronic helminth infection might protect against anemia remains to be elucidated, there is evidence to suggest that pro-inflammatory cytokines play an important role in mediating malarial anemia [23].
Prior studies from our group and others have demonstrated increased plasma levels of IL-10, as well as increases in regulatory T cell frequency and function, in children and young adults with helminth infection [4], [16], [24], [25]. More importantly, most studies to date have demonstrated increased production of IL-10 and/or decreased production of inflammatory cytokines in helminth-infected individuals in response to malarial antigen stimulation in vitro [4], [26], [27]. In the present study, we examined serum cytokines at the time of acute malaria and found a similar pattern in vivo with a decrease in cytokines/chemokines associated with severe malaria (IL-1ra, IP-10 and IL-8) and an increase, albeit not statistically significant, in IL-10. Interestingly, these results are the opposite of what has been reported in similarly conducted studies of coinfection with schistosomiasis and malaria in Mali, in which coinfected children (4–14 years of age) had increased plasma IL-10 levels at baseline but a blunted response in the setting of malaria infection as compared to children without schistosomiasis [25]. The explanation for this discordance is unclear, but may relate to differences in the age groups of the subjects or characteristics of the infecting helminths studied.
The association between high plasma IP-10 levels and more severe malarial anemia has been reported previously [23], [28], and was confirmed in the present study (Figure 7). Of note, in a recent study conducted in Mali, the Fulani, an ethnic group known to be less susceptible to clinical malaria than other ethnic groups, were shown to have significantly higher plasma levels of IP-10 at baseline than Dogon children living in the same area, but no significant change in IP-10 levels in the setting of malaria infection [29], suggesting that the change in IP-10 levels may be a more important determinant of malarial anemia than baseline levels. In the present study, FIL+ subjects were found to have decreased plasma levels of IP-10 compared to FIL− subjects at the time of acute clinical malaria. Although plasma levels of IP-10 prior to the malaria transmission season were not available for the present study cohort, data from a cross-sectional study of 38 FIL+ and FIL− subjects in the same village suggest that the magnitude of the increase in plasma IP-10 levels in the setting of acute malaria infection was likely to have been greater in the FIL− cohort [4]. Furthermore, as demonstrated in the prior study, this blunting of the inflammatory response to malarial antigen in FIL+ subjects was most likely due to increased IL-10 [4].
A major anticipated limitation of the present study was the relatively small sample size due to the difficulty in recruiting FIL+ subjects in the younger age groups. Consequently, in order to maximize the possibility of detecting significant differences between the FIL+ and FIL− groups, a matched design was selected. Potential confounding variables, such as baseline malaria parasitemia, intestinal helminth infection, G6PD deficiency, bednet use and location of residence, were also similar between the two groups. Although subjects were not tested for HIV infection, the prevalence of HIV is extremely low in rural Mali and no subjects with clinical evidence of immunodeficiency were identified in the screening. Additional factors that may have diminished differences between the FIL+ and FIL− groups include lasting effects of intestinal helminth infection following treatment, decreased severity of malaria infection due to earlier intervention as a result of the presence of health care personnel in the village (as has been seen in other studies) and the possibility that some FIL− subjects may have had occult (microfilaria-negative) Mp infection or have acquired filariasis during the course of the study. In fact, 8 FIL− subjects became CAg-positive during the study and 4 became microfilaria-positive for Mp. Although these subjects were likely infected at the start of the study, they were considered FIL− for the purposes of the study (intent-to-treat analysis). In view of these considerations, it is likely that small (clinically insignificant) differences in responses between the FIL+ and FIL− groups were missed.
In summary, although pre-existent filarial infection attenuates cytokine/chemokine responses known to be associated with severe malaria, it appears to have little effect on susceptibility to or severity of acute malaria infection in children and young adults living in a malaria-endemic area with seasonal transmission. The apparent protective effect of filarial infection on decreased hemoglobin levels at the height of malaria transmission is intriguing and warrants further study in a larger cohort.
|
10.1371/journal.pcbi.0030161 | Elucidating the Altered Transcriptional Programs in Breast Cancer using Independent Component Analysis | The quantity of mRNA transcripts in a cell is determined by a complex interplay of cooperative and counteracting biological processes. Independent Component Analysis (ICA) is one of a few number of unsupervised algorithms that have been applied to microarray gene expression data in an attempt to understand phenotype differences in terms of changes in the activation/inhibition patterns of biological pathways. While the ICA model has been shown to outperform other linear representations of the data such as Principal Components Analysis (PCA), a validation using explicit pathway and regulatory element information has not yet been performed. We apply a range of popular ICA algorithms to six of the largest microarray cancer datasets and use pathway-knowledge and regulatory-element databases for validation. We show that ICA outperforms PCA and clustering-based methods in that ICA components map closer to known cancer-related pathways, regulatory modules, and cancer phenotypes. Furthermore, we identify cancer signalling and oncogenic pathways and regulatory modules that play a prominent role in breast cancer and relate the differential activation patterns of these to breast cancer phenotypes. Importantly, we find novel associations linking immune response and epithelial–mesenchymal transition pathways with estrogen receptor status and histological grade, respectively. In addition, we find associations linking the activity levels of biological pathways and transcription factors (NF1 and NFAT) with clinical outcome in breast cancer. ICA provides a framework for a more biologically relevant interpretation of genomewide transcriptomic data. Adopting ICA as the analysis tool of choice will help understand the phenotype–pathway relationship and thus help elucidate the molecular taxonomy of heterogeneous cancers and of other complex genetic diseases.
| The amount of a given transcript or protein in a cell is determined by a balance of expression and repression in a complex network of biological processes. This delicate balance is compromised in complex genetic diseases such as cancer by alterations in the activation patterns of functionally important biological processes known as pathways. Over the last years, a large number of microarray experiments profiling the expression levels of more than 20,000 human genes in hundreds of tumor samples have shown that most cancer types are heterogeneous diseases, each characterized by many different expression subtypes. The biological and clinical goal is to explain the observed tumor and clinical heterogeneity in terms of specific patterns of altered pathways. The bioinformatic challenge is therefore to devise mathematical tools that explicitly attempt to infer these altered pathways. To this end, we applied a signal processing tool in a meta-analysis of breast cancer, encompassing more than 800 tumor specimens derived from four different patient cohorts, and showed that this algorithm significantly outperforms popular standard bioinformatics tools in identifying altered pathways underlying breast cancer. These results show that the same tool could be applied to other complex human genetic diseases to better elucidate the underlying altered pathways.
| Microarray technology is enabling genetic diseases like cancer to be studied in unprecedented detail, at both transcriptomic and genomic levels. A significant challenge that needs to be overcome to further our understanding of the relation between the quantitative transcriptome of a sample/cell and its phenotype is to unravel the complex mechanism that gives rise to the measured mRNA levels. The amount of a given mRNA transcript in a normal sample/cell is determined by a whole range of biological processes, some of which (e.g., transcription repression and degradation) act to reduce this number, while others (e.g., transcription factor induction) act to increase it. Therefore, it is natural to model the level of a given mRNA transcript as the net sum of a complex superposition of cooperating and counteracting biological processes, and, furthermore, to assume that disease is caused by aberrations in the activation patterns of these biological processes that upset the delicate balance between expression and repression in otherwise healthy tissue. Many distinct biological mechanisms that underlie the aberrations observed in human cancer have been identified, most notably copy-number changes [1] and epigenetic changes [2], yet it is the effect that these changes have downstream on the functional pathways that ultimately dictates whether these changes are pathological or not.
While several studies have recently characterised the altered functional pathways and transcriptional regulatory programs in human cancer, they have done so either by interrogating the expression data directly with previously characterised pathways, regulatory modules [3–6], and functionally related gene lists [7], or by interrogating derived “supervised” lists of genes for enrichment of biological function [8]. Hence, these studies have not attempted to infer the altered biological processes, which putatively map to alterations of known functional pathways and transcriptional regulatory programs. Thus, an unsupervised method that first infers the underlying altered biological processes and then relates these to aberrations in pathways or regulatory module activity levels is desirable.
A necessary property of such an algorithm is that it allows “gene-sharing,” so that a specific gene can be part of multiple distinct pathways. In this regard, it is worth noting that popular approaches for analysing transcriptomic data, such as hierarchical or k-means clustering, do not allow for genes to be shared by multiple biological processes, since they place a gene in a single cluster [9], and so they are not tailored to the problem of inferring altered pathways.
Algorithms that allow genes to be part of multiple processes/clusters have also been extensively applied [10–12]. Among these, Singular Value Decomposition (SVD) or Principal Components Analysis (PCA) provides a linear representation of the data in terms of components that are linearly uncorrelated [12]. While this linear decorrelation of the data covariance matrix can uncover interesting biological information, it is also clear that it fails to map the components into independent biological processes, since there is no requirement for PCA components to be statistically independent. Mapping the data to independent biological processes, whereby independence is measured using a statistical criterion, should provide a more realistic representation of the data, since it explicitly recognises how the data was generated in the first place. This assumption, which is to be tested a posteriori, underlies the application of Independent Component Analysis (ICA) to gene expression data [13,14]. Specifically, ICA decomposes the expression data matrix X into a number of “components” (k = 1,2,..K), each of which is characterised by an activation pattern over genes (Sk) and another over samples (Ak) (Figure 1 and Materials and Methods),
in such a way that the gene activation patterns (S1,S2,. . .,SK) are as statistically independent as possible while also minimising the residual “error” matrix E (in the above, ⊗ denotes the Kronecker tensor product). It is worth noting that while ICA also provides a linear decomposition of the data matrix, the requirement of statistical independence implies that the data covariance matrix is decorrelated in a non-linear fashion, in contrast to PCA where the decorrelation is performed linearly.
Many studies have shown the value of ICA in the gene expression context as a dimensional reduction and gene-functional discovery tool [15–20] and also as a potential tool for classification and diagnosis [21,22]. To validate the ICA model, most of these studies used the Gene Ontology (GO) framework [23]. However, GO does not provide the best framework in which to evaluate the ICA paradigm, since many genes with the same GO term annotation may not be part of the same biological pathway or may not be under the control of the same regulatory motif, and vice versa. In fact, to date no study has evaluated the ICA paradigm in the explicit context of biological pathways and regulatory modules.
In this work we apply various popular ICA algorithms to six of the largest available microarray cancer datasets. We focus on breast cancer for two reasons. First, for this type of cancer many large patient cohorts that have been profiled with microarrays are available. Second, breast cancer is a highly heterogeneous disease and hence it provides a more challenging (and hence suitable) arena in which to compare and evaluate different methodologies. We also use two large microarray datasets from two other cancer types to show that our results are valid more generally. The aim of our work is 2-fold. First, to test the ICA paradigm by showing that it significantly outperforms both a gene-sharing method that does not use the statistical independence criterion (PCA) and a traditional (“non–gene-sharing”) clustering method (k-means). We achieve this by using a pathway and regulatory module–based framework for validation. The second aim is to find the most frequently altered pathways and regulatory modules in human breast cancer and to explore their relationship to breast cancer phenotypes.
The main modelling hypothesis underlying the application of ICA to gene expression data is that the expression level of a gene is determined by a linear superposition of biological processes, some of which try to express it, while other contending processes try to suppress it (Figure 1). It is assumed that these biological processes correspond to activation or inhibition of single pathways or sets of highly correlated pathways, and that each of these pathways only affects a relatively small percentage of all genes. Because of the statistical independence assumption inherent in the ICA inference process, we would expect the identified independent components to map more closely to known pathways than an alternative linear decomposition method, like PCA, that does not use the statistical independence criterion. Similarly, we would expect ICA components to map closer to pathways than clusters derived from popular clustering algorithms such as k-means or hierarchical clustering.
To test the modeling hypothesis of ICA for expression data, we first asked how well the inferred components mapped to known pathways, as curated in the MSigDB pathway database [24] (Materials and Methods, Table S1). This strategy was initially applied to a total of six breast cancer microarray datasets (“Perou” [25], “JRH-1” [26], “Vijver” [27], “Wang” [28], “Naderi” [29], “JRH-2” [30]), summarised in Table 1, and for four different implementations of the ICA algorithm (“fastICA”, “JointDiag”, “KernelICA”, and “Radical”) [31–34] as well as for ordinary PCA and two versions of k-means clustering (PCA-KM and MVG-KM) (Materials and Methods and Protocol S1). For each of the ICA algorithms and PCA, we inferred ten components and selected the genes based on their weights in the corresponding column of the source matrix S (Materials and Methods). The average number of genes selected per component ranged from 50 to 200 depending on the cohort (Table S2). For the two k-means clustering algorithms, ten gene clusters were inferred on subsets of most variable genes to ensure that the average number of genes per cluster was similar to that of the PCA and ICA components. This step was necessary to ensure an objective comparison of the different algorithms. In what follows we also use the term component to denote clusters. To evaluate how close the inferred components of a given algorithm in a particular cohort mapped to existing pathways, we defined a pathway enrichment index, PEI, as follows. For each component i and pathway p, we first evaluated the significance of enrichment of genes in that pathway in the selected feature set of the component by using the hypergeometric test (see Materials and Methods). This yielded for each component i and pathway p a p-value Pip. Correction for multiple testing was done using the Benjamini-Hochberg procedure to obtain an estimate for the false discovery rate (FDR). A component i was then declared enriched for a pathway p if the Benjamini-Hochberg corrected p-value was less than 0.05. Hence, we would expect approximately 5% of significant tests to be false positives. Finally, we counted the number of pathways enriched in at least one component and defined the PEI as the corresponding fraction of enriched pathways.
The PEI for each of the seven methods (“PCA”, “MVG-KM”, “PCA-KM”, “fastICA”, “JointDiag”, “KernelICA”, “Radical”, and “PCA”) and the four largest breast cancer sets (“Vijver”, “Wang”, “Naderi”, “JRH-2”) are shown in Figure 2A (the results for all six breast cancer cohorts are presented in Figure S1). This showed that across the four major cohorts the PEI was higher for ICA algorithms when compared with PCA and the clustering-based methods. Interestingly, for the two largest cohorts (“Vijver” and “Wang”), the degree of improvement in the PEI of ICA over PCA, MVG-KM, and PCA-KM was highest. In contrast, for the smaller cohorts (e.g., “Perou” and “JRH-1”), the degree of improvement of ICA over PCA or KM was less marked. Hence, since we found that cohort size had a significant impact on the inferred components, we restricted all subsequent analysis to the four major breast cancer cohorts. It is also noteworthy that when comparing the various ICA algorithms with each other we didn't observe any appreciable difference in their respective PEI.
To investigate this further, we next compared the algorithms on the subset of nine cancer-signalling pathways from the curated resource NETPATH (http://www.netpath.org) and five oncogenic pathways [35]. These are pathways that are frequently altered in cancer and hence we would expect many of these to be captured by the ICA algorithm. Thus, for each method and study we counted the number of pathways that were enriched in any of the components (Figure 2B). This showed that in the three largest breast cancer studies (“Vijver”, “Wang”, and “Naderi”), PCA and the KM-methods captured the least number of pathways. In the two largest cohorts (“Vijver” and “Wang”), for example, the “RADICAL” ICA algorithm captured ten and six of the 14 pathways, while PCA captured eight and two pathways, respectively.
As a further validation that ICA outperforms PCA, we investigated the relation of the derived components with regulatory modules. Specifically, we tested the selected gene sets from each component for enrichment of genes with common regulatory motifs in their promoters and 3′ UTRs [36]. Under the ICA paradigm we would expect genes that are under the common regulatory control of a transcription factor to appear in the same ICA component. Thus, for each breast cancer cohort and method we counted the number of regulatory motifs whose associated genes were overrepresented in components (Figure 2C), using as before the hypergeometric test to test for significant enrichment (Materials and Methods). This showed that PCA performed worst out of all algorithms. In two cohorts (“Wang” and “Naderi”), none of the PCA components was associated with any of the 173 distinct regulatory motifs. In contrast, the components derived by ICA algorithms were consistently associated with regulatory motifs. Interestingly, the improvement of ICA over KM-based methods was less marked with only study (“Wang”) showing a substantial improvement (Figure 2C).
The results above show that ICA provided a more biologically meaningful decomposition of breast cancer expression data than PCA or KM-based methods. This suggested to us that similar results would hold in other types of cancer. To check this, we analysed two additional large microarray datasets, one profiling 221 lymphomas [37] (“Hummel”) and another profiling 132 gastric cancers [38] (“Chen”) (see Table 1). The same analysis on these two additional datasets confirmed that the PEI was higher for ICA when compared with PCA or KM-clustering methods (Figure 2A), and that ICA components also mapped closer to known regulatory motifs (Figure 2C).
To investigate the robustness of the algorithms, we next compared the ability of the algorithms to identify pathways and regulatory modules that were differentially activated independent of the breast cancer cohort used. Two important observations that were independent of the ICA algorithm and cohort used could be derived from the heatmaps of differential activation of pathways and regulatory modules (Figures S2–S5). First, ICA identified many more pathways that were consistently differentially activated across all four breast cancer cohorts (Figure 3A). This further confirmed that the associations between components and pathways as picked out by ICA were more robust and consistent between cohorts than those identified through PCA, MVG-KM, or PCA-KM. Among the pathways that were found to map most frequently and consistently to components were those related to estrogen signalling as well as to other important breast cancer–signalling pathways such as the EGFR1 and TGF-β pathways (Figures 3B and S2–S5). We also found cell-adhesion, immune-response, cell-cycle, and metabolic pathways to be commonly differentially activated across the cohorts. While breast cancer studies have found study-specific gene clusters associated with cell-cycle, estrogen-response, cell-adhesion, and immune-response functions, our results show that expression variation across breast tumours can be understood in terms of single pathways (i.e., a fixed common set of genes for all studies) that relate to these biological functions.
Second, we also observed that ICA outperformed PCA, MVG-KM, and PCA-KM in identifying regulatory modules that were consistently differentially activated across cohorts (Figure 3C). Specifically, the KernelICA algorithm identified the regulatory modules TATA, AACTTT, NFAT, IRF, and NF1, while MVG-KM only picked out TATA, with PCA and PCA-KM failing to capture any regulatory module. Among the motifs with regulatory gene modules that were most frequently captured by independent components, we found several with important general (e.g., TATA) and specific transcription factors (e.g., NF1 and ETS2) (Figures 3D and S2–S5).
We next asked whether components mapping into the various pathways/modules were associated with breast cancer phenotypes. Specifically, we considered three categorical phenotypes: estrogen receptor (ER) status (0,1), histological grade (1,2,3), and outcome (0,1). To evaluate statistical significance of any association between a component k and phenotype, we considered the distribution of weights from the corresponding row of the mixing matrix, i.e., Ak (Materials and Methods), across the different categories. We used the Wilcoxon rank-sum test for the two binary phenotypes and the Kruskal-Wallis test for histological grade. Because of the clustering nature of the MVG-KM and PCA-KM algorithms, in these two cases we first applied k-means over the genes in the cluster to partition the samples into two groups and subsequently used Fisher's exact test to determine whether the phenotype distribution across the two groups was significantly different from random or not.
This revealed a complex pattern of significant associations with several components differentiating breast tumours according to ER status and histological grade (Figures S2–S5). It is notable that in all cohorts ICA components associating with clinical outcome were also found, while PCA generally did not. Another feature was the fact that more and stronger phenotype associations were uncovered by using ICA as compared with PCA. On the other hand, MVG-KM performed as well as ICA in mapping to ER, grade, and outcome phenotypes.
Since we characterised each component in terms of the differential activation pattern of cancer-related pathways and regulatory modules, for those components associated with a phenotype we were able to link the corresponding pathways and regulatory motifs with the phenotype (Figure 4). This led to several well-known but also novel observations. First, as expected, ICA components that were strongly associated with ER status were frequently mapped to the estrogen signalling pathway. Second, ICA components that mapped to the CR (cancer related) cell-cycle pathway [39] were frequently associated with either grade or outcome. The association between cell-cycle genes and grade or outcome is well-known [26,30,40], and our finding further shows that an independently characterised cell-cycle pathway associates with these clinical variables across multiple studies. Third, we observed that pathways relating to immune response functions and the classical complement pathway were frequently correlated with ER status, grade, and, although less frequently, with clinical outcome. For example, we found in each of the four major breast cancer cohorts an ICA component that mapped to the CR immune response pathway [39], and which was consistently overactivated in ER− relative to ER+ tumours (Figure 5A and Table 2). We note that the same set of genes, when viewed over the measured expression matrix also separated the samples according to ER status (Figure 5B and Table 2). Fourth, in all studies where grade information was available, an ICA component mapping to either matrix-metalloproteinases (MMP) or the cell-adhesion pathway was found to be associated with histological grade. In three studies (“Wang”, “Vijver”, and “Naderi”), the MMP pathway was also found to be associated with outcome. Another interesting pathway we found to be associated with histological grade was an epithelial–mesenchymal transition (EMT) signalling pathway characterised in [41]. Specifically, ICA revealed a component driving upregulation of genes involved in EMT in poorly differentiated tumours relative to low-grade tumours across the three studies where grade information was available (Figure 6A and Table 3). When the same set of genes defining the EMT pathway was viewed over the measured expression matrix, their pathway coherence was less evident, although the association with grade was still revealed by k-means clustering (Figure 6B and Table 3).
The parallel analysis for regulatory motifs and breast cancer phenotypes provided direct links between the associated transcription factors and clinical variables (Figure 4B). Strikingly, we found that the interferon regulatory factor (IRF) showed the strongest associations with both the ER and grade phenotypes. The regulatory module associated with the TATA box was also frequently associated with ER, grade, and outcome. Interestingly, we found differential activation of the regulatory modules associated with the neurofibromin-1 (NF1), NFAT, and ETS2 transcription factors to be associated with clinical outcome, which is significant in view of the results of several recent studies linking these transcription factors with the metastatic and cell-growth properties of breast cancer cells [42–46].
It is important to point out that ICA facilitated the identification of many of the biological associations in comparison with PCA, MVG-KM, and PCA-KM (Figure 7). Thus, for example, we can see that the association between immune response and ER status was found in all cohorts by any one of the four ICA algorithms, whereas PCA and the KM methods were generally not as robust (Figure 7A). A similar observation could be made for the associations between the EMT pathway and grade, and that of the IRF module and ER status (Figure 7B and 7C). For the case of NF1 and clinical outcome, this association was not identified by PCA or the KM-based methods (Figure 7D).
Finally, we verified that in many cases the identified associations were independent, in the sense that the component(s) or genes linking a pathway with a phenotype could be different from the one(s) linking another pathway with the same phenotype. For example, we noted that this was the case for the associations of the cell-adhesion and estrogen-signalling pathways with grade (see Figures S2 and S4). Similarly, the associations of the immune response pathway and IRF module with ER status (Figure 7A and 7C) could not be attributed to a common gene subset selection, since the pathway and module gene sets shared no genes in common.
Networks are a useful tool for graphically representing relational structures between many layers of organisation. In our application, we sought to construct a network of associations, linking breast cancer phenotypes, pathways, and regulatory modules with each other as the nodes in the network. To represent only the most salient and robust features, we focused attention on those pathways and regulatory modules with most phenotypic associations (Figure 4) and on those associations that were most consistently predicted across cohorts. Thus, we constructed an average network over the networks for each study by defining a link between any two nodes in the network if there were at least three studies in which there was a link between the two nodes, as predicted by ICA (Figure 8) (KernelICA was used but the other ICA algorithms gave similar networks). This revealed a complex network of associations between transcription factors, pathways, and breast cancer phenotypes. Strengthening the association of immune response with ER status further, we found triangular relationships involving the NF-κβ, ETS2, and IRF transcription factors (Figure 8A), which is plausible in view of their role in regulating immune response pathways [47–49]. The corresponding network for clinical outcome showed that apart from the cell-cycle and estrogen-signalling pathways, only the EGFR1 and TGF-β pathways were consistently associated with outcome (Figure 8B).
In our view, it is most natural to analyse gene expression data in the context of a generative model, however approximate this model is to the true underlying mechanism that gives rise to the measured expression levels. ICA provides such a generative model since it explicitly recognises how the data was generated in the first place. By comparing ICA with PCA and clustering-based methods, we have shown that a more realistic representation of the data is obtained by allowing “gene-sharing” and using the statistical independence criterion (non-linear decorrelation) in the inference process (ICA), as opposed to not allowing gene-sharing (MVG-KM, PCA-KM) and only using a linear decorrelation criterion (PCA). We showed this on a total of six cancer microarray datasets, using existing pathway knowledge and gene regulatory module databases for evaluation. Specifically, we found that ICA components mapped closer to cancer-related pathways as well as to gene modules that are under the control of a common regulatory motif. It is worth pointing out though that the improvement of ICA over KM methods was less marked in the case of regulatory motifs, as we would expect, since a clustering method is partially tailored to finding co-regulatory structure. Importantly, when comparing the results across cohorts, we found that ICA algorithms were much more robust than PCA or KM-based methods, in the sense that pathways that were found to be differentially activated through ICA in one cohort were also consistently differentially activated in the other cohorts. A similar observation could also be made for the regulatory motifs and their regulatees. For example, using PCA or PCA-KM, no regulatory module was found to be differentially activated across all four major breast cancer studies, while the ICA algorithms found an average of four modules. The most likely explanation for the relatively smaller number of regulatory modules found in common across the four studies, as compared with pathways, is that many regulatory modules important to breast cancer have yet to be elucidated.
Of note, we also performed the enrichment analysis of the independent components for chromosomal bands (using the MSigDB database), which confirmed that the independent components were not capturing transcriptional programs localised to specific chromosomal regions. Instead, we believe that the inferred independent components encapsulate “net” transcriptional programs that act globally and downstream of the epigenetic and genetic modifications underlying cancer.
We also found that ICA components were associated more often with known breast cancer phenotypes, including clinical outcome, and that these associations were also much stronger for ICA than for PCA. While this result is to be expected, since ICA components map closer to pathways that have been characterised using phenotypic information, one should also bear in mind that these pathways were derived from independent experiments; hence, the stronger associations between components, pathways, and phenotypes as revealed by ICA provides a validation, not only of the algorithm itself, but also of the characterised pathways.
Another important observation was the presence of multiple components showing an association with a particular pathway, regulatory module, or phenotype. This suggests that a significant proportion of pathways are part of multiple biological processes. Alternatively, the presence of multiple components enriched for a given pathway may reflect distinct gene subset selection, which in turn suggests that the pathways in MSigDB and NETPATH may need to be refined further. In the context of phenotypes, the presence of multiple components correlating with ER status, grade, or outcome, is suggestive of tumour heterogeneity, since, more often than not, the differential distribution of the phenotype across samples is dependent on the precise component. Hence, the fingerprint patterns of pathway activation derived from ICA could potentially form the basis for further clinically relevant definitions of breast cancer subtypes.
In an exploratory analysis, ICA revealed many interesting associations between pathways and phenotypes that can form the basis for future investigations. While all methods were able to identify the expected relationships of the estrogen-signalling pathway with ER status and cell-cycle pathway with histological grade, ICA clearly outperformed PCA and KM-clustering in identifying many other biologically relevant associations (Figure 7). For example, ICA consistently found an expression mode involving immune response pathways that was upregulated in ER− versus ER+ tumours. Thus, while the relation between immune response and ER status is still poorly understood [50], our results clearly point at an important link between the immune response and estrogen signalling in breast cancer, which needs to be explored further. ICA also revealed interesting associations of the EMT-signalling, cell-adhesion, and MMP pathways with histological grade and clinical outcome. Specifically, we found a component upregulating EMT genes in high-grade versus low-grade tumours, and which was statistically significant in three major cohorts. The association between the activity level of the cell-adhesion and MMP pathways with clinical outcome as revealed by ICA is also noteworthy given that supervised approaches tend to only find genes related to cell-cycle pathways, as these are the strongest predictors of grade and outcome. While the association of cell-adhesion genes with outcome has been noted before in breast cancer [29] and to a lesser extent in gastric cancer [51], here we show that this result holds for a specific pathway and across several breast cancer cohorts. ICA, in contrast to PCA and KM-clustering, also identified interesting associations between transcription factor modules and phenotypes (Figure 7). For instance, it found strong associations between the IRF and ER status and between NF1 and clinical outcome, as well as an association between NFAT and outcome (Figure 4). These associations are plausible given that changes in NFAT have been shown to alter the metastatic and growth properties of breast cancer cells [42–44], and given the important role NF1 and IRF play in breast cancer generally [52–56].
It could be argued that both IR- and cell-adhesion pathways are differentially activated across tumours merely as a result of lymphocytic or stromal contamination, respectively. However, microarray studies profiling breast cancer cell lines (BCL) have shown that genes associated with IR- and cell-adhesion functions are also differentially regulated across cell lines [25,57]. In particular, it was shown that genes related to cell-adhesion functions were overexpressed in ER− compared with ER+ cell-lines [57]. While the study in [57] did not explicitly mention the differential expression of immune response genes, we verified, by applying ICA to this set of only 31 breast cancer cell lines (BCL), that an independent component enriched for immune response genes was present and that it correlated with the ER status of the cell lines (Figure S6). This provided further validation of the link between differential regulation of immune response pathways with the ER status of breast cancer cells, while also simultaneously confirming that the differential regulation of these genes across the tumour set is not necessarily related to varying degrees of lymphocytic infiltration.
Generally, we found that genes selected in the same independent component showed a relatively strong co-expression pattern (Figure 5B). It follows that ICA components can often be given a biological interpretation similar to that of clusters inferred through, say, hierarchical or k-means clustering. To illustrate this with another example, we considered the case of estrogen signalling and ER status. This showed that clustering over the genes selected in an IC that was associated with estrogen signalling and ER status yielded similar heatmaps for the measured expression matrix and the IC submatrix, and, furthermore, for both heatmaps the association with the phenotype was evident (Figure S7).
On the other hand, ICA also found “non-trivial” associations, such as the association of the EMT pathway with grade (Figure 6A), where the functional relationship of the genes in the same pathway was not as evident from the gene expression matrix (Figure 6B). Given that genes are shared by multiple pathways, the functional relationship of the genes may indeed not manifest itself as a strong co-expression pattern. Thus, it would appear that ICA, through the statistical independence criterion, which effectively uses non-linear correlation measures (as opposed to mere linear co-expression) to determine common functionality, is able to capture non-trivial functional relationships of genes in a common pathway, in spite of the fact that these genes may not exhibit strong co-expression.
In summary, this work is the first to our knowledge to validate the ICA paradigm using a framework based on existing pathway-knowledge and regulatory-module databases. Moreover, it confirms the added value of ICA over PCA and clustering-based methods in identifying novel associations of known pathways and regulatory modules with breast cancer phenotypes. Our results also indicate that larger datasets may be required before a more complete understanding of the ICA model in the gene expression context can be obtained, as well as to understand to what degree ICA can help in defining a more clinically relevant molecular taxonomy of breast cancer.
To test the ICA model, we first generated a comprehensive list of pathways, most of which are known to be directly or indirectly involved in cancer biology. To compile this list, we used the Molecular Signatures Database MSigDB [24], which included 522 distinct pathways curated from the literature and from other databases such as KEGG (http://www.genome.jp/kegg/) and CGAP (http://cgap.nci.nih.gov/). We augmented this list with known oncogenic pathways recently derived in [35] and cancer-signalling pathways from NETPATH (http://www.netpath.org), yielding a total of 536 pathways. Not all of these pathways had sufficient representation across the six major studies. Specifically, out of these 536 pathways, 277 had at least five genes represented on each of the six microarray platforms (probes on specific microarrays were also filtered based on quality, which explains why there wasn't a higher percentage of pathway gene lists with sufficient representation). The full list of pathways used are summarised in Table S1 in terms of their representation on each of the arrays.
We used the sequence-derived regulatory motifs in human promoters and 3′ UTRs from [36]. For each such motif we defined the associated regulatory gene module as the set of genes having this motif in their promoters or 3′ UTR, as provided in MSigDB [24]. The selected feature sets of the inferred components were tested for enrichment of regulatory modules, which provided us with putative links between components and the transcription factors that bind to these motifs.
Briefly, we review the ICA model [58] as used in this work. Let Xgs denote the normalised data matrix of expression values where g = 1,. . .,n denotes the genes and s = 1,. . . , N denotes the samples. We assume further that X has been normalised so that the mean of each column of X is zero. Then ICA (or PCA) produces an approximate decomposition of the matrix X into the product of two matrices S (the “source” matrix) and A (the “mixing” matrix):
where K ≤ min{n,N} is the number of components to be computed. When K is strictly smaller than min{n,N}, it is in general impossible to pick S and A such that the error matrix vanishes. Therefore, the algorithms aim at making E as small as possible, usually in the least squares sense. This condition on E still leaves much leeway to select the matrices S and A.
PCA consists of identifying an orthonormal matrix S (i.e.,
for all k ≠ k′, and
for all k) and an orthogonal matrix A (i.e.,
for all k ≠ k′) so that the data covariance matrix is diagonalised. In comparison, most ICA algorithms start with a preprocessing step, in which the means of the columns of X are set to zero, followed by a PCA. Thus, as with PCA itself, this first requires an orthonormal matrix S′ and an orthogonal matrix A′ such that X = S′A′ + E′. It should be noted that orthonormality of S′ implies a sample covariance between the columns of S′ that equals zero. The ICA step per se amounts to then finding a transformation W of S′,
such that the columns of S are “as independent as possible”. Most ICA methods consider that the zero covariance property of S′ is compatible with this goal, hence they preserve this property in S′ by restricting W to the set of K × K orthogonal transformations. The ICA algorithms, thus, search for an orthogonal matrix W that maximises the statistical independence of the columns of S′. The mixing matrix finally equals
and the error E is identical to E′.
A quantitative measure of independence between measurements of random variables, in this case the columns of S′, is provided by a contrast function. The only requirement on the contrast function is that it goes with probability one to a prescribed extremum (usually zero) if and only if the random variables are statistically independent and as the number of measurements n goes to infinity. This leaves many possibilities for the contrast function, leading to a variety of ICA algorithms, which may also differ in the numerical algorithm used for the optimisation procedure. Here, we considered four different ICA algorithms, which are described in more detail in Protocol S1: the JADE (or “JointDiag”) algorithm [59], the “FastICA” algorithm [31], the “KernelICA” algorithm [32], and the “RADICAL” algorithm [33].
The estimation of the number of sources in ICA is a hard outstanding problem. While approaches to estimating the number of sources exist, for example, the Bayesian Information Criterion (BIC) in a maximum likelihood framework [34] or using the evidence bound in a variational Bayesian approach [60–62], we decided to infer the same number of components for each algorithm. There are two reasons for this. First, because of the still relatively small sample sizes of microarray experiments, estimating the correct number of components is difficult. It has therefore been conventional to use a fixed number of components [15,16]. Second, since the aim with our work was to provide a comparison between the PCA-derived components and those derived from ICA algorithms, using the same number of components for each algorithm facilitated such a comparison.
For each component that is inferred, ICA and PCA yield a corresponding list of genes and signed weights. The ICA model is based on the premise that ICA modes selectively pick out a small percentage of genes (∼1%) that are strongly activated or repressed in response to the deregulation of a particular pathway, while the great majority of genes are unaffected. Mathematically, the distribution of inferred weights must be non-gaussian, and in the gene expression context they must be supergaussian (or leptokurtic), since most of the genes in a mode belong to a gaussian component centred at zero. Thus, to find the genes that are differentially activated, it is conventional to set a threshold, typically two or three standard deviations from the mean, and to pick out those genes whose absolute weights exceed this threshold. Although a more elegant method for determining an appropriate threshold, and which is based on measuring the deviation from normality of the weight distributions, is available [20], this method is not applicable to PCA components where deviation from normality is not a requirement. Hence, since the main aim was to provide an objective comparison of ICA with PCA, we decided to use the threshold method as this method would yield approximately the same number of features per component for PCA and ICA. To focus on the pathways that dominate an ICA mode, we used the more stringent threshold of 3 sigma on either side from the zero mean, which picks out the 0.2% of genes in the tails of the signed weight distributions. Robustness of our results to the choice of threshold was evaluated by considering less stringent thresholds of 2 and 2.5 sigma. Thus, for each inferred ICA mode or principal component, we obtained a list of selected features and associated signed weights. This resulted in a mean number of approximately 160 features (3 sigma threshold) selected per component, although this number varied significantly depending on study. Importantly though, while ICA algorithms did generally capture more features per component than PCA (as we would expect since ICA algorithms seek supergaussian components), the difference in selected feature numbers was not significant (Table S2).
To provide an objective comparison of ICA/PCA with clustering methods, the clustering step was preceded by a feature selection step which ensured that all methods selected an approximately equal number of genes. This feature selection step was performed in two different ways. For a given cohort, genes were first ranked according to their expression variance across samples. In the most-variable-genes (MVG) method, the top 15% variable genes were then selected. In the second method, we used all the distinct genes selected through PCA using the 3 sigma threshold. Since this number is less than the total number (i.e., not distinct) of features selected from the PCA components, the remaining distinct genes were selected from the ranked MVG list. Having selected the features via one of the above methods, clustering was then performed using a robust version of k-means clustering, known as partitioning around medoids [63], where k was set to 10 in order to match the number of components inferred by ICA and PCA. Thus, PCA-KM selected the same number of total features as PCA and approximately the same number as ICA, while the threshold of 15% was chosen to ensure that MVG-KM did not select less total number of features than ICA or PCA (Table S2).
For the genes selected in a ICA or PCA component or for the genes in a given cluster derived from either MVG-KM or PCA-KM, enrichment analysis evaluates whether there is statistically significant enrichment of genes from a given pathway or regulatory module. For a given study s and inference method m, let i denote a given inferred component (or cluster) and p a pathway (or regulatory module). In what follows, we also use “component” to refer to the clusters of the KM-algorithms, and also use “pathway” to refer to the regulatory modules. Let NS denote the number of genes on the array of data set s, and nsp denote the number of genes from pathway p on that same array. Similarly, let dsmi denote the number of genes selected in component i, and tsmi the number of genes from pathway p among the selected dsmi features. Then, under the null hypothesis, where the selected genes are chosen randomly, the number tsmi follows a hypergeometric distribution. Specifically, the probability distribution is
and a p-value can be readily computed as P(t > tsmi). Note that Vandermonde's identity implies that the probability distribution is correctly normalised. Thus, for a given study and method, we can compute a p-value for each component-pathway pair that evaluates how enriched the component is in terms of genes from that particular pathway. To correct for multiple testing, we used the Benjamini-Hochberg procedure [64] and called a component–pathway pair association significant if the p-value was less than a threshold determined by setting the false discovery rate (FDR) equal to 0.05.
|
10.1371/journal.pgen.1003231 | Delineating a Conserved Genetic Cassette Promoting Outgrowth of Body Appendages | The acquisition of the external genitalia allowed mammals to cope with terrestrial-specific reproductive needs for internal fertilization, and thus it represents one of the most fundamental steps in evolution towards a life on land. How genitalia evolved remains obscure, and the key to understanding this process may lie in the developmental genetics that underpins the early establishment of the genital primordium, the genital tubercle (GT). Development of the GT is similar to that of the limb, which requires precise regulation from a distal signaling epithelium. However, whether outgrowth of the GT and limbs is mediated by common instructive signals remains unknown. In this study, we used comprehensive genetic approaches to interrogate the signaling cascade involved in GT formation in comparison with limb formation. We demonstrate that the FGF ligand responsible for GT development is FGF8 expressed in the cloacal endoderm. We further showed that forced Fgf8 expression can rescue limb and GT reduction in embryos deficient in WNT signaling activity. Our studies show that the regulation of Fgf8 by the canonical WNT signaling pathway is mediated in part by the transcription factor SP8. Sp8 mutants elicit appendage defects mirroring WNT and FGF mutants, and abolishing Sp8 attenuates ectopic appendage development caused by a gain-of-function β-catenin mutation. These observations indicate that a conserved WNT-SP8-FGF8 genetic cassette is employed by both appendages for promoting outgrowth, and suggest a deep homology shared by the limb and external genitalia.
| Mammalian limbs and external genitalia are body appendages specialized for locomotion and internal fertilization, respectively. Despite their marked anatomical and functional differences, development of the limb and external genitalia appears to involve similar genetic controls, and some have suggested that regulatory mechanisms common to both might be evolutionarily linked. One essential aspect for appendage development is the establishment and maintenance of a separated proximodistal developmental axis apart from the main body axis, which is often instructed by a distal signaling epithelium. Herein, we adopted comprehensive mouse genetic approaches to investigate regulatory mechanisms underlying the distal signaling center in the limb and the GT, and uncovered a conserved genetic cassette that is utilized by both paired and unpaired appendages to establish a distal signaling center in the epithelium that mediates subsequent proximodistal outgrowth. Our results further suggested that the evolution of the external genital organ involved co-option of the same genetic program underpinning limb development.
| Development of the external genitalia is a crucial aspect of mammalian evolution that enables internal fertilization, a pivotal step towards land invasion. All therian mammals including metatherians develop external genitalia around their urogenital outlets. In mice, development of the embryonic anlage of external genitalia, the genital tubercle (GT), is identical in both sexes before androgen-mediated penile masculinization which occurs around embryonic day 16. The early androgen-independent phase of GT development is achieved through coordinated growth and patterning of cloacal endoderm-derived urethral epithelium (UE), mesoderm-derived para-cloacal mesenchyme (PCM) and the ventral ectoderm, which results in a cone-like structure with a ventral-medial positioned urethra surrounded by GT mesenchyme within an ectodermal epithelial capsule.
The development of the GT as an unpaired body appendage, is often compared to that of the paired-type appendages, the limbs [1], [2]. Despite their anatomical and functional differences, the morphogenesis of both structures appears to involve similar genetic controls. Regulatory genes/pathways including canonical WNT signaling [3], [4], HH signaling [5]–[7], BMP signaling [8], [9] and Hox genes [10]–[14] are essential for the development of both appendages. Some have suggested that the regulatory mechanisms common to both might be evolutionarily linked [11], [15], [16].
The first step in appendage outgrowth is the establishment of an independent proximodistal developmental axis apart from the main body axis. This process requires precise regulation from instructive signals, which often come from a distal signaling center. In addition to promoting and maintaining outgrowth, these signals also provide directional information that determines the orientation and shape of future structures. Moreover, genes required for subsequent patterning and differentiation are often regulated by the distal signaling center. For example, during limb development, the initiation and continuous outgrowth of the limb bud rely on growth factors secreted from a strip of ectodermal epithelium, termed the apical ectodermal ridge (AER), positioned at the distal edge of limb bud (Figure 1A and 1E). Fibroblast Growth Factors (FGFs) are crucial signals provided by the AER, as FGF-soaked beads can replace the AER to induce limb outgrowth [17]–[20]. Furthermore, the AER FGF signals are obligatory to maintain a positive feedback loop involving SHH and Gremlin that coordinates patterning and growth of the limb [21]–[23].
The early GT is built by two mesenchymal swellings at either side of the cloacal membrane, which is later joined by a third outgrowth anterior to the cloacal membrane (Figure 1B–1D). Unlike limb, subsequent outgrowth of the GT has to accommodate a continuous extension of the cloacal endoderm, which forms the epithelial lining of the future urethral tube. This unique pattern of GT development suggests that the centrally located distal cloacal endoderm (later the distal urethral epithelium or dUE, marked in red in Figure 1B–1D, also shown by Fgf8 in situ in Figure 1F–1H) is a strategic place for GT outgrowth. Consistently, an earlier study demonstrated that Fgf8 is expressed in a strip of cells located at the distal-most part of the cloacal endoderm right below the ventral ectoderm (Figure 1B inset). Subsequent functional analyses revealed that physically removing Fgf8-expressing cells or application of neutralizing FGF8 antibody could abolish GT growth in organ culture, and this growth inhibition could be reversed by adding FGF8-soaked beads [24].
Along with these studies, we uncovered that activity of the canonical WNT-β-catenin pathway is restricted to the Fgf8-expressing distal cloacal endoderm and later the dUE [4]. We found that abolishing β-catenin (β-Cat) in the cloacal endoderm caused GT agenesis/reduction, whereas ectopic activation of the same pathway resulted in GT over-development. Interestingly, the site and level of WNT signaling activity positively correlated with Fgf8 expression and the extent of outgrowth in both limbs and GT. These findings illustrated a parallel between dUE and AER signaling during appendage outgrowth. However, the exact mechanisms and functional relevance for this WNT-Fgf8 regulation remain to be elucidated.
Recently, two independent studies reported a surprising observation that abolishing Fgf8 [25], [26], or simultaneously abolishing Fgf4 and Fgf8 [26], in the cloacal endoderm does not affect GT outgrowth or GT-specific gene expression. These results questioned the relevance of FGF signals in external genitalia development and challenged the view that GT formation is organized and maintained by the dUE, which further suggested that the mechanisms underpinning limb and GT outgrowth are indeed divergent [2].
Herein, we adopted comprehensive genetic approaches to address the inductive signals in GT development, in comparison with that of the limb. We sought to define the role for FGF signaling and dUE-expressed Fgf8 in the GT, and explore mechanisms upstream of FGF activation in the dUE of the GT and AER of the limb. Results from our analyses revealed a remarkably conserved Wnt-Sp8-Fgf8 genetic circuitry that is crucial for proximodistal outgrowth of both paired- and unpaired-type appendages in mice.
Previous studies demonstrated that inactivating one or two FGF ligands did not affect genital development [25], [26]. We reasoned that the extensive genetic redundancy among FGF ligands might have undermined the power of these experiments. Therefore, we sought to re-evaluate the function of FGF signaling in GT development by conditionally abolishing FGF receptors. Fgfr1 and Fgfr2 are the only FGF receptors expressed in the developing GT [27]. Both Fgfr1 and Fgfr2IIIc are expressed in the para-cloacal mesenchyme (PCM) during GT outgrowth [27], and Fgfr2IIIb is expressed in both the PCM and the urethral epithelium (UE) [27], [28]. To abolish all FGF responsiveness in the developing GT, we employed an Msx2rtTA;tetO-Cre system which enables doxycycline (Dox)-inducible gene ablation in both the PCM and the UE [29], [30]. Dox was given to pregnant females on embryonic day (E) 9.5 and E10.5 to induce Cre-mediated recombination of the Fgfr1f/f [31] and Fgfr2f/f [32] alleles. The phenotypes of Msx2rtTA;tetO-Cre;Fgfr1f/f;Fgfr2f/f mutant embryos [hereafter referred to as GT-Fgfr1/2-double conditional knockouts (cKO), or dcKO] were analyzed by scanning electron microscopy (SEM). GT-Fgfr1;r2-dcKO genital tubercles were underdeveloped compared to their littermate controls at all stages examined starting from E11.5, when a clear tubercle structure could first be detected (Figure 2B). At E12.5, the dcKO GTs showed a clear deficiency in proximodistal outgrowth, as they were much shorter than the controls (data not shown). At E15.5, the mutant GTs were smaller in size, deformed (Figure 2D) and lacked the characteristic mesenchymal patterning present in controls (Figure S1A–S1B). To further explore the cellular basis for the reduction in GT size, we analyzed proliferation and cell-death in the GT of these mutants. We performed phospho-histone H3 (PHH3) staining on E11.0 coronal GT sections and counted the number of PHH3+ cells in a fixed area. A 28% reduction in PHH3-positive cell number in the genital mesenchyme of the dcKOs was observed (Figure S1C-S1E, n≥10, p = 0.0017). We also performed TUNEL analyses but did not observe any differences in the number of apoptotic cells between control and dcKO mutant GTs (data not shown).
An examination of genes known to mediate genital tubercle initiation revealed alterations in normal gene expression patterns as early as E11.5. Bmp4, Wnt5a and Msx1 were expressed in the PCM in control GTs (Figure 2E and 2G, and Figure S1F), and their expression was barely detectable in dcKO GTs (Figure 2F and 2H, and Figure S1G). Msx2 was expressed in both the PCM and UE in controls (Figure S1H). In the GT-Fgfr1;r2-dcKOs, Msx2 expression was absent in the PCM and downregulated in the UE (Figure S1I). Moreover, UE expression of Shh was also downregulated in the dcKOs (Figure 2I–2H). In contrast, dUE-Fgf8 expression remained unchanged in the GT-Fgfr1;r2-dcKOs (Figure 2K–2L), suggesting that maintenance of Fgf8 expression is independent of FGFR1 and FGFR2 during GT development. Collectively, these data demonstrate that FGF signaling is obligatory for promoting proximodistal GT outgrowth and for maintaining genital-specific gene expression.
The Msx2rtTA;tetO-Cre system mediates gene deletion in both the cloacal endoderm as well as the PCM. Therefore, it is not clear whether changes observed in the aforementioned Fgfr1;r2-dcKOs were direct results of compromised FGF responsiveness in the PCM, or secondary to loss of FGF receptors in the cloacal endoderm. Thus, we tested the requirement for FGF responsiveness in these two compartments by using a previously characterized endoderm-specific ShhEGFPCre allele [4], [33], and a PCM-specific Dermo1Cre allele [4], [32], respectively. ShhEGFPCre/+;Fgfr1f/f;Fgfr2f/f (UE-Fgfr1;r2-dcKO) and Dermo1Cre/+;Fgfr1f/f;Fgfr2f/f (PCM-Fgfr1;r2-dcKO) mutants were generated and their GTs analyzed. Intriguingly, GTs from UE-Fgfr1;r2-dcKOs did not exhibit any morphological abnormalities in the early outgrowth phase, and their size and shape were comparable to stage-matched controls (Figure S2A–S2D). Histological analysis revealed normal patterning of the genital mesenchyme in UE-Fgfr1;r2-dcKOs embryos (Figure S2E and S2F). The only phenotype observed in these mutants was the abnormal maturation of urethral epithelium similar to what was observed in Fgfr2IIIb mutants [28], which will be discussed in a separate manuscript. Furthermore, regulatory genes including Msx2, Bmp4, Wnt5a and Fgf8, were also properly expressed in these UE-Fgfr1;r2-dcKO embryos (Figure S2G–S2J).
In contrast, the GTs of PCM-Fgfr1;r2-dcKO were clearly smaller than their littermate controls (Figure S3A–S3B). Further analyses revealed that the distal GT mesenchymal expression of P-ERK1/2, a previously established FGF target gene [26], was also downregulated in these mutants (Figure S3C–S3D). In addition, PHH3 staining on E11.5 embryos revealed a 20% reduction in mitotic figure number in the PCM-Fgfr1;r2-dcKOs (Figure S3E–S3G). Collectively, these data indicate that the main target for FGF signaling during GT outgrowth is the PCM. A similar requirement for FGF signaling in the limb mesenchyme has been described previously [34].
Fgf8 is normally expressed in the distal-posterior cloacal endoderm at E10.5, and then in the dUE through E11.5–E14.5. We sought to determine whether the PCM could respond to dUE-expressed Fgf8 in vivo. We generated a conditional Fgf8 overexpressor mouse line by knocking the Fgf8 full-length cDNA (Accession: BC048734) into the ubiquitously expressed Rosa26 locus, preceded by a floxed transcriptional stop cassette (R26Fgf8). This design allows ectopic Fgf8 expression upon Cre-mediated recombination (Figure 3A). We used a tamoxifen (Tm)-inducible ShhCreERT2 allele [4], [33] and generated ShhCreERT2/+;R26Fgf8 gain of function (GOF) mutants (UE-R26Fgf8-GOF), to achieve UE-specific Fgf8 overexpression upon Tm treatment at E10.5. The Cre expression domain of this ShhCreERT2 allele recapitulates endogenous Shh expression, which includes all cloacal endodermal cells. Eight hours after Tm administration, we noted a clear upregulation (arrowhead in Figure 3C) and an anterior expansion (arrow in Figure 3C) of Fgf8 expression in the cloacal endoderm evidenced by whole-mount in situ hybridization. The ectopic expression in the anterior cloacal endoderm (arrow in Figure 3C and 3E) was the result of ectopic Fgf8 expression from the R26Fgf8 allele. Notably, a concurrently augmented Bmp4 expression was evident in the PCM (insets in Figure 3B and 3C). Sixteen hours after the initial tamoxifen injection, we observed a 19% increase in mitotic index in the PCM (Figure S4A–S4C, n = 10, p = 0.009). Consistently, the mutant GTs were larger than controls at E12.5 (Figure S4D–S4E) and E14.5 (Figure 3H–3I), respectively. These findings clearly demonstrate that endodermally expressed FGF8 can mediate gene expression and promote cell proliferation in the neighboring PCM.
Interestingly, we noted that endogenous Fgf8 expression was differentially regulated in the UE-R26Fgf8-GOF mutants. Sixteen hours after Tm treatment, we observed a distinct downregulation of Fgf8 expression in the GOF mutant dUE (arrowhead in Figure 3E) that persisted 24 (Figure 3G) and 48 hours after treatment (data not shown). It is also noteworthy that transcription from the R26 locus was much weaker than that from the endogenous Fgf8 locus, evidenced by the faint signal in the anterior cloacal endoderm (indicated by arrows in Figure 3C and 3E). Collectively, these findings demonstrate that the developing GT is sensitive to changes in FGF dosage, and a feedback loop is deployed to ensure proper signaling activity when misregulation occurs.
To compare the function of FGF8 in the limb and GT, we mated the R26Fgf8 allele with a transgenic Msx2-Cre line [35], which confers Cre expression in the forming and mature AER and the ventral limb ectoderm. As expected, the AER-R26Fgf8-GOF mutants exhibited excessive limb growth and developed extra digits (asterisk in Figure 3J), an enlarged calcaneus bone (arrowhead in Figure 3K), and ectopic skeletal elements (arrows in Figure 3J–3K, and Figure S4G) in both forelimbs and hindlimbs. These overgrowth phenotypes are similar but more severe than what has been observed in the AER-Fgf4-GOF embryos [36], and further support the concept that FGF8 plays a pivotal role in promoting the outgrowth of both appendages.
Our previous work has shown that the WNT-β-catenin signaling pathway and Fgf8 expression are tightly coupled in the distal signaling centers both in the limb and in the GT [4]. The above finding that Fgf8 was repressed by its own overexpression was in sharp contrast to our previous observation in the UE-β-Cat-GOF mutants where ectopic up-regulation Fgf8 was sustained in the UE [4]. This suggested that the canonical WNT pathway plays a key role in controlling the Fgf8 auto-regulatory feedback loop. Together, these observations suggest that the WNT-Fgf8 regulatory relationship is essential for appendage formation, and prompted us to test whether loss of Fgf8 was the critical event causing appendage reduction in embryos deficient in β-Cat in the AER and the UE. Therefore, we attempted to restore Fgf8 expression in β-Cat loss of function (LOF) embryos and analyze its effect on appendage formation. We generated ShhEGFPCre/+; β-Catf/f;R26Fgf8/+ (UE-β-Cat-LOF;R26Fgf8) mutants and analyzed their GT development. In contrast to absence of the GT in the UE-β-Cat-LOF embryos (Figure 4B), a cone-shaped tubercle structure was readily discernible in compound mutants carrying the R26Fgf8 allele (Figure 4A–4C). To determine whether this rescued structure bears GT characteristics, we performed both histological and gene expression analyses. Hematoxylin and Eosin (H&E) stained E12.0 transverse GT sections showed that the morphology of the rescued GT closely resembled that of the controls with the urethra properly positioned at the ventral side of the GT (Figure 4D). In addition, Hoxa13 and Hoxd13, both markers of the GT, were expressed in the rescued genital tubercles (Figure 4E and 4F). Altogether, these data indicate that restoring FGF8 rescued GT agenesis caused by β-catenin deficiency. It is noteworthy that this rescue of β-Cat-cKO by FGF8 is confined to the GT, as other caudal malformations observed in the β-Cat-cKO including failed cloaca septation and defective tail formation, were still present in the UE-β-Cat-LOF;R26Fgf8 embryos (data not shown).
To test whether forced Fgf8 expression can also rescue limb deficiency caused by β-Cat ablation, we generated Msx2-Cre; β-Catf/f (AER-β-Cat-LOF) and Msx2-Cre; β-Catf/f;R26Fgf8/+ (AER-β-Cat-LOF;R26Fgf8) mutants. Consistent with a previous investigation [3], all autopod elements, radius, and distal two-thirds of the ulna were missing from the forelimbs of AER-β-Cat-LOFs (Figure 4I), whereas the humerus was thinner and lacked the deltoid tuberosity (Figure 4I compared to G, arrow). In contrast, the LOF embryos with the R26Fgf8 allele developed normal humeri with the deltoid tuberosity (Figure 4K, arrow). The radius was evident and the ulna was longer and thicker (Figure 4K). Moreover, several small pieces of alcian blue-stained cartilage were observed distal to the ulna, indicating the presence of autopod rudiments (arrowhead and inset in Figure 4K). The hindlimbs of the AER-β-Cat-LOFs were largely absent except for a small remnant of the pelvic girdle (Figure 4J), whereas the R26Fgf8-expressing β-Cat-LOF embryos developed near normal pelvic girdles and femurs (Figure 4L) along with one or two ectopic cartilages (asterisk in Figure 4L). These phenotypes were consistently observed in all AER-β-Cat-LOF;R26Fgf8 embryos (n = 10), and indicated that exogenously supplying FGF8 can partially restore distal limb structures lost in the AER-β-Cat-LOF embryos. Collectively, these data suggest that FGF8 can promote outgrowth of both the GT and the limb in the absence of canonical epithelial WNT activity, suggesting that it functions as a downstream effector of WNT signaling during limb and GT outgrowth.
The positive feedback loop involving FGFs and SHH plays a critical role in appendage outgrowth, as evidenced by the down-regulation in Fgf expression in both the AER [21], [22], [35], [37] and the dUE [5], [6] of Shh-KOs, which display reductions in both appendages. We next examined whether forced Fgf8 expression could also rescue the severe appendage deficiencies caused by the absence of SHH. We generated ShhCreERT2/CreERT2;R26Fgf8/+ (Shh-KO;UE-R26Fgf8) embryos, which allowed us to induce Fgf8 expression in the UE of Shh null mutants (ShhCreERT2 allele is also a null allele). However, we detected neither tubercle formation (Figure S5B), nor Hoxa13 or Hoxd13 expression (Figure S5C and S5D) in the cloacal region of these compound mutants at E12.5, 48 hours after Tm treatment. We also generated Msx2-Cre; R26Fgf8/+; ShhCreERT2/CreERT2 (Shh-KO;AER-R26Fgf8) mutants, to test whether sustaining Fgf8 expression in the AER of Shh mutants can restore limb development. We carefully analyzed four Shh-KO embryos carrying both Msx2-Cre and R26Fgf8 alleles, and found no evidence of more advanced limb development (Figure S5G and S5H), compared to thirteen Shh-KO mutants without R26Fgf8 alleles (Figure S5E and S5F).Together, these results indicated that although Shh and Fgf8 expression are interdependent during appendage outgrowth, their function in promoting appendage outgrowth is independent and non-redundant.
We next analyzed GT gene expression in UE-β-Cat-LOF; R26Fgf8 mutants to further interrogate the genetic networks underlying GT development and identify downstream targets of dUE signaling. Fgf8 expression was detected in the distal cloacal endoderm in controls (Figure 5A), but was absent in UE-β-Cat-LOF mutant (Figure 5B) at E10.5. The R26Fgf8-expressing LOF embryos, on the other hand, showed very weak Fgf8 expression throughout the cloacal endoderm (arrow in Figure 5C). This low-expression was consistent with what we observed in the UE- R26Fgf8-GOF embryos, suggesting that these Fgf8 transcripts were transcribed from the R26 locus. Expression of Bmp4 and Wnt5a was normally detected in the PCM (Figure 5D and Figure S6A), absent in the UE-β-Cat-LOFs (Figure 5E and Figure S6B), and partially restored by ectopically supplying Fgf8 expression from the R26Fgf8 allele (Figure 5F and Figure S6C). The SHH pathway was also compromised in the β-Cat-LOF mutants. Shh (Figure 5H) and Ptch1 expressions (Figure S6E) were markedly decreased in the cloacal endoderm and the PCM, respectively. On the other hand, in the LOF embryos with the R26Fgf8 allele, the expression of Shh was partially restored in the UE (Figure 5I) and the PCM expressed Ptch1 restored to a level comparable to controls (Figure 5F).
Combined, these data indicate that most genes differentially regulated in the UE-β-Cat-LOF mutants were responsive to FGF8 induction, suggesting that their expression is controlled by the dUE-FGF signals. However, we found that Sp8, a transcription factor normally expressed in the cloacal endoderm (Figure 5J), was lost in the UE of UE-β-Cat-LOF mutant (Figure 5K), and did not respond to FGF8 supplementation (Figure 5L).
Sp8 is expressed throughout the cloacal endoderm and later in the UE (Figure 6A), overlapping with the Fgf8-expressing dUE during GT development. Previous studies have implicated SP8 in the transcriptional regulation of Fgf8 expression in the mouse commissural plate [38] and the chick limb ectoderm [39]. These findings prompted us to explore its role in mediating the WNT-Fgf8 pathway. We first examined Sp8 expression in the UE-β-Cat-LOF and GOF-mutants (for GOF analyses, we used a previously established β-CatΔex3 allele which produces stabilized β-catenin upon Cre-mediated recombination [40]). We found that Sp8 expression was reduced in the β-Cat-LOF UE (Figure 6B) and increased in the β-CatΔex3 GOF UE (Figure 6C). Similarly, robust Sp8 expression was also detected in the AER (Figure 6D), markedly reduced in the AER- β-Cat-LOF mutants (Figure 6E), and upregulated in the β-CatΔex3 GOF mutants where β-catenin activity was ectopically augmented in the limb ectoderm (Figure 6F, inset showing ventral view of a hindlimb). These data indicated that Sp8 is downstream of Wnt-β-catenin signaling in the UE and the limb ectoderm.
Next, we examined the GT phenotype of Sp8-null (KO) mutants. We found that 14/36 mutant embryos examined between E12.5–E15.5 exhibited GT agenesis (Figure 6H), while the rest demonstrated a range of GT defects including deformation, hypoplasia and proximal hypospadias (data not shown). Fgf8 expression was completely absent in the cloacal endoderm in all Sp8 KOs examined at E11.5 (Figure 6J, n = 9). Notably, these embryos also exhibited other caudal malformations such as deformed perineum, anal channel and tails, raising the concern whether the observed GT defects were secondary to earlier cloacal or neural tube malformations. Thus, we employed a conditional Sp8 null allele with the ShhCreERT2 line to generate UE-Sp8-LOF mutants, and induced Sp8 deletion by Tm treatment around the time of GT initiation (E10.5). GTs from the UE-Sp8-LOF mutants were smaller than their age-matched controls, especially at the distal tip (Figure 6L). We also found a clear reduction in Fgf8 expression in these mutants by in situ analyses (inset in Figure 6L). Consistently, the expression domains of Wnt5a (Figure S7B) and Msx2 (Figure S7D), both maintained by FGF8 from the neighboring dUE, were reduced.
Limb truncation, attributed to a failure to form the AER and consequently loss of Fgf8 expression, has been previously described in Sp8−/− embryos [41]. We generated AER-Sp8-LOFs using the Msx2-Cre line, and these mutants also showed a defect in limb outgrowth as evidenced by a loss of distal structures (Figure 6Q–6S). The stylopod and zeugopod developed normally in the forelimbs, while typically only one abnormal digit formed in these mutants (Figure 6Q–6R). The tibia and fibula were either missing or severely truncated, and no autopod was observed in the hindlimbs (Figure 6S). Notably, these limb defects can also be partially rescued by over expression of Fgf8. Skeleton preparation of E18.5 embryos showed that the AER-Sp8-LOF embryos carrying R26Fgf8 allele developed three digits in the forelimb (Figure 6U), and full-length tibia and fibula along with several irregular autopod elements in the hindlimbs (Figure 6V–6W). Altogether, these data indicated that Sp8 is required to maintain Fgf8 expression and appendage outgrowth in the distal signaling center of both the limb and GT.
To test whether Sp8 is in the same genetic pathway with β-catenin and Fgf8, we sought to determine whether SP8 mediates WNT-induced Fgf8 expression in the dUE and AER. We generated compound mutants using the β-CatΔex3 allele, which we have previously shown to activate Fgf8 expression in both the UE and the limb ectoderm, together with the floxed Sp8 allele and the corresponding Cre lines. We compared Fgf8 expression in embryos with one β-CatΔex3 allele and different numbers of functional Sp8 alleles by real-time RT-PCR and in situ hybridization. We found that the level of WNT-induced Fgf8 expression in the UE positively correlated with the number of functioning Sp8 alleles (Figure 7A–7E). Removing one wild type Sp8 allele caused a twofold reduction in Fgf8 expression, whereas ablating both wild type Sp8 alleles reduced Fgf8 expression by more than three-fold (Figure 7E). These results were further verified by Fgf8 in situ hybridization (Figure 7A–7D).
Similarly, deleting both Sp8 alleles abolished the ectopic Fgf8 expansion in the limb ectoderm of Msx2-Cre; β-CatΔex3/+ (AER-β-Cat-GOF) embryos (Figure 7F–7H), and consequently attenuated the polysyndactyly phenotype caused by constitutively active canonical WNT signaling (Figure 7I–7O). With two wild type Sp8 alleles, AER-β-Cat-GOF mutants developed an average of 6.8 digits in the forelimbs, and 6.1 digits in the hindlimbs (Figure 7J and 7M). In thirty percent of the embryos, ectopic limb in the flank ectoderm and ventral ectoderm were detected (Figure S8A and S8B). In comparison, AER-β-Cat-GOF;Sp8-null mutants only developed 4.9 digits in the forelimbs and 4.2 digits in the hindlimbs (Figure 7K and 7N, 28% and 31% reduction compared to AER-β-Cat GOFs, respectively). In addition, no extra limbs were observed at ectopic locations. Collectively, these results indicate that SP8 is responsible, at least in part, for the WNT-induced activation of Fgf8 expression during appendage outgrowth.
To test whether augmented Sp8 expression alone can induce Fgf8 overexpression, we generated a R26Sp8 allele to conditionally overexpress Sp8 using the same strategy as for the R26Fgf8 line (Figure 8A). Mice carrying ShhEGFPCre or Msx2-Cre alleles were used to generate corresponding UE-R26Sp8-GOF and AER-R26Sp8-GOF embryos. Overexpression of Sp8 in the UE and AER was confirmed by in situ hybridization (Figure 8C, and data not shown). However, unlike Fgf8- or β-Cat-GOF embryos, Sp8-GOF embryos showed normal development in both appendages (Figure 8E and 8I). Fgf8 expression in the dUE and the AER of the corresponding Sp8-GOF mutant was also comparable to their wild type counterparts (Figure S9A–S9D).
We also crossed the R26Sp8 allele into the UE- or AER-β-Cat-LOF mutants to test whether forced Sp8 expression can bypass epithelial β-catenin to induce Fgf8 expression and initiate/maintain appendage outgrowth. We carefully analyzed six compound mutants and did not observe phenotypic rescue in either the GT (Figure 8F) or the limb (Figure 8J–8M). In addition, no Fgf8 induction was detected in the dUE of the corresponding β-Cat-LOF mutants carrying R26Sp8 allele (Figure 8G). All of these results indicate that SP8 by itself is insufficient to activate Fgf8 expression, and suggest that SP8 is a facilitator of WNT-mediated Fgf8 activation during appendage formation.
In this study, we investigated the molecular cascade that regulates distal signaling centers in appendage development. Our study elicited a conservative genetic circuitry involving WNT-β-catenin signaling, the transcription factor Sp8, and the growth factor Fgf8, that underpins proximodistal outgrowth of limbs and external genitalia.
Our work provides in vivo evidence that FGF signaling is indispensible for early GT outgrowth. These findings are consistent with results from organ culture studies that inhibition of FGF signaling caused an arrest in GT development [24]. Our data demonstrates that the PCM's ability to respond to an FGF signal is essential for normal early genital tubercle outgrowth, as removing FGF receptors from the PCM caused impaired cell proliferation and perturbed normal gene expression patterns which led to severe GT reduction.
The obligatory role for FGF signaling during tubercle morphogenesis underscores the importance of identifying FGF ligands important for GT development. Our data are consistent with the views of Haraguchi et. al. [24], suggesting that dUE-expressed Fgf8 plays a key role in promoting GT outgrowth. Fgf8 is expressed at the correct time and place to signal to the PCM, which expresses both Fgfr1 and Fgfr2. This is in conflict with recent data by Seifert et. al. [25] in which they suggested that a normal GT could develop in the absence of Fgf8 expression. They did not detect FGF8 protein, which led to the conclusion that the Fgf8 mRNA may be present but not actively translated by the UE. They proposed that the ventral ectoderm may be an alternative source of other FGF ligands. In contrast, our results indicate that FGF8 protein can be made by the cloacal endoderm/UE as we demonstrated that a weakly-expressed R26Fgf8 allele can profoundly alter cell proliferation and gene expression in the neighboring PCM, and rescue GT agenesis in β-catenin mutants. In addition, our observation that the endogenous Fgf8 promoter is repressed upon forced Fgf8 overexpression revealed that the developing GT is equipped with mechanisms that can fine-tune the level of FGF8 signaling. All of these data indicate that the GT is not only responsive but also sensitive to FGF8 during normal development. The inability to detect FGF8 in the dUE by IHC is likely caused by the low expression level of Fgf8. We found that in E10.5 mouse embryos, robust AER Fgf8-expression was observed 2 hours after incubation with Alkaline Phosphatase substrates following standard in situ hybridization procedures, whereas dUE-expression was not observed until 12–18 hours later. In addition, the extensive genetic redundancy among FGFs has to be carefully considered. A recent study demonstrated ectopic Fgf4 expression in the dUE of Fgf8-cKO embryos, and ectopic Fgf3 expression in the dUE of Fgf4;Fgf8-dcKO embryos [26]. Both FGF4 and FGF3 can efficiently induce mitogenic activity when paired with FGFR1 or FGFR2 in vitro [42]. These observations indicate that not only FGFs endogenous to the dUE can compensate for the loss of Fgf8, other FGFs can also be ectopically activated to fulfill the requirement for FGF signaling. Activities from these ectopic FGFs could well explain the lack of GT phenotype in Fgf8-cKO mutants. Intriguingly, induction of Fgf3 and Fgf4 was not observed in either β-Cat- or Sp8-cKOs (Figure S10A–S10D), while the dUE expression of Fgf9 was downregulated in Sp8-cKOs (Figure S10E–S10F). These results suggest that the upregulation of these compensatory FGF factors also requires WNT and SP8. Alternatively, the hypothesis that FGFs can be produced by the ventral ectoderm is plausible [25]. However, one has to keep in mind that the GT is built and patterned around the UE. Most regulatory genes expressed distally including Msx1, Msx2, Wnt5a and Lef1, showed peri-dUE expression but not sub-ectodermal expression. Consistently, we have shown that the expression of these genes is orchestrated by UE-specific WNT and FGF signaling. Considering all the available evidence, we conclude that FGF8 produced by the dUE is most likely the endogenous ligand that mediates FGF responses during GT development.
Notably, the GT phenotype of the UE-Fgfr1;r2-dcKO embryos is less severe than what was observed previously in Fgfr2IIIb-KOs [28]. This difference is likely attributable to the method of gene ablation since in the Fgfr2IIIb-KO embryos, Fgfr2-IIIb is abolished from not only the UE, but also from the ventral ectoderm. The underdeveloped phenotype described in Fgfr2IIIb-KO embryos reflects a deficiency from ectodermal FGF responsiveness as it can be phenocopied by conditional ablation of both Fgfr1 and Fgfr2 using the ectodermally restricted Msx2-Cre allele (Lin et. al., unpublished data).
We have demonstrated a conserved WNT-SP8-FGF8 pathway in the distal signaling epithelia that functions to promote proximodistal outgrowth of both limbs and genitalia. We and others have shown that the canonical WNT pathway is a master molecular switch in the signaling epithelia during appendage formation, as epithelial WNT activation is not only necessary but also sufficient to induce Sp8 and Fgf8 expression and appendage outgrowth. FGF8 is the downstream signal output for the WNT pathway during this process as it acts directly on recipient mesenchymal cells to promote cell proliferation and establish patterns of gene expression. This genetic hierarchy has been supported by our observation that forced overexpression of Fgf8, even at a level much lower than endogenous expression, can bypass the requirement for epithelial WNT-β-catenin signaling to activate gene expression and initiate/maintain appendage outgrowth. Notably, in both appendages, the rescue of β-catenin deficiency by ectopic Fgf8 expression is only partial. This could be due to either weak Fgf8 expression from the R26 locus compared to the endogenous level of expression, or the possibility that in addition to regulating Fgf8 expression, canonical WNT signaling is also required for the formation of the AER structure, independent of FGF signaling [20]. The regulation of Fgf8 by the canonical WNT-β-catenin signaling pathway is in part mediated by SP8, as Sp8 expression is regulated by WNT signaling in both limb ectoderm and the UE, and is necessary for Fgf8 expression and subsequent appendage formation. However, unlike WNT activity and Fgf8 expression, the expression domain of Sp8 is not restricted to the AER and dUE but also includes the limb ectoderm [41] as well as proximal UE (Figure 6A), suggesting that SP8 is a permissive but not inductive factor for the establishment of Fgf8 expression. In support of this notion, overexpressing Sp8 in the limb ectoderm or UE did not cause any perturbation in Fgf8 expression or appendage formation. In addition, forced Sp8 expression failed to rescue Fgf8 expression and appendage outgrowth in the epithelial β-Cat-LOF mutants.
Notably, even with both Sp8 alleles mutated, the dUE and AER Fgf8 expression in the β-Cat-GOF mutants is still higher than in the controls. This finding apparently is counterintuitive to the obligatory role for SP8 in maintaining Fgf8 expression during normal outgrowth processes. One possible explanation is that the level of WNT signaling induced by the stabilized β-catenin protein from the β-CatΔex3 allele might be too high, and not subjected to endogenous regulatory mechanisms. This could potentially trigger ectopic events that lead to Fgf8 overexpression. It is also noteworthy that the other members of SP/KLF transcription factor family Sp6, was upregulated in the β-Cat-GOF mutants (Figure S11). Sp6 is expressed in the cloacal endoderm and the AER in early appendage development (Figure S11B and Figure S6A, respectively), and loss of Sp6 causes abnormal AER formation and Fgf8 expression [43]. Similar to Sp8, the expression of Sp6 is also downstream of the canonical WNT signaling pathway but independent of FGFs in the AER [43]. The exact function of SP6 in regulating GT development and dUE-Fgf8 expression remains to be determined. However, it is plausible that high levels of SP6 induced by ectopic WNT signaling might compensate for the absence of SP8 in the regulation of Fgf8 expression.
The exact molecular mechanisms through which Fgf8 expression is regulated by SP8 and β-catenin, and how Sp8 expression is regulated by WNT-β-catenin remain to be determined. Direct binding of the β-catenin/LEF1 complex to cis-regulatory elements within the Fgf8 promoter has been reported in dental epithelial cell lines [44] and nephron progenitors [45], suggesting that this WNT-Fgf8 pathway also functions in the development of other organs involving epithelial-mesenchymal interactions. In our preliminary studies, we identified several novel LEF binding sites around Fgf8 and Sp8 genes (data not shown). The functional relevance of these binding sites in GT and limb development will be the goal of future investigations.
Although the distal signaling cascades appear to be similar in both appendages, the upstream events leading to their initiation are likely different. In the limb, the ectoderm is the site for both Wnt3 expression [3], [46], and the induction of canonical Wnt downstream targets as evidenced by TOPGAL activity [47]. On the other hand, the establishment of the dUE signaling center in the cloacal endoderm appears to involve signal transduction between the cloacal endoderm and the ventral ectoderm. Both TOPGAL and Fgf8 expression are confined to the endodermal cells adjacent to the genital ectoderm [4], [24], and the contact between ectoderm and endoderm appears to be a prerequisite for Fgf8 induction and GT initiation [25]. In support of this notion, candidate WNT ligands responsible for activating canonical WNT signaling in the dUE, Wnt3 and Wnt7a, are both expressed in the genital ectoderm [4], [26].
Altogether, our results demonstrate extensive parallels between genetic networks regulating the outgrowth of both the limb and GT. These findings strongly support the notion initially brought up by two pivotal studies on the role of Hox genes in appendage development [10], [11], that mammalian GT development appears to be achieved through co-option of the limb outgrowth program, i.e. the mammalian GT and limb share deep homology [48], [49].
Clones containing full length Fgf8 (BC048734) and Sp8 (BC082582) cDNAs were obtained from Invitrogen (Carlsbad, CA). The cDNAs were released and subcloned into the NotI site of the pBigT vector [50]. The insert containing either Fgf8 or Sp8 cDNA and a floxed transcription stop cassette was released by PacI/AscI double digestion and cloned into pRosa26-PA [51]. The targeting construct was linearized by SwaI and subjected to electroporation (performed by ES cell core at Washington University). The ES cells were screened for recombination by PCR and southern blotting. Five out of seventy-two clones were positive for recombination for R26Fgf8 allele, and six out of sixty clones were positive for R26Sp8 allele. Positive clones were expanded, karyotyped and used for blastocyst injection. For both lines, at least three chimeras were able to pass the knock-in alleles through germline transmission, and the phenotypes resulting from expression of the knock-in alleles were identical. For all experiments described in the manuscript, three embryos with the same genotype were examined if not otherwise specified.
All animals were maintained according to NIH guidelines and in compliance with animal protocol approved by Washington University. Msx2rtTA [29], Sp8f/f and Sp8−/− [41], tetO-Cre [52], β-Catf/f [53], β-Catex3/ex3 [40], ShhCreERT2 and ShhEGFPCre [33], Msx2-Cre [35], Fgfr1f/f [31], Dermo1Cre and Fgfr2f/f [32] alleles were previously described. Tamoxifen treatment and doxycycline treatment were performed as previously described [30].
For all experiments, we used three independent biological samples from each genotype. Each sample contained a pool of RNA isolated from two E12.5 GTs with the corresponding genotype. The results were analyzed using the delta-CT method. Expression of the corresponding genes was normalized to that of housekeeping gene Rpl7.
Whole mount in situ hybridization was performed using a standard protocol. The probes were previously described [4].
Paraformaldehyde (4%)-fixed embryos were dehydrated and embedded in paraffin. Five-micron transverse GT sections were generated using a standard microtome. Scanning electron microscopy analyses were performed as previously described [4].
Skeleton preparation was performed as previously described [54].
The Genbank accession numbers for genes included in the manuscript are as follows: Fgf8 (NM_010205.2), Sp8 (NM_177082.4), β-catenin (NM_007614.3), Fgfr1 (NM_010206.2), Fgfr2 (NM_010207.2), Shh (NM_009170.3).
|
10.1371/journal.pbio.0060310 | SUMO-Specific Protease 2 Is Essential for Modulating p53-Mdm2 in Development of Trophoblast Stem Cell Niches and Lineages | SUMO-specific protease 2 (SENP2) modifies proteins by removing SUMO from its substrates. Although SUMO-specific proteases are known to reverse sumoylation in many defined systems, their importance in mammalian development and pathogenesis remains largely elusive. Here we report that SENP2 is highly expressed in trophoblast cells that are required for placentation. Targeted disruption of SENP2 in mice reveals its essential role in development of all three trophoblast layers. The mutation causes a deficiency in cell cycle progression. SENP2 has a specific role in the G–S transition, which is required for mitotic and endoreduplication cell cycles in trophoblast proliferation and differentiation, respectively. SENP2 ablation disturbs the p53–Mdm2 pathway, affecting the expansion of trophoblast progenitors and their maturation. Reintroducing SENP2 into the mutants can reduce the sumoylation of Mdm2, diminish the p53 level and promote trophoblast development. Furthermore, downregulation of p53 alleviates the SENP2-null phenotypes and stimulation of p53 causes abnormalities in trophoblast proliferation and differentiation, resembling those of the SENP2 mutants. Our data reveal a key genetic pathway, SENP2–Mdm2–p53, underlying trophoblast lineage development, suggesting its pivotal role in cell cycle progression of mitosis and endoreduplication.
| Genome replication is essential for both expansion of stem cell numbers through mitosis and their maturation into certain specialized cell types through endoreduplication, a unique mechanism for multiplying chromosomes without dividing the cell. An important function of p53 as a guardian of the genome ensures that the genetic information is properly propagated during these processes. In this study, we discovered that mice with disruption of SENP2, an enzyme that removes small molecular signals (called SUMO) that modify a protein's behavior and stability, are unable to form a healthy placenta as a result of deficiencies in the formation of various trophoblast cell types that give rise to the placenta. In the mutants, SUMO modification of Mdm2, a protein that monitors the cellular levels of p53, is deregulated. The loss of SENP2 causes dislocation of Mdm2, leading to aberrant stimulation of p53. The precursor cells known as trophoblast stem cells rely on p53 to proliferate and differentiate into specialized polyploid cells, which contain multiple copies of chromosomes. In SENP2 mutants, all three trophoblast layers were substantially defective, with the layer containing mainly the polyploid cells most severely affected and diminished. This study reveals a key genetic pathway, SENP2–Mdm2–p53, which is pivotal for the genome replication underlying trophoblast cell proliferation and differentiation.
| The first two distinct lineages to form in the mammalian embryos are the outer trophectoderm and the inner cell mass (ICM) of the blastocyst [1]. The trophectoderm initiates implantation and invasion of the uterus, processes that are essential for placental development [2]. This process depends on the differentiation of trophoblasts, the main and most important cell types in the placenta [3,4]. The trophoblast stem (TS) cells in the mural trophectoderm, distal to the ICM, stop dividing but continue to duplicate their genomes, a mechanism known as endoreduplication. The polyploid trophoblast giant cells (TGCs) then develop and eventually surround the entire fetus [5]. As development proceeds, the trophoblast progenitors give rise to three distinct layers in rodents—labyrinth, spongiotrophoblast and TGCs—to form a functional placenta acting as the maternal–fetal interface [6]. The fetal–placental blood vessels grow in from the allantois to generate the fetal parts of the placental vasculature where the chorioallantoic fusion has occurred [7]. The labyrinth is formed by extensive branching morphogenesis of the labyrinth trophoblast and endothelial cells [8]. The maternal blood passes through the small spaces of the labyrinth, directly contacting the fetal trophoblast cells to ensure exchange between the two blood systems. The labyrinth layer is supported structurally by the spongiotrophoblast cells, which are mainly derived from the ectoplacental cone and which form a layer separating the labyrinth from the TGC. The simplicity of placental cell lineages makes the placenta a valuable model system for understanding general aspects of development, including branching morphogenesis, lineage-specific determination, cell invasion, and polyploidy, crucial for cancer development and metastasis.
SENP2 belongs to a family of proteases that remove a small ubiquitin-related modifier (SUMO) from protein substrates. SUMO (also known as sentrin), which regulates posttranslational modification of proteins, is a member of the ubiquitin-like modifier family [9]. This covalent conjugation process is reversible and highly evolutionary conserved from yeasts to humans [10]. Unlike ubiquitination, which has a well-established role in targeting protein degradation, SUMO modification is involved in protein trafficking, cell cycle, cell survival, and cell death [11]. SUMO conjugation of proteins can alter their function, activity, or subcellular localization. Many sumoylated proteins have been shown to accumulate preferentially in specific complexes such as the nuclear pore and PML (promyelocytic leukemia) bodies [12]. Similar to ubiquitination, sumoylation requires processing, conjugation, and transfer. The transfer process, which covalently conjugates SUMO polypeptides to their targets, is catalyzed by E3 ligases [13]. The reverse desumoylation process is mediated by SUMO proteases. The hallmark of these proteases is the highly conserved carboxyl-terminal SENP domain of ∼200 amino acids. SENP2, which is found in three different alternatively spliced forms, has been localized to the nucleus, cytoplasmic vesicles and PML nuclear bodies [14–16]. Although SENPs have been shown to catalyze SUMO modification in various physiological systems, their roles in mammalian development and pathogenesis are mostly unknown.
We previously discovered an interaction of SENP2 with Axin [17,18], a key signaling regulator for the canonical Wnt pathway. To determine the role of SENP2 in cellular signaling and the importance of SUMO modification in trophoblast development, we initiated a genetic analysis in mice. A SENP2-null mouse strain was created by gene targeting in embryonic stem (ES) cells. We found that the disruption of SENP2 leads to developmental defects in all three trophoblast layers. SENP2 is essential for the G–S transition of both the mitotic and the endoreduplication cell cycles, which control the expansion of trophoblast precursors and the maturation of TGCs, respectively. In the mutants, the loss of SENP2 caused a deregulation of Mdm2, resulting in p53 stimulation. We also present evidence to support an essential role of SENP2 in modulating the p53–Mdm2 circuit that underlies genome replication in mitosis and polyploidy during trophoblast development.
To determine the role of SENP2 and the importance of SUMO modification in trophoblast development, we first examined its expression pattern. Strong expression of SENP2 was observed in extraembryonic tissues, including extraembryonic ectoderm, chorion and ectoplacental cone, at embryonic day (E)7 (Figure 1A). In extraembryonic ectoderm, its expression started diminishing by E7.5 (Figure 1B). In addition to these stem cell niche sites, we also detected its transcript in TS cells (Figure 1C). At E8.5, SENP2 maintained its ubiquitous expression in trophoblast cells located in the chorion and ectoplacental cone (Figure 1D–1F). By E9.5 and E10.5, the SENP2 transcript was detected in all three trophoblast layers: labyrinth, spongiotrophoblast and TGC (Figure 1H and 1L). SENP2 was expressed in the labyrinth trophoblast cells, which derive from the extraembryonic ectoderm and chorion, upon chorioallantoic fusion at E9.5 (Figure 1I). In the E10.5 labyrinth layer, its expression was specifically localized to cytotrophoblasts (mononuclear trophoblasts), adjacent to the maternal blood cells (Figure 1M). Syncytiotrophoblasts, as well as endothelial and blood cells, appeared to be negative for the staining. In contrast, we found a uniform expression of SENP2 in spongiotrophoblasts and TGCs (Figure 1J, 1K, 1N, and 1O), which are derivatives of the ectoplacental cone. TGCs include primary and secondary cells, derived from mural trophectoderm and ectoplacental cone (derivatives of polar trophectoderm), respectively. The SENP2 transcript was detected in both the primary and secondary TGCs (Figure 1G, 1K, and 1O). These results imply an important function of SENP2 in trophoblast progenitors and their development into all three major layers.
A SENP2-null allele was created by the targeted insertion of a lacZ reporter with pgk-neo cassette into exon 2 and the deletion of exons 3 to 5 to inactivate all different forms of the SENP2 gene product (see Materials and Methods for details). The targeted mouse ES cell clones heterozygous for SENP2 (Figure S1A), obtained by homologous recombination, were then used to obtain the SENP2lacZ mouse strain (Figure S1B). Mice carrying the targeted allele were subsequently bred with a Zp3-Cre transgenic strain to remove the pgk-neo cassette (SENP2-null allele), as confirmed by PCR genotyping analysis (Figure S1C). RT-PCR analyses further showed that the SENP2-null allele does not express the SENP2 transcript, but instead expresses the inserted lacZ gene (Figure S1D).
The SENP2-null heterozygous (hereafter referred to as SENP2+/–) mice were viable and fertile without any noticeable abnormalities. However, we were unable to find SENP2-null homozygous (hereafter referred to as SENP2–/–) newborns, implying that they died prematurely. These results prompted us to investigate whether the loss of SENP2 causes embryonic lethality. The SENP2–/– embryos appeared to be morphologically indistinguishable from their SENP2+/+ and SENP2+/– littermates at E9.5 (Figure 2A and 2B). However, the SENP2–/– embryos were significantly smaller or underdeveloped compared with the SENP2+/+ and SENP2+/– littermates at E10.5 (Figure 2C and 2D). We could not recover the SENP2–/– embryos after E11.5. This phenotype is often associated with placental deficiencies, as the embryos begin to rely on maternal supplies upon allantoic fusion at mid gestation. Indeed, the SENP2–/– placentas were smaller and paler than the controls (Figure 2E–2H). The average diameter of E10.5 placentas reduced from 5.2 mm in controls to 3.8 mm in mutants (Figure 2S, p < 0.0001, n = 7). Histological analyses revealed a reduction of the TGC layer by E9.5 (Figures 2I and 2J). By E10.5, the thickness of all three trophoblast layers decreased drastically in the SENP2-null mutants (Figures 2K and 2L). The TGC layer, which is the layer most severely affected by the SENP2 mutation, is almost completely missing. The data suggest that SENP2 has a pivotal role in development of all three trophoblast layers.
The placental defects caused by SENP2 deficiency suggested that it is critical for trophoblast development. The stem cells derived from the trophectoderm develop into progenitors, which reside in the ectoplacental cone, the extraembryonic ectoderm, and the chorion. We therefore examined whether the SENP2 deletion interferes with formation of these niche sites. In situ hybridization of Tpbpa, a marker for the ectoplacental cone [19], revealed a drastic reduction of trophoblast progenitors in the mutants (Figure 2M and 2P). The number of trophoblast progenitors, marked by Cdx2 expression [20] was also decreased in the SENP2–/– chorion and extraembryonic ectoderm (Figure 2N, 2O, 2Q, and 2R). The apparent developmental defects of trophoblast niche sites suggested that SENP2 might have a role in trophoblast stem cell development.
A closer examination of the labyrinth layer was performed by analyzing the expression of Gcm1, a labyrinth trophoblast marker that is specifically detected in the chorioallantoic invasion sites and later in the differentiated syncytiotrophoblasts [21]. No obvious difference between SENP2+/+ and SENP2–/– was observed at E9.5 (Figure 3A and 3D). Gcm-1-positive trophoblast progenitors were clearly identified at the invasion sites. Therefore, fetal vascular invasion was not affected by the deletion. However, deficiencies in labyrinth development of SENP2–/– embryos were evident at E10. The syncytiotrophoblasts positive for Gcm-1 exhibited punctated staining in the mutant instead of the continuous thin layers seen in the wild type, suggesting that their differentiation is defective (Figure 3B and 3E). By E10.5, the number of the Gcm1-expressing cells was dramatically reduced in the mutants (data not shown). At this stage, SENP2 expression is restricted to the cytotrophoblasts (Figure 1M), a subtype of TGCs [22]. Therefore, we examined whether cytotrophoblast development was affected by analyzing a cytotrophoblast marker, Ctsq [23]. Indeed, the Ctsq-positive cytotrophoblasts identified in the wild-type labyrinth were completely missing in the mutants (Figure 3C and 3F). Histological analysis further showed that both maternal and fetal blood spaces were enlarged without formation of capillary structures in the SENP2 mutants (Figure 3G and 3J). Immunostaining of laminin [24], a basement membrane protein expressed by endothelial cells that highlight fetal blood spaces, further revealed a failure of branching morphogenesis in the SENP2-null fetal vasculature (Figure 3H and 3K). This might be attributed to a deficiency in endothelial proliferation as the number of cyclin D1-positive cells (proliferation marker only detected in endothelial cells) was decreased in the mutants (Figure 3I and 3L). These data demonstrated that SENP2 is essential for labyrinth trophoblast development in establishment of the maternal and fetal blood spaces. The presence of SENP2 in early trophoblast precursors might regulate the differentiation of specialized cell types at later stages. Alternatively, its function in the cytotrophoblasts could be crucial for proper development of syncytiotrophoblasts and endothelial cells. SENP2 is necessary for development of the labyrinth layer during placentation.
We next examined the spongiotrophoblast layer that is affected by the SENP2 deletion. In situ hybridization analysis of Tpbpa [19], a marker for the spongiotrophoblast, revealed that its expressing cells diminished significantly in the mutants at E9.5–E10.5 (Figure 3M, 3N, 3P, and 3Q). Histology confirmed that a rapid expansion of this layer, found in the wild-type placenta, did not occur in the mutants (Figure 3O and 3R). As a result, the SENP2-null spongiotrophoblast layer decreased significantly in volume. Based on the expression of SENP2 in spongiotrophoblasts (Figure 1J and 1N) and earlier in their precursors at the ectoplacental cone (Figure 1A, 1B, and 1F), it is most likely that the abnormalities are primarily due to its deletion in these tissues. Therefore, spongiotrophoblast development requires SENP2 and its disruption induces abnormalities in the spongiotrophoblast layer.
Consistent with its expression in early trophoblast development, histological analyses revealed a severe abnormality in the TGC layer (Figure 4A–4H). The SENP2-null primary TGCs were reduced at E8.5 and completely missing at E9.5 (Figure 4A, 4B, 4E, and 4F). Similarly, the number of secondary TGCs was decreased at E9.5 and almost disappeared at E10.5 (Figure 4C, 4D, 4G, and 4H). In addition, the size of TGCs was significantly smaller in the SENP2 mutants (Figure 4D and 4H). The analyses of TGC markers [19,25], including PL-I (Figure 4I–4P), PL-II (unpublished data), and p450scc (Figure 4Q–4V), confirmed that the TGC cell numbers were dramatically decreased in the SENP2 mutants at all stages examined. We next examined the initiation of TGC differentiation by in situ hybridization of Hand1. Hand1 is required for cell fate determination of TGC, as mice without Hand1 lack TGCs [26]. Hand1 expression was detected in the SENP2-null TGCs, suggesting that the initial induction of TGCs was not affected by the loss of SENP2 (Figure 4W–4Z and 4W′–4Z′). However, later developmental processes of TGC were impaired in the mutants.
The abnormal development of TGC caused by SENP2 deficiency was further tested using an in vitro differentiation analysis. The SENP2+/+ and SENP2–/– blastocysts were isolated at E3.5, and cultured to induce TGC differentiation. TS cells growing out from the trophectoderm soon attached to the cultured plates, differentiated, and formed a single trophoblast layer. No noticeable difference was observed between the SENP2+/+ and SENP2–/– blastocysts before hatching (Figure 5A and 5B). About equal amounts of ICM and trophoblast cells developed after 3 d in culture (Figure 5C and 5D). However, although the differentiated TGCs were evident in the SENP2+/+ cultures, their number was significantly reduced in the SENP2–/– cultures after 6 d (Figure 5E–5H). The average number of TGC dropped from 40 in the SENP2+/+ culture to 15 in the SENP2–/– (Figure 5I, p = 0.005, n = 6). Consistent with our in vivo findings, these data suggest that TGC differentiation is severely affected by the loss of SENP2. The results suggest an essential role for SENP2 in TGC development during early placentation.
The SENP2 mutation led to abnormalities in trophoblast progenitors at niche sites and their development into all three major trophoblast lineages. These findings imply that SENP2 might have a general role in cellular regulations important for expansion of precursors and their differentiation. We speculated that decreases in the numbers of trophoblast progenitors and specialized cell types might be due to alterations in cell survival. However, we failed to detect differences in apoptosis caused by the mutation in trophoblast stem cell niches and all three major trophoblast layers in vivo, or in TS cell culture in vitro (Figure S2). We then examined whether SENP2 has an important function in the cell cycle. Investigating the expansion of trophoblast progenitors at the niches revealed a deficiency in their cell cycle progression. The expression of Ki67, a marker detected in all phases of mitotic cells [18], was detected in virtually all trophoblast progenitors in stem cell niches, including extraembryonic ectoderm, chorion, and ectoplacental cone (Figure 6A, 6C, 6E, and 6G), suggesting that they are actively cycling cells. We next examined the cell cycle progression rate among actively cycling cells by measuring the DNA synthesis rate at S phase using BrdU labeling [18] for 1 h (Figure 6B, 6D, 6F, and 6H). BrdU incorporation specifically measures the rate of cell cycle progression at S phase, whereas Ki67 identifies all phases of mitotic cells. The cell cycle progression index (% BrdU-positive cells / % of Ki67-positive cells × 102) among actively cycling cells decreased 18 units in the mutants (SENP2+/+ and SENP2+/–, 67; SENP2–/–, 49; p = 0.0001, n = 6) in the stem cell niches (Figure 6M). These data suggest a delay in cell cycle progression of trophoblast progenitors caused by the SENP2 deletion.
Next, we determined whether similar deficiencies also affect development of the spongiotrophoblast layer. We found that this layer expanded rapidly in the wild-type placenta, but not in the mutants, between E9.5 and E10.5 (data not shown). A portion of the SENP2+/+ spongiotrophoblasts exited the cell cycle at E10.5 (Figure 6I), whereas almost all of the SENP2–/– spongiotrophoblasts remained Ki67-positive (Figure 6K). In E10 SENP2+/+ and SENP2–/– placentas, spongiotrophoblasts were all positive for Ki67, indicating that they are actively cycling cells (Figure 6J and 6L). However, the cell cycle progression index, which mainly reflects the BrdU incorporation rate, was reduced from 72 in the controls to 53 in the SENP2 mutants (p = 0.0005, n = 3) (Figure 6N). To examine whether cycling of TGC was also affected by SENP2 deficiency, we determined its cell cycle progression index (Figure 6O). The cell cycle progression index of TGC decreased from 60 in SENP2+/+ to 41 in SENP2–/– (p = 0.0034, n = 4). Taken together, these results suggest that cell cycle progression was defective in all stem cell niches, and the spongiotrophoblast and TGC layers, of SENP2 mutants. The SENP2 mutant cells were trapped or arrested in the cell cycle.
To further examine stem cell expansion and development, we derived a number of SENP2–/– TS cell lines from blastocysts. Immunostaining analyses of Oct4 (an ES cell marker) [27] and Cdx2 (a TS cell marker) [20] confirmed that we were able to successfully establish the SENP2-null TS cell lines (Figure S3). The proliferation rate (BrdU labeling for 1 h) of the SENP2-null TS cells in vitro was also reduced, compared to that of the wild-type cells (p = 0.013, n = 3) (Figure 7A). Although the deficiency in cell cycle progression was also demonstrated using the TS cells in vitro, the degree of severity was reduced compared with that seen in the in vivo studies. As we were aware, the in vitro system does not always recapitulate the dynamic developmental processes that occur in vivo. Nevertheless, because of the limited materials available from the early stages of placenta, the TS cell culture does provide a valuable system to further those of our investigations that are otherwise impossible to perform in vivo.
To investigate whether a specific phase of the cell cycle was defective, we then determined the cell cycle profiles of the SENP2+/+ and SENP2–/– TS cells by flow cytometry analysis of PI (propidium iodide) stained cells. There was no significant difference in the cell population of G2–M between SENP2+/+ and SENP2–/– cells (Figure 7B). However, in the SENP2 nulls, the percentage of cells in G0–G1 was increased (p < 0.0001, n = 4) but the percentage in S was decreased (p = 0.0024, n = 4) (Figure 7B and 7C). This implied that the mutant cells were affected at the G1–S transition. To test this hypothesis, we used nocodazole, a microtubule depolymerizing agent, to block cell division at M phase. Nocodazole was effective in synchronizing the SENP2+/+ TS cells at G2–M after 6 h (Figure 7D). However, if cells were arrested or trapped in the G1–S phase and unable to pass through the cell cycle, there would be a delay in synchronizing cells by the nocodazole treatment. Indeed, there were still ∼7% of the G0–G1 cells in SENP2–/–, but none in SENP2+/+, 3 h after the treatment. After the 6 h treatment, a significant number (6.16%) of the SENP2–/– TS cells remained in G0–G1 (Figure 7D). Even after 24 h, this population arrested in G0–G1 was still present (data not shown). The results suggest that SENP2 has a pivotal role in TS cell cycle progression and the G1–S checkpoint might be affected by the SENP2 ablation.
Immunostaining of nuclear envelopes with lamin B [28] revealed that nuclei of the SENP2–/– TGCs were significantly smaller (Figure 8A–8F). In addition, the mutant TGC nuclei contained smaller and fewer blue dots upon hematoxylin staining (Figure 8A–8F), suggesting that the DNA content might be reduced. These abnormalities are likely caused by a deficiency in endopolyploidy. An important specialized process for TGC maturation is endoreduplication, whereby the genome is amplified without a complete mitosis. The endoreduplication cycle requires only the G and S phases [29]. To examine the possibility of a defect in endoreduplication, we induced the TS cells to undergo TGC differentiation in vitro by removal of FGF4, heparin, and mouse embryonic fibroblast (MEF)-conditioned medium (see Materials and Methods and [30]). Flow cytometric analysis of the differentiated cells stained with PI showed that the percentage of cells with higher DNA contents (>4N) was drastically reduced in the SENP2 mutants (Figure 8G). The average percentage of polyploid cells reduced from 25% (SENP2+/+) to 7% (SENP2–/–) (p < 0.0001, n = 5) (Figure 8H). Therefore, the loss of SENP2 induced a severe deficiency in endopolyploidy. SENP2 apparently has a dual role in regulating the G–S transition of mitotic division and endoreduplication during TS cell proliferation and differentiation, respectively.
The cell cycle defects led us to investigate potential downstream targets involved in trophoblast development. We specifically focused on those regulators shown to be conjugated by SUMO. Previous reports showed that SENP2 (also known as Axam) modulates the canonical Wnt pathway by interacting with its signaling molecules [14,31]. Even though this led us to identify SENP2 through its binding to Axin initially, we failed to detect any alteration of Wnt signaling in the SENP2 mutants. Nor were we able to show other alternative pathways critical for placentation, e.g., MAPK and SAPK [32–36], to be involved in the SENP2-dependent developmental processes. However, when p53 was examined by immunostaining, we detected an aberrant accumulation in the nuclei of the developing SENP2–/– TGCs at E8.5–E10.5 (Figure 9D–9F). In contrast, the SENP2+/+ TGCs showed no detectable, or very low if any, p53 at these stages (Figure 9A–9C). The results implied that there is a deficiency in p53 regulation caused by the SENP2 deletion. Degradation of p53 is mediated by ubiquitination-dependent proteolysis. Mdm2, a RING finger E3 ubiquitin ligase that binds to p53, has an essential role in this process [37–40]. We therefore tested whether the loss of SENP2 had an effect on Mdm2. Immunostaining of Mdm2 revealed its localization in both the SENP2+/+ cytoplasm and the nucleus during early stages (E8.5–E9.5) of TGC development (Figure 9G and 9H). However, Mdm2 was mainly located to the nuclei of the terminally differentiated TGCs at E10.5 (Figure 9I). The differential subcellular distribution of Mdm2 implies that it might be critical for development of TGCs. In contrast, Mdm2 accumulated in nuclei throughout TGC development in the absence of SENP2 (Figure 9J–9L). The prominent cytoplasmic staining was lost in the mutants at E8.5–E9.5 (Figure 9J and 9K). Furthermore, the loss of SENP2 also affected Mdm2 localization in the stem cell niche sites, such as extraembryonic ectoderm and chorion. Mdm2 clearly accumulated in the nuclei of the SENP2–/– trophoblast progenitors, but was evenly distributed in the whole cells of the controls (Figure 9M and 9N). Similar nuclear accumulations of Mdm2, affecting the p53 level, were also detected in the SENP2–/– labyrinth and spongiotrophoblast layers (data not shown). Therefore, Mdm2 appeared to be aberrantly localized in the stem cell niches and all three major layers of trophoblast during early embryogenesis. The data suggest that SENP2 is required for proper localization of Mdm2 and degradation of p53. Disturbance of SUMO modification by the SENP2 deletion thus causes deregulation of the p53–Mdm2 pathway, leading to deficiencies in mitotic and endoreduplication cell cycle progression and abnormal trophoblast development.
The accumulation of p53 in the nuclei of SENP2-null placentas implied that SENP2 negatively modulates the p53–Mdm2 circuit. To determine the role of p53–Mdm2 in trophoblast development, we investigated whether SENP2 modulates Mdm2 and p53 at the posttranscriptional level. In addition to altering the subcellular distribution of Mdm2, the loss of SENP2 had an effect on posttranslational modification of Mdm2. The loss of SENP2 disturbed desumoylation of Mdm2. In the SENP2–/– TS cells, Mdm2 accumulated in the SUMO conjugated state (Figure 9O). The loss of SENP2 disturbed the ratio of Mdm2 and Mdm2–SUMO. The sumoylated Mdm2 could also be detected by an anti-SUMO-1 antibody (Figure 9O) as well as immunoprecipitation–immunoblot analysis using anti-Mdm2 and anti-SUMO-1 antibodies (data not shown). We encountered a technical problem in determining the actual amount of the sumoylated Mdm2 by immunoprecipitation–immunoblot analysis. This is likely because desumoylation occurs rapidly in isolated cell extracts whereas immunoprecipitation requires proteins in a native conformation. Therefore, a straight immunoblot assay appears to be better suited for quantitative measurements. To determine whether SUMO modification of Mdm2 is regulated by SENP2, a plasmid expressing a Myc-tagged SENP2 (MT–SENP2) under the control of a CMV promoter was transiently transfected into the mutants. The reintroduction of SENP2 altered the ratio of Mdm2 and Mdm2–SUMO and diminished the level of Mdm2–SUMO, suggesting that its desumoylation is modulated by SENP2 (Figure 9O). Immunoblot analysis also revealed an elevation of p53 caused by the SENP2 deletion in TS cells (Figure 9P). Although p53 is known to be sumoylated, we did not detect obvious accumulations of the SUMO-conjugated form caused by the SENP2 ablation. We then tested whether SENP2 is required to mediate the downregulation of p53 by overexpression of MT–SENP2. Consistent with our hypothesis, p53 levels were significantly reduced in the SENP2-null cells transiently transfected by MT–SENP2 (Figure 9P).
To further confirm that the loss of SENP2 was the primary cause of the trophoblast defects, we reintroduced MT–SENP2 into SENP2–/– cells. To determine the differentiation process affected by SENP2 at a more quantitative level, we examined the expression of a TGC marker, p450scc, by immunoblot analysis. The expression of p450scc was drastically reduced in SENP2–/– placentas, confirming the TGC developmental defects (Figure 9Q). The expression of p450scc was not detectable in SENP2+/+ TS cells but was highly increased in the differentiated TGCs, suggesting the success of the in vitro culture system (Figure 9Q). We did not detect a great induction of p450scc in the differentiated SENP2–/– cells, consistent with our in vivo findings (Figure 9Q). The reintroduction of MT–SENP2 in the SENP2 mutants led to an induction of p450scc upon TGC differentiation (Figure 9Q). The p450scc induction level did not reach that of the SENP2+/+ TGCs, most likely due to the transfection efficacy, in that not all of the mutants were transfected. Nevertheless, these data demonstrate that reintroducing SENP2 into the SENP2–/– TS cells can promote their differentiation into TGCs. This suggests that SENP2 inactivation is the cause of the trophoblast developmental defects observed in the mutants. An aberrant stimulation of p53 might be responsible for the SENP2-null defects in mitotic division and polyploidy.
In the SENP2 mutants, the dislocation of Mdm2 implied that its distribution is regulated by the SUMO pathway. We therefore investigated whether Mdm2 localization is affected by SUMO. First, immunoblot analysis after cell fractionation showed that sumoylated Mdm2 is found preferentially in the nuclear fraction of SENP2–/– cells (Figure 9R). Next, we examined whether SUMO conjugation alters the subcellular distribution of Mdm2 in live cells. GFP analysis of TS cells transiently expressing GFP-tagged Mdm2 or Mdm2–SUMO-1, revealed their preferential localization. We found that Mdm2 mainly accumulated in the cytoplasm (Figure 9S and 9V), with occasional distribution to the whole cell (Figure 9T). However, Mdm2–SUMO-1 displayed a clear nuclear accumulation (Figure 9U), with either a punctated (Figure 9W) or a nucleolar (Figure 9X) staining pattern. Similar results were also obtained by the use of Mdm2–SUMO-1GG96–97Δ, a mutant lacking the last two glycine residues of SUMO-1, which prevent further conjugation that might affect subcellular distribution (data not shown). Therefore, the SENP2 mediated SUMO modification of Mdm2 appears to be crucial for its subcellular trafficking.
To address the importance of p53 in mediating the SENP2-null phenotype, we tested whether p53 activation is necessary and sufficient to affect trophoblast proliferation and differentiation. We used both gain-of-function and loss-of-function analyses. Nutlin-3 is a potent small-molecule antagonist of Mdm2, which binds to the p53-binding pocket of Mdm2 and prevents its interaction, thereby stabilizing p53. We first determined that the Nutlin-3 treatment of the SENP2+/+ cells could elevate p53 in a dosage-dependent manner, but, most importantly, to reach the level detected in the SENP2–/– TS cells (Figure 10A). To examine whether the p53 elevation induced G1–S arrest, TS cells were treated with Nutlin-3. A cell cycle profiling assay showed that the Nutlin-3 treatment caused the wild-type TS cells to accumulate in G0–G1 phase, similar to the SENP2–/– TS cells (Figure 10C). Next, we examined whether the elevated level of p53 interfered with the differentiation process. In the SENP2+/+ TS cells induced for TGC differentiation, Nutlin-3 significantly reduced the expression of the TGC marker p450scc (Figure 10E), and prevented TGC differentiation (Figure 10F–10K). The average number of TGC decreased significantly in the presence of Nutlin-3 (Figure 10L, p = 0.006, n = 4). These results support the hypothesis that stimulation of p53 by alteration in Mdm2 activity induces phenotypic defects in trophoblast proliferation and differentiation, resembling those observed in the SENP2 mutants.
To determine whether downregulation of p53 was able to alleviate the trophoblast deficiencies caused by the SENP2 ablation, we knocked down its cellular levels using an RNA interference (RNAi) approach. First, immunoblot analysis showed that the p53 RNAi treatment successfully diminished its levels in the SENP2–/– TS cells (Figure 10B). The p53 RNAi treatment also promoted the G1–S transition of the SENP2–/– TS cells arrested in G0–G1 (Figure 10D). Furthermore, downregulation of p53 enhanced TGC differentiation of the SENP2–/– cells, as determined by the expression of p450scc (Figure 10E). These data demonstrated that stimulation of p53 is not only necessary to mediate the SENP2-null defects, but is also sufficient to induce deficiencies in expansion of trophoblast stem cells and their maturation.
This study demonstrates an essential role of SENP2 in trophoblast lineage development during placentation. All three major trophoblast layers were affected by SENP2 deficiency. Our data provide an important connection between SENP2 and the p53–Mdm2 pathway in trophoblast development. The loss of SENP2 caused a deficiency in the G–S transition, which is required for both the mitotic cell cycle (containing G1, S, G2, and M phases) and the endocycle (containing only the G and S phases) during trophoblast proliferation and differentiation, respectively. The cell cycle regulators p53 and Mdm2 appear to be critical for SENP2-dependent trophoblast mitosis and polyploidy. We propose that the SENP2–Mdm2–p53 pathway has a dual role in the G–S checkpoint of mitotic division and endoreduplication (Figure 11A). Although high levels of p53 induce a G1 arrest, a low level may be necessary to go through the rest of mitosis, such as through the tetraploid checkpoint. Because of the omission of M phase in endoreduplication, repression of p53 is essential to produce polyploid cells. Our findings further suggest that SENP2-dependent SUMO modification controls the subcellular localization of Mdm2 (Figure 11B). Sumoylated Mdm2, which preferentially accumulates in the nucleus, likely cannot modulate p53, whereas desumoylated Mdm2, which can move freely to the cytoplasm, is capable of p53 degradation.
This study provides evidence to support an important function of p53, as a guardian of the genome to control polyploidy. An endoreduplication deficiency was previously observed in embryos lacking cyclin E proteins [41]. In contrast to the SENP2-null deficiencies, the loss of cyclin E proteins did not affect TGC differentiation. It is conceivable that cyclin E, which functions in late G1 phase to promote S-phase entry, acts further downstream of the SENP2–Mdm2–p53 pathway. In the SENP2 mutants, we detected alterations of this regulatory pathway not only in the stem cell niche site, but also in the differentiated trophoblast layer. A recent report found that an increased number of TGCs were detected in the p53-null placentas [42], further supporting our hypothesis. SENP2 might also be involved in a crucial step of p53-dependent aneuploidy, genome instability and tumorigenesis [43]. Polyploid cells have several different fates. They can arrest in the cell cycle mediated by the tetraploidy checkpoint, which then triggers apoptosis. However, the lack of p53 allows these cells, as they escape from the arrest to undergo multipolar mitosis, to become aneuploid [44–46]. The nature of trophoblast development provides a system to elucidate the regulatory mechanism underlying polyploidy. Because of the biochemical activity of SENP2, the SENP2-null model offers a unique opportunity to further investigate the modulation of the p53–Mdm2 circuit by SUMO in normal developmental programming of polyploidy. The knowledge obtained here might be applicable to malignant transformation processes associated with polyploidy.
SENP2 is also known as Axam, which has been shown to modulate Wnt signaling by interacting with Axin, a scaffold protein involved in targeting β-catenin for degradation [14,17]. Although biochemical studies suggested that SENP2 could regulate the canonical Wnt pathway by SUMO modulation of a LEF/TCF transcription factor [31], there was no in vivo evidence to support this idea. We failed to detect alterations in Wnt–β-catenin signaling in the SENP2 mutant placentas (SC and WH, unpublished data) although this might occur in other tissues. SUMO modification of Axin has been shown to modulate its effects on JNK signaling [36]. Neither JNK, nor the related p38 and Erk1/2 factors that are important for placental function [32–35], seem to be involved in the SENP2-mediated trophoblast development (SC and WH, unpublished data). However, we identified the p53–Mdm2 pathway as a downstream target of SENP2. Our data imply that SUMO modification mediated by SENP2 is required for proper localization and function of Mdm2, which in turn controls p53 stability during trophoblast development. Not only does stimulation of p53 induce phenotypic defects resembling those of the SENP2 inactivation, but downregulation of p53 alleviates the trophoblast deficiencies caused by SENP2 deficiency. It is conceivable that Wnt or JNK/SAPK signaling regulated by SENP2 is critical for another cell type and lineage development. The generation of mouse models permitting conditional inactivation of SENP2 will aid these studies and determine its essential role in other developmental processes.
The loss of SENP2 disturbs the balance of SUMO modification. Although sumoylation of Mdm2 has been described [47], it was not clear whether this modification dictates subcellular distribution. Our data provide evidence that cellular distribution of Mdm2 is regulated by the SUMO pathway. Disruption of SENP2, leading to an accumulation of Mdm2 in a hyper-sumoylated state, induces its mislocalization. Many sumoylated proteins, including PML, preferentially accumulate in specific complexes called PML nuclear bodies [12]. Sumoylation of PML is essential not only for these nuclear bodies to form but also for other sumoylated proteins to concentrate there. Although the biological function of PML nuclear bodies remains largely elusive, subsequent recruitment of proteins can modulate transcription activity. It has been shown that sumoylation of PML directs p53 to nuclear bodies, leading to a stimulation of its transcriptional and pro-apoptotic activities [48,49]. These effects can be regulated by sumoylation of p53 [11,50,51]. Because of technical limitations and, more importantly, SUMO regulation of a number of p53 regulators (Mdm2, MdmX, and PML), the functional consequences of sumoylation have been difficult to elucidate. As SUMO modification of PML and p53 is a key determinant for maintaining genome integrity [12], our data imply that SENP2 might mediate this maintenance.
Using a mouse model with disruption of SENP2, this study suggests a novel role of SUMO modification in cell cycle progression and induction of polyploidy. Sumoylation, which dictates Mdm2 trafficking, is crucial for modulation of the p53–Mdm2 circuit. Further studies focusing on the detailed mechanistic switch of the SENP2–Mdm2–p53 pathway and its implications in other developmental and pathogenic processes promise important insights into the role of SUMO modification in mammalian development and disease.
Genomic DNA fragments containing the SENP2 gene (Accession number NC_000082) were isolated by PCR and cloned into the pGEM vector. The 5′ arm contained sequences from the first coding exon to the beginning of the second coding exon, which encodes the first 49 amino acids of SENP2. The 3′ arm included parts of the fifth intron and the sixth coding exon. A β-galactosidase cDNA was fused in-frame to the second coding exon of SENP2. The SENP2lacZ /+ mutant ES cell lines were generated by electroporation of the targeting vector into CSL3 ES cells [52]. Correct homologous recombination at the SENP2 locus was confirmed by Southern blotting (Figure S1A). ES cell clones were injected into blastocysts to generate chimeras that were bred to obtain mice carrying the targeted allele. Mice were genotyped by PCR analysis using primers (G1: 5′-ctgttttctactgcagtggacac-3′, G3: 5′-gatacttgtagaaaggcctagtat-3′ and K1: 5′-taaccgtgcatctgccagtttga-3′) to identify the wild-type and mutant SENP2 locus (Figure S1B). To delete the neo cassette flanked by two loxP sites, the SENP2lacZ/+ strain was crossed with the Zp3-Cre strain as described [53]. PCR genotyping was performed to confirm the removal of neo and the presence of lacZ as described (Figure S1C) [52]. Care and use of experimental animals described in this work comply with guidelines and policies of the University Committee on Animal Resources at the University of Rochester.
The pCS2-SENP2 clone, containing the Myc-tagged SENP2 cDNA, was generated by inserting a blunt-ended 1.7 kb Not1–Spe1 fragment into the blunt-ended Xho1–Xba1 sites of pCS2 vector [54]. The GFP-tagged Mdm2 expression vector (pGFP-Mdm2) was generated by ligation of a full length Mdm2 [55] and GFP (BD bioscience) cDNA fragments. The GFP-tagged Mdm2–SUMO expression vector was created by insertion of a SUMO-1 fragment [51] into the pGFP-Mdm2 plasmid. To generate the pBS-SENP2 clone for making the RNA probes, a 400 bp BamH1–EcoR1 fragment of the pCS2-SENP2 clone was cloned into the same restriction sites in pBS vector (Stratagene). To generate RNA probes for in situ hybridization, DNA plasmids pBS-Gcm1, pBS-Hand1, pCR4-PL-I, pCR4-Tpbpa, pBS-Ctsq, and pBS-SENP2 [19,21,23,56] were linearized and transcribed in vitro using RNA polymerases T3, T7, and SP6 (Promega). Plasmid DNA transfection was performed by Lipofectamine 2000 (Invitrogen)-mediated transfer with 4 μg pCS2-MT–SENP2, 1 μg pGFP-Mdm2, 1 μg pGFP-Mdm2–SUMO-1, or 10–100 nM p53 siRNA (Santa Cruz). Cells were plated (1.5 × 105 cells in a 30 mm dish for protein extraction, 2 × 104 cells in a 24-well dish for GFP analysis, and 5 × 105 cells in a 60 mm dish for flow cytometry) 24 h prior to the transfection procedure. The transfected cells were harvested after 48 or 72 h for further analyses. Total RNA, isolated using Trizol (Invitrogen), was used to produce cDNA according to the manufacturer's instructions (SuperScript III, Invitrogen). The reverse transcription products were subject to PCR amplifications of the SENP2-lacZ fusion transcript using primers 5′-cagtctctacaatgctgcc-3′ and 5′-ctgtcactctgatctttgg-3′ (exons 3–5), primers 5′-gtgagctgatgagttctgg-3′ and 5′-gtcgctccaataactttcg-3′ (exons 4–6), primers 5′-ggaggagcagaatcatgg-3′ and 5′-ctcaaaatctcatctggtgg-3′ (exons 8–11) and primers 5′-cattaccagttggtctggtg-3′ and 5′-gctgcaataaacaagttccg-3′ (lacZ). The PCR reaction was performed by denaturation at 94 °C for 5 min and 30 cycles of amplification (94 °C for 30 s, 53 °C for 30 s, and 72 °C for 45 s), followed by a 7-min extension at 72 °C.
Mouse blastocysts were recovered and cultured in DMEM medium containing 15% FBS, 100 μM β-mercaptoethanol, 100 μM non-essential amino acid, and 100 μg/ml penicillin-streptomycin, in a humidified 5% CO2 incubator at 37 °C. Cultured embryos were hatched and attached to dishes after 24–36 h. The differentiated trophoblasts became identifiable in a few days. For genotyping, cultured cells were incubated in 10 μl buffer containing 25 mM NaOH and 0.2 mM EDTA, pH 12 for 1 h at 95 °C, followed by the addition of 10 μl buffer containing 40 mM Tris-HCl, pH 5.0. Lysates were subject to PCR analysis. The SENP2 wild-type allele was detected by a nested PCR assay. Primers 5′- ctgttttctactgcagtggacac-3′ and 5′-gctgcctggagtttatctactgtag-3′ were used for the first PCR reaction, performed with 35 cycles of amplification (94 °C for 30 s, 60 °C for 30 s, and 72 °C for 2 min 30 s), followed by a 7-min extension at 72 °C. Subsequently, the first PCR products were subject to a second PCR reaction using the method described for genotyping the SENP2 wild-type mouse strain. For genotyping the SENP2 mutant culture, the same method for the SENP2 mutant mouse strain was used.
To establish the TS cell lines [30], blastocysts were recovered in TS medium (RPMI-1640 medium containing 20% fetal bovine serum, 1 mM sodium pyruvate, 100 μM β-mercaptoethanol, 100 μg/ml penicillin–streptomycin), plus 25 ng/ml FGF4 and 1 ng/ml heparin. Briefly, each blastocyst was placed in a culture dish with mitomycin C-treated MEF feeders and cultured in a humidified 5% CO2 incubator at 37 °C. The blastocysts were hatched and attached to the dishes in 24–36 h. After 48 h, a small outgrowth from a blastocyst was formed and cultured in TS medium containing 25 ng/ml FGF4 and 1 ng/ml heparin. After 72–96 h, the outgrowths were ready to be disaggregated by the addition of 0.25% trypsin/EDTA and incubation for 3 min at 37 °C. The disaggregated cells were continuously cultured in TS medium with the presence of FGF4 and heparin. The TS cell colonies began to appear after days 6 to 10, and continued to be cultured until they were about 50% confluent. After expanding the cultures on the feeders for one or two passages, MEF-free TS cells were obtained and maintained in media containing 70% MEF-conditioned medium, 30% TS medium, 37.5 ng/ml FGF4, and 1.5 ng/ml heparin. To differentiate TS cells into TGC, cells were cultured in TS medium with no additions [30]. For BrdU labeling of the cultured cells, 30 μg/ml BrdU (Sigma) was added in the media for 1 h. The labeled cells were then fixed with methanol/acetone (1:1), followed by immunostaining analysis. For cell cycle analysis by flow cytometry, 8 × 105 (for mitotic cell cycle) or 105 (for endoreduplication cycle) TS cells were cultured in 6 cm dishes in TS media plus FGF4, heparin, and MEF-conditioned medium (undifferentiated medium) for 2 d, and TS media only (differentiated medium) for 6 d, respectively. Cells were then harvested by trypsinization and fixed in 70% ethanol at 4 °C for at least 24 h. Cells were then treated with RNase (1 mg/ml) for 30 min, followed by PI staining (20 μg/ml) for 10 min at room temperature. Samples were analyzed by an Epics Elite ESP (Coulter Electronics) set to collect 10,000 events. The percentage of cells in G0–G1, S, G2–M or with polyploidy were determined using ModFit LT software. For synchronizing cells in M phase, 3 μM nocodazole was added to the media. Nuclear and cytoplasmic fractionations of TS cells were extracted using an NE-PER extraction kit according to the manufacturer's protocol (PIERCE).
Paraffin sections were treated with buffer containing 0.1 M Tris-HCl and 0.1 M EDTA (pH 8.0) plus 1 μg/ml proteinase K for 30 min, and washed with the same buffer without proteinase K for 5 min at 37 °C. Samples were then incubated with buffer containing 0.2 M Tris-HCl (pH 8.0) and 0.1 M glycine for 10 min at room temp, followed by post-fixing with 4% paraformaldehyde in PBS buffer for 20 min and a 20-min wash in PBS buffer at room temperature. The sections were incubated in buffer containing 0.1 M triethanolamine (pH 8.0) for 10 min, followed by 0.25% (v/v) acetic anhydride in 0.1 M triethanolamine (pH 8.0) buffer for 10 min and by 2× SSC (1× SSC: 0.15 M sodium chloride and 15 mM sodium citrate, pH 5.5) buffer for 10 min. After dehydration through ethanol gradients and air drying for 2 h, sections were incubated with digoxygenin-labeled probes (1 μg/ml) in 5× SSC buffer containing 50% formamide, 50 μg/ml yeast tRNA and 1% SDS overnight at 70 °C. Samples were then washed three times with 5× SSC buffer for 15 min at 70 °C and 2× SSC buffer containing 50% formamide for 10 min at 45 °C before incubating with buffer containing 20 μg/ml RNase A, 5 U/ml RNase T1, 0.5 M sodium chloride, 10 mM Tris (pH 8.0) and 1 mM EDTA (pH 8.0) for 30 min at 37 °C. After washing with 2× SSC for 10 min at 37 °C and 0.1× SSC for 10 min at 45 °C, samples were incubated in MBST buffer containing 60 mM maleic acid, 0.15 M sodium chloride, and 0.1% Tween-20, pH 7.5 for 10 min and blocked with 10% goat serum in MBST for 2 h at room temperature. After incubating with anti-digoxygenin antibody (Roche) in the blocking buffer for overnight at 4 °C, sections were washed with NTMT buffer (100 mM sodium chloride, 100 mM Tris, pH 9.5, 50 mM magnesium chloride and 0.1 % Tween 20) and incubated in NTMT plus 2 mM levamisole overnight at 4 °C. To visualize the bound signals, samples were incubated with BM-purple (Roche) for 2 h to several days. The reaction was stopped by incubating in PBS buffer, followed by counterstaining with nuclear fast red.
Samples were fixed, paraffin embedded, sectioned, and stained with hematoxylin/eosin for histological evaluation as described [57]. Tissue sections were subject to immunological staining with avidin:biotinylated enzyme complex as described [18,58]. Proteins were extracted from TS cells using M-PER reagent (PIERCE) with the addition of protease inhibitor cocktail (Sigma-Aldrich), 1 mM sodium molybdate, 1 mM sodium vanadate, and 10 mM N-ethylmaleimide, or SDS lysis buffer (2% SDS, 10% glycerol, and 50 mM Tris, pH 6.8). Protein extracts were subject to immunoblotting as described [54]. Bound primary antibodies were detected with horseradish peroxidase-conjugated secondary antibodies (Vector Lab), followed by ECL-mediated visualization (GE HealthCare) and autoradiography. Mouse monoclonal antibodies anti-actin (Thermo Fisher; 1:1,000), anti-BrdU (Thermo Fisher; 1:300), anti-Cdx2 (BioGenex; 1:1), anti-MDM2 (Santa Cruz; 1:100), and anti-SUMO-1 (Zymed; 1:2,000); rabbit polyclonal antibodies anti-calnexin (Stressgene; 1:2,000), anti-cyclin D1 (Neomarker; 1:100), anti-Ki67 (Neomarker; 1:400), anti-laminin (Sigma-Aldrich; 1:25), anti-Myc tag (CalBioChem; 1:400), anti-Oct4 (Santa Cruz; 1:200), anti-p53 (Santa Cruz; 1:50), and anti-p450scc (Chemicon; 1:200); and goat polyclonal antibody anti-lamin B (Santa Cruz; 1:100) were used as primary antibodies. BrdU incorporation analysis was performed by intraperitoneal injection of BrdU (250 μg/g of body weight) into pregnant females for 1 h. Placentas were recovered, fixed, embedded, sectioned, and subject to immunostaining as described [18,57].
|
10.1371/journal.pntd.0005912 | Use of a novel antigen expressing system to study the Salmonella enterica serovar Typhi protein recognition by T cells | Salmonella enterica serovar Typhi (S. Typhi), the causative agent of the typhoid fever, is a pathogen of great public health importance. Typhoid vaccines have the potential to be cost-effective measures towards combating this disease, yet the antigens triggering host protective immune responses are largely unknown. Given the key role of cellular-mediated immunity in S. Typhi protection, it is crucial to identify S. Typhi proteins involved in T-cell responses. Here, cells from individuals immunized with Ty21a typhoid vaccine were collected before and after immunization and used as effectors. We also used an innovative antigen expressing system based on the infection of B-cells with recombinant Escherichia coli (E. coli) expressing one of four S. Typhi gene products (i.e., SifA, OmpC, FliC, GroEL) as targets. Using flow cytometry, we found that the pattern of response to specific S. Typhi proteins was variable. Some individuals responded to all four proteins while others responded to only one or two proteins. We next evaluated whether T-cells responding to recombinant E. coli also possess the ability to respond to purified proteins. We observed that CD4+ cell responses, but not CD8+ cell responses, to recombinant E. coli were significantly associated with the responses to purified proteins. Thus, our results demonstrate the feasibility of using an E. coli expressing system to uncover the antigen specificity of T-cells and highlight its applicability to vaccine studies. These results also emphasize the importance of selecting the stimuli appropriately when evaluating CD4+ and CD8+ cell responses.
| Salmonella enterica serovar Typhi (S. Typhi) is the causative agent of the life-threatening typhoid fever that affects 11.9–20.6 million individuals annually in low-income and middle-income countries. The T-cells, CD4+ and CD8+ T cells, play a significant role in protection against S. Typhi infection. Yet, the antigens triggering host protective immune responses recognized by these cells are largely unknown. To address this shortcoming, in this study we used an E. coli expression system methodology for identifying immunogenic proteins of S. Typhi. We found that although the pattern of response to individual S. Typhi proteins was variable among the typhoid vaccinees, the E. coli expressing system uncovered the antigen specificity of T-cells, and highlight its applicability to vaccine studies.
| Typhoid fever is caused by Salmonella enterica serovar Typhi (S. Typhi), a human-restricted pathogen that enters the host through the gut-associated lymphoid tissue. Recent calculations of the typhoid burden estimated that 11.9–20.6 million new cases of typhoid fever occur annually in low-income and middle-income countries with about 129,000–223,000 mortality [1–4]. Based on data provided by the World Health Organization, 90 percent of these typhoid deaths occur in Asia, and most victims are children under five years of age [5]. Furthermore, antimicrobial treatment of enteric fever and asymptomatic carriers has become increasingly complicated due to the emergence of multidrug-resistant strains of S. Typhi [6–8]. Thus, there has been an increased emphasis on control measures, such as vaccination to fight S. Typhi infection [9, 10]. It has also become evident that a better understanding of the host immune responses against S. Typhi will be required to achieve this task. Currently, two typhoid vaccines are licensed in the USA for use in humans, the purified Vi (“virulence”) parenteral polysaccharide vaccine and the oral live attenuated S. Typhi strain Ty21a vaccine. Although these vaccines are moderately protective and able to induce herd immunity [11, 12], they also have some significant shortcomings. Since Vi is a T-cell independent antigen, Vi vaccine does not confer “memory,” and there are no robust data to suggest that the efficacy of Vi persists beyond three years [11, 13, 14]. The Ty21a vaccine, which does not elicit anti-Vi antibodies, requires the administration of three to four doses spaced at 48-hour intervals [12, 13, 15]. Moreover, recently, Vi-protein-conjugate vaccines that consist of the S. Typhi Vi polysaccharide covalently bound to a carrier protein have been developed [5, 16–19]. However, issues have been raised about selective pressure for the development and spread of S. Typhi Vi antigen-negative strains due to the generalized use of Vi and Vi-protein-conjugate vaccines [20, 21]. As a result, novel approaches to typhoid vaccination are critically needed [22].
It is now widely accepted that cellular-mediated immunity (CMI) plays a significant role in protection against S. Typhi infection [8]. These host responses rely mainly on two types of T-cells, CD4+ and CD8+ T cells [23–26]. The presence of both CD4+ helper T-cells and classical class Ia and non-classical HLA-E-restricted S. Typhi-specific CD8+ T cells have been observed in individuals who recover from typhoid fever [25] or immunized with Ty21a and other attenuated leading typhoid vaccine candidates, including CVD 906, CVD 908, CVD 908-htrA and CVD 909 [26–33]. Moreover, our group recently provided the first evidence that S. Typhi-specific CD8+ responses correlate with clinical outcome in humans challenged with wild-type S. Typhi [34]. However, the antigen specificity of these T cells remains largely unknown. Moreover, most of the S. Typhi proteins described as being involved in human protection have been derived from studies using mouse models of Salmonella infection [35, 36]. One of the reasons for this is the inherent problems of working with humans as experimental models.
Here, we used an innovative antigen expressing system, originally developed by the Higgins laboratory [37, 38] and based on the infection of B-cells with recombinant E. coli to evaluate T cell responses to four S. Typhi proteins: SifA, FliC, GroEL, and OmpC (Table 1). These proteins are known to confer survival properties to Salmonella and therefore might be evaluated as vaccine antigens [27, 39–44]. Briefly, in this system, EBV-transformed lymphoblastoid B-cell lines (B-LCL) were used as antigen-presenting cells (APCs). These B-LCL were infected with E. coli expressing both S. Typhi proteins and cytoplasmic listeriolysin O (Hly). Hly is a pore-forming hemolysin from Listeria monocytogenes, which allows leakage of E. coli antigen from the phagolysosomal compartment into the APC cytosol, there gaining access to the MHC class I antigen processing and presentation pathway [37, 38]. This system also allows the identification of S. Typhi-specific CD4+ T cell as the expression on E. coli also results in antigen presentation in the context of MHC class II molecules [45]. Additionally, this approach has the advantage of assessing T-cell responses to full-length proteins before initiating more expensive and time-consuming procedures, such as synthesizing overlapping peptides [46]. Due to HLA diversity in humans, host responses to subunit vaccines have a greater chance to be successful if they encompass specific protein antigens rather than specific epitopes within those proteins [45, 46].
By using this innovative antigen expression system, we found that the pattern of response to individual S. Typhi proteins was variable. Some individuals responded to all four proteins while others responded to only one or two proteins. When comparing T cells responses to B-LCL exposed to recombinant E. coli to those to purified proteins from the same genes, we observed that the CD4+ cell responses, but not CD8+ cell responses, to recombinant E. coli were significantly associated with the responses to purified proteins. Thus, our results demonstrate the feasibility of using an E. coli expressing system to uncover the antigen specificity of T-cells, and highlight its applicability to vaccine studies. These results also emphasize the importance of selecting the stimuli appropriately when designing experiments aimed at evaluating CD4+ and CD8+ cell responses.
To show the feasibility of our E. coli expressing system, we evaluate four S. Typhi proteins (i.e., SifA, FliC, GroEL, and OmpC) (Table 1) known to confer survival properties to Salmonella then potentially promising as vaccine antigens [27, 39–44]. As shown in Fig 1, proper E.coli protein expression for all four proteins, SifA, OmpC, FliC, and GroEL, as well as the Hly was detected by Western blot.
We next evaluated the effect of the recombinant E. coli infection on B-LCL viability. Briefly, we assessed cell viability by measuring the levels of Yevid viability staining on 2-hour-E. coli infected B-LCLs that have been rested overnight in the presence of gentamicin. As shown in Fig 2A, regardless of the protein being expressed, the infection did not adversely affect the viability of E. coli-infected B-LCLs. After infection, the percentage of live cells in cultures with recombinant E. coli was comparable to control cultures with media only (uninfected). By using the same experimental conditions as for determinations of cell viability, we also detected the expression of bacterial antigens on B-LCLs. Similarly to the viability, regardless of the type of protein being expressed in the recombinant E. coli, we found similar levels of E. coli-expressing cells as assessed by surface staining with anti-E. coli antigen polyclonal antibody using flow cytometry (Fig 2B–2D).
As described above, we reasoned that Hly should promote the phagosomal escape of bacterial antigens thereby improving MHC class I processing of S. Typhi antigens presented by B-LCLs and hence recall immune responses from both CD4+ and CD8+ primed T cells [8, 26, 27, 29, 30, 39, 47]. To test this assumption, Hly-recombinant E. coli strain BL21, or wild type E. coli strain BL21 were used to infect B-LCL cells. Cells were infected for 2 hours using two different multiplicity of infection (MOI, 1:30 and 1:100). After 2 hours, cells were collected, washed to remove extracellular bacteria and cultured in the presence of gentamicin for 2 additional hours. Thus, the ability to detect E. coli proteins in B-LCL infected cells was assessed over time by flow cytometry (up to 120 minutes) using polyclonal anti-E. coli antibodies. As shown in Fig 3, at all-time points evaluated, we observed higher expression of E. coli antigens on B-LCL cells infected with the recombinant E. coli strain expressing Hly as compared to the wild-type E. coli strain. Thus, the hly gene appears functional. These results are very significant since based on our previous study [48], we expect to see background responses against the vector itself (E. coli antigens). Further antigen expression might help to better discriminate T-cell responses to S. Typhi antigens from those directed to E. coli antigens.
In order to demonstrate the feasibility of using an E. coli expression system to uncover the antigen specificity of T-cells, PBMC obtained before and 42 days after immunization were exposed to autologous B-LCL infected with recombinant E. coli expressing Hly only or co-expressing one of the four Salmonella gene products: Hly/SifA, Hly/OmpC, Hly/FliC and Hly/GroEL. Specifically, we studied the ability of ex-vivo PBMC from seven Ty21a-immunized volunteers to express IL-17A, IFN-γ and TNF-α cytokines and/or CD107a and b molecules against autologous infected targets. T-cell responses (i.e., CD4+ and CD8+ T-cells) were evaluated by multichromatic flow cytometry using a 10-color surface/intracellular staining panel. Unstimulated and Staphylococcus enterotoxin B (SEB)-stimulated effector cells were used as negative and positive controls, respectively. We observed that the pattern of response to individual S. Typhi proteins was variable, with some individuals responding to all four proteins while others were responding to only one or two proteins. We also observed differential CD4+ and CD8+ T cells responses to the S. Typhi proteins (Figs 4–7). In six individuals, although the magnitude of responses varied considerably, both CD4+ and CD8+ T cells responded to at least one protein. In one individual, we were unable to detect CD4+ T cells responses to any of the four protein, but CD8+ T cells responded to 3 out of 4 proteins tested (Fig 8). Representative responses from selected volunteers are shown in Figs 4–6.
Because previous results from our group have shown that multifunctional T-cells might contribute to S. Typhi immunity [30, 34, 49], we then investigated the multi-functionality patterns of CD4+ T-cells and CD8+ T-cells after exposure to infected B-LCL infected with recombinant E. coli. We measured simultaneously four T-cell functions (i.e., expression of IL-17A, IFN-γ and TNF-α cytokines, or CD107a and b molecules) by multichromatic flow cytometry using the FCOM feature of the WinList software, which provides the % of T-cells expressing each of the possible function combinations. Analyses of multiple function patterns (i.e., single, double, triple or quadruple functions) revealed that, albeit different for different proteins, both CD4+ T-cells and CD8+ T-cells were multi-functional (Fig 8).
An important hypothesis to evaluate is whether the T-cells from volunteers who responded to B-LCL cultured with recombinant E. coli expressing Salmonella gene products also possess the ability to respond to exogenous proteins. To this end, we next investigated whether there was an association between T-cell responses after exposure to autologous B-LCL cultured with recombinant E. coli expressing one of the four Salmonella gene products (Hly/SifA, Hly/OmpC, Hly/FliC and Hly/GroEL) and those T-cell responses after exposure to B-LCL cultured with individual purified recombinant proteins (SifA, OmpC, FliC, and GroEL). To this end, B-LCLs were cultured overnight with purified SifA, OmpC, FliC or GroEL proteins, and used as targets for T-cells. Representative responses of CD8+ and CD4+ cells from one volunteer are shown in Figs 9 & 10 respectively. We found that the CD4+ cell responses to B-LCLs cultured with exogenous proteins were significantly associated with the CD4+ cell responses to B-LCLs cultured with recombinant E. coli expressing one of the four S. Typhi gene products (Fig 11; p = 0.0111, Pearson Product Moment Correlation). However, no significant association was observed between CD8+ cell responses to B-LCLs cultured with recombinant E. coli expressing one of the four Salmonella gene products and CD8+ cell responses to B-LCLs cultured with exogenous recombinant proteins (Fig 11; p = 0.0790, Pearson Product Moment Correlation). On the other hand, CD8+ cell responses to B-LCLs cultured with exogenous proteins were consistently higher than CD4+ cell responses (Figs 9 and 10 & S1 Table). Thus, as expected, CD4+ and CD8+ cell responses against S. Typhi antigens depend on the nature of the stimulant [39, 50]. These results emphasize the importance of selecting the stimuli appropriately when designing experiments aimed at evaluating CD4+ and CD8+ cell responses cell responses [8].
One of the characteristics that make the E. coli expression system methodology employed in this manuscript highly relevant for identifying immunogenic proteins of S. Typhi is the use of a translational research approach using human T cells and autologous APC to identify S. Typhi-specific T-cell immune responses. Most published methods have relied heavily on “proof-of-concept” studies performed in mice. However, S. Typhi is a human-restricted pathogen and there are no good animal models that faithfully recapitulate S. Typhi infection [51]. To partially address this shortcoming, the infection of susceptible mice with S. Typhimurium has been used as a model for the pathogenesis of human typhoid fever [51]. Although these murine models have provided considerable knowledge regarding host-pathogen interactions, they do not adequately recapitulate S. Typhi infection in humans [52]. The recent availability of full genome sequences from various Salmonella enterica serovars has uncovered many differences in pseudo genes which can explain, at least in part, the dissimilarities observed in the immune and other host responses to these enteric bacteria [52]. Thus, samples from human participants exposed to the licensed Ty21a oral typhoid vaccine have the potential to provide a better characterization of key Salmonella antigens involved in T cell responses than murine models.
The main novelty of our system is that we engineered the hly gene encoding the pore-forming cytoplasmic listeriolysin O (Hly) protein onto the backbone of the recombinant protein expression plasmid pET-DEST-Hly (see methods). In contrast, Higgin’s E. coli expressing system [37, 38] used two expression plasmids inside individual E. coli cells: one expressing Hly and the other expressing the recombinant protein of interest. The reason for this difference stems from our preliminary results showing that the use of two expression plasmids reduced B-LCL infectivity (S1 Fig). We speculate that this difference is due to the fitness cost for E. coli to maintain, replicate and propagate two plasmids instead of one. There may also have been a negative effect on bacterial cell preparations resulting from growth on media containing two selective antibiotics instead of one.
Herein, using this methodology, we found that all the tested individuals had increased T-cell responses over baseline (before immunization) to at least one of the four S. Typhi proteins evaluated (i.e., SifA, OmpC, FliC, and GroEL). Moreover, multifunctional CD4+ and CD8+ T cells that expressed two or more cytokines (IL-17A, IFN-γ and TNF-α) and/or CD107a/b molecules were detected. These results are particularly significant since we have previously demonstrated that these two T-cell population might play a role in controlling Salmonella infection [39]. These results also support previous data showing that the depletion of either CD4+ or CD8+ T-cells had impaired recall immunity to oral challenge with the virulent S. Typhimurium at different times after vaccination [53].
The reason underlying the observation that the responses to the different antigens are variable among the vaccinees are unclear. However, it is reasonable to speculate that this phenomenon might be due to the HLA‐haplotype variation between individuals. In fact, antigen processing, together with defined MHC genes, are known to shape the individual immune responses to a wide array of pathogens [54]. Furthermore, the differential responsiveness among the participants supports the development of multi-component vaccines by introducing many antigenic determinants into vaccine formulations. Alternatively, these results might encourage a renewed focus on whole-cell live attenuated preparations, especially since they may overcome some of the inherent weaknesses associated with sub-unit vaccines such as the need for considerable amounts of antigens and the use of adjuvants.
We also observed that the magnitude of the responses against S. Typhi SifA, OmpC, FliC and GroEL varied among participants. Since S. Typhi GroEL has a significant degree of homology with self-heat shock proteins in humans [27], these results provide additional information that T-cells can discriminate between self and foreign antigens during the immune response. Interestingly, in contrast to CD8+ cell responses, we found that the CD4+ cell responses to B-LCLs cultured with exogenous proteins were significantly associated with the CD4+ cell responses to B-LCLs cultured with recombinant E. coli expressing S. Typhi genes. These results confirm and extend previous findings from our group and others that the balance of CD4+ and CD8+ cell responses against S. Typhi antigens are likely to depend on the nature of the stimulant [8, 39, 50]. Indeed, previous work from our group has shown that CD4+ cells respond differently to soluble antigen stimulation than CD8+ T cells [39, 50]. Future studies will be directed to use this novel antigen discovery system to evaluate in humans the immune responses to other S. Typhi proteins expressed in this E. coli expression system.
Since these studies were performed in immunized volunteers, it is also important to note that the relative contributions of CD4+ and CD8+ cell responses against these S. Typhi proteins in protection cannot be ascertained. We are directly addressing this critical issue in separate studies in which we are evaluating whether responses to these S. Typhi proteins correlate with protection in volunteers who have been immunized with Ty21a and subsequently challenged with wild-type S. Typhi. Nevertheless, the results presented herein demonstrate the feasibility of using a novel antigen discovery platform. This system could be used, for example, to systematically assess the specificity of T-cell immune responses against the entire proteome of a human pathogen and generate a database of the repertoire of these human T-cell antigen specificities. This should help narrow down the proteins of interest which correlate with defined phenotypes (e.g., responses associated with protection in human challenge studies with S. Typhi, those directed to pathogenic determinants), ultimately leading to the identification of candidate vaccine antigens. Finally, these results emphasize the importance of selecting the appropriate antigens when designing experiments aimed at evaluating CD4+ and CD8+ cell responses.
The human experimentation guidelines of the US Department of Health and Human Services and those of the University of Maryland, Baltimore, were followed in the conduct of the clinical research. All blood specimens were collected from volunteers who participated in the University of Maryland Institutional Review Board approved protocol number HP-00040022 that authorized the collection of blood samples from healthy volunteers for the studies included in this manuscript. Volunteers were explained the purpose and possible consequences of participating in this study and gave informed, signed consent before the blood draws. This protocol has been conducted in accordance with the ethical standards laid down in the 1964 Declaration of Helsinki and the principles of the International Conference on Harmonization Good Clinical Practice guidelines [55].
Seven healthy adult volunteers, 20–50 (39 ± 10) years old, recruited from the Baltimore-Washington area and the University of Maryland at Baltimore campus, participated in this study. They were immunized with four spaced doses of 2–6 x 109 CFU of oral live attenuated Ty21a at an interval of 48 hours between doses [12, 56]. Blood collection was performed before and 42 days after Ty21a immunization. Peripheral blood mononuclear cells (PBMC) were isolated from the blood by density gradient centrifugation and cryopreserved in liquid N2 following standard techniques [28]. These PBMC were used ex vivo as effector cells or to prepare the target cells.
The entire S. Typhi strain Ty2 ORFeome was constructed as a part of the former National Institute of Allergy and Infectious Disease (NIAID)-funded Pathogen Functional Genomics Resource Center (PFGRC). The PFGRC provided a library of entry clones by cloning 3,381 ORFs, out of the 4,323 ORFs predicted in the Ty2 genome (78%), into the pDONR 221 vector using methods described in Peterson et al. [57]. Entry clones can then be shuttled into a variety of destination vectors using in vitro site-specific recombination (http://www.ncbi.nlm.nih.gov/pubmed/11076863). For expression of the S. Typhi proteins, we engineered the pET161-DEST destination vector (Invitrogen, Carlsbad, CA). The pET161-DEST, a commercial plasmid from Invitrogen, is a robust E. coli destination vector that has been used for several years and carries many desirable characteristics for this study. Gene expression is under the tight control of the T7 promoter and lac repressor, and an optimized ribosome binding site is provided by the inserted gene. The vector contains the V5 and His-tag epitopes downstream of the attR sites into which inserts from entry clones were recombined resulting in the addition of these tags as C-terminal fusions to the S. Typhi protein. These tags were used to monitor recombinant protein expression and, if necessary, for protein purification. pET161-DEST also contains the ccdB and chloramphenicol (CmR) genes which help the recovery of correct recombination products after the LR recombinase reaction via selection on ampicillin agar plates. We also inserted the hly gene coding for the cytoplasmic Hly protein from L. monocytogenes into the BglII restriction site. The hly gene was PCR-amplified with a His-tag at the 3’end of the reverse primer, and a BgIll site was added to the forward 5’ end, and reverse 3’ end of each primer, respectively (sense: GCGCAGATCTAGCAAGCATATAATATTGCG, anti-sense: GCGCAGATCTTTAGTGATGGTGATGGTGATGTTCGATTGGATTATCTAC). The resulting vector pET-DEST-Hly was verified by Sanger sequencing. It carries a T7 promoter and the lac operator preceding the gene to be cloned; the cloning site harbors attR1 and attR2 site for cloning from entry clones through the LR clonase reaction (see below). The plasmid was transformed into One Shot ccdB survival competent cells (Invitrogen).
Entry clone glycerol stocks were streaked on LB agar with 50 μg/ml kanamycin and incubated at 37°C overnight. Single colonies were incubated overnight at 37°C in 4ml LB broth with 50 μg/ml kanamycin. Then the entry clone plasmid was extracted using the QIA prep kit (Qiagen, Valencia, CA) and DNA concentration measured using a NanoDrop instrument (Thermo Scientific, Waltham, MA). The pET-DEST-Hly glycerol stock was streaked on LB agar plates with 100 μg/ml carbenicillin and incubated at 37°C. After overnight incubation, single colonies were isolated, incubated overnight in 50 ml LB broth with 100 μg/ml carbenicillin at 37°C and the destination plasmid was extracted with the HiSpeed plasmid preparation kit (Qiagen). DNA concentration was measured by NanoDrop. We chose to evaluate four S. Typhi proteins: SifA, FliC, GroEL, and OmpC (Table 1). Each of these four S. Typhi proteins was then transferred in vitro from the entry clone plasmid into the destination plasmid through LR clonase reactions (Gateway LR clonase II Enzyme mix, Invitrogen) following the manufacturer’s protocol. We also cloned a short non-coding sequence called “pmark” that carried stop codons in all six frames of translation into pET-DEST-Hly as a negative control of protein expression. Finally, 3 μl of LR clonase reaction product was transformed into 50 μl of One Shot BL21(DE3) competent cells (Invitrogen) by chemical transformation.
Proper cloning of the target S. Typhi gene was confirmed by PCR amplification using the T7 forward and reverse primers and sequencing of the PCR products. The presence of the hly gene was verified by BgIll digestion of the plasmid.
A single colony from verified BL21 (DE3) expression vector clones was cultured at 37°C overnight in 2 ml LB broth with 100 μg/ml carbenicillin. Then 30 μl of the overnight culture was inoculated to 3 ml fresh LB broth with 100 μg/ml carbenicillin. After ~4h of incubation at 37°C (OD~0.6), the culture was induced for protein expression with 100 μM IPTG and incubated for an additional 2 h. Bacteria were then spun down at 2,800 rpm for 10 min and the supernatant discarded. E. coli were lysed with of 0.3 μg/ml lysozyme in Tris-HCl buffer, pH 7.5 (Thermo Scientific), then proteins were heat-denatured at 100°C for 10 minutes and evaluated on SDS-PAGE gels by Coomassie staining. Specificity of protein expression was confirmed by western blot using mouse anti-His monoclonal antibodies (Sigma, St. Louis, MO), and rabbit anti-mouse IgG conjugated to horseradish peroxidase (HRP) (Sigma) as secondary antibodies; as well as rabbit anti-Listeriolysin polyclonal antibodies (Abcam, Cambridge, MA, USA), and goat anti-rabbit IgG-HRP (Millipore, Billerica, MA) as secondary antibodies. His-tagged Annexin 3 (A3) protein (Abcam) was used as positive control for antibody detection.
PBMC obtained from Ty21a vaccines before immunization were used to generate autologous Epstein-Barr Virus (EBV)-transformed B-LCLs as previously described [27, 31]. Briefly, B-LCLs were established by using B95-8 cell line (ATCC# CRL1612) supernatants as the EBV source. After transformation, B-LCL were maintained in culture in RPMI 1640 (Gibco, Grand Island, New York) supplemented with 100 U/ml penicillin, 100 μg/ml streptomycin, 50 μg/ml gentamicin, 2 mM L-glutamine, 2.5 mM sodium pyruvate, 10 mM HEPES buffer and 10% heat-inactivated fetal bovine serum (R10) or cryopreserved until used in the experiments.
Target/stimulator cells were infected as previously described [27–30, 39, 47, 48] with slight modifications. Briefly, target cells were infected by incubation in RPMI (without antibiotics) at 37°C for 2 hours with any of the recombinant E. coli strains at 1:30 or 1:100 multiplicity of infection (MOI). After incubation, cells were washed and incubated for an additional 2 or 16–18 hours (overnight) in complete R10 containing gentamicin (100 μg/ml) to kill extracellular and/or to detach cell-bound bacteria. For co-culture experiments, targets were then gamma-irradiated (6,000 rads), surface stained with anti-CD45, a marker abundantly expressed on the surface of hematopoietic cells [58], and used as stimulators. To confirm E. coli infection, aliquots of targets were surface stained with rabbit anti-E. coli antigen polyclonal antibody (1:1000, Abcam).
Four Salmonella purified proteins were tested: SifA, FliC, GroEL, and OmpC (Table 1). The region encoding residues 53–450 of FliC were subcloned from S. Typhi ISP1820 into pET15b. The plasmid was used to transform E. coli Tuner (DE3) for overexpression after being induced with IPTG. The overexpressed protein was purified by standard immobilized metal affinity column chromatography (IMAC) methods. OmpC was purified based on the protocol of Kumar and Krishnaswamy [59]. The sequence encoding OmpC from S. Typhi ISP1820 was subcloned into pET15b, which was used to transform E. coli Tuner (DE3). Overexpression was induced using IPTG, and the OmpC was purified from inclusion bodies using IMAC after solubilizing in buffer containing urea. Refolding was then allowed to occur in 50 mM Tris, pH 8.5, 0.1 M NaCl, 10% (v/v) glycerol and 0.2% (v/v) polyoxyethelene-9-lauryl ether. The sequence for GroEL was introduced into pBAD TOPO TA such that it was in-frame. The protein was overexpressed by the addition of arabinose, and the HT-GroEL purified initially by standard IMAC, which provides good yields at high purity. Removal of contaminating peptides could be achieved by adding acetone to 45% (v/v) to precipitate the GroEl which could then be collected by centrifugation followed by solubilizing using PBS (10 mM phosphate, pH 7.4, 150 mM NaCl). The sequence encoding SifA from Salmonella Typhi ISP1820 was subcloned into pET15b and used to transform E. coli c-41 (DE3). The bacteria were grown to an absorbance of about 0.8 at 600 nm, protein expression induced with IPTG, and the cultures immediately moved to 20°C for overnight growth. The bacteria were then lysed and the SifA purified by standard IMAC methods.
B-LCLs were incubated overnight in 24-well plates at a density of 2 x 106/ml/well in the absence or presence of 1 μg/ml of each of the purified proteins at 37°C in a 5% CO2 atmosphere. After incubation, cells were washed and used as stimulators for T-cells.
Cells were stained with monoclonal antibodies (mAbs) to CD69 (clone TPI-55-3) (Beckman-Coulter, Miami, FL), CD4 (clone RPA-T4), CD8 (clone HIT8a), CD107a and b (clones H4A3 and H4B4 respectively), interferon (IFN)-γ (clone B27), tumor necrosis factor (TNF)-α (clone MAb11) (BD Pharmingen, San Diego, CA, USA), CD14 (clone TuK4), CD19 (clone SJ25-C1), CD45 (clone H130) (Invitrogen), interleukin (IL)-17A (clone eBio64DEC17) (eBioscience, San Diego, CA), and CD3 (clone OKT3)(Biolegend, San Diego, CA). Antibodies conjugated to the following fluorochromes were used in these studies: Fluorescein isothiocyanate (FITC), PE-Cy5.5, PE-Cy7, V450, Brilliant Violet (BV)570, BV650, Energy Coupled Dye or PE-Texas-Red conjugate (ECD), allophycocyanin (APC)-Alexa 700 and Quantum Dot (QD) 800.
Ex vivo PBMC from immunized volunteers collected before and 42 days after immunization were used as effectors as previously described [48]. Briefly, PBMC were co-cultured with autologous B-LCL cells at an effector to stimulator cell ratio of 5:1 in the presence of mAbs to CD107a and CD107b (15 μl of each/1 x 106 cells in 500 μl of R10 medium). The CD107 a and b antibodies were used to measure degranulation, a mechanism essential for the killing of S. infected targets by the cytotoxic CD8+ cells [60]. PBMC cultured with uninfected target cells or Staphylococcus enterotoxin B (SEB) (10 μg/ml, Sigma) were used as negative and positive controls, respectively. After ~2 hours of stimulation, protein transport blockers, Monensin (1 μg/ml, Sigma) and brefeldin-A (BFA) (2 μg/ml, Sigma), were added to the co-culture. After overnight (16–18 hours) incubation, cells were harvested, stained with a dead-cell discriminator, yellow fluorescent viability dye (Yevid, Invitrogen)[61, 62], followed by surface staining with mAbs against surface antigens (CD3, CD4, CD8, CD14, and CD19) and fixation and permeabilization with Fix & Perm cell buffers (Invitrogen, Carlsbad, CA). Cells were then stained intracellularly for IFN-γ, TNF-α, IL-17A and CD69. Finally, cells were fixed and analyzed by flow cytometry on an LSR-II instrument (BD Biosciences). Data were analyzed with WinList 7.0 (Verity Software House, Topsham, ME). Lymphocytes were gated based on their scatter characteristics. Single lymphocytes were gated based on forward scatter height vs. forward scatter area. A “dump” channel was used to eliminate dead cells (Yevid+) as well as macrophages/monocytes (CD14+), B lymphocytes (CD19+) and targets (CD45+) from analysis. Additional gating on CD3, CD4, and CD8 was performed to identify cytokine-producing (IFN-γ, TNF-α and IL-17A) and CD107 expressing T cell subsets. Net responses were calculated by subtracting the number of positive events of the experimental (Salmonella-Hly proteins) from the negative control (Hly only). Functional responses were considered specific for S. Typhi if the differential in the number of positive and negative events between experimental (Salmonella-Hly proteins) and negative control (Hly only) cultures were significantly increased (P < 0.01) using Z-test. Volunteers were considered responders if the net responses from the PBMC collected 42 days after immunization were greater than 0.1% from the net responses of PBMC collected before immunization.
All statistical tests were performed using Prism software (version 7, GraphPad Software, La Jolla, CA). Comparisons between two groups were carried out by Student’s t tests. Correlation analysis was achieved by Pearson Product Moment Correlation tests. P values <0.05 were considered significant.
|
10.1371/journal.pcbi.1003253 | Target Prediction for an Open Access Set of Compounds Active against Mycobacterium tuberculosis | Mycobacterium tuberculosis, the causative agent of tuberculosis (TB), infects an estimated two billion people worldwide and is the leading cause of mortality due to infectious disease. The development of new anti-TB therapeutics is required, because of the emergence of multi-drug resistance strains as well as co-infection with other pathogens, especially HIV. Recently, the pharmaceutical company GlaxoSmithKline published the results of a high-throughput screen (HTS) of their two million compound library for anti-mycobacterial phenotypes. The screen revealed 776 compounds with significant activity against the M. tuberculosis H37Rv strain, including a subset of 177 prioritized compounds with high potency and low in vitro cytotoxicity. The next major challenge is the identification of the target proteins. Here, we use a computational approach that integrates historical bioassay data, chemical properties and structural comparisons of selected compounds to propose their potential targets in M. tuberculosis. We predicted 139 target - compound links, providing a necessary basis for further studies to characterize the mode of action of these compounds. The results from our analysis, including the predicted structural models, are available to the wider scientific community in the open source mode, to encourage further development of novel TB therapeutics.
| Mycobacterium tuberculosis is a major worldwide pathogen infecting millions individuals every year. Additionally, the number of antibiotic resistant strains has dramatically increased over the last decades. Trying to address this challenge, the pharmaceutical company GlaxoSmithKline has recently published the results of a large-scale high-throughput screen (HTS) that resulted in the release of 776 chemical compound structures active against tuberculosis. We have used this dataset of compounds as input to our computational approach that integrates historical bioassay data, chemical properties and structural comparisons. We propose 139 targets alongside their respective hit compounds and made them open to the wider scientific community. Our hope is that the availability of the experimental data from GSK and our computational analysis will encourage further research providing validated therapeutically targets against this devastating disease.
| One third of the world's population is infected with Mycobacterium tuberculosis (MTB), the causative agent of tuberculosis [1]. Approximately 95% of infected individuals are thought to have persistent, latent MTB infections that remain dormant until activated by specific environmental and host response events. Approximately 10% of latent infections eventually progress to active disease, which, if left untreated, kills more than half of the infected patients [2]. Moreover, there is an increasing clinical occurrence of MTB strains with extensive multi-drug-resistance (eg, MTB MDR and MTB XDR), where mortality rates can approach 100% [3]. In some countries, the MTB MDR and XDR strains may account for up to 22% of infections [1]. In addition, current TB therapeutic regimes involve a combination of antibiotics, administered at regular intervals over a 6-month period, which makes patient compliance an issue, especially in developing countries [1], [2].
The discovery and development of new antibiotics is widely recognized as one of the major global health emergencies, yet it is also a major pharmaceutical challenge. Most currently used antibiotics were discovered during the golden era from the 1940s to 1960s through large scale screening of compound collections for anti-bacterial activity – the so-called whole cell or phenotypic screens [4]. The emergence of bacterial molecular genomics technologies and the availability of whole genome sequences in the 1990s led to dramatic changes in anti-bacterial drug discovery, where the emphasis was placed on screening essential targets for inhibitory compounds. However, despite intensive efforts, target-based screening has been largely unsuccessful in producing clinical candidate molecules [5]. As a result, a return to whole cell screening has been widely advocated, in combination with novel technologies and bioinformatics to rapid identify targets associated with a compound's mechanism of action (MOA) [4], [6].
Recently, the pharmaceutical company GlaxoSmithKline (GSK) completed an anti-mycobacterial phenotypic screening campaign against M. bovis BCG, a non-virulent, vaccine Mycobacterium strain, with a subsequent secondary screening in M. tuberculosis H37Rv (MTB H37Rv) for hit confirmation [7]. A total of 776 potent compound hits (including 177 MTB H37RV hits with limited human cell line toxicity) were made openly available to the wider scientific community through the ChEMBL database (http://dx.doi.org/10.6019/CHEMBL2095176). The aim of this release was to stimulate mechanism of action analyses using chemical genetics/proteomics approaches, as well as to provide many potential new starting points for synthetic lead generation activities. To attain these goals, it is essential to identify the likely protein targets of these active compounds. Here, we introduce an integrative computational analysis towards the genome-wide characterization of targets for selected compounds against tuberculosis. Our approach is in contrast to the classical target-based experiments, widely used in drug discovery, that suffer from very high attrition rates in anti-infective molecules [8]. This study should also serve the wider anti-tuberculosis research community by providing a list of genes and pathways that are more likely to be validated as TB targets for drug discovery and development.
We applied computational approaches using three domains of knowledge, namely the “assay space”, “chemogenomics space” and “structural space”, to identify new targets that are likely to interact with the active compounds from the GSK collection. We characterized the structural and chemical spaces of the recently released set of 776 compounds active against tuberculosis [7] and grouped the compounds into a total of 551 structural families. Subsequently, we predicted their likely targets using three orthogonal and complementary computational approaches. Jointly, we identified several amino-acid biosynthesis proteins as possible targets of several compounds in the dataset. A total of 207 unique pairs of compounds and potential MTB targets have been predicted. These compounds constitute a basis for further hypothesis-led exploration of their mode of action. We briefly outline the possible impact and contribution of our findings to Open Drug Discovery Initiatives [9], [10], [11], in particular against tuberculosis.
GSK recently released the data from a phenotypic screen against tuberculosis (available at ChEMBL http://dx.doi.org/10.6019/CHEMBL2095176) [7]. This open access dataset contains a total of 776 compounds active against M. bovis BCG, a non-virulent Mycobacterium species widely used in experimental studies as a vaccine component, and a subset of 177 confirmed compounds active against MTB strain H37Rv. The compound collection had been pre-filtered to remove known anti-bacterial compounds to maximize the discovery of novel compounds with anti-Mycobacterium activities. About 90% of the compounds have a quantitative estimate of drug-likeness (QED) value above 0.35 [12], herein called optimal drug-like compounds (Figure 1). The remaining 10% of compounds, which are highlighted by red bars in Figure 1, have higher molecular weights (>400 KD) and slightly higher hydrophobicity, expressed as the calculated logarithm of the 1-octanol/water partition coefficient (ALogP) [13]. For the subset of 177 compounds active against H37Rv, the average molecular weight is statistically smaller than for the entire dataset (Figure 1), consistent with known trends of lipophilicity and cytotoxicity/polypharmacology. The molecular PSA (polar surface area), ALogP (octanol–water partition coefficient) and wQED (weighted QED) scores result in statistically indistinguishable average values and distributions for both datasets. To assess the diversity of the dataset, we applied our Random Forest Score (RFS) to identify pairs of similar compounds (Methods). An all-against-all comparison was performed by nAnnolyze [14] and any pair of compounds with an RFS higher than 0.9 were considered similar. The resulting network of compound similarities was layered using Cytoscape [15] (Figure 1E). The entire dataset of 776 compounds was clustered into a total of 551 compound families, primarily composed of two large compound families and 481 singleton families. The two large families of compounds (GSKFAM_1 and GSKFAM_2) included 38 compounds each connected by 156 and 80 links, respectively (Figure 1F). In summary, the active compound set released by GSK is composed of drug-like molecules with non-redundant and diverse scaffolds.
The 776 compounds released by GSK were used as input to our integrative computational analysis approach that combines the results from a chemogenomics space search (CHEM), a structural space search (STR) and a historical assay space search (HIST). First, the exploration of the chemical space allowed us to identify likely targets for the input compounds based on their structural similarity to compounds with experimentally validated targets deposited in the ChEMBL database [16]. The approach employed a multi-category Naïve Bayesian classifier, which has been successfully used in ligand-based target prediction efforts [17], [18], [19]. Second, the exploration of the structural space allowed for the identification of likely targets based on the structural similarity of compounds and protein targets with known three-dimensional structures. The method was based on an improved version of the AnnoLyze program [14]. Finally, the exploration of the historical data on screening assays resulted in testable hypotheses for the anti-Mycobacterium mode of action of the selected compounds, based on the historical data from internal GSK screening experiments. This integrative approach allowed us to predict targets for the set of released compounds in the absence of known structural data (CHEM and HIST) or the absence of knowledge of the binding site (STR). When the three-dimensional structure of the target and the localization of the binding site are known or predicted, it is often helpful to follow up with molecular docking (see [20] and examples below). However, such an approach would be prohibitive for large numbers of compounds against a large number of targets, because molecular docking results still need to be interpreted manually for best impact. The three methods used in our integrative approach are further detailed in the Methods section of this manuscript.
We applied a multi-category Naïve Bayesian classifier (MCNBC) that was built and trained using structural and bioactivity information from the ChEMBL database [16]. Given a new compound, the model calculates a likelihood score based on the molecule's individual sub-structural/fingerprint features and produces a ranked list of likely targets. In total, the 776 compounds in the M. bovis BCG dataset resulted in 2,179 statistically significant target associations (at a Z-score >2.0) to proteins in the ChEMBL database from 62 different organisms (63% of hits are to human proteins). A simple orthology search against the MTB proteins from this set resulted in 1,401 compound-target relationships for 84 MTB proteins, with detectable orthology to 34 organisms. The specific predictions from the chemical space search are available at http://www.tropicaldisease.org/TCAMSTB (CHEM type).
We applied a Random Forest Score that identified structural similarities between any compound in the dataset and ligands from the Protein Data Bank (PDB) [14]. Each compound in the M. bovis BCG dataset is compared to ∼2,500 ligands for which there are known complex structures in the PDB, identifying structural similarities to be included in a pre-built network of structural relationships between ligands and targets. In total, the 776 compounds resulted in 207 significant target associations (RFS score >0.4) to proteins in a set of modeled three-dimensional structures from the MTB proteome. The specific predictions from the structural space search are available at http://www.tropicaldisease.org/TCAMSTB (STR type).
We used the historical GSK bioassay data to develop hypotheses for the anti-Mycobacterium mode of action for the active compounds. Using conservative activity thresholds, we found among the compounds active against MTB H37Rv unambiguous annotations for 49 compounds and their previously measured activity in 120 biochemical assays against 63 human targets (i.e., sub-micromolar IC50 or EC50). Overall, the M. bovis BCG screens resulted in a considerably larger number of active compounds and thus have a correspondingly greater amount of historical assay information. A total of 240 compounds were found to have activity recorded in 642 assays involving 209 human targets, with the largest human target classes being GPCRs and protein kinases, as expected. We then searched for orthologous sequences of the human assayed proteins in the MTB H37Rv and M. bovis BCG genomes using conservative criteria for assigning human-Mycobacterium homology (BLAST E-value ≤1.0e−10). Although there are significant evolutionary differences between bacterial and mammalian genomes, we still found 19 M. bovis BCG homologous genes (Table S1) in different target classes (Figure S1), including kinases (8 genes), cytochrome P450s (2 genes) and nine other enzymes such as a putative D-amino acid oxidase, an amidase, a putative flavin-containing monoamine oxidase, a NAD-dependent deacetylase, a putative catechol-O-methyltransferase, a protease, a putative epoxide hydrolase, a 3-ketoacyl-(acyl-carrier-protein) reductase, and a dihydroorotate dehydrogenase 2. While these M. bovis BCG genes had orthologous sequences in MTB H37Rv, fewer compounds were associated with putative targets in the latter species. For example, two Mycobacterium kinases and five enzymes were exclusively associated with M. bovis BCG positive compounds. Two kinases (pknA and pknB) and one enzyme (fabG) were experimentally characterized as essential for the survival of MTB [21], [22]. A total of 20 and 94 compounds were indirectly mapped by human protein target homology to 12 MTB H37Rv and 19 M. bovis BCG genes, respectively. The specific predictions from the historical assay space search are detailed in Supporting Information and are available at http://www.tropicaldisease.org/TCAMSTB (HIST type).
Of the 776 compounds in the GSK dataset, only one compound (GSK445886A) was predicted to hit diverse targets from different pathways by the three independent methods (Figure 2A). A total of 25 and 9 compounds were jointly predicted to hit a target by CHEM/STR and CHEM/HIST searches, respectively. The majority of predictions were obtained by the CHEM approach (404 compounds with predicted targets), followed by the STR approach (38 compounds with a predicted target) and the HIST approach (20 compounds with predicted targets). Such results were expected because the available information on biological activity shrinks as we move from the general “chemical” to the more specific “structural” and “historical” spaces. Interestingly, as an indication of the orthogonality of the three approaches, most of the redundancy of compounds with a predicted target was specific to each approach. In other words, each of the three approaches covered different parts of the space of compound-target predictions. For example, the CHEM approach predicted a target for 300 compound families (compared to a total of 404 unique compounds), of which it still shared 34 with either the STR or the HIST approaches (Figure 1B). A similar trend was observed for the other two approaches, indicating that the common compounds mostly occurred in small compound families or even singletons. Indeed, the GSK445886A compound, which was predicted to have a target by all three approaches, corresponded to a singleton compound family (GSKFAM_293).
To identify whether the three different approaches predicted targets for specific families in the dataset, we calculated the log odds probability (LogOdd) of a given compound family to appear in the list of selected compounds, given their different distributions in the original dataset (Figure 2C). This analysis aimed at identifying possible biases or artifacts specific to each of the three independent methods used in our integrative approach. Eleven compound families were under-represented in the selected dataset and 18 families were over-represented (with LogOdd values smaller than −0.5 and greater than 0.5, respectively). Interestingly, GSKFAM_551, which is a singleton with the SKF-67461 compound, was over-represented in the subset of selected compounds. Such predictions were based mostly on the STR and CHEM searches and may correspond to the chemical properties of the compound, resulting in a high false-positive rate for those two approaches. Conversely, the GSKFAM_4, which contains 15 compounds, is under-represented in the final subset of selected compounds, with only 1 hit identified by the CHEM approach.
There are a total of 1,044 unique MTB targets associated with a total of 112 pathways annotated in the KEGG database [23] (the mtu identifiers below refer to the relevant KEGG pathway id). Of those, the three orthogonal approaches identified targets for the selected set of compounds in a total of 84 pathways (Figure 3A). The STR search resulted in hits to 71 unique pathways, while the CHEM and the HIST searches resulted in hits to 35 and 16 pathways, respectively. These results were expected, because the target information is reduced from the STR space to the HIST space. A total of 11 unique pathways were predicted by the three approaches (Figure 3A and Table 1); these include many pathways associated with amino acid and nucleotide metabolism, such as arginine and proline metabolism (mtu00330), tryptophan metabolism (mtu00380), phenylalanine metabolism (mtu00360), tyrosine metabolism (mtu00350), histidine metabolism (mtu00340), glycine/serine/threonine metabolism (mtu00260) and pyrimidine metabolism (mtu00240). The results indicate that the GSK compounds potentially target proteins associated with primary metabolism. Interestingly, another seven pathways, not identified by the HIST approach, were found over-represented in the final set of predicted targets (Figure 3B). Those include some further primary and secondary metabolism systems, including streptomycin biosynthesis (mtu00521), folate biosynthesis (mtu00790), nitrogen metabolism (mtu00910), aminoacyl-tRNA biosynthesis (mtu00970), purine metabolism (mtu00230), penicillin and cephalosporin biosynthesis (mtu00311), D-arginine and D-ornithine metabolism (mtu00472), and one carbon pool by folate (mtu00670).
To assess the significance of our predictions using the three different approaches, we calculated a t-statistics p-value of any compound family - KEGG pathway pair (Methods). The search identified 8 different compound families with significant links (p-value <1×10−5) to 14 different KEGG pathways (Table 2). The GSK compound family 1, through its compounds GSK975784A, GSK975810A, GSK975839A, GSK975840A and GSK975842A, was predicted to target the glycerolipid (mtu00561) and glycerophospholipid metabolisms (mtu00564), with significant over-representation through 6 different targets including Rv2182c and Rv2483c, both acyltransferases essential for the survival of the bacteria [21]. The GSK compound family 3 was predicted to target the ABC transporters (mtu02010) through its compounds GSK547481A, GSK547490A, GSK547491A, GSK547499A, GSK547500A, GSK547511A, GSK547512A, GSK547527A, GSK547528A and GSK547543A. Similarly, it was also predicted to target the aminoacyl-tRNA biosynthesis (mtu00970) pathways, through 3 different targets including Rv1640c, a lysyl-tRNA synthetase essential for the survival of the bacteria [21]. The GSK compound family 7, was predicted to target several pathways through 2 different targets Rv0053 (30S ribosomal protein S6) and Rv0650 (a glucokinase), none considered essential for the survival of the bacteria [21]. The GSK compound family 9 through its compounds GSK1188379A and GSK1188380A, was predicted to target the ABC transporters (mtu02010) pathway through the Rv0194 target (ATP-binding cassette, subfamily C) considered non-essential for the survival of the bacteria [21]. Identical results were obtained with the GSK compound family 16 through its compounds GSK1825940A and GSK1825944A. The GSK compound family 35 through its compounds BRL-10143SA and BRL-51093AA was predicted to target the one carbon pool by folate (mtu00670) pathway through the Rv2763c and Rv2764c targets (a dihydrofolate reductase and a thymidylate synthase, respectively) considered non-essential for the survival of the bacteria [21]. The GSK compound family 173 through its compound GSK14022909A was predicted to target the aminoacyl-tRNA biosynthesis (mtu00970) pathway through three essential targets [21], Rv1640c, Rv3598c and Rv3834c (a lysyl-tRNA synthetase, a lysyl-tRNA ligase, and a seryl-tRNA ligase, respectively), which are essential for the survival of the organism [21]. Interestingly, this family is also predicted to target Rv3105c and Rv3135 genes (a peptide chain release factor 2 and a PPE family protein), which are also essential for the survival of the organism [21]. Finally, the GSK compound family 334 through compound GSK270671A was predicted to target the nitrogen metabolism (mtu00910) pathway through the Rv1284 and Rv3588 targets (carbonic anhydrases) considered essential for the survival of the bacteria [21].
Even though target Rv0014c, a serine/threonine-protein kinase, was not identified as belonging to an enriched pathway (it is not annotated in the KEGG database), it was predicted by the HIST approach to be a target for the GSK1365028A, GSK1598164A, GSK275628A and GW664700A (all singleton families in our compound clustering). Kinases are the most prominent human target class having identifiable orthologs in both M. tuberculosis H37Rv and M. bovis BCG genomes (Figure 4A). The human genome encodes over 450 kinases, while Mycobacterium contains between 4 and 24 serine/threonine kinases, depending on the exact species (M. tuberculosis and M. bovis have 11 conserved kinases each). At least two of these kinases, pknA and pknB, have been determined to be essential for in vitro viability of M. tuberculosis [21]. To further evaluate potential MoA of kinase inhibitors, we computationally docked several compounds into the adenine-binding portion of the ATP binding pockets of the two available experimental structures for the essential kinase pknB. The criteria for choosing the compounds were whole cell screening activity of MIC90 less than 10 µM and IC50 less than 8 µM. Two structures (PDB IDs: 2PZI and 3F69) were selected because both were co-crystallized with an inhibitor, clearly detailing their ATP binding pockets.
An empirical docking score threshold of −8.5 kJ/mol was chosen to identify putative positive bindings of the active compounds across the two pknB PDB models (Table S2). GSK1598164A, an inhibitor of several human serine/threonine protein kinases, was positive in both H37RV and BCG whole cell screens, based on favorable docking scores (−9.19 and −8.96 kJ/mol against 2PZI and 3F69, respectively). Both GSK1598164A and the enzymatic product ADP in the crystal structure were found to interact with the Glu93 of pknB, where the nitrogen atoms on the ‘head’ unit form the hydrogen bond with Glu93 (Figure 4B). Glu93 is conserved across both human and TB kinases (Figure 4A). Several residues in the putative hydrophobic binding pocket (Leu17, Gly18, Phe19, Val25, Ala38, Val72, Met92, Glu93 and Val95) were also found to be within 4 Å of both GSK1598164A and ADP. In conclusion, our analysis suggests that several bactericidal compounds in the published phenotypic screen act by inhibiting essential M. tuberculosis kinases.
The CHEM and STR methods identified Rv3598c (lysS1 lysine-tRNA ligase 1) and Rv3834c (serS serine-tRNA ligase) as possible targets for the GSK1402290A compound, respectively. Both enzymes are part of the aminoacyl-tRNA biosynthesis pathway (mtu00970) and are essential in in vitro experiments [21]. Moreover, the mtu00970 pathway was selected in our analysis as being significantly associated with GSKFAM_173 (GSK1402290A compound).
The CHEM approach predicted that the human lysyl-tRNA synthetase (UniProt ID Q15046) was a likely target of GSK1402290A, with a likelihood score of 11.3 and a Z-score of 2.4. Furthermore, the model indicated that the individual fragments contributing to this prediction were derived by its fused triazole ring (e.g., pyrazole and imidazole features), as well as by its aniline group. In fact, the model for this target was trained using 47 active compounds from ChEMBL and almost all of them contained the aforementioned fragments (Figure 5A). Moreover, the predicted human target shared in OrthoMCL [24] the ortholog group (OG5_126972) with MTB's lysine-tRNA ligase 1 (UniProt ID P67607).
The STR method predicted a link between the compound and the target through a 3D model of the Rv3834c protein built based on the known structure of a seryl-tRNA synthetase from Aquifex aeolicus. The Rv3834c target and the seryl-tRNA synthetase template aligned with 43% sequence identity and resulted in good quality models (MPQS>1.5) [25]. To further evaluate potential MoA of the GSK1402290A compound, we computationally docked it into the nAnnoLyze predicted binding site for Rv3834c (Figure 5). The AutoDock run resulted in a best pose with −8,4 kJ/mol, indicating interactions between the GSK1402290A compound and the Rv3834c target (Figure 5B). In support of this model, the interactions occur with conserved protein residues, given the curated multiple sequence alignment for PFAM family PF00587 (tRNA synthetase class II core domain).
In summary, our CHEM and STR predictions suggest that GSK1402290A could act as an inhibitor of the aminoacyl-tRNA biosynthesis pathway and provide the basis for further chemical optimization of this compound.
The recent publication of a large-scale screening effort for identifying drug-like small molecule compounds active against tuberculosis has been used as starting point for our research. Here, we predicted the likely mode of action of a selected set of compounds active against tuberculosis, based on a computational approach that integrates data from historical assay results, chemical features and their relationship to activity, and structural comparisons. Our integrated approach resulted in prediction of several compound-target pairs, which can be further tested using genomics, genetics and biochemical assays. More broadly, our approach can be applied to whole cell screens for any pathogen, provided sufficient datasets are available.
We have predicted a wide range of MTB specific as well as more evolutionary conserved targets. While compounds with known activity against a human protein could be compromised by toxicity, and therefore should be eliminated from further study, empirical evidence suggests that existence of a human orthologous sequences is not a strong filter for selecting pathogen targets. Many clinically used antibiotics have targets with human orthologs, such quinolones (DNA gyrase and topoisomerases), rifampicin (RNA polymerase), mupirocin (isoleucyl-tRNA synthetase) and the latest anti-TB drug now in Phase II testing, bedaquiline (F1F0 ATPase) [4], [6]. The associated side effects of antibiotics are mostly due to high doses treatments affecting off-target proteins (including human ortologs) and not specifically to on-target effects. The billion plus years of evolutionary distance between prokaryotes and mammals has lead to significant divergence between orthologous proteins such that there is sufficient structure activity relationship or SAR bandwidth to develop specific inhibitors of the pathogen target, in our case MTB.
It is important to note that we also had a subset of compounds with historical data indicating activity against human protein targets with no known homologs in MTB, such as the GPCRs. Thus, their mechanism of action against MTB must be due to non-human target related interactions. These compounds must be pursued with caution as drug candidates given their known in vitro interaction with a human protein. Nevertheless, such compounds could be valuable tools for understanding MTB viability. In general, knowledge of potential human protein interactions adds to the design of effective counter-screens to drive compound SAR specificity and potency towards the pathogen.
The public availability of the data and compounds [7] as well as our predictions (http://www.tropicaldisaes.org/TCAMSTB/ or ftp://ftp.ebi.ac.uk/pub/databases/chembl/tb) will facilitate further research on drug discovery against tuberculosis. A major goal of our work is to encourage other researchers to experimentally validate the described targets and make their findings publicly available as soon as possible, thus optimizing the process of developing a safe and well tolerated novel therapy for tuberculosis.
All compound datasets used in this study (that is, BCG dataset of 776 GSK compounds including the H37Rv sub-dataset of 177 compounds) were obtained directly from the ChEMBL database (as deposition set http://dx.doi.org/10.6019/CHEMBL2095176). Chemical properties of the compounds (Figure 1) were calculated as previously described [12].
A multi-category Naïve Bayesian classifier (MCNBC) was built using structural and bioactivity information from the ChEMBL database (version 14) [16]. In brief, the classifier learns the various classes (in this case protein targets) by considering the frequency of occurrence of certain sub-structural features for the different chemical compounds. Given a new, unseen compound, the model calculates a Bayesian probability score based on the molecule's individual features and produces a ranked list of likely targets. The model was built in Accelrys Pipeline Pilot (version 8.5). The structure and bioactivity data were extracted from the ChEMBL database and conformed the following filters: (i) the activity value was better than 10 uM (pIC50>5), (ii) the target type was a protein, (iii) the activity type was IC50, Ki or EC50, and (iv) the target confidence score was above 7.0. The last filter ensured that there was a reported direct interaction between the ligand and the protein target. The script resulted in 489,056 distinct compound-target pairs. To increase the robustness of the model, only targets with 40 or more active compounds were considered further, thus reducing the number of unique compound-target pairs to 466,686, spanning 1,258 distinct targets and 271,918 distinct compounds.
Two multiple-category models were subsequently built. Firstly, a model was created by choosing at random 85% of the compound records as the training set, so that the remaining 15% could be used as a test set for model validation, ensuring no overlapping structures in the 85-15 partition [17]. The MCNBC trained on 85% of the 271,918 ChEMBL compounds and associated targets was then used to predict the targets for the remaining 15% of the ChEMBL subset, containing 40,788 distinct compounds, unseen by the model. Standard ECFP_6 fingerprints were employed as molecular descriptors for the classifier [26]. These fingerprints encode a molecular structure as a series of overlapping features/fragments of a diameter of up to three bond lengths.
For each compound in the test set, the Pipeline Pilot model generated a likelihood score Ptotal for all possible targets. This is derived by the Laplacian-corrected Bayes rule of conditional probability P(A|Fi) for each fingerprint feature i of the compound.where Fi is the ith fingerprint feature; A is the number of active molecules for a target; T is the total number of molecules; AFi is the number of active molecules containing feature i; and TFi is the number of all molecules containing feature i.
For the purposes of this validation, only the top five target predictions were considered (i.e., the ones with the highest positive likelihood score). This reflects a real-life situation where only a small number of target predictions can be practically and economically tested experimentally. To test the accuracy of the method, the five target predictions were then compared to the actual target reported for that particular compound.
The model derived by the training set ranked the correct target highest among all 1,258 possible targets for 82% of the compounds in the test set (Figure 6A). The target is correctly predicted on the second guess for 6% of the compounds and correctly predicted on the third guess for 2% of the compounds. In total, 92% of the compounds in the test set are correctly assigned to their known targets within the top five predicted targets. The ChEMBL database groups most of the individual protein targets into a hierarchical classification of target family names. Given this information, further analysis was done to examine the accuracy of the target classification predictions. Individual targets were replaced by their respective protein classification annotation using a lookup dictionary. In total, 568 unique protein classification labels were considered. The model's predictive power improves, returning the correct protein family as the top ranked prediction in 88% of the compounds and within the top five predictions in 94% of the compounds (Figure 6A). After the successful validation of the method, a second model was created utilizing 100% of the data and keeping the rest of the parameters intact. The derived model was then used for predicting the targets of all GSK compounds.
A network of structural similarities between compounds and targets was built to identify the most likely target of a given compounds in our GSK dataset. To explore the structural space we used an improved version of our previously published AnnoLyze algorithm [14], which was based on homology detection through structural superimposition of targets and their interaction networks to small compounds similarly to previously published approaches [27], [28]. Briefly, the new nAnnoLyze algorithm relies in four pre-built layers of interconnected networks, First, the “GSK Ligand” network where nodes are GSK compounds and edges correspond to their similarity as measured by a Random Forest classifier score (RFS) (see below). Second, the “PDB Ligand” network where nodes are ligands in the Protein Data Bank (PDB) [29] and edges correspond to their similarity also measured by the RFS. The “GSK Ligand” network is linked to the “PDB ligand” network by edges corresponding to the compound similarity measure by the RFS. Third, the “PDB Protein” network where nodes are proteins in PDB and edges corresponds to their structural similarity as measured with the MAMMOTH structural superimposition [30]. Fourth, the “MTB Models” network where nodes are structure models of MTB targets and edges corresponds to their structural similarity after superimposition by the MAMMOTH program. The two central networks (that is, “PDB Ligand” and “PDB Protein” networks) are connected by co-appearance in any solved structure in the PDB and the “PDB Protein” and the “MTB Models” networks are also linked by the structural comparison between any protein in the PDB and all models from MTB. Finally, once all the networks are constructed, we identified the closest path between any GSK compounds and a MTB target and scored their relationship as the sum of all similarities scores in the network. Such score was then normalized between 0 (non-similar) and 1 (similar) and only pairs of GSK compounds and their MTB targets with scores higher than 0.4 were kept.
To identify whether two compounds could be considered similar, we developed a new Random Forest classifier (RFS), which was trained with a dataset of “similar” and “non-similar” ligands. Two ligands were similar if they bind the same binding site as defined by the LigASite database, a gold-standard dataset of biologically relevant binding sites in protein structures [31]. To avoid overestimation in the validation of our approach, all ligands in the database that were included in a testing set of 2,380 ligands from the PDB were removed. Our training set of similar ligands included 197 pairwise comparisons considered as “true similar” and a set of randomly paired ligands as “true non-similar” comparisons. The SMSD program [32] was then used to compare all pair of selected ligands to obtain their Tanimoto score, bond breaking energy, Euclidian distance for equivalent atoms, stereochemical match, substructure fragment size, and finally the molecular weight difference. Such scores were then normalized and constituted a vector defining the similarity between any two compared ligands, which was then used as input for the Random Forest classifier. The aim of the classifier was thus to identify hidden relationships between the six scores to maximize its capacity to identify true pairs of similar ligands and discern them from non-similar ligands. The classifier was then tested with a 10-fold cross validation procedure and resulted in an area under the ROC curve of 0.97 and a very small false positive rate of 1.6% (Figure 6B).
To populate the “MTB Model” network with structures of MTB targets, we built all possible comparative structure models for any protein in the M. tuberculosis H37Rv, M. bovis BCG, and M. smegmatis genomes using the ModPipe program [25]. All sequences were obtained from the Genomes Web site of the NCBI database. Such modeling resulted in a total of 34,894 comparative models for which 5,008 were predicted to be reliable models (that is, 1.1 or higher ModPipe quality score and ga341 higher than 0.7). Next, we structurally compared this set of selected models to any non-redundant (90% sequence identity) structure in the PDB that contained at least one known ligand. Structural comparisons between two proteins were performed using the MAMMOTH algorithm [30]. Four different scores were stored for each structural superimposition: percentage of sequence and structure identity for the entire protein and percentage of sequence and structure identity for the residues involved in the binding site defined as any residue in the PDB template structure within 6 Ångstroms of any atom in the ligand. A binding site in a model was considered then similar to a binding site in a known PDB structure if at least the binding site sequence and structure similarity were higher than 40%. This similarity cut-off was previously validated in a large-scale comparison of known ligand-protein pairs [14].
The final entire network of comparisons included the 776 compounds from the GSK dataset, ∼2.500 unique ligands from the PDB, ∼16,000 unique protein structures from the PDB and a total of ∼5,000 structure models from MTB. Such network resulted in 207 pairs of GSK compound to MTB target short paths (i.e., score >0.4).
GSK proprietary compound screening databases were queried for any historical assay data associated with both Mycobacterium species active compounds. The majority of these screens were against human protein targets. The threshold above which compound efficacy against specific human targets was considered significant was defined as pIC50≥5.0 for inhibition or antagonist assays, and pEC50≥5.0 for agonist, activation or modulator assays. Activities at more than 600 target-result type combinations (some targets are assayed in both an antagonist and agonist mode) were analyzed amongst the BCG and H37Rv active compounds, representing potential modes of action. The target activities for the screened compounds were analyzed to identify targets over-represented amongst the anti-malarial actives vs. inactives.
Using BLASTP [33] we queried the protein complement of published MTB H37Rv and M. bovis BCG genomes with RefSeq proteins [34] for all human targets accepting a homology cut-off of an E-value ≤1.0e-10 and visual inspection of the alignments. Putative homologous relationships were confirmed by reciprocal BLASTP searches of identified Mycobacterium homologues against the human RefSeq protein databases. Initial multiple sequence alignments were performed using the program CLUSTALW v1.8 [35] with default settings and subsequently refined manually using the program SEQLAB of the GCG Wisconsin Package v11.0 software package (Accelrys, San Diego, CA, USA).
We measured two different statistics to assess the significance of a particular link between a chemical compound and a target pathway. Firstly, we calculated the LogOdds (that is, the odds of an observation given its probability). A feature i (in our case, a compound in Figure 2C or a pathway in Figure 3B) has a probability (pi,c) in the entire dataset and a probability (pi,r) of being at the subset of selected compounds/pathways. Their LogOdds are defined as the logarithm of its Odds (Oi):Therefore, Odds higher than 1 (or positive LogOdds) indicate over-occurrence of the compound/pathway in the selected subset. Odds smaller than 1 (or negative LogOdds) indicate under-representation of the compound/pathway in the selected subset. Secondly, a p-value score was calculated for each predicted link between a compound and a target pathway using a Fisher's exact test for 2×2 contingency tables comparing two groups of annotations (i.e., the group of compounds in a given pathway and the group of compounds in the entire dataset) [36].
Autodock 4.2 was used for docking studies [37]. The ga_num_evals were set at 250,000 to balance docking performance and CPU consumption. Thirty replicates were run for each chemical-protein pair and the binding conformation with the lowest docking score was chosen for visualization using PyMOL.
|
10.1371/journal.pgen.1006805 | Overexpression of the essential Sis1 chaperone reduces TDP-43 effects on toxicity and proteolysis | Amyotrophic lateral sclerosis (ALS) is a devastating neurodegenerative disease characterized by selective loss of motor neurons with inclusions frequently containing the RNA/DNA binding protein TDP-43. Using a yeast model of ALS exhibiting TDP-43 dependent toxicity, we now show that TDP-43 overexpression dramatically alters cell shape and reduces ubiquitin dependent proteolysis of a reporter construct. Furthermore, we show that an excess of the Hsp40 chaperone, Sis1, reduced TDP-43’s effect on toxicity, cell shape and proteolysis. The strength of these effects was influenced by the presence of the endogenous yeast prion, [PIN+]. Although overexpression of Sis1 altered the TDP-43 aggregation pattern, we did not detect physical association of Sis1 with TDP-43, suggesting the possibility of indirect effects on TDP-43 aggregation. Furthermore, overexpression of the mammalian Sis1 homologue, DNAJB1, relieves TDP-43 mediated toxicity in primary rodent cortical neurons, suggesting that Sis1 and its homologues may have neuroprotective effects in ALS.
| Many neurodegenerative diseases are associated with aggregation of specific proteins. Thus we are interested in factors that influence the aggregation and how the aggregated proteins are associated with pathology. Here, we study a protein called TDP-43 that is frequently aggregated in the neurons of patients with amyotrophic lateral sclerosis (ALS). TDP-43 aggregates and is toxic when expressed in yeast, providing a useful model for ALS. Remarkably, a protein that modified TDP-43 toxicity in yeast successfully predicted a new ALS susceptibility gene in humans. We now report a new modifier of TDP-43 toxicity, Sis1. We show that expression of TDP-43 in yeast inhibits degradation of damaged protein, while overexpression of Sis1 restores degradation. Thus suggests a link between protein degradation and TDP-43 toxicity. Furthermore we show that a mammalian protein similar to Sis1 reduces TDP-43 toxicity in primary rodent neurons. This identifies the mammalian Sis1-like gene as a new ALS therapeutic target and possible susceptibility gene.
| Amyotrophic lateral sclerosis (ALS), often referred to as "Lou Gehrig's Disease", is a progressive neurodegenerative disorder characterized by degeneration and ultimately death of motor neurons in the brain and the spinal cord [1,2]. The most common pathologic characteristic of ALS is the formation of cytoplasmic inclusions rich in the transactive response element DNA/RNA binding protein of 43 kDa (TDP-43) [3–7]. Furthermore, mutations in the gene encoding TDP-43 (TARDBP) cause familial ALS [8–11]. Cytoplasmic TDP-43 inclusions are also common in the majority of individuals with frontotemporal dementia (FTD) [3,4], suggesting that ALS and FTD are linked by a common disease mechanism involving TDP-43 mislocalization and aggregation.
TDP-43 is an RNA/DNA-binding protein [12,13] that is found in the nucleus under normal conditions [3,4], and is associated with cytoplasmic stress granules during stress [14]. TDP-43 harbors two RNA Recognition Motif (RRM) domains and a C-terminal glycine-rich region, where most of the ALS-linked mutations are found [12,13,15]. An algorithm to identify human proteins with aggregation prone, prion-like domains ranked TDP-43 69th among the entire human proteome, because of its C-terminus (residues 277–414) ([16] also see [17–19]). The abnormal localization of wild-type or mutant versions of this protein to cytoplasmic foci in the absence of stress is associated with ALS [3,5,6,20].
Prions are self-seeding conformational variants of particular proteins [21]. The conversion of largely α-helical cellular prion protein PrPC into self-seeding fibrous β-sheet-rich ordered aggregates (amyloids) called PrPSc (associated with scrapie) is the causative agent of prion diseases in mammals [22]. A conformational change of a disease-specific soluble cellular protein to an aggregated form that seeds further aggregation was also shown for Alzheimer’s disease, Parkinson’s disease and ALS [17,18,23–25].
In addition, several yeast proteins have been shown to convert from soluble to self-seeding amyloid conformations called prions, and are associated with transmissible phenotypes in yeast [21,26–33]. Interestingly, yeast prions can enhance the rate of de novo aggregation of heterologous prion proteins, presumably by a cross-seeding mechanism. Indeed, the endogenous yeast prion, [PIN+] [29,34], is required for the aggregation and toxicity of human Huntington's disease Htt protein with an expanded polyglutamine (polyQ) tract when expressed in yeast [35,36].
Chaperones, including Sis1, an Hsp40 chaperone, are required for the propagation of yeast prions [37–42]. In addition to its high nuclear concentration, Sis1 is also diffusely localized in the cytoplasm in exponentially growing cells [43]. However, when prion or other amyloid cytoplasmic aggregates are present, Sis1 also co-localizes with them creating cytoplasmic Sis1 foci [44,45]. Sis1 is required for the fragmentation of prion oligomers. Indeed, reduced or overexpressed levels of Sis1 lead to larger or smaller prion oligomers [37–41]. The new aggregate ends created by this fragmentation are required for prion propagation. Furthermore, the smaller aggregates created by fragmentation are more easily transmitted to daughter cells, also promoting prion propagation [45–47].
Overexpression of Sis1 rescues yeast from toxicity associated with overexpression of several prion or prion-like proteins. For example, overexpression of Sis1 reduces [PIN+] dependent polyQ aggregation and toxicity [48]. The mechanism of rescue is in part because the polyQ aggregates sequester Sis1, inhibiting it from its normal role of transporting ubiquitinated misfolded proteins in the cytoplasm to the nucleus for proteasomal degradation [44].
Similarly, overexpression of Sis1 rescues cells from [PIN+] dependent toxicity of an overexpressed prion-like protein Pin4C. When overexpressed in the presence of [PIN+], Pin4C formed large hyperphosphorylated aggregates that co-localized with Sis1 and reduced degradation of a ubiquitin proteasome system (UPS) reporter protein. Aggregation, hyperphosphorylation and toxicity were suppressed by overexpression of Sis1 [45,49].
Overexpression of Sis1 also rescues [PIN+] dependent toxicity caused by overexpression of the [PIN+] prion protein, Rnq1. Here, overexpressed Sis1 appears to enhance aggregation of toxic detergent soluble Rnq1 oligomers into apparently non-toxic larger detergent resistant aggregates in the nucleus [50,51].
Yeast is a useful organism to study the conversion of soluble disease-specific protein into amyloid or prion-like aggregates [35,48,52–56]. Overexpressed human TDP-43 forms cytoplasmic aggregates in yeast associated with toxicity [53]. Both wild-type TDP-43, which accumulates in the majority of sporadic ALS, and mutant TDP-43, associated with familial ALS, elicit cell death when overexpressed. However, at near-endogenous expression levels mutant TDP-43 is significantly more toxic than wild-type TDP-43 [57] consistent with the near complete penetrance of TARDBP mutations in familial ALS [8].
Unlike most other prion-like aggregating proteins, TDP-43 aggregates do not appear to be typical amyloids [58]. Unbiased screens for overexpression or deletion modifiers of TDP-43 toxicity identified numerous yeast proteins as candidates for involvement in the TDP-43 toxicity cascade. The identification of one such modifier, Pbp1, with a human homologue ATXN2, led to the discovery that the length of polyglutamine expansions in ATXN2 is associated with increased risk for ALS [59]. This clearly established the power of the yeast model in understanding human disease.
Here, we identify a new modifier by showing that excess Sis1 reduces the toxicity of overexpressed TDP-43. Likewise, overexpression of the mammalian Sis1 homologue, DNAJB1, reduces TDP-43-mediated toxicity in primary rodent cortical neurons, suggesting that Sis1 and its homologues may have neuroprotective effects in ALS. Finally, we provide evidence that TDP-43 impedes the UPS-mediated degradation of cytosolic misfolded proteins and that overexpression of Sis1 restores degradation even in the presence of excess TDP-43.
Although overexpressed polyQ or Pin4C only form large aggregates and causes toxicity in [PIN+] cells [35,49], overexpressed TDP-43 does so in both [PIN+] and [pin-] yeast [53,60]. Likewise, we saw that overexpressed TDP-43 inhibited growth in both [PIN+] and [pin-] cells (Fig 1A and S1 Fig). Although we noted some variability between transformants and strain backgrounds, there was often a slightly stronger inhibition in [PIN+] vs. isogenic [pin-] cells (Fig 1A).
In addition, we found that overexpressed TDP-43 altered cell shape, and again the effect was slightly but reproducibly stronger in [PIN+] cells. We examined this effect in three strain backgrounds for cells grown on 2% galactose plates for 4 days (S2 Fig). In strain 74D-694 [PIN+] nearly every cell showed altered morphology, with less extreme numbers and less elongation in [pin-] cells. Elongated cells were also easily seen in BY4741 and W303 strain backgrounds although especially in BY4741 many cells did not have an altered shape. Likewise, 74D-694 overexpressing TDP-43 for 24 h in liquid were elongated and bloated relative to control cells without TDP-43 overexpression (Fig 1C and 1D). Using cells with the endogenous nuclear Htb1 protein tagged with mCherry [61], we could see that the bloated/elongated cells each have only one non-fragmented nucleus and so did not result from a failure of cytokinesis (Fig 2). Finally, the cell morphological change was proportional to the TDP-43 expression level (S3 Fig).
Since overexpression of Sis1 suppressed toxicity of other overexpressed aggregating proteins (polyQ, Rnq1 and Pin4C) [44,49–51] we decided to look at the effect of Sis1 on TDP-43 toxicity even though previous screens for modifiers of TDP-43 and FUS toxicity in yeast did not uncover Sis1 [54,59,62].
We compared growth of cells overexpressing TDP-43-YFP with those co-overexpressing Sis1 and TDP-43-YFP. Excess Sis1 reduced growth inhibition due to TDP-43 in both [pin-] and [PIN+] cells in three strain backgrounds. The rescue appeared to be weaker in 74D-694 [PIN+] cells where TDP-43 is more toxic (Fig 1A). While overexpression of Sis1 did not completely suppress TDP-43 toxicity, it altered TDP-43 toxicity more efficiently than previously reported modifiers of TDP-43 toxicity: overexpression of Hsp104, Pbp1 or human Upf1 [59,63–65] (S1 Fig).
Overexpression of Sis1 also reduced cell death (Fig 1B) and the level of cell bloating/elongation associated with TDP-43 overexpression (Figs 1D and 2). Sis1 overexpression did not cause these effects by reducing the level of TDP-43 or by curing cells of [PIN+], because TDP-43 levels remained unchanged (Fig 3) and Sis1 overexpression does not cure cells of [PIN+] [66] (S4 Fig).
Previous results showed that overexpression of polyQ or Pin4C in [PIN+] cells reduced degradation of a UPS reporter, CG* [44,49], and that the overexpression of Sis1 reversed this polyQ effect [44,49]. Thus we asked if TDP-43 overexpression likewise affects clearance of CG*. CG* is the mutant version of the secretory protein carboxypeptidase Y lacking its signal sequence and tagged with GFP (ΔssCPY*). Normally, CG* is rapidly degraded via the UPS, dependent on Hsp40 and Hsp70 chaperones [44]. After inhibition of new protein synthesis with cycloheximide (CHX), CG* degradation was inhibited by TDP-43 overexpression (Fig 4). The effect was seen in both [pin-] and [PIN+] 74D-694 cells but was more dramatic in [PIN+] cells where TDP-43 is more toxic. Furthermore, simultaneous overexpression of Sis1 and TDP-43 resulted in rapid degradation of CG*. This is consistent with the hypothesis that TDP-43 toxicity in part results from interference with the UPS system.
Since deletion of HSP104 prevents propagation of several yeast prions [21] we looked for effects of hsp104Δ on the ability of Sis1 overexpression to reduce TDP-43 toxicity. No effects were detected when comparing cells with wild-type HSP104 vs. hsp104Δ (S5A Fig). Also, as Sis1 is an Ssa1 co-chaperone [67,68] we looked for effects of lowered Ssa1 activity on the ability of Sis1 overexpression to reduce TDP-43 toxicity. This experiment was complicated by the fact that in addition to Sis1 being a co-chaperone for Ssa1, it is presumed to be a co-chaperone of the other members of the Ssa family (Ssa2, Ssa3 and Ssa4). Furthermore, deletion of all members of this family is lethal and deletion of one member is compensated for by alteraed expression of other members. Thus, we used the dominant negative SSA1-21 mutant in an ssa2Δ background which has a dramatic effect on Ssa family activity causing loss of the [PSI+] prion [69]. We found that this alteration in Ssa activity did not detectably affect the ability of Sis1 overexpression to reduce TDP-43 toxicity (S5B Fig).
TDP-43 aggregates are soluble in 2% SDS at room temperature and to fail to stain with the amyloid dye thioflavin T [53]. We asked if the TDP-43 aggregates had the amyloid-like property of being insoluble in a different detergent, sarkosyl, because some pathological TDP-43 aggregates were resistant to sarkosyl [3,4] and because in vitro fibrillized TDP-43 remains insoluble in sarkosyl [70]. Indeed, when we treated lysates of cells harboring TDP-43 aggregates with sarkosyl at room temperature (Fig 3), TDP-43-YFP oligomers were visible in unboiled samples. The size and abundance of these oligomers was unchanged by Sis1 overexpression and Sis1 was not detected in these oligomers.
Not only does overexpression of Sis1 inhibit toxicity of polyQ and Pin4C but it also dramatically reduces aggregation of these proteins [44,49]. In contrast, TDP-43 formed large aggregates even in the presence of overexpressed Sis1. However, the frequency of small aggregates was influenced by the level of Sis1: when the level of Sis1 was reduced, TDP-43-YFP formed more smaller aggregates that appeared in addition to larger aggregates that were comparable in size and number to that seen with the higher level of Sis1 (Fig 5A). In this experiment, Sis1 expression was controlled via a TET promoter, whereby the level of Sis1 was above normal in the absence of doxycycline and below normal in the presence of doxycycline. In control cultures with endogenous Sis1 but no TET controlled Sis1 (GF820), aggregation patterns of overexpressed TDP-43-YFP were not affected by doxycycline.
Despite the above aggregation pattern differences, no reproducible differences were seen in the level of soluble vs. insoluble TDP-43 in the presence or absence of Sis1 overexpression. About 55–60% of TDP-43 was found in the pellet whether or not Sis1 was overexpressed (Fig 5B).
The finding that the essential chaperone, Sis1, is sequestered by PolyQ and Pin4C aggregates suggested that the lack of functional Sis1 could contribute to PolyQ and Pin4C-mediated toxicity [45,49]. We thus asked if the aggregates formed by overexpression of TDP-43-GFP co-localized with endogenous Sis1 tagged with mCherry (Sis1-mCh). Sis1-mCh is largely diffuse in the nucleus but also forms some cytoplasmic dots [71,72]. However, TDP-43-GFP and Sis1-mCh dots failed to show any co-localization (Fig 6A). We also investigated the interaction of Sis1 and TDP-43 with co-immunocapture. Sis1 failed to associate with TDP-43-GFP captured with TDP-43 antibody (Fig 6B).
There are many Hsp40 chaperones with distinct structures and functions in mammalian cells [73]. One of these proteins, DNAJB1, has the most similarity to the yeast Sis1. To determine if DNAJB1 extended neuronal survival in mammalian neurons, we overexpressed wild-type (WT) and mutant TDP-43 (A315T) linked with familial ALS in rodent primary cortical neurons [8,57]. This model recapitulates several features of human disease, including neurodegeneration, the formation of ubiquitin-positive neuronal aggregates, and cytoplasmic TDP-43 mislocalization in affected cells [3,57]. For these assays, individual neurons expressing WT or mutant TDP-43-EGFP or EGFP alone were monitored at regular intervals for 10 days by fluorescence microscopy, and the time of death for each cell used to create survival or cumulative hazard plots (Fig 7A and 7B). Both WT and mutant TDP-43-EGFP were highly toxic when overexpressed, effectively doubling the risk of death over that of control neurons transfected with EGFP (hazard ratio (HR) 2.02 and 2.06, respectively, p < 2x10-16, Cox hazards analysis). Although DNAJB1 expression caused mild toxicity in control neurons (HR 1.24, p = 0.0001), it also extended the survival of neurons transfected with TDP-43(A315T) by 15% (HR 0.85 compared to TDP-43(A315T)-EGFP alone, p = 0.001). In contrast, DNAJB1 co-expression had no significant effect upon TDP-43(WT)-EGFP expressing cells (HR 0.98 compared to TDP-43(WT)-EGFP alone, p = 0.7). Since TDP-43 levels are directly proportional to survival [74], we questioned whether DNAJB1 expression may affect neuronal survival by acting on TDP-43(A315T) levels. However, we detected no significant alterations in TDP-43(WT)-EGFP or TDP-43(A315T)-EGFP levels associated with DNAJB1 expression (Fig 7C). Thus, as in yeast, expression of DNAJB1 partially mitigates TDP-43-dependent toxicity.
To determine if deficiencies in DNAJB1 may potentiate neurodegeneration due to the accumulation of TDP-43, we used siRNA to knockdown DNAJB1 in rodent primary cortical neurons overexpressing TDP-43(WT)-mApple or mApple alone (S6 Fig). Similar to its effects when overexpressed, DNAJB1 knockdown increased the risk of death by 20% in control neurons expressing mApple alone (HR 1.2 compared to scrambled siRNA, p 0.004). Knockdown of DNAJB1 likewise enhanced the risk of death by 21% in neurons overexpressing TDP-43(WT)-mApple (HR 1.21 compared to scrambled siRNA, p = 2.3x10-5). These results show that DNAJB1 is essential for maintaining neuronal health, and despite the fact that DNAJB1 can help prevent TDP-43 mediated neurotoxicity, its deficiency does not exacerbate neurodegeneration due to TDP-43 overexpression.
Aggregation of TDP-43 is frequently associated with ALS and other neurological diseases [75–79] and overexpressed human TDP-43 is toxic and forms cytoplasmic aggregates in yeast and mammalian neurons [53,59,64,65]. Our finding in yeast that TDP-43 aggregates inhibit clearance of a misfolded reporter protein, CG* by the ubiquitin proteasome system (UPS) [80], is consistent with the hypothesis that some of the TDP-43 associated toxicity is caused by proteolytic dysfunction. Indeed, failure of proteolysis and the consequent accumulation of misfolded proteins is associated with a variety of diseases [23,24,44]. Furthermore, we found that the effect of TDP-43 on toxicity is correlated with its effect on UPS: TDP-43 is more toxic in [PIN+] than [pin-] 74D-694 cells and likewise has a greater effect on UPS in [PIN+] vs. [pin-] cells. By analogy, the presence vs. absence of heterologous aggregates in mammalian cells may also influence the proteostasis of disease-causing aggregates.
Why and how [PIN+] effects toxicity and UPS inhibition is unknown. While [PIN+] dramatically enhances aggregation and toxicity of polyQ and Pin4C [44,49], the visible appearance of TDP43 aggregates formed in [PIN+] vs. [pin-] cells are indistinguishable. Nonetheless, the fact that [PIN+] prion aggregates sequester some essential chaperones including Sis1 [66] could make [PIN+] cells more vulnerable to TDP-43 overexpression, e.g. the lower level of Sis1 available for the UPS in [PIN+] cells could make cells more sensitive to the inhibitory effect of TDP-43 on the UPS. Indeed, we found that overexpression of Sis1 reduces the deleterious effects of TDP-43 on cell growth, cell shape and UPS inhibition. Several genetic screens in yeast have identified modifiers of TDP-43 toxicity; however, Sis1 was not among them [59,62,81]. The importance of such screens was clearly established when one such modifier, Pbp1, led to the discovery of a new human ALS risk factor, ATXN2 harboring intermediate length polyglutamine expansions, the human homolog of yeast Pbp1 [59]. Sis1 may have been missed in the overexpression screens because of their high throughput nature. Although the Sis1 effect was weak in the strains used, it was at least as strong as that of several other modifiers that were detected (Hsp104, Pbp1 and hUpf1) (S1 Fig). Since the presence or absence of [PIN+] and other unknown differences between strains alter the toxicity of TDP-43, repeating screens for toxicity modifiers in additional yeast strain backgrounds may reveal new modifiers.
The mechanism by which overexpression of Sis1 rescues TDP-43 toxicity is unknown. Sis1 has a C-terminal substrate binding domain, an N-terminal J domain that regulates the ATPase activity of Hsp70, and a dimerization domain at the end of its C-terminus. It is largely in the nucleus, but is involved with other proteins in nucleocytoplasmic shuttling that transports ubiquitinated proteins to the nucleus for degradation by the proteasome [44]. When cells are subjected to stress, cytoplasmic diffuse Sis1 accumulates in juxtanuclear and peripheral punctate compartments [71], as well as in stress granules, which it helps target for autophagy [72].
Overexpression of Sis1 also rescues yeast from toxicity and inhibition of UPS associated with amyloid aggregates formed upon overexpression of polyQ or Pin4C [44,49]. However, unlike for TDP-43, in these cases Sis1 clearly co-localizes with the aggregates, suggesting that inhibition of Sis1 activity is the source of the toxicity and/or that Sis1 renders the aggregates non-toxic. Indeed, in these cases overexpression of Sis1 inhibits aggregate formation. In contrast, TDP-43 formed large aggregates independently of the Sis1 level, although there were also numerous small TDP-43 aggregates at reduced levels of Sis1. Possibly Sis1 helps to shear small TDP-43 aggregates so that with lower levels of Sis1 the smaller aggregates continue to grow without being sheared and become visible.
In addition, Sis1 overexpression rescues cells from toxicity associated with overexpression of Rnq1 in [PIN+] cells [50]. In that case, Sis1 overexpression caused a reduction in small soluble RnqQ1 oligomers that were presumed to be the toxic species. Also, Sis1 has been shown to help shear prion aggregates reducing the size of their detergent resistant oligomers [37,38,66,71]. No such effects of Sis1 on TDP-43 oligomer size were detected in our study.
Thus, the mechanism by which Sis1 rescues cells from TDP-43 toxicity appears to be distinct from its rescue of polyQ, Pin4C or Rnq1 toxicity. Previous studies implicated a direct effect of a different HSP40 chaperone, DNAJB6 (a member of the DNAJB7 and 8 family) on TDP-43 via interactions with the TDP-43 C-terminal domain [82,83]. In contrast, we did not detect any direct interaction between TDP-43 and Sis1 (homolog of DNAJB1 and part of the B2, B4, B5, B9 and B11 HSP40 family). Possibly TDP-43 causes toxicity and inhibits UPS for reasons unrelated to Sis1. Overexpression of Sis1 could still rescue by enhancing UPS through an independent mechanism. In support of this hypothesis, Sis1 overexpression in the presence of TDP-43 enhanced UPS activity above that seen in cells without TDP-43 (Fig 4).
Our findings that overexpression of the human homolog of Sis1, DNAJB1, reduced TDP-43 A315T toxicity in primary cortical neurons suggests that the mechanism by which Sis1 reduces cell death is conserved. Indeed, other Hsp40 chaperones have also been reported to affect the cellular formation of amyloid protein aggregates in mammalian cells [83–87]. The magnitude of the Sis1 effect seen here is similar to that seen with other genetic modifications (i.e. DBR1 knockdown) [88].
Mutations in different Hsp40 family member, DNAJB6, have been shown to cause limb-girdle muscular dystrophy (LGMD) and to inhibit induced nuclear and cytoplasmic TDP-43 aggregation induced by stress, although no effect on toxicity was found [82,83,89]. Our new results show that Sis1 (and in neurons its mammalian homolog, DNAJB1) reduces toxicity arising from TDP-43 accumulation, and furthermore, this occurs in the absence of a direct stressor such as heat. It remains to be determined if the protection afforded by Sis1, and DNAJB1 in mammalian cells, can be an effective means of extending neuronal survival in ALS and in related conditions characterized by cytoplasmic mislocalization and aggregation of DNA/RNA binding proteins such as TDP-43.
Yeast strains and plasmids used in this study are listed in Tables 1 and 2, respectively. In L3478 the C-terminus of SIS1 was endogenously tagged with mCherry by screening L1749 transformants of a PCR mcherry-HIS5 amplified fragment from p2268 from SD-His plates for the Sis1-mCherry signal. In L3491, the nuclear marker HTB1 was mCherry tagged by transforming L1749 with Cla1 linearized p2278 (pRS305-ADH1-HTB1-mCherry). Isogenic [pin-] versions of strains were obtained on YPD plates with 5 mM GuHCl and were confirmed to be [pin-] by checking for lack of aggregates after transforming with an Rnq1-GFP plasmid [90].
Unless otherwise stated, all overexpression plasmids were driven by GAL1. The TDP-43 entry clones (p2273 and p2367) were made by BP reactions between pDONR221 and a PCR amplified TDP-43 fragment with (p2273) or without a stop codon (p2367) using Gateway Technology [96] and the TDP-43 fragment in p2273 was further transferred to p2039, p2189 and p2245 to build p2055 (pAG426 GAL1-TDP43), p2275 (pAG414 GAL1-TDP43) and p2368 (pAG415-GAL1-TDP43), respectively. pDONR221-TDP-43 without stop codon (p2367) was used as a donor plasmid to build pGAL1-TDP-43-DsRed (p2173) on a destination vector (p2302). Plasmid p2195 (pAG413-GAL1-TDP43-EYFP) was constructed by a LR reaction between pDONR221-TDP-43-EYFP (p2187) and p2186 (pAG413-GAL1-ccdB). Plasmid p2288 (pRS416 GAL1-TDP-43-GFP) was generated by switching TDP-43-YFP in p2042 (pRS416 GAL1-TDP43-YFP) with TDP-43-GFP from p2041 (pRS426 GAL1-TDP43-GFP). TET promoter driven TDP-43-YFP (p2223) was constructed on pCM184, which was transformed into a Gateway expression vector (p1576).
Yeast strains were cultivated using standard media and growth conditions [97]. Rich media contained 2% dextrose (YPD). Synthetic complete media contained all amino acids except for those used for selection and 2% dextrose (SD); 2% glycerol (SG); 2% galactose (SGal); 2% raffinose (SRaf) or 2% galactose + 2% raffinose (SGal/Raf). To avoid collecting suppressor mutants that reversed TDP-43 toxicity, pGAL1-TDP-43 transformants were routinely plated and patched on plasmid selective SD medium where TDP-43 was not expressed. Patches were then replica-plated onto SG. Cells that failed to grow on SG were dropped from further study because they were petites. For analysis of growth, non-petite transformants taken from SD plates and suspended in water were normalized to an OD600 of 1.5. Then, 15 μl of 10X serially diluted cell suspensions were spotted on plasmid selective SD and SGal.
Cells containing the pGAL1-TDP-43 or control plasmids grown overnight in selective SRaf were resuspended and grown in SGal/Raf for 72 h to allow for TDP-43 overexpression. Then cells were resuspended in 1/10th volume TE and 1.5 μl were mixed on slides with 1.5 μl of 0.4% trypan blue. Different fields of the cells were then photograph under a Nikon Eclipse fluorescent microscope. The number of blue vs. unstained cells were counted blind. Between 820 and 1106 cells were counted for each sample.
All vertebrate animal work was approved by the Committee on the Use and Care of Animals at the University of Michigan (protocol # PRO00007096). All experiments were carefully planned so that we use as few animals as possible. Pregnant female wild-type, non-transgenic Long Evans rats (Rattus norvegicus) were housed singly in chambers equipped with environmental enrichment. They were fed ad libitum a full diet (30% protein, 13% fat, 57% carbohydrate; full information available at www.labdiet.com), and cared for by the Unit for Laboratory Animal Medicine (ULAM) at the University of Michigan. Veterinary specialists and technicians in ULAM are trained and approved in the care and long-term maintenance of rodent colonies, in accordance with the NIH-supported Guide for the Care and Use of Laboratory Animals. All rats were kept in routine housing for as little time as possible prior to euthanasia and dissection, minimizing any pain and/or discomfort. Pregnant dams were euthanized by CO2 inhalation at gestation day 20. For each animal, euthanasia was confirmed by bilateral pneumothorax. Euthanasia was fully consistent with the recommendations of the Guidelines on Euthanasia of the American Veterinary Medical Association and the University of Michigan Methods of Euthanasia by Species Guidelines. Following euthanasia, the fetuses were removed in a sterile manner from the uterus and decapitated. Primary cells from these fetuses were dissected and cultured immediately afterwards.
Mixed primary cortical neurons were isolated from E20 rat pups as described previously [98] and transiently transfected on day 4 in vitro (DIV4) with plasmids encoding TDP-43 variants, DNAJB1, or siRNA targeting DNAJB1 (Dharmacon ON-TARGETplus Rat Dnajb1 (361384) siRNA—SMARTpool) using Lipofectamine 2000 (Invitrogen) [74]. We then tracked neuronal survival using a system of fully automated fluorescence microscopy [8,74,99]. Background was digitally subtracted from raw images and the adjusted images were stitched together in Fiji. Stitched images were sequenced and registered (aligned) using the multistackreg plugin in Fiji, and neuron survival determined by custom-designed code written in Python. In this method cell death is marked by disruption of the cell membrane, loss of neuronal processes, or rounding of the cell body. We plotted the cumulative risk of death for neurons in each population, and compared survival between groups using Cox proportional hazards analysis. All statistical analyses were performed in R using the survival analysis package.
Imaging was accomplished using a Nikon TiE-B inverted microscope equipped with PerfectFocus3, a high-numerical aperture 20X objective lens and a 16-bit Andor iXon electron multiplied ultra-cooled charge-coupled device digital camera. A Lambda XL Xenon lamp (Sutter) illuminated the samples via a 5-mm liquid light guide. All x- and y- stage movements were performed using an ASI 2000 stage and rotary encoders. An environmental chamber that maintained 37°C and 5% CO2 was used for all experiments. All movements of the stage, shutters, and PerfectFocus3 system were controlled by BeanShell code from μManager, publicly available software that runs in ImageJ.
Rodent primary cortical neurons were isolated and transfected on DIV4 as described above. Forty-eight hours later the cells were rinsed twice in PBS, then fixed in 4% paraformaldehyde for 10 min. Following 2 more rinses, neurons were permeabilized with 0.1% Triton X-100 in PBS for 20 min at room temperature, equilibrated with 10 mM glycine in PBS for 10 min at room temperature, then blocked in 0.1% Triton X-100, 3% BSA and 0.2% goat serum in PBS for 1 hour at room temperature. Primary antibodies against DNAJB1 (Abcam ab69402, rabbit polyclonal anti-Hsp40, 1:100) were added directly to the block and the samples incubated overnight at 4°C. All cells were rinsed twice quickly and 3 times for 10 min with PBS, then placed back in block solution containing the appropriate secondary antibodies (Donkey anti-rabbit Cy5, Jackson ImmunoResearch, 711-175-152, whole IgG) at a dilution of 1:250. The cells were rinsed twice quickly in PBS, and 3 times for 10 min each in PBS containing Hoescht dye (33342, Invitrogen) at 100 nM, then twice more in PBS before imaging by automated microscopy.
Equal amounts of total proteins in precleared lysates were analyzed by Western blot as described previously [100]. To analyze TDP-43 aggregates on semi-denaturing Detergent Agarose Gel electrophoresis (SDD-AGE) proteins were extracted from cells expressing GAL1-controlled TDP-43-GFP [101,102]. Around 60 μg of crude lysate was treated with 2% sarkosyl sample buffer (25 mM Tris, 200 mM glycine, 5% glycerol, and 0.025% bromophenol blue) for 7 min at room temperature and electrophoretically resolved in a horizontal 1.5% agarose gel in a standard tris/glycine/SDS buffer, transferred to a PVDF membrane and probed with indicated antibodies. To visualized monomer TDP-43, the crude lysates were boiled at 95°C for 5 min in 2% SDS sample buffer containing 80 mM dithiothreitol (DTT) prior to electrophoresis. Blots were developed with TDP-43 rabbit polyclonal antibody from Proteintech Group (Rosemont, IL); Sis1 antibody kindly provided by E. Craig (Madison, WI); Sup35C monoclonal antibody (made in our lab by Viravan Prapapanich). GFP and PGK antibodies were purchased from Roche Applied Science (Indianapolis, IN) and Novex (Frederick, MD), respectively.
To compare levels of proteins in supernatant vs. pellet, normalized cell lysates [103] (300 μl at a concentration of 1 mg of protein /ml) were centrifuged at 90,000 rpm for 30 min at 4°C to separate supernatant and pellet fractions. After the supernatant was removed and saved, pellets were washed with the lysate buffer containing a protease inhibitor cocktail, recentrifuged at 90,000 rpm for 10 min and resuspended in 300 μl of lysate buffer. Boiled proteins in total (T), supernatant (S) and pellet (P) fractions were resolved by PAGE that was immunoblotted with anti-TDP-43. The blots were then stripped and reprobed with anti-Sis1 and the loading control anti-Pgk1.
Aggregates formed in cells by fluorescently labeled proteins were examined with a Nikon Eclipse E600 fluorescent microscope (100X oil immersion) equipped with FITC, YFP and mCherry (respectively, chroma 49011, 49003 and 49008) filter cubes. Cells were sometimes fixed with formaldehyde 3.7% final concentration for 15 minutes before washing with 0.1M potassium phosphate (pH 7.5) and were imaged over the next few days (http://openwetware.org/wiki/McClean:_Fixation_of_Yeast,Bisaria_Protocol)). Fixed and unfixed cells looked identical in shape and GFP and mCherry fluorescence in control experiments. Cells were always examined and photographed in the mCherry channel before being exposed to FITC excitation and all experiments controlled for very bright GFP fluorescence that also appeared in the mCherry channel even in the absence of an mCherry tag.
Cells containing TDP-43-GFP (p2288) made after 24 h growth in SRaf/Gal-Ura medium to induce TDP-43-GFP were harvested, and washed in ice cold water. As described previously [100], lysates were made by vortexing 800 μl cells resuspended in LB2 buffer [40 mM Tris-HCl (pH 7.5), 150 mM KCl, 5 mM MgCl2, 10% glycerol] containing anti-protease cocktail and PMSF. Then 500 μl of the precleared lysate containing 0.5–1.0 mg/ml proteins were incubated with 2 μl of TDP-43 antibody for 2 h on ice; samples were then mixed with 50 μl magnetic beads with immobilized G protein (Miltenyi Biotec Inc., San Diego, CA) and incubated on ice for 1 h. Nonspecifically bound proteins were removed in the following order of washing steps. Beads were washed with 1.0 ml of each of the following solutions at room temperature: LB2 with 1% Triton X-100 with LB2, 210 mM KCl, 1% Triton X-100; LB2 with 1% Triton X- 100; LB2; LB1 [40 mM Tris-HCl (pH 7.5), 50 mM KCl, 5 mM MgCl2, 5% glycerol]; 20 mM Tris-HCl (pH 7.6). Proteins were eluted with hot sample buffer (50 mM Tris-HCl, pH 6.8, 5% glycerol, and 0.05 and 2% β-mercaptoethanol), and then they were analyzed by electrophoresis and immunoblotting. Total (Input) protein was boiled at 95°C for 5 min in 2% SDS sample buffer containing 80 mM DTT.
|
10.1371/journal.pgen.1004534 | Accumulation of a Threonine Biosynthetic Intermediate Attenuates General Amino Acid Control by Accelerating Degradation of Gcn4 via Pho85 and Cdk8 | Gcn4 is a master transcriptional regulator of amino acid and vitamin biosynthetic enzymes subject to the general amino acid control (GAAC), whose expression is upregulated in response to amino acid starvation in Saccharomyces cerevisiae. We found that accumulation of the threonine pathway intermediate β-aspartate semialdehyde (ASA), substrate of homoserine dehydrogenase (Hom6), attenuates the GAAC transcriptional response by accelerating degradation of Gcn4, already an exceedingly unstable protein, in cells starved for isoleucine and valine. The reduction in Gcn4 abundance on ASA accumulation requires Cdk8/Srb10 and Pho85, cyclin-dependent kinases (CDKs) known to mediate rapid turnover of Gcn4 by the proteasome via phosphorylation of the Gcn4 activation domain under nonstarvation conditions. Interestingly, rescue of Gcn4 abundance in hom6 cells by elimination of SRB10 is not accompanied by recovery of transcriptional activation, while equivalent rescue of UAS-bound Gcn4 in hom6 pho85 cells restores greater than wild-type activation of Gcn4 target genes. These and other findings suggest that the two CDKs target different populations of Gcn4 on ASA accumulation, with Srb10 clearing mostly inactive Gcn4 molecules at the promoter that are enriched for sumoylation of the activation domain, and Pho85 clearing molecules unbound to the UAS that include both fully functional and inactive Gcn4 species.
| Transcriptional activator Gcn4 maintains amino acid homeostasis in budding yeast by inducing multiple amino acid biosynthetic pathways in response to starvation for any amino acid—the general amino acid control. Gcn4 abundance is tightly regulated by the interplay between an intricate translational control mechanism, which induces Gcn4 synthesis in starved cells, and a pathway of phosphorylation and ubiquitylation that mediates its rapid degradation by the proteasome. Here, we discovered that accumulation of a threonine biosynthetic pathway intermediate, β-aspartate semialdehyde (ASA), in hom6Δ mutant cells impairs general amino acid control in cells starved for isoleucine and valine by accelerating the already rapid degradation of Gcn4, in a manner requiring its phosphorylation by cyclin-dependent kinases Cdk8/Srb10 and Pho85. Interestingly, our results unveil a division of labor between these two kinases wherein Srb10 primarily targets inactive Gcn4 molecules—presumably damaged under conditions of ASA excess—while Pho85 clears a greater proportion of functional Gcn4 species from the cell. The ability of ASA to inhibit transcriptional induction of threonine pathway enzymes by Gcn4, dampening ASA accumulation and its toxic effects on cell physiology, should be adaptive in the wild when yeast encounters natural antibiotics that target Hom6 enzymatic activity.
| Cells undergo rapid transcriptional reprogramming in response to environmental changes by mobilizing transcriptional activators and repressors. Transcriptional activators function by binding to specific DNA sequences (UAS elements in yeast) and recruiting transcriptional cofactor proteins/complexes that remove repressive chromatin structure and directly recruit the transcriptional machinery to the promoters of genes under their control. Various mechanisms have been elucidated for stimulating activator function in response to environmental signals, including dissociation from a repressor, as in the case of yeast Gal4 [1], or increased entry into the nucleus as described for Pho4 and Gln3 [2]. The yeast activator Gcn4 is regulated by a unique translational control mechanism that rapidly increases the rate of Gcn4 synthesis in response to limitation for any amino acid—the conditions where increased transcription of amino acid biosynthetic genes under Gcn4 control is essential to maintaining cell growth. Gcn4 is also negatively regulated by a pathway that evokes its phosphorylation, ubiquitylation, and degradation by the proteasome, such that continued high-level translation of GCN4 mRNA is required to sustain induction of Gcn4 protein and its target genes. Together, these systems provide for reversible, short-lived induction of Gcn4, except under conditions of extreme starvation—in which protein synthesis is strongly impaired—where Gcn4 turnover is attenuated (reviewed in [3]). In addition to stimulating the transcription of genes encoding enzymes representing all of the amino acid biosynthetic pathways—the regulatory response dubbed general amino acid control (GAAC)— one-tenth or more of the yeast genome is induced by Gcn4, including genes involved in producing amino acid precursors, mitochondrial carrier proteins, vitamins and cofactors, amino acid transporters, autophagy, or the metabolism of purine, glycogen, and trehalose [4], [5].
The induction of Gcn4 expression at the translational level in amino acid-starved cells requires the protein kinase Gcn2, which is activated by uncharged tRNAs cognate to the limiting amino acid. Gcn2's sole substrate in yeast is the α subunit of general translation initiation factor 2 (eIF2). In its GTP-bound form, eIF2 delivers charged methionyl initiator tRNA (Met-tRNAiMet) to the small (40S) ribosomal subunit in the first step of translation initiation. The inactive eIF2-GDP complex is released at the end of the process and must be recycled to eIF2-GTP by the guanine nucleotide exchange factor eIF2B. Phosphorylation of eIF2α on serine-51 by Gcn2 converts eIF2-GDP from substrate to inhibitor of eIF2B, impeding the formation of the eIF2-GTP-Met-tRNAiMet ternary complex (TC). While this reduces the rate of bulk protein synthesis and limits amino acid consumption, it specifically induces translation of GCN4 mRNA owing to specialized regulatory sequences (upstream ORFs) present in the mRNA leader that couple reduced TC concentration to increased initiation at the GCN4 AUG start codon [3]. The newly synthesized Gcn4 enters the nucleus—a constitutive process for this activator [6], binds to the UAS elements of its target genes and recruits multiple cofactors to the promoter. The recruited cofactors include the nucleosome remodeling complexes SWI/SNF and RSC; and the SAGA and Mediator complexes, which carry out histone acetylation and/or function as adaptors to recruit general transcription factors and RNA polymerase II (PolII), culminating in increased assembly of transcription initiation complexes and elevated transcription of the coding sequences (CDS) [7]–[10].
In nutrient-replete yeast cells, and under conditions of moderate amino acid limitation, where Gcn2 is activated and translation of GCN4 mRNA induced, Gcn4 is a highly unstable protein owing to its ubiquitylation by ubiquitin ligase SCFCDC4 and attendant degradation by the proteasome [11]–[13]. This process helps to maintain Gcn4 at a low, basal level in nonstarved cells, and allows rapid restoration of the basal level when the translation rate of GCN4 mRNA is repressed by replenishing amino acids in starved cells. Rapid degradation of Gcn4 in sated or moderately starved cells requires its phosphorylation by the CDKs Cdk8/Srb10 and Pho85, with Pho85 making the greater contribution [12], [14]. In severely starved cells, Pho85's contribution to Gcn4 turnover is essentially eliminated, owing to the destabilization and consequent disappearance of its cyclin Pcl5 [15], [16], which accounts in large part for the stabilization of Gcn4 under these conditions. By contrast, Srb10 contributes to Gcn4 turnover under all conditions examined, making the minor contribution in sated or moderately starved cells but the major contribution in severely starved cells (where Pho85 is inactive) [14].
Despite its lesser importance in Gcn4 turnover, Srb10 appears to be responsible for clearing the fraction of Gcn4 that is sumoylated on Lys residues 50 and 58. It appears that sumoylation of Gcn4 on K50/K58 reduces its occupancy at the UAS elements of target genes in the early stages of GAAC induction during moderate starvation for Ile/Val imposed with the inhibitor sulfometuron (SM). However, the higher levels of UAS-bound unsumoylated Gcn4 that result from Arg substitutions of K50/K58 do not evoke increased PolII occupancy or higher transcription rates under these conditions in otherwise WT cells [17]. As sumoylation of transcription factors can inhibit transcriptional activation by impairing their ability to recruit RNA polymerase [18], sumoylation of Gcn4 might impair its activator function to dampen the general control response.
There is evidence that phosphorylation of Gcn4 by Srb10 or Pho85 reduces its activation function and that the phosphorylated species must be ubiquitylated and degraded by the proteasome to maintain WT basal expression of Gcn4 target genes under nonstarvation conditions. Thus, blocking proteasomal degradation of Gcn4 reduces target gene transcription in non-starved cells in a manner suppressed by eliminating CDK phosphorylation sites in Gcn4 or deleting both Srb10 and Pho85 [19]. It is unclear whether the accumulation of phosphorylated Gcn4 also impairs transcriptional activation under inducing conditions of amino acid starvation. The phosphorylation of Gcn4 by both kinases appears to be nucleus-localized, as Gcn4 mutants lacking nuclear localization signals are stabilized [6]. Srb10 is associated with the Mediator coactivator complex [20], which phosphorylates the CTD of the largest subunit of RNA polymerase II, Rpb1 [21]. The fact that Gcn4 recruits Mediator to the promoter [7], [8] is consistent with the possibility that Gcn4 participates in down-regulating its own function and stability by recruiting at least one of its inactivating CDKs [14].
Biosynthesis of threonine in yeast, as in other microorganisms and several plants, is a five-step pathway initiated with L-aspartic acid as the primary substrate [22],[23] (Fig. 1A). The absence of the threonine biosynthetic pathway in humans makes it a valuable target for drug development against fungal pathogens [24]. Transcription of at least 4 genes of the threonine pathway, HOM3, HOM2, THR1 and THR4 is under Gcn4 control (Fig. 1A) [4], [25]. Threonine biosynthesis is also subject to feedback inhibition by threonine, which inhibits the activity of the first enzyme in the pathway, aspartate kinase (Hom3), and also partially inhibits homoserine kinase (Thr1) (Fig. 1A) [26]. Peptide prolyl isomerase FKBP12 participates in the feedback inhibition of Hom3 by physical interaction between these two proteins [27], [28].
Besides threonine auxotrophy, thr1Δ and thr4Δ mutants exhibit myriad phenotypes that result from accumulation of the pathway intermediate homoserine (HS), as they are mitigated in thr1Δ hom3Δ double mutants that cannot produce HS (Fig. 1A) [29]. Accumulation of the substrate of Hom6, β-aspartate semialdehyde (ASA), also is toxic, as releasing feedback inhibition of Hom3 is lethal in hom6Δ cells (that accumulate ASA) in a manner rescued by simultaneously blocking ASA synthesis by eliminating Hom2 or Hom3 [27], [29]. However, hom6Δ mutants do not share all phenotypes of thr1Δ and thr4Δ mutants, and hom6Δ suppresses those unique to the latter mutants, implicating HS and ruling out a role for ASA in conferring many defects displayed by thr1Δ and thr4Δ cells. There is circumstantial evidence that HS toxicity results from its incorporation into proteins in place of threonine, which might evoke increased degradation of the HS-substituted proteins by the proteasome [29].
Previously, we screened the entire library of viable haploid deletion mutants of Saccharomyces cerevisiae for sensitivity to SM (SMS phenotype) to identify genes required for a robust GAAC, which allowed us to implicate various cofactors in the mechanism of transcriptional activation by Gcn4 [7], [30], and certain vacuolar sorting proteins (Vps) in maintaining high-level Gcn4 activation function in cells starved for Ile/Val [13]. In the course of that work, we also discovered that hom6Δ, thr1Δ, and thr4Δ mutants are also SMS, and undertook here to elucidate the mechanisms underlying this phenotype. In fact, it had been shown previously that thr1Δ mutants are SMS, and that this phenotype is suppressed by deleting HOM3. As SM evokes derepression of threonine pathway enzymes by Gcn4 [4] (Fig. 1A), the SMS phenotype of thr1Δ mutants was attributed to Hom3-dependent accumulation of HS, and its attendant toxicity to cellular processes, when HOM3 and HOM2 transcription is induced by Gcn4 [29]. This explanation would not apply to hom6Δ cells, however, which cannot produce HS, leading us to examine whether the SMS phenotype in this instance results from ASA accumulation and impairment of the GAAC response. The results of our analysis indicate that ASA accumulation indeed attenuates GAAC, by accelerating further the already rapid degradation of Gcn4 triggered by the CDKs Pho85 and Srb10. They further suggest that Srb10 functions primarily in efficient clearance of inactive Gcn4 molecules, enriched for sumoylated species, whereas Pho85 clears unsumoylated, highly functional Gcn4 in addition to defective species.
As noted above and displayed in Fig. 1B, yeast deletion mutants lacking HOM6, THR1, or THR4 are sensitive to sulfometuron methyl (SM), which evokes starvation for isoleucine and valine (Ile/Val) by inhibition of the ILV2-encoded biosynthetic enzyme [4]. At the SM concentration employed, growth of the hom6Δ, thr1Δ and thr4Δ strains is impaired to an extent similar to that of the gcn4Δ strain, lacking the activator of GAAC. Unlike these mutants, the hom3Δ and hom2Δ mutants grow like the wild-type (WT) strain on SM-containing medium (Fig. 1B, SC+SM). These findings indicate that thr1Δ, thr4Δ, and hom6Δ strains, but not hom3Δ and hom2Δ mutants, are sensitive to Ile/Val starvation imposed by SM. Moreover the hom6Δ mutant grows more slowly than WT (Slg- phenotype) even on medium lacking SM (Fig. 1B, SC).
To determine whether the SMS phenotypes of the thr1Δ, thr4Δ, and hom6Δ mutants reflect defective transcriptional activation by Gcn4, we measured induction of a UASGCRE-CYC1-lacZ reporter, driven by the CYC1 promoter and tandem Gcn4 binding sites from HIS4 (the UASGCRE) replacing the endogenous CYC1 UAS; and of a HIS4-lacZ reporter containing the native HIS4 5′-noncoding region. (HIS4 is a known Gcn4 target gene [5], [31].) As expected, treatment with SM for 6 h evokes a strong increase in UASGCRE-CYC1-lacZ reporter expression in WT, but not in gcn4Δ cells (Fig. 1C). Smaller induction ratios were observed for all five mutants of the threonine pathway, with the largest defect seen for the hom6Δ strain (∼75% reduction of induced UASGCRE-CYC1-lacZ expression) and the smallest defects observed for the hom3Δ and hom2Δ mutants (∼25% reductions) (Fig. 1C, left). In the case of the HIS4-lacZ reporter, the hom6Δ mutant, but not the hom3Δ or thr1Δ strains, displayed a marked (∼75%) reduction in induction by SM (Fig. 1C, right).
To confirm these findings, we measured induction of native mRNAs for HIS4 and ARG1 (another known Gcn4 target gene). Consistent with the HIS4-lacZ data, we observed induction defects for the hom6Δ mutant, but not the hom3Δ or thr1Δ strains, for both mRNAs (Fig. 1D). The magnitude of the induction defect in the hom6Δ mutant was considerably greater after 120 min versus 30 min of SM treatment, displaying ∼60% and ∼67% reductions for HIS4 and ARG1 mRNAs, respectively, at the longer incubation time, even though full induction of both mRNAs was achieved by 30 min of SM treatment in WT cells (Fig. 1D).
The foregoing results indicate that the SM-sensitivity of the hom6Δ mutant reflects a substantial defect in GAAC resulting from reduced transcriptional activation by Gcn4, which becomes more severe as starvation proceeds. By contrast, the other four threonine pathway mutants exhibit smaller defects in transcriptional activation, and the hom3Δ and thr1Δ strains actually display no detectable impairment of HIS4 and ARG1 induction by SM. The strong SMS phenotypes of the thr1Δ and thr4Δ mutants (Fig. 1B) can be reconciled with their moderate GAAC defects (Figs. 1C–D) by recalling that they accumulate the toxic intermediate HS, and that Gcn4-mediated induction of HOM2 and HOM3 under SM-induced starvation conditions is expected to elevate HS production in these strains (Fig. 1A), in the manner proposed previously for thr1Δ cells [29]. By contrast, induction of the threonine pathway during SM treatment in hom3Δ or hom2Δ mutants should have no effect on cell growth (as observed in Fig. 1B) because they cannot produce HS.
The HOM6 product, homoserine dehydrogenase (HSD), converts β-aspartate semialdehyde (ASA) into HS. If the GAAC defect in hom6Δ cells results from the absence of this reaction, then hom6 mutants that produce catalytically defective HSD should display a strong GAAC defect. Based on a crystal structure of yeast HSD, 4 active site substitutions were generated that were previously characterized for their effects on HSD catalytic activity in vitro [32]. We introduced the corresponding mutations into plasmid-borne HOM6 and examined the ability of the mutant alleles to complement the transcriptional activation defects of hom6Δ cells. As expected, introduction of WT HOM6 complemented the Slg- and SMS phenotypes on media containing threonine, and the failure to grow on medium lacking threonine, of the hom6Δ strain (Fig. 2A, SC, SC+SM and SC-Thr, respectively). Except for the E208D allele, the plasmid-borne hom6 alleles encoding HSD active site substitutions abolished complementation of the threonine auxotrophy and SM-sensitivity of the hom6Δ strain (Fig. 2A). Consistent with this, the three defective alleles failed to restore SM-induction of the UASGCRE-CYC1-lacZ reporter, whereas E208D restored a WT level of induction (Fig. 2B). Interestingly, the previously determined kinetic parameters of the hom6-E208D product indicated a reduced substrate affinity, but high-level catalytic activity, in comparison to WT HSD [32]. Accordingly, our results demonstrate that HSD catalytic activity is required for a robust GAAC response. We presume that the diminished substrate affinity of the hom6-E208D mutant does not significantly reduce the rate of converting ASA to HS in living cells.
We asked next whether the requirement for HSD activity for the GAAC response reflects a requirement for HS synthesis or, rather, the need to prevent accumulation of ASA. If the inability to produce HS is the salient defect, then supplementing hom6Δ cells with HS should restore their GAAC response. We found that a supplement of 1 mM HS restores growth on SC-Thr medium for the hom3Δ, hom2Δ and hom6Δ strains, but not for the thr1Δ or thr4Δ strains (Fig. 3A), consistent with the position of Thr1 and Thr4 downstream of HS production in the Thr pathway (Fig. 1A). A supplement of 5 mM HS was required to confer growth of the hom3Δ, hom2Δ and hom6Δ strains indistinguishable from that of WT; although this elevated HS concentration retards the growth of WT cells (Fig. 3A), presumably reflecting HS toxicity [29]. Importantly, HS supplementation did not rescue the defective SM-induction of the UASGCRE-CYC1-lacZ reporter in the hom6Δ mutant (Fig. 3B), indicating that its GAAC defect does not result from the inability to produce HS.
If accumulation of the Hom6/HSD substrate (ASA) in hom6Δ cells is responsible for the GAAC defect, then the GAAC response should be restored by preventing ASA production by eliminating Hom2 or Hom3; moreover, the GAAC defect should be exacerbated by eliminating feedback inhibition of the Hom3 product (Fig. 1A). Indeed, deleting HOM2 or HOM3 in the hom6Δ mutant restored SM-induction of the UASGCRE-CYC1-lacZ reporter essentially to the same levels observed in the hom2Δ or hom3Δ single mutants (Fig. 3C). Introducing WT HOM3 into the hom3Δ hom6Δ strain reinstated a defect in SM-induction of UASGCRE-CYC1-lacZ similar to that seen in the hom6Δ single mutant (Fig. 3D). Importantly, a relatively greater induction defect was observed when the feedback-resistant allele hom3-E282D (dubbed HOM3fbr) was introduced instead into the hom3Δ hom6Δ strain (Fig. 3D, cf. last two columns). As expected, introduction of HOM3fbr into the hom3Δ hom6Δ strain confers a strong Slg- phenotype (Fig. S1), owing to accumulation of ASA and its toxic effects on cell growth [27]. These findings demonstrate that the GAAC defect in hom6Δ cells results from ASA accumulation.
We sought next to determine whether ASA accumulation impairs the GAAC by reducing Gcn4 abundance. Starvation for Ile/Val by SM rapidly increases Gcn4 synthesis by inducing the translation of GCN4 mRNA [3]. Western analysis of WT cells reveals the expected rapid induction of Gcn4 after only 30 min of SM treatment, with a gradual decline in abundance as starvation continues up to 120 min [9] (Fig. 4A and B). Gcn4 abundance was decidedly reduced over much of the time course of SM treatment in vector transformants of the hom6Δ strain, again reaching its lowest level at 120 min of SM treatment. This reduction in Gcn4 abundance was mitigated by the absence of HOM3 in the hom6Δ hom3Δ double mutant, and exacerbated in transformants of the double mutant harboring feedback-resistant HOM3fbr, in which ASA accumulation is eliminated or exacerbated, respectively (Fig. 4A and B). Even after only 30 min of SM treatment, the hom6Δ HOM3fbr strain displayed low-level Gcn4 similar to that observed in hom6Δ/vector transformants after prolonged SM treatment for 120 min. The gradual decrease in Gcn4 abundance in hom6Δ cells (Fig. 4B) is consistent with the greater reduction in Gcn4 target gene transcription seen at 120 min versus 30 min of SM treatment (Fig. 1D). Moreover low-level HIS4 mRNA was observed in the hom6Δ HOM3fbr strain even after only 30 min of SM treatment (Fig. S2).
To determine whether the reduced Gcn4 abundance on ASA accumulation reflects decreased translation of GCN4 mRNA, we assayed a GCN4-lacZ fusion shown to be a faithful reporter of GCN4 transcription and the translational efficiency of GCN4 mRNA [33], [34]. Expression of this reporter shows the expected ∼10-fold induction in WT cells after 6 h of SM treatment, which is dampened somewhat both in the hom6Δ strain and in HOM3 transformants of the hom6Δ hom3Δ double mutant, but not in the vector transformants of the same strain (Fig. 4C). However, the HOM3fbr and HOM3 transformants of the double mutant exhibit indistinguishable levels of GCN4-lacZ expression. Similar results were obtained after only 1 h or 2 h of SM treatment (Fig. 4D), with the hom6Δ strain and HOM3fbr transformants of the hom6Δ hom3Δ double mutant both exhibiting similar reductions in GCN4-lacZ expression of ∼33% compared to the WT strain. As expected, a gcn2Δ mutant, lacking the key activator of GCN4 mRNA translation [3], is completely defective for GCN4-lacZ expression (Fig. 4D). While these findings suggest a reduction in Gcn4 synthesis on ASA accumulation in cells lacking HOM6, the ∼33% reductions in GCN4-lacZ expression observed in the hom6Δ and hom6Δ HOM3fbr strains do not account for the 60–70% reductions in Gcn4 abundance observed after 2 h of SM treatment in the same strains. These findings suggest that Gcn4 is also degraded more rapidly than usual in response to ASA accumulation.
To provide direct evidence supporting this last conclusion, we measured the turnover of newly synthesized Gcn4 by a pulse-chase experiment. Cells were cultured with SM for 30 min and pulse-labeled with [35S]-methionine/cysteine for the last 15 min of the starvation period, and then chased with excess nonradioactive methionine/cysteine. Consistent with previous reports, Gcn4 is normally a highly unstable protein and decays with a half-life of ∼10–12 min in SM-treated WT cells (Fig. 4E–F) [13]. Importantly, Gcn4 decay was markedly accelerated in the hom6Δ HOM3fbr strain, with the Gcn4 half-life dropping below 5 min, thus confirming that Gcn4 is degraded more rapidly in response to ASA accumulation (Fig. 4E–F).
As noted above, rapid degradation of Gcn4 is dependent on its phosphorylation by Pho85 and Srb10 in the nucleus, leading to its ubiquitylation and degradation by the proteasome [3]. It was also shown that the DNA-binding activity of Gcn4 is required for its sumoylation [17]. While it has been assumed that phosphorylation of Gcn4 by Srb10 likewise requires its binding to the UASGCRE [35], this has not been directly demonstrated. We hypothesized that the increased rate of Gcn4 turnover on ASA accumulation results from its increased phosphorylation by Pho85 or Srb10 and attendant degradation by the proteasome; and wished to determine whether, like sumoylation, the increased phosphorylation occurs when Gcn4 is bound to the UASGCRE. To this end, we asked whether inactivating the DNA-binding ability of Gcn4 would suppress the effect of ASA accumulation on its abundance by conducting Western analysis of Gcn4 variants described previously [36] lacking the C-terminal basic region or leucine zipper, which are both required for DNA binding by Gcn4 [37]. The variant lacking the DNA binding domain, gcn4-Δ235-250, also lacks one of two nuclear localization sequences (NLS2) identified in Gcn4, whereas the variant lacking the leucine zipper, gcn4-Δ251-281, retains both NLSs, and it was shown that the leucine zipper is dispensable for nuclear localization of GFP-tagged Gcn4 [6]. We verified that both gcn4 alleles are indistinguishable from deletion of the entire GCN4 coding sequence in the inability to permit growth on SM-containing medium (Fig. S3). Western analysis of SM-treated cells revealed that both variants differ dramatically from WT Gcn4 and display no detectable reduction in abundance in hom6Δ cells treated for 2 h with SM (Fig. 5A). (Note that both truncated variants are well expressed and show the expected increased electrophoretic mobility compared to WT Gcn4 (Fig. 5A)). Furthermore, introducing HOM3fbr into hom6Δ cells, which severely diminishes WT Gcn4 after only 30 min of SM treatment (Fig. 4A), has no effect on abundance of the gcn4-Δ235-250 and gcn4-Δ251-281 mutant proteins (Fig. S4A). While these results suggested that ASA evokes accelerated degradation only when Gcn4 is capable of UASGCRE-binding, it was possible that the greater stability of the mutant variants results from a failure to accumulate ASA on SM treatment owing to the absence of Gcn4-mediated derepression of threonine biosynthetic enzymes required for ASA production.
To address this last possibility, we repeated the experiment with the gcn4-Δ251-281 strains containing or lacking HOM6 after introducing WT GCN4 to reinstate the GAAC. Now we observed that the abundance of the truncated gcn4-Δ251-281 product was strongly reduced in hom6Δ cells, mirroring the behavior of full-length WT Gcn4 present in the same cells (Fig. 5B). The same results were observed in the corresponding strains also containing HOM3fbr (Fig. S4B). These results indicate that UAS-binding by Gcn4 is not required for its rapid degradation on ASA accumulation. Considering that Pho85 is responsible, whereas UAS-binding is dispensable, for the bulk of Gcn4 turnover under normal growth conditions [6], [12], [14], these findings are consistent with the possibility that Pho85 plays a prominent role in the accelerated degradation of Gcn4 evoked by excess ASA.
We sought next to determine the contributions of Srb10 and Pho85 to the enhanced degradation of Gcn4 in response to excess ASA. Consistent with previous findings [12], [14], deletion of either SRB10 or PHO85 increases the abundance of Gcn4 in otherwise WT cells treated with SM, with pho85Δ evoking a somewhat greater increase than srb10Δ (Fig. 5C, lanes 1,5,9; Fig. 5D, vector transformants). As already shown above, Gcn4 abundance is severely diminished after 120 min of SM treatment in hom6Δ or hom6Δ HOM3fbr cells compared to the isogenic HOM6 cells (Fig. 5C, lanes 3–4 vs. 1–2). Importantly, eliminating SRB10 almost completely eliminates this reduction in Gcn4 abundance in both hom6Δ and hom6Δ HOM3fbr cells treated with SM (Fig. 5C, lanes 7–8 vs. 3–4; & Fig. 5D). The slower migrating Gcn4 species evident in WT cells is considerably reduced in srb10Δ cells, suggesting that it represents a phosphorylated form of Gcn4 that depends on Srb10, which is generally consistent with previous results [14].
Deletion of PHO85 also suppresses the reduction in Gcn4 abundance evoked by SM treatment of hom6Δ or hom6Δ HOM3fbr cells (Fig. 5C, cf. lanes 3–4 vs. 11–12). In fact in pho85Δ strains, Gcn4 abundance is higher in hom6Δ and hom6Δ HOM3fbr cells (where ASA accumulates) compared to the isogenic HOM6 cells (Fig. 5C, cf. lanes 11–12 vs. 9–10; and Fig. 5D). Interestingly, deleting PHO85 seems to increase the relative abundance of the slower migrating Gcn4 species, which presumably represent products of Srb10 phosphorylation that are not efficiently cleared in pho85Δ cells (Fig. 5C, cf. lanes 9–12 vs. 1–4). The results in Figs. 5C–D suggest that both Srb10 and Pho85 are required for the strong depletion of Gcn4 that occurs on ASA accumulation. The stronger effect of pho85Δ versus srb10Δ on Gcn4 abundance observed on ASA accumulation in these experiments is consistent with previous results indicating a relatively greater contribution of Pho85 to Gcn4 degradation under normal growth conditions [12], [14].
Gcn4 was found to be phosphorylated in vitro by Srb10 on multiple CDK consensus sites, including Ser17, Ser210, Thr61, Thr105 and possibly Thr165 [14], and by Pho85 both in vivo and in vitro on Thr165 [12]. Moreover, the T165A substitution alone was sufficient to confer marked stabilization of Gcn4 in vivo [12]. Importantly, we found that the Gcn4-T165A variant showed no reduction in abundance in SM-induced hom6Δ cells compared to HOM6 cells (Fig. 5E–F). Moreover, replacing WT GCN4 with the GCN4-T165A allele in hom6Δ cells restored UASGCRE-CYC1-lacZ reporter (Fig. 5G) and ARG1 mRNA expression (Fig. 5H) to levels essentially equivalent to those seen in HOM6 GCN4 cells. Expression of HIS4 mRNA also was boosted by GCN4-T165A in hom6Δ cells, although expression remained below that seen in HOM6 GCN4 cells (Fig. 5H), suggesting either that the Gcn4-T165A variant is not functionally equivalent to WT Gcn4 or that a fraction of Gcn4-T165A rescued in hom6Δ cells has a lower than WT specific activity. In any event, these findings provide strong evidence that phosphorylation of T165 by Pho85 and/or Srb10 is required for the pronounced depletion of Gcn4 evoked by ASA accumulation in hom6Δ cells.
Having found that removing either Srb10 or Pho85 restores high-level Gcn4 abundance during ASA accumulation in hom6Δ cells, we expected to find that transcriptional activation by Gcn4 would likewise be restored in both hom6Δ srb10Δ and hom6Δ pho85Δ strains, particularly since these CDKs have been implicated in reducing Gcn4 activation function via phosphorylation of Gcn4 [19]. However, we observed distinct differences in the activation function of Gcn4 in cells lacking Srb10 versus Pho85.
First, we found that eliminating PHO85 restores the ability of hom6Δ cells to grow on SM containing plates (Fig 6A). By contrast, hom6Δ srb10Δ cells cannot grow on SM medium, even though HOM6 srb10Δ cells grow at the WT rate on SM medium (Fig. 6A). These findings suggest that deletion of SRB10 does not rescue the defective GAAC response to SM treatment in hom6Δ cells, whereas deletion of PHO85 does. Consistent with the growth assays, we found that eliminating PHO85 fully restores transcriptional activation of HIS4 and ARG1 in hom6Δ cells, conferring even higher than WT levels of both transcripts in the hom6Δ pho85Δ double mutants (Fig. 6B). It is noteworthy that deleting HOM6 provokes no reduction in HIS4 or ARG1 mRNAs, and even seems to elevate HIS4 mRNA, in pho85Δ cells (Fig. 6B). By contrast, deleting SRB10 evokes little or no increase in HIS4 or ARG1 mRNA levels in SM-treated hom6Δ cells (Fig. 6B). The failure of srb10Δ to rescue activation of these genes in hom6Δ cells cannot be attributed simply to the loss of a coactivator function of Srb10 [10], as srb10Δ had little or no effect on levels of HIS4 or ARG1 mRNAs in otherwise WT HOM6 cells (Fig. 6B, srb10Δ vs. WT), nor on the ability to grow in SM medium (Fig. 6A, srb10Δ vs. WT). These findings suggest that the Gcn4 molecules rescued from accelerated degradation on ASA accumulation by elimination of Srb10 are relatively nonfunctional in transcriptional activation. By contrast, the Gcn4 molecules rescued from degradation by elimination of Pho85 from hom6Δ cells appear to include highly functional species capable of evoking a greater than WT level of transcriptional activation.
Having found above that deleting PHO85 restores a higher level of Gcn4 in hom6Δ cells than does deleting SRB10 (Fig. 5D), it was important to determine whether the higher level of transcriptional activation seen in hom6Δ pho85Δ versus hom6Δ srb10Δ cells (Figs. 6A–B) arises simply from relatively greater UAS occupancy by Gcn4 in hom6Δ pho85Δ cells. To address this possibility, we conducted ChIP analysis to measure the occupancy of Gcn4 at the ARG1 UAS and the occupancies of Rpb3 (a PolII subunit) at the promoter (TATA element) and the 5′ or 3′ ends of the CDS at ARG1 after 2 h of SM treatment. It was shown previously that SM treatment of WT cells evokes large increases in occupancies of Gcn4 and Rpb3 at ARG1 that are completely absent in gcn4Δ cells [9], [10]. Importantly, these increases in occupancy are strongly diminished in SM-treated hom6Δ cells (Fig. 6C), providing direct evidence that the GAAC defect in the hom6Δ mutant results from low-level Gcn4 occupancy of the UAS with attendant reduced recruitment of PolII to the promoter.
As expected from the ability of srb10Δ to restore cellular Gcn4 abundance in hom6Δ cells (Fig. 5C–D), Gcn4 occupancy of the ARG1 UAS is substantially higher in hom6Δ srb10Δ versus hom6Δ SRB10 cells (Fig. 6C, blue bars). However, this increase in Gcn4 occupancy is associated with much smaller increases in Rpb3 occupancies at all three locations at ARG1 (Fig. 6C, orange, green, purple bars), consistent with the idea that the Gcn4 recovered in hom6Δ srb10Δ cells is relatively inactive.
Deletion of PHO85 had strikingly different consequences on Gcn4 activity. In HOM6 cells, the pho85Δ mutation evokes a reduction in UAS occupancy of Gcn4, but actually increases Rpb3 occupancies compared to the WT strain (Fig. 6C, pho85Δ vs. WT), which is consistent with the higher than WT levels of ARG1 mRNA in pho85Δ cells shown above (Fig. 6B). This effect of pho85Δ was noted previously [17], and is not understood mechanistically; however, it might indicate that Pho85 clears fully functional Gcn4 molecules as a homoeostatic mechanism to prevent hyperinduction of the GAAC response, such that UAS-bound Gcn4 has a greater than WT specific activity in pho85Δ cells.
Despite the much higher total cellular abundance of Gcn4 observed in hom6Δ pho85Δ versus hom6Δ srb10Δ cells (Fig. 5D), Gcn4 occupancy of the ARG1 UAS is comparable in these two strains (Fig. 6C, blue bars). In contrast, the Rpb3 occupancies at all three locations at ARG1 are substantially higher in the hom6Δ pho85Δ versus hom6Δ srb10Δ cells (Fig. 6C, orange, green, purple bars). In fact, the Rpb3 occupancies observed in hom6Δ pho85Δ cells exceed those in WT cells despite a lower than WT level of Gcn4 UAS occupancy in the mutant cells (Fig. 6C). These findings suggest that the Gcn4 molecules rescued in hom6Δ pho85Δ cells that are capable of UAS binding have a greater than WT specific activity.
As elaborated in the Discussion, the reduced ability of UAS-bound Gcn4 to activate transcription in the hom6Δ srb10Δ double mutant could be explained by proposing that Gcn4 is rendered less functional in response to ASA accumulation and that Srb10 is required to clear the inactive Gcn4 molecules from the promoter by targeting them for degradation. The apparent hyperactivity of UAS-bound Gcn4 in hom6Δ pho85Δ cells could be explained by proposing that Pho85 targets both fully functional Gcn4 and defective species rendered incapable of UAS-binding on ASA accumulation in hom6Δ cells.
It was shown recently that Gcn4 is sumoylated at target gene promoters, and Srb10 was implicated in clearing these sumoylated Gcn4 molecules [17]. We considered the possibility that sumoylation of Gcn4 bound to promoters increases on ASA accumulation and enhances the clearance of inactive Gcn4 by Srb10. If so, we would expect to find elevated sumoylation of Gcn4 in srb10Δ hom6Δ strains, but not in pho85Δ hom6Δ strains. To examine this possibility, we immunoprecipitated Gcn4 from whole cell extracts (WCEs) and probed the immune complexes with antibodies against Smt3 (yeast SUMO). After normalizing the Smt3 signal for Gcn4 abundance in the immune complexes, we observed that the Gcn4 present in hom6Δ srb10Δ cells after 2 h of SM treatment has an ∼2-fold higher level of sumoylation than observed in WT or hom6Δ cells under the same conditions, whereas sumoylation of Gcn4 is ∼2-fold lower in hom6Δ pho85Δ compared to WT or hom6Δ cells (Fig. 7A–B).
The finding that sumoylated Gcn4 is elevated specifically in the srb10Δ hom6Δ strain supports the idea that sumoylation of Gcn4 increases during SM-starvation of hom6Δ cells and that Srb10 targets the sumoylated Gcn4 for clearance from target promoters. As a result of Srb10 function, the proportion of Gcn4 that is sumoylated should not increase in response to ASA accumulation in hom6Δ SRB10 cells, as we observed (Fig. 7B). By the same token, the fact that elimination of PHO85 from hom6Δ cells does not significantly alter the sumoylation of Gcn4 following SM treatment implies that Pho85 plays little role in clearing sumoylated Gcn4 and is therefore restricted primarily to clearing unsumoylated Gcn4, which would include Gcn4 molecules not bound to the UASGCRE.
It was shown that sumoylation of Gcn4 at Lys50 and Lys58 contributes to clearing Gcn4 from the UAS via Srb10 phosphorylation in the early stages of SM induction; and this process is eliminated by arginine substitutions at both Lys residues [17]. To determine whether sumoylation of Gcn4 stimulates the clearing of Gcn4 from promoters on ASA accumulation, we examined the effects of the K50R and K58R substitutions on Gcn4 abundance in SM-treated hom6Δ cells. We found that the Gcn4-K50R, K58R mutant displayed a reduction in abundance on SM-treatment of hom6Δ cells very similar to that observed for WT Gcn4 (Fig. 7C–D). The K50R, K58R substitution also had no effect on Gcn4 abundance in SM-treated pho85Δ and pho85Δ hom6Δ cells (Fig. S6), where turnover of Gcn4 is dependent on Srb10, thus suggesting that Srb10-dependent degradation of Gcn4 on ASA accumulation is not enhanced by sumoylation of Lys50/Lys58. We also found that, in otherwise WT cells, the Gcn4-K50R, K58R variant confers essentially WT SM-induction of the UASGCRE-CYC1-lacZ reporter and HIS4 and ARG1 mRNAs, in accordance with previous findings [17]; and that the Gcn4-K50R, K58R variant resembles WT Gcn4 in being unable to sustain efficient SM-activation of UASGCRE-CYC1-lacZ, HIS4 and ARG1 expression in hom6Δ cells (Fig. 7E–F). Thus, although our data suggest that sumoylation of promoter-bound Gcn4 increases on ASA accumulation, and that the sumoylated Gcn4 molecules are cleared from the promoter primarily by Srb10, as concluded previously [17], the sumoylation of Lys50/Lys58 is not critically required for the enhanced degradation of Gcn4 that occurs under conditions of ASA excess.
In this report we have shown that accumulation of ASA in hom6 mutants lacking functional homoserine dehydrogenase (HSD) impairs the GAAC response to starvation for Ile/Val by accelerating degradation of the activator Gcn4. It is remarkable that ASA accumulation increases the rate of Gcn4 turnover considering that Gcn4 is already exceedingly short-lived under normal growth conditions [11], [12], [14]. The effect of ASA accumulation in reducing Gcn4 abundance and occupancy of the ARG1 UAS was mitigated in mutants lacking either of the CDKs, Srb10 and Pho85, known to phosphorylate Gcn4 and target it for ubiquitylation and rapid degradation by the proteasome. Deletion of SRB10 restored an essentially WT level of cellular Gcn4 in Ile/Val-starved hom6Δ cells, whereas deletion of PH085 conferred an even greater than WT level of cellular Gcn4 in starved hom6Δ cells. Similarly, mutating a key phosphorylation site of Pho85 and possibly Srb10, Thr-165, also rescued WT Gcn4 abundance in Ile/Val-starved hom6Δ cells. These findings are consistent with the model that ASA accumulation evokes an increased rate of phosphorylation-dependent degradation of Gcn4 by the proteasome. While both Srb10 and Pho85 are required for the accelerated Gcn4 turnover, it appears that Pho85 plays the larger role—just as observed under normal growth conditions [12]. This last conclusion is consistent with our finding that UASGCRE-binding by Gcn4 is dispensable for its rapid turnover on ASA accumulation, which is also the case under normal growth conditions [6].
Interestingly, the outcome on the GAAC response differed significantly depending on which of the two CDKs was eliminated in hom6Δ cells. On removal of Srb10 from hom6Δ cells, the recovery of UAS-bound Gcn4 was accompanied by only a small increase in transcriptional activation of ARG1, such that srb10Δ hom6Δ cells cannot grow on SM medium. By contrast, hom6Δ cells lacking Pho85 can grow on SM medium, and we observed an even greater than WT activation of ARG1 transcription conferred by essentially the same level of UAS-bound Gcn4 seen in hom6Δ srb10Δ cells, which is actually less than the UAS occupancy of Gcn4 found in fully WT cells. It could be argued that the low-level activation of ARG1 transcription seen in the srb10Δ hom6Δ double mutant reflects the requirement for Srb10 for efficient activation by Gcn4 observed previously [7], [10]. However, here we observed no effect of deleting SRB10 on cell growth, and little or no effect on the induction of ARG1 and HIS4 mRNAs or PolII occupancy of ARG1 CDS in otherwise WT SM-treated cells; and the small defects we observed seem inadequate to explain the nearly complete absence of increased ARG1 and HIS4 transcription and PolII occupancy at ARG1 occurring in SM-treated hom6Δ srb10Δ cells. Hence, we favor the alternative explanation that the specific activity of the UAS-bound Gcn4 rescued by eliminating Srb10 in hom6Δ cells is lower than that rescued by eliminating Pho85 in hom6Δ cells. This in turn suggests that these CDKs target different populations of Gcn4.
The notion that Srb10 and Pho85 recognize different populations of Gcn4 also fits with our demonstration that Gcn4 is more highly sumoylated in Ile/Val-starved srb10Δ hom6Δ cells than in starved WT or hom6Δ cells, whereas Gcn4 is hypo-sumoylated in Ile/Val-starved pho85Δ hom6Δ cells. This finding is consistent with the previous conclusion that Srb10 is required to clear sumoylated Gcn4 from promoters [17]. Hence, we suggest that the putative population of defective Gcn4 molecules that are phosphorylated by Srb10 and subsequently degraded also tend to be hyper-sumoylated. However, we found that sumoylation of the known sites of this modification in Gcn4, Lys50/Lys58, was unimportant for the accelerated degradation of Gcn4 in Ile/Val-starved hom6Δ cells. Thus, while sumoylation appears to be a characteristic of Gcn4 molecules that are phosphorylated by Srb10 and subsequently cleared from the promoter, we have no evidence that sumoylation enhances the unusually rapid degradation of these Gcn4 molecules that occurs during ASA accumulation.
To explain in greater detail our proposal that Srb10 and Pho85 target distinct populations of Gcn4, we begin by positing that phosphorylation of the Gcn4 activation domain (AD) by Srb10 and Pho85 occurs most rapidly when the AD is not engaged with coactivators at the promoter. Hence, both functional and non-functional Gcn4 molecules not bound to the UAS would be susceptible to rapid turnover, whereas UAS-bound Gcn4 would turn over more slowly unless it harbors a damaged or modified AD that cannot engage with coactivators. Pho85 is located in the nucleus [6]; however, we have observed only low-level recruitment of Pho85 to the ARG1 UASGCRE by ChIP analysis, at a level decidedly smaller than that seen for Srb10 (Fig. S7) or other Mediator subunits [38]. Moreover, Pho85 is responsible for the majority of Gcn4 degradation under both nonstarvation conditions and moderate-starvation conditions where the Pho85/Pcl5 complex is abundant [12], which includes our SM-induction conditions. This can explain the previous finding that Gcn4 DNA binding activity is dispensable for rapid Gcn4 turnover under such conditions [6]. Thus, we envision that Pho85 primarily targets Gcn4 molecules when they are not bound to a UASGCRE. In contrast, Srb10 is recruited by Gcn4 to the ARG1 promoter and, hence, likely plays a prominent role in the degradation of UAS-bound Gcn4 molecules that become disengaged from coactivators either stochastically or because of damage or modification of the AD (Fig. 8A). Again, this proposal is consistent with the previous finding that Srb10 is required to clear sumoylated Gcn4, as sumoylation is impaired by mutations that impair UAS-binding by Gcn4 [17].
There is evidence that Gcn4 is deactivated under normal growth conditions by Srb10 and Pho85, and that the phosphorylated, inactive protein must be degraded by the proteasome to prevent a reduction in the specific activity of UAS-bound Gcn4 [19]. We suggest that ASA accumulation in hom6Δ cells provokes damage or modification of Gcn4 that increases its rate of phosphorylation and subsequent turnover by the proteasome. The inability of the putative damaged or modified Gcn4 molecules to bind to the UAS or engage with coactivators could be responsible for their enhanced phosphorylation. Because eliminating the DNA binding activity of Gcn4 did not abolish its rapid turnover in hom6Δ cells harboring an intact GAAC, and DNA binding is not required for the Pho85-dominated turnover of Gcn4 under normal conditions [6], [12], we propose that Pho85 plays a predominant role in targeting the putative defective Gcn4 molecules generated under ASA excess, presumably when they are dissociated from the promoter; whereas Srb10 would make a lesser contribution and mediate the rapid degradation of defective, UAS-bound Gcn4 species (Fig. 8B). Accordingly, eliminating Srb10 will spare from degradation defective Gcn4 molecules that are capable of UAS binding, and because Pho85 will continue to target functional molecules when they become disengaged from the UAS, the specific activity of UAS-bound Gcn4 should decline in srb10Δ cells, as we observed. By contrast, eliminating Pho85 will rescue both damaged molecules incapable of UAS binding as well as functional Gcn4 molecules that are phosphorylated by Pho85 when they disengage from the UAS; and because Srb10 will continue to clear activation-defective Gcn4 species capable of UAS binding, the specific activity of UAS-bound Gcn4 should increase in pho85Δ cells, as we observed (Fig. 8B). Even in HOM6 pho85Δ cells, where no ASA accumulation occurs, the specific activity of UAS-bound Gcn4 exceeds that in WT cells, as seen both here and previously [17], and this phenomenon is also explained by our model (Fig. 8A). The proposal that Pho85 is responsible for clearing defective molecules incapable of binding to the UAS can also explain why a sizeable fraction of the Gcn4 spared from degradation in pho85Δ cells appears to be incapable of UAS binding, as indicated by the lower than WT UAS occupancy despite higher than WT cellular abundance of Gcn4 seen in pho85Δ and pho85Δ hom6Δ cells (Fig. 6C vs. Fig. 5C–D). As noted above, it is also possible that the lower than WT UAS-occupancy of Gcn4 in pho85Δ cells reflects an unknown feedback regulatory mechanism that limits Gcn4 binding to the UAS as a way to prevent hyperactivation of Gcn4 target genes beyond the elevated levels seen in pho85Δ cells.
An intriguing observation not anticipated by the model in Fig. 8 is that ASA accumulation provoked by SM-treatment of hom6Δ pho85Δ cells leads to a higher level of Gcn4 than occurs in HOM6 pho85Δ cells where ASA does not accumulate (Fig. 5C–D). We recently obtained evidence that most of this effect can be accounted for by an unexpected increase in GCN4 transcription or translation, as expression of the GCN4-lacZ reporter was found to be ∼2-fold higher in SM-treated hom6Δ pho85Δ versus HOM6 pho85Δ cells (Fig. S8).
A final interesting question is whether attenuation of the GAAC evoked by ASA accumulation is adaptive in WT yeast in the wild. Perhaps the enzyme HSD is frequently targeted for inhibition by plants, animals, or other microorganisms as a means inhibiting yeast growth. Indeed, the threonine pathway does not exist in mammals and has been identified as a valuable target for developing new antifungal therapeutics [24]. Moreover, it was shown that a strain of Streptomyces produces a natural antibiotic that targets HSD [39]. Reducing threonine biosynthesis by inhibiting HSD should activate eIF2α phosphorylation by Gcn2 and thereby reduce general protein synthesis, which is an appropriate response to limitation for threonine as a means of reducing the rate of threonine consumption. However, the concurrent transcriptional induction of Gcn4 target genes, including threonine biosynthetic pathway genes, evoked by translational upregulation of GCN4 mRNA might not be adaptive in this instance, owing to the toxic effects of ASA on cell physiology, including cytokinesis [27]. This toxicity of ASA provides a plausible rationale for the ability of this intermediate to suppress GAAC by accelerating Gcn4 turnover in the manner discovered here.
Yeast strains were grown at 30°C in rich YPD medium (1% yeast extract, 2% peptone and 2% glucose) or defined synthetic complete (SC) medium (1.45g yeast nitrogen base, 5g ammonium sulfate, 2% glucose and 2g amino acid mix per liter) lacking leucine, uracil or histidine wherever appropriate for selection of plasmids; and lacking isoleucine and valine (Ile/Val) for treatment with sulfometuron (SM) at 0.5 µg/ml. Increasing threonine in SC medium from approximately 1 mM to 2.5 mM diminished the slow growth phenotype of hom6Δ cells, and eliminated that of thr1Δ and thr4Δ cells, in SC medium lacking Ile/Val. Therefore, overnight growth to saturation was achieved in SC medium supplemented with 2.5 mM threonine and thereafter yeast strains were cultured in SC with 1 mM threonine. Accordingly, a moderate threonine limitation was imposed in our experiments and, as threonine is a precursor in Ile/Val biosynthesis (Fig. 1A), this should intensify the limitation for Ile/Val provoked by SM treatment.
Yeast strains used in this study are listed in Table 1. Yeast strains purchased from Research Genetics or previously reported were verified for all auxotrophic requirements indicated in the genotype; and gene deletions were confirmed by PCR amplification of predicted deletion junctions using primers described in Table S1. To generate HOM6 deletion strains, the appropriate hphMX4 gene deletion cassette conferring hygromycin B resistance [40] was PCR-amplified from plasmid pAG32 using primers HOM6-MX4-F and HOM6-MX4-R, thus introducing homologous flanking sequences upstream and downstream of HOM6 coding sequences, and used to delete HOM6 by transforming the appropriate strains to hygromycin B resistance on YPD agar plates. HOM6 deletion was further confirmed by demonstrating acquisition of threonine auxotrophy, except when deleted in hom2Δ or hom3Δ strains F2057 and F1929, respectively; and by PCR-amplification of predicted junction fragments containing hphMX4 and sequences upstream or downstream of HOM6 coding sequences using primer pairs HOM6-A/HphMX-R1 and HOM6-DN-R/HphMX-F1 respectively. To generate SRB10-myc13 and PHO85-myc13 strains, a myc13::HIS3MX6 cassette was PCR-amplified from plasmid pFA6a-13myc-HIS3MX6 using primer pairs SRB10-MYC13-F/SRB10-MYC13-R or PHO85-MYC13-F/PHO85-MYC13-R, respectively, and used to transform strains BY4741 and YR001to His+. Cassette insertions were confirmed by PCR analysis of genomic DNA using the appropriate primers specific for the myc13::HIS3MX6 cassette and SRB10 or PHO85, and by Western analysis of whole cell extracts (WCEs) using anti-Myc antibodies (Roche). To generate GCN4 deletion strains, plasmid pHQ1240 containing a gcn4Δ::hisG::URA3::hisG cassette was digested with SspI and used to transform strains F947 and YR006 to Ura+. Deletion of GCN4 was indicated by acquisition of SM-sensitivity and verified by PCR amplification of gcn4Δ::hisG::URA3::hisG from chromosomal DNA using primer pairs specific for sequences upstream and downstream of the GCN4 CDS. The URA3 gene was subsequently evicted by selecting for growth on medium containing 5-fluoroorotic acid.
All plasmids used in this study are listed in Table 2, and primers used in plasmid constructions are listed in Table S1. To construct pYPR010, HOM6 (chrX:689,322.690,749) was PCR-amplified from chromosomal yeast DNA of strain BY4741, using primers HOM6-HindIII-F and HOM6-BamHI-R, and inserted between the HindIII and BamHI sites in YCplac111. To construct pYPR018, pYPR020, pYPR022 and PYPR024, HOM6 in pYPR010 was mutagenized using the QuikChange II XL Site-Directed Mutagenesis Kit (Stratagene) to produce hom6 mutant alleles encoding the K117A, E208D, E208L and D219L substitutions, respectively, using sets of complementary primer pairs harboring the corresponding mutations (Table S1).
To construct pYPR028, HOM3 (chrV:256,132.258,737) was PCR-amplified from chromosomal yeast DNA of BY4741 using primers HOM3-F1 and HOM3-R1 and inserted between the SpeI and EcoRI sites in pRS313. pYPR030, containing HOM3-E282D (HOM3fbr), was generated by fusion-PCR using HOM3-F1 and HOM3-R1 as outside primers and complementary primers HOM3-E282D-F and HOM3-E282D-R encoding the appropriate mutation, and inserted between the SpeI and EcoRI sites of pRS313. To construct pYPR013, the ApaI-SpeI fragment containing GCN4 was isolated from plasmid p164, polishing the ApaI end using Klenow polymerase exonuclease activity, and inserted between the SmaI and SpeI sites of YCplac111. pYPR038 and pYPR047 were constructed by fusion-PCR using Gcn4c-SphI-F and GCN4c-SpeI-R as outside primers in combination with primers GCN4-K50,58R-F and GCN4-K50,58R-R or primers GCN4-T165A-F and GCN4-T165A-R, respectively, and pYPR013 as PCR template. The PCR products were inserted between the SphI and SpeI sites in YCplac111.
Yeast strains transformed with plasmids pHYC2 (UASGCRE-CYC1-lacZ), p367 (HIS4-lacZ), or p180 (GCN4-lacZ) were grown to saturation and diluted in two identical cultures in SC-Ura/Ile/Val at A600 = 0.5, and after 2.5 h of growth, 0.5 µg/ml SM was added to one set of cultures. Cells were harvested from untreated (unstarved) cultures after a total 6 h of growth and SM-treated cultures grown for 6 h in the presence of SM [7]. Whole cell extracts (WCEs) were prepared and assayed for β-galactosidase activity as previously described [41]. Mean specific activities were calculated from results obtained from three independent transformants.
Yeast strains were cultured to an A600 of 0.4–0.6 in SC-Ile/Val, achieving at least two cell doublings, and treated with 0.5 µg/ml SM for the indicated times or left untreated. Total RNA was isolated by hot phenol extraction as previously described [42]. RNA concentration was quantified by Nanodrop spectroscopy and analyzed for integrity by agarose gel electrophoresis and ethidium bromide staining. An aliquot of 1 µg total RNA was used for cDNA synthesis using SuperScript III First-strand Synthesis Supermix for qRT-PCR (Invitrogen) and the resulting cDNA was diluted 10-fold. qRT-PCR was performed using Brilliant III Ultra-Fast qPCR Master Mix (Agilient Technologies) using the diluted cDNA in multiplex PCR and the appropriate TaqMan probes (Table S1) to quantify ACT1 (labelled with FAM), ARG1, or HIS4 (both labelled with HEX). qRT-PCR reactions were performed in triplicate using cDNA synthesized from RNA extracted from at least two independent cultures. ARG1 or HIS4 cDNA abundance was normalized to that of ACT1 by calculating 2(−ΔCt), where ΔCt is (Ct (Target)- Ct (ACT1)). Fold changes in mRNA abundance were normalized to those measured in uninduced WT cells, or as indicated, and plotted.
ChIP assays were conducted essentially as described previously [7], [43]. Yeast strains were cultured in 100 ml SC-Ile/Val as described above for RNA isolation, treated with 0.5 µg/ml SM for 2 h or as indicated, cross-linked for 15 min with 10 ml formaldehyde solution (50 mM HEPES KOH, pH 7.5, 1 mM EDTA, 100 mM NaCl and 11% formaldehyde) and quenched with 15 ml of 2.5 M glycine. WCEs were prepared by glass beads lysis in 400 µl FA lysis buffer (50 mM HEPES KOH, pH 7.5, 1 mM EDTA, 150 mM NaCl, 1% TritonX-100 and 0.1% Na-deoxycholate) with protease inhibitors for 45 min at 4°C and the supernatant collected after removing the beads was pooled with 600 µl FA lysis buffer used for washing the beads. The resulting lysate was sonicated to yield DNA fragments of 300–500 bp and cleared by centrifugation. 50 µl aliquots of lysates were immunoprecipitated for 2 h at 4°C with α-Gcn4, (Rabbit) [13] or α-Rpb3 antibodies (Mouse, Neoclone) coupled with α-rabbit IgG or α-mouse IgG conjugated magnetic beads (Dynabeads, Invitrogen), respectively, or with α-c-Myc (Rabbit, Roche) coupled with α-rabbit IgG conjugated magnetic beads. Recovered immune complexes were washed and eluted as described [43]. For matched input and IP samples, the crosslinks were reversed by incubation at 65°C overnight, treated with proteinase K, extracted twice with phenol:chloroform:isoamyl alcohol (25∶24∶1) and once with chloroform:isoamyl alcohol (24∶1), and ethanol precipitated, resuspending the resulting pellets in 30–40 µl TE containing RNAase as described earlier [43]. Quantitative PCRs were performed in the presence of [33P]-dATP with undiluted IP DNA and 500-fold diluted input DNA and further analyzed as previously described [7], [43]. The primers employed for ChIP analysis are listed in Table S1.
WCEs were prepared in denaturing conditions with trichloroacetic acid, as described previously [44] and analyzed by immunoblotting with α-Gcd6 [45] and affinity purified α-Gcn4 antibodies [13]. Western signals were quantified by ImageJ software.
The analysis was performed essentially as previously described [13], [46]. Yeast cells collected from a 10 ml culture at A600 = 0.4–0.6 were washed with SC-Met/Ile/Val, inoculated into 0.5 ml SC-Met/Ile/Val containing 1 µg/ml SM and incubated for 15 min in a shaking water bath at 30°C; after which 1.0 mCi [35S]methionine/cysteine labelling mix was added and incubation continued for an additional 15 min. Cells were collected, transferred to 5 ml of pre-warmed SC-Ile/Val containing 10 mM methionine and 10 mM cysteine, and an 1 ml aliquots were removed immediately or after appropriate times of chase. Aliquots were denatured with 170 µl of 1.85 M NaOH, 7.4% 2-marcaptoethanol and precipitated with 70 µl 100% TCA on ice, washed with chilled acetone and dried under vacuum in a SpeedVac. The dried pellets were resuspended in 120 µl of 2.5% SDS, 5 mM EDTA, 1 mM PMSF by vortexing, boiled for 1 min, and cleared by centrifugation. Incorporation of label was measured by scintillation counting [13] and aliquots of extract containing equal amounts of radioactivity (5.7×105 cpm) were combined with 1 ml of immunoprecipitation (IP) buffer (50 mM Na-HEPES [pH 7.5], 150 mM NaCl, 5 mM EDTA, 1% Triton X-100, 1 mM PMSF) containing 1 mg/ml BSA and 1 µl affinity-purified α-Gcn4 antibodies and mixed by rotating at 4°C for 2 h. Twenty µl of a 50% slurry of protein A-agarose beads pretreated with IP buffer containing BSA (1 mg/ml) was added, and mixing continued for 2 h. The beads were washed thrice with 500 µl cold IP buffer containing 0.1% SDS, resuspended in loading buffer, boiled, and resolved by SDS-PAGE using 4 to 20% gels. The gel was dried and subjected to autoradiography, and the [35S]-labeled Gcn4 was quantified by phosphorimaging analysis.
A modification of a previously described protocol was employed [47], as follows. Yeast strains were cultured and treated with SM as described above for ChIP analysis. 40–60 A600 units of cells were lysed at 4°C with glass beads by 10 cycles of vortexing, 30s-on and 30s-off, in 500 µL of chilled lysis buffer (50 mM Tris-HCl [pH 8.0], 5 mM EDTA, 150 mM NaCl, 0.2% Triton ×100 and 1 mM PMSF) containing 10 mM sodium ethyl maleimide (NEM) and protease inhibitors. The resulting lysate was cleared by centrifugation at 13,000 rpm for 30 min at 4°C and soluble protein concentration was determined by the Bio-Rad protein assay. For each sample, a 40 µl suspension of magnetic beads conjugated with α-Rabbit IgG (Dynabeads, Invitrogen) was washed twice with lysis buffer containing 5 mg/mL BSA and rotated with 1 µl affinity purified α-Gcn4 antibody [13] in 200 µl lysis buffer/BSA for 3 h at 4°C. The magnetic beads coupled with α-Gcn4 antibody were washed twice with lysis buffer/BSA to remove unbound antibody and resuspended in 200 µl lysis buffer/BSA. Aliquots containing 1 mg of protein were added to the magnetic beads suspension, adjusting the final volume to 500 µl with lysis buffer/BSA, and further rotated for 2 h at 4°C. IP samples were washed thrice with lysis buffer containing 0.1% SDS and resuspended in 30 µl 1× Novex tris-glycine SDS sample buffer (Invitrogen) and boiled for 3 min. Aliquots of 5 µl and 25 µl were subjected to Western analysis with α-Gcn4 antibodies [13] and α-SUMO (α-Smt3) polyclonal antibodies [47].
|
10.1371/journal.pcbi.1004157 | Ligand-Target Prediction by Structural Network Biology Using nAnnoLyze | Target identification is essential for drug design, drug-drug interaction prediction, dosage adjustment and side effect anticipation. Specifically, the knowledge of structural details is essential for understanding the mode of action of a compound on a target protein. Here, we present nAnnoLyze, a method for target identification that relies on the hypothesis that structurally similar binding sites bind similar ligands. nAnnoLyze integrates structural information into a bipartite network of interactions and similarities to predict structurally detailed compound-protein interactions at proteome scale. The method was benchmarked on a dataset of 6,282 pairs of known interacting ligand-target pairs reaching a 0.96 of area under the Receiver Operating Characteristic curve (AUC) when using the drug names as an input feature for the classifier, and a 0.70 of AUC for “anonymous” compounds or compounds not present in the training set. nAnnoLyze resulted in higher accuracies than its predecessor, AnnoLyze. We applied the method to predict interactions for all the compounds in the DrugBank database with each human protein structure and provide examples of target identification for known drugs against human diseases. The accuracy and applicability of our method to any compound indicate that a comparative docking approach such as nAnnoLyze enables large-scale annotation and analysis of compound–protein interactions and thus may benefit drug development.
| Description of the “mode-of-action” of a small chemical compound against a protein target is essential for the drug discovery process. Such description relies on three main steps: i) the identification of the target protein within the thousands of proteins in an organism, ii) the localization of the binding interaction site in the identified target protein, and iii) the molecular characterization of the compound’s binding mode in the binding site of the target protein. Here, we introduce a new computational method, called nAnnoLyze, which uses graph theory principles to relate compounds and target proteins based on comparative principles. nAnnoLyze aims at addressing two of the three previous steps, that is, target identification and binding site localization. Our results suggest that the nAnnoLyze accuracy and proteome-wide applicability enables the large-scale annotation and analysis of compound–protein interaction and thus may benefit drug development.
| The number of newly approved drugs has been significantly decreasing over the last two decades [1]. To make things worse, the therapeutic dogma that has prevailed over the years aimed at single target-specific ‘magic bullets’ against each disease. However, proteins act in complex interconnected networks, and thus, this ‘one gene, one drug, one disease’ paradigm is now clearly challenged [2,3]. The polypharmacology concept, which relies on the fact that a drug can modulate its activity by interacting with multiple targets rather than just one, was proposed to address these limitations [2]. Polypharmacology is especially valid in complex diseases like cancer or central nervous system disorders where the modulation of the activity of one single protein is not sufficient to obtain a therapeutic effect [4–6]. Therefore, identification of all possible targets of a chemical compound is critical in the drug discovery process.
Many in silico methods have been published for drug target identification using network approaches [7,8]. Broadly, we can distinguish two different classes of methods, structure-free methods and structure-based methods. Within the first group, there are methods based on ligand features [9] that have been successfully used to identify numerous experimentally validated interactions. However, they have difficulties in identifying interactions for drugs with novel scaffolds [10] or for targets with no bioactivity information. Others, named network-based approaches, exploit network properties to provide the drug target interactions and drug repositioning opportunities [11–18]. Although the accuracy of predictions by these methods has significantly increased, the majority cannot explain the mode of action of the drug over the predicted target due to the lack of three-dimensional (3D) information about the ligand and/or the target. The use of 3D structural data helps addressing such limitation. The most popular structured-based methods rely on molecular docking approaches performing a virtual screening of a compound against a limited number of protein targets or of several compounds against one protein target [19–21]. As a result, they provide structurally detailed information about the likely interaction between the compound and its target/s. However, the computational requirements of such approaches make them not generally applicable at proteomic scales. An exception to this limitation is the recent massive human screening of 600,000 drugs against 7,000 human protein pockets by Cardozo and colleagues whose results are available online [22]. To overcome the computational limitations, new structure-based methods use the so-called “comparative docking” approaches that solely rely on structural comparisons, both of compounds and protein targets, to infer new interactions [23,24]. Other methods use local structural comparisons of small molecule binding sites to infer the localization and specificity of binding pockets [25,26] as well as to infer new ligand interactions in known binding pockets [27]. Finally, several other methods that rely on 3D structure comparisons that aim at functionally annotating structures [23,24,28].
Here we introduce nAnnoLyze, a network-based version of the comparative docking method AnnoLyze [23]. Our new method predicts interactions for any query compound against an entire 3D proteome by relying on a bi-partite network of interactions and similarities. Unlike Annolyze, nAnnolyze can predict interactions for any compound regardless if they have been previously co-crystallized with a protein. We have benchmarked nAnnoLyze against a dataset composed by all the interactions for approved drugs present in the Protein Data Bank (PDB) [29]. The method outperforms AnnoLyze precision by 27 folds. Both Annolyze and nAnnolyze have been already successfully applied. Annolyze was used in an open source drug discovery initiative against neglected tropical diseases [30] while nAnnoLyze has been applied to a set of anti-tubercular drugs against the Mycobacterium tuberculosis proteome [31]. Here, we describe the method alongside the predictions for all the small molecule drugs present in DrugBank [32] against the human 3D proteome. To our knowledge, this is the first screening of almost 6,000 drugs against the entire human structure proteome predicted by comparative approaches. The nAnnoLyze network, method and predictions are available online at http://www.marciuslab.org/services/nAnnoLyze.
The correct selection of a benchmark dataset is one of the most important steps in assessing the accuracy of a newly developed method. Unfortunately, there were no available and adequate datasets for benchmarking structure-based network methods for ligand-target prediction. The “Yamanishi-2008” dataset [11], which has widely been used previously, could not be used here due to the limited structural coverage of its targets, which added to the increasing concern on biases of the current drug-target interaction datasets [33]. To address these issues, we have generated a benchmark set consisting of a “positive” and a “negative” set. The “positive” set contains all drug-protein annotated pairs between any structure in the PDB and any compound approved by the FDA. The “positive” benchmark set resulted in a total of 6,282 interactions and is considered the “true” set of interactions. The “negative” set was generated by randomly selecting pairs of compounds and targets that have never been annotated in the DrugBank or PDB databases. To assess how many of these drug-protein negative pairs could result as a potentially miss-annotated negative interactions we looked for similar compounds interacting with the “negative” target of each compound. The search resulted in 118 (∼2%) out of the 5,981 pairs that could result in a miss-annotated negative interaction. However, the removal of these pairs of putative miss-annotated “negative pairs” from the set had no effect on the assessment of the nAnnoLyze accuracy. Our final benchmark dataset included thus a total of 6,282 drug-target in the “positive” interactions and 5,981 negative pairs.
The nAnnoLyze precision varies at different Z-score thresholds (Fig. 1A) with an optimal threshold at −2.5 local Z-score resulting in a precision of 0.63 and coverage of 0.19 corresponding to 1,148 true positive predictions (Fig. 1B). It is important to note that both the precision and coverage of our method depend dramatically on the definition of false positives for our predictions. Given that our benchmark set relies only on deposited data in the PDB, many of the predictions by nAnnoLyze are likely to be correct despite not being present in our benchmark. For example, the drug Enalapril (DB00584 DrugBank identifier) has been co-solved in only two PDB entries (i.e., 2X90 and 1UZE). However, nAnnoLyze predicts interactions between Enalapril and three other targets in the PDB (i.e., 2X91, 1J36 and 2X8Z). Those structures actually correspond to the same target sequence (Q10714 UniProt id) being solved with no ligands.
To further increase the accuracy of our predictions, we implemented a Random Forest Classifier (RFC) that classifies pairs of compound-protein as binders or not by combining several of the nAnnoLyze scores (that is, the raw score, the Local Z-score, and the Global Z-score). The RFC correctly recalled 66% of the pairs with a precision of 0.73 and an AUC of 0.71 using a 10-fold cross validation (Table 1). The tested RFC did not include the DrugBank ID as input feature to simulate a situation where a completely new compound not deposited in the databases was tested. However, by using the DrugBank ID as an input feature, the accuracy of nAnnoLyze dramatically improves to a 0.93 precision, 0.93 recall and a 0.97 of AUC (Fig. 1A and Table 1). These results suggest that predictions for known drugs already in our dataset are much more precise than those for unknown or anonymous compounds. The RFC outperformed the use of any of the single scores from nAnnoLyze (Table 1).
Comparatively, nAnnoLyze reached a 0.61 increase in precision at the optimal cut-off (from 0.02 to 0.63) at the expenses of a decrease in recall by 0.38 with respect to AnnoLyze (Table 2). Finally, It is important to note that the benchmark set used for this test resulted more difficult for AnnoLyze than the original test-set used to benchmark it [23] (Table 2).
The human Cyclooxygenase-1 is targeted by NSAID drugs. Cyclooxygenase (COX) is the enzyme responsible for the formation of prostanoids, which are classified in 3 different groups: prostaglandins, prostacyclins, and thromboxanes, each of them is involved in the inflammatory response, among other processes. There are two COX isoenzymes. COX-1 promotes the production of the natural mucus that protects the inner stomach lining while COX-2, is primarily present at sites of inflammation [34]. Traditional non-steroidal anti-inflammatory drugs (NSAIDs) such as Aspirin, Ibuprofen or Flurbiprofen are considered non-selective because they inhibit both COX-1 and COX-2. The inhibition of COX-2 by NSAIDs results in the anti-inflammatory effect, while the inhibition of COX-1 can lead the undesired side effects such as damage to the gastrointestinal tract [35]. nAnnoLyze predicted interactions for several NSAIDs with the 3D model of the human COX-1. Specifically, nAnnoLyze predicted 21 (out of the 44 approved drugs against COX-1) as binders of the COX-1 target (Table 3). In particular, nAnnoLyze predicted the binding of Flurbiprofen (DB00712) and Ibuprofen (DB01050) to COX-1, which are known inhibitory drugs of the human COX-1 (Fig. 2A). The nAnnoLyze path between Flubiprofen and COX-1 starts from a ligand node composed by tripotassium (1R)-4-biphenyl-4-yl-1-phosphonatobutane-1-sulfonate (B70) and two stereoisomers of Flubiprofen (FLR and FLP). Thorough the binding site of FLP to ovine COX-1 (1QEH), nAnnoLyze predicts its binding site of the COX-1 human 3D model. Conversely, the path between Ibuprofen and COX-1 starts in the ligand node composed by 1-(4-ethylphenyl)propan-1-one (I3E) and two stereoisomers of Ibuprofen (IBP and IZP). Those ligands are predicted to bind the same predicted binding site of the human COX-1 thanks to its similarity to the crystal structure of the ovine COX-1 (1EQG). Remarkably, the human COX-1 predicted binding site includes the tyrosine 385, which is known to be responsible of the catalytic reaction with the NSAID drugs (Fig. 2A). However, not all the NSAIDs performed with the same accuracy. Aspirin (DB00945), also a known inhibitor of the human COX-1 and COX-2, results in false positive predictions (Table 4 and Fig. 2B). The nAnnoLyze search with Aspirin as input molecule results in many proteases predicted targets. This false-positive pathway starts from the ligand node composed by two Benzoic Acids, the 4-Guanidinobenzoic Acid (GBS) and the Acetylsalicylic acid (AIN). GBS has been crystallized with different trypsin proteins so the pathway goes thorough the GSB binding site of the guanidinobenzoyl-trypsin acyl-enzyme (2AH4) reaching eventually the predicted binding site for the human Trypsin-2 (P07478). The same pathway is used to find other proteases like the Airway trypsin-like protease 4 (Q6ZWK6) or the Trypsin-3 (P35030) resulting in several false positive predictions. Conversely, the Aspirin-COX1 network pathway starts from the ligand node composed by 3,6-dichloro-2-methoxy-benzoic acid (D3M) and Salicylic acid (SAL) (Fig. 2B). The RFC classifier identified a network link between Aspirin and the SAL compound with a similarity score of 0.86. This SAL mediated pathway guides the nAnnoLyze search towards its binding site in the ovine COX-1 (3N8Y), which is homologous to the human COX-1 binding site. This pathway is also the responsible of the link between Aspirin and the human COX-2 with a score of 0.77. However, the lower similarity between the predicted human COX-2 binding site and the ovine COX-1 (3N8Y) introduces a penalty that significantly decreases the score of the link.
Sorafenib pathway targeting through binding of several proteins. Sorafenib, which is marketed as Nexavar, is an approved drug for the treatment of advanced renal cell carcinoma. It is also in Phase III trials for Hepatocellular carcinoma, Non-small-cell lung carcinoma (NSCLC) and melanoma and in Phase II trials for Myelodysplastic syndrome, Acute Myeloid Leukemia (AML), head and neck, breast, colon, ovarian and pancreatic cancers. Arising as one of the most promising anticancer drugs, Nexavar is known to perform its activity by targeting the Raf/Mek/Erk pathways [36,37]. Specifically it is known to inhibit Raf kinases, Receptor-type tyrosine-protein kinase (FLT3), platelet-derived growth factor (PDGF), Vascular endothelial growth factor receptor 2 & 3 (VEGF2/3) and the Mast/stem cell growth factor receptor Kit. Within our predictions, we found 4 of these links alongside other interesting links for targets involved in the same pathways (Table 5 and Fig. 3A). Interestingly, most of the links have been previously annotated either in DrugBank, PubChem or in the PDB as a crystal structure. However, there are two links not annotated within the predictions, the serine/threonine-protein kinase A-Raf (ARAF) and the Cyclin-dependent kinase 10 (CDK10). ARAF is involved in several pathways, including AML and FoxO signaling and together with FLT3, BRAF, MAPK14 could be a good opportunity to exploit the polyphamarcological profile of Sorafenib against AML. In fact, Phase II trials are showing very promising results in AML combining Sorafenib with other marketed drugs [38,39]. Of the ten predicted targets, only 3 have been co-crystallized with Sorafenib (BRAF, MAPK14 and CDK8), while in the other seven nAnnoLyze proposes the binding site localization of the drug providing insights into the mode of action of the compound. nAnnoLyze predicted the correct binding site for the three targets (Fig. 3B). The predicted binding sites were 75%, 62%, and 86% correct (i.e., % of predicted residues defined as binding site in LigBase) for CDK8, BRAF, and MAPK14, respectively.
Since structurally similar binding sites are more likely to bind the same small molecule. We wanted to assess if the 7 predicted binding sites (i.e., FLT3, CDK10, ARAF, MAPK15, FLT1, RAF1, and CDK19) have similarity with the 3 Sorafenib known binding sites (i.e., BRAF, MAPK14, and CDK8). All of the 7 predicted binding sites are similar to at least one of the already known (Fig. 3C). Within the annotated interactions with non-crystallized structure, FLT3 is the one with lowest similarity to a known structure (ProBiS Z-score of 1.04 with the CDK8 binding site). Unlike FLT3, FLT1 binding site has MAPK14 as the most similar binding site with a higher score (2.25 ProBiS Z-score). Regarding the Cyclin dependent kinases CDK10 and CDK19 proposed binding sites, CDK10 binding site has a high similarity (ProBiS Z-score of 2.09) with the MAPK14´s one while the CDK19 binding site is almost identical to that of CDK8 (ProBiS 2.9). As expected, RAF predicted protein binding sites ARAF and RAF1 have BRAF binding site as the most similar (3.5 and 3.94 ProBiS Z-scores, respectively). Following the same trend, the MAPK14 binding site is the most similar to MAPK15 (2.51 ProBiS Z-score). Although small changes in the catalytic site could have a dramatic impact on the binding-affinity of a small molecule, the overall high similarity among the Sorafenib predicted binding sites shows a clear trend towards binding site conservation within this set of proteins. This example shows not only the capability of the method to find drug targets but also the possibility to explore pathways rather than individual proteins as targets.
The increase of compound phenotypic screenings over the last years has dramatically increased the number of small molecules with non-annotated protein targets [40–42]. Because target annotation is a crucial step when developing a drug, and specifically the elucidation of the amino acids involved in the interactions is key to understand the mode of action of the compound, many methods have been developed to annotate drug protein targets. However, most of them do not provide any structural information about the link, and for those providing it, the application at proteome scale for any query compound is unfeasible. Here we introduced nAnnoLyze a method for drug target interaction prediction that provides structural details at proteome scale. nAnnoLyze relies on a pre-built network of structural similarities to perform its prediction for any query molecule providing not only the connection between the molecule and its predicted target but also the binding site of the ligand in the protein. It is important to note that nAnnoLyze has been specifically tested for drug-target interaction prediction. The accuracy of our method on less studied compounds, such as non-drug like molecules, could lead to a reduction of the precision and the coverage.
The lack of crystal structure for several proteins in other datasets prompted us to build a new dataset of approved drugs. The reduction of the precision by our previous method [23] with this dataset is indicative of the complexity of the new benchmark. The new dataset includes real set of interactions that better simulates a scenario where the different molecules have different affinities to one or many targets. This addressed a current concern about the possible bias of artificial datasets [33]. Unfortunately, the lack of a real “negative” set of drug-protein pairs (i.e., pairs of molecules known not to interact) hampered the creation of the complete dataset. To overcome this issue, we generated a set of drug-protein pairs that, so far, are not annotated as interactions. The nAnnoLyze benchmark using these newly created datasets resulted in satisfactory accuracies, especially in light of the fact that the dataset is bound to produce an overestimation of the false-positive rate (i.e., a drug and a protein are not interacting if they have not been crystallized together) [43]. The limitation of the maximum distance in the search for the shortest pathway can explain some of the missed drug-protein pairs and, consequently, limits the recall reached by the method. Analysis of the precision and recall of specific compounds in the benchmark dataset indicate that nAnnoLyze results in higher accuracy for moderate promiscuous compounds compared to highly promiscuous compounds. Indeed, promiscuous compounds have high-degrees of connectivity in our network, which makes it very difficult to identify specific targets. A similar analysis to identify trends in the accuracy of nAnnoLyze for targets for different protein Pfam families did not result in any clear trend. The usage of binding site to represent a family of targets instead of whole protein domain structures may explain the homogeneity in the performance for different protein families.
Several scores for each prediction permits to explore the effect of the selection of different thresholds values depending of the user needs. For instance, when extracting only the most confident targets for a drug, very low values of Global Z-score will be suitable; while when retrieving the most specific targets for a compound filtering by low values of Local Z-score will be the best option. This, of course, makes it difficult to provide a specific score threshold for the predictions. Despite this, we studied the variation of the performance at different thresholds measured by a ROC curve. The AUC was excellent when using drug names and scores as input feature for the predictions. When only the scores of the predictions were used (that is, treating the compound as anonymous), there was a clear decrease in the AUC suggesting that the method performs better for already known chemical entities rather than for new unseen compounds. This fact makes sense since the method is based upon comparative approaches relating compounds by their structural similarities.
The comparison of the nAnnoLyze method against the original AnnoLyze indicates that our network-based approach predicts drug-protein complexes with higher precision. Importantly, nAnnolyze is a clear progress over Annolyze by improving not only the performance (27-fold higher precision) but also the applicability, since it can be applied to any compound regardless whether it has been previously deposited in the PDB. Moreover, the network-based paradigm implemented in nAnnoLyze allows for the integration of other types of additional information such as the diseases linked to the protein targets, which may eventually allow for drug indication predictions. A successful example of a method for predicting drug-like targets using the modelable human proteome with medical data integration is the Computational Analysis of Novel Drug Opportunities (CANDO) platform [43]. While the aim of our work is accurately predicting drug-protein interactions, future developments of nAnnoLyze could include medical indications of drugs.
To demonstrate the applicability of the method, we screened all the drugs in the DrugBank database against the entire human 3D proteome that could be modeled by comparative protein structure prediction. We not only provided the drug-protein predictions but also the structural binding localization of the interaction. We carefully described two examples of this screening. The first example illustrates the nAnnoLyze ability to correctly (or incorrectly) predict the binding of a NSAIDs set of drugs to the COX-1 human protein. Within the correctly predicted interactions (i.e., true positives), we included Flurbiprofen and Ibuprofen detailed information about the network routes. In the case of the incorrectly predicted interaction between Aspirin and proteases proteins, the analysis indicates that the clustering in a ligand node of two similar Benzenoids compounds lead to the undesired drug-target association. It is thus likely that adding extra information beyond the chemical similarity during the clustering of the core-network may result in more functionally homogeneous clusters of compounds. Even though, nAnnoLyze was able to reach the two main targets of aspirin (i.e., COX-1 and COX-2) through alternatives network pathways. However, the lower similarity of the human predicted COX-2 binding site with the ovine COX-1 included in the core network penalized the score of the hit. This example also illustrates the nAnnoLyze capacity of predicting interactions when no crystal structure is available for the target.
The second example studied the polypharmacological profile of the anticancer drug Sorafenib. The method correctly retrieved most of the known targets and proposed some others with structural similarities in the binding site and that are involved in the same metabolic pathways as the known ones. This example shows the possibility of studying pathways rather than individual proteins as drug targets, which could be even more interesting in complex diseases such as cancer or Alzheimer where multiple factors play a role in the progress of the disease.
The major limitation of the method is the restricted applicability because is based on structural data, which is still scarce compared to sequence data. In spite of it, we were able to cover 42% of the human proteome with either a crystal structure or a reliable model. Moreover, the amount of crystal structures in the PDB has significantly increased over the past years [44] and the percentage of a proteome that can be modeled by homology has increased thanks to initiatives like the Protein Structure Initiative [45,46]. The more structural information we have, the more information can be extracted and therefore applied in nAnnoLyze. Indeed, the underlying network in nAnnoLyze can continue growing with the integration of new molecules or sets of biomolecules (both compounds and protein targets). To this end, we have developed a Web server that allows everyone to submit their own sets of compounds and check the predictions against pre-built networks for the human and Mycobacterium proteomes. So far, we have applied the method in an open source drug discovery initiative against Mycobacterium tuberculosis [31] and are currently working in other projects and initiatives. Our goal is to encourage open source drug discovery by releasing the method with all the predictions expecting that other researchers can benefit from our work. Finally, the scientific community could experimentally validate the predictions providing us a feedback to improve the quality of this tool and of future ones.
Next, we describe the different steps (Fig. 4A) performed to build a bi-partite network of structural similarities and interactions (Fig. 4B). We continue by describing the methods used to assess the accuracy of nAnnoLyze.
To build the ligand sub-network, only compounds with a pharmaceutical or a biological function on their co-crystallized proteins were retrieved from the PDB. To perform the filtering, we calculated, for each compound in the PDB, the weighted quantitative estimate of drug-likeness (wQED). Briefly, the wQED is calculated by combining a set of the chemical features of the compound (i.e., molecular weight, octanol-water partition coefficient as LogP, polar surface area, number hydrogen bond donors, number of hydrogen bond acceptors, number of rotatable bonds, number of aromatic rings, and number of possible toxic scaffolds) to quantify its drug-likeness given the pre-calculated value for that chemical features in a gold standard set of drugs [47]. Compounds with good drug-like properties (i.e., wQED ≥0.35) were selected resulting in 7,609 PDB compounds. Each selected compound was then represented as a vertex in the ligand network. Links between vertices of the network (i.e., edges) were obtained by structurally comparing all compounds. The weights of the edges were obtained using a Random Forest Classifier (RFC) developed to identify compound similarities [31]. Briefly, the RFC classifier predicts whether two small molecules are likely to bind the same target-binding site by comparing their structural and chemical properties. The usage of a classifier allows for an automatically determination of optimal thresholds after the RFC has been trained with the training-set. Therefore, the all-against-all comparison performed by the RFC resulted in 134,493 pairs of similar compounds. To reduce redundancy in the network we created groups of connected compounds by identifying k-cores in the network. A k-core in a network N, is a maximal connected sub-graph of N in which all vertices have degree at least k. Thus, every k-core in the non-redundant network represents a vertex and edges between vertices indicate the existence of at least one similar compound between the two k-cores. In the ligand network, a k-core would be a set of ligands such every two ligands within the set are similar to each other (i.e., they have an edge in the network). An edge between two k-cores vertices was given the maximum weight of all possible edges between their constitutive compounds. The resulting non-redundant ligand sub-network had 4,101 vertices connected by 24,856 edges.
We first downloaded from the LigBase database (February 19th, 2013) [48], a database containing all ligand-binding sites of known protein structures, all unique protein binding sites composed of at least seven residues within a radius of 5 Å, binding any of the selected 7,609 highly drug-like compounds in the ligand sub-network. We defined “highly drug-like” compounds as those compounds with very good absorption, distribution, metabolism, and excretion properties (i.e., with an wQED ≥0.35). This initial protein binding site sub-network resulted in 28,299 binding sites from 22,959 different proteins in the PDB. Next, we populated the network with links (edges) between two proteins by structurally comparing their binding sites. The structural comparison of the binding sites was performed using ProBiS [49], a tool for local structural alignment of binding sites based on geometry as well as physicochemical properties. We defined two binding sites as similar if their similarity Z-score is higher than 2.0. An all-against-all structural comparison of the selected binding sites was performed resulting in 579,155 pairs of similar binding sites. Next, we removed redundancy from the sub-network by applying a similar filtering that is used for the ligand sub-network. The final non-redundant sub-network for binding sites contained 19,487 vertices and 29,811 edges.
Finally, we joined the two sub-networks by creating edges between protein binding sites and ligands. A binding site was linked to a ligand if both have been experimentally observed to interact (i.e., a solved structure with the target and the ligand exists in the PDB). The two sub-networks were linked by 22,832 edges and the final nAnnoLyze bi-partite network contained 23,588 vertices and 54,667 edges.
To populate the nAnnoLyze network with structures for human targets, we downloaded all human 3D models deposited in ModBase (November 11th, 2013) [50–52] with at least a 1.1 ModPipe Protein quality score [53]. ModBase is a database of comparative protein structure models calculated by the automatic modeling pipeline ModPipe [53]. The likely accuracy of the ModPipe models is predicted by the ModPipe Protein Quality score defined as a composite score that includes sequence identity to the template, coverage, and the three individual scores: the alignment e-value, z-dope [54], and GA341 [55]. This resulted in a total of 31,734 reliable 3D models from 16,694 unique human target sequences. Next, we structurally compared this set of selected models to any non-redundant (90% sequence identity) set of 29,772 structures from the PDB solved with at least one ligand compound. Structural comparisons between two proteins were performed using the MAMMOTH algorithm, which is based on a fast and accurate heuristic method to find, in a sequence-independent mode, the maximal structural subset between two proteins structures [56]. Four different scores were stored for each structural superposition: percentage of sequence and structure identity for the entire protein and percentage of sequence and structure identity for the residues involved in the binding site of the known structure as defined by LigBase. The structure identity between two structures was defined as the percentage of residues with their Cα atoms within 4 Å after optimal superposition. A binding site in a model was considered then similar to a binding site in a known PDB structure if at least the binding site sequence and structure identity were higher than 40%. This identity cut-off was previously validated in a large-scale comparison of known ligand-protein pairs [23]. A total of 576,675 binding sites were predicted for the human proteins (that is, ∼18 binding sites per model). Due to the high redundancy in the predicted binding sites, we excluded binding sites fulfilling the following requisites: redundant binding sites (i.e., more than 80% sequence identity to any other binding site) or small binding sites (i.e., with less than 6 residues). A total of 64,275 binding sites (∼2 binding sites per model) remained after the redundancy and size filtering. Next, we compared all human predicted binding sites against all binding sites in our network using ProBiS resulting in 459,356 similarity links (Z-score > 1.0) between any of the human 64,275 binding sites and the 28,299 binding sites in the network. Every significant pair became an edge with a weight equal to the normalized Z-score of the comparison. The final human network included the 7,609 compounds, the 28,299 known binding sites and the 64,275 human predicted binding sites.
A total of 6,540 small compounds were downloaded from the DrugBank database (May 15th, 2013). We then looked for similarity with the compounds present in the PDB ligand sub-network by using our trained RFC classifier as described above. Next, all the drugs were integrated in our network by making an edge between every DrugBank compound and their similar PDB compounds retaining the link with higher RFC when more than one link between a DrugBank compound and one network vertex (i.e., a k-core of PDB compounds) was found. A total of 5,824 drugs were integrated into the network through 149,538 edges.
Once the network was completed, to predict all possible interactions between DrugBank compounds and any of the modeled targets of the human proteome, we simply calculated the shortest path in the network from every queried DrugBank compound to any human binding sites. We implemented a version of the Dijkstra algorithm that limits the maximum reachable distance in order to speed up the computational time of the search [57]. Each hit was then scored by using the inverse of the sum of all edge weights of the path between the compound and the human target. Such score was then normalized and Z-scored. Specifically, two different Z-scores were calculated for each prediction.
The “Global Z-score” (Gz) is obtained by running the predictions of all drugs present in DrugBank against all targets, obtaining a global mean (μG) and a global standard deviation (σG) to Z-score a specific predicted pair. The “Global Z-score” represents how good is a prediction given its score in the constructed network.
The “Local Z-score” (Lz), is similarly calculated by running the predictions of all drugs present in DrugBank retrieving the mean (μL) and the standard deviation (σL) of the score for a specific target. The “Local Z-score” represents how good is a prediction for a specific binding site or target. For example, highly promiscuous binding sites tend to have higher local Z-scores.
Finally, we combined the three scores (that is, the inverse of the sum of all edge weights, the global Z-score and the local Z-score through a Random Forest Classifier that aims at predicting the interaction of a compound and a target. Two RFCs were trained with and without the DrugBank ID as an input feature of the compound. The RFC classifier, thus, results in a single Boolean score indicating interaction or non-interaction between the compound and the target. To train the RFC, we used the Weka software for data mining tasks [58].
To benchmark nAnnoLyze, we retrieved all the compound-protein complexes for DrugBank approved drugs from the PDB. A total of 213 approved drugs were uniquely mapped into compounds bound to a protein deposited in the PDB. Next, we retrieved all the proteins binding to those compounds resulting in a protein-compound set of 6,282 entries. To test the method, we first created the benchmark network: the 213 compounds were integrated in the clustered network by using the RFC classifier. To avoid overestimation in the benchmark, we did not create any edge between a ligand in the benchmark and any identical (i.e., RFC score of 1.0) ligand in the network. Next, we extracted from LigBase the 7,074 protein binding sites of the 213 aforementioned compounds and integrated them in the network following the procedure used for the human binding sites. Similarly, we did not create links between identical binding sites in the benchmark and any protein in the network. We then selected all interactions between the 213 compounds and any of the 7,074 binding sites. To assess the accuracy of our method in finding real interactions, we then calculated two different statistics. First, the precision defined as the ratio between the true positives (TP; true drug-protein interactions found by nAnnoLyze) and the sum of TP and false positives (FP, a link between a drug and a protein not in the PDB). Second, the sensitivity (or recall) defined as the ratio of TP and the TP+ false negatives (FN, a link between a compound and protein not found by nAnnoLyze).
We have implemented a Web server where an end user can retrieve all pre-calculated predictions for the DrugBank and human protein as well as submit its own set of compounds. The server takes as input a compound ID and its SMILE in case of a new compound or only the DrugBank ID in case of a DrugBank drug. Then the user needs to select which organism proteome should be searched against. Currently nAnnoLyze has pre-calculated networks for the human and three Mycobacterium proteomes. The server search results in a list of all the predicted compound-protein pairs presented as a sortable table for easy filtering depending on the Global Z-score cut-off. A graphical enrichment of the Gene Ontology Terms [59] and KEGG pathways [60] of the predicted targets is also shown above the result table. Each prediction is further detailed by providing a GLMol based visualization (http://webglmol.sourceforge.jp) of the compound and the protein structure alongside the predicted binding site. All the structural data and all the predictions can be downloaded from the nAnnoLyze Web server at http://www.marciuslab.org/services/nAnnoLyze.
|
10.1371/journal.ppat.1007834 | ATP6V0d2 controls Leishmania parasitophorous vacuole biogenesis via cholesterol homeostasis | V-ATPases are part of the membrane components of pathogen-containing vacuoles, although their function in intracellular infection remains elusive. In addition to organelle acidification, V-ATPases are alternatively implicated in membrane fusion and anti-inflammatory functions controlled by ATP6V0d2, the d subunit variant of the V-ATPase complex. Therefore, we evaluated the role of ATP6V0d2 in the biogenesis of pathogen-containing vacuoles using ATP6V0d2 knock-down macrophages infected with the protozoan parasite Leishmania amazonensis. These parasites survive within IFNγ/LPS-activated inflammatory macrophages, multiplying in large/fusogenic parasitophorous vacuoles (PVs) and inducing ATP6V0d2 upregulation. ATP6V0d2 knock-down decreased macrophage cholesterol levels and inhibited PV enlargement without interfering with parasite multiplication. However, parasites required ATP6V0d2 to resist the influx of oxidized low-density lipoprotein (ox-LDL)-derived cholesterol, which restored PV enlargement in ATP6V0d2 knock-down macrophages by replenishing macrophage cholesterol pools. Thus, we reveal parasite-mediated subversion of host V-ATPase function toward cholesterol retention, which is required for establishing an inflammation-resistant intracellular parasite niche.
| V-ATPases control acidification and other processes at intracellular vesicles that bacteria and parasites exploit as compartments for replication and immune evasion. We report that the protozoan intracellular parasite Leishmania amazonensis resists inflammatory macrophage immune responses and upregulates an alternative isoform of subunit d of V-ATPase (ATP6V0d2). Leishmania are still sequestered within acidified parasitophorous vacuoles (PVs) in cells lacking ATP6V0d2, but these PVs do not enlarge in volume, a distinguishing feature of intracellular infection by these parasites. Cholesterol levels in ATP6V0d2-deficient cells are reduced and exogenous cholesterol repletion can restore vacuole size, leading to enhanced parasite killing. This study demonstrates the ATP6V0d2-mediated interplay of macrophage cholesterol retention and control of the biogenesis of large pathogen-containing vacuoles. The study provides grounds for the development of new therapeutic strategies for diseases caused by intracellular pathogens sheltered in host cell compartments.
| Vacuolar H+-ATPases (V-ATPases) are membrane-associated ATP-dependent multimeric enzymes responsible for pumping protons from the cytosol into the lumen of intracellular organelles, thus controlling the acidification of lysosomes, endosomes, the trans-Golgi network and other intracellular vesicles [1, 2]. V-ATPases display two functionally distinct domains composed of several subunits: the cytosolic domain V1, composed of eight subunits (A, B, C, D, E, F, G and H) and that is implicated in ATP hydrolysis, and membranal domain V0, which is composed of subunits a, d, e, c, c’, and c” and is implicated in proton transport across the vesicle membrane [1].
Acidification of intracellular compartments is the canonical function of V-ATPases, which are largely implicated in diverse cellular processes, such as maturation and degradation of proteins, receptor-mediated endocytosis, receptor recycling and endocytic traffic [3, 4]. At the crossroads of innate immunity and endocytosis, V-ATPases are responsible for phagolysosome acidification in macrophages and other professional phagocytes, a key feature in the immune response against intracellular pathogens [5]. Maintenance of an acidic pH controlled by V-ATPases is required for the optimal activity of lysosomal digestive enzymes and production of hydrogen peroxide and other reactive oxygen species directly involved in pathogen killing [6].
Pathogens have nevertheless evolved strategies to evade phagolysosome acidification and killing, including targeting and subverting V-ATPase functions, thus improving their adaptation inside the hostile environment of host cells [7]. The pathogen-mediated subversion of V-ATPases may involve the interference of one or several subunits that compose the two functional domains, inhibiting proton pump activity or driving V-ATPases to target different organelles. The bacterial pathogens Legionella pneumophila and Mycobacterium tuberculosis, for instance, have the ability to secrete virulence factors that directly target the H-subunit of the V1 domain of host cell V-ATPases, blocking the acidification of bacteria-containing vacuoles in which they multiply by V-ATPase exclusion [8–10]. Conversely, Yersinia pseudotuberculosis does not exclude V-ATPases from the bacteria-containing vacuole but decreases their activity during intracellular infection [11].
In addition to coupling with the V1 domain and its proton translocation canonical function, the V0 membrane domain interacts with Soluble NSF Attachment Protein Receptors (SNAREs), thus being implicated in membrane fusion and exocytosis [12, 13]. These noncanonical functions of V-ATPases can take place when V0 domains are dissociated from V1 and directed to different organelles or when V-ATPases are composed of alternative isoforms of some of their subunits [4, 14, 15], a feature that could be exploited by intracellular pathogens.
The a subunit from the V0 domain, for example, has four different isoforms, each one expressed in different specialized cell types and distinct organelles [16]. The d subunit, also from the V0 domain, is expressed either as a ubiquitous isoform d1, which is implicated in the regular proton pumping activity of V-ATPases, or as an alternative isoform d2 (ATP6V0d2), which is highly expressed in restricted tissues, such as bones, kidney and lungs [17], and specialized cell types, such as osteoclasts [18] and macrophages [19], where it acts as a membrane fusogen [20–22].
The isoform ATP6V0d2 is implicated in counteracting macrophage inflammatory responses [23, 24]; therefore, the pathogen-induced production of this subunit isoform may constitute a mechanism by which intracellular pathogens multiply in macrophages despite inflammatory stimuli. Accordingly, ATP6V0d2 is upregulated in macrophages upon in vitro intracellular infection with the protozoan parasite Leishmania (Leishmania) amazonensis [25]. Leishmania spp. are trypanosomatid parasites, which induce tegumentary or visceral leishmaniasis in humans and other animals, a major health problem in poor and developing countries [26]. They are dimorphic parasites found extracellularly in the midgut of insect vectors as flagellated and elongated promastigotes and intracellularly in mammalian host macrophages, neutrophils and dendritic cells as round-shaped amastigotes [27]. Species from the L. mexicana complex, such as L. amazonensis, L. mexicana and L. pifanoi, are known to multiply within large and fusogenic pathogen-containing vacuoles or parasitophorous vacuoles (PV) [28], which are acidic compartments displaying functional V-ATPases [29]. Compared to other species, they also display, at least in vitro, a remarkable resistance to parasite killing mechanisms mediated by interferon-γ (IFN-γ) and lipopolysaccharide (LPS) within macrophages or by direct treatment with reactive oxygen species (ROS) [30–32]. A causal relationship between large PV development and parasite resistance to inflammatory macrophages remains elusive especially in vivo.
Considering that ATP6V0d2 participates in both membrane fusion and anti-inflammatory processes, we evaluated the participation of this subunit isoform in the biogenesis of pathogen-containing vacuole formation. ATP6V0d2 participation in L. amazonensis resistance to inflammatory macrophages upon stimulation with IFN-γ/LPS or treatment with inflammatory, oxidized lipoproteins (ox-LDL) was also approached. Here, we demonstrate that ATP6V0d2 is upregulated by intracellular parasites as a countermeasure to macrophage inflammatory immune responses, controlling the volumetric expansion of the pathogen-containing vacuole by regulating macrophage intracellular cholesterol levels. ATP6V0d2 does not participate in parasite survival within inflammatory macrophages classically activated by IFN-γ/LPS. ATP6V0d2 is required, however, for parasite survival within macrophages that scavenge ox-LDL via parasite-mediated increased expression of LOX-1 and CD36 scavenger receptors.
The subunit d (ATP6V0d) connects the two functionally distinct subunit V-ATPase complexes V0 and V1, which are responsible for the acidification of intracellular compartments. The subunit d from V-ATPase V0 complex occurs as two variants, ATP6V0d1 (ubiquitous) and ATP6V0d2, which expression is restricted to certain tissues and cells, expressed in parallel with ATP6V0d1 variant [17, 21]. V-ATPases will be thus composed of either d1 or d2 variant filling the space for the d subunit of V0 complex. To evaluate the role of isoform d2 in this canonical function of V-ATPases, we stably knocked-down ATP6V0d2 in RAW 264.7 macrophages (ATP6V0d2-KD) and evaluated phagolysosomal acidification using fluorescein (FITC)-tagged latex beads ingested by the phagocytes [33, 34]. We have stably and specifically knocked down the d2 variant (ATP6V0d2), not the ubiquitous ATP6V0d1 variant which predominates over ATP6V0d2 on nonsilenced control macrophages (Fig 1A). The expression of another V-ATPase subunit, ATP6V0a1, remains unaltered upon ATP6V0d2 knock-down (Fig 1B), demonstrating that this and likely all other subunits compose a functional V-ATPase in ATP6V0d2-KD macrophages. After phagosomal pH measurements using FITC-tagged beads internalized by nonsilenced and ATP6V0d2-KD macrophages (S1 Fig), we observed that, although ATP6V0d2 is efficiently knocked-down (Fig 1A and 1B), phagolysosomes containing FITC-tagged beads reach an acidic pH of approximately 5.2 in both nonsilenced and ATP6V0d2-KD macrophages, activated or not by IFN-γ/LPS treatment (Fig 1C–1E). Thus, the knock-down of ATP6V0d2 does not interfere in V-ATPase canonical function of phagolysosomal acidification as corroborated by others using different methods [21, 24].
Despite demonstrating that ATP6V0d2 does not participate in the V-ATPase canonical function of phagolysosome acidification, ATP6V0d2-KD macrophages display impaired lysosomal functions as assessed by analysis of the activity of some lysosomal enzymes. Cathepsin D (CTSD), one of the most well-studied lysosomal enzymes whose activity is a direct indicator of lysosomal functions [35, 36], was more abundantly associated with lysosome-associated membrane protein 1 (LAMP-1)-positive compartments as assessed by fluorescence colocalization analysis (S2A Fig), although cleaved, “mature” functional forms of CTSD were absent in ATP6V0d2-KD (S2B Fig). The activity of enzymes involved in lysosomal storage diseases that could indicate lysosome impairment was also evaluated: lysosomal acid lipase (LAL), implicated in Wolman and cholesteryl ester storage diseases, displayed the same activity in both nonsilenced and ATP6V0d2-KD macrophages; activity of α-galactosidase (α-Gal), implicated in Fabry Disease, was increased in ATP6V0d2-KD macrophages, while β-glucocerebrosidase (GCase) activity, whose activity deficiency is observed in Gaucher Disease, was decreased compared to nonsilenced macrophages (S2C Fig). All tested enzymes are acid hydrolases only active at acidic pH; considering that LAL activity does not depend on ATP6V0d2, we excluded an impairment of lysosome acidification in the lysosome dysfunction displayed by ATP6V0d2-KD macrophages. Therefore, ATP6V0d2 does not participate in the canonical V-ATPase function of phagolysosome acidification, instead exerting a pH-independent regulation of lysosomal enzymatic functions.
To evaluate the participation of ATP6V0d2 in the innate immune response of macrophages, we assessed the expression of ATP6V0d2 mRNA transcripts (relative to expression of its alternative ubiquitous isoform ATP6V0d1), in nonsilenced and ATP6V0d2-KD macrophages (Fig 2A). Macrophages were activated or not by IFN-γ/LPS treatment and cultured with or without the intracellular parasite L. amazonensis (S3A Fig). In nonsilenced macrophages, expression of ATP6V0d2 was upregulated upon Leishmania infection. We reproduced the remarkable decrease of ATP6V0d2 expression upon classical activation with IFN-γ/LPS as demonstrated by others [24], to levels comparable to those obtained in ATP6V0d2-KD macrophages. ATP6V0d2 expression is partially recovered by Leishmania intracellular infection, suggesting that Leishmania stimulates the expression of ATP6V0d2 as a countermeasure to the macrophage immune response.
However, ATP6V0d2 is not directly implicated in the macrophage responses related to parasite intracellular multiplication, namely: i) production of nitric oxide (NO) inferred by expression of the inducible isoform of nitric oxide synthase (iNOS, NOS2), the main effector of innate immunity against intracellular pathogens [37]; and ii) expression of arginase, which is involved in polyamine synthesis and is exploited by pathogens to establish intracellular infection [38]. NOS2 expression was increased upon IFN-γ/LPS treatment in ATP6V0d2-KD as compared with nonsilenced macrophages, indicating that ATP6V0d2 buffers this activation pathway in non-infected macrophages (Fig 2B, first graph). In infected macrophages, however, NOS2 expression was equally decreased upon IFN-γ/LPS treatment in nonsilenced and ATP6V0d2-KD macrophages harboring Leishmania, indicating that other host factors induced by the parasite, such as arginase, are more determinant in downregulating iNOS expression. Since arginase expression was increased in macrophages hosting the parasite independently of ATP6V0d2 knock-down or macrophage activation with IFN-γ/LPS (Fig 2B, second graph), the previous data showing decreased NOS2 expression upon IFN-γ/LPS treatment may be related to this increased arginase expression due to the presence of Leishmania.
Multiplication of intracellular Leishmania was assessed by quantitative live imaging and microscopic counting (Fig 2C–2E). Cultures of macrophages infected with Leishmania were recorded by live imaging for 36 hours, and the numbers of macrophages per microscopic field and parasites per macrophage were quantified by image segmentation (Fig 2C). Independently of ATP6V0d2 knock-down, activation with IFN-γ/LPS inhibited RAW 264.7 cell proliferation (Fig 2D, upper graph) but increased Leishmania intracellular multiplication (Fig 2D, lower graph), as demonstrated by others upon IFN-γ-only treatment [30]. At the end of 72 hours after administration of parasites to macrophage cultures, samples were fixed, and the numbers of macrophages and parasites hosted per macrophage were converted into an infection index, which revealed that activation with IFN-γ/LPS increased parasite multiplication independently of ATP6V0d2 (Fig 2E).
Next, we evaluated L. amazonensis PV features, such as acidification and PV volumetric enlargement [28], in nonsilenced and ATP6V0d2-KD macrophages. Intracellular parasites are sequestered within acidified PVs independently of ATP6V0d2, as assessed by lysosomotropic probes retained in acidic compartments (Fig 3A). Complete abrogation of probe fluorescence of the L. amazonensis PV in macrophages treated with the alkalinizer agent ammonium chloride (NH4Cl) functionally confirmed the acidified content of PVs formed independently of ATP6V0d2. In addition, the trafficking of LAMP-1 to the L. amazonensis PV membrane, a distinguishing feature of lysosomes, phagolysosomes and Leishmania PVs [28], was not altered by ATP6V0d2 knock-down in control or IFN-γ/LPS-activated macrophages (Fig 3B). In addition, the frequency of L. amazonensis PVs displaying the late endosomal SNARE VAMP8 in their membranes is not altered by ATP6V0d2 knock-down (S4D Fig). Concerning PV morphology, however, L. amazonensis PV developed in ATP6V0d2-KD macrophages did not enlarge in size as compared with nonsilenced macrophages according to three-dimensional projections of images obtained from infected samples (Fig 3C and S4A Fig).
To further investigate this impairment in PV enlargement, ATP6V0d2-KD macrophages hosting L. amazonensis PV were dynamically tracked by live imaging (Fig 3D, S1 Movie). The parasite developed enlarging PVs in nonsilenced macrophages (Fig 3D, arrowheads, upper row); this was in contrast to ATP6V0d2-KD macrophages, in which PV dimensions are smaller and often fit parasite size, promoting PV fissions as the parasite multiplies (Fig 3D, arrowheads, lower row). Using fluorescent lysosomal probes and image segmentation analysis [28], we dynamically assessed PV volumetric enlargement in parasite-infected macrophages activated or not with IFN-γ/LPS, demonstrating that L. amazonensis PV enlargement depends on ATP6V0d2 (Fig 3E–3G). On average, infected nonsilenced and ATP6V0d2-KD macrophages do not differ in or change their cell sphericity over the course of 36 hours of multidimensional (S4B Fig) and, in contrast to PV area measurements, PV volumetric assessment is nevertheless not influenced by cell sphericity effects (S4B and S4C Fig). These results demonstrate the participation of ATP6V0d2 in controlling L. amazonensis PV volumetric expansion.
The biogenesis of large L. amazonensis PVs is accompanied by upregulation of host macrophage genes implicated in lipid metabolism, specifically cholesterol homeostasis [25], suggesting the participation of cholesterol in the intracellular establishment of this parasite. Therefore, we evaluated the intracellular levels of free cholesterol/cholesteryl esters in the studied macrophages, demonstrating that macrophages displayed a 40% decrease in cholesterol levels when ATP6V0d2 was knocked-down as detected by ELISA (Fig 4A, nontreated group) and confirmed by mass spectrometry (S5A Fig).
To functionally assess the participation of cholesterol in the ATP6V0d2-dependent biogenesis of L. amazonensis PVs, we envisioned a protocol for cholesterol repletion by adding oxidized low-density lipoprotein (ox-LDL) to macrophage cultures (S3B Fig), as performed previously [39–41]. Modified LDL, such as ox-LDL, is more efficiently taken up by macrophages through scavenger receptors and induces higher accumulation of intracellular cholesterol than native LDL [41, 42]. Among three different strategies to replenish macrophage intracellular cholesterol levels decreased in ATP6V0d2-KD–namely, treatment with methyl-β-cyclodextrin/cholesterol complexes [43], with LDL [41, 42] or with ox-LDL [39, 41]–ox-LDL was the most effective method to replenish intracellular cholesterol with less cytotoxicity in both nonsilenced and ATP6V0d2-KD macrophages (Fig 4A and S5B and S5C Fig).
Accumulation of ox-LDL-derived cholesterol in macrophages leads to the formation of foamy macrophages, which are full of lipid-laden vacuoles (lipid droplets) [44, 45] that could reconstitute L. amazonensis PV volumes in ATP6V0d2-KD macrophages. Accordingly, exogenous ox-LDL traffics into PVs independently of ATP6V0d2 (Fig 4B, arrowheads), and the ox-LDL-mediated intracellular cholesterol repletion in ATP6V0d2-KD macrophages hosting L. amazonensis increased the PV volume to dimensions comparable to those measured in nonsilenced macrophages (Fig 4C and S4C FIg). There is a negative correlation between PV size and the amount of ox-LDL accumulated within PVs, demonstrating that smaller PVs like those formed in ATP6V0d2-KD macrophages accumulate more ox-LDL (Fig 4D). Importantly, PVs formed in ATP6V0d2-KD macrophages—which recover their dimensions by ox-LDL treatment—retain more ox-LDL per μm3 as compared with PVs formed in nonsilenced macrophages (Fig 4E).
This ox-LDL-mediated PV dimensional recovery was accompanied by a decrease in the intracellular survival of L. amazonensis specifically within ATP6V0d2-KD macrophages, as assessed by comparing infection indexes under two different concentrations of ox-LDL (Fig 4F and 4G). Parasites hosted within PVs formed in ATP6V0d2-KD macrophages and enlarged after treatment with ox-LDL displayed aberrant morphology suggestive of parasite killing [46] in contrast to parasites multiplying in nonsilenced macrophages under the same ox-LDL treatment (Fig 4F and S2 Movie).
The ox-LDL-mediated PV size recovery observed in ATP6V0d2-KD macrophages is not related to differential expression of ATP6V0d subunit isoforms d1 and d2 (S6A Fig) or the differential expression of the lysosomal traffic regulator LYST/Beige (S6B Fig, right graph) involved in PV biogenesis [47]. In addition, the impaired intracellular establishment of L. amazonensis in ATP6V0d2-KD macrophages treated with ox-LDL was not due to increased production of reactive oxygen species [48] or inflammatory cytokines upon cellular uptake of ox-LDL [49] at the evaluated ox-LDL concentration (S6C Fig). Finally, the enzymatic activities of α-Gal and GCase lysosomal enzymes after ox-LDL-mediated cholesterol replenishment were assessed and do not explain neither the ox-LDL-mediated recovery of PV dimensions in ATP6V0d2-KD macrophages (compare infected macrophages treated or not with ox-LDL, S2D Fig).
The cholesterol intracellular homeostasis in macrophages can be regarded as a balance between cholesterol biosynthesis that generates cholesterol precursors involved in the cholesterol biosynthetic pathways, cholesterol catabolism, and cholesterol uptake/efflux promoted by receptors for non-modified LDL and scavenger receptors for modified LDL [50]. To approach the participation of ATP6V0d2 in cholesterol homeostasis, we have evaluated the mRNA levels of scavenger receptors and of the sterol regulatory element-binding protein 2 (SREBP2) which controls expression of genes involved in cholesterol synthesis [51], in the context of ATP6V0d2 knock-down, infection with Leishmania and treatment with ox-LDL. The non-altered mRNA expression of SREBP2 observed in the conditions studied (S6B Fig, left graph) and the non-altered abundance of the cholesterol biosynthetic precursors squalene and lanosterol observed by mass spectrometry comparing nonsilenced and ATP6V0d2-KD macrophages (S5A Fig) indicate that ATP6V0d2 does not associate with cholesterol biosynthesis.
An increased gene expression for LDL receptor (LDL-R) in ATP6V0d2-KD macrophages as compared with nonsilenced ones was observed independently of the conditions studied, with ox-LDL treatment decreasing the mRNA levels (Fig 5A, upper left graph). This is compatible with LDL-R stimulated expression upon lower intracellular cholesterol levels as displayed by ATP6V0d2-KD [52–54] and reinforces the role of ATP6V0d2 in the influx of cholesterol.
Considering the scavenger receptors for modified LDL, CD36 is decreased by ATP6V0d2 knock-down (Fig 5A upper right graph and 5B-C). RT-qPCR for CD36, covering the detection for all 5 isoforms of murine CD36, was the more efficient technique to detect these differences. The decrease of total (Fig 5B) and membrane surface (Fig 5C) CD36 levels was not so marked as the decrease observed in mRNA levels (Fig 5A). Recovery of PV dimensions by ox-LDL-mediated cholesterol replenishment in ATP6V0d2-KD occurs in parallel with increasing in CD36 gene expression specifically in infected ATP6V0d2-KD macrophages (Fig 5A and 5B, red arrowhead) in both mRNA and protein levels (Fig 5A upper right graph and 5B). Considering that the ox-LDL-mediated parasite killing occurs exclusively in ATP6V0d2-KD macrophages (parasites hosted by nonsilenced macrophages are resistant to ox-LDL intake) and that CD36 is known to control PV enlargement [55], we infer that CD36 participates in the recovery of PV dimensions upon ox-LDL uptake, what is detrimental to the parasite only in the absence of ATP6V0d2.
Other scavenger receptors implicated in ox-LDL intake display a non-altered expression in the conditions studied (Scavenger Receptor class A, Msr1/SRA, Fig 5A lower left graph) or display an increased expression specifically in infected ATP6V0d2-KD macrophages, although independent of ox-LDL treatment, such as the lectin-type oxidized LDL receptor 1, LOX-1 (Fig 5A, lower right graph). The membrane surface expression of scavenger receptors involved in cholesterol efflux, namely Scavenger receptor class B type 1 (SR-BI) and its alternative isoform SR-BII, was not altered by ATP6V0d2 knock-down (Fig 5D). Again, it reinforces the role of ATP6V0d2 in cholesterol intake in infected macrophages.
We report the participation of an alternative isoform of the V-ATPase subunit d, the isoform d2 (ATP6V0d2) in controlling the biogenesis of pathogen-containing vacuoles generated by L. amazonensis in macrophages. ATP6V0d2, whose expression is restricted to certain cell lineages, including macrophages, does not participate in phagolysosome acidification, indicating that the ubiquitous isoform d1 (ATP6V0d1) participates exclusively in the canonical function of this V-ATPase, while isoform d2 switches the V-ATPase toward noncanonical, acidification-independent functions, such as membrane fusion, regulation of lysosome enzymatic activities and downregulation of macrophage inflammatory burst [4, 21, 24, 56]. Therefore, the variant ATP6V0d1 is still expressed in ATP6V0d2 knock-down macrophages (ATP6V0d2-KD), capable of composing functional V-ATPases that acidify phagolysosomes and parasite-containing vacuoles. The preservation of phagolysosome acidification in the absence of the d2 variant demonstrated by us here and by others [21, 24] is a solid evidence that V-ATPases in ATP6V0d2-KD macrophages are functional and thus composed of all subunits required for their canonical functions.
ATP6V0d2 is involved in the function of important lysosomal enzymes, such as cathepsin D (CTSD), whose cleavage into mature forms depends on this V-ATPase subunit isoform. Inhibition of CTSD activity was demonstrated to either increase [57] or decrease [58] cholesterol intracellular levels depending on the studied models and a definitive participation of CTSD in cholesterol homeostasis remains to be established. Sphingolipid metabolism is also likely to be disturbed by ATP6V0d2 knock-down: β-glucocerebrosidase (GCase), whose activity is decreased in ATP6V0d2-KD macrophages and is responsible for breaking down glucosylceramide into ceramide [59], is also implicated in CTSD processing [60, 61], and α-galactosidase (α-Gal), whose activity is increased in ATP6V0d2-KD macrophages, participates in the production of glucosylceramide [62]. Hence, in addition to a 40% decrease in intracellular cholesterol levels, ATP6V0d2-KD macrophages could accumulate glucosylceramide (glucocerebroside) in detriment to ceramide and its incorporation into macrophage membranes. The data therefore indicate that ATP6V0d2 participates in lysosomal metabolic processes involved in the homeostasis of important membrane components, such as cholesterol and ceramide, which ultimately interfere in the biogenesis of pathogen-containing vacuoles in macrophages.
The regulation of lysosome function is coordinated by multiple factors, including proper assembly, trafficking and function of V-ATPases in the membrane of lysosomes and phagolysosomes. These lysosome-associated V-ATPase features could be controlled by ATP6V0d2 in macrophages reacting to pathogens and/or inflammatory stimuli. ATP6V0d2 is implicated in buffering inflammatory responses in macrophages, particularly upon TLR4 stimulation by LPS treatment [23]; however, the conclusion that this anti-inflammatory role of ATP6V0d2 is due to an ATP6V0d2-dependent vesicle acidification contrasts with our results and previous works showing that ATP6V0d2 depletion does not interfere in V-ATPase canonical functions such as ATP hydrolysis and H+ transport [21, 24] and that depletion of one particular subunit isoform does not interfere in V-ATPase-mediated phagosomal acidification, what would be compensated by expression with other variants (the case of subunit ATP6V0a3 [63]).
We demonstrated that ATP6V0d2 is upregulated by the parasite in IFN-γ/LPS-treated classically activated or M1-differentiated macrophages [64], e.g., macrophages that trigger an intra and extracellular inflammatory environment producing nitric oxide (NO) and reactive oxygen species (ROS), which is recognized as the most effective macrophage response against intracellular pathogens both in vitro and in vivo [30]. In contrast with Leishmania major parasites, which multiply in macrophages sheltered by tight-fitting pathogen-containing vacuoles and are sensitive to NO and ROS generated by classical macrophage activation, L. amazonensis and L. mexicana multiply within spacious and communal vacuoles and are resistant to M1 macrophage activation, that exerts cytostatic effects on intracellular L. amazonensis [28, 30, 31, 65, 66]. Conversely, our in vitro study demonstrated that macrophage stimulation with IFN-γ/LPS increased parasite multiplication independently of ATP6V0d2.
The persistence of this intracellular parasite despite inflammatory scenarios could be related to parasite-mediated counteraction of macrophage innate immune responses and microbicidal activities, e.g., by production of antioxidant enzymes to cope with oxidative burst [67] and establishment of a safe, customized intracellular niche where the parasite multiplies sheltered from ROS activity and antigen presentation [68, 69]. We reproduced the drastic downregulation of ATP6V0d2 expression upon LPS stimulation of macrophages as demonstrated by others [24], what is partially recovered by Leishmania infection. ATP6V0d2 is thus one of the several factors upregulated by the parasite in response to (or counteracting) the hostile environment of inflammatory macrophages. The ATP6V0d2-dependent volumetric expansion of pathogen-containing vacuoles may represent one additional countermeasure, possibly diluting phagolysosome hydrolases to concentrations innocuous to the parasite [70], thus favoring L. amazonensis multiplication. However, we observed that inhibition of PV volumetric enlargement by ATP6V0d2 knock-down did not interfere with parasite multiplication in either non-activated or IFN-γ/LPS-activated macrophages, suggesting that PV enlargement is not crucial for parasite intracellular multiplication and does not account for parasite persistence in NO-producing inflammatory macrophages, at least for a short 72-hour in vitro infection.
The ATP6V0d2-dependent PV expansion and parasite-mediated upregulation of ATP6V0d2 in IFN-γ/LPS-activated macrophages indicate that intracellular pathogens exploit ATP6V0d2 as a countermeasure to inflammatory scenarios. Although ATP6V0d2 does not participate in parasite resistance to the classical in vitro IFN-γ/LPS model of inflammatory macrophages, this V-ATPase subunit isoform was required for parasite survival in macrophages stimulated with ox-LDL, a potent inflammatory stimulus mainly studied in the context of atherosclerotic lesions but that has also been implicated in chronic psoriatic skin inflammation [71, 72].
Our results contrast with other mechanistic studies of L. amazonensis PV enlargement, which have established that interfering with the expression of host macrophage genes, such as the lysosomal traffic regulator LYST/Beige or some members of membrane fusion SNAREs machinery impact PV expansion and directly influence parasite multiplication [47, 73]. Parasite factors also account for this direct correlation between PV expansion and intracellular multiplication, as L. mexicana establishment in macrophages depends on Cysteine Peptidase B-mediated modulation of host cell membrane fusion machinery via the parasite GPI-anchored metalloprotease GP63 [73]. The observed PV impairments in these studies could be, however, the effect rather than the cause of parasite killing or inhibition of multiplication. We demonstrate that recruitment of late endosome-associated VAMP8 [74] to PVs and expression of LYST/Beige [47] are not associated with PV size impairments nor in the ox-LDL-mediated PV recovery observed in ATP6V0d2-KD macrophages. On the other hand, the main scavenger receptor for ox-LDL, CD36, was demonstrated to participate in the complex machinery that regulates PV biogenesis [55] and might be implicated in the ox-LDL-mediated PV dimensional recovery. The decreased CD36 expression in ATP6V0d2-KD macrophages together with increased LDL-R expression reinforce the central role of ATP6V0d2 gene on cholesterol intake and PV size. In addition, ATP6V0d2 knock-down, infection or ox-LDL treatment do not influence expression of SREBP2, which controls expression of genes involved in cholesterol synthesis [51]. Therefore, the ATP6V0d2-dependent PV biogenesis is unlikely to be related to cholesterol biosynthetic pathways but rather to cholesterol flux mechanisms. The similar expression of receptors involved in cholesterol efflux (SR-BI and SR-BII) in non-silenced and ATP6V0d2-KD macrophages, and the differences observed in the expression of receptors involved in cholesterol uptake strongly suggest that ATP6V0d2 participates in cholesterol influx.
While the precise molecular mechanisms controlling ox-LDL-mediated PV dimensional recovery and parasite killing working in cooperation with ATP6V0d2 remain to be elucidated, a model summarizing our results is presented in Fig 6. ATP6V0d2-KD macrophages displayed a 40% reduction in intracellular cholesterol levels, suggesting that the d2 subunit participates in cholesterol influx, which impacts the biogenesis of host cell membranes, including the formation of pathogen-containing vacuoles. Replenishment of ATP6V0d2-KD macrophage intracellular cholesterol levels with ox-LDL, modified LDL known to be more readily absorbed by macrophages compared with native LDL [42], partially reconstituted PV enlargement in parallel with parasite killing. The smaller the volume of PVs, the more ox-LDL is retained in these compartments, suggesting that as pathogen-containing vacuoles expand in volume, exogenous modified LDL internalized by macrophages are filtered out from or diluted within PVs. In this scenario, we speculate that, rather than induce an inflammatory cytokine microenvironment ultimately beneficial to the parasite [49, 75], the uptake of ox-LDL at the concentrations employed may induce the intracellular accumulation of oxygen radicals [76], oxidized phospholipids [77] and cholesterol crystals [71]. These compounds could access the parasites, and the potential anti-parasitic effects would be controlled by ATP6V0d2. The hypothesis that ATP6V0d2 induced by parasites during inflammation would, at the PV membrane level, restrict the access of LDL-derived components potentially toxic to intracellular parasites is in line with the demonstration that Leishmania does not have de novo cholesterol synthesis [78]. Furthermore, similar to other protozoan parasites, such as Toxoplasma gondii, Trypanosoma cruzi and Cryptosporidium parvum [79, 80], the parasite is able to salvage and incorporate host cell cholesterol through endocytosis of LDL [81, 82]. Importantly, L. mexicana is able to sequester host cell cholesterol directly from the large PV membrane built from exogenous LDL-derived components [83]. Therefore, PVs reconstituted in size by ox-LDL-mediated cholesterol influx in ATP6V0d2-KD macrophages (but not in nonsilenced macrophages) would be built up from ox-LDL-derived components potentially absorbed by the parasite, leading to parasite killing. ATP6V0d2 would participate in the selective features of Leishmania PV biogenesis, sparing the parasite from contacting and incorporating inflammation-derived toxic macrophage cargo.
This ATP6V0d2-mediated PV selectivity for ox-LDL-derived components could play an important role in vivo: Leishmania parasites developing large PVs are clinically associated with persistent diffuse granulomatous lesions in humans (diffuse cutaneous leishmaniasis), causing chronic damage to skin deep tissues despite only moderate inflammation in terms of NOS2 and IFN-γ expression compared to other disease manifestations [84, 85]. This context of persistent inflammation may favor the oxidative damage of proteins and lipids, resulting in oxidation and accumulation of modified LDL in tissues [48, 86], thus promoting an environment in which the ATP6V0d2-mediated selective PV biogenesis would account for Leishmania intracellular persistence. Therefore, ATP6V0d2 interference represents an unexplored therapeutic target for chronic diseases caused by inflammation-resistant intracellular pathogens.
Altogether, our results demonstrate that host macrophage V-ATPase functions can be subverted by the intracellular protozoan parasite L. amazonensis, thus establishing an intracellular niche in macrophages and allowing parasites to persist despite inflammatory environments.
All experiments involving animal work were conducted under the guidelines approved by the Committee on the Ethics of Animal Experiments of the Institutional Animal Care and Use Committee at the Federal University of Sao Paulo (CEUA/UNIFESP n° 3398150715) in accordance with the Guide for the Care and Use of Laboratory Animals of the Brazilian National Council of Animal Experimentation (http://www.cobea.org.br/).
Wild-type MHOM/BR/1973/M2269 or DsRed2-transfected MPRO/BR/72/M1841 L. (L.) amazonensis amastigote parasites were derived from BALB/c mice footpad lesions and were maintained and obtained as described [87].
RAW 264.7 cells (macrophage-like cells, BALB/c origin and donated by Prof. Michel Rabinovitch, EPM-UNIFESP, São Paulo) were cultivated in RPMI medium supplemented with 10% fetal bovine serum (FBS), 100 U/ml penicillin, 100 μg/ml streptomycin (complete medium) and were incubated at 37°C in a humidified air atmosphere containing 5% CO2. Macrophages were stably silenced for ATP6V0d2 using GIPZ Lentiviral shRNAi transduction following the manufacturer’s instructions (Dharmacon, Inc.). Efficient transduction was monitored by GFP reporter gene expression. From three oligonucleotides tested (V2LMM_88448, V2LMM_194889 and V2LMM_88451), oligonucleotide V2LMM_88451 yielded >90% of ATP6V0d2 silencing, thus providing the preferred model of ATP6V0d2 knock-down (ATP6V0d2-KD) macrophages. Nonsilenced macrophage controls are macrophages stably expressing the GFP reporter gene and a nonsilencing shRNA which is processed by the endogenous RNAi pathway but its processed siRNA will not target any mRNA in the mammalian genome. The nonsilencing shRNA sequence is verified to contain no homology to known mammalian genes. Nonsilenced or ATP6V0d2-KD macrophages were cultivated in complete medium supplemented with 10 μg/ml puromycin until intracellular infection experiments. ATP6V0d2 efficient knock-down was confirmed up to 72 hours of intracellular infection or up to 96 hours after puromycin removal.
L. amazonensis amastigotes were added to nonsilenced or ATP6V0d2-KD macrophages at a multiplicity of infection (MOI) of 20 parasites to 1 macrophage (20:1) for 6 hours of interaction at 34°C, 5% CO2. Macrophages were washed with phosphate-buffered saline (PBS) for the removal of non-internalized parasites, and complete medium was replenished without puromycin. Infected macrophages were maintained at 34°C, 5% CO2. The infection index was calculated 72 hours post-infection (p.i.) by multiplying the percentage of macrophages containing at least one parasite (% of infected macrophages) and the number of parasites per macrophage, as quantified after Giemsa counterstaining performed as described [88].
Macrophages were treated with 20 ng/ml interferon-γ (IFN-γ) (R&D Systems, Inc.) and 1 μg/ml lipopolysaccharide (LPS) (Sigma-Aldrich Inc.) overnight and washed out before adding parasites to the macrophage cultures. Macrophages were infected for 24 hours prior to treatment with human high-oxidized low-density lipoprotein (ox-LDL, Kalen Biomedical, LLC, USA) diluted in complete medium for an additional 48 hours. Macrophage cultures were then washed with PBS and either incubated for 30 minutes with 200 nM Lysotracker Red DND-99 Invitrogen probe (for assessment of the volume of parasitophorous vacuoles) or proceeded to Giemsa staining for assessment of infection index. When indicated, infected macrophages were incubated with 50 μg/ml of fluorescent Dil-ox-LDL (Invitrogen L34358) for 48 hours.
Images of paraformaldehyde 4%-fixed (PFA, Electron Microscopy Sciences) or live macrophage cultures infected with L. amazonensis were acquired with a Leica SP5 II Tandem Scanner System confocal unit (Leica Microsystems IR GmbH) coupled to a microincubator controlling the temperature and CO2 pressure conditions to 34°C, 5% CO2 (Tokai Hit Co., Japan). Fluorescence and Differential Interference Contrast (DIC) were acquired in the resonant scanning mode at 512 x 512 or 1024 x 1024 resolution using the 63× (HCX PL APO 63×/1.40–0.60 CS) or 100× (HCX PL APO 100×/1.44 CORR CS) immersion oil objectives, z-stacks between 0.5 to 0.8 μm and hybrid detectors enabled. During live imaging acquisitions, the lasers were adjusted to levels below 5% of laser power, and the duration of z-stacks was reduced to less than 30 seconds per recorded position to minimize phototoxicity. Images were processed by Imaris v.7.4.2 software (Bitplane AG, Andor Technology). Cells were stained for 15 minutes with Hoechst 33342 live cell nuclear dye (Thermo Fisher Scientific Inc.) as indicated.
Macrophages cultivated in ibiTreat-sterile tissue culture-treated HiQ4 multichamber dishes (ibidi GmbH) were infected with fluorescent L. amazonensis expressing DsRed2. These multichamber units allow for acquisition of four different experimental conditions at the same live imaging session, namely, infected nonsilenced or ATP6V0d2-KD macrophages activated or not with IFN-γ/LPS. Macrophage cultures were placed in the microincubator coupled to the confocal unit, and serial images of live, infected macrophages were acquired each 30 minutes during 36 hours in 8 microscopic fields per microchamber. A counting algorithm adapted from previous studies [28] was established using Imaris software as follows: i) isospots built based on parasite DsRed2 signals allowed for dynamic quantification of parasites per microscopic field during the acquisition period; ii) isosurfaces built based on macrophage GFP signals allowed for dynamic quantification of macrophages per microscopic field in the same acquisition period; iii) the ratio between these two variables per microscopic field provided the dynamic quantification of parasites per macrophages in infected cultures. The number of parasites in each analyzed macrophage was graphically represented by a color scale applied to each macrophage isosurface, ranging from cyan (no parasite) to magenta (>8 parasites per macrophage).
Macrophages cultivated in the HiQ4 multichamber dishes and infected with DsRed2-expressing L. amazonensis for 24 hours were incubated with 200 nM of Lysotracker Red DND-99 probe (Invitrogen) for a pulse of 30 min, washed and given fresh medium in the microincubator coupled to the confocal unit. The dynamic measurement of PV volumetric enlargement was performed as described [28], acquiring 10 microscopic fields per experimental condition. PV volumes in μm3 in each analyzed macrophage were graphically represented by a color scale applied to each PV isosurface, ranging from cyan (smaller) to magenta (larger PV). PV volume isosurfaces were also obtained from Dil-ox-LDL fluorescence for correlations between PV size and ox-LDL PV accumulation, using the same methodology. Similar to volume, cell sphericity is a measure obtained from three-dimensional image reconstructions assessed as described [87].
Macrophages cultivated on 13 mm circular coverslips were fixed with 4% PFA in PBS and blocked for 30 minutes with 0.25% gelatin, 0.1% NaN3 and 0.1% saponin PBS solution prior to 1-hour incubation with primary antibodies, including 1:2 (v/v) rat anti-LAMP-1 (Developmental Studies Hybridoma Bank 1D4B) or 1:1000 (v/v) anti-cathepsin D (Abcam ab75852). Next samples were treated for 1 hour with a 1:100 (v/v) solution of anti-rat or anti-rabbit AlexaFluor-568 secondary antibodies (Invitrogen). Samples processed for confocal microscopy were treated for 15 min with 10 μM 4’,6-diamindino-2-fenilindol hydrochloride (DAPI) to stain macrophages and parasite nuclei. The coverslips were mounted with Dako Fluorescent Mounting Medium (Dako) before image acquisition under the confocal unit. Zymosan (Zymosan A Z-4250, Sigma-Aldrich Inc.) were administrated to macrophage cultures for 6 hours (50 particles per macrophage) for generation of 48-hours phagolysosomes used as positive control for VAMP8+ phagosomes, immunostained as described[68].
Samples processed for flow cytometry analysis were centrifuged 300g at 4°C for 5 minutes and incubated with BALB/c mouse serum for 1 hour to block Fc receptors in MACS buffer (PBS pH 7.2, 0.05% BSA, 2 mM EDTA). Then, cells were fixed by adding 400 μl of 1% PFA in 100 μl of MACS buffer for 30 minutes, washed and incubated with primary antibodies anti-CD36 (cat 552544 BD) 1:40 (v/v), anti-SR-BI (bs-1186R Bioss) 1:50 (v/v) or anti-SR-BII conjugated with AlexaFluor-647 (bs-7545R Bioss) 1:100 (v/v) in MACS buffer for 1 hour at 4°C. Fluorescence-coupled secondary antibodies were incubated for additional 1 hour at 4°C and include biotin anti-mouse IgA (cat 556978 BD) 1:500 (v/v) plus streptavidin-APC (cat 17-4317-82 eBioscience) 1:500 (v/v) (for CD36 antibody) and anti-rabbit AlexaFluor-568 1:100 (v/v) (for SR-BI antibodies). Then, cells were washed, centrifuged and resuspended in MACS buffer for analysis on LSR Fortessa cytometer (BD Biosciences). Unstained cells and cells treated with secondary antibodies alone were used as controls.
Macrophage lysates were obtained by treating cultures with lysis buffer (Tris-HCl 50 mM pH 7.4, NaCl 150 mM, EDTA 1 mM, Triton X-100 1%) supplemented with a protease inhibitor cocktail (Halt Protease Inhibitor Cocktail, Thermo Fisher Scientific Inc.) at 4°C for 30 min and processed as described [88]. The membranes were blocked with TBS-Tween 0.1% buffer supplemented with 5% bovine serum albumin (BSA) for 1 hour. The primary antibodies rabbit anti-ATP6V0d2 (Sigma-Aldrich Inc. SAB2103220) 1:1000 (v/v), rabbit anti-ATP6V0a1 (Synaptic Systems cat 109 003) 1:1000 (v/v), rabbit anti-LAMP-1 (Cell Signaling 9091S) 1:1000 (v/v), mouse anti-CD36 (BD cat 552544) 1:1000 (v/v), mouse anti-β-actin (Cell Signaling 8H10D10 #3700) 1:5000 (v/v) and rabbit anti-cathepsin D (Abcam ab75852) 1:1000 (v/v) were incubated in TBS-Tween 0.1% supplemented with 5% bovine serum albumin overnight at 4°C. Anti-rabbit (A6154, Sigma-Aldrich Inc.) and anti-mouse (Sigma-Aldrich Inc. A4416) IgG peroxidase 1:8000 (v/v) secondary antibodies were incubated with 5% BSA in TBS-Tween 0.1% for 1 hour at room temperature. Biotin anti-mouse IgA (BD cat 556978) 1:8000 and Streptavidin-HRP (Southern Biotechnology Assoc. Inc cat 7100–05) 1:8000 (v/v) secondary antibodies were used to detect CD36 and incubated with 5% BSA in TBS-Tween 0.1% for 1 hour at room temperature. The membrane images were acquired using ECL Prime reagent (GE Healthcare Life Sciences) and analyzed on a UVITEC photodocumentator (Cleaver Scientific Ltd). Protein bands were quantified by densitometry using AlphaEaseFC software 3.2 beta version (Alpha Innotech Corporation, San Leandro, CA, USA), and the results are expressed in arbitrary units, which were calculated by integrating the intensity of each pixel over the spot area and normalizing to the gel background.
Macrophage messenger RNA (mRNA) was obtained and processed for quantitative RT-PCR as described [89]. The following primers for mouse sequences were employed in the RT-PCR analysis: Mus musculus ATPase, H+ transporting, lysosomal V0 subunit D2 (Atp6v0d2)—GenBank (access number: NM_175406.3), Forward: 5'- TGT GTC CCA TTC TTG AGT TTG AGG -3' and Reverse: 5'- AGG GTC TCC CTG TCT TCT TTG CTT -3'; subunit d1 (NM_013477.3), Forward: 5’-ATT GGC CAG GAA GTT GCC ATA AT-3’ and Reverse: 5’-GTC GTT CTT CCC GGA GCT CTA TTT-3’; Arginase 1 (NM_007482.3) Forward: 5′-AGC ACT GAG GAA AGC TGG TC- 3′ and Reverse: 5′-CAG ACC GTG GGT TCT TCA CA-3′; Nos2 (NM_010927.4) Forward: 5′- AGA GCC ACA GTC CTC TTT GC- 3′ and Reverse: 5′- GCT CCT CTT CCA AGG TGC TT- 3′; Lysosomal trafficking regulator (NM_010748.2) Forward: 5´- GCC TGG ATG AAG AAT TTG ATC TGG-3´and Reward: 5´- ATT AGT CCG AGA ACG GGA ATG ACA-3´; Sterol regulatory element binding factor 2 (Srebf2) (NM_033218.1) Forward: 5´- ACC AAG CAT GGA GAG GTA GAC ACC-3´ and Reverse: 5´- GGA AGA CAG GAA AGA GAG GGG AAG-3´; CD36 molecule (NM_001159558.1) Forward: 5´- GGC TAA ATG AGA CTG GGA CCA TTG-3´ and Reverse: 5´- AAC ATC ACC ACT CCA ATC CCA AGT-3´; Low density lipoprotein receptor (LDLR) (NM_010700.3) Forward: 5´- AAC CTG AAG AAT GTG GTG GCT CTC-3´and Reverse: 5´- CAT CAG GGC GCT GTA GAT CTT TTT-3´; Lectin-like oxidized low-density lipoprotein receptor-1 (Lox-1) (NM_138648.2) Forward: 5’- TCT TTG GGT GGC CAG TTA CTA CAA -3’ and Reverse: 5’-GCC CCT GGT CTT AAA GAA TTG AAA-3’; Scavenger receptor class A (SRA) (NM_031195.2)Forward: 5’- CTA CAG CAA AGC AAC AGG AGG ACA– 3’ and Reverse: 5’–TGC GCT TGT TCT TCT TTC ACA GAC- 3’. For all experiments, β-actin and HPRT were used as the endogenous gene. β-actin (NM_007393.5) Forward: 5´-GCC TTC CTT CTT GGG TAT GGA ATC-3´ and Reverse: 5´-ACG GAT GTC AAC GTC ACA CTT CAT -3´; HPRT (NM_013556.2) Forward: 5´-TCA GTC AAC GGG GGA CAT AAA AGT-3´and Reverse: 5´- ACC ATT TTG GGG CTG TAC TGC TTA-3´.
Gene expression analysis is in accordance with the MIQE guidelines [90]. We present results using two endogenous genes, i.e. β-actin and HPRT, showing that the profile of the results is similar using both endogenous genes (S7D Fig). The efficiency of all the primers used is shown as values of slope, R2 and percentage of efficiency (S7A and S7B Fig). The parameter between the curves of target and endogenous genes of a standard curve is used to calculate the amplification efficiency of the reaction, according to the equation: E = [10(-1 / slope)– 1] x 100. The standard curve is obtained by linear regression of the Ct amplification (cycle threshold) value on the log of the initial cDNA amount. An angular coefficient of the standard curve of -3.32 indicates a reaction with 100% efficiency. Reactions are considered efficient when amplification efficiencies of the target and endogenous gene are very close, with a tolerance of ± 10% of variation [91]. The specificity of the qPCR reaction was demonstrated by the melt curves of each gene (S7C Fig). The data were presented as a relative quantification and were calculated using 2−ΔΔCt [92].
To confirm acidification of L. amazonensis PVs, macrophages cultivated in HiQ4 multichamber dishes were infected for 24 h and then incubated for 20 minutes with 200 nM Lysotracker Red DND-99 or 100 μg/ml of Neutral Red dye before direct observation by confocal or bright-field microscopy, respectively. To test the specificity of the Lysotracker lysosomal probe for acidic pH, macrophages were treated with 10 mM ammonium chloride (NH4Cl) during probing.
To assess phagolysosomal pH, ATP6V0d2-KD or nonsilenced macrophages were cultivated in HiQ4 multichamber plates in the presence of FITC-coated latex beads (20 beads per macrophage) for 24 hours at 34°C, 5% CO2. Fluorescein fluorescence intensity decreases in direct correlation with acidic pH [33] and we have explored the differences in the excitation maximum of turboGFP (ex. max = 482 nm) and FITC (ex. max = 495 nm) to specifically detect fluorescence from FITC using Leica hybrid photodetectors (Leica HyD). When excited by a 496 nm laser (400 Hz frequency and 10% laser power), FITC is detected by Leica HyD 2.7 more efficiently then turboGFP using an emission range of 520–537 nm (S1A Fig), allowing us to adjust the voltage (gain) of photodetectors to threshold out turboGFP emission (S1B Fig). The raw acquired image of FITC beads are cleared from turboGFP fluorescence overlap (S1B Fig), and the fluorescence intensities per FITC-tagged bead are retrieved (in arbitrary units generated by Leica system, Fig 1C–1E). For each field, a z series of 18 images (steps) in resolution of 512 x 512 pixels and an average of 3 scans per line (line average) were established. FITC fluorescence intensity per bead was retrieved from bead isospots built using Imaris software as described [28]. This approach was applied to FITC-tagged beads internalized by GFP-expressing non-silenced and ATP6V0d2-KD macrophages incubated in complete medium adjusted to different pH ranging from 6.5–5 (buffered with 15 mM HEPES) to 4.5–3.0 (buffered with 30 mM citrate buffer) and in the presence of 10 μM of the ionophore nigericin (Sigma-Aldrich Inc.), which will rapidly equilibrate the pH within phagosomes with that of the extracellular medium [5]. A standard curve of pH measurement was then obtained using both non-silenced and ATP6V0d2-KD macrophages (Fig 1D), generating very similar functions positively correlating pH and the FITC fluorescence acquired that validate the method applied in this particular condition (i.e., FITC-tagged beads within GFP-expressing cells). The mean FITC fluorescence intensities retrieved in each experimental group were applied to the standard curve to obtain phagosomal pH.
α-galactosidase and β-glucocerebrosidase activities were determined as described [93, 94], with modifications. The determination of the activity of these enzymes is based on its action on the fluorogenic substrate 4-methylumbiliferiferone-D-galactopyranoside/4-methylumbiliferone-D-glucopyranoside (Sigma-Aldrich Inc.), resulting in release of the 4-methylumbiliferone molecule (4MU) and allowing for inference of the enzymatic activity in nmol per mg of protein per hour. Determination of the activity of the lysosomal acid lipase (LAL) enzyme in cells was performed as described [95], with modifications. For this, the fluorogenic substrate 4-methylumbiliferone palmitate (4MU palmitate, Santa Cruz Biotechnology) was used in the presence of an LAL activator, cardiolipin, and an inhibitor, Lalistat (Sigma-Aldrich Inc.), that allows quantification of the enzymatic activity in nmol per mg of protein per hour.
Replenishment of macrophage intracellular cholesterol levels was performed as previously described using methyl-β-cyclodextrin/cholesterol complexes [43], with LDL [41, 42] or ox-LDL [39, 41]. Methyl-β-cyclodextrin/cholesterol complexes were obtained by mixing 5 mM cholesterol (Sigma Aldrich C-8503) and 40 mM MβCD (Sigma Aldrich M-4555) in serum-free and non-antibiotic medium (macrophage-SFM 1X Gibco 12065–074). The solution was subjected to sonication for complete solubilization and incubated under shaking at 37°C overnight. Next, solution was filtered through a 0.45 μm filter and used in macrophage cultures. The concentrations of methyl-β-cyclodextrin/cholesterol complexes employed in the study refers to the 5mM cholesterol concentration used for composing the complexes. LDL was generously provided by Dr. Magnus Gidlund and Dr. Henrique Fonseca (University of São Paulo).
For intracellular cholesterol measurement, macrophages were lysed with lipid buffer (0.5 M potassium phosphate pH 7.4, 0.25 mM cholic acid and 0.5% Triton X-100) and sonicated in three high intensity cycles for 10 seconds [96], and cell lysates were then assessed for cholesterol levels by the Amplex Red Cholesterol Assay Kit (Thermo Fisher Scientific Inc.) according to the manufacturer's instructions. The results were normalized by the amount of protein obtained in lysates, as assessed by the Bradford method [97].
Total lipids were obtained from 2 x 107 macrophages as described [98]. Purification of sterols was performed in a 10 x 2.5 cm silica gel 60 column (Merck Millipore). Samples were prepared using 10 μL of the sterol fraction (resuspended in 100 μL of methanol for each 107 cells) in 2 ml acetonitrile:water (3:1 v/v) solution and infused with a syringe pump at flow rate of 30 μl/minute. The analyses were performed on a triple quadrupole instrument (model 310, Varian Inc./Agilent Technologies) with atmospheric pressure chemical ionization (APCI) source. The data were scanned in the range of 360–450 m/z. Nitrogen was used as nebulizer (275.8 KPa) and drying gas (68.9 KPa). Vaporization temperature was set at 300°C with the following conditions: capillary voltage set at 56 V, housing temperature set at 50°C, corona at 1μA and shield at 600 V. Sterol masses were retrieved from values of [M+H–H2O] and sterol abundance was assessed in non-saturated conditions. Data were acquired and analyzed with the Varian Workstation software version MS 6.9 and the amount of cholesterol and its precursors was assessed qualitatively comparing nonsilenced and ATP6V0d2-KD macrophages.
Nonsilenced and ATP6V0d2-KD macrophages cultivated in 96-well plates were treated or not with different concentrations of ox-LDL for 48 hours. Next, samples were cultivated in a solution of 1 mg/ml 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT, Sigma-Aldrich Inc.) for 2 hours in 37°C and 5% CO2. Macrophage supernatant chromogenic reaction was read at 540 nm in a micro ELISA reader (Multiskan MS–LabSystems, Finland). Cytotoxicity was assessed at cellular level by FACS using 1:1000 (v/v) of the viability dye eFluor780 (eBiosciences) following manufacturer instructions. For cell death positive control, macrophages were first fixed with 4% PFA for 15 minutes and then labeled with viability dye.
Nonsilenced and ATP6V0d2-KD macrophages were cultivated in 24-well plates in complete medium stimulated or not with IFN-γ/LPS or ox-LDL for 48 hours. Cell culture supernatants were collected and stored at -80°C until analysis. Nitric oxide concentrations from 25 μl of supernatants were assessed as described [99] using the chemoluminescence reader Nitric Oxide Analyzer (NOA 208i –Sievers). To determine cytokine concentrations, supernatants were loaded with the Milliplex Map Mouse Cytokine/Chemokine Magnetic Bead Panel and Milliplex Map TGFβ1 Single Plex Magnetic Bead Kit (MCYTOMAG-70K and TGFBMAG-64K-01, Merck Milipore), following the manufacturer’s instructions. Samples were then analyzed by Luminex MAGPIX System 40–072 (Merck Millipore). The NO and cytokine concentrations were normalized according to the macrophage protein lysate concentration, as assessed using the Bradford method.
The experiments were repeated independently at least twice using experimental replicates. The results were represented as the means with respective standard errors. Statistical tests were performed by SPSS software (IBM), considering normal (parametric tests) or nonnormal distributions (nonparametric tests), and significant differences were indicated by p values below 0.05. Data were normalized by nonsilenced or nontreated controls as indicated.
|
10.1371/journal.pntd.0004562 | Cyclical Patterns of Hand, Foot and Mouth Disease Caused by Enterovirus A71 in Malaysia | Enterovirus A71 (EV-A71) is an important emerging pathogen causing large epidemics of hand, foot and mouth disease (HFMD) in children. In Malaysia, since the first EV-A71 epidemic in 1997, recurrent cyclical epidemics have occurred every 2–3 years for reasons that remain unclear. We hypothesize that this cyclical pattern is due to changes in population immunity in children (measured as seroprevalence). Neutralizing antibody titers against EV-A71 were measured in 2,141 residual serum samples collected from children ≤12 years old between 1995 and 2012 to determine the seroprevalence of EV-A71. Reported national HFMD incidence was highest in children <2 years, and decreased with age; in support of this, EV-A71 seroprevalence was significantly associated with age, indicating greater susceptibility in younger children. EV-A71 epidemics are also characterized by peaks of increased genetic diversity, often with genotype changes. Cross-sectional time series analysis was used to model the association between EV-A71 epidemic periods and EV-A71 seroprevalence adjusting for age and climatic variables (temperature, rainfall, rain days and ultraviolet radiance). A 10% increase in absolute monthly EV-A71 seroprevalence was associated with a 45% higher odds of an epidemic (adjusted odds ratio, aOR1.45; 95% CI 1.24–1.69; P<0.001). Every 10% decrease in seroprevalence between preceding and current months was associated with a 16% higher odds of an epidemic (aOR = 1.16; CI 1.01–1.34 P<0.034). In summary, the 2–3 year cyclical pattern of EV-A71 epidemics in Malaysia is mainly due to the fall of population immunity accompanying the accumulation of susceptible children between epidemics. This study will impact the future planning, timing and target populations for vaccine programs.
| Enterovirus A71 (EV-A71) is a major cause of hand, foot, and mouth disease (HFMD) in children. Since the first outbreak in Malaysia in 1997, EV-A71 epidemics have occurred every 2–3 years, in 2000, 2003, 2006, 2008/2009, and 2012. As the reasons for this cyclical pattern are not known, we hypothesize that it is due to changes in population immunity in children. In this study, we measured the EV-A71 neutralizing antibody prevalence in serum collected from children ≤12 years old between 1995 and 2012, covering 18 years and 6 epidemics. HFMD incidence was highest in children <2 years, and seroprevalence increased with age, and was higher during epidemics compared to non-epidemic periods. Peaks in EV-A71 genetic diversity coincided with reported EV-A71 epidemics. Decreases in EV-A71 seroprevalence over time were significantly associated with subsequent epidemic periods. This suggests that epidemics lead to high levels of population seroprevalence; but during the 2–3 years between epidemics, the population of young children with no immunity is replenished and increases, making it more likely that a new epidemic will occur. This is the first study to show that the cyclical pattern of EV-A71 epidemics is associated with changes in EV-A71 seroprevalence.
| Hand, foot and mouth disease (HFMD) is a common childhood disease, characterized by vesicles on the hands and feet, and ulcers in the mouth. Enterovirus A71 (EV-A71) is one of the main causative agents of HFMD apart from coxsackieviruses (CV) A6, A10, and A16 [1–3]. EV-A71, which belongs to the genus Enterovirus of the family Picornaviridae, is a small, non-enveloped, positive-stranded RNA virus. Substantial genetic diversity is observed in EV-A71. EV-A71 can be divided into three major genotypes, A, B and C, based on a cut-off nucleotide divergence value of 17–22% [4]. Genotypes B and C can be further subdivided into subgenotypes B0-B5 and C1-C5 respectively based on cut-off nucleotide divergence of 10–14% [4,5].
In addition to self-limiting HFMD or herpangina, EV-A71 is also rarely associated with more severe neurological diseases such as encephalitis, meningitis, acute flaccid paralysis and neurogenic pulmonary edema [6,7]. Since its first isolation in California in 1969, numerous epidemics of EV-A71 have been reported, mainly in Asia, including Singapore [8,9], Malaysia [10,11], Taiwan [12,13] and mainland China [14,15].
In Malaysia, HFMD epidemics due to EV-A71 were first documented in Sarawak, East Malaysia and Peninsular Malaysia in 1997, with over 2600 children affected and 48 deaths [10,11,16]. EV-A71 epidemics with fatalities recurred in Peninsular Malaysia in late 2000 [10] and in late 2005 [11], followed by an epidemic in Sarawak in early 2006, with 6 deaths reported [10,11]. Further epidemics of EV-A71 occurred in 2008/09 and 2012 [17–19]. In Sarawak, a clear 3-year recurrent cyclical pattern has been shown, with EV-A71 epidemics occurring in 1997, 2000, 2003 and 2006 [1,20]. A cyclical pattern of EV-A71 epidemics occurring every 3–4 years has also been described in Japan [21,22] and Singapore [23].
Cyclical epidemics may be due to various factors, including changes in pathogen antigenicity, variations in host population immunity, and environmental drivers [24]. Shifts in genotype often accompany new epidemics [18,21], but it is unclear whether these antigenic changes are the cause of recurrent epidemics. Most EV-A71 studies showed presence of cross-protective immunity against other genotypes following infection with a given genotype and high concordance in neutralization titers between the same genotypes [25–29]. Hence, changes in herd immunity are likely to be more important. As EV-A71 disease mainly affects young children below the age of 5, cyclical patterns of epidemics could be due to the accumulation of a new generation of susceptible children every few years, enabling sustained transmission [29]. However, direct evidence for this is scanty, as most EV-A71 seroprevalence studies are point prevalence studies.
In this study, we used 2,141 serum samples collected from children over an 18-year period encompassing 6 epidemics, and determined the association between seroprevalence rates and cyclical patterns of reported EV-A71 epidemics in Kuala Lumpur, Malaysia. We hypothesized that falls in EV-A71 seroprevalence rates were associated with new epidemics.
Overall national incidence rates of notified HFMD from 2006 to 2012 were available from the Ministry of Health, Malaysia. However, as the statutory notification of HFMD came into enforcement only in October 2006, cases prior to this were underreported. The monthly numbers of HFMD cases for each of the 13 states and 2 federal territories were available only between 2008 and 2014. The case definition for reporting HFMD is a child with mouth/tongue ulcers and/or maculopapular rash/ vesicles on the palms and soles, with or without a history of fever. This data is syndromic, without laboratory confirmation of the viral agent. As diagnostic virology facilities are not widely accessible, there is scanty data on causative viral agents. Consequently, EV-A71 epidemic years, with limited laboratory confirmation, were obtained from published reports [10,11,16,19] and defined as 1997, 2000, 2003, 2006, 2008/2009, and 2012.
Population data for Kuala Lumpur and Malaysia was generated based on the 2010 Population and Housing Census performed by the Department of Statistics, Malaysia [30]. Monthly climatic data for Kuala Lumpur consisting of temperature (°C), rainfall (mm), number of rain days and ultraviolet radiance (MJm2) were provided by the Malaysian Meteorological Department.
Serum samples were randomly picked from archived residual sera collected for routine virology and bacteriology tests in the Diagnostic Virology Laboratory, University of Malaya Medical Center, in Kuala Lumpur, the capital of Malaysia. Samples from patients with suspected HFMD were excluded. A total of 1,769 sera from children aged between 1 to 12 years old, collected between 1995 and 2012, were tested for EV-A71 neutralizing antibodies. Between 52 and 200 samples were collected for each year, except for 2009, when only 30 samples from children were available. Samples were divided into 1–6 years (pre-school) and 7–12 years (primary school) age groups for most analyses. A further 372 serum samples from children <1 year were analyzed separately, as these may contain maternal antibodies.
The study was approved by the hospital’s Medical Ethics Committee (reference number 872.7) and the Medical Research and Ethics Committee of the Ministry of Health, Malaysia (reference number NMRR-12-1038-13816). Our institution does not require informed consent for retrospective studies of archived and anonymised samples.
The selected serum samples were heat-inactivated at 56°C for 30 minutes. The neutralizing titer of each serum was determined by a microneutralization assay as described previously [8], with modifications. Two-fold serial dilution of each serum sample was performed from 1:8 to 1:32. An aliquot of 90 μl of each dilution was mixed with 90 μl of 1000 tissue culture infective dose (TCID50) of EV-A71 strain UH1/PM/97 (GenBank accession number AM396587) from subgenotype B4. The serum-virus mixture was then incubated at 37°C for 2 hours in 5% CO2. Each serum dilution was transferred into a 96-well plate in triplicate. A suspension of 100 μl containing 1 x 104 rhabdomyosarcoma (RD, ATCC no. CRL-2061) cells was then added. Pooled positive sera of known titer were included in each assay as positive controls, using previously described criteria for reproducibility [8]. Wells containing diluted serum, virus alone, and uninfected RD cells were also included as controls. The plates were incubated at 37°C in 5% CO2 and examined for cytopathic effects (CPE) after 5 days. Neutralizing antibody titer was defined as the highest dilution that prevents the development of CPE in 50% of the inoculated cells. A sample was considered positive if the neutralizing titer was ≥1:8 [31,32]. Good cross-neutralization of serum against EV-A71 of different subgenotypes has been observed (S1 Table). Nevertheless, sera from children ≤3 years collected in 2013 were used to verify the concordance of neutralization titers between the UH1 strain and a clinical virus isolate from subgenotype B5 (GenBank accession number JN316092) isolated in 2006 (39 sera) and a clinical isolate from subgenotype C1 (GenBank accession number JN316071) cultured in 1997 (32 sera). These were the genotypes circulating in Malaysia between 1997–2012 [17]. High concordance in seropositive/seronegative status was obtained between UH1 and B5 virus (97%, 38/39 sera) and UH1 and C1 virus (81%, 26/32 sera); hence, this supports our use of the B4 virus alone for all the neutralization assays.
We used phylogenetic analysis and selective pressure to investigate the role of genetic diversity in the cyclical patterns of EV-A71 epidemics. EV-A71 VP1 gene sequences of Malaysian isolates were retrieved from GenBank and aligned with Geneious R6 (Biomatters Ltd, New Zealand). A total of 275 VP1 sequences reported from Malaysia between 1997 and 2012 were available for analysis (S2 Table). The best substitution model was determined using jModelTest v0.1.1 [33] as the general time reversible model with rate variation among sites (GTR+G). Phylogenetic trees were constructed using the Bayesian Markov Chain Monte Carlo method in BEAST 1.7.4 [34], run for 30 million iterations with a 10% burn-in. All runs reached convergence with estimated sample sizes of >200. The clock model was uncorrelated lognormal relaxed and the tree prior was coalescent GMRF Bayesian Skyride, allowing the generation of a plot of relative genetic diversity, which reflects the change in effective population size over time [35]. The maximum clade credibility tree was viewed using FigTree 1.4 [36]. Selective pressure analysis was performed using codon-based maximum likelihood methods implemented in the Datamonkey web server [37]. Amino acids were only selected when positively identified by the two different codon-based maximum likelihood methods, which were single likelihood ancestor counting and fixed effects likelihood.
We used cross-sectional time series analysis to determine the associations between epidemic periods and changes in seroprevalence. EV-A71 epidemic periods were categorized based on reported epidemic years obtained from limited laboratory confirmation and published reports [10,11,16,19]. We categorized time of our study into six distinct clusters: 1995–1997, 1998–2000, 2001–2003, 2004–2006, 2007–2009, and 2010–2012. A cluster of time consists of months before the epidemic and during the epidemic year. Epidemic periods were defined as the years 1997, 2000, 2003, 2006, 2008, 2009, and 2012. The seroprevalence rate was defined as the proportion of the samples tested which had a neutralization titer ≥1:8. The association of seroprevalence and epidemic months were modeled using generalized estimated equations population average models adjusted for confounders which are biologically plausible or have been previously described; these factors were age (1–6 years and 7–12 years), and climatic variables (monthly temperature, rainfall, rain days, and ultraviolet radiance). Interaction between seroprevalence, epidemic periods and age was also evaluated for possible heterogeneous effects or associations.
Geometric mean titers (GMT) were calculated by log-transforming the positive neutralization titers, using a value of 64 for titers >1:32. A two-sided type I error of 0.05 was used for statistical significance. Statistical analyses were performed using SPSS software version 22 (IBM SPSS Software, USA) and Stata version 12 (Stata Corp, College Station, Texas, USA), and graphs were drawn using GraphPad Prism 5 (GraphPad Software, USA).
Malaysia consists of Peninsular Malaysia, where most of the country's 16 states and federal territories are located, and East Malaysia, which consists of Sabah, Sarawak and Labuan. The monthly notified HFMD cases in each state and federal territory were available from 2008–2014 (Fig 1). The total annual HFMD cases in 2008–2014 were 15,564, 17,154, 13,394, 7,002, 34,519, 23,331 and 31,322 respectively. In 5 of the 7 years, HFMD cases increased around March and peaked around May-June. Sarawak had the highest number of HFMD cases nationwide, while Selangor reported the highest number of cases among the states in Peninsular Malaysia.
Only national overall HFMD incidence rates were available from 2006. Details of causative viruses are generally not available, as most HFMD cases are clinically diagnosed and diagnostic virology facilities are not widely accessible. However, published reports of laboratory-confirmed EV-A71 epidemic years [2,3,11] were in accordance with cyclical reported HFMD activity from the available surveillance data, showing that EV-A71 epidemics occurred in Malaysia every 2–3 years, in 1997, 2000, 2003, 2006, 2008/9, and 2012.
Age-specific incidence data was available only from 2011 to 2014 for the total HFMD cases nationwide and for the cases in Kuala Lumpur (Fig 2). The incidence rate of HFMD was the highest in those <2 years in both Kuala Lumpur (8.3, 43.7, 19.0 and 31.7 per 1000 population in 2011, 2012, 2013 and 2014, respectively) and Malaysia (4.8, 22.9, 17.9 and 20.6 per 1000 population in 2011, 2012, 2013 and 2014, respectively). The rates decreased with increasing age, with the 7–12 years age group having the lowest incidence rates; in Kuala Lumpur, rates were 0.07, 1.1, 0.4 and 1.2 per 1000 population in 2011, 2012, 2013 and 2014 respectively, and overall in Malaysia, rates were 0.2, 0.9, 0.4 and 0.7 per 1000 population in 2011, 2012, 2013 and 2014, respectively.
EV-A71 seroprevalence was higher in the primary school 7–12 years age group (71.6%, 95% CI 68.2–74.7%) compared to the preschool 1–6 years age group (52.8%, 95% CI 49.8–55.9%; P<0.001) overall, and in 16 out of the 18 years analyzed (significantly different in 3 years; S3 Table). The overall seroprevalence and GMT were significantly higher in epidemic years (seroprevalence 67.4%, 95% CI 63.8–70.9%; GMT 23.6, 95% CI 21.8–25.5) compared to non-epidemic years (seroprevalence 56.6%, 95% CI 53.6–59.5%; GMT 17.8, 95% CI 16.7–19.0; P<0.001) (Fig 3). During epidemic years, the seroprevalence of children aged 1–2 years (52.5%, 95% CI 44.8–60.0%), 3–5 years (66.1%, 95% CI 58.7–72.8%), and 6–9 years (75.4%, 95% CI 69.1–80.8%) were significantly higher compared to non-epidemic years (1–2 years old: 39.6%, 95% CI 33.9–45.5%; 3–5 years old: 51.6%, 95% CI 45.8–57.4%; and 6–9 years: 64.4%, 95% CI 59.1–69) (Fig 3A). GMT also rose significantly during epidemic years (3–5 years old: 23.3, 95% CI 19.8–27.3; 6–9 years: 26.7, 95% CI 23.4–30.4; and 10–12 years old: 28.0, 95% CI 23.5–33.4) compared to non-epidemic years (3–5 years old: 18.1, 95% CI 15.7–20.9; 6–9 years: 18.1, 95% CI 16.2–20.2; and 10–12 years old: 20.4, 95% CI 18.0–23.1) (Fig 3B).
This is consistent with the observed general trend of EV-A71 seroprevalence spiking during reported EV-A71 epidemic years, and seroprevalence falling between epidemics (Fig 4A). These results showed that younger children aged 1–6 years old had lower seroprevalence in non-epidemic years, indicating greater susceptibility, which may explain the higher HFMD incidence in this age group (Fig 2). The higher seropositive rates and GMT levels seen during epidemic years are likely to reflect recent infection. HFMD incidence in older children aged 7–12 years is considerably lower; thus, the higher GMT levels seen during epidemics are more likely to represent re-exposure to EV-A71 or milder infection resulting in under-reporting. Taken together, both the incidence rates and the seroprevalence data suggested that HFMD caused by EV-A71 affects susceptible children aged 1–12 years, and most frequently affects younger children aged 1–6 years.
Many subgenotypes were co-circulating during EV-A71 epidemics. Subgenotypes B3, B4, C1 and C2 were present during the 1997 epidemic, but only subgenotypes B4 and C1 continued to circulate till 2001 and 2003, respectively (Fig 4C). After 2003, subgenotype B5 became the sole genotype circulating in Malaysia. A Bayesian Skyride plot was used to estimate the evolutionary dynamics of EV-A71 in Malaysia over time (Fig 4B). Sharp, transient rises of genetic diversity were observed in the reported epidemic years 1997, 2000, 2003, 2006, 2008/2009, and 2012. The decline in the effective population seen after the 1997 epidemic may coincide with purifying selection against subgenotypes B3 and C2. The decline in the effective population observed after the 2000 and 2003 epidemics may indicate purifying selection against subgenotypes B4 and C1, respectively. After 2006, when only subgenotype B5 was circulating, interepidemic viral diversity showed overall decline punctuated by spikes during the epidemic years of 2008/2009 and 2012.
To further investigate the driving force of diversification in EV-A71, selective pressure on the Malaysian EV-A71 VP1 was examined. The mean dN/dS (ratio of nonsynonymous substitution rate to synonymous substitution rate) was 0.058, and evolution of EV-A71 was driven by strong purifying selection with over 62% of the analyzed codons under negative selection pressure. Two codons at positions 98 and 145 were under positive selective pressure, likely resulting in the emergence of new virus variants or lineage extinction. About 9.8% (27/275) of the sequences showed E98K, 3.6% (10/275) were E145Q, and 9.1% (25/275) were E145G and these could be observed in different genotypes across different EV-A71 epidemics (S2 Table).
These results showed that EV-A71 epidemics are characterized by peaks of increased genetic diversity, often with genotype changes. Evidence of strong negative selection and 2 codons with positive selection may explain the emergence of immune escape though the role in cyclical patterns of EV-A71 epidemics remains unclear.
Interaction between seroprevalence, age and epidemic periods was first evaluated in the model for possible heterogeneous effects in different strata of age groups. The association between a 10% increase in monthly seroprevalence and odds of an epidemic was 1.41 (95% CI 1.16–1.71) in those aged 1–6 years and 1.23 (95% CI 0.94–1.61) in those aged 7–12 years. The association between a 10% decrease between the preceding and current months and odds of an epidemic was 1.13 (95% CI 0.95–1.35) in those aged 1–6 years and 1.05 (95% CI 0.82–1.33) in those aged 7–12 years. The model incorporating the interaction terms of monthly seroprevalence with age groups and changes between seroprevalence of preceding and current months with age groups was tested, but showed non-significant interactions (P = 0.30). This suggests that age did not modify the association between the two measures of seroprevalence and epidemic period.
To further understand the relationship between recurrent EV-A71 epidemics and other factors such as seroprevalence, age and climatic variables, time series analysis was performed (Table 1). The monthly seroprevalence was positively associated with the odds of an epidemic period in the univariate analysis (OR for every 10% increase in seroprevalence, 1.40; 95% CI 1.18–1.65; P<0.001) and multivariate analysis after adjusting for plausible confounding factors such as age, temperature, rainfall, rain days, and ultraviolet radiance (adjusted OR for every 10% increase in seroprevalence, 1.45; 95% CI 1.24–1.69; P<0.001). This means that every 10% increase in monthly seroprevalence is associated with 45% higher odds of an epidemic, which is consistent with the observation that seroprevalence rates are higher during epidemics. We then examined whether relative changes in seroprevalence over time were associated with epidemics. Every 10% decrease in EV-A71 seroprevalence between preceding and current months was not significantly associated with epidemics in univariate analysis; but there was a significant association in multivariate analysis (aOR, 1.16; CI 1.01–1.35; P<0.034). This shows that every 10% fall in monthly seroprevalence compared to the preceding month is associated with 16% higher odds of an epidemic.
In Asia, recurring epidemics of HFMD with associated severe neurological disease is a major public health concern. In Malaysia, HFMD became a statutorily notifiable disease only from October 2006, although national surveillance data does not include the causative viral agents. A notable exception is Sarawak, the worst affected state in Malaysia, which established sentinel and laboratory-based surveillance of HFMD in 1998, and clearly showed recurrent EV-A71 epidemics coinciding with large spikes in HFMD rates occurring at 2–3 year intervals [3,38]. We have found that national HFMD rates, which were not virus-specific, accorded with EV-A71 seroprevalence, spikes in genetic diversity of EV-A71, and published reports of laboratory-confirmed epidemic years. Together, this showed that EV-A71 epidemics also occurred in similar 3 year cycles in Malaysia. We found clear support for our hypothesis, showing that statistically significant decreases in population seroprevalence (as a proxy for immunity) are temporally associated with subsequent epidemics, after adjustment for age, temperature, rainfall, rain days, and ultraviolet radiance.
We identified seropositive children from as early as 1995 and 1996, suggesting that EV-A71 was already circulating before the first documented epidemic in 1997. The presence of seropositive young children in interepidemic years shows that ongoing transmission occurs between epidemics. This is supported by laboratory reports of EV-A71 isolated in low numbers during interepidemic years [3,12,17,39]. Based on the HFMD monthly distribution from 2008–2014, a seasonal pattern was observed, with incidence peaking between May to June. In USA, HFMD epidemics occur during summer and autumn months [40]. Taiwan has also showed higher incidence in the summer months [41] and in Guangzhou, incidence peaked in April/May and September/October [42]. The location-specific factors leading to seasonal epidemics have not been clearly defined, but could include climatic factors, such as the association with relative humidity and mean temperature in Taiwan [43], which may affect environmental survival of enteroviruses. In the present study, the overall likelihood of an epidemic was influenced by temperature and rain days, but not rainfall or ultraviolet radiance. The effects of these climatic factors on virus survival and spread will require further investigation. The relationships of HFMD with climatic variables remain to be explored in detail in Malaysia, particularly as individual states may have widely varying weather.
The highest age-specific incidence of HFMD is seen in children <2 years old (Fig 2). This is consistent with the significant differences in age-specific EV-A71 seroprevalence seen between non-epidemic and epidemic years in those <2 years old, particularly in the <6 month (from 47.7% to 64.0%, p = 0.016) and 6 months to 1 year age groups (from 35.9% to 64.3%, p = 0.0016). If an EV-A71 vaccine, such as the inactivated vaccine that has recently shown promise in phase 3 trials [44], were introduced into routine immunization programs, children would have to be vaccinated at least by the age of 6 months, and possibly earlier [45,46]. As most children in Malaysia and other Asian countries [9,31] are seropositive by 5 years, an effective vaccine could prevent EV-A71 HFMD, as well as the severe associated neurological complications that mainly affect the very young [47].
The well-recognized cyclical pattern of EV-A71 epidemics seen in some countries has been attributed to the time taken for accumulation of enough susceptible children in the population. In Tokyo, the overall EV-A71 seroprevalence dropped to its lowest point in 6 years during the months just preceding an epidemic in 1973, including an absence of antibodies in children <4 years old [48]. In Guangdong, China, seroprevalence gradually dropped from 2007 to 2009, before a large epidemic in 2010 [49]. In Taiwan, there was evidence of fewer EV71 seroconversions in 1994–1997, before the 1998 epidemic [47]. Our study is unusual as it charts seroprevalence over a long period of time, covering 18 years and 6 epidemics, and we showed that changes in population immunity in children appear to be the major driving force of the observed cyclical epidemics. Specifically, we demonstrated that falls in seroprevalence were clearly associated with higher odds of a subsequent epidemic. Seroprevalence in both 1–6 and 7–12 years age groups increased in epidemic years, suggesting that both groups are involved in disease burden and transmission. The higher HFMD rates seen in children aged 1–6 years is most likely due to their greater susceptibility (as shown by their lower seroprevalence rates in non-epidemic years), but it may also be due to under-reporting in older children, who often have milder disease [38,50].
Estimation of the basic reproduction ratio (R0), or the number of secondary cases arising from an infectious case, has been widely used to study the dynamics of transmission of infectious diseases such as SARS and influenza [51]. The R0 of EV-A71 has been estimated as 5.48, which is considered as moderately infectious [52]. The EV-A71 R0 was higher than the estimated CV-A16 R0 of 2.5, suggesting that EV-A71 is more transmissible. For such a transmissible virus, the epidemic size is mainly dependent on the size of the susceptible population [53]. Following a viral epidemic, most of the population at risk would become immune. It may then take 2–3 years for the susceptible population to be replenished by newborns, and to be large enough for the R0 to increase to >1, hence leading to a cyclical pattern of EV-A71 epidemics every 2–3 years. A similar study should be conducted to determine the R0 to further understand EV-A71 transmission dynamics in Malaysia.
The present study also showed that Malaysian epidemics are characterized by peaks of increased genetic diversity, often with genotype changes. While the increased diversity may simply reflect a larger number of infections, we cannot exclude that new variants with antigenic changes may escape population immunity and contribute to cyclical epidemics. Although found in less than a quarter of Malaysian EV-A71, the positive selection pressure sites found at positions 98 and 145 of the VP1 protein have been previously reported [5]. These mutations appeared at the terminal branches with changes from E98K, E145Q and E145G. Amino acid position 98 is part of the BC loop and position 155 is part of the DE loop, both of which are immunogenic loops of VP1 [17]. A recent study measured cross-reactive neutralizing antibody titers against viruses with mutations at residues 98, 145 and 164 [54]. Up to 4-fold neutralization reduction was seen in sera from children, adults and rabbits tested against an EV-A71 VP1-98K/145Q/164E mutant, and all neutralization titers were ≥ 1:16. However, viruses with all three mutations concomitantly have not yet been seen in nature. The significance of the antigenic variation will require more detailed longitudinal serological studies. If immune escape is not needed or plays only a minor role to produce the cyclical pattern of EV-A71 epidemics, a significant accumulation of susceptible children between epidemics will be enough to support large-scale transmission and another epidemic.
Overall, in other published studies, EV71-infected children have detectable neutralizing antibody titers against all the EV71 genotypes [27], and cross-protective immunity between genotypes is generally considered to be high [28,55]. Previous studies in humans, monkeys, rabbits and mice showed that neutralization antibody levels against different genotypes may vary, but overall human anti-serum generally does cross-neutralize strains of different genotypes (S1 Table). Lower neutralization titers may not reflect antigenic shift sufficient to lead to immune escape. To date, no cases of recurrent EV-A71 infection have been reported, suggesting the presence of life-long protective immunity against EV-A71. While enteroviruses clearly undergo antigenic evolution, complete immunological escape in EV-A71 seems to be rare, thus EV-A71 is generally considered to be a single serotype antigenically. Any possible clinical significance and contribution of reduced cross-protective immunity towards new epidemics will require further confirmation.
Our study's findings may be a useful basis for future efforts to forecast EV-A71 HFMD epidemics in Malaysia. Occurrence of EV-A71 epidemics may be predicted by seroprevalence rates in children and influenced by temperature and number of rain days. The changing population immunity, the effects of climate variables on the survival and spread of EV-A71 in the environment, the change in virus genetic diversity, and changing probability of transmission of EV-A71 due to changes in host behavior under certain climatic conditions may explain the seasonal cyclical patterns. Time-series analysis of real-time, high-quality surveillance and seroprevalence data may provide efficient detection and effective forecasting of EV-A71 epidemics. Future research may also focus on the potential influence of other HFMD enteroviruses in the cyclical pattern of EV-A71 epidemics.
The main limitation of this study is that we used a convenience sample of residual diagnostic sera from a single hospital. However, it would be difficult to otherwise obtain such an extensive collection of serum samples from healthy children over many years. When compared to a random cluster survey, convenience sampling has also been shown to give similar estimates of seroprevalence to 5 vaccine-preventable viral diseases [56]. The convenience sample used here is likely to be appropriate for this study.
In conclusion, falls in seroprevalence in children aged 1–12 years old are the major driving force of the cyclical pattern of EV-A71 epidemics seen in Malaysia over 18 years. Nevertheless, possible interplay between seroprevalence with climatic variables and virus antigenic variations is evident and warrant future study. The highest age-specific incidence of disease, as shown by surveillance figures and seroprevalence rates, occurred in children <2 years. Together with the seasonal and cyclical patterns observed, this study has provided important data which will impact vaccine planning, timing and target populations for vaccine programs.
|
10.1371/journal.pntd.0001674 | Sodium Stibogluconate (SSG) & Paromomycin Combination Compared to SSG for Visceral Leishmaniasis in East Africa: A Randomised Controlled Trial | Alternative treatments for visceral leishmaniasis (VL) are required in East Africa. Paromomycin sulphate (PM) has been shown to be efficacious for VL treatment in India.
A multi-centre randomized-controlled trial (RCT) to compare efficacy and safety of PM (20 mg/kg/day for 21 days) and PM plus sodium stibogluconate (SSG) combination (PM, 15 mg/kg/day and SSG, 20 mg/kg/day for 17 days) with SSG (20 mg/kg/day for 30 days) for treatment of VL in East Africa. Patients aged 4–60 years with parasitologically confirmed VL were enrolled, excluding patients with contraindications. Primary and secondary efficacy outcomes were parasite clearance at 6-months follow-up and end of treatment, respectively. Safety was assessed mainly using adverse event (AE) data.
The PM versus SSG comparison enrolled 205 patients per arm with primary efficacy data available for 198 and 200 patients respectively. The SSG & PM versus SSG comparison enrolled 381 and 386 patients per arm respectively, with primary efficacy data available for 359 patients per arm. In Intention-to-Treat complete-case analyses, the efficacy of PM was significantly lower than SSG (84.3% versus 94.1%, difference = 9.7%, 95% confidence interval, CI: 3.6 to 15.7%, p = 0.002). The efficacy of SSG & PM was comparable to SSG (91.4% versus 93.9%, difference = 2.5%, 95% CI: −1.3 to 6.3%, p = 0.198). End of treatment efficacy results were very similar. There were no apparent differences in the safety profile of the three treatment regimens.
The 17 day SSG & PM combination treatment had a good safety profile and was similar in efficacy to the standard 30 day SSG treatment, suggesting suitability for VL treatment in East Africa.
www.clinicaltrials.gov NCT00255567
| Visceral leishmaniasis (VL) is a parasitic disease with about 500,000 new cases each year and is fatal if untreated. The current standard therapy involves long courses, has toxicity and there is evidence of increasing resistance. New and better treatment options are urgently needed. Recently, the antibiotic paromomycin (PM) was tested and registered in India to treat this disease, but the same dose of PM monotherapy evaluated and registered in India was not efficacious in Sudan. This article reports the results of a clinical trial to test the effectiveness of injectable PM either alone (in a higher dose) or in combination with sodium stibogluconate (SSG) against the standard SSG monotherapy treatment in four East African countries—Sudan, Kenya, Ethiopia and Uganda. The study showed that the combination of SSG &PM was as efficacious and safe as the standard SSG treatment, with the advantages of being cheaper and requiring only 17 days rather than 30 days of treatment. In March 2010, a WHO Expert Committee recommended the use of the SSG & PM combination as a first line treatment for VL in East Africa.
| The parasitic disease visceral leishmaniasis (VL) has an incidence of 500,000 new cases annually occurring primarily in India, Bangladesh, Nepal, Sudan, and Brazil and is fatal if untreated [1]. However, it is also an important disease in several other East African countries, with an incidence rate of 30,000 cases per year and a mortality rate of 4,000 deaths per year [2], [3].
VL treatment options in East Africa are primarily limited to the antimonial sodium stibogluconate (SSG), which is efficacious, but requires 4 weeks of hospitalization for daily intra-muscular injections and has been associated with serious adverse events such as cardiotoxicity; a concern in areas of HIV co-infection [3], [4], [5]. In India, leishmania parasites have developed resistance to SSG, with up to 65% of the population in the hyper endemic region of Bihar being unresponsive [6], [7]. SSG unresponsiveness is emerging in East Africa and treatment with a combination of SSG & PM may limit the spread of the emerging resistant strains of leishmania parasites [8].
The efficacy of paromomycin sulphate (PM) monotherapy for the treatment of VL has been demonstrated in India, where it is now registered for the treatment of VL [9] and the safety and efficacy of the combination of SSG and PM has been demonstrated in trials in India and a small Kenyan study [10], [11]. A large case series of 4,263 VL patients carried out by Médecins sans Frontières – Holland (MSF) in South Sudan showed that treating patients with a combination of SSG & PM for 17 days yielded better results than treating them with SSG alone: the initial cure rate was 97% for SSG & PM for 17 days versus 92% for SSG alone for 30 days [12].
For registration of PM and evaluation of the SSG & PM combination therapy throughout East Africa, efficacy and safety data were required from a Phase III trial. A multi-centre phase III trial has been conducted in six clinical trials sites in 4 East Africa countries (Ethiopia, Kenya, Sudan and Uganda). The trial started in 2004 with three arms; SSG monotherapy (20 mg/kg/day for 30 days: reference arm), PM monotherapy (15 mg/kg/day for 21 days) and SSG & PM combination (SSG: 20 mg/kg/day, PM: 15 mg/kg/day both given for 17 days).
The aim was to compare safety and efficacy of PM monotherapy and SSG & PM combination therapy with standard SSG treatment. An interim analysis showed that the PM monotherapy had an efficacy of <50% parasite clearance 6 months after the end of treatment in Sudan [13]. This arm was discontinued while a separate dose-finding trial of alternative PM regimens was conducted in Sudan [14]. The original Phase III trial was then restarted with a higher PM monotherapy dose (20 mg/kg/day for 21 days), while the other two arms remained unchanged.
The objectives remained the same; to compare the efficacy and safety of PM monotherapy and SSG & PM combination therapy to SSG. The results of this trial are reported here.
The trial was conducted in accordance with the Declaration of Helsinki (2002 version) relating to the conduct of research on human subjects and followed the International Committee on Harmonization guidelines for the conduct of clinical trials. All trial site personnel received training in Good Clinical Practice (GCP).
The relevant ethics committees from each country approved the study and the details are listed in the attached supporting text document. Patients and their legal guardians (if they were minors) provided signed informed consent prior to being randomized to the different treatment arms. GCP-trained monitors recruited from all four participating countries regularly monitored the trial at all sites.
An open label, parallel-arm multi-centre individually randomized controlled trial.
Patients were enrolled from six clinical trials sites: Médecins Sans Frontières (MSF) Holland treatment centre, Um el Kher, Gedaref State, Sudan; Ministry of Health Hospital Kassab, Gedaref State, Sudan; Gondar University Hospital, Amhara State, Northern Ethiopia; Arba Minch Hospital, SNNPR state, Southern Ethiopia; Centre for Clinical Research, Kenya Medical Research Institute (KEMRI), Nairobi, Kenya; and Amudat Hospital, Nakapiripirit Region, Uganda.
Inclusion and exclusion criteria have been described previously [13]. Briefly, patients aged 4–60 years with parasitologically confirmed VL were included, but patients with very severe VL or those with contraindications were excluded (Figures 1 and 2).
The three arms were SSG monotherapy (20 mg/kg/day for 30 days: reference arm), PM monotherapy (20 mg/kg/day for 21 days) and SSG & PM combination (SSG: 20 mg/kg/day, PM: 15 mg/kg/day both administered for 17 days).
Administration of PM (Gland Pharma, India) was intramuscular (IM). SSG (Albert David, India) was administered IM, or intravenous (IV) in Kenya. Patients requiring rescue medication were given liposomal amphotericin B, (manufactured as Ambisome®, Gilead, USA) according to national dosage guidelines of the participating countries. Patients were hospitalized for treatment and weekly monitoring of clinical and biological parameters. Follow-up visits were conducted 3 months and 6 months post end of treatment (Figures 1 and 2).
The primary efficacy endpoint was definitive cure, defined as parasite clearance from splenic, bone marrow or lymph node tissue aspirates 6 months after the end of treatment. Any patient who died from VL, received rescue medication during the trial, or had parasites detected at the 6-month assessment was considered a treatment failure. The secondary efficacy endpoint was parasite clearance from tissue aspirates at the end of treatment (SSG: day 31, PM: day 22, SSG & PM: day18). Treatment failure at the end of treatment was defined as death or receipt of rescue medication during initial hospitalization or presence of parasites at end of treatment necessitating rescue treatment. The presence of parasites at the end of the treatment, subsequently cleared without need for rescue treatment was considered a treatment success for primary outcome (definitive cure at 6 months follow-up), but a treatment failure for secondary outcome (cure at end of treatment). Slow responders were defined as patients with detectable parasites at end of treatment and parasite clearance at 6 months follow-up, without need for rescue treatment at any time. Parasitology was performed and reported according to an approved World Health Organization (WHO) method [1]. The numbers of parasites in slide fields were counted under oil emersion at 100× magnification and counts recorded.
Safety was evaluated based on the occurrence of adverse events (AE), laboratory parameters (haematology and biochemistry), electrocardiogram (ECG) readings, and audiometry. AEs were classified according to the Medical Dictionary for Regulatory Activities (MedDRA) version 10 [15]. A treatment emergent AE (TEAE) was defined as an AE with onset between the first day of treatment and 30 days after end of treatment.
ECGs were performed at all sites using a portable self-reporting ECG machine (Cardiofax, Model ECG 9620, Nihon Kohden) with patients resting supine on their beds. Trial physicians reviewed tracings and reported any abnormality.
Post-kala-azar dermal Leishmaniasis (PKDL) was recorded actively as an adverse event during patient follow-up or reported directly by the patients in between follow-up dates.
Audiometric testing was performed at all trial sites except Um el Kher using Voyager 522 Portable Diagnostic Audiometer (Madsen, Taastrup, Denmark). In recruitment period 1, investigators reported audiometric data as normal, clinically insignificant or clinically significant [13]. In period 2, hearing levels were recorded in detail for each ear at six frequencies. The following definitions were used to measure abnormalities; 1) disabling hearing impairment (DHI): an average hearing level, over frequencies 500, 1000, 2000, 4000 Hz, of ≥31 dB in both ears for those <15 years and ≥41 dB for those aged ≥15 years; 2) audiometric shift: a change in hearing level from baseline of ≥25 dB at ≥1 threshold frequency or ≥20 dB at ≥2 adjacent threshold frequencies.
All patients were offered counselling and HIV testing in accordance to national guidelines at screening.
The trial was designed to have 90% power (β = 0.1) to detect, at the 5% significance level (α = 0.05), an absolute difference in efficacy of 15% between PM and SSG and 10% between SSG & PM and SSG regimens [16]. An 85% efficacy was assumed in the reference arm and adjusting for 10% HIV co-infection and 10% loss to follow-up at 6 months post end of treatment, it was estimated that 404 and 195 patients per arm were required for the respective comparisons.
Being HIV-positive was not an exclusion criteria but the original protocol stated that there was to be a sufficient number of patients for a subgroup analysis excluding HIV patients (if deemed necessary).
As described at the end of the Introduction, recruitment and randomisation was carried out during two periods. In the first period, patients were randomised to SSG or SSG & PM combination arms, as part of a randomisation into three arms. Data from the third arm, a lower dosage regimen of PM found to be ineffective are not included here. In the second period, randomisation continued into one of three arms; SSG, SSG & PM arms as per period 1 and a PM monotherapy arm at a higher dosage regimen than previously (see Introduction and Interventions sections.)
In recruitment period 2 (using the higher 20 mg/kg dose of PM), randomization into 3 arms was continued until the desired sample size was reached for the PM versus SSG comparison. Randomization was then continued into one of two arms (SSG or SSG & PM) until reaching the sample size for the SSG versus SSG & PM comparison. Um el Kher site participated in period 1 only and Amudat site in period 2 only (during the two-arm randomization).
A computer-generated randomization list was produced with stratification by centre and block sizes of 15 until recruitment in the PM arm was completed, and block sizes of 10 thereafter. Allocation was concealed using opaque, sequentially numbered sealed envelopes. The randomization list and envelopes were prepared and stored securely at the LEAP Data Centre, based at the trial co-ordination centre in Nairobi.
Blinding of patients and investigators was not possible due to the different treatment durations and additional placebo injections were considered inappropriate.
Data were double-entered and validated in Epi-Info. Bespoke query generation programs were developed using Stata software, version 11 [17]. All statistical analyses were performed using Stata. Baseline data were summarized using mean and standard deviation (SD) or proportions where appropriate. Nutritional status was classified as normal, underweight, or severely underweight according to WHO Child Growth Standards in those <19 years and body mass index (BMI) in those ≥20 years [18].
For the SSG vs. PM comparison, patient data from randomisation during period 2 are included in this comparison. For the SSG vs SSG & PM comparison, patient data from randomisation into these arms in periods 1 and 2 are included in this comparison.
Efficacy data were analysed according to Intention-to-Treat (ITT) and Per-Protocol (PP). The PP population excluded those with pre-specified major protocol deviations (i.e. consent withdrawal after taking a dose of study medication, receipt of under 70% or over 130% of the expected treatment dosage, or receipt of alternative treatment to that of random allocation). Missing efficacy data were handled in two ways for each analysis population; complete-case analysis, where patients with missing data were excluded and worst-case analysis, where missing outcomes were considered treatment failures.
Efficacy is measured as the percentage of patients cured per arm. The treatment effect is the difference in efficacy between each test treatment (PM or SSG & PM) and the reference (SSG). Unadjusted treatment effects were calculated with exact binomial 95% confidence intervals (CI). Adjusted treatment effects were obtained using generalized linear models with a binomial distribution and identity link function. To assess possible effects of centre, age group (<18 years and ≥18 years) and recruitment period on efficacy after accounting for treatment allocation, regression models including treatment but with and without the covariate of interest were compared using the likelihood ratio test (LRT).
Treatment emergent adverse event (TEAE) rates were calculated as the number of TEAE, divided by the person-days at risk for each arm, and comparisons made using rate ratios. The treatment emergent period was defined as between day 1 of treatment and 30 days after the pre-defined treatment period, inclusive, therefore person-time at risk was as follows; SSG arm: 60 days, PM: 51 days, SSG & PM: 47 days. An adverse drug reaction was defined where an investigator recorded a probable, possible or unlikely relationship between the AE and study drug for VL.
The study was initiated in November 2004 and was completed in January 2010. A total of 2862 patients were screened for entry into the trial. Of these, 1755 were excluded (Figures 1 and 2), mainly due to negative parasitology. For the PM monotherapy versus SSG comparison, 205 patients per arm were recruited during period 2 (Figure 1). The total sample size for the SSG versus SSG & PM comparison was 386 patients in the SSG arm and 381 for SSG & PM (Figure 2): 135 patients per arm from period 1; 251 and 246 per arm respectively, from period 2.
Treatment arms were balanced for both comparisons with respect to demographic characteristics, vital signs, and physical measurements (combined arm data shown in Table 1). There were more male than female patients and more than 65% of patients were under the age of 18 years. All biological data except for nutritional status were balanced between arms at baseline; more patients in the PM and SSG & PM arms were classified as severely underweight but, overall combined percentages of underweight and severely underweight were balanced by arm. Overall, for all recruited patients, the HIV co-infection frequency was 1.4% (95% CI: 0.8–2.4%).
In the population analysed for the SSG versus PM comparison (n = 205 per arm), one patient in each arm did not receive the correct treatment allocation (Figure 1). Two patients in the PM arm withdrew consent after 4 and 6 days of treatment. For the SSG versus SSG & PM analysis population, three (0.8%) patients in the SSG arm and eight (2.0%) in the SSG & PM arm received a partial or incorrect dose (Figure 2). One SSG & PM patient withdrew consent after 6 days on treatment. Patients who had their 6-month follow-up at or before 4.5 months after the end of treatment were considered lost to follow-up since these visits were too early to assess definitive cure. For the SSG versus PM comparisons, outcome data for one (0.5%) patient in the SSG arm and two (1.0%) in the PM arm were considered missing. For the SSG versus SSG & PM comparison, outcome data for 13 (6.5%) SSG patients and nine (2.5% )SSG & PM patients were treated as missing data based on early follow-up. Data for patients whose primary endpoint assessment was later than 6 months were included in the analysis.
Efficacy in the SSG (reference) arm was 94% at 6 months after the end of treatment and 84% in the PM arm, according to the ITT complete-case population. All pre-specified primary endpoint analyses (ITT complete-case and worst-case, PP complete-case and worst-case) suggest that the efficacy of PM monotherapy was significantly lower than SSG - up to 17% less efficacious (Table 2). There were negligible differences in estimates of treatment effect and corresponding 95% CIs in these four pre-specified analyses. After adjustment for arm, efficacy did not differ between adults (≥18 years) and children (p>0.4 for both ITT and PP complete-case analyses).
There were 8 (4.0%) slow responders of the 198 ITT complete-case PM patient population at 6 months after the end of treatment and none in the SSG arm. Secondary endpoint treatment effects measured at the end of treatment were again very similar to 6 months primary endpoint data (Table 2).
In ITT complete-case primary endpoint analyses, the efficacy of SSG was 94% and for SSG & PM, 91% (Table 3). No difference in efficacy was noted between treatments. After adjusting for arm, no additional differences in efficacy were found between centres, age groups or recruitment periods (p>0.1, Table 3).
Worst-case analyses in the ITT and PP populations did suggest some additional variation by centre, age group and period after accounting for arm; due to some imbalance in losses to follow-up by age group and centre. However, treatment effects and corresponding 95% CIs were very similar in all four pre-specified primary endpoint analyses (Table 3).
In the SSG arm, 3 (0.8%) of 359 ITT complete-case analysis patients were slow responders, compared to 7 (1.9%) of the 359 SSG & PM patients. End of treatment secondary endpoint efficacy data were in agreement with primary endpoint data (Table 3).
The proportion of patients with SAE and non-serious TEAEs was similar in comparisons of both test treatment regimens to SSG (Table 4). Approximately 3% of patients in each arm in each comparison experienced an SAE deemed to be an adverse drug reaction (Table 4). One death occurred during the treatment period in each arm in the SSG versus PM comparison. In the SSG & PM versus SSG comparison, there were 3 deaths during initial hospitalization and a death of unknown cause during follow-up in the SSG arm. In the SSG & PM arm, there was a treatment period death and an unrelated death during follow-up (Tables 4 and 5). Of the 5 cases of renal impairment, 3 led to death, whilst 2 resolved after some time. Patients were withdrawn from treatment in all cases. Important cardiac events occurred in two patients: one in the SSG-PM arm and one in the SSG arm. In the former, a long QT interval appeared on Day 7, leading to treatment withdrawal. The long QT interval resolved 3 days later. In the second case, the patient died due to cardiotoxicity on Day 11 of treatment.
Rates and rate ratios, adjusted for centre, in both comparisons show no difference in safety based on analysis of TEAEs; adjusted rate ratio between the SSG and PM arm: 1.13, (95% CI: 0.93 to 1.38, p = 0.225) and between the SSG and Combination arms: 1.01, (95% CI: 0.88 to 1.17, p = 0.993). All of the non-fatal SAEs in the SSG and Combination arms resolved by the 6-month follow-up and all except one (pulmonary tuberculosis) in the PM arm resolved by the 6-month follow-up.
Treatment emergent adverse drug reactions (TEADRs) occurring in ≥10% of patients in the PM arm were injection site pain (13.2%), increase in aspartate aminotransferase (10.7%), and epistaxis (13.2%). In the subset of SSG patients analysed in the SSG versus PM comparison, TEADRs occurring in ≥10% of patients were aspartate aminotransferase increases (10.2%) and epistaxis (11.2%). For the population in the SSG versus Combination arms, no TEADR occurred in ≥10% of patients in the larger group of SSG patients. In the Combination arm, the most common TEADRs were injection site pain (17.3%) and increases in aspartate aminotransferase (10.5%).
Two patients in the Combination arm and one in the SSG arm had abnormal ECG findings that were considered clinically significant at end of treatment. These were, respectively, QT-wave inversion in V1–V4, arrhythmia and QT interval prolongation, which had normalized by 6 months follow-up.
In the SSG vs. PM comparison, 26 (12.7%) out of 205 patients developed PKDL in the SSG arm and 18 (8.9%) out of 203 patients randomised to PM. In the SSG vs SSG & PM comparison, 48 (12.4%) out of 386 patients in the SSG group and 23 (6.1%) out of 380 patients in the SSG-PM group developed PKDL. Two patients were given SSG for PKDL during their three months follow-up visit. DHI was reported in one patient in the PM and one patient in the Combination arm at the end of treatment, both of which resolved by the 6-month follow-up. None of the patients in the SSG arm had DHI. Thirty-six patients had audiometric shift at end of treatment (11 patients in the SSG arm, nine in the PM arm, and 16 in the SSG & PM arm). Audiometric shifts had still not resolved at the 6-month follow-up in three of the SSG, four of the PM and eight of the Combination patients.
This phase III GCP-compliant RCT investigated the safety and efficacy of PM both as monotherapy (20 mg/kg/day for 21 days) and as short course treatment in combination with SSG (PM at 15 mg/kg/day and SSG at 20 mg/kg/day for 17 days) for VL treatment in four East African countries, with the ultimate goal of determining if the SSG & PM combination treatment has acceptable safety and efficacy profiles to support its introduction in the region.
Definitive cure at six months follow-up in patients treated with SSG or SSG & PM was comparable with greater than 90% efficacy, despite PM monotherapy having significantly lower efficacy (84% cured) compared to SSG. Efficacy of the 20 mg/kg/day PM monotherapy at the 33% higher dose used in this study was better than that of the 15 mg/kg/day dose used earlier [13] (6-month cure rate of 84% vs. 64%), and is consistent with the dose-finding study conducted by the authors in Sudan [14]. However, the efficacy at this higher dose was still lower than that of SSG alone. By contrast, studies performed in India had shown that the efficacy of PM was consistently >90% at 15 and 20 mg/kg day for 21 days [19], [20], with PM showing better efficacy than SSG (20 mg/kg/day for 30 days) in the Jha et al. study [19] and non-inferior efficacy compared with amphotericin (1 mg/kg/day every 2 days for 30 days) in the Sundar et al. study [20]. Pharmacological differences in the East African and Indian populations that may explain these results were explored and will be reported separately. Geographical variation in efficacy of PM seen for the lower daily dose (15 mg/kg) was not apparent in this study with the higher daily dosage (20 mg/kg), though it must be noted that sufficient numbers of patients were not enrolled at all sites to perform a by-site analysis.
Secondary endpoints were performed at different times for each of the treatments (day 18 for the combination, day 22 for PM and day 31 for SSG), assumed comparable by design but potentially leading to bias in clinical and parasitological evaluations. Similarly, lack of blinding also may have led to bias in reporting, especially once lack of PM efficacy at the 15 mg/kg dose was suspected. As numerous sites and countries were involved, differentiation of reporting, particularly of adverse events was possible. Nonetheless, using a standard primary endpoint at 6 months and an objective measurement of efficacy based on parasitology, high rates of follow up were achieved. This is reflected in the relatively robust and comparable findings of the ITT, per protocol, complete case and worst case analyses.
The trial was powered to evaluate efficacy at the primary endpoint of 6 months follow-up and had limited power to detect differences in safety outcomes. However, almost identical rates of TEAEs and proportions of patients with adverse drug reactions were observed in patients treated with each regimen in the trial. The study was not powered to perform a subgroup analysis in HIV-positive patients assuming a 10% co-infection rate and HIV positive patients were not excluded. HIV co-infection was lower than expected, which may be due to the relatively small number of patients enrolled in Northern Ethiopia, where up to 35% co-infection had previously been reported [21]. In this study, 3 out of 5 and 5 out of 9 HIV co-infected patients had parasite clearance at 6 months after treatment with SSG and SSG & PM respectively. It was not possible to conclude on the difference in toxicity of either treatment among HIV co-infected patients.
Almost all of the SAEs that emerged in the three arms during treatment had resolved by the 6-month follow-up. There was no evidence of any new or important safety events, in either the PM or Combination arm. Although slightly more audiometric shifts remained at the 6-month follow in the PM and SSG & PM arms compared with the SSG arm, the trial was not powered to test for differences. With a larger sample size, percentages of patients with shifts remaining may have been balanced. Although not statistically significant, three deaths in the SSG arm were considered to be treatment-related (cardiotoxicity and renal disorders), whereas there were no treatment-related deaths in the Combination arm.
These results, together with those of a retrospective comparison of a 17 day regimen of SSG & PM versus 30 days of SSG alone carried out among 4,263 primary VL patients in South Sudan [12] support the use of a shorter course Combination therapy for VL in East Africa, which would be consistent with the long-term goal of reducing reliance on SSG monotherapy.
The reduced duration of treatment with the Combination compared with SSG (17 versus 30 days) will also reduce burden on hospitals and patients and other associated costs. The cost of drugs alone compares favourably for the Combination in comparison to SSG (44US$ versus 55.8 US$ respectively for a patient weighing 35 kg) [1]. Finally, the potential risk of development of parasite resistance to the treatment could be reduced.
In conclusion, our results show that SSG & PM combination treatment has comparable efficacy and safety profiles to conventional SSG monotherapy in a Phase III setting, and support its introduction for treatment of primary VL in East Africa.
|
10.1371/journal.pcbi.1002676 | Speeded Reaching Movements around Invisible Obstacles | We analyze the problem of obstacle avoidance from a Bayesian decision-theoretic perspective using an experimental task in which reaches around a virtual obstacle were made toward targets on an upright monitor. Subjects received monetary rewards for touching the target and incurred losses for accidentally touching the intervening obstacle. The locations of target-obstacle pairs within the workspace were varied from trial to trial. We compared human performance to that of a Bayesian ideal movement planner (who chooses motor strategies maximizing expected gain) using the Dominance Test employed in Hudson et al. (2007). The ideal movement planner suffers from the same sources of noise as the human, but selects movement plans that maximize expected gain in the presence of that noise. We find good agreement between the predictions of the model and actual performance in most but not all experimental conditions.
| In everyday, cluttered environments, moving to reach or grasp an object can result in unintended collisions with other objects along the path of movement. Depending on what we run into (a priceless Ming vase, a crotchety colleague) we can suffer serious monetary or social consequences. It makes sense to choose movement trajectories that trade off the value of reaching a goal against the consequences of unintended collisions along the way. In the research described here, subjects made speeded movements to touch targets while avoiding obstacles placed along the natural reach trajectory. There were explicit monetary rewards for hitting the target and explicit monetary costs for accidentally hitting the intervening obstacle. We varied the cost and location of the obstacle across conditions. The task was to earn as large a monetary bonus as possible, which required that reaches curve around obstacles only to the extent justified by the location and cost of the obstacle. We compared human performance in this task to that of a Bayesian movement planner who maximized expected gain on each trial. In most conditions, but not all, movement strategies were close to optimal.
| Imagine that you are sitting at your desk with a nice, hot cup of coffee in front of you and your laptop keyboard roughly behind it. In reaching out to hit the return key, you plan a trajectory that takes into account the possibility that you might jostle the cup and spill your coffee – that is, you plan a movement trajectory that you would not pick if there were no coffee cup in the way. Whatever trajectory you pick, however, will typically deviate from the one that you planned due to noise/uncertainty in the neuro-motor system. This noise has two important consequences: a risk of inadvertently spilling your coffee, and a risk of missing the key altogether. Your choice of plan involves a tradeoff between the costs and rewards associated with the possible outcomes of your planned movement.
The motor system, in planning any speeded movement, is selecting a stochastic “bundle” of possible trajectories [1], [2] and the particular bundle chosen determines the probabilities of favorable and unfavorable outcomes. There is no basis for selecting one planned trajectory as “best” without knowing the consequences of these different outcomes. If you are reaching to prevent your laptop from deleting your morning's work, you may be quite willing to put your coffee in peril and clean up later. In this article, we consider the problem of obstacle avoidance within the framework of Bayesian decision theory.
In this first investigation of obstacle avoidance within the framework of Bayesian decision theory, we translate the above example to one where there is an explicit reward for touching targets and an explicit cost for inadvertently intersecting intervening obstacles. We examine human obstacle-avoidance reach trajectories relative to the benchmark performance of an optimal Bayesian reach planner that chooses motor strategies to maximize expected gain as described next.
The experimental task illustrated in Figure 1 contains many of the elements of our coffee-cup example, and is reminiscent of the kind of obstacle avoidance behavior that has been studied extensively both in terms of its neurophysiological substrates [3], [4], [5] and in identifying sensory/motor factors that influence the movement trajectory [6], [7], [8], [9], [10], [11], [12], [13]. We will describe it in detail in the next section.
To study obstacle-avoidance reaches within the framework of Bayesian decision theory, we translated the above example to one where there is an explicit reward () associated with touching a target and an explicit cost () associated with inadvertently intersecting an obstacle that is placed between the starting point of the hand and the target. Contact with a physical obstacle placed along the reach path might change the physical character of the reach and such an obstacle would constitute an intrinsic cost whose value we could not easily measure or manipulate. To avoid these issues, we used virtual obstacles that could not impede the reach.
Although the virtual obstacle is invisible, a visual indication of its leftmost edge (at ) is presented on the monitor prior to each reach. Figure 1A shows a front view of the experimental apparatus with the virtual obstacle shown in transparent blue. The blue line on the monitor marks its edge (at ). The subject incurs the cost if the fingertip passes through the virtual obstacle while reaching toward the target (centered on , with width ). One part of training will allow subjects to become familiar with the location of the obstacle in depth and how its edge relates to the visual marker (the blue line). Across experimental conditions we varied the location of the obstacle and target and the cost incurred by passing through the obstacle as described in the next section. In all conditions there was a constant relative distance between the obstacle edge and the center of the target . Figure 1B show the same setup but from an overhead viewpoint. The left and right panels differ in the location of the obstacle-target pair.
Seven naive subjects participated in the experiment. Subjects were paid for their time ($10/hr.) and also received a bonus based on points earned during the experiment that amounted to $.01 per point (an additional $5–$10 over the hourly rate). All participants provided informed consent and research protocols were approved by the local Institutional Review Board.
Subjects were seated in a dimly lit room 42.5 cm away from a fronto-parallel transparent polycarbonate screen mounted flush to the front of a 21″ computer monitor (Sony Multiscan G500, 1920×1440 pixels, 60 Hz). Reach trajectories were recorded using a Northern Digital Optotrak 3D motion capture system with two three-camera heads located above-left and above-right of the subject. Subjects wore a ring over the distal joint of the right index finger. A small (0.75×7 cm) wing, bent 20 deg at the center, was attached to the ring. Three infrared emitting diodes (IREDs) were attached to each half of the wing, the 3D locations of which were tracked by the Optotrak system. Further details of the apparatus are given in a recent report [16]. The experiment was run using the Psychophysics Toolbox software [17], [18] and the Northern Digital API (for controlling the Optotrak) on a Pentium III Dell Precision workstation.
Subjects attempted to touch targets on a computer screen, represented visually as a vertical [6.5 mm×15 cm] strip, whose locations were chosen randomly and uniformly from a set of three locations [0, 38, 75 mm] relative to the monitor center. Rewards and penalties were specified in terms of points. Hits on the target earned subjects two points, and passing through the obstacle incurred a cost of one, two or five points. Missing the target earned no points, and too-slow reaches incurred a cost of ten points.
All reaches. All trials proceeded as follows: subjects brought their right index finger to a fixed starting position at the front edge of the table (15 cm to the right of screen center), triggering the start of the trial. Next, the target (and obstacle) was displayed (Figure 1A), followed 50 ms later by a brief tone indicating that subjects could begin their reach when ready. Movement onset was defined as the moment the fingertip crossed a frontal plane 3 mm in front of the table edge, itself located 35 cm from the screen; the fingertip was required to reach the screen within 600 ms of movement onset. Both the fingertip endpoint, the location where the fingertip passed through the plane of the obstacle (during obstacle practice and experimental reaches) and a running total of points (during experimental reaches) were displayed on-screen at reach completion.
Before each experimental session, subjects (fitted with IREDs) touched their right index finger (pointing finger) to a metal calibration nub located to the right of the screen while the Optotrak recorded the locations of the six IREDs on the finger 150 times. Linear transformations converting a least-squares fit of the three vectors derived from the 3 IREDs on each wing (left and right; each defining a coordinate frame) into the fingertip location at the metal nub were computed.
During each reach we recorded the 3D positions of all IREDs at 200 Hz and converted them into fingertip location using this transformation. The 3 IREDs on the left and right wings were used to obtain fingertip location independently, and the two estimates were averaged when all IRED locations were available for analysis. This redundancy allowed data to be obtained even if IREDs on one wing or the other were occluded during some portion of a reach.
Because we cannot predict the biomechanical costs associated with reach speed and overall length of reach trajectory that might accompany the longer and faster reaches necessary to reach targets within the timeout interval for, e.g., midline vs. right-of-midline target locations, we restrict the cost function that must be minimized by an optimal reach planner to the target and obstacle costs defined by and . Thus, the only factors entering into the optimal reach plan are fingertip positional uncertainty (i.e., the standard deviation of fingertip position in the relevant plane), average fingertip coordinates at the two critical planes, , and target and obstacle costs. To compute optimal reach plans, we model the empirical relationship between mean excursion, , and the remaining kinematic variables, the two sample standard deviations, and at the obstacle and target planes, respectively. The relationships were close to linear and we thus fit three lines relating empirical fingertip standard deviation to mean excursion separately for each of the three obstacle positions because we allowed for the possibility that fingertip standard deviations will change differently for excursions around nearby and further-away obstacles. Similarly we fit three lines relating empirical fingertip standard deviation to mean excursion. These six lines allowed us to predict and as a function of any planned excursion . While it is plausible that we could develop a single equation to predict each of the standard deviations, and by incorporating the obstacle location itself we could only do so at the cost of additional assumptions; the equations we use are sufficient for our purposes.
After having obtained a function relating excursion size and fingertip uncertainty (at both the target and obstacle planes, for all three obstacle positions), it is possible to predict fingertip standard deviations for theoretical excursions () not observed experimentally, around any of our obstacles. This in turn allows one to compute the expected gain associated with any theoretical excursion. Maximizing the expected gain function yields the prediction of the optimal reach planner (i.e., the theoretical excursion maximizing expected gain, ) in each of the 9 conditions of the experiment.
In the previous section we outline our method of predicting the obstacle avoidance behavior of an optimal Bayesian reach planner based on modeled changes in uncertainty, both at the obstacle plane and the target plane, of making reaches that deviate from their natural unobstructed trajectory. Because we parameterize the expected gain function in terms of obstacle-plane excursion, we can test the hypothesis that data conform to the predictions of the optimal Bayesian reach planner by comparing predicted and observed () obstacle-plane excursions. Data conforming to the Bayesian (optimal planning) model will fall along the identity line of a plot showing observed vs. predicted excursions.
Notice that we manipulated value to get the range of data needed to predict the standard deviations and given the planned excursion , and we then use these equations to predict the optimal excursion for each condition. The reader may be concerned that there is an apparent circularity in our use of the Dominance Test. The circularity is only apparent, not actual; This is because, no matter how well the empirical fits (relating planned excursion to standard deviations and ) fit the data, there is no guarantee that the average excursion () observed in a particular condition, of all possible excursions, will produce the largest possible gain; i.e., that it happened to fall at the theoretical MEG excursion () for that condition. Suppose, for example, that the subject consistently chose excursions that are 80% of the way between the edge of the obstacle and the theoretical MEG excursion (). While the observer has failed to maximize expected gain in every condition, the fits relating planned excursions to standard deviations and will be little affected. We refer the reader to the second experiment of Hudson et al. [15], which used a similar Dominance Test and demonstrated such a patterned failure.
We compare performance to that predicted by the optimal planning model using standard Bayesian model comparison techniques (see Supplemental Text S1). This analysis yields a measure of evidence [19] (given in decibels) for the optimal planning model (or conversely, against non-optimal planning models), based on the odds ratio comparing the probability of the optimal planning model given the observed data and the probability of any of the non-optimal models on the same data. For example, evidence of between 3 and 4.75 dB (or odds of between 2: and 3∶1 favoring one model over the alternative[s]) is usually considered a lower bound for statistically significant evidence [see e.g.], [ 15], [16], [19], [20,21].
Several features of the data can be observed directly in the value diagrams (Figure 3). First, higher costs lead subjects to avoid the obstacle region by a greater margin: there is an increasing deviation between obstacle-plane crossing points and the obstacle edge as magnitudes increase, across all targets. However, this change in crossing-point is not accompanied by within-target changes in average target-relative endpoints: no matter how great an excursion the finger took around the obstacle, the location of the distribution of target endpoints was unchanged. This relation of endpoint error with target position alone (i.e., independent of excursion) allowed us to model as the average endpoint error in each condition (), regardless of excursion size. In addition, covariance ellipses consistently increase in size as magnitudes increase (within each target location). These four functions, relating changes in positions and standard deviations to magnitude, are plotted in Figure 4.
One can also see a slight positive correlation (“counterclockwise tilt”) in value diagram covariance ellipses (Figure 3). That is, a rightward deviation from the mean in the obstacle plane tends to be paired with a rightward deviation in the target plane. This correlation implies that there is a component of the trial-to-trial variation in trajectories that affects the entire reach, and is therefore detectable at both obstacle and target planes. This tendency is quite small, however, and is ignored in our modeling.
We have developed a simple empirical model of the relationship between horizontal excursion within the obstacle plane and horizontal variance. While the model allows us to predict optimal behavior, we make no claims regarding the factors affecting horizontal variance.
Our study was not designed to determine the origins of positional uncertainty, a separate and intriguing question. There are very likely many factors that contribute separately to sensory and motor uncertainty and we implicitly assume that those factors (in our task, direction of gaze, body posture, etc.) are selected by the visuo-motor system so as to provide the best possible tradeoffs between hitting the target and avoiding the obstacles.
To compute optimal reach plans based on the data available in the value diagrams, we re-organize the plots in Figure 4 to predict target- and obstacle-plane fingertip positional uncertainty as functions of the observed obstacle-relative fingertip excursion (Figures 5A and 5B, respectively). Fitting straight-line functions to these data by linear regression (i.e., a line was fit to the data from each obstacle condition separately; R2 ranged from 0.8 to 0.99), we can predict target- and obstacle-plane uncertainties at unobserved fingertip excursions. By varying the theoretical planned excursion (), we compute the expected gain (Equations 1–3) at the obstacle plane (), the target plane () and overall, predicted as a function of any possible (i.e., non-positive) planned obstacle-plane excursion for each obstacle location and magnitude. An illustration of the computation is given in Figure 5C, corresponding to the middle target location and the middle obstacle cost. The maximum of the expected gain curve as a function of theoretical excursion, , corresponds to the excursion in the obstacle plane that maximizes expected gain, denoted .
The mean observed excursion across subjects is plotted versus the excursion maximizing expected gain in Figure 5D. The confidence intervals are 95% confidence intervals across subjects. An optimal reach planner would produce data along the identity line of this plot. Overall, the Bayesian evidence measure we computed is 12.99 dB (about 20∶1 odds) favoring the hypothesis that data do, in fact, fall along the identity line. However, there are deviations when the predicted MEG excursion () is large in magnitude (leftmost point in Figure 5D) where the mean observed shift is almost a factor of two smaller than the predicted shift. While human performance for smaller excursions is not far from optimal, there is a clear failure of optimality for the largest predicted excursion. Subjects passed too close to the obstacle in following their trajectory to the target.
The optimal reach planning model described here assumes that the distribution on is stationary (does not change across time). We considered the possibility that subjects might employ a within-block “hill-climbing” strategy designed to discover the MEG excursion by initially making too-large excursions around the obstacle and reducing their size over the following few reaches until an appropriate point was found. We verified that this was not the case in the Supplement (Supplemental Figure S2). There, we show that the distribution of excursions does not vary appreciably over the course of each block. To further investigate the possibility of similar cognitive strategies, we computed autocorrelations for each subject and block up to lag 15. No significant autocorrelations were found, suggesting that cognitive “contamination” was not present in our results.
We developed a model of obstacle avoidance within the framework of Bayesian decision theory and tested that model experimentally. We considered the possibility that reach trajectories around an obstacle can be explained quantitatively by a reach planner that minimizes the overall negative effect of an intervening obstacle. Such a reach planner would optimize the trade-off that increases excursion extent to reduce the expected cost of contacting the obstacle, but also decreases excursion extents so that the probability of contacting the eventual target is not drastically reduced.
This work represents a different approach to the problem than is traditionally taken: We are not attempting to determine how specific elements of the display determine changes in the details of the obstacle-avoidance reach or affect the possible covariance structures at the two points along the trajectory of interest. The Bayesian decision-theoretic approach [22], [23], [24], [25], [26] allows us to model and consider a wider range of tasks, of which simply hitting the target or avoiding the obstacle are at the extremes of a continuum. We frame the problem as a tradeoff among possible value-weighted outcomes with the motor system able to select among movement plans that assign probabilities to those outcomes [15].
We focused on a task where the key tradeoff is between the uncertainties at two locations (depth planes) along a reach trajectory, and we examined the covariance structure induced by a virtual obstacle placed between the subject and the goal. We employed a method for testing whether subjects maximize expected gain (the Dominance Test) based on an empirical characterization of relevant movement strategies available to the subject followed by a test, in each experimental condition, of whether the subject has selected the movement strategy that maximizes expected gain.
Studies aimed at identifying the visual [e.g.], [ 7,11], proprioceptive and biomechanical [e.g.], [ 8,27] elements that affect the specific form of a reach around an obstacle provide valuable contributions to solving the engineering problem of how these variables interact to modify reach trajectories planned around an obstacle. Our goal was different. We asked why, out of all possibilities, reach trajectories during obstacle avoidance have the form that they do.
Reaches have goals. Although particularly obvious when reaching around an obstacle, this aspect of reach planning in the presence of an intervening obstacle has previously been ignored. This has created something of a dilemma for subjects, who must choose how much ‘weight’ to assign to accidentally contacting an obstacle vs. successfully touching the target (reminiscent of studies where one is instructed to perform a task ‘as quickly as possible without sacrificing accuracy’). Subjects must resolve the conflict created by these contradictory goals by choosing a relative weighting, a weighting that cannot generally be inferred from the data alone. Here, we avoid these problems; obstacles are assigned a cost, giving a clear indication of the relative ‘importance’ of accidentally contacting an obstacle and of contacting the reach target.
Not only does our value manipulation allow us to avoid the uncertainty associated with arbitrary target and obstacle weightings that change by subject (and possibly by experimental condition), it is also a necessary element of an optimal model of obstacle-avoidance reach trajectories. The value component of (1) allows us to quantitatively predict the excursion magnitudes that form the basis of the comparison shown in Figure 5D. This in turn allows us to separate the optimal planning model (data on the unity line of Figure 5D) from other models of trajectory planning around the virtual obstacle that might make the same qualitative predictions, but are nevertheless quantitatively sub-optimal (though not observed, such data would lie along a non-unity-line in Figure 5D). Such a separation between qualitative and quantitative optimal performance is demonstrated in Tassinari et al. [28] and in Hudson et al. [15].
Our data have implications for a class of popular models of obstacle avoidance and reach planning in general based on optimal linear feedback control [29], [30], [31]. One important prediction of these models is that 2D and 3D variance may be partitioned among the axes to produce the best task performance; for example more precision may be required along the horizontal than along the vertical dimension, as in the current experiments. Such a system is capable of partitioning more variance to the dimension requiring less precision. Here, for the first time, we are looking at a task where variance at two points along the trajectory of a reaching movement affects the outcome of the movement. We find no evidence that subjects partition their covariance in response to rewards or costs. Had they done so, there should have been increased vertical variance, not increased horizontal variance. That is, any manipulation that in fact increased horizontal variance should have been ‘referred’ to the vertical dimension, where it would not have adversely affected performance.
We confined analysis to the intersection of trajectories with the obstacle and target planes. The subject's reward is determined by these two points: fingertip position at the intersection of the obstacle and target plane, nothing more. The subject should select a movement plan, , with the criteria that means and covariances in passing through these two critical planes maximize expected gain. Movement plans that satisfy these criteria clearly form a subset of all possible plans, but are they unique? Does the choice of the movement plan that maximizes expected gain in our experiment determine the entire trajectory bundle? Or, are there multiple planned trajectories (,, etc.) that match in mean and covariance at the two critical planes, but that deviate from elsewhere? We cannot exclude this possibility nor can we exclude the possibility that a subject chooses now , now , now , as he pleases. All would count as optimal choices of movement plan. The constraints we impose in the obstacle and target planes serve to select a set of equivalent optimal movement plans but further research is needed to determine the effect of the constraints we impose on the trajectory outside of the obstacle and target plane. In particular we avoided using data from outside the obstacle and target plane precisely because measured means and covariances at points along the trajectory outside of the obstacle and target planes may not reflect any single movement plan and it would be inappropriate to analyze them as if they were determined by the constraints of our task.
In our task the location of the fingertip at just two points along the trajectory determines the resulting reward or cost. We can readily generalize the task by adding additional obstacles along the path to create tasks for which the subject must consider his covariance at many points along the trajectory. This sort of generalization would allow investigation of the possible covariance structures along the reach trajectory available to the motor system. It also serves as a model task mimicking the constraints of many natural tasks where the goal is to maneuver around multiple obstacles to reach a goal, as in reaching into a computer chassis to extract one component.
We found that subjects' performance was close to that of a Bayesian decision-theoretic movement planner maximizing expected gain except for the most extreme conditions where the optimal choice of trajectory required a large excursion (“detour”) around the virtual obstacle. One possible explanation is that such movements entail a large biological cost and that the subject includes biological costs in the computation of expected gain. In effect he “prices” biological cost and is willing to reduce his monetary gain in order to reduce biological cost as well (see discussion in [32]). Although our current data cannot speak to this possibility, one might predict that separate measurements of biomechanical cost would allow these extreme conditions to be predicted as well.
The costs in our task are monetary but in theory would also apply to tasks where movement constraints are the results of injury or disease to the motor system [33], [34]. Patients might limit their motor repertoire in order to prevent undesirable outcomes such as pain or clumsiness, leading to long-term, conditioned motor deficits. This idea forms the basis of a now well-established rehabilitation approach, Constraint Induced Movement Therapy, in which the reward/cost structure of the environment is manipulated in ways that encourage the use of the previously avoided regions of motor space [35].
The conclusions we draw are based on movements confined to a narrow, clearly visible region of space immediately in front of the reviewer. Subjects presumably have considerable experience in coordinating eye and hand in this region of space before they begin the experiment. It would be interesting to investigate in future work with a full range of arm movements, including whether movement plans tend to avoid awkward or unusual movements.
We examined the problem of obstacle avoidance from the standpoint of Bayesian decision theory. Our approach is different from other work in the area of obstacle avoidance. Previously, this problem has been approached from the standpoint of theories that suggest that the CNS minimizes kinematic or dynamic variables (e.g., total force production), with the constraint that the hand path not intersect an obstacle. Of course, this approach fails to take account of two major contributions to real-world movement plans: the uncertainty of visual estimates and motor outcomes (even for the same real-world obstacle and planned trajectory), and variable costs associated with intersecting different kinds of obstacles (accidentally toppling a cup of water is very different from toppling a cup of scalding coffee). Instead, such models always predict the smallest possible trajectory deviation that does not contact the obstacle (with no ‘room for error’, so to speak). Moreover, the approach confounds the effect on trajectory of hitting an impenetrable obstacle and the cost to the subject. To return to the example we began with, it is easy to imagine circumstances where one would smash through the coffee cup to grasp something on the other side, such as a child in danger of falling. We see that obstacle avoidance, when viewed from the standpoint of Bayesian decision theory, can explain the amount of deviation around a virtual obstacle based on the cost of accidentally intersecting it, and the visuo-motor uncertainty in predicting the location of the fingertip when it passes the obstacle and when it reaches the target.
|
10.1371/journal.pntd.0003399 | Functional Activity of Monocytes and Macrophages in HTLV-1 Infected Subjects | The Human T lymphotropic virus type-1 (HTLV-1) infects predominantly T cells, inducing proliferation and lymphocyte activation. Additionally, HTLV-1 infected subjects are more susceptible to other infections caused by other intracellular agents. Monocytes/macrophages are important cells in the defense against intracellular pathogens. Our aims were to determine the frequency of monocytes subsets, expression of co-stimulatory molecules in these cells and to evaluate microbicidal ability and cytokine and chemokine production by macrophages from HTLV-1 infected subjects. Participants were 23 HTLV-1 carriers (HC), 22 HAM/TSP patients and 22 healthy subjects (HS) not infected with HTLV-1. The frequencies of monocyte subsets and expression of co-stimulatory molecules were determined by flow cytometry. Macrophages were infected with L. braziliensis or stimulated with LPS. Microbicidal activity of macrophages was determined by optic microscopy. Cytokines/chemokines from macrophage supernatants were measured by ELISA. HAM/TSP patients showed an increase frequency of intermediate monocytes, but expression of co-stimulatory molecules was similar between the groups. Macrophages from HTLV-1 infected individuals were infected with L. braziliensis at the same ratio than macrophages from HS, and all the groups had the same ability to kill Leishmania parasites. However, macrophages from HTLV-1 infected subjects produced more CXCL9 and CCL5, and less IL-10 than cells from HS. While there was no correlation between IFN-γ and cytokine/chemokine production by macrophages, there was a correlation between proviral load and TNF and CXCL10. These data showed a dissociation between the inflammatory response and microbicidal ability of macrophages from HTLV-1 infected subjects. While macrophages ability to kill an intracellular pathogen did not differ among HTLV-1 infected subjects, these cells secreted high amount of chemokines even in unstimulated cultures. Moreover the increasing inflammatory activity of macrophages was similar in HAM/TSP patients and HC and it was related to HTLV-1 proviral load rather than the high IFN-γ production observed in these subjects.
| HTLV-1 predominantly infects T cells, inducing cell proliferation and activation. While there is a larger amount of studies regarding T cells functions in HTLV-1 infected subjects, little is known about innate immunity. We evaluated monocyte and macrophage functions in HTLV-1 infected subjects. We observed that HAM/TSP patients have an increased frequency of intermediate monocytes, but expression of co-stimulatory molecules in these cells was similar between HTLV-1 infected subjects and healthy subjects (HS). Additionally, the microbicidal ability of macrophages from HTLV-1 infected subjects to kill Leishmania braziliensis is preserved, and these cells showed inflammatory profile, producing more CXCL9 and CCL5, and less IL-10 than macrophages from HS. It was important to determine if the exacerbated ability of macrophages to secrete cytokine was due to IFN-γ production. While there was no correlation between IFN-γ levels by PBMCs and cytokine/chemokine production by macrophages, there was a direct correlation between proviral load and TNF and CXCL10 levels. Our data indicate that despite the high production of proinflammatory mediators, macrophages from HTLV-1 infected subjects kill an intracellular pathogen in similar levels than cells from HS and pointed out for the role of viral factors inducing the inflammatory response in these cells.
| Human T lymphotropic virus type 1 (HTLV-1) infects about 15 to 20 million people worldwide, with endemic foci in virtually all continents [1], [2]. A large proportion of individuals remain asymptomatic until the end of life, but a subgroup of infected individuals will develop a malignant lymphoproliferative disease called adult T cell leukemia/lymphoma (ATLL) [3], [4] or a chronic neurodegenerative inflammatory disease called HTLV-1 associated myelopathy/tropical spastic paraparesis (HAM/TSP) [5]. Additionally, more than 40% of infected individuals will present clinical manifestations, such as infectious dermatitis [6], polymyositis [7], sicca syndrome [8], [9], overactive bladder and/or erectile dysfunction [10], [11], chronic periodontitis [12] and HTLV-1 associated arthropathy among other diseases [13], [14], [15]. The pathogenesis of diseases associated to HTLV-1 is related predominantly to the proviral load and the exaggerated inflammatory response in HTLV-1 infection [16], [17].
HTLV-1 infects predominantly CD4+ T cells, but CD8+ T cells [18], monocytes/macrophages [19], [20] and dendritic cells [21] can also be infected by the virus. The infection is characterized by a high spontaneous proliferation and activation of T cells, leading to high production and secretion of inflammatory mediators, such as TNF, IFN-γ, CXCL9 and CXCL10 [16], [22]. Previous immunological studies have directed attention to the role of T cells in HTLV-1 infection, and seek to correlate the dysfunctions of the adaptive immune system with the development of diseases or clinical manifestations associated with the virus. Very few studies have evaluated the role of the innate immune response in HTLV-1 infection.
It is well known that HTLV-1 infection increases susceptibility and severity to other infectious diseases [23], [24], [25]. The mechanism involved in the increased susceptibility of HTLV-1 infected subjects to other infectious agents is only partially known [23], [26]. Regarding intracellular pathogens, despite the high IFN-γ and TNF production there is an increased susceptibility to Mycobacterium tuberculosis [27], [28], [29] and fungal infections [30]. It is known that cells of the innate immunity response, such as neutrophils and macrophages, are important effectors cells against infectious agents. However very few studies have evaluated monocytes, macrophages or neutrophils functions in HTLV-1 infection. It is known that HTLV-1 infection results in spontaneous activation of neutrophils, as indicated by increasing in the number of positive cells in the nitroblue tetrazolium test (NBT) (indicating high burst oxidative activity), and by the decreasing in the number of neutrophils expressing CD62L and higher expression of CD66b [31], [32]. Regarding dendritic cells (DCs), some studies showed an increased expression of molecules involved in virus internalization process and T cells adhesion (DC-SIGN) [33], [34], and a decrease in CD14 and CD1a, molecules related with derived-monocytes DCs maturation have been described [35]. Moreover, DCs from HTLV-1 infected patients show an impaired expression of CD83, CD86 and HLA-DR after stimulation with TNF and reduced ability to stimulate T cells not infected with the virus [35]. A recent study developed in Jamaica, with a cohort of HTLV-1 infected subjects, documented a decreased frequency of plasmocytoid DCs (pDCs) and expression of HLA-DR in ATL and HAM/TSP patients compared to ACs and HC. Myeloid DCs (mDCs) also showed a lower expression of HLA-DR in HAM/TSP patients. However, the expression of CD86 in both plasmocytoid and mDCs was higher in HAM/TSP patients compared to the other groups. They also demonstrated that the programmed death ligand 1 (PD-L1) is high-expressed in DCs from HAM/TSP compared to ACs [36].
These and others dysfunctions in the myeloid cell lineage may modify the immune response of HTLV-1 infected subjects to antigens. However, studies regarding the inflammatory response and microbicidal activity of monocytes and macrophages in HTLV-1 infected subjects have not been performed. The aims of the present study were to evaluate monocytes and macrophages functions in HTLV-1 infected subjects, by comparing the frequency of monocyte subsets in HTLV-1 infected subjects and the ability of monocyte-derived macrophages from HTLV-1 infected subjects to produce cytokines and chemokines and to kill the intracellular pathogen Leishmania braziliensis. Moreover we evaluate if there are correlations between the frequency of monocytes subsets and cytokines/chemokines produced by macrophages with IFN-γ and proviral load in these individuals.
All HTLV-1 subjects have been followed at the HTLV-1 clinic of the Complexo Hospitalar Universitário Professor Edgard Santos (COM-HUPES), Federal University of Bahia, Brazil. The study was approved by the Ethics Committee from the Federal University of Bahia and all patients signed a document of informed consent.
This is a cross-sectional study with the purpose of evaluating the role of myeloid lineage cells (monocytes and macrophages) from HTLV-1 infected subjects. Participants included 45 HTLV-1 infected subjects, being 23 HTLV-1 carriers (HC), 22 patients diagnosed with HAM/TSP and 22 individuals not infected with HTLV-1 constituted the healthy subjects group (HS). Pregnant woman and individuals in use of immunosupressing drugs were excluded. The diagnosis of HTLV-1 infection was established by antibody detection by ELISA (Murex HTLV-I+II, Abbot, Dartford, UK) and confirmed by Western blot (HTLV blot 2.4, Genelabs. Singapore). Motor dysfunction and neurological involvement were determined by Osame's motor disability score (OMDS) [37] and Expanded disability status scale (EDSS) [38]. Individuals with an OMDS and EDSS equal to 0 were considered HC. Patients with OMDS ≥1 and presence of specific antibodies against HTLV-1 in the cerebrospinal fluid were diagnosed with HAM/TSP.
Peripheral blood mononuclear cells (PBMCs) were obtained from heparinized blood of HTLV-1 infected subjects and healthy controls, and separated by density gradient with Ficoll-Hypaque (GE Healthcare Bio – Sciences, Uppsala, Sweden). PBMCs from the interface were aspirated and washed with saline. After that, these cells were resuspended in RPMI 1640 culture medium with L-glutamine and 25 mM HEPES (Gibco BRL, Grand Island, New York, USA) supplemented with 10% fetal bovine serum (FBS) and 0.5% gentamicin at 10 mg/mL (Gibco BRL, Grand Island, New York, USA). PBMCs were then directed to three separate experiments: a) staining with specific antibodies for analyses of monocyte by flow cytometry; b) culture for determination of spontaneous production of IFN-γ by PBMCs. 3×106 cells/mL were incubated without stimulus or stimulated with PHA (5 µg/mL) at 37°C in 5% CO2 for 72 hours and then the supernatant was frozen for later determination of IFN-γ; or c) use for the differentiation of cultured monocytes into macrophages, 5×106 cells/mL. These cells were added to 4-well plates (Lab-Tek Permanox Chamber Slide, Electron Microscopy Sciences, Hatfield, PA) and incubated for 2 hours at 37°C and 5% CO2. Cells that did not adhere to the slides were removed by washing. The adherent cells (monocytes) were differentiated into macrophages after 6 days of culture at 37°C in 5% CO2, with absence of stimulus, in the presence of LPS or L. braziliensis.
The ex vivo frequency of monocyte subsets and expression of HLA-DR, CD80 and CD86 was determined using PBMCs from HC, HAM/TSP patients and HS. Cells were stained with monoclonal antibodies (anti-CD14-FITC, anti-CD16-PE-Cy5, anti-HLA-DR-PE, anti-CD80-PE e anti-CD86-PE, from eBioscience, San Diego, CA or R&D Systems, Minneapolis, MN) for 20 minutes at 4°C. PBMCs were washed with PBS and then fixed with 2% paraformaldehyde. Cells were then analyzed on the flow cytometer (II FacsCanto, BD Biosciences, San Jose, CA). Analysis was performed using FlowJo software version 7.6 (TreeStar, Ashland, OR). The monocyte population was selected based on size and cell granularity and then subdivided into classical (CD14++CD16-), intermediate (CD14+CD16+) and non-classical monocytes (CD14+CD16++).
A strain of L. braziliensis isolated from a patient with cutaneous leishmaniasis from the endemic area of Corte de Pedra, Salvador, Bahia, is maintained, cryopreserved, by the Immunology Service. Parasites were initially cultivated in tubes with biphasic medium (NNN) supplemented with 10% fetal bovine serum and maintained in culture in Schneider medium (LGC Biotechnology, São Paulo, Brazil) supplemented with 10% FBS and 1% penicillin streptomycin and glutamine (Gibco BRL, Grand Island, New York, USA) for expansion and proliferation of protozoa.
L. braziliensis promastigotes were maintained in Schneider medium until reaching stationary (infectious) growth phase. Parasites were than centrifuged and resuspended in RPMI 1640 medium and used to infect macrophages cultured from both HTLV-1 infected subjects and HS. Unstimulated or macrophages stimulated with lipopolysaccharide (LPS) from Escherichia coli (100 ng/mL) were used as controls. Infection with L. braziliensis was performed at a ratio of 5 parasites to 1 cell for 2 hours at 37°C in 5% CO2. Following the incubation period, extracellular parasites were removed by washing, and then the cells were incubated at 37°C and 5% CO2. The percentage of macrophages infected with L. braziliensis and the number of amastigotes per 100 macrophages were evaluated by optical microscopy after 2, 48 and 72 hours of infection and staining with Giemsa. Counts were performed by two independent observers who were unaware if the slides were from an HTLV-1 infected subject or from a healthy control. The results expressed are the average of the results from both observers.
Culture supernatants from unstimulated PBMCs or from macrophages were collected after 72 and 48 hours of incubation, respectively, and frozen at −20°C until used for determination of cytokines and chemokines. The IFN-γ levels (supernatants of PBMCs) and TNF, IL-10, CXCL9, CXCL10 and CCL5 (macrophages supernatants) were determined by ELISA, using commercial kits and following the manufacturer's instructions (DuoSet R&D Systems, Minneapolis, MN, USA and BD Pharmingen, San Diego, CA, USA). Due to the limited amount of cells some experiments did not included all patients. Data about the production of cytokines and chemokines produced by macrophages after infection by L. braziliensis were not represented in the figures, because this pathogen did not induced the production of these molecules.
DNA was extracted from 106 cells using proteinase K and salting-out method. The HTLV-1 proviral load was quantified using a real-time TaqMan PCR method [39]. Albumin DNA was used as an endogenous reference. Amplification and data acquisition were carried out using the ABI Prism 7700 Sequence detector system (Applied Biosystems). Standard curves were generated using a 10-fold serial dilution of a double-stranded plasmid (pcHTLV-ALB). All standard dilutions and control and individual samples were run in duplicate for both HTLV-1 and albumin DNA quantification. The normalized value of the HTLV-1 proviral load was calculated as the ratio of (HTLV-1 DNA average copy number/albumin DNA average copy number) ×2×106 and expressed as the number of HTLV-1 copies/106 cells.
Mann-Whitney test was used to compare IFN-γ production by PBMCs and proviral load between ACs and HAM/TSP patients. Kruskal-Wallis followed by Dunn's post test was used to assess differences between the three groups studied under the same conditions. Wilcoxon test was used to evaluate the influence of stimuli (LPS and L. braziliensis) compared to condition without stimulation. Spearman correlation test was used in the correlations results. Data were expressed as median and range (minimum and maximum values). GraphPad Prism 5 (San Diego, CA) was used to carry out the statistical evaluation and a P<0.05 was considered to indicate a significant difference.
Demographic characteristics, IFN-γ production, proviral load and degree of HAM/TSP severity of the participants on this study are shown in Table 1. Of the 45 HTLV-1 infected individuals from an HTLV-1 cohort, 23 were HTLV-1 carriers (HC) and 22 patients had HAM/TSP. There was a predominance of female in both HTLV-1 infected subject's group. Patients with HAM/TSP showed a higher production of IFN-γ by PBMCs than HC (1,979 pg/mL, ranging 38–3,661 pg/mL vs 863 pg/mL, ranging 0–3,666 pg/mL, respectively), P = 0.03, and also a greater proviral load (231,016 copies/106cells, ranging 932–1.186,254 copies/106cells, vs 22,665 copies/106cells, ranging 81–255,319 copies/106cells, respectively), P = 0,0004. Patients with HAM/TSP showed a median of OMDS of 6, ranging 3–9 (Table 1).
The frequencies of monocyte subsets (classical, intermediate and non-classical monocytes) in HC, HAM/TSP patients and HS were determined by flow cytometry and are shown in Fig. 1. The HC group showed a similar frequency of monocyte subsets as observed in the HS group. However, patients with HAM/TSP exhibit lower frequency of classical monocytes and higher frequency of intermediate monocytes than HC and HS. 89.4% of monocytes from HS were classical, while 6.4% were intermediate and 5.1% were non-classical. 83.9% of monocytes from HC were classical, 6.8% were intermediate and 7.5% were non-classical monocytes. HAM/TSP patients showed 75.1% of classical monocytes (P = 0.0005 compared to HC and HS), 23.9% of intermediate monocytes (P<0.0001 also compared to HC and HS) and 5% of non-classical monocytes (Fig. 1). There was no difference in the expression of HLA-DR, CD80 and CD86 by monocytes between HC, HAM/TSP patients and HS (P>0.05). The increasing in frequency of intermediate monocytes in HAM/TSP patients was not associated with IFN-γ production by PBMCs (Fig. 2).
To evaluate the susceptibility of macrophages from HTLV-1 infected subjects to be infected by an intracellular pathogen and the ability of these cells to kill it, macrophages were infected by L. braziliensis at a 5∶1 ratio. The percentage of infected macrophages and the number of amastigotes/100 macrophages were evaluated by optic microscopy after 2, 48 and 72 hours of infection, as shown in Fig. 3. There was no difference in the microbicidal activity between macrophages from the groups studied at any of the three time points following infection by L. braziliensis. Macrophages from HTLV-1 infected subjects (both HC and HAM/TSP patients) were initially infected by the parasite at the same proportion as macrophages from HS (Fig. 3A), and there were similar amounts of L. braziliensis amastigotes inside the cells after 2, 48 and 72 hours of infection in the groups (Fig. 3B).
Levels of TNF and IL-10 were evaluated in the supernatants from HTLV-1 infected subjects (HC and HAM/TSP) and HS macrophages cultured after 48 hours of incubation with or without LPS. These assays were performed by ELISA and data are shown in Fig. 4. Neither macrophages from HS, HC nor from HAM/TSP patients produced spontaneously significant detectable levels of TNF (0 pg/mL, 0 pg/mL, 16 pg/mL, respectively). When macrophages from three groups were stimulated with LPS, high levels of TNF were detected (P<0.0002). Macrophages from HS produced 2,145 pg/mL while macrophages from HC produced 2,102 pg/mL, and macrophages from HAM/TSP produced 1,965 pg/mL, with no statistically significant differences between those three values (Fig. 4A).
HC, HAM/TSP patients and HS macrophages did not produced IL-10 in significant levels either spontaneously (0 pg/mL, 32 pg/mL and 0 pg/mL, respectively). Upon LPS stimulation, all groups showed increased IL-10 production, but macrophages from HS produced more of this cytokine (265 pg/mL) than macrophages from HC and HAM/TSP patients (46 pg/mL and 31 pg/mL, respectively, P = 0.003) (Fig. 4B).
Levels of CXCL9, CXCL10 and CCL5 were evaluated in the supernatants from macrophage cultures of HTLV-1 infected subjects (HC and HAM/TSP patients) and HC after 48 hours of incubation, with or without LPS, as shown in Fig. 5. Macrophages from HC and HAM/TSP patient spontaneously produced more CXCL9 than macrophages from HS (32,115 pg/mL and 25,558 pg/mL vs 6,992 pg/mL, P = 0.003). Macrophages from HC and HAM/TSP patients also produced more CXCL9 than HS's macrophages after LPS stimulus (31,080 pg/mL and 26,834 pg/mL vs 9,648 pg/mL, P = 0.001). Furthermore, stimulation with LPS did not induce the production of CXCL9 in macrophages from HS and HTLV-1 infected subjects (Fig. 5A).
Macrophages from HC and HAM/TSP patients produced spontaneously similar levels of CXCL10 (2,458 pg/mL and 2,288 pg/mL) than macrophages from HS (255 pg/mL), After stimulus with LPS, we also did not observed differences between the production of this cytokine by macrophages from HS (2,785 pg/mL), HC (3,418 pg/mL) and HAM/TSP patients (3,201 pg/mL). However, while LPS increased the production of CXCL10 by macrophages from HS compared to the unstimulated condition (P<0.03), this stimuli did not increased CXCL10 production by macrophages from HC and HAM/TSP patients (Fig. 5B).
Macrophages from HAM/TSP patients produced more CCL5 than macrophages from HS (1,050 pg/mL vs 254 pg/mL, P<0.0001). LPS was responsible to increase the production of CCL5 in all group studied (P<0.003), but we did not observe statistically significant differences between those groups (Fig. 5C).
To evaluate if the spontaneous production of IFN-γ by PBMCs was associated with proviral load, correlations were performed using Spearman correlation and r significance test. We observed a direct correlation between IFN-γ and proviral load in HTLV-1 infected subjects, r = 0.43 and P = 0.004 (analyzing HC and HAM/TSP patients together) (Fig. 6).
To evaluate if the spontaneous production of IFN-γ by PBMCs was associated with cytokines/chemokines produced by macrophages from HTLV-1 infected subjects, correlations were performed using Spearman correlation and r significance test. We found no correlation between spontaneous production of IFN-γ and production of TNF (r = 0.36 and P = 0.08), IL-10 (r = 0.09 and P = 0.74), CXCL9 (r = −0.02 and P = 0.90), CXCL10 (r = 0.41 and P = 0.09) and CCL5 (r = 0.20 and P = 0.44) in HTLV-1 infected patients.
Correlations between proviral load and cytokines/chemokines produced by macrophages were performed using Spearman correlation and r significance test. We observed a positive correlation, although weak, between proviral load and TNF (r = 0.51 and P = 0.01) and CXCL10 (r = 0.63 e P = 0.05) (Fig. 7). There was no correlation between proviral load and the IL-10 (r = −0.34 and P = 0.26), CXCL9 (r = 0.26 and P = 0.24) and CCL5 (r = 0.35 and P = 0.21) production by macrophages from HTLV-1 infected individuals.
The activity and phenotype of T cells in HTLV-1 infection have been well studied. These cells are characterized by the increased expression of proinflammatory cytokines, such as TNF and IFN-γ, and increased production of IL-2, which helps maintain CD4+ and CD8+ T cell proliferation [16], [22]. In contrast very little is known about innate immunity during HTLV-1 infection. We observed that HAM/TSP patients exhibit a higher frequency of intermediate (inflammatory) monocytes than HC and HS, but it was not associated with IFN-γ levels. While the microbicidal ability from HTLV-1 infected subject's macrophages was preserved, macrophages from HTLV-1 infected individuals produced more CXCL9 and CCL5, and less IL-10 than macrophages from HS. Indeed while there was no correlation between IFN-γ and cytokine levels, there was a correlation between proviral loads and TNF and CXCL10 production.
It is known that monocytes are a heterogeneous population of cells and can be classified, based on the expression of CD14 and CD16, as classical, intermediate or inflammatory and non-classical. Here we documented that HAM/TSP patients have a higher frequency of intermediate monocytes than HC and HS. It is known that intermediate monocytes are the main source of TNF among the three subpopulations [40]. As PBMCs from HTLV-1 infected subjects, especially from HAM/TSP patients, produce more proinflammatory mediators such as CXCL9, CXCL10 and TNF than PBMCs from HS [16], [41], we hypothesized that monocytes may play an important role in the inflammatory response and in the pathogenesis of HAM/TSP. Agreeing with this hypothesis, we observed that there was no correlation between IFN-γ production and increasing intermediate monocytes.
Macrophages are capable of killing infectious agents but may also serve as habitat for intracellular pathogens. As HTLV-1 infection increases the susceptibility to infections caused by intracellular agents such as M. tuberculosis, we sought to evaluate macrophage microbicidal function in HTLV-1 infection. To evaluate macrophage killing we used L. braziliensis, an intracellular pathogen knowing to interact with TLR2, TLR4 and TLR9 [42], [43] and with the ability to multiply in macrophages. As the number of intracellular parasites inside the macrophages after 2 hours of infection was similar to that observed in HS, it was concluded that penetration and/or phagocytosis of L. braziliensis was equal in the three groups. Moreover the leishmania killing, evaluated at 48 hours and 72 hours by quantifying the number of intracellular amastigotes in macrophages, was similar. This data extend our previous observations that the ability of neutrophils from HTLV-1 infected subjects to kill leishmania parasites is preserved [32]. As IFN-γ is the main cytokine known to activate macrophages and high IFN-γ production is observed in HTLV-1 infected subjects, one could expect that macrophages from HAM/TSP patients had greater ability to kill an intracellular pathogen, but we did not find that the marked Th1 environment observed in these individuals modified the killing ability of myeloid cells.
Macrophages activation has been used to indicate both, increasing ability of killing and secretion of molecules such as chemokines and cytokines. Macrophages are also a heterogeneous cells population and macrophage's subsets have been defined as classical macrophages that are associated with a type 1 immune response, and alternative macrophages that secret IL-4 and IL-10 [44]. Here we showed that the killing ability and secretion of cytokines by macrophages are not necessarily associated. While we did not observe an increase in the killing ability of macrophages from HTLV-1 infected subjects, manly in unstimulated cells or after LPS stimulation, macrophages produced higher levels of CXCL9 and CCL5 than HS's macrophages. This indicates that at the level of innate immunity, there was an enhancement of chemokines related with both Th1 and Th2 immune responses. This is an agreement with the observation that atopic diseases may occur in HTLV-1 infection [30] and that PBMC from HC produce higher amount of Th2 cytokines than cells from HS [16]. In this study we did not observed a higher production of CXCL10 by macrophages from HC and HAM/TSP patients compared to HS, but while the cells from HS were stimulated by LPS to produce this chemokine, it was not observed in macrophages from HC and HAM/TSP patients. A reasonable explanation for this observation is that cells from HTLV-1 infected subjects could have already reached the limit of production of these chemokine even before the addition of LPS in the cultures.
It's known the ability of LPS to induce strong TNF production which is followed by IL-10 synthesis. The observation that HS's macrophages stimulated with LPS produced more IL-10 than cells from HTLV-1 infected subjects, both HC and HAM/TSP patients, suggests an impairment on macrophages of these individuals to secret this regulatory cytokine. Although the high production of proinflammatory mediators documented in HTLV-1 infection, especially in HAM/TSP patients, such as IFN-γ, IL-1, IL-6, could contribute to the increased production of chemokines and TNF by macrophages, we did not find a correlation between the IFN-γ production by PBMCs and TNF, IL-10, CXCL9, CXCL10 and CCL5 produced by macrophages. In contrast with the absence of correlation between IFN-γ and cytokines/chemokines levels, there was a direct correlation between proviral load and TNF and CXCL10 levels. This indicates that the HTLV-1 by itself or by inducing soluble mediators play a key role in the increased ability of macrophages to produce cytokine during HTLV-1 infection. Therefore it is important that future studies evaluate not only the role of viral proteins and other viral factors in activate innate immunity cells but also in disease expression associated to HTLV-1.
It is already known that PBMCs from HAM/TSP patients produced more IFN-γ and have a higher proviral load compared to HC [16], [17], [45]. In agreement with previous observations, we also documented a positive correlation between proviral load and IFN-γ production by PBMCs from HTLV-1 infected subjects. However our data clearly show while the IFN-γ production did not correlate with the increased cytokine/chemokine production by macrophages in HTLV-1 infected subjects, there was a correlation between proviral load and TNF and CXCL10.
This is the first study to evaluate the monocyte subsets and activation, as well as the inflammatory and microbicidal activity from macrophages in HTLV-1 infection. Our data indicate that patients with HAM/TSP have an increase frequency of intermediate monocytes. We also observed that macrophages from HTLV-1 infected subjects have the same ability to phagocytize and kill an intracellular pathogen as healthy subject's macrophages, but that proinflammatory activity was enhanced in HTLV-1 infected subjects. The dissociation between microbicidal activity and production of proinflammatory cytokines and chemokines is a relevant subject and show that inflammation and killing may be independent functions. As IFN-γ is the main cytokine that activate macrophages we expected at some extension a relationship between this cytokine and the inflammatory profile observed in monocytes and macrophages. However the absence of correlation between IFN-γ and cytokine/chemokine production and a direct correlation between proviral load and TNF and CXCL10 produced by macrophages suggests innate immune cells triggered by viral factors may play an important role in the inflammatory response and in the pathogenesis of HAM/TSP.
|
10.1371/journal.pgen.1002660 | The Probability of a Gene Tree Topology within a Phylogenetic Network with Applications to Hybridization Detection | Gene tree topologies have proven a powerful data source for various tasks, including species tree inference and species delimitation. Consequently, methods for computing probabilities of gene trees within species trees have been developed and widely used in probabilistic inference frameworks. All these methods assume an underlying multispecies coalescent model. However, when reticulate evolutionary events such as hybridization occur, these methods are inadequate, as they do not account for such events. Methods that account for both hybridization and deep coalescence in computing the probability of a gene tree topology currently exist for very limited cases. However, no such methods exist for general cases, owing primarily to the fact that it is currently unknown how to compute the probability of a gene tree topology within the branches of a phylogenetic network. Here we present a novel method for computing the probability of gene tree topologies on phylogenetic networks and demonstrate its application to the inference of hybridization in the presence of incomplete lineage sorting. We reanalyze a Saccharomyces species data set for which multiple analyses had converged on a species tree candidate. Using our method, though, we show that an evolutionary hypothesis involving hybridization in this group has better support than one of strict divergence. A similar reanalysis on a group of three Drosophila species shows that the data is consistent with hybridization. Further, using extensive simulation studies, we demonstrate the power of gene tree topologies at obtaining accurate estimates of branch lengths and hybridization probabilities of a given phylogenetic network. Finally, we discuss identifiability issues with detecting hybridization, particularly in cases that involve extinction or incomplete sampling of taxa.
| Species trees depict how species split and diverge. Within the branches of a species tree, gene trees, which depict the evolutionary histories of different genomic regions in the species, grow. Evolutionary analyses of the genomes of closely related organisms have highlighted the phenomenon that gene trees may disagree with each other as well as with the species tree that contains them due to deep coalescence. Furthermore, for several groups of organisms, hybridization plays an important role in their evolution and diversification. This evolutionary event also results in gene tree incongruence and gives rise to a species phylogeny that is a network. Thus, inferring the evolutionary histories of groups of organisms where hybridization is known, or suspected, to play an evolutionary role requires dealing simultaneously with hybridization and other sources of gene tree incongruence. Currently, no methods exist for doing this with general scenarios of hybridization. In this paper, we propose the first method for this task and demonstrate its performance. We revisit the analysis of a set of yeast species and another of Drosophila species, and show that evolutionary histories involving hybridization have higher support than the strictly diverging evolutionary histories estimated when not incorporating hybridization in the analysis.
| A molecular systematics paradigm that views molecular sequences as the characters of gene trees, and gene trees as characters of the species tree [1] is being increasingly adopted in the post-genomic era [2], [3]. Several models of evolution for the former type of characters have been devised [4], while the coalescent has been the main model of the latter type of characters [5], [6]. However, hybridization, a process that is believed to play an important role in the speciation and evolutionary innovations of several groups of plant and animal species [7], [8], results in reticulate (species) evolutionary histories that are best modeled using a phylogenetic network [9], [10]. Further, as hybridization may occur between closely related species, incongruence among gene trees may also be partly due to deep coalescence, and distinguishing between the two factors is hard under these conditions [11]. Therefore, to enable a more general application of the new paradigm, a phylogenetic network model that allows simultaneously for deep coalescence events as well as hybridization is needed [12]. This model can be devised by extending the coalescent model to allow for computing gene tree probabilities in the presence of hybridization. In this paper we focus on gene tree topologies and analyze the signal they contain for detecting hybridization in the presence of deep coalescence.
Applications of probabilities of gene tree topologies given species trees include determining statistical consistency (or inconsistency) of topology-based methods for inferring species trees 13–15, testing the multispecies coalescent model [13], [16], determining identifiability of species trees using linear invariants of functions of gene tree topology probabilities [17], [18], delimiting species [19], designing simulation studies for species tree inference methods [20]–[22], and inferring species trees [23], [24]. We expect that similar applications may be useful for probabilities of gene tree topologies given species networks. In particular, it will be useful to be able to evaluate the performance of methods that infer species trees in the presence of hybridization as well as the performance of methods for inferring species networks. Knowing the distribution of gene tree topologies could also be useful for estimating the probability that two gene trees have the same topology, a quantity that is used in constructing the prior which models gene tree discordance in BUCKy [25], a program that is often used to estimate species trees or concordance trees.
A method for computing the probability mass function of gene tree topologies in the absence of hybridization (i.e., under the multispecies coalescent model is assumed) is given by Degnan and Salter [26]. However, to handle hybridization and deep coalescence simultaneously, this method has to be extended to allow for reticulate species evolutionary histories.
Indeed, attempts have been made recently for this very task [27]–[30], all of which have focused on very limited special cases where the phylogenetic network topology is known and contains one or two hybridization events, and a single allele sampled per species. However, a general formula for the probability of a gene tree topology given a general (any number of taxa, hybridizations, gene trees, and/or alleles) phylogenetic network has remained elusive.
A binary phylogenetic network topology contains two types of nodes: tree nodes, each of which has exactly one parent (except for the root, which has zero parents), and reticulation nodes, each of which has exactly two parents. The edge incident into a tree node is called tree edge, and the edges incident into a reticulation node are called reticulation edges. In our context, we associate with a phylogenetic network a vector of branch lengths (in units of generations, where is the effective population size in that branch) and a vector of hybridization probabilities (which indicates for each allele in a hybrid population its probability of inheritance from each of the two parent populations); see Text S1 for formal definition. The gene tree topology can be viewed as a random variable with probability mass function . In this paper, we solve the aforementioned open problem by reporting on a novel method for computing the probability of a gene tree topology given a phylogenetic network, .
We illustrate the use of gene tree topology probabilities to estimate the values of species network parameters using the likelihood of the gene tree topologies. This application allows for disentangling hybridization and deep coalescence when analyzing a set of incongruent gene trees, as both events can give rise to similar incongruence patterns. Given a collection of gene tree topologies, one per locus, in a set of sampled loci, the likelihood function is given by(1)This formulation provides a framework for estimating the parameters and of an evolutionary history hypothesis , given a collection of gene trees . Estimates of 0 or 1 for the entries in the vector reflect the absence of evidence for hybridization based on the gene tree topology distribution.
As gene tree topologies are estimated from sequence data, there is often uncertainty about them. In our method, we account for that in two ways: (1) by considering a set of gene tree topology candidates, along with their associated probabilities (produced, for example, by a Bayesian analysis), and (2) by considering for each locus the strict consensus of all optimal tree topologies computed for that locus (produced, for example, by a maximum parsimony analysis).
Finally, to account for model complexity, we employ a simple technique based on three information criteria, AIC [31], AICc [32] and BIC [33]. While these criteria have their shortcomings for model selection, the question of how to account for phylogenetic network complexity is still wide open and no methods exist for addressing it systematically [10].
We have implemented our method in the publicly available software package PhyloNet [34] and demonstrated its broad utilities in three domains. First, we reanalyze a Saccharomyces data set and a Drosophila data set, and find support for hybridization in both data sets. Second, we show the identifiability of the parameter values of certain reticulate evolutionary histories. Third, we highlight and discuss the lack of identifiability of the parameters in other scenarios that involve extinctions.
We begin by reviewing Degnan and Salter's method for computing the probability gene tree topologies on species trees, and then describe our novel extension to the case of species networks.
Degnan and Salter [26] gave the mass probability function of a gene tree topology for a given species tree with topology and vector of branch lengths as(2)which is taken over coalescent histories from the set of all coalescent histories . The product is taken over all internal branches of the species tree. The term is the probability that lineages coalesce into lineages on branch whose length is . And the terms and represents the probability that the coalescent events agree with the gene tree topology. In particular, is the number of ways that coalescent events can occur consistently with the gene tree and is the number of sequences of coalescences that give the number of coalescent events specified by . However, this equation assumes that is a tree and as such is inapplicable to reticulate evolutionary histories. Recently, this equation was adapted to very special cases of species phylogenies with hybridization [28]–[30]. However, none of these adaptations is general enough to allow for multiple hybridizations, multiple alleles per species, or arbitrary divergence patterns following hybridization. We present a novel approach for generalizing this equation to handle hybridization. Our approach is general enough in that it allows for computing gene tree probabilities on any binary phylogenetic network topology, thus overcoming limitations of recent works.
Our approach for computing the probability of a gene tree given a species network has three steps. First, is converted into a multilabeled (MUL) tree (a tree whose leaves are not uniquely labeled by a set of taxa; see Text S1); second, the alleles at the tips of are mapped in every valid way to the tips of ; and, finally, the probability of is computed as the sum, over all valid allele mappings, of probabilities of given (see Figure 1).
Thus far, we have assumed that we have an accurate, fully resolved gene tree for each locus. However, in practice, gene tree topologies are inferred from sequence data and, as such, there is uncertainty about them. In Bayesian inference, this uncertainty is reflected by a posterior distribution of gene tree topologies. In a parsimony analysis, several equally optimal trees are computed. We propose here a way for incorporating this uncertainty into the framework above. Assume we have loci under analysis, and for each locus , a Bayesian analysis of the sequence alignment returns a set of gene trees , along with their associated posterior probabilities (). Now, let be the set of all distinct tree topologies computed on all loci, and for each let be the sum of posterior probabilities associated with all gene trees computed over all loci whose topology is . Thus, and . Then, we replace Eq. (1) by(7)We note that if or for each and , then Eq. (7) is equivalent to Eq. (1), and both are multinomial likelihoods. This multinomial approach has also been used elsewhere for both species networks under simple hybridization scenarios [28] and species trees [24]. We additionally allow the terms to be between 0 and 1 (and therefore to be non-integer values) in order to reflect uncertainty in the estimated gene trees.
In the case where a maximum parsimony analysis is conducted to infer gene trees on the individual loci, a different treatment is necessary, since for each locus, all inferred trees are equally optimal. For locus , let be the strict consensus of all optimal gene tree topologies found. Then, Eq. (1) becomes(8)where is the set of all binary refinements of gene tree topology .
Using our method to compute the likelihood function given by Eq. (1), we reanalyzed the yeast data set of [35], which consists of 106 loci, each with a single allele sampled from seven Saccharomyces species S. cerevisiae (Scer), S. paradoxus (Spar), S. mikatae (Smik), S. kudriavzevii (Skud), S. bayanus (Sbay), S. castellii (Scas), S. kluyveri (Sklu), and the outgroup fungus Candida albicans (Calb). Given that there is no indication of coalescences deeper than the MRCA of Scer, Spar, Smik, Skud, and Sbay [36], we focused only on the evolutionary history of these five species (see Text S1). We inferred gene trees using Bayesian inference in MrBayes [37] and using maximum parsimony in PAUP* [38] (see Text S1 for settings).
The species tree that has been reported for these five species, based on the 106 loci, is shown in Figure 2A [35]. Further, additional studies inferred the tree in Figure 2B as a very close candidate for giving rise to the 106 gene trees, under the coalescent model [36], [39]. Notice that the difference between the two trees is the placement of Skud, which flags hybridization as a possibility. Indeed, the phylogenetic network topologies in Figure 2C and 2D have been proposed as an alternative evolutionary history, under the stochastic framework of [40], as well as the parsimony framework of [30].
Using the 106 gene trees, we estimated the times , , , and for the six phylogenies in Figure 2 that maximize the likelihood function (we used a grid search of values between 0.05 and 4, with step length of 0.05 for branch lengths, and values between 0 and 1 with step length of 0.01 for ). Table 1 lists the values of the parameters computed using Eq. (7) on the gene trees inferred by MrBayes and Table 2 lists the values of the parameters computed using Eq. (8) on the gene trees inferred by PAUP*, as well as the values of three information criteria, AIC [31], AICc [32] and BIC [33], in order to account for the number of parameters and allow for model selection.
Out of the 106 gene trees (using either of the two inference methods), roughly 100 trees placed Scer and Spar as sister taxa, which potentially reflects the lack of deep coalescence involving this clade (and is reflected by the relatively large values estimated). Roughly 25% of the gene trees did not show monophyly of the group Scer, Spar, and Smik, thus indicating a mild level of deep coalescence involving these three species (and reflected by the relatively small values estimated). However, a large proportion of the 106 gene trees indicated incongruence involving Skud; see . This pattern is reflected by the very low estimates of the time on the two phylogenetic trees in Figure 2. On the other hand, analysis under the phylogenetic network models of Figure 2C and 2D indicates a larger divergence time, with substantial extent of hybridization. These latter hypotheses naturally result in a better likelihood score. When accounting for model complexity, all three information criteria indicated that these two phylogenetic network models with extensive hybridization and larger divergence time between Sbay and the ( Smik,( Scer,Spar)) clade provide better fit for the data. Further, while both networks produced identical hybridization probabilities, the network in Figure 2D had much lower values of the information criteria than those of the network in Figure 2E. The networks in Figure 2E and 2F have lower support (under all measures) than the other four phylogenies. In summary, our analysis gives higher support for the hypothesis of extensive hybridization, a low degree of deep coalescence, and long branch lengths than to the hypothesis of a species tree with short branches and extensive deep coalescence. It is worth mentioning that while the three networks in Figure 2C–2E were reported as equally optimal under a parsimonious reconciliation [36], our new framework can distinguish among the three, and identifies the network in Figure 2D as best, followed by the one in Figure 2C (the network of Figure 2E is found to be a worse fit than either of the two species tree candidates).
We reanalyzed the three-species Drosophila data set of [41], which includes D. melanogaster ( Dmel), D. yakuba ( Dyak), and D. erecta ( Dere).
The data set consisted of loci supporting the three possible gene tree topologies as follows:
For a species tree with three species and one individual sampled per species, the multispecies coalescent predicts that the two gene trees with topologies different from that of the species tree each occur with probability , where is the length of the one internal branch in coalescent units [42]. Two important predictions under the coalescent are therefore that the two nonmatching gene trees are expected to be tied in frequency and that both occur less than of the time, with the matching gene tree topology occurring more than of the time. This tie in the expected frequency of nonmatching gene trees is observed in some three-taxon data sets, but not in others, including the Drosophila data set.
Although this deviation from symmetry can be explained by a model of population subdivision, where the subdivision must occur in the internal branch as well as the population ancestral to all three species [43], the asymmetry can also be explained by the simplest hybridization network on three species with just one hybridization parameter (Figure 3).
We considered six candidates for the species phylogeny: three with no hybridization, and three with hybridizations involving different pairs of species (see Figure 3). For the three phylogenetic trees, we estimated the time that maximizes the probability of observing all gene trees, and for the three phylogenetic networks, we additionally estimated the hybridization probability .
The results in Table 3 show that of the three phylogenetic trees, the one in Figure 3A provides the best fit of the data, which is in agreement with the analysis in [41]. In fact, the value of we estimated on the other two trees was the lowest value we used in the estimation procedure. Clearly, this value can be arbitrarily small for these two trees, since the unresolved phylogeny ( Dmel, Dere, Dyak) fits the data better.
Among the three network candidates, the one in Figure 3D has the best fit of the data. This network, with a value of , indicates that of the alleles sampled from Dere shared a common ancestor first with alleles from Dyak (reflecting the tree in Figure 3A), while of the alleles from Dere shared a common ancestor first with alleles from Dmel (reflecting the tree in Figure 3B). Indeed, this network is the smallest network (in terms of the number of reticulation nodes) that reconciles both trees. Further, the change in AIC for this network is , indicating a much better fit than the best tree (Figure 3A). As noted previously [43], a -square test will also strongly reject the hypothesis that the species relationships are tree-like with random mating.
This three-taxon example can be analyzed analytically. Fitting a hybridization parameter allows a perfect fit to any observed frequencies of gene tree topologies for three species for one of the three networks in Figure 3. We let , , and represent the probabilities of topologies (Dmel,( Dere, Dyak)), ((Dmel, Dere), Dyak), and ((Dmel, Dyak), Dere) under the network in Figure 3D. Then
This system has the unique solution(9)for and (either at least one of the gene tree probabilities is less than if since they sum to 1.0; or if they are all exactly 1/3, then a star tree with and any exactly fits the data). Thus we can estimate and using the observed and in equation (9), and this also maximizes the likelihood.
For the simulated data, we evolved gene trees within the branches of phylogenetic networks, while varying branch lengths and hybridization probabilities, and investigated two questions: (1) how much data (gene trees) is needed to obtain accurate inference of the parameters (branch lengths and/or hybridization probabilities)? (2) are the parameters always identifiable? To answer these two questions, we investigated six different phylogenetic network topologies that involved single reticulation scenario, two reticulation scenarios (dependent and independent), and cases with extinctions involving the species that hybridize (see Text S1).
Our results show that both hybridization probabilities and branch lengths can be estimated with very high accuracy provided that no extinction events were involved in the parents of hybrid populations (see Text S1). Further, this accuracy can be achieved even when using the smallest number of gene trees we used in our study, which is 10. Under these settings, estimates using our framework seemed to converge quickly to the true values.
We also investigated the performance of the method, as well as identifiability issues when phylogenetic signal from at least one of the species involved in the hybridization is completely lost. Figure 4 shows the results for one such scenario (see Text S1 for another scenario that involves the loss of phylogenetic signal from both species involved in the hybridization).
Panels Figure 4B–4D show that when the true values of and are assumed to be known in the estimation procedure (the value of is irrelevant in the case when a single allele is sampled per species), the estimates of the hybridization probabilities converge to the true values. However, unlike the cases that did not involved extinctions, a larger number of gene trees is now required to obtain an accurate estimate (while there are only three possible gene tree topologies, a large number of gene trees need be sampled in order for the three topologies' frequencies to be informative). The time intervals of coalescent units amount to a large extent of deep coalescence events, which blurs the phylogenetic signal, and results in slight over- or under-estimation of the hybridization probabilities (Text S1 shows the results for the time interval with ).
If the topology of the network in Figure 4A is assumed to be known, but both the branch lengths and hybridization probabilities are to be estimated, then these parameters are unidentifiable; that is, two different pairs of vectors of branch lengths and hybridization probabilities can be found to explain the observed data with exactly the same probability (see Text S1). If at least two alleles are sampled from species B, then the parameter values become identifiable; however, an extremely large, and potentially infeasible, number of gene trees need to be sampled to uniquely identify the parameter values in practice (see Text S1).
Furthermore, in the special case where , a phylogenetic tree, with appropriate branch lengths can be found, to fit the data exactly with the same probability that the phylogenetic network would. Let be the branch lengths vector with , , and , and let be the hybridization probabilities vector with . Now, consider the phylogenetic tree in Figure 4E. Then, if we set as a function of , , and , using , then, for any gene tree . The values of are shown in Figure 4F–4H. These results show that as increases, the value of becomes unaffected by , and that increasing proportionally to the increase in always maintains identical probabilities of gene trees under both species phylogenies (see Text S1).
Our method for computing the probability of gene trees under hybridization and deep coalescence allows for analyzing data sets with arbitrary complexity of evolutionary histories in terms of the hybridization scenarios. When parameters are identifiable, our method estimates their values with high accuracy from a relatively small number of loci. Further, our method can be used to show lack of identifiability of model parameters for other cases. Our method supports a hypothesis of larger divergence time coupled with hybridization over short divergence times (with extensive deep coalescence) in a yeast data set. Finally, for a large Drosophila data set, our method indicated no hybridization based on the sampled loci.
We have focused on calculating probabilities of gene tree topologies and using these probabilities to infer species networks. In addition, the joint density of the coalescence times and topology in the gene trees could be used to infer species networks. Indeed, this approach has been used for networks where reticulation nodes have one descendant which is an extant species [29], using the density for coalescence times derived by Rannala and Yang [44]. This approach is computationally faster than computing gene tree topology probabilities because it is not necessary to sum over a large number of coalescent histories. To compute this joint density, each gene sampled can potentially have to trace through up to possible paths through the network, where is the number of hybridization events ancestral to the sampled gene from species , and the density will take the form of a sum over possible paths through the network. (In contrast, computing the probability of a topology will require mappings of alleles to the MUL-tree, and each gene topology calculation will require summing over coalescent histories.) This joint density for the gene trees with coalescence times could then be used in either maximum likelihood or Bayesian frameworks to infer the species network.
An important advantage of using coalescence times is that certain networks might be identifiable using coalescence times when probabilities of topologies might not identify the network. In the example of Figure 3A, although the gene tree topology probabilities can be obtained by a tree, the distribution of the coalescence times between lineages sampled from B and C is a mixture of three shifted exponential distributions if , but a mixture of two shifted exponential distributions if . For example, if , and are known but and are unkown, then the likelihood of observing a coalescence between a B and C lineage for times slightly greater will be very low if , and much higher for , thus making it possible to test whether when coalescence times are used.
Another identifiability issue is that both population subdivision and hybridization can lead to the asymmetry in gene tree topology probabilities in the 3-taxon case such as observed in the Drosophila example discussed earlier, where the two least frequently observed topologies are not tied in frequency. Either population subdivision, with a parameter describing the probability that the two most closely related species fail to coalesce in the ancestral population due to population structure, or hybridization can fit the data for the gene tree topologies. However, the two models could imply different distributions on coalescence times, which might therefore be useful in distinguishing the models. We note that identifiability in the case of three species with one individual per species might be especially limited due to the small number of gene tree topology probabilities that can be used to estimate parameters. In the case of identifying rooted species trees from unrooted gene trees with one lineage per species, for example, identifiability is achieved only with 5 or more species [17].
We consider it desirable to develop many methods for inferring species trees and species networks so that their properties and performances can be compared. In the case of species tree inference, there are advantages and disadvantages to using topology-based methods versus methods that include branch lengths, and in using likelihood versus Bayesian methods. We expect that many of these strengths and weaknesses may carry over to the case of inferring networks. For moderately sized data sets, Bayesian methods that model branch lengths and uncertainty in the gene trees such as BEST [45] and *BEAST [46] often have the best performance [47]. However, these methods require estimating the joint posterior distribution of the species tree and gene trees and therefore are difficult to implement for large numbers of loci. Maximizing the likelihood of the gene trees and their coalescent times (but without accounting for uncertainty in the gene trees), as in STEM [48], is fast and has very good performance on known gene trees but seems to be very sensitive to the assumption that branch lengths are estimated correctly [24], [49]. Maximizing the likelihood of the species tree using only gene tree topologies using the program STELLS, even while not accounting for uncertainty in the gene trees, tended to have better performance than STEM for a large simulated data set ( loci on 8 taxa) and worse performance on fewer loci [24]. Which method is optimal for inferring species trees or networks might depend on many factors such as the number of loci, the number of lineages sampled per species, the accuracy with which branch lengths can be estimated, the extent to which there are model violations, and the speciation history [49].
Two common assumptions in multispecies coalescent models are that there is no recombination within loci (and free recombination between loci) and that ancestral population sizes are constant.
Recombination can lead to different portions of a gene alignment effectively having distinct gene tree topologies. Ideally, alignments should be chosen so that recombination within genes is unlikely. This can be achieved by testing alignments beforehand for recombination using many available methods [50]–[52], or for whole genome data, choosing the cutoffs for loci such that they are unlikely to occur at recombination breakpoints [53]. In addition, recombination may lead to greater violations of the coalescent model for branch lengths than for topologies [53], so that topology-based methods might be less sensitive to the assumption that there is no recombination within loci. In addition, a recent simulation study found that recombination within loci did not have much impact on species tree inference methods for a wide range of recombination rates [54].
Coalescent models often assume that ancestral populations have constant size for the duration of the population (i.e., a constant size for a given branch of the species tree, but not necessarily the same on different branches). The program *BEAST [46] allows for ancestral population sizes to change linearly with time. Nonconstant population sizes will tend to result in branch lengths that make topologies more (or less) star-like for populations that are increasing (or decreasing) in size [55]. One approach to modelling a changing population size would be to break up a branch into intervals that are relatively constant in size. Suppose, for instance that a branch consists of an interval of generations with population size , and generations with size . The total time of the branch in coalescent units is . Although unequal values of can affect the distribution of coalescence times (for example, if but , then coalescence events might be more likely to occur in the interval with size ), the probabilities of topologies arising in this branch are not affected and can be calculated just using the total time . In particular, for the functions , which are the terms that depend on time in the calculations for gene tree topology probabilities, we havewhich is an instance of the Chapman-Kolmogorov equations because the number of lineages is a continuous time Markov chain (a death chain) [56].
We expect that topology-based methods may show more robustness to recombination and changing population sizes than approaches which explicitly model coalescence times. However, for estimating species trees and networks from gene trees, as in other areas of statistical inference, there is likely to be a tradeoff between power and robustness for methods that do and do not model branch lengths of the gene trees.
A current limitation to the procedure we have outlined for estimating hybridization is that we require a set of candidate networks on which to perform model selection. In some cases, such a set of candidate networks can be obtained by considering specific hypotheses related to biogeographical information. Candidate networks can also be generated using supernetworks from gene trees [57] or other network methods [9]. Often these methods will generate very complicated networks if there are many conflicts in the data, so it might be useful to choose different random subsets of well-supported (or frequently occurring) gene tree topologies to generate candidate species networks. In the future it will be desirable to develop algorithms that directly search the space of species networks in order to automate searching for optimal species networks.
|
10.1371/journal.pntd.0007777 | Ultraviolet sensitivity of WASH (water, sanitation, and hygiene) -related helminths: A systematic review | Helminthiases are a group of disabling neglected tropical diseases that affect billions of people worldwide. Current control methods use preventative chemotherapy but reinfection is common and an inter-sectoral approach is required if elimination is to be achieved. Household and community scale water treatment can be used to provide a safe alternative water supply for contact activities, reducing exposure to WASH (water, sanitation, and hygiene) -related helminths. With the introduction of ultraviolet light emitting diodes (UV-C LEDs), ultraviolet (UV) disinfection could be a realistic option for water treatment in low-income regions in the near future, to provide safe alternative water supplies for drinking and contact activities such as handwashing, bathing, and laundry, but currently there is no guidance for the use of UV or solar disinfection against helminths.
A qualitative systematic review of existing literature was carried out to establish which WASH-related helminths are more susceptible to UV disinfection and identify gaps in research to inform future studies. The search included all species that can infect humans and can be transmitted through water or wastewater. Five online databases were searched and results were categorized based on the UV source: sunlight and solar simulators, UV-A and UV-B (long wavelength) sources, and UV-C (germicidal) sources.
There has been very little research into the UV sensitivity of helminths; only 47 studies were included in this review and the majority were carried out before the standard protocol for UV disinfection experiments was published. Only 18 species were studied; however all species could be inactivated by UV light. Fluences required to achieve a 1-log inactivation ranged from 5 mJ/cm2 to over 800 mJ/cm2. Larval forms were generally more sensitive to UV light than species which remain as an egg in the environment. This review confirms that further research is required to produce detailed recommendations for household or community scale UV-C LED or solar disinfection (SODIS) of water for preventing helminthiases.
| Helminth infections are currently controlled by mass administration of anthelmintic drugs which are effective at treating the diseases but cannot prevent reinfection. As we work to eliminate these diseases, complimentary control methods such as improving access to water, sanitation, and hygiene will be crucial to reduce re-exposure and cut transmission. UV disinfection is a widely used form of water treatment but it is often seen as incompatible with low income regions. Recently developed UV-C LEDs and SODIS offer alternative sources of UV light that may be more suitable for this context, but there is little guidance about how we can use this technology to prevent helminth infections. We carried out a systematic review to establish which helminths are more sensitive to UV light and identify the areas which need further research. This will enable the production of design guidelines for household and community scale UV water treatment, so that the WASH community will be able to take full advantage of the recent developments and standardizations in UV disinfection technology.
| In 2016, WASH-related helminth infections (e.g. schistosomiasis, soil-transmitted helminthiases, taeniasis) were responsible for over 9.5 million years lost due to ill-health, disability or early death [1]. They are transmitted through contact with (or consumption of) water, food, and soil that contain the human infective stages of the parasite. Current control methods for combating these neglected tropical diseases (NTDs) are primarily focused on preventative chemotherapy with anthelmintic drugs, which has been effective at reducing the global health burden [2, 3]. However, reinfection is common and it is now widely recognized that an inter-sectoral approach is required for combatting many of these diseases [4–7]. In 2015 the World Health Organization (WHO) published their global strategy for WASH for NTDs, confirming that whilst WASH was one of the five key interventions in the global NTD roadmap published in 2012, little progress has been made in linking WASH and NTD programs [8]. More recently the WHO published “WASH and Health working together”, a toolkit for WASH and NTD programs based on the BEST (Behavior, Environment, Social inclusion, Treatment and care) framework [9]. Access to sanitation and clean water, and promotion of safe water practices are key interventions under the behavior and environmental components of the framework for many of the NTDs, including six helminthiases. Yet 29% of the global population do not have access to managed water supplies and 61% lack access to sanitation services [10]. Whilst piped water requires significant developments in regional infrastructure, household and community scale water treatment processes can be used to treat water collected from contaminated water bodies. This reduces exposure to helminth eggs and larvae by providing safe alternative water supplies for contact activities such as hand washing, bathing, and laundry.
UV disinfection is widely used for water and wastewater treatment in many parts of North America, Asia, and Europe. It has the benefits of forming no trihalomethanes or haloacetic acids, regulated by-products of chlorination, and can be successfully used against chlorine resistant pathogens such as Cryptosporidium parvum and Giardia lamblia [11, 12]. UV radiation is the part of the electromagnetic spectrum between 100 and 400 nm, which can be categorized into four types: UV-A (400–315 nm), UV-B (315–280 nm), UV-C or the germicidal range (280–200 nm), and Vacuum UV (200–100 nm). Unlike chlorination, UV disinfection does not necessarily kill pathogens. When a microorganism is exposed to UV light, most of the photons pass through it but some are absorbed by various cellular components. In the germicidal range, proteins and the nucleotide bases that make up DNA and RNA account for most of the absorption. Absorption by proteins is highest below 230 nm, but in this range water also strongly absorbs UV light, and high fluences are generally required for protein damage to occur. Lower fluences are required for absorption by DNA or RNA, which peaks at about 260 nm. All nucleotide bases absorb UV light, but absorption by the pyrimidine base thymine is the most critical for UV inactivation of microorganisms. When two thymine bases are adjacent to each other on a DNA chain, the absorption of a photon by one of the bases leads to a new chemical bond with the neighbouring thymine base, known as a dimer. In viruses that only contain RNA a similar reaction occurs between neighbouring uracil bases. The dimer changes the structure of the DNA or RNA and prevents the formation of new chains during replication, thereby inactivating the pathogen [13].
Conventional UV technologies use low pressure mercury filled arc lamps which produce near-monochromatic light at 253.7 nm, very close to the absorption maximum of DNA [13]. However, these lamps are made of fragile quartz and contain toxic mercury, which requires specialist handling and disposal. They also require relatively high input power and a reliable AC electricity supply. As a result, UV disinfection is often seen as incompatible with small scale water treatment in low income regions. The recent rapid development of UV-C LEDs offers an alternative source of UV light that may be more suitable for this context. UV-C LEDs are mercury free, durable, have a lower drive voltage than conventional mercury lamps, and can be powered by battery or photovoltaic supplies, so they can be used in rural or remote settings. UV-C LEDs are also much smaller than mercury lamps, which allows for novel design of water treatment systems, particularly point-of-use applications [14, 15]. The optical power of UV-C LEDs is currently relatively low, meaning devices need to be run for long periods of time to achieve sufficient inactivation of pathogens. The best wall plug efficiency (WPE, ratio of optical output power to electric input power) for a commercially available UV-C LED device is currently 4.1%, compared to 30 to 40% for low pressure mercury lamps [16, 17]. However, efficiency is improving and the WPE of commercial UV-C LED devices is expected to exceed 10% by 2021 [16]. Furthermore, in the last 15 years the cost of commercially available UV-C LEDs has decreased from over 1000 USD/mW to less than 1 USD/mW [18]. If these trends continue, and UV-C LED technology follows the path of visible LEDs, household and community scale UV disinfection of water may become a realistic option for low-income regions.
Sunlight is an alternative source of UV light and SODIS is now widely recognized as a sustainable form of small scale, e.g. household level, drinking water treatment. SODIS typically involves filling 2-liter polyethylene-terephthalate (PET) drinks bottles with water and placing them on a reflective surface for a minimum of six hours in direct sunlight (24 hours on overcast days). Pathogens are inactivated through a combination of heating and UV-A and UV-B disinfection [19].
Conventional UV disinfection and SODIS have been shown to be effective against a wide range of waterborne pathogens, but there is no guidance for their use against helminths, even though many can be spread through water. The aim of this research is to review existing literature on the UV sensitivity of WASH-related helminths, determine which helminths are more susceptible to this form of water treatment, and identify gaps in research which will inform future studies regarding the proper use of UV and SODIS for minimizing the spread of these diseases via water in low-income regions.
This systematic review follows the guidelines of the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) [20]. The search took place between 1st and 6th June 2018 and included five databases: Web of Science, PubMed, The British Library, Scopus, and Google Scholar. All languages and document types were included, and databases were searched from inception to present day. The databases were searched for any combination of species name (Table 1) and common UV disinfection terms (UV, Ultra-violet, Ultraviolet, SODIS, Sunlight, Solar Disinfection) in the title. The search was not limited to NTDs; it included all WASH-related helminths that can infect humans (including zoonotic species) as listed on the Centre for Disease Control index of Parasites of Public Health Concern [21]. This includes species that are not necessarily waterborne but that can be transmitted through water if they enter water or wastewater (such as Soil-Transmitted Helminths which are spread through faeces). Class names cestode, nematode, and trematode, and common names, such as hookworm, were also included in the search. An example search strategy can be found in S1 Supporting Information.
The studies were reviewed and classified according to the process flow diagram shown in Fig 1. First duplicates were removed and assigned code 1, then the papers were classified by title. Titles which suggested the studies were not about a relevant species or UV disinfection were removed and assigned codes 2–4 (animal species of the listed genera were included). The remaining abstracts were read and those that were about a non-waterborne life stage or primarily about the host response to a UV-attenuated vaccine were assigned codes 5 and 6, and removed. Papers for the remaining studies were obtained and read in full. Studies that provided limited information about the effect of UV light on the helminth or that contained significant errors (such as using a wavelength outside the UV range) were assigned codes 7 and 8 and were excluded; the remainder (code 9) were included in the review. Additional studies that were referenced in the papers in a way that suggested they were relevant to this review were also obtained, read in full, and assigned the relevant code.
Papers were obtained from Imperial College London Library, The British Library, and The Wellcome Trust and were read independently by the first and second authors. Papers that were not written in English were either translated using online translation software (Google Docs translation tool) or by Imperial College London students who were native speakers of the language. Notes were made on the studies and relevant information was extracted and included in a table (S1 Table). Any discrepancies between the first and second authors about which studies should be included were discussed and resolved.
Where possible, the log reduction was calculated using the inactivation data presented in the studies and the equation LogReduction=−log10(NN0), where N = proportion of viable organisms in the experimental sample and N0 = proportion of viable organisms in the control sample. If the survival percentage of the control sample was not stated in a paper, it was assumed that 100% survived (except for studies assessing the worm burden). If 100% of the experimental sample was inactivated, it was assumed that one organism survived in order to calculate the minimum log reduction; if the study only reported the percentage of organisms, then it was assumed that 1% survived. The log reduction values were then interpolated to calculate the UV fluence required to achieve a 1-log and 2-log reduction. If a 1-log reduction was not achieved in a study, then the data were extrapolated.
In total 704 papers were returned by the search, resulting in 252 individual studies, once duplicates were removed. After classifying the papers by title and abstract, 59 studies were selected to be read in full, but two were unavailable. 43 studies from the search were ultimately included in the review and four additional studies that were referenced in the original papers were added, resulting in a total of 47 studies.
Whilst 52 species of 23 genera were included in the search, results were returned for only 18 species of 10 genera: Ancylostoma spp, Angiostrongylus spp, Ascaris spp, Echinococcus spp, Fasciola spp, Hymenolepis spp, Opisthorchis spp, Schistosoma spp, Taenia spp, and Trichuris spp (Table 2).
Most studies used low pressure mercury arc lamps but other sources include: sunlight, solar simulators, fluorescent lamps emitting in the UV-A and UV-B range, medium pressure mercury lamps emitting over a broad spectrum in the UV-C range, and monochromatic excimer lamps emitting in the UV-C range. It is difficult to directly compare studies that used sunlight or simulated sunlight with mercury lamps, as sunlight contains almost no radiation in the UV-C (germicidal) range. Studies using sunlight and long wavelength sources (UV-A and UV-B) have therefore been reviewed separately to studies using UV-C sources. Where the source or wavelength was not stated in a paper it has been reviewed alongside the UV-C studies, as these are the most common.
The amount of UV light applied to a water sample is known as the fluence (mJ/cm2), which is a product of the exposure time (s) and fluence rate (mW/cm2). The protocol for calculating the fluence from low pressure mercury arc lamps in laboratory experiments was standardized only in 2003, using a bench top collimated beam apparatus. The method involves applying a series of corrections to the irradiance measured by a radiometer at the center of the beam, to account for reflection of light from the water surface, variation in irradiance over the surface area of the liquid, absorption of UV by the water column, and divergence of the “quasi-collimated” beam. Application of these factors to the measured irradiance will give the average germicidal fluence rate in the water sample. The method also requires that mercury lamps are allowed to warm up for at least 10 minutes to allow the output to stabilize and samples must be stirred during exposures to ensure all microorganisms receive the same fluence [22]. Only one study used this method to calculate the fluence, therefore the fluences stated for all the other studies should be considered approximate. Some studies only recorded the exposure time and the fluence could not be calculated.
Exposures were carried out in a number of different containers and the sample depth also varied between studies, from droplets on a glass cover slip to 25 mm deep samples in a culture dish [23, 24]. Different water matrices were also used for the exposures, for example deionized water, salt solutions, and filtered wastewater treatment plant effluent [25–32]. The sample depth and water matrix can affect the amount of UV light that is absorbed by the water; if this is not accounted for in the fluence calculation it may result in an overestimation of the average fluence in the sample. In the case of samples being exposed in very small volumes of water (e.g. droplets on coverslips), this may cause the samples to dry out, and it is difficult to separate the effect of drying from the effect of UV light on the inactivation of the target organism. Similarly, some UV sources are known to produce a considerable amount of heat, and not all experiments controlled the temperature of the samples, which may have also contributed to inactivation of the target organism.
Many microorganisms (e.g. bacteria) have the ability to reverse the damage caused by UV light through photoreactivation, dark repair, or excision repair [13]. However, the repair potential of helminths was explicitly examined in only one study and not all studies kept samples in the dark after UV exposure.
A variety of methods were used to determine the viability of helminths following exposure to UV light, the most common was to assess the ability of eggs or larvae to reach the next stage of development inside an animal host (also known as in vivo methods). In vitro methods such as assessing the motility or morphology of larvae, and the ability of eggs to embryonate in culture dishes, were also used. The most appropriate method may vary between genera. The in vivo method is often seen as the most definitive way to establish viability although it may not always produce the most reliable fluence-response curves. This is because helminths have complex lifecycles and often the number of organisms collected from a host is not directly proportional to the number of organisms in the inoculant. For example, in one study there was a considerable difference in the number of mice that developed infections depending on whether they were inoculated with 500 or 2,000 eggs (0% and 75%, respectively) and in another the number of control organisms recovered from the host varied notably between experiments (4–30%), possibly as a result of incomplete recovery or because of an unknown underlying issue which caused a reduction or increase in the intensity of infection in some of the host animals [33, 34]. Using the in vivo method is also likely to result in a higher level of inactivation than if an in vitro method is used to assess viability. This is because UV light inactivates pathogens by altering nucleic acids, and not all damage is immediately evident but can show up later in the development of the organism, which has been demonstrated in a number of studies in this review [24, 35–40]. Furthermore, as migration of eggs and larvae through the body can still pose a health risk, some authors suggest it is preferable to prevent entry to the bloodstream of the host and that it is necessary to demonstrate the organisms have been inactivated in vitro [41].
It is difficult to compare results between studies that use in vivo and in vitro methods. As the infection mechanism varies between genera, in vitro viability assessments may allow for better comparison, however further research is required to establish standardized methods for in vitro viability assessments of helminths. Selective dyes which stain only alive or dead cells may be suitable for this purpose and have previously been used to determine the viability of schistosome schistosomula [42]. One study in this review used methylene blue to identify dead schistosome cercariae which were stained violet blue, whilst live cercariae were left colorless [43].
Twelve studies investigated the effect of sunlight and long wavelength UV light (313 and 390 nm) on seven species of helminth: Ancylostoma caninum, Ancylostoma ceylanicum, Angiostrongylus cantonensis, Ascaris suum, Ascaris lumbricoides, Schistosoma mansoni, and Schistosoma haematobium. In sunlight experiments the eggs or larvae were exposed for a minimum of 15 minutes to over six hours of continuous sunlight. Some studies also investigated the effect of exposure to intermittent sunlight over a period of days. Much shorter time periods were required when using artificial UV-A and UV-B sources.
S. mansoni cercariae were the most sensitive to natural sunlight, requiring 60 minutes for all cercariae to be rendered motionless in one study, even on cloudy days [44]. Prah and James found S. mansoni and S. haematobium miracidia were equally sensitive to sunlight, however longer exposures were required than in studies using cercariae [44, 45]. This suggests sensitivity to UV light may vary between different life-stages of the same species. Similarly, Spindler found single cell A. suum were more sensitive to sunlight than embryonated eggs [46] (Table 3). A. suum was the most resistant to sunlight; in one study single cell eggs were exposed to simulated sunlight between 290 and 800 nm at a fluence rate of 55 mW/cm2 for over six hours and only a 1.42-log reduction was achieved. However, it must be noted that a fluence rate over such a broad spectrum cannot be directly compared to a fluence rate in the UV range, as not all wavelengths have the same germicidal effectiveness. Furthermore, a high concentration of approximately 1 million eggs/mL was used and the study did not consider the effect of shielding, where organisms higher in the water column may protect lower ones from UV exposure [27]. The results of this study should therefore be considered conservative (i.e. under-estimate the true sensitivity of the eggs to UV light). Jones and Hollaender investigated the effect of simulated sunlight on A. lumbricoides, using a mercury source lamp which emitted light between 350 and 490 nm at a fluence rate of 0.1 to 30 mW/cm2. In this experiment the highest inactivation achieved was a 0.98-log reduction, but the authors noted that they would expect natural sunlight to be more damaging due to the presence of infrared radiation and higher temperatures. The samples were not mixed during the exposures; an effort was made to expose the eggs in single layers but there was an issue of “clumping” in some of the experiments [47].
Two studies investigated the effect of intermittent sunlight on A. lumbricoides; both found that eggs were able to survive for much longer periods (up to 60 hours) than in other studies that used continuous sunlight [27, 46, 48–50]. However, it should be noted that these experiments were carried out in Russia (at a high latitude) whereas other studies either used solar simulators or tropical sunlight containing higher levels of UV radiation (high fluence rates), which increases with proximity to the equator. Nenow confirmed that the germicidal effect of sunlight varies with altitude, suggesting that SODIS may be a more effective form of disinfection in communities located at higher altitudes, as shorter exposure times are required [50].
Of the hookworm species, no A. caninum larvae were able to survive 180 minutes exposure to sunlight [51], and only 60 seconds exposure to UV-A (390 nm) radiation was required for larvae of A. ceylanicum to become visibly sluggish. After 30 minutes exposure to UV-A light, larvae began to lose motility completely and when hamsters were orally infected with a dose of 100 larvae, no worms were able to develop [28] (Table 4). Similarly, larvae of the nematode A. cantonensis exposed for 15 minutes to UV-A light were unable to develop inside an animal host [52]. Only one study used UV-B light, which was shown to be relatively effective against S. mansoni miracidia. There were limited details on how the fluence was measured, but 86.1 mJ/cm2 (approximately 2 minutes 30 seconds) was sufficient to achieve a 2-log reduction in the number of daughter sporocysts in snails, even though the miracidia did not appear harmed. When exposed to fluorescent white light, immediately after irradiation, the miracidia were able to photoreactivate, with significantly higher numbers of sporocysts than in snails that were kept in the dark [23].
The eggs and larvae in these experiments were exposed to sunlight or long wavelength UV in small amounts of water, with no more than 3 mL used in any of the experiments. However, SODIS is generally carried out in 2 L bottles, which are laid on their side and left in direct sunlight. The depth of the water column will therefore be much higher than in the studies included in this review, increasing the amount of UV light that is absorbed by the water. It is therefore recommended that SODIS experiments are also carried out in the containers that will be used by households and local communities.
SODIS is currently mainly used for drinking water treatment, and 2 L bottles are therefore appropriate reactors as the treated water can be drunk straight from the bottle. However, some helminthiases can be transmitted through poor personal hygiene and through contact activities such as bathing and laundry which require larger amounts of water and alternative reactors maybe required to effectively treat these volumes. Previous studies have used transparent plastic bags of various sizes as effective SODIS reactors although they have not been tested against helminths [53]. These have the advantage of a higher surface area to depth ratio whilst also allowing a larger volume of water to be treated. The underside of the bag can be coated with a reflective surface to increase the reflection of UV light into the water and they are cheap and easy to transport [54]. As the bags can be made to any size they may be more suitable for treating water for contact activities, however more research is required in this area.
Germicidal mercury lamps (253.7 nm unless stated otherwise) were used in 24 studies and an additional 12 studies used other UV-C light sources or did not specify the wavelength. There was a very large range in the inactivation data for A. suum and A. lumbricoides, with fluences from 11 to 3,367 mJ/cm2 required to achieve a 1-log reduction (Table 5). Only one study by Brownell and Nelson used the industry standard protocol to evaluate the fluence though Lucio-Forster et al. applied some factors, correcting for reflection, absorption, and divergence of the UV beam. The results of these two studies are reasonably similar with fluences of 100 and 84 mJ/cm2 required to achieve 1-log inactivation of intact single cell eggs, respectively, although the difference is greater for 2-log inactivation [41, 55].
A study by Tromba suggested A. suum is more sensitive to UV light, achieving a 2.21-log reduction at 24 mJ/cm2 [24], even though the eggs used were in a later stage of development and other studies reported that resistance to UV light increases with development stage [46, 50, 56]. However, Tromba determined viability of the eggs by assessing the worm burden in animal hosts, which may have resulted in a higher level of inactivation than if an in vitro method was used, as not all damage is immediately evident [24, 41, 57, 58]. Peng et al. found that deformities began to show two weeks after single cell eggs were exposed to UV light (unknown wavelength) for 10–20 minutes, even though during the first week of incubation development of irradiated eggs matched that of the controls. Eggs that were exposed for less than 10 minutes appeared to develop normally for longer periods, with deformities showing only after three weeks [35].
0.77-log reduction of A. lumbricoides was achieved at 20.3 mJ/cm2 using a prototype flow-through reactor, but in this study eggs were dissected from worm uteruses rather than collected from faeces or isolated from host intestines [31]. It is possible that these eggs were more sensitive to UV light because the eggshells may not have fully developed [41]. Furthermore, there was no information on how the fluence was calculated for the flow-through reactor.
One study compared the use of UV light and microwave radiation for disinfection of soil containing A. lumbricoides eggs, although experiments were also carried out in water. The authors found fluences over 3,000 mJ/cm2 were required to achieve any significant inactivation in water. Whilst some factors were applied during the fluence calculation, the collimating tube used in the experiments was 6 cm in diameter but only 10 cm long [59]. In this region the beam from mercury lamps is divergent, and radiometers can produce errors if they are used to measure the irradiance very close to the source. It is generally recommended that a collimating tube four times as long as the diameter is used with mercury sources [22]. It is therefore possible that the fluence was actually less than was stated in the paper.
Another study suggested that exposure to UV light actively increased the larval development of A. lumbricoides, even when exposed to fluences greater than 15,000 mJ/cm2 [30]. This contradicts the results of all the other studies included in this review and goes against the general understanding of the effect of UV light on microorganisms and UV disinfection. As with the previous study the irradiance was measured only 5 cm from the source and it is unclear if a collimating tube was used at all. The number of eggs in each sample was not specified and it is unclear if the suspensions were stirred. Furthermore, eggs were suspended in filtered secondary wastewater effluence, which may have absorbed a considerable amount of UV light, for which there was no correction applied. However, some inactivation would still have been expected. It is possible that some repair occurred (it is unclear if the samples were kept in the dark following exposure) or the accelerated development may be a result of the increase in temperature, which was not measured during the experiments, and has previously been shown to increase the rate of development in Ascaris spp [30].
An early study by Nolf compared two species of soil-transmitted helminth, and found that Trichuris trichiura was less sensitive to UV light (unknown wavelength) than A. lumbricoides [60]. Non-standard units were used to measure the extent of the exposure to UV light which means this paper cannot be compared to other studies, and there have been no further studies using Trichuris spp support this. The hookworm A. caninum appeared to be the most sensitive soil-transmitted helminth studied but the fluence was not recorded in these experiments. The initial log reductions achieved were relatively low, only 0.38 after five minutes exposure, however exposed larvae were not able to survive for as long as controls. Larvae exposed for five minutes did not live more than five days, whereas 52% of larvae exposed for 30 seconds were able to survive five days or more [36]. Unlike Ascaris spp and Trichuris spp, Ancylostoma spp larvae hatch from the egg in the environment, not inside the host, possibly explaining why this species is more sensitive to UV light. No studies were carried out using Ancylostoma spp eggs.
Taenia taeniaeformis eggs were very resistant to UV light, requiring 720 mJ/cm2 to achieve a 0.65-log reduction in the number of cysts recovered from the host when compared to control eggs. However, only one study investigated Taenia spp and there are few details of how the fluence was calculated. Furthermore, there was a notable difference in the number of cysts recovered from the controls in each of the experiments (4–30%) and it is unclear why this occurred. In the same study only 30 mJ/cm2 was required for 3-log reduction when the embryophore had been removed, suggesting that as with the eggshell for Ascaris spp, the embryophore is key to Taenia spp resistance to UV light [34, 41, 55].
The importance of the eggshell is less clear for other helminths in this review due to the lack of studies. A flow-through reactor with excimer lamps at 222 and 282 nm achieved a 0.92-log reduction in Opisthorchis felineus eggs in wastewater at a fluence of 25 mJ/cm2, suggesting it is relatively sensitive to UV light. However, a very small sample size was used, and it is unclear how the fluence was determined. Only two experiment samples were tested and the number of eggs in the samples prior to UV exposure was not known, only one control sample was tested to calculate the log-reductions [32]. Hymenolepis diminuta ova that were exposed to UV light for a minimum of 30 minutes were unable to develop into cysts. At 15 minutes exposure one cyst was able to develop, though this was deformed [29]. No infections developed in mice injected with 500 exposed Echinococcus granulosus eggs that had been exposed to UV light (unknown wavelength) for 24 hours. Yet, when mice were injected with 2,000 exposed eggs, 75% were able to develop infections, although significantly less eggs were able to develop into cysts than in control mice (0.15% compared to 0.7%). This was probably due to the proportion of viable embryos in each of the doses [33].
Of the trematodes, only one study used miracidia of Fasciola gigantica. There was a significant reduction in cercariae shed from snails when they were infected with one exposed miracidium, compared to one control miracidium, even at very short exposure times less than 70 seconds [61]. The effect of UV light on Schistosoma spp has been the most widely studied and it is the most sensitive helminth in this review. The majority of papers were immunization studies, investigating the use of UV-attenuated cercariae to produce a vaccine against human schistosomiasis. In these experiments cercariae were exposed to a fluence high enough to cause damage and prevent development into adult worms, but that still allowed penetration of the host’s skin. The focus was therefore on the worm burden, and details of the exposure methods were often limited. In the immunization papers a 1-log reduction in worm burden was achieved with fluences of 5–14 mJ/cm2 [25, 26, 43, 62–68] or exposure times of less than one minute [39]. The direct effect of UV light on cercariae was first studied by Krakower in 1940 who found 45 minutes exposure to a mercury lamp (unknown wavelength) was required to kill the whole sample. Shorter exposures were still able to cause damage, making the cercariae less motile than the control samples, though they were able to recover from their injuries within 30 minutes and survived for as long as the controls, suggesting schistosome cercariae have some repair potential [44]. Standen and Fuller found that only four minutes was required to kill 100% of S. mansoni cercariae in their study, but the mercury lamp used was very near to the sample (2-cm) and it is unclear if the authors controlled the water temperature [69]. Older mercury lamps are known to have produced a lot of heat and cercariae are inactivated within minutes at 45°C and almost instantly at temperatures above 50°C [70].
Ghandour and Webbe studied the effect of UV light on the ability of S. mansoni and S. haematobium cercariae to penetrate skin. There was a significant increase in mortality during skin penetration when cercariae were exposed for 5–20 seconds, even though they did not appear to be harmed. 10–11% of exposed cercariae were unable to penetrate at all compared to 2–3% of the control sample [37, 38]. Another study found that short exposure times caused a reduction in the motility of cercariae, but this only became apparent four hours after exposure [39]. Cercariae penetrate skin through enzyme activity and mechanical action, a combination of motility reduction and inhibition of enzymes may have prevented cercarial penetration.
Two studies used scanning electron microscopy (SEM) to examine the physical damage caused by UV light to S. mansoni. Mohamed showed that adult worms developed from irradiated cercariae had lost their spikes and suffered from torn tubercles and lesions, causing sexual anomalies and sterility, possibly explaining the reduction in fecundity of worms derived from irradiated cercariae in other studies [39, 64, 71]. Later Dajem & Mostafa used SEM to examine the damage on the surface of cercariae and discovered that irradiated samples appeared to be physically the same as control cercariae, suggesting the damage observed in adult worms is either a result of mutagenic effects of UV light which only appear later in development, or as a result of the hosts immune response to irradiated cercariae [40]. Another study found that UV exposure modified the structure of molecules on the surface of S. mansoni cercariae, even though no morphological changes occurred. This may have caused an enhanced immune response by the host [72].
Significantly more male S. mansoni worms were able to develop from irradiated cercariae in one study, suggesting that males may be more resistant to UV light than females [73]. This also may explain the reduction in fecundity observed in other studies, but further research is required to confirm this [39, 64]. S. mansoni and S. japonicum were shown to be equally sensitive to UV light, suggesting the inactivation mechanism is the same in both species [66]. Only one study used S. haematobium cercariae, which were found to be slightly more resistant that S. mansoni cercariae, however no statistical analysis was performed [38]. Prah and James found there was no difference in the response of S. mansoni and S. haematobium miracidia to UV light from mercury source lamps. Experiments in this study were repeated with 1% turbid water, and a 15.4% reduction in the rate of movement of miracidia was observed, compared to a 60.3% reduction when distilled water was used [45]. However, it should also be noted that distilled water has been shown to kill schistosome cercariae, and it may have a similar effect on miracidia [74]. Whilst many different suspension media have been used, this was the only study in the review that investigated the impact of turbidity or the water matrix on UV disinfection. If UV or solar disinfection is to be used effectively for household and community scale water treatment this aspect requires further research, preferably using water samples collected from the environment. If water collected from local waterbodies is of particularly poor quality (e.g. high turbidity, iron, or organic matter content), consideration may need to be given to pre-treatment, such as filtration or sedimentation.
With the recent introduction of UV-C LED technology into the water sector, UV disinfection could be a realistic option for sustainable water treatment in low-income regions in the near future, to provide safe water supplies for water contact activities such as bathing, laundry, and to improve hygiene. Compared to bacterial and viral pathogens there has been little research into the effectiveness of UV light at inactivating helminth eggs or larvae, which are endemic to many developing countries. The majority of studies in this review investigated the effect of UV light on either Schistosoma spp or Ascaris spp, and many were immunization studies used for developing UV-attenuated vaccines, with a focus on the host response to irradiated larvae or eggs, not complete inactivation of the target organism or applications to water treatment.
There were limitations to almost all of the studies, the most significant being the lack of a standardized procedure for calculating the UV fluence to which samples were exposed. 68% of studies were carried out before the industry standard protocol for fluence measurement was published in 2003 [22]. In the SODIS studies, experiments were carried out using very small amounts of water which is not representative of how disinfection will take place in practice. Very few studies considered the impact of water quality or accounted for the absorbance of UV by the water column or the effect of shielding caused by suspended particles and other organisms. Mercury lamps are known to produce considerable amounts of heat and it is not clear which of the studies controlled the water temperature. In some studies the fluence was not recorded at all and only one study investigated the repair potential. The methods for determining the viability of larvae or eggs varied, even between papers using the same genera, and this resulted in large ranges in the fluence response, most notably for Ascaris spp. Furthermore, the survival percentage of control samples was not stated in all studies and assumptions were made to calculate the log reductions presented in this review.
These limitations make it difficult to directly compare the studies, however some conclusions can be drawn. All helminths included in this review could be inactivated by UV light at certain fluences and wavelengths, but the number of species studied was limited. Helminths which hatch from the egg in the environment were generally more sensitive to UV light than species which stayed in the egg until after they had infected the host. Studies found that eggs were much more sensitive to UV when the shell or embryophore had been removed, suggesting they play a key role in the resistance to UV light for some species. Fluences in excess of 80 mJ/cm2 were required to achieve a 1-log inactivation of Ascaris spp and Taenia spp eggs, over twice the current minimum fluence required by some European countries for the treatment of publicly supplied drinking water [75, 76]. UV disinfection may therefore not be the most efficient form of water treatment for these helminths. UV disinfection may be particularly effective against Schistosoma spp which was consistently the most sensitive to UV light in this review, however further experimental research is required using the standard fluence measurement protocol.
This systematic review has demonstrated that evidence exists to suggest that UV disinfection is effective against some helminths, but the data covers a limited number of species and is insufficient to produce detailed recommendations for household or community scale UV or solar disinfection of water in endemic regions. To aid the design of these water treatment systems we recommend the following for future studies on UV disinfection of WASH-related helminths:
|
10.1371/journal.pntd.0002989 | Presence of Antigen-Experienced T Cells with Low Grade of Differentiation and Proliferative Potential in Chronic Chagas Disease Myocarditis | The main consequence of chronic Trypanosoma cruzi infection is the development of myocarditis in approximately 20–30% of infected individuals but not until 10–20 years after the initial infection. We have previously shown that circulating interferon-γ-secreting T cells responsive to Trypanosoma cruzi antigens in chronic Chagas disease patients display a low grade of differentiation and the frequency of these T lymphocytes decreases along with the severity of heart disease. This study thought to explore the expression of inhibitory receptors, transcription factors of type 1 or regulatory T cells, and markers of T cell differentiation, immunosenescence or active cell cycle in cardiac explants from patients with advanced Chagas disease myocarditis.
The expression of different markers for T and B cells as well as for macrophages was evaluated by immunohistochemistry and immunofluorescence techniques in cardiac explants from patients with advanced chronic Chagas disease submitted to heart transplantation. Most infiltrating cells displayed markers of antigen-experienced T cells (CD3+, CD4+, CD8+, CD45RO+) with a low grade of differentiation (CD27+, CD57−, CD45RA−, PD-1−). A skewed T helper1/T cytotoxic 1 profile was supported by the expression of T-bet; whereas FOXP3+ cells were scarce and located only in areas of severe myocarditis. In addition, a significant proliferative capacity of CD3+ T cells, assessed by Ki67 staining, was found.
The quality of T cell responses and immunoregulatory mechanisms might determine the pattern of the cellular response and the severity of disease in chronic Trypanosoma cruzi infection.
| Chagas disease is a neglected tropical disease affecting approximately 10 million people in the world. It is caused by infection with the protozoan Trypanosoma cruzi. As a consequence of migration flows, the disease has been also become established in non-endemic countries. In this study, the functional and phenotypic profile of inflammatory T cells were evaluated in heart tissues of patients with end-stage chronic Chagas disease by analyzing the expression of inhibitory receptors, transcription factors of type 1 or regulatory T cells, and markers of T cell differentiation, immunosenescence or active cell cycle. Most infiltrating cells displayed markers of antigen-experienced T cells with a low grade of differentiation and a significant proliferative capacity. A skewed T helper1/T cytotoxic 1 profile was supported by the expression of T-bet; whereas FOXP3+ cells were scarce. The quality of T cell responses and immunoregulatory mechanisms might determine the pattern of the cellular response and the severity of disease in Trypanosoma cruzi infection.
| The main consequence of chronic Trypanosoma cruzi (T. cruzi) infection is the development of myocarditis in approximately 20–30% of infected individuals but not until 10–20 years after the initial infection [1]. Advanced chronic Chagas heart disease (cChHD) is characterized by dilated cavities with high degree of fibrosis and inflammation [2], [3]. The analysis by immunohistochemical, but mainly by molecular techniques, of cardiac samples from chronically T. cruzi-infected subjects provided evidence of the association between parasite persistence and tissue damage in cChHD [2], [4], [5]. Reis et al. showed that inflammatory lesions were dominated by CD8+ lymphocytes, many of which expressed granzyme A [6]. Lymphocytes in these lesions express lymphocyte function antigen-a (LFA-1), CD44, very late antigen-4 (VLA-4) [7] and cytotoxic lymphocyte antigen 4 (CTLA-4) [8]. A Th1 cytokine pattern predominated in the cardiac inflammatory cell infiltrate of Chagas disease patients with heart failure [9].Whereas some authors have shown increased peripheral levels of IFN-γ in patients with severe heart disease [10]–[12], other studies have demonstrated an inverse association between disease severity and IFN-γ production [8], [13], [14].We have previously shown that most IFN-γ-secreting T cells in response to T. cruzi display a low grade of differentiation but high expression of the inhibitory receptor CTLA-4 in the circulation of subjects with chronic T. cruzi infection [8], [15], [16]. Conversely, the total T cell compartment in Chagas disease patients is enriched in highly differentiated T cells compared to uninfected controls [15]–[17].
There is very limited data available on the degree of differentiation of T cells in heart lesions of cChHD, and a comprehensive analysis of the relationship of different T cell functions in Chagas disease myocarditis has not been performed. This study thought to explore the expression of inhibitory receptors, transcription factors of type 1 or regulatory T cells and markers of T cell differentiation, immunosenescence or active cell cycle in cardiac explants from patients with advanced cChHD submitted to heart transplantation.
This study was approved by the Institutional Review Boards of the Hospital Universitario Fundación Favaloro (UIC (863) 1604), and all patients gave written informed consent for the heart transplant procedure.
Eight patients with severe cChHD (4 men and 4 women; mean age ± SD, 51.4±7.3 years) were admitted at Hospital Universitario Fundación Favaloro in Buenos Aires, Argentina, during the period 1998–2008 to undergo orthotopic heart transplantation. Diagnosis of T. cruzi infection was confirmed when at least 2 out of 3 standard serological tests (enzyme-linked immunosorbent assay, indirect hemagglutination and immunofluorescence).were positive. Determination of cChHD was based on clinical, echocardiographic and electrocardiographic findings. Explanted hearts from patients with Giant cell myocarditis (GCM; n = 2) and idiopathic dilated cardiomyopathy (IDCM; n = 1) were also included as controls. Human lymph node and placental tissues from the Tissue Bank of the Pathology Lab were employed as positive staining controls. The cChHD patients included in this study had not received immunosuppressive drugs, trypanocidal therapy or prophylactic benznidazole by the time of this study.
Eight explanted hearts were weighed and fixed for 72 h in 10% phosphate-buffered formaldehyde. After fixation, transmural sections of the whole circumference of the left and right ventricle at a plane equidistant from the base to the apex were collected and embedded in paraffin. A 5-mm-thick section from each region was stained with hematoxylin and eosin and Masson's trichrome solution. The interventricular septum of each heart was selected for histological and immunohistochemistry analysis. The diagnosis of myocarditis was defined according to the Dallas criteria taking into account the inflammatory infiltrate of the myocardium and the presence of necrosis and/or degeneration of adjacent myocytes [18]. The distribution of the inflammatory infiltrate was classified as focal, confluent or diffuse [19].
The median number of lymphocytes plus macrophages identified by the expression of CD3 and CD68, respectively, was calculated for 8 tissue samples from cChHD, 2 GCM samples and 1 IDCM sample assessed, as described in “Quantification of cells” [20]. Myocarditis recorded in each tissue sample was considered as severe when the number of lymphocytes plus macrophages was over the median number of these cell types in cChHD, moderate when the number of lymphocytes plus macrophages was between 25th and 50th percentile, and mild when the number of lymphocytes plus macrophages was under percentile 25th.
For quantitative assessment of fibrosis an interventricular septum block that was embedded in paraffin and sectioned at 5 µm was stained with picrosirius red. After obtaining digital images with a digital scanner (UMAX Technologies Inc., USA) at 2× magnification and a 1200 ppp resolution, the percentage of the surface area occupied by collagen was established by morphometric analysis using the digital analysis system Image Pro Plus 4.5 (Media Cybernetics, Silver Spring, USA) [21]. The percentage of fibrosis was semi-quantified as mild (<10%), moderate (10%–20%), or severe (>20%) [22]. Molecular detection of T. cruzi in the same interventricular tissue sections by PCR was previously performed [23]. The presence of T. cruzi was also analyzed by direct observation of intracellular amastigotes.
Formalin fixed paraffin embedded tissue sections were rehydrated. Heat induced antigen retrieval, incubation time and antibody concentrations were selected following the manufacturer's recommendations. For CD3 (Mouse monoclonal, Santa Cruz Biotechnology, USA), CD8 (Leica Microsystems, Germany), CD68 (Biogenex, USA), CD20 (Biogenex, USA), CD45RA (Biogenex, USA), CD21 (Biogenex, USA), PD-1 (Abcam plc, UK), CD27 (Leica Microsystems, Germany), Ki67 (Rabbit monoclonal from Abcam plc, UK, and mouse monoclonal HLA-G from Leica Microsystems, Germany) and T-bet (BD, Biosciences, USA), heat induced antigen retrieval was done using Antigen Retrieval Citra Plus (Citrate buffer buffer based Ag retrieval solution pH = 6; Biogenex, USA). Antigen retrieval solution EZ-AR2™ (EDTA based retrieval buffer pH = 9; Biogenex, USA) was used to detect CD4 (Leica Microsystems, Germany) and FOXP3 (BD, Biosciences, USA) expression. For CD57 (BD, Biosciences, USA) and CD45RO expression (Biogenex, USA), no antigen retrieval after rehydration was done. Biotinylated anti mouse immunoglobulin G, peroxidase labeled streptavidin, and AEC (3-amino-9-ethyl carbazole) as chromogen were used as secondary detection system (Biogenex, USA). All immunohistochemistry slides were counterstained with hematoxylin. The list of antibodies and clones used, as well the function of each marker are depicted in Table S1 [24]–[37].
Tissue sections with high inflammation from patients 1 to 4 were selected for double labeling studies by immunofluorescence. Double labeling assays were carried out by staining with a combination of anti-Ki67 (rabbit monoclonal antibody)/anti-CD8; anti-KI67/anti-CD20 (mouse mAbs), anti-Ki67 (mouse mAb)/anti-CD3 (rabbit polyclonal) and anti-Ki67/anti-CD21 (rabbit monoclonal) (Table 1). Fluorescein labeled anti-rabbit goat immunoglobulin (Vector, USA) and Alexa fluor 594 labeled anti-mouse immunoglobulin (Invitrogen, USA) were used as secondary detection system. Nuclei staining were done with ready to use mounting medium for fluorescence with DAPI (Vectashield, Vector, USA). Antibody dilutions were used according to the manufacturer's instructions. Observations were made with a 100 W ultraviolet lamp and photographed with an AXIOCAM camera (Carl Zeiss AG, Germany).
For quantification of the total number of mononuclear inflammatory cells (i.e. cells with positive and negative staining for each marker assessed), 10 High Power Field (HPF) at 400× were counted for each section (1 section for each one of the 13 markers assessed). The median number of mononuclear inflammatory cells in 130 HPF was calculated for each patient.
The total number of mononuclear inflammatory cells with positive staining for each marker assessed was counted in 10 HPF, Magnification 400×. The percentage of cells expressing each marker was calculated by the ratio between the median number of positive cells for each marker and the median number of total mononuclear cell count in 10 HPF.
Cell counting was manually implemented using the Cell Counter plug-in of Image J 1.45b software (National Institutes of Health, USA). The position of the cells in each HPF was determined by Cell Counter plug-in and utilized thereafter for spatial histological analysis. This analysis was carried out in each one of the eight samples from Chagas disease patients, as well as in control samples.
Continuous variables are reported as means (SDs) or medians (interquartile range –IQR-), while categorical variables are presented as the percentage of subjects out of total subjects evaluated. Continuous variables with non-Gaussian distribution were analyzed by Mann-Whitney U test. Correlation analysis between the percentage of cells positive for each marker and the total number of mononuclear inflammatory cells was done using the Spearman correlation test.
In order to estimate the spatial pattern of infiltrating cells surrounding a T. cruzi –infected cardiomyocyte, a spatial point pattern analysis was applied. A focal analysis of the distribution of CD45RO+ cells around an infected cardiomyocyte was performed by the bivariate Wiegand-Moloney O-ring statistic [38]. This test allows the characterization of a spatial pattern around a point at varying distances to the point, detecting aggregation, repulsion or randomness. The analysis was implemented with software Programita, using a random labeling null model. Confidence envelopes were calculated using 999 Montecarlo simulations [39]. A p value<0.05 (2-tailed) was considered statistically significant. For statistical analysis SPSS 11.0 statistical software (SPSS Inc, USA) was used.
The clinical and demographical characteristics of cChHD patients are depicted in Table 2. Patients 1–5 and 7–8 were in end-stage cChHD, whereas patient 6 displayed cChHD and concomitant valvular heart disease. This patient suffered from rheumatic fever during her childhood, but no evidence of active endocarditis was observed by the time of the study. All patients were in the New York Heart Association classes III and IV by the time of transplantation. Echocardiographic studies revealed moderate to severe dilation of the cavities with a mean left ventricular end-diastolic diameter of 67.0±10.2 mm. The mean left ventricular ejection fraction, determined by radionuclide ventriculography, was 19.2±7.1%.
The mean weight of cChHD explanted hearts was 450±65 g. Four∶8 cChHD interventricular septum samples showed severe diffuse myocarditis, 2∶8 moderate myocarditis and 2∶8 had mild focal myocarditis. Interstitial fibrosis was moderate to severe in all except for one interventricular septum sample. Amastigote nests were only recorded in 3 sections with severe diffuse myocarditis of patient 1 (Table 1, Figure 1A). However, the presence of T. cruzi by PCR in the interventricular septum of patients 1–5 and 7–8 but not in patient 6 was demonstrated in a previous report [23]. The mean weight of hearts with GCM was 375 g and exhibited severe diffuse myocarditis and severe fibrosis. The IDCM heart weighted 510 g, presenting moderate fibrosis and absence of myocarditis (Table 1).
Inflammatory mononuclear cell infiltrates in heart tissue samples from cChHD were evaluated by immunohistochemistry- As observed in Figure 1A and Table 2, patients 1 to 4 exhibited severe diffuse inflammatory infiltrate, while patients 5 to 8 presented lower number of total infiltrating cells within the interventricular septum. The percentage of T (CD3+, CD4+, CD8+) and B cells (CD20+) was correlated with the total cell count in inflammatory cell infiltrates (Figure 1B and 1C). Conversely, the percentage of macrophages/dendritic cells (CD68+ cells) in regions with the highest levels of infiltration was similar to those observed in areas with low inflammation (Figure 1B and 1C). In 7 out of the 8 cChHD, CD8+ T cells were prevalent among infiltrating mononuclear cells compared to CD4+, CD68+ and CD20+ cells (Table 2).
Since T cells were the prevalent cell population in heart tissues with high degree of inflammation, the expression of markers of antigen-experienced T cells and T cell differentiation (Table S1) was evaluated in heart samples. A high percentage (>30%) of infiltrating cells in heart tissues from cChHD with severe myocarditis express CD45RO, a marker of antigen-experienced T cells (Figure 2A, patients 1–4 in Table 2). Likewise, CD27 expression, a marker of low grade T cell differentiation was higher in cChHD with severe myocarditis than in patients with lower degree of inflammation (Figure 2B, patients 1–4 vs. patients 5–8 in Table 2). In contrast, the expression of the inhibitory receptor PD-1, the immunosenescence marker CD57 and CD45RA (i.e. expressed by naïve and terminally differentiated T cells; Table S1), was generally very low in heart tissues from cChHD (Figure 2C, 2D, Figure S1, Table 2). Since CD57 is expressed in terminally differentiated effector T cells and also in NK, the low expression of CD57 confirms the scarcity of NK in heart tissue samples from chronic Chagas disease patients [3]. PD-1 expression was detected in intramyocardial lymphoid follicles (tertiary lymphoid follicles) in the heart of patients with severe Chagas myocarditis (i.e. patients 1–4) [Figure 2C and 2F, left panel]. Of note, tertiary lymphoid follicles were randomly observed in myocardial tissues, as confirmed by the expression of CD21 and PD-1. These findings indicate that antigen-experienced T cells with low grade of differentiation are abundant in Chagas disease myocarditis.
To better characterize the phenotype of the cell infiltrate recruited by infected myocytes, tissue samples bearing amastigotes nests along with high inflammation were selected and analyzed for the expression of CD45RO, CD45RA, CD57, CD68 and CD20 (Figure S1). Most cells surrounding the infected cardiomyocyte were CD45RO+, CD45RA−, CD57−, CD20−; while a lower number expressed CD68+ (Figure S1A and S1B).
With the aim to evaluate whether the different cell types were recruited with the same efficacy towards the infected cardiomyocyte, the spatial pattern of CD45RO+ and CD45RO− cells was analyzed by using the O-ring test for 2-dimensional point patterns. A statistically significant aggregation of CD45RO+ cells (blue line in Figure S1D) was observed near the infected cardiomyocyte (i.e. CD45RO+ cells are over the confidence envelope) [Figure S1C and S1D, P<0.001]. The proportion of CD45RO+ cells decreases at longer distances from the infected cell.
The functional profile of the cell infiltrate was assessed by the expression of T-bet, a marker of type 1 T cell responses, the regulatory molecules FOXP3 and HLA-G, and Ki67, a marker of proliferative potential (Table S1). T-bet expression was broadly detected in cell nuclei in areas of high cell infiltrate and was correlated with the degree of cell infiltration (P<0.001, rho = 0.669) (Figure 3A; Table 2). In contrast, HLA-G+ and FOXP3+ cells were found in a very low proportion or were absent in cChHD (Figure 3B, Table 2). Cells with proliferative potential were observed in tissue samples from patients with severe myocarditis (median percentage of proliferating cells = 9%) in areas with high inflammation (Figure 3C, Table 2 and Table 3). In order to identify the proliferating lymphocyte subsets, double immunofluorescence staining was carried out using Ki67, CD3, CD8 and CD21 antibodies (Figure 4). Most cells expressing Ki67+ were CD3+ cells [median percentage (interquartile range) = 74 (66–89)], while a minor proportion of Ki67+ cells expressed CD8 [median percentage (interquartile range) = 21 (15–19)], (Figure 4A, 4B and 4C). In contrast, a very low proportion of Ki67+ cells expressed CD21 (data not shown), suggesting that KI67+/CD21− proliferating cells belong to the T cell compartment.
To assess whether the phenotypic and functional profiles found in Chagas disease myocarditis were associated with parasite persistence, the same set of markers were assessed in heart tissues from patients suffering from idiopathic GCM (patient A and B, Table 2), and compared to the profile in cChHD with severe myocarditis (patients 1–4, Table 2). GCM is characterized by an intense myocarditis with multifocal cardiomyocyte damage. Although T lymphocytes were a major cell type in both severe cChHD and GCM, the latter was enriched in macrophages/dendritic cells and CD4+ T cells (Table 3), whereas CD8 was prevalent in cChHD with severe myocarditis.
Regarding the differentiation status of the cell infiltrate in these two types of myocarditis, no differences were found in PD-1 and CD45RO expression, whereas CD57 expression was low in both types of myocarditis. cChHD with severe myocarditis presented lower levels of CD27+ cells (i.e. 2/4 patients with severe myocarditis in Table 2) compared to GCM (Table 2 and Table 3). Higher counts of Ki67+ and T-bet+ cells were found in cChHD as compared with GCM, whereas the expression of FOXP3 was very low in both myocarditis. HLA-G expression was only recorded in cardiomyocytes and mononuclear inflammatory cells from GCM (Table 2 and Table 3). Altogether, these findings support the notion that chronic infection with T. cruzi drives T cell differentiation and proliferation in chronic Chagas disease myocarditis.
Most studies concerning the characterization of inflammatory cells in heart tissues in chronic Chagas disease have comprised the evaluation of different cell types, and the cytokine profile, while the differentiation status of T cells is less known. Herein, we report that antigen-experienced (CD45RO+) T cells with a low degree of differentiation (CD27+/CD57−), a Th1 profile (Tbet+) and proliferative capacity (Ki67+) are recruited into the heart of patients with chronic advanced Chagas disease myocarditis. As previously described, the inflammatory cell infiltrate in most patients was dominated by CD8+ T cells [2], [3], [6]. Although less prominent than CD8+ T cells, appreciable counts of CD4+ T cells were also found in intracardiac infiltrates.
The T cell phenotype in heart tissues concurs with that observed in circulating CD8+ and CD4+ T cells responsive to T. cruzi antigens in patients with chronic Chagas disease, regardless the clinical status [15], [16], but it is at odds with the phenotype of total peripheral CD4+ and CD8+ T cells [15]–[17], [Table S2, [8], [10]–[13], [15]–[17], [40]–[43]. Whereas most circulating IFN-γ-producing CD8+ and CD4+ T cells responsive to T. cruzi display a low degree of differentiation (CD27+/CD28+/CD57−/LIR-1−) [8], [16], high differentiated (CD27−/CD28−/CD57+/LIR-1+) CD8+ and CD4+ T cells are increased in the total peripheral T cell compartment of patients with severe cardiomyopathy [8], [15]–[17]. Subjects with chronic T. cruzi infection with severe cardiomyopathy also displayed lower frequencies of T. cruzi-responsive IFN-γ-producing T cells and lower levels of IFN-γ production compared to subjects with no signs of heart disease [8], [13], [16], leading us to propose that long-term parasite persistent might dampen parasite specific T cell responses [44]. Thus, recently developed effector T cells that bear a low degree of differentiation appear to be an important source of T. cruzi-specific T cells in the periphery of chronic Chagas disease patients. In contrast, the bulk of the total CD8+ and CD4+ T cells might reflect the effect of persistent exposure to the parasite.
The low degree of differentiation of T cells found in the heart of chronically T. cruzi-infected subjects was somehow unexpected, since T cells at target tissues are more likely to be stimulated by antigen and induced to further differentiation. However, CD27− T cells also comprised a substantial fraction of the inflammatory infiltrate, indicating a heterogeneous composition of T cells with different degree of differentiation in heart tissues. The presence of CD27+/CD57− cells might account for proliferating CD3+ T cells, as proliferation is a hallmark of T cells with low grade of differentiation (CD27+ cells) and low rounds of antigen stimulation (CD57− T cells) [45].
We also demonstrate that antigen-experienced (CD45RO+) T cells were more efficiently recruited than other mononuclear cells towards the infected cardiomyocyte, supporting that T. cruzi is a driving force for T cell recruitment. Activated VLA-4+/LFA-1+/granzyme A+ T lymphocytes have been observed in cardiac infiltrates of patients with chagasic heart failure [7], while the local production of IL-7 and IL-15 was claimed to be associated with the maintenance and predominance of CD8+ T cells in heart tissues [46]. However, the recruitment of early effector T cells might be also involved in the maintenance of inflammatory cells at sites of chronic localized infection.
Tertiary lymphoid organs associated to severe myocarditis might have been developed in the context of chronic inflammatory conditions [47]–[49]. Since the formation of TLO involves the recruitment of lymphocytes not normally associated with inflammatory infiltrates, notably naive T cells and central memory T cells, these structures might be a source of recently recruited effector T cells from the naïve or central memory pool. To the best of our knowledge, there are not reports describing these structures in chronic chagasic cardiomyopathy. TNF-α has been pointed out as a key molecule in the generation of ectopic TLO which might facilitate the perpetuation of inflammation in the heart. Cells expressing TNF-α and IFN-γ have been identified in heart tissues from patients with chronic chagasic cardiomyopathy [9], [50]. Our findings showing T-bet expression in areas of high inflammation and proliferation further confirmed a type 1 T cell response in patients with severe heart disease. The low levels of PD-1 counts concur with the presence of T-bet+ cells, since T bet inhibits PD-1 expression [51].
Recent reports have demonstrated that regulatory T cells (FOXP3+) are absent in Chagas disease myocarditis, indicating that deficiency in IL-10 producing T regs may led to a regulatory imbalance, perhaps rendering heart tissues increasingly susceptible to type 1-dependent pathology [9], [52]. In line with these findings, scarce FOXP3+ cells were found in heart tissues from patients with severe heart disease. Nevertheless, we have shown that CD3+ T lymphocytes infiltrating heart tissues express CTLA-4, another negative regulator of effector T cell responses [8]. In addition to be involved in the regulatory T cell function of T regs, CTLA-4 is also upregulated on activated T cells [53]. This might explain the concomitant presence of proliferating and CTLA-4+ T cells. However, it is likely that proliferating T lymphocytes do not CTLA-4 or likely express dysfunctional CTLA-4.
The predominant role for T. cruzi in the differentiation and functional T cell profile observed in the heart of subjects with chronic T. cruzi infection becomes apparent at comparing the inflammatory cell infiltrates between chagasic and GCM myocarditis which is a rare and frequently fatal type of myocarditis with features of autoimmune disease [54], [55]. Higher levels of infiltrating cells with low degree of differentiation (i.e. CD27− expressing cells), lower T-bet expression and lower proliferative capacity were observed in GCM compared with severe Chagas disease myocarditis. Of note, HLA-G, a non-classical class I major histocompatibility complex molecule playing a tolerogenic role in innate and adaptive responses [56] was not expressed in inflammatory cell infiltrates of chronic Chagas disease patients, while it was expressed by cardiomyocytes in GCM. These findings support that immunoregulation is different between an infectious and a non-infectious myocarditis. Coxsackievirus infection is another common cause of infectious myocarditis, in which a pathogen driven inflammatory process was supported by the reduction in inflammation and resolution of the infection after treatment with IFN-beta [57], [58].
The development of tissue damage might depend on parasite burden; the effectiveness of the host immune response in controlling parasite replication, and the effectiveness of the host immune response in limiting peripheral damage. The effectiveness of the host immune response in controlling parasite replication without the induction of tissue damage will largely rely on the quality of T cell responses [59]. We have recently reported that children in early stages of T. cruzi infection maintain polyfunctional and stronger T cell responses to T. cruzi in their circulation, contrasting with the prevalent monofunctional T cell profile in long-term T. cruzi-infected adults [60]. Although, we cannot rule out that recruitment of IFN-γ-producing T cells into the heart might account for the decrease of IFN-γ-producing T cells from the periphery, it is also possible that clonal exhaustion occurs overtime. Consequently, in the phase of an ineffective T cell response, the development of tissue damage might occur in order to get parasites under control. Similarly, immunoregulatory pathways might be also dampened overtime.
In summary the quality of T cell responses and immunoregulatory mechanisms might determine the pattern of the cellular response and the severity of disease in chronic T. cruzi infection.
The main limitation of the study is the sample size. However, the detailed analysis conducted and the difficulty to get tissue samples can mitigate, at least partially, such limitation.
|
10.1371/journal.pgen.1002122 | Cooperative and Antagonistic Contributions of Two Heterochromatin Proteins to Transcriptional Regulation of the Drosophila Sex Determination Decision | Eukaryotic nuclei contain regions of differentially staining chromatin (heterochromatin), which remain condensed throughout the cell cycle and are largely transcriptionally silent. RNAi knockdown of the highly conserved heterochromatin protein HP1 in Drosophila was previously shown to preferentially reduce male viability. Here we report a similar phenotype for the telomeric partner of HP1, HOAP, and roles for both proteins in regulating the Drosophila sex determination pathway. Specifically, these proteins regulate the critical decision in this pathway, firing of the establishment promoter of the masterswitch gene, Sex-lethal (Sxl). Female-specific activation of this promoter, SxlPe, is essential to females, as it provides SXL protein to initiate the productive female-specific splicing of later Sxl transcripts, which are transcribed from the maintenance promoter (SxlPm) in both sexes. HOAP mutants show inappropriate SxlPe firing in males and the concomitant inappropriate splicing of SxlPm-derived transcripts, while females show premature firing of SxlPe. HP1 mutants, by contrast, display SxlPm splicing defects in both sexes. Chromatin immunoprecipitation assays show both proteins are associated with SxlPe sequences. In embryos from HP1 mutant mothers and Sxl mutant fathers, female viability and RNA polymerase II recruitment to SxlPe are severely compromised. Our genetic and biochemical assays indicate a repressing activity for HOAP and both activating and repressing roles for HP1 at SxlPe.
| Eukaryotic genomes are organized into two distinct classes of chromatin, euchromatin and heterochromatin. The former is less condensed to enable transcription, whereas heterochromatin, which is marked by Heterochromatin Protein 1 (HP1), remains compact and mostly transcriptionally silent throughout the cell cycle. The viability of Drosophila males is known to be preferentially compromised in mutants for HP1 and some HP1-associated proteins, suggesting more generalized roles for these proteins in sex-specific gene expression. We now describe a male viability defect for the telomeric partner of HP1, HOAP, and misregulation of the sex determination pathway. Key to the sex determination process is the activation of the X chromosome dose sensing promoter of Sex-lethal, SxlPe. We provide genetic and biochemical evidence that HOAP, the telomere binding partner of HP1, has repressing activity; while HP1 has both activating and repressing roles at this critical promoter. Chromatin immunoprecipitation assays show both proteins are associated with SxlPe sequences. Additionally, RNA polymerase II association with SxlPe shows a requirement for HP1, suggesting a transcription initiation role for HP1. Combined, our data implicate HP1 and HOAP at a euchromatic gene, functioning in a developmental context, and provide the first evidence for a non-telomeric function for HOAP.
| Eukaryotic genomes are organized into two distinct classes of chromatin [1]. The major class (euchromatin) can undergo decondensation to enable transcription during interphase, whereas a minor fraction (heterochromatin) remains compact and mostly transcriptionally silent throughout the cell cycle. Pericentric and telomeric regions of chromosomes from fungi to humans are organized into a constitutive form of heterochromatin, marked by heterochromatin protein 1 (HP1a in Drosophila [2] and humans [3], Swi6 in S. pombe [4]) and lysine 9-methylated histone H3 (MeK9H3) [5], [6]. Lysine 9 methylation of histone H3 is catalyzed by the Drosophila SU(VAR)3-9 protein [7] (human SUV39H1 [8], S. pombe clr4 [9]) and provides a chromatin-binding site for HP1. Both heterochromatin marks have also been observed in euchromatic genes [10], where their roles in gene activation, as well as repression, have recently been uncovered [11]–[13].
The question of how Drosophila HP1a (designated HP1 throughout this text) is targeted to specific chromosomal regions prompted our biochemical characterizations of HP1 complexes in the maternally loaded cytoplasm of early embryos [14], [15]. HP1/ORC-Associated Protein (HOAP) was identified as a component of a complex that also contains Drosophila origin recognition complex (ORC) subunits [16]. Similarity of the HOAP N-terminus to the HMG-box of mammalian SRY (sex-determining region of the Y chromosome) proteins suggested a role for its DNA-binding activity and that of the ORC, in targeting HP1 to constitutive heterochromatin. Recent data in Drosophila and S. pombe point to a role for small interfering RNAs (siRNA) from heterochromatin-enriched transposable elements in targeting SU(VAR)3-9 (clr4) and HP1 (Swi6) to these regions [17]–[21]. RNAi-independent mechanisms also operate in recruiting HP1 to heterochromatin and to euchromatic genes [5], [8], [22]–[25]. Indeed, recent data point to a role for the DNA binding activity of KAP-1 (TIF1-β) in targeting HP1 to SRY-regulated genes in repressing transcription of testis-specific genes in the ovary [26].
HOAP is best known for its cooperative role with HP1 in forming a capping complex over Drosophila telomeres [27]–[29]. Immunostaining for HOAP also shows the protein at multiple non-telomeric sites in both heterochromatin and euchromatin of larval salivary gland polytene chromosomes [16], [30]. This study was undertaken to examine the non-telomeric functions of HOAP through microarray expression profiling of a mutant for it in order to identify candidate HOAP-regulated genes in these regions [27]. Contrary to our expectation, the majority of genes with altered expression in the mutant had reduced, rather than elevated, transcript levels. The majority of those with reduced transcript levels were found to normally be expressed only in the testis. This led us to uncover an underlying effect of the mutation on male viability and a role for both HOAP and HP1 in regulating the establishment promoter for the master sex determination gene in Drosophila, Sex lethal (Sxl) [31]–[33].
This establishment promoter of Sxl, SxlPe, is critical to the sex determination decision which is made early in embryogenesis (see Figure 4A for an overview of the Sxl locus). SxlPe is only transcribed in females, which have two X chromosomes. In counting the X chromosome number, also known as the X∶A ratio, SxlPe responds to five X-linked activating genes (sisterless-a, sisterless-b, runt, myc and unpaired) working in conjunction with positive maternal factors such as Daughterless. These activating components have their dose measured against the negative effect of maternal factors, such as Groucho and Extramacrochetae, and genes on the autosomes (deadpan is the only known member). Firing of SxlPe generates functional SXL protein which initiates the female-mode of splicing of transcripts from the maintenance promoter, SxlPm. SxlPm is transcribed in both sexes, soon after SxlPe shuts down and its mRNAs are being turned over. In female embryos, the SXL protein from SxlPe transcripts inhibits inclusion of the male-specific exon, which would otherwise prematurely terminate translation, of SxlPm transcripts. This autoregulatory splicing loop maintains SXL expression for the rest of the female life cycle. As males do not activate SxlPe they make no SXL protein and the splicing of SxlPm transcripts includes the male exon by default. Through this autoregulation, the binary sex determination decision is maintained.
SXL in females, through splicing and translational regulation, controls the downstream sex determination genes. A vital effect is turning off dosage compensation (DC), which equalizes X chromosome gene dose between the sexes by upregulating transcription of the male X by about two-fold [32], [34]. Failure to activate SxlPe thus leads to the improper, male mode of splicing of SxlPm transcripts and female lethality. Conversely, inappropriate activation of it in males results in the female mode of splicing of SxlPm transcripts and male lethality.
Our data show roles for both HOAP and HP1 in regulating the critical decision of whether activation of SxlPe will occur. These data support HOAP acting as a repressor, and HP1 as an activator at this promoter. They also suggest that HP1 first cooperates with HOAP in repressing SxlPe before switching to an activation mode. This is the first report of a non-telomeric function for HOAP and the most precisely defined role to date for HP1 in developmental control of a euchromatic gene.
In an effort to identify candidate HOAP-regulated genes, the Affymetrix Drosophila Genome Array 1 was used to compare the expression profile of wild type larvae to those that were mutant for the HOAP-encoding gene (caravaggio or cav). The original recessive lethal cav1 allele was used in the study. This allele contains a 5 bp insertion that causes truncation of the HOAP protein after two of three copies of a C-terminal HP1-binding repeat [27], [30]. The y1w67c23 stock, which provides the genetic background for all genetic manipulations in the lab, was used as the wild type control. Larvae of each genotype were collected at the first and second instar stage, prior to the lethal phase of the cav1 mutant. Out of 13,500 transcription units represented on the array, 183 genes were found to have significantly altered expression (log2R+2.0, p<0.01) in cav1 mutant larvae. Within this set, 142 genes had reduced transcript levels and 41 had increased levels.
Two strategies were then used to catalogue the normal expression profiles of genes in each data set. A gene's relative representation in publically available tissue-specific cDNA libraries [35]–[37] provided the first method for assessing its normal tissue distribution. This analysis was later complemented by data from two published microarray profiling studies of Drosophila sex- or tissue-specific gene expression [38], [39]. The results of these analyses are summarized in Tables S1 and S2. Genes categorized as “multiple” were represented in cDNA libraries of at least three different developmental specificities and enriched in at least three different tissue types in the Chintapalli et al. study [38]. Those categorized as “rare” were represented by a single or few cDNA clones and detected at low levels in multiple tissues by Chintapalli et al. [38]. Those categorized as tissue-specific (e.g., testis or midgut) were represented only, or predominantly, in cDNA libraries of that tissue-specificity and also specifically enriched in that tissue in Chintapalli et al. [38]. The most striking pattern to emerge from these analyses was the relative enrichment of testis-enriched genes (67%) in gene set with reduced transcript levels in the cav1 mutant. The Parisi et al. [39] study of sex-specific gonad expression provided corroboration for the testis-specificity of 67% of these genes, as they were both enriched in testes relative to whole animals and in testes relative to ovaries. This is in contrast to a complete absence of testis-specific genes in the gene set with elevated transcript levels, and also far exceeds the ∼12% of Drosophila genes reported to have testis-specific expression [40].
The enrichment of testis-specific genes in the reduced transcript level data set could indicate a requirement for HOAP in testis-specific gene expression. Alternatively, it could reflect an early lethal phase for cav1 mutant males and, thus, under-representation of male-specific transcripts in the cav1 RNA sample. RNA interference (RNAi) against the HP1-encoding gene (Su(var)205) in Drosophila has revealed enhanced vulnerability of males to partial HP1 knockdown [41]. To determine if males were under-represented in our cav1 larval sample, we used the X-linked yellow (y+) marker to sex individual cav1 mutant larvae, allowing us to sex the animals through a phenotype other than gonad size. By this criterion, 2.83-fold fewer male cav1 larvae were observed than female cav1 larvae. Using PCR with Y-linked primers to sex individual cav1 homozygous embryos, we also observed approximately two-fold more male embryos fail to progress to the larval stage.
We then used RNAi to partially knockdown HOAP and determine its effect on adult male viability. The GAL4/UAS binary expression system was used to drive expression of a cav RNAi transgene (Figure 1). Ubiquitous expression of cav RNAi from two transgenic lines (F8 and 543) through a maternally introduced Actin5C GAL4 driver resulted in a 2.5- and 1.5-fold reduction in adult male viability relative to females, respectively (Figure 1A). A third cav RNAi transgenic line (662) resulted in lethality of both males and females. The severity in effect of cav RNAi expression in different transgenic lines correlated with the degree to which endogenous cav mRNA was reduced as monitored through semi-quantitative RT-PCR, with ∼80% cav knockdown resulting in a 2.5-fold reduction in male viability (Figure 1B). The dose-dependent effects of cav RNAi expression on male viability are similar to those observed with Su(var)205 RNAi expression (2.54-fold reduction with 60–90% Su(var)205 knockdown and lethality of both sexes with >90% knockdown) [41]. A less severe, but significant, reduction in viability of male progeny was also observed in the reciprocal cross in which the Actin5C GAL4 transgene was introduced through the fathers. Although the results in both sets of crosses indicate a zygotic requirement for HOAP in male viability, the more pronounced effect was with maternally contributed Actin5C GAL4. This suggests maternally expressed GAL4 is needed to drive earlier expression of the cav RNAi for maximum effect.
A dominant effect on male viability was also observed for a newly recovered cav allele (cav2248). Heterozygosity for this allele reduces male viability 1.84-fold (n≥312, p<0.01). The cav2248 carries a nonsense mutation at nucleotide 111 of the cav-PB coding sequence within the region of similarity to the SRY HMG box. This allele appears to exert dominant negative activity on the wild type protein, as its effect on male viability is similar to that caused by RNAi-induced HOAP knockdown but is more pronounced than a deficiency for the locus (Df(3R)F89-4). The prematurely truncated protein in this mutant, or a reinitiation product from the next Met within the HMG box, is apparently responsible for this dominant negative effect. Smaller, but significant, reductions in male viability were also observed in progeny from wild type crossed to either parent carrying the cav2248 allele (data not shown). The lethal phase for cav2248 heterozygous males was determined to be late in embryogenesis after denticle belt formation; cav2248 homozygous embryos had an earlier lethal phase which is apparently due to telomeric defects (Figure S1).
Although the Y chromosome is not required for viability or sexual differentiation in Drosophila males, its heterochromatic composition enables it to act as a sink for heterochromatin proteins in suppressing position effect variegation of euchromatic genes artificially juxtaposed to heterochromatin [42], [43]. A limited pool of heterochromatin proteins might then render males more vulnerable to reductions in HOAP (and other heterochromatin proteins). To test whether the Y chromosome has a role in reducing viability of males that are deficient for HOAP, a compound X-Y chromosome [C(1;Y)] was introduced into animals expressing a cav RNAi transgene (Figure 1C). While the C(1;Y) chromosome modestly reduced viability of both male and female progeny by 30% relative to siblings lacking it, the sex of the animal had a pronounced effect on viability. Males expressing cav RNAi showed a 2.3-fold decrease in viability, regardless of whether they also carried C(1;Y).
Defects in the sex determination pathway, causing inappropriate dosage compensation in Drosophila, result in sex-specific lethality. Sex-lethal (SXL) protein acts as the master switch regulator of this pathway at the level of splicing and translation [44]. As described earlier, the critical decision is made early in embryogenesis by the activation of the Sxl establishment promoter (SxlPe) in only females, to generate functional SXL protein [32]. Failure to activate SxlPe, thus leads to improper male splicing of SxlPm transcripts and female lethality. Conversely, inappropriate activation of SxlPe in males results in the female mode of splicing of SxlPm transcripts and male lethality.
To determine if the enhanced male lethality of cav mutants is associated with inappropriate SXL activity in males, RT (reverse transcriptase)-PCR assays were used to monitor sex-specific splicing of SxlPm transcripts in heterozygous cav2248 and homozygous cav1 male embryos that failed to progress to the larval stage. Individual embryos of the appropriate genotype were identified through a GFP-marked wild type balancer chromosome (described more fully in Materials and Methods) and sexed by PCR with Y-linked primers (Figure S2). RNA was then purified from pools of male or female embryos and used in RT-PCR assays with Sxl primer pairs (P1/P3 and P2/P3) designed to discriminate between SxlPm transcripts spliced in the male vs. female mode (Figure 2A). Assays of RNA from wild type (y1w67c23) embryos showed correct sex-specific splicing of SxlPm transcripts in both males and females (Figure 2B). By contrast, a minor fraction of transcripts had undergone the female-mode of splicing in heterozygous cav2248 and homozygous cav1 male embryos (arrowhead in Figure 2B).
If inappropriate SXL expression is the cause of the enhanced male lethality in the cav2248 heterozygous males, we would expect a loss of function Sxl allele to rescue them. To this end, the Sxlf1 point null allele was introduced into heterozygous male and female cav2248 progeny through their mothers. Closely linked recessive markers, cut and carmine, which flank Sxl were used to follow the Sxlf1 allele in males, while in females white+, which should segregate with Sxlf1 82.3% of the time, was used as a close approximation. As shown in Table 1, the viability of cav2248 males carrying a defective Sxl allele (Sxlf1) was not significantly different from their female siblings, unlike their Sxl+ brothers (yw). It should be noted that in this cross, the effect of the cav2248 mutation on male viability was somewhat reduced; we ascribe this difference to the temperature at which the cross was done. For reasons that are not clear at present, all progeny from this cross were nonviable at 25°C, so the cross was performed at room temperature. An overall rescue in male viability was also observed by the introduction of the SxlfP7BO allele, although for this cross the markers on the deletion allele chromosome did not allow identification of the different progeny classes.
As HP1 and HOAP interact, and HP1 reduction also preferentially affects male viability, we wondered whether male larvae mutant for the HP1-coding gene [Su(var)2055/Su(var)2054] would also show altered splicing of SxlPm transcripts. GFP-marked balancer chromosomes were used to identify individual larvae of the correct genotype, and both an X-linked genetic marker (y+) and PCR assays of a Y-linked gene were used to sex them. Surprisingly, we found aberrant SxlPm transcripts in multiple individual Su(var)2055/Su(var)2054 larvae of both sexes (arrowheads in Figure 2C), suggesting HP1 has a dual role, of opposite consequence in each sex.
We therefore explored whether a maternal mutation in either cav or Su(var)205 would affect the viability of female progeny that also have a reduction in Sxl dose (Table 2). Female viability is not compromised in a cross between either wild type females and Sxl− males (fP7B0 or f1), or between cav mutant females and Sxl−/Y fathers (Note: the modest reduction in male viability in the progeny from cav2248 females and Sxl− males is essentially the same as that observed in progeny from cav2248 females and wild type males, as described earlier). However, in progeny from Su(var)205 heterozygous females crossed to Sxl−/Y males, female viability was dramatically reduced, particularly with the strong loss of function Su(var)2055 allele. The effect of Su(var)205 mutations is strictly maternal; no significant reduction in female viability is observed in the reciprocal cross (Table 2). Interestingly, this effect is allele-specific. Whereas mothers carrying the Su(var)2055 null allele or the Su(var)2054 carboxyl-terminally deleted allele had greatly reduced viability in their female progeny, mothers carrying a point mutation in the MeK9-H3-binding chromodomain allele [Su(var)2052] had a modest, but significant, effect. Although neither cav allele affected sex-specific viability of progeny from Sxl mutant fathers, the introduction of either allele into Su(var)205 mothers rescued the Su(var)205 maternal effect on female viability (Table 3). Both the dominant negative cav2248 allele and the C-terminally truncated cav1 allele, with compromised HP1-binding activity [27], [30], were capable of rescuing the Su(var)205 maternal effect, and this rescue was also strictly maternal.
The rescue of cav2248 male viability by the loss of Sxl and the antagonistic maternal effects of cav and Su(var)205 on the viability of their Sxl−-bearing female progeny, suggested a role for maternal HOAP and HP1 in regulating SxlPe. To directly assess the effect of reduced HOAP or HP1 on SxlPe, in situ hybridizations were performed with probes that distinguish the early from maintenance Sxl mRNAs on 0–4 hr embryos from cav2248 and Su(var)2055 heterozygous parents. For SxlPe, wild type embryos show two dots on the chromosomes, one for each female X, beginning in cycle 12 through to early cycle 14. Cycle 14 is also when the maintenance promoter, SxlPm, is activated and the autoregulatory splicing loop set in motion in females. In embryos from cav2248 heterozygous parents, two key changes were observed (Figure 3A). First, expression of SxlPe in females began two cycles earlier than normal – cycle 10, and overall expression appeared more robust than in wild type embryos. Second, male embryos also showed SxlPe activation (single in situ dot), although the expression was not as strong or as early as in females. The level of expression in male embryos was more variable both between embryos and at the level of individual nuclei. The majority had sporadic or scattered positive nuclei, while others had much greater numbers, as shown in Figure 3A. A count of early cycle 14 male embryos suggests >95% of the male embryos (n = 33) had at least some positive nuclei. In embryos from Su(var)2055 heterozygous parents, SxlPe expression was not observed in male embryos (single in situ signal) and was weaker and more variable than normal in females (Figure 3B). Analysis of the female embryos indicates that ∼85% express the promoter weakly during cycles 12 and 13 (n = 14–20 for each cycle) and less than half the embryos have normal levels of expression at cycle 14 (n>25). These results are consistent with the early expression of SxlPe being heavily reliant on the maternal deposit of HP1 protein and/or mRNA, with the zygotic contribution becoming more apparent at cycle 14. Expression of the maintenance SxlPm transcripts did not show significant changes in embryos from either genotype (Figure S3). Although HP1 binds to SxlPm in 4–18 hr embryos, at cellular blastoderm the promoter does not appear to be sensitive to a reduction in maternal HP1.
Chromatin immunoprecipitation assays were then used to determine if the effects of cav2248 and Su(var)2055 on SxlPe activity are mediated through direct physical association of the proteins with the Sxl locus (Figure 4). Cross-linked chromatin was prepared from developmentally staged collections of wild type (y1w67c23) embryos and immunoprecipitated with antibodies against HP1, HOAP, and non-immune IgG. Embryos were staged to monitor both Sxl promoters: before and during SxlPe activation (1–3 hr), during SxlPe activation (2–3 hr), and during SxlPm expression (4–18 hr). Quantitative real time PCR was then used to measure enrichment of Sxl sequences in each ChIP fraction. As summarized in Figure 4A, significant enrichment of sequences in the immediate vicinity of SxlPe (+138 and −101 fragments) was observed in the HP1 and HOAP ChIP fractions but not in the IgG ChIP fraction from 1–3 hr embryos. No enrichment of SxlPm sequences was observed at this stage. Significant enrichment of sequences upstream of SxlPe (−1132,−1800, and −2224) was also observed in the HOAP ChIP fraction. Using an anti-Myc antibody to probe chromatin from a stock expressing Myc-tagged HOAP from the endogenous cav locus [29], similar results were obtained (data not shown).
A comparison of the ChIP data from 1–3 hr versus 2–3 hr embryos shows a significant decrease (p<0.5) in the associations of both HOAP and HP1 with the −101 region of SxlPe. Both proteins retained significant associations with the +138 region of SxlPe in 2–3 hr embryos at a time that is coincident with SxlPe expression. HOAP enrichment in both the +138 region and the −1132 region appeared reduced at this time, although only at the 90% confidence level. No enrichment of either protein was found more 3′ in Sxl gene coding sequences or in the SxlPm region in either the 1–3 hr or 2–3 hr embryos. Also, SxlPe sequences were not enriched in either the HP1 or HOAP ChIP fractions from 4–18 hr embryos. However, HP1 enrichment in the vicinity of SxlPm (−5094 and −5331) and a slight, but significant, enrichment of HOAP in the −1132 region was observed in later staged embryos.
The ChIP assays described above confirmed physical association of HOAP and HP1 with the SxlPe promoter, but do not distinguish between the sexes. We, therefore, took advantage of the feminizing and masculinizing effects of maternal mutations in cav and Su(var)205 to gain insight into the requirement for each protein in regulating SxlPe activity.
The in situs, as well as the lethal effect of a maternal mutation in Su(var)205 on female progeny lacking one functional copy of Sxl, indicated an activation function for HP1 at SxlPe. To further test this hypothesis and examine its nature, we determined the effect of reducing maternal HP1 on RNA polymerase II (RNAP II) association with SxlPe (Figure 5A). Chromatin was prepared from 2–3 hr. embryos from wild type or heterozygous Su(var)2055 mothers mated to Sxlf1 fathers, as SxlPe is active at this stage in wild type embryos. Antibodies that recognize three different phosphoisoforms of RNAP II, which mark distinct stages of transcription initiation and elongation, were used to gain insight into the level of transcriptional activation at SxlPe at which HP1 is functioning. In embryos from wild type mothers, both unphosphorylated RNAP II (blue bar) and Ser 5-phorphorylated carboxyterminal domain RNAP II (green bar), which marks early steps of transcription elongation, were enriched in the proximal regions of SxlPe and the hunchback (hb) promoter. Both genes are active at this stage. Ser 2-phosphorylated RNAP II (red bar) which identifies polymerase in its elongation phase was mostly absent from both promoters. All RNAP II isoforms, particularly those associated with transcription initiation (unphosphorylated RNAP II, blue bars) and early events in transcription elongation (Ser 5-phosphorylated RNAP II, green bars), were drastically reduced from SxlPe proximal sequences in embryos from Su(var)205 mutant mothers and Sxlf1/Y fathers (Figure 5A). By contrast, the Su(var)205 mutation did not affect RNA polymerase II association with the hb promoter (Figure 5A).
The feminizing effects of reducing maternal HOAP, most notably the effect of the cav2248 mutation on transcription from SxlPe, support a repressive role for HOAP at this promoter. The partial rescue of the maternal Su(var)205 masculinizing effect by the HP1-binding-defective cav1 allele implicates HP1 in this repression as well. To examine the interdependency of HOAP and HP1 in their association with SxlPe sequences, we performed ChIP assays of each protein in 1–3 hr embryos with a reduced maternal dose of the other protein. As shown in Figure 5B, a maternal mutation in one copy of the gene for either protein significantly reduced association of its encoded protein with SxlPe sequences. These data also indicate a strong reliance of HP1 on HOAP for its association with SxlPe proximal sequences in embryos of this stage. HP1 association with these sequences was significantly reduced in embryos from cav2248 heterozygous parents. HOAP association with SxlPe sequences, also appeared educed in embryos from Su(var)2055 heterozygous parents, but this reduction was not statistically significant. These data support a role for HOAP in recruiting HP1 to SxlPe prior to the time of SxlPe activation; failure to form this repressive chromatin in embryos from heterozygous cav2248 mothers apparently negates the need for HP1 in SxlPe activation slightly later in embryonic development.
K9-methylated histone H3 has a well known conserved role in HP1 association with constitutive pericentric heterochromatin [7]–[9] and some euchromatic genes [5]. The activating function of HP1 at SxlPe does not appear to rely on this histone modification, as a maternal mutation for the Su(var)2052 allele, which encodes a protein that is defective for MeK9H3-binding, does not strongly affect the viability of female progeny from Sxl mutant fathers. It is still possible, however, that HP1-binding to MeK9H3 has a role in HP1 repression prior to the time of SxlPe activation. We, therefore, used ChIP assays to monitor the presence of both di- and tri-methylated histone H3 at the major site of HP1enrichment in the −101 region of SxlPe. The hb promoter, which is active at the same time as SxlPe but does not appear to require HP1 for activation, was used as the normalizing standard in these experiments.
Di-MeK9H3 was detected at above background levels (p<0.05) at SxlPe, but not at the simultaneously expressed hb promoter, in embryos from wild type mothers (Figure 5C). Although no significant enrichment was observed at SxlPe relative to hb in wild type embryos, a maternal cav2248 or Su(var)2055 mutation resulted in significant enrichment of both di- and tri-MeK9H3 at SxlPe (Figure 5C). Assuming, as in pericentric heterochromatin, a repressive activity for these histone modifications at SxlPe, their increased enrichment in this region in mutants for HP1 and HOAP may reflect a feedback mechanism to compensate the loss of HP1.
The canonical heterochromatin protein HP1 is most commonly associated with constitutive heterochromatin and gene repression. Here we report a critical role for it in regulating one of the earliest decisions in metazoan development, whether to embark on a female or male path of sexual differentiation and dosage compensation. The role of heterochromatin in mammalian dosage compensation has been recognized since the early work of Lyon [45]. Although Drosophila utilizes a different mechanism to equalize X-linked gene dose, through hyper-activation of the single male X chromosome via chromatin modification [46], this study provides the first evidence of a role for heterochromatin proteins in the early events of Drosophila sex determination. HP1, together with its telomere partner HOAP, influence the critical decision in sex determination - activation of SxlPe, the Sxl establishment promoter.
We find that reductions in HOAP preferentially compromise male viability. This was observed for two different cav mutant alleles and by reducing HOAP through RNAi. The presence of SxlPm-derived transcripts that have been spliced in the female mode in cav mutant males suggested inappropriate Sxl activation to be responsible for this reduced viability. In situ data indicating inappropriate firing of SxlPe in male embryos from cav2248 heterozygous parents support this view, as does the rescue of the cav2248 male viability defect by Sxl loss of function mutations. The more pronounced male lethality observed from reducing HOAP by RNAi expression driven by maternal, versus paternal, contribution of Actin5C GAL4 is consistent with such an early requirement for HOAP for male viability (Figure 1).
Previous reports have shown that reducing HP1 by RNAi similarly reduces male viability preferentially [41]. RT-PCR assays of SxlPm transcripts in HP1 mutants, however, suggested a more complex scenario as incorrect sex specific transcripts were observed in both sexes (Figure 2C). This pointed to an activation, as well as repressor, role for HP1. Consistent with an activation role, reduction of maternal HP1 severely compromised female viability when the dose of Sxl was also reduced in the progeny (Table 2), and ChIP assays of embryos from this cross showed recruitment of RNAP II to SxlPe to be impaired (Figure 5A). This effect of reducing HP1 on female viability was strictly maternal, as was the antagonizing effect of simultaneously reducing maternal HOAP (Table 3). Moreover, the partial rescue of the Su(var)205 maternal effect by the C-terminally truncated cav1 allele, which produces a protein that is compromised for HP1-binding [27], [30], points to an involvement of HP1 in the antagonizing activity of HOAP. Finally, ChIP assays show a dependence of HP1 on HOAP for its association with SxlPe. Combined, these data indicate both antagonistic and cooperative roles for these heterochromatin proteins in regulating SxlPe, whereby HOAP acts as a repressor and HP1 acts as both an activator and repressor. The reliance of HP1 on HOAP for recruitment to the promoter would suggest HOAP may also have a role in the activation function of HP1 at the promoter, although this was not readily apparent in our assays.
Although our data clearly show maternal roles for HOAP and HP1 in regulating the activity of SxlPe, for both HOAP and HP1 [41] the RNAi knockdown data indicate a substantial zygotic component in their effects on male viability. These zygotic effects, observed only in progeny carrying both an interference RNA transgene and a GAL4 driver transgene (Figure 1A), suggest additional later sex-specific roles for both proteins. Such roles could be related to those observed for HP1 and SU(VAR)3-7 in male dosage compensation [47]. Because the effect of reducing these proteins on the chromosomal distribution of DCC proteins [47] is the opposite of those observed for males that are deficient for DCC proteins [48], as predicted to occur with inappropriate SxlPe expression, the activities of heterochromatin proteins in dosage compensation appear to be distinct from the early roles of HP1 and HOAP at SxlPe. In addition, there may be zygotic roles for heterochromatin proteins in sex-specific gene expression, as proposed for HP1 by Liu et al. [41].
Previous analysis of SxlPe indicated that 400 bp immediately upstream of the promoter are sufficient for sex-specific regulation, but distal sequences, extending to −1700 bp, are required for wild type levels of expression [33]. As shown in Figure 4A, E-box binding sites for antagonistically acting bHLH proteins, which are encoded by zygotically expressed X-linked and autosomal signal elements (XSE and ASE) and direct an X counting mechanism, are distributed throughout both regions [49].
Both HP1 and HOAP are enriched in the region proximal to SxlPe which contains binding sites for both positive and negative E-box proteins. Within the SxlPe promoter distal region, HOAP alone is enriched in two peaks where there is a striking relationship with E-box binding sites for positive factors, but those for negative factors appear essentially devoid of HOAP. HOAP may antagonize positive factors but permit negative factors to bind in the SxlPe distal region, in an HP1-independent repressing role. Whereas loss of HOAP de-represses SxlPe in males, the strength and uniformity of expression does not approach that in wild type females. This indicates continued influence from the X counting mechanism in cav mutant males. SxlPe is also expressed prematurely in female embryos. This de-repression by reduced levels of maternal HOAP in both sexes indicates that HOAP is present at SxlPe in both sexes of wild type embryos. However, whether the proximal and distal SxlPe regions have the same or different compositions of HOAP and HP1 in the two sexes cannot be determined from our ChIP assays, as the embryos are of mixed sexual identity.
The interdependency of HOAP and HP1 for their binding to the SxlPe proximal region, most notably the dependence of HP1 on HOAP, also indicates both proteins are in this region in, at least, wild type female embryos. In spite of this interdependency, the genetic data show HOAP repression antagonizes HP1 activation. HOAP repression appears to also be partly HP1-dependent; the mutant HOAP protein from the cav1 allele which lacks HP1-binding also antagonizes HP1 activation. This combination of antagonistic and cooperative interactions suggests a model in which maternal HOAP and HP1 first cooperate to repress SxlPe prior to its activation (Figure 6A). The repressive structure formed by maternal HOAP and HP1 likely serves to reduce the sensitivity of SxlPe to spurious fluctuations in zygotic XSE levels, ensuring it is only activated in females where an effective ratio of activating to repressing transcription factors exists (Figure 6B). HP1 is retained at SxlPe during its activation in females, where it presumably switches into an activation role. In early embryos constitutive heterochromatin proteins may be more appropriate for such regulation than the Polycomb Group of facultative heterochromatin proteins, as they would not be subject to cross regulatory signals from body plan specification pathways.
How HP1 switches over to transcriptional activation mode in the SxlPe proximal region is unclear. Changes in HP1 phosphorylation and/or association with other factors could alter its activity [50], [51]. Several XSE binding sites are nearby, making them strong candidates. Presumably, this would only occur in females where the XSE dose surpasses a threshold and SxlPe is activated.
This report provides the most clearly defined role for HP1 in developmental control of a euchromatic gene in a metazoan species, and the first evidence of a bifunctional regulatory role for it in such a context. Prior reports describing HP1 in transcriptional activation have focused on it in the context of transcription elongation [11]–[13], [52], [53]. Our ChIP data at SxlPe, however, show a requirement of it for association of RNAP II with the promoter, more consistent with a role in transcription initiation. A role in initiation is also in keeping with the position of HP1 on the gene; we find very little HP1 elsewhere on the Sxl gene, even during the time of SxlPe activity. This dependence of RNAP II association on HP1 is similar to what is observed in the accumulation of noncoding RNAs at S. pombe centromeric repeats and mating type locus [54]. Nonetheless, it is possible that the loss of RNAP II at SxlPe reflects reduced stability of all RNAP II isoforms as a consequence of an early defect in transcription elongation, rather than a defect in RNAP II recruitment to the promoter.
Pausing of RNAP II in promoter proximal regions prior to activation has been observed in a high proportion of genes under developmental control in Drosophila embryos [55], and such pauses have also been implicated in regulation of alternative splicing [56]. While SxlPm appears to have the features of a promoter with paused RNAP II in a genome wide RNAP II ChIP study of 0–4 hr embryos (Flybase MODENCODE), RNAP II was absent from SxlPe. It is likely that the collection window for this study did not precisely coincide with the time of SxlPe activity. Our more narrowly timed collection indicates paused RNAP II at SxlPe, suggesting that, like SxlPm, it is a pre-loaded promoter. A preloaded SxlPe also readily explains how generalized up-regulation of phosphorylation of the RNAP II CTD by the loss of Nanos, causes SxlPe activation in males with an unchanged X∶A ratio [57].
Finally, the dominant negative activity of the cav2248 allele suggests a role for the partially deleted SRY-like HMG box in HOAP association with SxlPe. Our ChIP data show HOAP association with the SxlPe proximal region is required for HP1 association. This proposed role for the HMG box of HOAP in SxlPe regulation is of particular interest with regards to a recent report linking HP1 and KAP-1 (TIF1β) to SRY-dependent repression of testis-specific genes in the ovary [26]. Because mammalian sex determination is inextricably linked to gonad sex determination, SRY and HOAP each appear to constitute early decision points in their respective sex determination pathways. There are, perhaps, unexpected parallels between these divergent pathways.
The cav RNAi lines were obtained by germline transformation of the pUAST vector containing an inverted repeat of a near full length cDNA sequence for the cav-RB transcript (See Table S3 for primer sequences). The cav2248 allele was recovered in a screen of progeny from ethane methyl sulfonate (EMS)-mutagenized males which failed to complement the original cav1 allele, and the sequence of the allele was determined from PCR products obtained from homozygous cav2248 embryos identified through their lack of the homologous twi-GAL4, UAS-GFP marked balancer chromosome.
Flies in all crosses were reared under uncrowded conditions on standard cornmeal medium enriched with active dry yeast. Unless otherwise noted, all crosses were done at 25°C; Ore R or y1w67c23 were used as the wild type control. Progeny were counted for 8 days beginning on the first day of eclosion. The Z test was used in statistical analyses of distributions in two populations [58]. Description of genes can be found in Flybase (http://www.flybase.org/).
RNA was TRIzol-extracted from y1w67c23; cav1 homozygous (those lacking the Tm6B,Tb balancer chromosome used to maintain the cav1 mutation in heterozygous condition) and wild type first and second instar larvae and purified through Qiagen RNeasy columns. The RNA was then used by the University of Kentucky Microarray Facility to prepare biotin-labeled cRNA that was hybridized to separate Affymetrix Drosophila Genome microarrays (version 1). The y1w67c23 stock to which all mutant stocks are out-crossed during genetic manipulations was used as the wild type control in these experiments. The data were obtained from two biological replicate samples from each stock. The statistical analyses of the arrays were carried out by A.J. Stromberg, PhD (UK Dept. of Statistics) associated with this facility according to standard Affymetrix specifications. Of 13,982 affytags to Drosophila genes on the array, an average of 46% and 53% were present in the cav1 and wild type y1w67c23 samples, respectively. Data from publically available tissue-specific cDNA libraries [35]–[37] and from two published microarray profiling studies of Drosophila sex- or tissue-specific gene expression [38], [39] were used to catalogue normal gene expression profiles of genes with reduced or elevated transcript levels (log2R>2.0, p<0.01).
RNA was similarly prepared from pools of male and female cav2248 heterozygous and cav1 homozygous embryos and from individual Su(var)2054/Su(var)2055 larvae. The cav2248 heterozygous embryos were identified from a pool of tightly staged embryos that failed to progress to the larval stage and contained intermediate levels of GFP expression from the P{GAL4-twi.C}2.3, P{UAS-2xEGFP}AH2.3-marked TM3Sb balancer chromosome. The cav1 homozygous embryos were similarly identified through their lack of GFP expression from this balancer chromosome. Su(var)2054/Su(var)2055 larvae were identified through their lack of GFP expression from the P{Act5C-GAL4}25F01-marked CyO balancer chromosome. The sex of individual cav2248 heterozygous and cav1 homozygous embryos was first determined through PCR of the Y-linked kl-2 gene from a total nucleic acid extraction from individual embryos (See Table S3 for primer sequences). RNA was then purified from pools of nucleic acids from individual male or female embryos. The sex of individual Su(var)2054/Su(var)2055 larvae was first determined through the presence or absence of the y+-marker from the paternally derived X chromosome and then substantiated through PCR assays of the Y-linked kl-2 gene. RT-PCR assays were carried out with RNA purified from these pools using the Qiagen One-Step RT-PCR kit (Qiagen 210212). The sequences of all primers used are shown in Table S3. Primers for the Drosophila RpA-70 were used in PCR reactions to determine that all RNA samples were DNA-free before they were used in RT-PCR as a positive control and in normalization of all RT-PCR data (Table S3).
These were done as described in [59]. Digoxygenin labeled RNA probes complementary to Sxl exon E1 or L1 region were prepared using T7 RNA polymerase in vitro transcription of plasmid- or PCR-derived templates. The establishment (407 nt) and maintenance (1039 nt) transcript specific probes were generated by the primers shown in Table S3. All in situs were repeated at least once. Each batch was done simultaneously with an Ore R control and had sufficient embryos so that several representatives of each cycle could be examined.
A modification of the protocol described in [60] was used to prepare cross-linked chromatin from embryonic progeny from parents of the following genotypes: wild type (y1w67c23 or yw), yw; cav2248/TM3Sb, yw; Su(var)2055/CyO, yw females×Sxlf1/Y males, and yw; Su(var)2055/CyO females×Sxlf1/Y males. Embryos from yw parents were collected at the following developmental stages: 0.75 to 2.75 hr (labeled 1–3 hr), 2–3 hr, and 4–18 hr. Embryos were collected from yw; Su(var)2055/CyO females crossed to Sxlf1/Y males in parallel to those from yw females crossed to Sxlf1/Y males (at 2–3 hr stage). Embryos from yw; cav2248/TM3Sb and from yw; Su(var)2055/CyO parents were also collected in parallel (at 0.75 to 2.75 stage). All embryo collections and staging were done at 22°C. Chromatin was prepared from 6.0 g yw embryos in 6.0 ml homogenization buffer (50 mM Hepes at pH 7.6, 60 mM potassium chloride, 0.25 M sucrose, protease inhibitor cocktail [15]). The homogenate was first clarified by centrifugation at 500× g for 10 minutes before the addition of formaldehyde to 2%. Cross-linked chromatin was then washed 3 times in phosphate buffered saline (150 mM sodium chloride (NaCl), 10 mM sodium phosphate, pH 7.6) (with centrifugation at 3,000× g for 10 minutes after each wash) and re-suspended in 6.0 ml RIPA buffer (50 mM Tris-HCl, pH 7.6, 1 mM ethylene diamine tetraacetic acid (EDTA), 0.5 mM ethylene glycol tetraacetic acid (EGTA), 140 mM NaCl, 1% Triton X-100, 0.1% Na-deoxycholate, 0.1% SDS) and sonnicated to an average length of 500 bp. A scaled down version of the protocol was carried out with 0.5 g mutant embryos in 1.5 ml homogenization buffer.
Chromatin immunoprecipitations were performed with 0.2 ml clarified chromatin, 20–40 µg antibody [anti-HOAP [61], anti-HP1 [15], anti-RNA polymerase II antibody (Covance MMS-126R, MMS-134, or MMS-129R), anti-di and tri MeK9 histone H3 (Millipore 05-1249 and 05-1242) or non-immune IgG (Santa Cruz Biotech. SC-2027 or SC-2025)] and 100 µl anti-rabbit (Sigma A914) or anti-mouse (Sigma A6531) IgG agarose in 1.5 ml RIPA buffer. Washes were performed as described in Alekseyenko et al., [60]. The immunoprecipitated material was eluted from the beads by incubation at 37°C for 1 hr in 500 µl TE (1 mM EDTA, 50 mM Tris, pH 8.0) containing 0.5% sodium dodecyl sulfate (SDS) and proteinase K (0.1 mg/ml), followed by 12 hr at 65°C after the addition of NaCl to 0.3 M and SDS to 1%. The samples were then extracted once with phenol/chloroform, once with chloroform before ethanol precipitation in the presence of glycogen.
The iCycler iQ real-time PCR detection system (Bio Rad) was used to quantitate Sxl sequences in the precipitated DNA from each ChIP fraction. The primer pairs shown in Table S3 were used to amplify fragments spanning the Sxl locus as shown in Figure 4 and RpA-70 normalizing standard (average length 288 bp). Similar enrichment values were obtained for all data when calculated as % of total in ChIP vs. input fraction. PCR amplification was performed in duplicate in 50 µl SYBR Green qPCR SuperMix (Bio-Rad 170–8880) on two biological replicates of each ChIP fraction. Dissociation curve analysis was performed at the end of 40 cycles, and quantification was carried out by Bio-Rad comparative CT methodology with standard curves constructed for each primer pair with a serial dilution of input DNA having PCR efficiencies of 80–120%. A one sample t-test was performed to identify sequences that were enriched in ChIP fractions above background (i.e., >1); a student's t test was used to determine significance of differences between two samples of equal variance.
|
10.1371/journal.pntd.0003620 | Distribution of Lutzomyia longipalpis Chemotype Populations in São Paulo State, Brazil | American visceral leishmaniasis (AVL) is an emerging disease in the state of São Paulo, Brazil. Its geographical expansion and the increase in the number of human cases has been linked to dispersion of Lutzomyia longipalpis into urban areas. To produce more accurate risk maps we investigated the geographic distribution and routes of expansion of the disease as well as chemotype populations of the vector.
A database, containing the annual records of municipalities which had notified human and canine AVL cases as well as the presence of the vector, was compiled. The chemotypes of L. longipalpis populations from municipalities in different regions of São Paulo State were determined by Coupled Gas Chromatography – Mass Spectrometry. From 1997 to June 2014, L. longipalpis has been reported in 166 municipalities, 148 of them in the Western region. A total of 106 municipalities were identified with transmission and 99 were located in the Western region, where all 2,204 autochthonous human cases occurred. Both the vector and the occurrence of human cases have expanded in a South-easterly direction, from the Western to central region, and from there, a further expansion to the North and the South. The (S)-9-methylgermacrene-B population of L. longipalpis is widely distributed in the Western region and the cembrene-1 population is restricted to the Eastern region.
The maps in the present study show that there are two distinct epidemiological patterns of AVL in São Paulo State and that the expansion of human and canine AVL cases through the Western region has followed the same dispersion route of only one of the two species of the L. longipalpis complex, (S)-9-methylgermacrene-B. Entomological vigilance based on the routes of dispersion and identification of the chemotype population could be used to identify at-risk areas and consequently define the priorities for control measures.
| Information on the geographical distribution, dispersal mechanisms and dispersion route of insect-borne diseases can help to identify ongoing transmission areas, new risk areas and guide surveillance and control activities. Lutzomyia longipalpis, the principal vector of American visceral leishmaniasis disease in the Americas, is a group of closely related species that can be separated according to the type of pheromone produced by male individuals. It is still unclear how many members there are in this complex, how they are related and if some are more important vectors than others. In the present study, we show that the expansion of human visceral leishmaniasis in the state of São Paulo, Brazil, has followed the same dispersion route of only one of the two related species of L. longipalpis found in the state. The maps that we produced in the present study will allow us to determine risk areas for the occurrence of human visceral leishmaniasis, and reinforces our hypothesis that within São Paulo state these chemotype populations could have different biological capacities to act as a vector.
| Recording the geographic distribution and identifying the possible routes of expansion of both arthropod-borne diseases and their associated vectors is essential information for surveillance as well as the execution and elaboration of control strategies [1].
In Brazil, the expansion of the geographic range of Lutzomyia longipalpis (Lutz & Neiva), the principal vector of Leishmania (Leishmania) infantum chagasi (Cunha & Chagas), and its adaptation to domiciliary habitats in the urban areas throughout Brazil has resulted in an increase in the incidence of both canine and human visceral leishmaniasis (VL) in the last 25 years [2–6]. According to the Brazilian Ministry of Health, in the period from 2009 to 2011, there were 251 municipalities classified as having moderate (mean number of human cases > = 2.4 and < 4.4) or intense (mean number of human cases > = 4.4) VL transmission in the country [7].
Before 1998, São Paulo State was considered free of autochthonous cases of this zoonotic disease and records of the vector’s presence were restricted to some rural areas of municipalities in the Northeast region of the state [8]. Two human cases had been reported in Greater São Paulo, but possible reservoirs and vectors were not described [9]. The first record of L. longipalpis in an urban area in São Paulo State was in 1997 from Araçatuba in the West of São Paulo State [8]. Canine and autochthonous human cases occurred in the same municipality in 1998 and 1999 respectively [10]. Since then, the appearance of L. longipalpis in urban areas of other municipalities has been linked to an increase in both canine and human visceral leishmaniasis within the State [11, 12]. From 1999 to April 2013, São Paulo State recorded 2204 autochthonous human cases of disease, with 192 deaths [13]. In São Paulo State, 18 municipalities were classified as having moderate or intense transmission in the period from 2010 to 2012 [11].
Based on genetic and behavioural studies it is generally accepted that L. longipalpis is a species complex, but it is unclear how many members there are and how they are related [14, 15]. Chemical, behavioural and ecological analysis of male produced sex pheromones suggests that L. longipalpis is a complex of at least four different, reproductively isolated members [16–19]. In Brazil two of these are represented by members where the males produce either 3-methyl-α-himachalene [20], a novel bicyclic methylsesquiterpene (C16; mw 218) found in Jacobina, Bahia State, or (S)-9-methylgermacrene-B [21], a novel monocyclic methylsesquiterpene (C16; mw 218) that is widely distributed throughout Brazil but typically represented by L. longipalpis from Lapinha Cave, Minas Gerais State. The other two members of the complex produce novel diterpenes, cembrene-1 and cembrene-2 (C20; mw 272) and are represented by the Sobral-2S population from Ceará State and the Jaíbas-1S population from Minas Gerais State [22]. Two of these chemotype populations, (S)-9-methylgermacrene-B and cembrene-1, have been identified in São Paulo State [23].
Considering the remarkable epidemiological differences between the two municipalities of Araçatuba and Espírito Santo do Pinhal (mainly the number of human cases notified as well as the abundance, and chemotype of the L. longipalpis population present in each urban area), Casanova et al (2006) [23] suggested that the (S)-9-methylgermacrene-B and cembrene-1—chemotype populations had different vectorial capacities. If this is true then it is important to have more detailed information on the distribution of the chemotypes to produce more accurate risk maps and to direct more effective control programs. With this in mind, the present study is aimed at determining the chemotypes of a greater number of L. longipalpis populations from different regions of São Paulo State.
São Paulo State is located in Southeast region of Brazil, and shares borders with Minas Gerais to the North and Northeast, Paraná to the South, Rio de Janeiro to the East and Mato Grosso do Sul to the West, and to the Southeast, the Atlantic Ocean (Fig. 1). It is divided into 645 municipalities totalling 248,209 km2.
Its climate can be divided into seven distinct types, most are classified as humid. According to Koeppen’s climate classification, the predominant climate type is Cwa, which includes Central and Eastern São Paulo, defined as high-altitude tropical climate, where summer is the rainy season, winter is dry, and the average temperature in summer is above 22°C. In the West region (Aw climate type), the rainy season is delayed until autumn, the winters are dry (the precipitation for the driest month is less than 60mm) and the average temperature for the coldest month is above 18°C [24].
In Brazil, including São Paulo State, American visceral leishmaniasis is a compulsory notifiable disease. Data used in the present study were obtained from Epidemiological Surveillance Centre of Secretary of Health of São Paulo State [13]
Since the beginning of visceral leishmaniasis surveillance and control activities in São Paulo State, all the municipalities are expected to notify the first- confirmed (laboratory based parasitological identification) occurrence of L. i. chagasi. Data were obtained from canine surveys carried out by the municipalities, Adolfo Lutz Institute and the Secretary of Health of São Paulo State.
L. longipalpis distribution data was obtained from both published data and, principally, from entomological collections carried out by Secretary of Health of São Paulo State, during the performance of their epidemiologic surveillance activities [8, 11, 23, 25–27]. These activities included annual or biannual collections, with CDC light traps, in a minimum of 4 dwellings (more where possible) of all the municipalities considered silent—i.e. without canine or human cases, non-receptive—i.e. where L. longipalpis has not yet been found—and those considered vulnerable—i.e. those municipalities that are located near to or that connected because of intense transportation of goods and people by road and railway with municipalities with transmission [25]. Annual entomological collections were also carried out in at least 42 dwellings in areas where proven or suspected human or canine transmission occurs but where the vector has not yet been registered.
Male L. longipalpis from different municipalities were collected manually with an aspirator or CDC electric light trap from peridomiciliary habitats within urban, peri-urban and rural areas, always with permission from local homeowners. The attempts to collect males were made in at least three evenings in four peridomicilies of each sampled municipality. Samples from western São Paulo State were obtained in 11 municipalities that were selected so as to represent all the vector distribution area. For the eastern area, where only 25 municipalities have registered the presence of the vector, nine municipalities, including those with canine transmission were sampled. The sampled municipalities, the geographic coordinates, the number of males chemically analysed and the collection year were respectively: Araçatuba (21°12'14” S; 50°42'61” W), >100, 2005 and 2009; Promissão (21°32'18'' S; 49°51'28'' W), 35, 2009; Bauru (22°18'55'' S; 49°03'41'' W), 23, 2009; Dracena (21°29'00'' S; 51°32'01'' W), 22, 2009; Adamantina (21°40'32'' S; 51°03'47'' W), 13, 2011; Oswaldo Cruz (21°47'38'' S; 50°53'08'' W), 17, 2011; Jales (20°16'06'' S; 50°32'56'' W), 10, 2013; Presidente Prudente (22°07'39'' S; 51°23'08'' W), 13, 2010; Marília (22°13'15'' S; 49°56'55'' W), 9, 2012; Salmourão (22°13'15'' S; 49°56'55'' W), 2, 2013; Lourdes (20°58'01" S; 50°13'27" W), 2, 2013; Espírito Santo do Pinhal (22°10'60'' S; 46°45'45'' W), >20, 2004 and 2009; Socorro (22°35'50'' S; 46°31’04'' W), 1, 2012; Salto (23°12'10'' S; 47°17'11'' W), 2, 2012; São Pedro (22°36'00'' S; 47°52'31'' W), 28, 2009; Indaiatuba (23°5'18'' S; 47°13'24'' W), 3, 2012; Campinas (22°54'23'' S; 47°03'42'' W), 10, 2009; Águas da Prata (21°56'18'' S; 46°42'54'' W), 4, 2013; Sorocaba (23°30'22'' S; 47°27'21'' W), 13, 2013; Votorantim (23°32'26'' S; 47°26'38'' W), 5, 2013. All males were observed under a stereomicroscope to identify to the species level by examination of external morphological characteristics (pale spots on the 4th or 3rd and 4th abdominal tergites and a pair of spikes on the paramere).
All males were killed by placing them in a freezer at-20° C for 10 minutes. They were then placed individually in a glass vial and then covered with hexane (ca. 20 μl). Analysis of male sex pheromone extracts was on a HP-5MS capillary column, 30 m x 0.25 mm i.d., 0.25 μm film thickness (Agilent, Stockport, Cheshire) in a Hewlett Packard 5890 II+ Gas Chromatograph coupled to a Hewlett Packard 5972A bench-top mass spectrometer (electron impact, 70 eV, 180°C). Injection and chromatography conditions were as previously described [19].
Before 1997, L. longipalpis had been found only in the rural areas of six municipalities of São Paulo State, all of which are in the East and Northeast regions of the state. The first report of the vector in an urban area was in 1997 in the municipality of Araçatuba, in the Western region near the border with Mato Grosso do Sul State (Fig. 2). From 1998 to June 2014, L. longipalpis has been reported in another 164 municipalities (Fig. 2, S1 Database). During this period, between 2 and 21 new municipalities per year reported the presence of L. longipalpis, with more than 45 reporting the presence of the vector in the last 3 years (Fig. 3). The biggest expansion in the distribution of L. longipalpis happened in the western part of Sao Paulo where 146 municipalities have recorded their presence in urban areas during this 17.5 year period.
The spatial and temporal distribution of L. longipalpis and human and canine cases in general shows that the presence of the vector preceded the canine cases, and these in turn preceded the human cases (Figs. 2, 4, 5, S1 Database). Up until 2014, there have been 105 municipalities with canine and/or human VL transmission. The majority of these municipalities (93.3%), are placed in the Western part of São Paulo State, and the cases show an expansion route in a Southeasterly direction, towards the Central region, and from there, an expansion both to the North and the South (Figs. 4 and 5). In 71 of these municipalities, there has been both human and canine cases, in 23 only canine transmission and in five only human cases (Figs. 4 and 5). It is interesting to note that in the East region, L. longipalpis has been found in 25 municipalities only, with canine cases reported in seven, and no known human cases.
The chemical analysis of all samples of L. longipalpis males from 11 municipalities in the West region (Araçatuba, Promissão, Bauru, Dracena, Adamantina, Oswaldo Cruz, Jales, Presidente Prudente, Marília, Lourdes and Salmourão) have been shown to contain (S)-9-methylgermacrene-B (Fig. 6). On the other hand, all samples of male L. longipalpis collected in eight municipalities of the Eastern region (Espírito Santo do Pinhal, Socorro, Salto, Indaiatuba, Campinas, Águas da Prata, Sorocaba and Votorantim) contained cembrene-1 (C-20) (Fig. 6). In the municipality of São Pedro, situated in Central region of the State, it was found a (S)-9-methylgermacrene-B producing population and in addition, two flies that produced both (S)-9-methylgermancrene-B, and cembrene-1.
The argument in favour of the hypothesis of the recent introduction of L. longipalpis into the Western region of São Paulo State can be supported by its absence, for decades, from various sporadic rural collections of sand flies [28–30]. These collections were done in areas where autochthonous cases of cutaneous leishmaniasis, caused by Leishmania (Viannia) braziliensis, had been reported. The contrary hypothesis, that L. longipalpis has always been there, hidden in the primitive natural vegetation habitat, could be supported because of the existence of several areas where collections have never been done [28–30] Therefore, there are gaps in our knowledge of its distribution and in future, it would be interesting to collect samples from the few natural vegetation areas of São Paulo State.
It is very difficult to pinpoint the year when L. longipalpis first reached the urban areas of the municipalities in the Western region, however it is likely that, when detected for the first time in the municipality of Araçatuba, in 1997 [8], L. longipalpis was already present in the urban areas of surrounding municipalities, as the species was found there in the first collections of the following year. From that point on, the spatial and temporal distribution leaves no doubt as to the west-east progression (as far as the central region of the state). This progression can be inferred from the annual urban entomological collection results, which showed that in various municipalities the vector detection only occurred after successive annual negative collections. It is possible that factors related to the economic development of the country, such as the increase in transportation of goods and people by road and railway, could have been responsible for the dispersion of the vector and, consequently, the expansion in the vector’s range in the West.
The higher number of municipalities from the West of the state which reported the presence of L. longipalpis for the first time since 1997 indicates a rapid inter-municipality dispersion rate. The greatest number of municipalities reported in 2012 and 2013 (21 and 14, respectively) is a further indication of this ongoing, fast expansion. The fact that the expansion of canine and human cases through the Western São Paulo has followed the same dispersion route as that of the vector with a temporal delay cannot be considered to be merely coincidental because it has long been observed that in VL epidemiology the vector precedes canine and subsequently human cases [31]. The lower number of municipalities notifying the presence of the vector in the East region of the state, in contrast to the West, does not show a recognisable dispersion route. Spread of the disease is more likely therefore to be due to the expansion of urban areas into rural or wild areas. This hypothesis is further supported by the observation that L. longipalpis was only found exclusively in rural habitats in nine of the 25 studied municipalities. In the other 16 municipalities, L. longipalpis was found in both rural and urban areas in four, and in periurban areas (i.e. those that have the characteristics of rural areas) in 12 municipalities.
It is clear that there are two distinct epidemiological patterns of VL in these two regions of São Paulo State. In the western region it is defined by the occurrence of human cases [12], frequent high prevalence of canine cases [32, 33], and a greater number of municipalities where L. longipalpis is present. Generally, a great number of flies is frequently found in both manual and CDC light trap collections carried out in peridomiciliary environments associated with food sources, such as chickens and dogs [33, 34]. In this area, all the males analysed have been shown to be the (S)-9-methylgermacrene-B chemotype population, including those collected in the 6 of the 18 municipalities currently classified as having moderate or intense transmission in the period from 2010 to 2012. In contrast, the eastern region, can be characterized by the absence of notified human cases—even where the presence of L. longipalpis and canine cases have been reported for at least 12 years—low prevalence in dogs and a smaller number of municipalities where the vector is present. The populations of sand flies are generally low in abundance in manual and CDC light trap collections carried out in peridomiciliary and rural environments associated with the feeding sources, such as chickens and dogs [35]. All samples of males analysed were the cembrene-1 chemotype.
These observations support the hypothesis of Casanova et al [2006] [23] which proposes that the (S)-9-methylgermacrene-B chemotype population has a greater vectorial capacity than the cembrene-1 chemotype in São Paulo State. Differences in ecological parameters of the vector capacity (e.g. vector abundance, survival, host feeding pattern and blood feeding rate) could vary between the two chemotype populations. Furthermore, susceptibility and coevolutionary interactions with Leishmania genotypes, which can influence Leishmania transmission, are parameters involved in vectorial competence and can vary between different species of L. longipalpis complex [15, 36, 37]. It is interesting to note that the two main genetic clusters of L. i. chagasi, identified by multilocus microsatellite typing, isolated from dogs from Northwest and Southeast regions of the São Paulo State [38] show distribution coinciding with the two L. longipalpis chemotype populations distribution presented here. The present study indicates that the chemotype of L. longipalpis populations as well as the effect of spatial and temporal environmental heterogeneity (reviewed in Belo et al 2013 [39]), should also be considered when explaining the variety of eco-epidemiological transmission scenarios in São Paulo State. However, it is also important to mention that the available data for pheromone types populations from municipalities of other states of Brazil and classified by the Brazilian Ministry of Health as having moderate and intense transmission of VL in the period from 2009 to 2011, shows that in three of these municipalities (Marajó-PA, Natal-RN, Pancas-ES) the chemotype population is cembrene-1, in two (Terezina-PI, Campo Grande-MS) it is (S)-9-methylgermacrene-B chemotype, and in one (Sobral-CE) both occur [7, 15]. Further studies are required to access the parameters of vectorial capacity of the longipalpis species complex.
The (S)-9-methylgermacrene-B, distributed throughout São Paulo State’s western region has previously been found and characterized in males from Lapinha Cave-MG [21] and later in populations from Sobral-CE, Terezina-PI, Campo Grande-MS, Barra de Guaratiba-RJ, Montes Claros-MG and Araçatuba-SP [15, 19, 23, 40]. The diterpene (C-20) found in populations from municipalities of the East region of the state was previously found and characterized as cembrene-1 in L. longipalpis from Sobral-CE, and later also found in populations from Marajó-PA, Natal-RN, Estrela-AL, Jaíba-MG, Pancas-ES and Espírito Santo do Pinhal-SP [15, 19, 23]. The presence of a cembrene-1 population in São Paulo State is the southernmost extension of this chemotype.
Based on the results of our present study we suggest that the L. longipalpis cembrene-1 populations are of rural origin and native of the Eastern region of the São Paulo State, while (S)-9-methylgermacrene-B is an introduced chemotype population. Although L. longipalpis has been found in urban areas of several municipalities of Mato Grosso do Sul State [41], we are not aware of any publications on L. longipalpis distribution through time in this State. However, although human cases are not a good space-time indicator of parasite circulation, the occurrence of the first human autochthonous case in a municipality usually indicates that the vector and the canine transmission were already established in the area. The West-to-East expansion of human VL in Mato Grosso do Sul was properly demonstrated by Correa-Antonialli et al (2007) [42] who pointed to the construction of the Bolivia-Brazil gas pipeline as a possible cause for the VL time and space expansion. This same hypothesis was considered to explain the spread of the canine and human disease in the west region of São Paulo [12, 38]. Finding the same pheromone type in males in the same temporal expansion route from West to East in São Paulo State may also support the hypothesis that the (S)-9-methylgermacrene-B L. longipalpis chemotype has been introduced from Mato Grosso do Sul.
More samples of males from São Pedro, in the Central region of São Paulo State, should be analysed to allow a better understanding of the possible presence and distribution of sympatric populations of the two pheromone chemotypes. Molecular and behavioural analyses such as those done by Araki et al (2009, 2013) [15, 43] may also help to clarify this question.
Information on the dispersion route and distribution of L. longipalpis chemotype populations is essential to understand the epidemiological patterns observed in São Paulo State. It may be used to identify areas at risk and consequently define priorities for control measures. In addition, identifying the distribution of the different chemotype populations is helpful in the application of appropriate synthetic male sex pheromone, in pheromone-baited traps or other appropriate “attract-and-kill” approaches [40, 44–49].
|
10.1371/journal.pgen.1007262 | Zinc transporters belonging to the Cation Diffusion Facilitator (CDF) family have complementary roles in transporting zinc out of the cytosol | Zinc is an essential trace element that is required for the function of a large number of proteins. As these zinc-binding proteins are found within the cytosol and organelles, all eukaryotes require mechanisms to ensure that zinc is delivered to organelles, even under conditions of zinc deficiency. Although many zinc transporters belonging to the Cation Diffusion Facilitator (CDF) families have well characterized roles in transporting zinc into the lumens of intracellular compartments, relatively little is known about the mechanisms that maintain organelle zinc homeostasis. The fission yeast Schizosaccharomyces pombe is a useful model system to study organelle zinc homeostasis as it expresses three CDF family members that transport zinc out of the cytosol into intracellular compartments: Zhf1, Cis4, and Zrg17. Zhf1 transports zinc into the endoplasmic reticulum, and Cis4 and Zrg17 form a heterodimeric complex that transports zinc into the cis-Golgi. Here we have used the high and low affinity ZapCY zinc-responsive FRET sensors to examine cytosolic zinc levels in yeast mutants that lack each of these CDF proteins. We find that deletion of cis4 or zrg17 leads to higher levels of zinc accumulating in the cytosol under conditions of zinc deficiency, whereas deletion of zhf1 results in zinc accumulating in the cytosol when zinc is not limiting. We also show that the expression of cis4, zrg17, and zhf1 is independent of cellular zinc status. Taken together our results suggest that the Cis4/Zrg17 complex is necessary for zinc transport out of the cytosol under conditions of zinc-deficiency, while Zhf1 plays the dominant role in removing zinc from the cytosol when labile zinc is present. We propose that the properties and/or activities of individual CDF family members are fine-tuned to enable cells to control the flux of zinc out of the cytosol over a broad range of environmental zinc stress.
| All organisms require homeostasis mechanisms to maintain sufficient levels of zinc for normal cell metabolism and to avoid toxicity. As zinc-binding proteins are located in the cytosol and within intracellular compartments, all cells have to balance intracellular zinc ion distribution so that there are sufficient, but non toxic levels of zinc in the cytosol as well as organelles. Although much is known about the mechanisms that control cytosolic zinc levels, relatively little is known about the mechanisms that maintain organelle zinc homeostasis. As proteins belonging to the CDF family transport zinc into organelles, here we used a fission yeast model system to determine if the expression or function of zinc transporters belonging to this family was regulated by zinc. We find that two CDF family members, Cis4 and Zrg17, facilitate the transport of zinc out of the cytosol of zinc-deficient cells, whereas the CDF family member Zhf1 preferentially transports zinc out of the cytosol when zinc is not limiting. As the expression of the genes encoding these transport proteins is not regulated by zinc, the results suggest that different CDF family members have complementary roles in transporting zinc out of the cytosol that are independent of changes in transcription. These results provide new insights into the regulatory mechanisms that control cytosolic and organelle zinc homeostasis.
| Zinc is an essential trace metal that is required for the structure and activity of a large number of proteins. In eukaryotes these proteins include transcription factors containing structural domains stabilized by zinc ions, such as the C2H2-type and C4-type zinc fingers [1]. Zinc is also a cofactor for many enzymes that are located in the cytosol (e.g. alcohol dehydrogenase 1), and in subcellular compartments such as the nucleus (e.g. RNA polymerases), mitochondria (e.g. cytochrome c oxidase), and endoplasmic reticulum (e.g. calreticulin) [2–4]. Due to the essential nature of some of these proteins, all organisms are challenged with obtaining sufficient levels of zinc for incorporation into newly synthesized proteins. A further complicating factor is that excessive levels of zinc are toxic to cells. As a consequence, zinc acquisition, compartmentalization, storage, and efflux need to be tightly regulated to maintain zinc at a level that is sufficient, but not toxic to cell metabolism.
In many organisms zinc-responsive transcription factors maintain zinc homeostasis by controlling the expression of genes that are required for the transport of zinc into and out of the cytosol. In eukaryotes these zinc transport proteins commonly belong to either the Zrt- Irt- like protein family (ZIP) or CDF family. Members of the ZIP family typically facilitate zinc uptake or the release of zinc from intracellular stores, whereas the CDF family members usually transport zinc into the lumens of intracellular compartments or out of a cell [5]. As zinc transport by a ZIP family member typically results in an increase in cytosol zinc levels, the expression of genes encoding ZIP family members is often up-regulated when zinc is limiting [6]. As an example, in Saccharomyces cerevisiae the transcriptional activator Zap1 controls the expression of genes encoding ZIP family members required for zinc uptake (Zrt1 and Zrt2) and release of zinc from the vacuolar stores (Zrt3) [7]. As Zap1 is active in zinc-limited cells and is inactive when zinc is in excess, the expression of ZRT1-3 increases when cells need zinc. Importantly, as zinc transport into the cytosol by the ZIP proteins inactivates Zap1, a negative feedback loop is created that prevents zinc from reaching toxic levels.
Negative feedback circuits also control the expression of CDF family members. In humans, the metal-responsive transcription factor 1 (MTF-1) regulates the expression of ZnT1, an essential CDF family member that is required for zinc efflux from cells [8]. MTF-1 is activated by excess zinc in the cytosol, which in turn transcriptionally induces ZnT1 expression when zinc is high. Similarly, when dietary zinc levels are high in the nematode Caenorhabditis elegans, the high zinc-responsive factor 1 (HIZR-1) induces the expression of CDF family members required for the excretion of zinc from intestinal cells (ttm-1b) and storage of zinc in intestinal gut granules (cdf-2) [9]. As the end result of these transcriptional changes is a reduction in cytosolic zinc levels, thereby inactivating MTF-1 and HIZR-1, a negative feedback loop is created that prevents the cytosol from being depleted of zinc.
Although a number of genes encoding transporters required for zinc uptake, storage, and efflux are subject to negative feedback control, the expression of some CDF family members increases in zinc-limited cells. For example, Zrg17 and Msc2 are two CDF family members from S. cerevisiae that form a heterodimeric complex that transports zinc into the endoplasmic reticulum [10]. Although this complex transports zinc out of the cytosol, ZRG17 is a Zap1-target gene that is expressed at higher levels in zinc-deficient cells [11]. While this regulation at first seems counterintuitive, as it would further deplete zinc from the cytosol, the induction of ZRG17 by Zap1 is critical for preventing the unfolding of proteins in the endoplasmic reticulum under this condition [11]. As zinc transport by the Zrg17/Msc2 complex would also further increase Zap1 activity, the zinc-dependent regulation of ZRG17 presumably results in a positive feedback circuit to supply zinc to compartmentalized proteins when the cytosol is limited for zinc.
The regulation of ZRG17 by Zap1 illustrates a mechanism of how zinc can be supplied to an intracellular compartment in a zinc-limited environment. As few other studies have examined the regulatory circuits that maintain zinc levels in organelles during periods of zinc starvation, the goal of this work was to determine if related mechanisms were present in the distantly related yeast S. pombe. We chose to use S. pombe because multiple aspects of zinc homeostasis differ between fission and budding yeast. These differences include the transcription factor used to control zinc homeostasis (Loz1 vs. Zap1), the primary site for the storage of excess zinc (endoplasmic reticulum vs. vacuole), and the presence of metallothioneins that preferentially bind divalent metal ions such as zinc (Zym1 from S. pombe) or monovalent ions such as copper (Cup1 from S. cerevisiae) [12–16]. Another difference between fission and budding yeasts is the subcellular localization of zinc transporters within the secretory pathway. In S. cerevisiae two CDF family members, Zrc1 and Cot1, transport zinc into the vacuole [17]. In S. pombe, the homolog of Zrc1, named Zhf1, transports zinc into the endoplasmic reticulum, and the homologs of Msc2 and Zrg17 (named Cis4 and Zrg17 respectively) form a complex that localizes to the cis-Golgi [18, 19]. The biological significance of these differences in subcellular localization of the CDF family members between budding and fission yeasts is currently unclear.
To gain a better understanding of the mechanisms that control the supply of zinc to organelles, we used multiple genetic approaches to determine the extent to which three CDF family members from S. pombe (Zhf1, Zrg17, and Cis4) facilitate zinc transport out of the cytosol under conditions of zinc deficiency and zinc excess. We found that deletion of zhf1 results in a strong growth defect when zinc is in excess and that deletion of zrg17 or cis4 leads to a mild growth defect in the presence of the zinc chelator EDTA. These latter results suggest that the Cis4/Zrg17 complex may play an important role under zinc deficiency conditions. To further investigate whether transport via Cis4 and Zrg17 is affected by cellular zinc status we developed methods to monitor changes in cytosolic zinc availability in fission yeast. These analyses revealed that that cis4Δ and zrg17Δ cells accumulate higher levels of zinc in the cytosol under conditions of zinc deficiency, while zhf1Δ cells accumulate higher levels of zinc in the cytosol when zinc is not limiting. We also show that the transcription of zhf1, cis4, and zrg17 genes is not dependent upon zinc. These results reveal that different CDF family members have complementary roles in transporting zinc out of the cytosol. They also suggest that either the activities or the properties of different CDF family members are fine-tuned to transport zinc out of the cytosol under different environmental zinc stresses.
Three members of the CDF family transport zinc into the secretory pathway in fission yeast: Zhf1, Cis4, and Zrg17 [13, 15, 18, 19]. Although previous studies have shown that Zhf1 is required for growth in the presence of high zinc, relatively little was known about the roles of Cis4 and Zrg17 in zinc homeostasis. To determine if Cis4 and Zrg17 were necessary for growth under low or high zinc conditions, serial dilutions of cis4Δ and zrg17Δ cells were plated onto zinc-limiting (EMM + 100 μM EDTA) or zinc-replete medium (EMM + 0–200 μM zinc) (Fig 1). Cells lacking the Zrt1 or Zhf1, which are required for survival during zinc deficiency or zinc toxicity respectively [18], were also plated as controls. In the presence of 100 μM EDTA, cis4Δ cells exhibited a slight growth defect relative to the wild-type. zrg17Δ cells had a modest growth defect under all conditions, but grew more slowly in the presence of EDTA relative to cis4Δ. As Cis4 and Zrg17 form a heteromeric complex, these results are consistent with Cis4 and Zrg17 playing an important role in supplying zinc to the secretory pathway when zinc is limiting. The slower growth of zrg17Δ relative to cis4Δ also suggests that Zrg17 may have additional functions that are independent of Cis4 and zinc.
To determine if Cis4 and Zrg17 were required for zinc transport out of the cytosol of zinc-limited cells, we developed constructs to express the genetically encoded ZapCY1 and ZapCY2 zinc-responsive FRET sensors in the cytosol of fission yeast. The ZapCY1/2 sensors have been widely used to monitor dynamic changes in intracellular zinc levels [20]. Both sensors contain the regulatory zinc fingers 1 and 2 from the transcription factor Zap1, flanked by citrine YFP, hereafter referred to as YFP, and eCFP (Fig 2A). In zinc-limited environments, the Zap1 zinc finger domain 1 and 2 are largely unstructured. However, in the presence of zinc, the zinc finger domains fold together to form a single structural unit [21]. As this closed conformation brings together YFP and eCFP increasing FRET, the FRET signal in cells expressing ZapCY1 and ZapCY2 is directly coupled to intracellular zinc availability. Previous studies have shown that ZapCY1 binds zinc in vitro with an apparent dissociation constant of ~ 2.5 pM, while ZapCY2 contains substitutions within the zinc finger domains, which result in it binding zinc with an ~ 300 fold lower affinity [20].
To determine if the ZapCY1 and ZapCY2 FRET reporters were stably produced in S. pombe, strains expressing ZapCY1 and ZapCY2 from the constitutive pgk1 promoter were grown overnight in ZL-EMM, and the levels of each sensor examined by immunoblotting. Both reporters accumulated to similar levels in zinc-limited and zinc-replete cells indicating that their stability was not affected by zinc (Fig 2B). We also assessed the subcellular localization of each reporter in response to cellular zinc status using fluorescent microscopy. As shown in Fig 2C, there was strong fluorescence in cells expressing ZapCY1 or ZapCY2, which was absent from cells transformed with the empty vector. The ZapCY1 and ZapCY2 proteins were both localized to the cytosol and nucleus, and were excluded from the vacuole. For unknown reasons, higher levels of the ZapCY2 reporter accumulated in the nucleus of zinc-replete cells.
A potential concern with using ZapCY1 and ZapCY2 to assess alterations in the labile pools of zinc in yeast is that both sensors bind zinc ions, which in turn might reduce the levels of zinc that are normally available for cellular metabolism. In previous studies we have shown that the expression of the Loz1 target genes zrt1 and zym1 is dependent upon intracellular zinc levels [22]. Specifically, zrt1 is expressed in zinc-limited cells and zym1 is expressed in zinc-replete cells. We therefore predicted that zrt1 and zym1 expression would be altered if the ZapCY1/2 FRET sensors interfere with zinc homeostasis. As shown in Fig 2D, the introduction of the ZapCY1/2 FRET sensors had no major effect on zym1 and zrt1 mRNA levels. In addition, when the FRET reporters were co-expressed with a zrt1-lacZ reporter, there were no differences in β-galactosidase activity when compared to cells expressing the vector (Fig 2E). Taken together the above results show that the ZapCY1/2 FRET sensors accumulate within the cytosol and nucleus of cells without any substantial effect on zinc homeostasis.
To determine if the ZapCY1 and ZapCY2 FRET sensors are able to detect dynamic changes in the labile pool of zinc in fission yeast, we measured the activity of each reporter in vivo following a ‘zinc shock’. In a zinc shock experiment cells are initially depleted of zinc by growing overnight in ZL-EMM, which leads to increased expression of zrt1 and high levels of Zrt1 on the plasma membrane. As zinc-limited cells are primed and ready to uptake zinc, zinc rapidly enters cells in a dose-dependent manner when it is added to the growth medium (Fig 3A). To examine the response of the high affinity FRET sensor to zinc shock, wild-type cells expressing ZapCY1 were grown overnight in ZL-EMM before being transferred to temperature-controlled microplate wells. Cells were excited at 434 nm and a FRET ratio calculated by dividing the intensity of the emission at 535 nm by the emission at 475 nm. The growth overnight in ZL-EMM resulted in a FRET ratio of 2.29 +/- 0.20 (Fig 3B and 3C, t = -5 min). This ratio remained constant until the addition of zinc, which led to a rapid increase in FRET in a dose-dependent manner (Fig 3B, 0.01–1 μM zinc). These changes in FRET occurred without affecting the stability of ZapCY1 (Fig 3D), consistent with the ZapCY1 sensor binding zinc and forming the closed conformation which brings YFP and CFP closer together. Zinc shocks with higher levels of zinc (10–1000 μM Zn2+) also led to a rapid increase in the FRET ratio (Fig 3B and 3C). However, the magnitude of the response was similar to that seen following a zinc shock with 1 μM zinc. We conclude from these results that a zinc shock with 1 μM Zn2+ results in sufficient levels of zinc accumulating within the cytosol and nucleus of cells to saturate the high affinity ZapCY1 sensor.
To assess the effects of a zinc shock on the low affinity reporter, similar experiments were performed with wild-type cells expressing ZapCY2. In these cells, growth overnight in ZL-EMM resulted in an initial FRET ratio of 1.57 +/- 0.05 (Fig 3E and 3F, t = 0 min). As ZapCY2 binds zinc with a lower affinity than ZapCY1, we predicted that higher levels of zinc would be needed to saturate ZapCY2. Consistent with this hypothesis, no significant increase in FRET was observed following a zinc shock with 0.1 μM Zn2+ and a modest response was seen with 1 μM Zn2+. A rapid increase in the FRET ratio to 1.92 +/- 0.03 was detected following a zinc shock with 10 μM Zn2+ (Fig 3E, t = 2 min). However, this FRET signal subsequently decreased until it reached a more constant ratio of ~ 1.75 after 15 minutes (Fig 3E, t = 15–90 min). As the zinc shock did not affect the stability of ZapCY2 (Fig 3G), these results are consistent with ZapCY2 rapidly binding zinc, and then zinc being lost to higher affinity zinc binding sites in the surrounding environment. Zinc shocks with higher levels of zinc (Fig 3E and 3F, 100 μM zinc) were sufficient to saturate the FRET sensor, but in contrast to ZapCY1, the FRET ratio slowly decreased with time. Thus, the ZapCY1 and ZapCY2 sensors are both able to detect changes in cytosolic zinc levels. However, higher levels of zinc are necessary to saturate ZapCY2 and the zinc bound to ZapCY2 is more readily lost to the surrounding environment.
To gain further evidence that the sensors were measuring changes in cytosolic zinc levels, ZapCY1 and ZapCY2 were introduced into cells lacking zrt1. In the absence of Zrt1, higher levels of zinc are needed during a zinc shock experiment to see an increase in total cellular zinc because cells rely on low affinity systems for zinc uptake (compare Fig 3A to Fig 4A). We therefore predicted that higher levels of zinc would be necessary to saturate ZapCY1 and ZapCY2 in zrt1Δ. Consistent with this prediction, higher levels of zinc were required to obtain a maximal FRET ratio change in zrt1Δ cells compared to the wild-type (Fig 4B–4E). As an example, a zinc shock with 1 μM zinc was sufficient to saturate ZapCY1 in the wild-type, but did not affect the activity of the ZapCY1 reporter in zrt1Δ (Fig 4F). The differences in FRET response were a result of loss of zrt1, as both reporters were expressed at similar levels to the wild-type, and a zinc shock had no effect on the stability of either reporter (Fig 4G–4I). Together, these results indicate that the FRET signal in cells is dependent upon the expression of zrt1 and also is consistent with the activity of both reporters being directly regulated by cytosolic zinc levels.
In our previous studies we found that genetic mutations that disrupt Loz1 function result in the constitutive de-repression of zrt1 transcription, leading to increased expression levels. Therefore, to assess the effects of overexpression of zrt1 on cytosolic zinc levels, ZapCY1 and ZapCY2 were introduced in loz1Δ cells and the FRET response was measured during a zinc shock experiment. Following the growth of loz1Δ ZapCY1 cells overnight in ZL-EMM, an initial FRET ratio of 3.3 +/- 0.5 was detected (Fig 5A and 5B), which is higher than the initial FRET ratio in cells expressing Loz1. Further, only a minor increase in FRET was seen following a zinc shock with 0.1–1000 μM zinc. As loz1Δ cells constitutively express zrt1, one explanation for the high initial FRET ratio in this mutant is that they have higher levels of zinc uptake leading to the saturation of ZapCY1. It was also possible that the ZapCY1 reporter was unable to respond to zinc in this genetic background. To distinguish between these possibilities, we used sodium pyrithione (NaPT) to artificially lower cytosolic zinc levels. Pyrithione is a membrane permeable ionophore that readily forms complex with zinc [23]. When 50 μM NaPT was added to wild-type ZapCY1 cells grown overnight in ZL-EMM, a small decrease in the FRET ratio consistent with this molecule binding or releasing accessible zinc within the cytosol (Fig 5C). Importantly, a rapid increase in FRET was seen when zinc was added to the NaPT treated cells. When a similar experiment was performed with loz1Δ cells, a large decrease in the FRET ratio was seen upon the addition of NaPT, which could be reversed by the addition of zinc (Fig 5D). These results indicate that the ZapCY1 reporter is functional in loz1Δ cells, and suggest that in the absence of strong chelators and ionophores it is saturated with zinc under all conditions.
To test whether the saturation of the ZapCY1 reporter in loz1Δ cells was a result of high zrt1 expression, we examined the activity of the ZapCY1 reporter in double mutants lacking loz1 and zrt1. In these cells a low FRET ratio of 1.8 +/- 0.2 was detected following growth overnight in ZL-EMM (Fig 5E). These results suggest that the high expression of zrt1 significantly contributes to the saturation of ZapCY1 in loz1Δ cells. We also noted that higher levels of total zinc accumulated in loz1Δ zrt1Δ when compared to zrt1Δ following a zinc shock (Fig 5F). These results suggest that Loz1 controls the expression of a second lower affinity zinc uptake system and/or regulates the expression of other genes that affect cytosolic zinc availability. Consistent with this hypothesis, a zinc shock with 1 μM zinc did not result in an increased FRET ratio in zrt1Δ, but did lead to a slow increase in FRET in loz1Δ zrt1Δ (compare Figs 4B and 5E). Similarly, the ZapCY1 reporter was close to saturation after a 30 min zinc shock with 10 μM Zn in loz1Δ zrt1Δ cells, and yet in zrt1Δ, a zinc shock with 10 μM Zn zinc only led to a slow gradual increase in FRET over 60 min.
As loz1Δ cells accumulate higher levels of zinc within the cytosol, we also assessed the effects of this allele on the response of the low affinity ZapCY2 reporter. In contrast to the response of ZapCY1, the basal FRET ratio in zinc-limited loz1Δ ZapCY2 cells was similar to the wild-type (compare Figs 3F to 5H). The responses of the ZapCY2 reporter to zinc also resembled those of the wild-type (Fig 5G and 5H). Thus, under conditions of zinc deficiency, loz1Δ cells accumulate higher levels of zinc in the cytosol/nucleus relative to the wild-type. However, when zinc is not limiting in these cells it is effectively buffered and/or transported out of the cytosol.
The above results show that the ZapCY1 and ZapCY2 sensors can be used in S. pombe to measure dynamic changes in the levels of labile zinc in the cytosol and nucleus. As deletion of cis4 or zrg17 resulted in a growth defect on low zinc medium, we used these sensors to test whether Cis4 and Zrg17 were necessary for zinc transport out of the cell under this condition. When a zinc shock experiment was performed with cis4Δ ZapCY1 cells, the starting FRET ratio was higher than that observed in the wild-type. Additionally, only a small increase in the FRET ratio was seen with 0.1 μM zinc (from 2.8 +/- 0.23 to 4.0 +/-0.2) and also when cells were shocked with higher levels of zinc (1–1000 μM) (Fig 6A–6C). In contrast, the addition of NaPT resulted in a large decrease in the FRET ratio, which could be reversed by the addition of zinc (Fig 6D). For the most part, similar trends were seen with zrg17Δ ZapCY1 and cis4Δ zrg17Δ ZapCY1. However, for zrg17Δ cells the maximum FRET ratio was slightly higher than that observed with cis4Δ cells; and a zinc shock with 0.1 μM zinc was not sufficient to totally saturate ZapCY1 (Fig 6E–6G and S1 Fig). As ZapCY1 was largely saturated in cis4Δ and zrg17Δ cells following growth overnight in ZL-EMM, these results are consistent with Cis4 and Zrg17 being required for the transport of zinc out of the cytosol under conditions of zinc deficiency.
To determine if the higher levels of zinc that accumulated in cis4Δ and zrg17Δ also affected zinc binding to low affinity sites, similar experiments were performed with cells expressing ZapCY2. A zinc shock with 10 or 1000 μM zinc resulted in smaller increase in FRET compared to the wild-type (Fig 6H and 6I and S1 Fig). The signal also decreased over time. While it is possible that Cis4 and Zrg17 transport zinc out of the cytosol following a zinc shock, the decrease in FRET response in these mutants suggests that other mechanisms that are independent of Cis4 and Zrg17 protect the cytosol from accumulating high levels of labile zinc.
As Zhf1 is predicted to play the primary role in protecting cells from zinc toxicity, we next examined the activity of ZapCY1 and ZapCY2 in strains lacking zhf1. In zhf1Δ ZapCY1 cells grown overnight in ZL-EMM, the basal FRET ratio and response of this sensor to zinc shocks with 0.1 and 1 μM zinc were similar to those seen in wild-type cells (Fig 7A and 7B). In contrast, a zinc shock with 10 μM zinc led to a rapid increase in the FRET ratio, followed by an immediate decrease. After these rapid changes the FRET ratio slowly increased for the remainder of the experiment. A similar response was seen with zinc shocks with higher levels of zinc, with the exception that it took longer (~30 min) to see the increase in FRET ratio. To test whether this atypical response was a result of the instability of the ZapCY1 reporter in zhf1Δ cells, immunoblot analysis was used to examine the stability of ZapCY1 during a zinc shock with 100 μM Zn2+. As shown in Fig 7C and 7D, elevated levels of a lower molecular weight band accumulated in this strain (see asterisk), suggesting that ZapCY1 was more prone to degradation in this strain relative to others. Despite this higher level of degradation, there were no changes in the levels of the full-length reporter and experiments using the zinc chelator NaPT resulted in FRET profiles that resembled those observed in the wild-type (Fig 7E). Although we do yet understand the zinc-dependent changes in the FRET response in zhf1Δ cells, the observation that they are not observed in the presence of NaPT suggests that they result from zinc accumulating in the cytosol of this strain.
To gain further evidence that Zhf1 protects the cytosol from excess zinc, similar experiments were performed with zhf1Δ ZapCY2. In these cells, zinc had no effect on the stability of the full length reporter and a zinc shock with 1 μM zinc was sufficient to saturate ZapCY2 (Fig 7F–7I). The FRET ratio after a zinc shock with 1 μM zinc also remained high for the duration of the experiment. These results reveal that higher levels of zinc accumulate in the cytosol of zhf1Δ following a zinc shock and indicate that Zhf1 has a central role in removing labile zinc from the cytosol.
The above results suggest that the Cis4/Zrg17 complex plays a primary role in the transport of zinc out of a zinc-limited cytosol, while Zhf1 has the dominant role in transporting labile zinc from the cytosol. In S. cerevisiae the expression ZRG17 increases under conditions of zinc deficiency and this increase is critical for normal endoplasmic reticulum function under this condition [11]. To determine if the expression of cis4 and zrg17 was dependent on zinc in fission yeast we used S1 nuclease analysis to examine mRNA levels in wild-type and loz1Δ cells grown under zinc-limiting and zinc-replete conditions. As shown in Fig 8A, cis4 and zrg17 transcripts accumulated under all conditions indicating that their expression is not affected by zinc or Loz1. The levels of zhf1 mRNAs were also not regulated by zinc and Loz1, consistent with previous studies that have demonstrated experimentally that the expression of zhf1 is not affected by cellular zinc status [13, 24].
As deletion of cis4 or zrg17 resulted in increased saturation of the high affinity ZapCY1 reporter, we also used S1 nuclease analysis to test whether these mutants accumulated sufficient levels of zinc in the cytosol to trigger increased Loz1-mediated gene repression. The rationale for these experiments is that Loz1 represses target gene expression when zinc levels are high. As a consequence, if higher levels of zinc accumulate in the cytosol of cis4Δ and zrg17Δ, this could result in increased repression of Loz1 target genes. When cells were grown under zinc-limiting conditions, lower levels of zrt1 transcripts accumulated in cis4Δ and zrg17Δ cells relative to the wild-type control (Fig 8B). These results are consistent with the Cis4/Zrg17 complex transporting zinc out of cytosol of zinc-limited cells.
Yeast are useful model systems to study zinc homeostasis, as they are able to survive in low zinc environments and rapidly adapt to conditions of zinc excess. In this work we took advantage of these properties by examining the activity of the zinc-responsive ZapCY FRET reporters following overnight growth in a zinc-limited medium and during a zinc shock. We show that ZapCY1 and ZapCY2 are both able to measure dynamic changes in cytosolic zinc levels in fission yeast and that higher levels of zinc are necessary to saturate ZapCY2. We also show that there is a transient increase in FRET following a zinc shock in wild-type cells expressing ZapCY2, suggesting that zinc bound to this sensor exchanges with other ligands within the cytosol that can bind or buffer zinc. As ZapCY1 is able to detect zinc ions binding to high affinity sites within proteins, and ZapCY2 detects binding to low affinity sites, these sensors create useful tools for monitoring the factors that influence cytosolic zinc ion availability and zinc ion binding within the cytosol.
To identify additional factors that affect the levels and availability of zinc within the cytosol, we used ZapCY1/2 to test whether Cis4, Zrg17, and Zhf1 have redundant or complementary roles in zinc transport out of the cytosol. We found that deletion of cis4 or zrg17 resulted in higher levels of saturation of ZapCY1 under conditions of zinc deficiency, whereas deletion of zhf1 had little effect on the saturation of ZapCY1 under this condition. In contrast, significantly lower levels of zinc were necessary to saturate ZapCY2 in zhf1Δ cells compared to cis4Δ or zrg17Δ following a zinc shock. We propose that the Cis4/Zrg17 heterodimer preferentially transports zinc out of the cytosol into the secretory pathway under zinc-limiting conditions, whereas Zhf1 has the dominant role in transporting labile zinc out of the cytosol when zinc is not limiting (Fig 9). In this model, the transport activity of the Cis4/Zrg17 heterodimer ensures that zinc is supplied to the secretory pathway under zinc-limiting conditions. As a reduction in cytosolic zinc levels also triggers the inactivation of Loz1 and increased expression of the zrt1 zinc uptake system, cells are able to balance the levels of zinc uptake with zinc flux out of the cytosol. Cells face a different challenge when zinc is in excess, as too much zinc is toxic to cell metabolism. Under these conditions, the dominant role of Zhf1 results in excess zinc being directed to intracellular stores, protecting the cytosol and other organelles from the toxic effects of too much zinc.
A key question that our studies raise is what is the mechanism by which individual CDF family members preferentially transport zinc under zinc-limiting or zinc-replete conditions? Studies with S. cerevisiae have revealed much of what we know about the ability of CDF proteins to transport zinc under varying conditions of zinc stress. In this yeast, the Msc2/Zrg17 complex facilitates the transport of zinc into the endoplasmic reticulum, whereas Zrc1 and Cot transport zinc into the vacuole [10, 14]. One factor that affects zinc transport via the Msc2/Zrg17 complex is ZRG17 expression. ZRG17 is a Zap1 target gene that is expressed at higher levels in zinc-deficient cells [7, 11]. Importantly, in the absence of the Zap1-dependent induction of ZRG17, zinc-deficient cells experience greater levels of ER stress [11]. These results suggest that the levels of Zrg17 protein limit zinc transport by the Msc2/Zrg17 complex and that the increase in ZRG17 expression is critical for normal ER function under conditions of zinc-deficiency. Recent studies have also revealed that higher levels of MSC2 mRNAs accumulate in zinc-deficient cells [25]. The levels of Msc2 may also be an important factor that limits zinc transport by the Msc2/Zrg17 complex. While changes in gene expression are an integral part of zinc transport into the ER under zinc-deficient conditions, it is also important to note that ZRC1 is a Zap1 target gene, and yet overexpression of ZRC1 does not affect cytosolic zinc availability in zinc-deficient cells [26]. These results suggest that Zrc1 does not play a significant role in transporting zinc out of the cytosol under this condition. They also reveal that increased expression of a zinc transport gene does not necessarily result in more zinc being transported out of the cytosol, and that other factors likely affect zinc transporter function.
As the expression of cis4, zrg17, and zhf1 is not dependent upon zinc, what other factors could affect their ability to transport zinc out of the cytosol? One possibility is that there are intrinsic differences in the ability of Zhf1 and Cis4/Zrg17 to transport zinc. For example, if the affinities of the zinc binding sites on the cytosolic face of Zhf1 were weaker than those of the Cis4/Zrg17 heterodimer, this latter complex may only be able to acquire zinc when the cytosolic zinc pools are less saturated. An alternative possibility is that the labile pool of zinc that is accessible to Zhf1 under zinc-limiting conditions is dependent on the presence of an active Cis4/Zrg17 heterodimer. Potential mechanisms that could lead to a pool of labile zinc that is inaccessible to Zhf1 include an increase in the total number of buffering components in the cytosol (i.e. the total amount of zinc remains the same and the buffering capacity increases) and/or tighter buffering of cytosolic zinc (preventing zinc from being available to the weaker binding sites of Zhf1). While future experiments are necessary to determine the precise nature of the buffering components of the cytosol, and the affinity of different CDF transporters for zinc, it is noteworthy that Cis4/Zrg17 and Msc2/Zrg17 complexes both appear to have a primary role in transporting zinc out of the cytosol into the secretory pathway under zinc-limiting conditions. In addition, Zhf1 and Zrc1 both facilitate the transport of zinc from the cytosol when zinc availability is not limited. These similar functions suggest that at least some features of these transporters are conserved in fission and budding yeast.
In addition to the potential differences above, multiple other factors could affect the function of the Cis4/Zrg17 heterodimer and Zhf1. For example, in yeast and humans, zinc transporters from the ZIP family are targeted for degradation in response to high zinc [27, 28]. Although it is currently not known if the stability or activity of each of the S. pombe CDF proteins are regulated at a post-translational level in response to cellular zinc levels, proteomic studies that compared the copy numbers of proteins in fission yeast during vegetative growth in minimal medium revealed the presence of ~12,000–14,000 Zhf1 molecules/cell, ~3000–6000 Zrg17 molecules/cell, and ~ 1300–5500 Cis4 molecules/cell [29, 30]. The higher levels of Zhf1 relative to Zrg17 and Cis4 may therefore be one factor that contributes to Zhf1 having the principal role in transporting zinc out of the cytosol in a zinc-replete environment. In addition to factors that directly affect the function of zinc transport proteins, it is unclear if the subcellular localization of zinc transporters or their local environment affects their function. Moreover, relatively little is known about the molecules that buffer zinc within organelles and the cytosol, and whether the buffering capacity of an organelle for zinc, and/or the number and affinity of zinc-binding proteins within a compartment influences zinc transport. Thus, future studies with CDF proteins from S. pombe and other organisms are warranted to identify additional factors that alter zinc transport function.
We also examined the effects of the loz1Δ allele on cytosol zinc availability. We had previously found that loz1Δ cells constitutively express zrt1 and hyperaccumulate zinc when excess zinc is present in the growth medium [22]. Although loz1Δ cells hyperaccumulate zinc, paradoxically they have a more severe growth defect under zinc-deficient conditions compared to zinc replete (Fig 1). Here we find that the loz1Δ allele results in the saturation of the high affinity ZapCY1 sensor following growth overnight in ZL-EMM, indicating that this mutant accumulates higher levels of zinc in the cytosol relative to the wild-type. We also find that the response of the ZapCY2 reporter was similar to that of the wild-type, revealing that labile zinc entering loz1Δ cells is rapidly removed into stores and/or is effectively buffered. These latter results reveal that other mechanisms that are independent of Loz1 help S. pombe to maintain zinc homeostasis. They also provide an explanation for the viability of the loz1 mutant in high zinc. Another observation that we made was that ZapCY1 was not saturated in double mutants lacking zrt1 and loz1, and that this double mutant accumulated higher levels of zinc in the cytosol relative to zrt1. These results indicate that the constitutive derepression of zrt1 is the primary reason for the saturation of ZapCY1 in loz1Δ cells. They also suggest that Loz1 regulates other genes that affect cytosolic zinc availability. Known Loz1 target genes include zrt1, as well as adh4 (alcohol dehydrogenase 4), gcd1 (glucose dehydrogenase 1), and SPBC1348.06c, which encodes a small fungal protein of unknown function [22, 31]. Loz1 also represses the expression of non-protein coding RNAs that interfere with the expression of the adh1 (alcohol dehydrogenase 1) and zym1 (zinc metallothionein 1) genes [22, 32]. The expression of adh1 and zym1 is therefore inverse to that of other Loz1 targets, in that they are repressed under conditions of zinc deficiency. Although no known Loz1 target gene other than zrt1 has a role in transporting zinc, altered expression of some of its targets could affect intracellular zinc availability. For example, as the loz1Δ allele results in the constitutive repression of adh1, which encodes the abundant zinc binding protein Adh1, the lower levels of this protein could result in higher levels of zinc being available for other proteins. Thus, future studies to identify new Loz1 target genes and to examine the roles of existing target genes in controlling intracellular zinc availability may provide additional insight into factors affecting zinc homeostasis.
The ability of some CDF proteins to transport zinc is a manner that is dependent upon the levels of ‘labile’ or ‘readily available’ zinc in the cytosol could be of particular importance in organisms that express large numbers of CDF family members. For example, humans express 10 CDF family members (named ZnT1-10), while Arabidopsis thaliana and C. elegans each express 14 family members [33]. Potential reasons for why these organisms have so many CDF proteins include that they have unique subcellular localizations, different metal ion specificities, and/or that they have more specialized roles in supplying zinc to smaller subsets of proteins [5, 34–38]. Some genes encoding CDF proteins also show tissue- or developmental- specific expression patterns, while others are regulated by zinc and/or by hormonal or stress-responsive signaling pathways [5, 39, 40]. Although it is currently unclear in other organisms if the activity of specific CDF proteins is dependent upon cellular zinc status, the conserved role of this family in supplying zinc to organelles and storage compartments raises the possibility that the activity of other CDF proteins may also be fined tuned according to cytosolic zinc ion availability.
To generate the strains used for the FRET analysis, the plasmids pZapCY1 and pZapCY2 were linearized with NruI and were integrated into the leu1-32 locus of the wild-type strain JW81 (h- ade6-M210 leu1-32 ura4-D18) [41]. All other strains expressing ZapCY1 or ZapCY2 were generated from genetic crosses with the wild-type ZapCY1 (ABY795) or WT ZapCY2 (ABY797) with the respective mutant. The strains co-expressing the ZapCY1 and ZapCY2 with the zrt1-lacZ reporter were generated from genetic crosses with JW81 containing the integrated reporter TN-zrt1-lacZ [24]. To generate zinc-deficient and zinc-replete cells, yeast strains were initially grown to exponential phase in the nutrient rich YES medium. Cells were then spun down and washed twice in ZL-EMM, a derivative of Edinburgh minimal medium that lacks zinc (ZL-EMM). Washed cells were then diluted to 0.02 OD600 with fresh ZL-EMM and were grown for 16 hrs at 31°C in ZL-EMM or in this medium supplemented with 1, 10, or 100 μM ZnCl2. For all zinc shock experiments, cells were grown as described above in ZL-EMM without zinc. The indicated amount of zinc (0.01–1000 μM ZnCl2) was then added to induce the zinc shock.
The plasmids pZapCY1 and pZapCY2 were generated by PCR amplifying the coding regions for ZapCY1 and ZapCY2 from the vectors pcDNA3.1-ZapCY1 and pcDNA3.1-ZapCY2 respectively, with primers containing EcoRI and BamHI restriction sites. The ZapCY1/2 PCR products were then digested with EcoRI and BamHI and cloned into similar sites into the vector JK-pgk1-adh4T. The vector JK-pgk1-adh4T is a derivative of JK148 that contains 840 bp of the pgk1 promoter inclusive of its 5’UTR and 726 bp of the adh4 terminator. It was generated by initially PCR amplifying the pgk1 promoter with primers containing KpnI and EcoRI restriction sites. KpnI- and EcoRI- digested PCR products were then cloned into the vector JK148 to generate JK-pgk1. The adh4 terminator was cloned using a similar approach with the exception that primers were designed to introduce the adh4 PCR product into the BamHI/SacI sites of JK-pgk1.
β-Galactosidase assays were performed as described previously [42]. Activity units were calculated as follows: (ΔA420 x 1000)/(min x ml of culture x culture absorbance at 600 nm). For AAS 10 ml of cells were grown in ZL-EMM as described above. After the OD600 was measured, the indicate amount of zinc was added at t = 0 min. Cells were then grown at 31°C with shaking and 1.5 ml aliquots removed at the indicated time point. To remove extracellular zinc, cells were washed twice with 0.5 M EDTA and twice with ddH2O. Cell pellets were then digested by boiling in 150 μl of metal free nitric acid for 45 min and the zinc content measured using a SpectrAA 220FS Atomic Absorption Spectrometer. The final zinc concentration/cell was calculated by comparing the readings to a standard curve generated using a zinc standard (Sigma 18827). All values are the average from three independent experiments and error bars represent standard deviations.
For FRET experiments, cells were grown for 16 hrs in ZL-EMM as described above. ~2.5 x 106 cells were directly transferred to temperature controlled 96 well plates and the FRET emission intensities measured using spectrofluorometry using the following excitation and emission wavelengths: eCFP excitation 434 nm / emission 474 nm, and FRET excitation 434 nm / emission 535 nm. The FRET ratio was calculated by dividing the FRET emission intensity by the eCFP emission intensity. All average values show the mean FRET ratio from three independent experiments that were performed on independent days. Error bars show standard deviations.
For immunoblotting, total protein extracts were prepared by a trichloroacetic acid precipitation. Proteins were separated by SDS/PAGE analysis using a 10% resolving gel before transfer to nitrocellulose membranes. Proteins were detected using anti-GFP (Sigma G1544) or anti-Actin (Abcam ab3280), and secondary antibodies IR-Dye800CW conjugated anti-mouse IgG (LICOR) and IRDye680 conjugated anti-rabbit IgG (LICOR). Signal intensities were measured using an Odyssey infrared imaging system. For RNA analysis, total RNA was purified using hot acidic phenol method. RNA blots and S1 nuclease analyses were performed as described previously [32, 43]. Probes for the RNA blot analyses were generated using the MAXISCRIPT T7 kit (Ambion) according to manufacturers instructions, whereas probes for the S1 nuclease analyses were generated by 5’ end labeling the following oligonucleotides: zrg17 5’-GATCACTAATAGTTACAGAGACATTATTATTTATAGGGTTTTGAATCTGAATAGCAGTCGGGATG- 3’, cis4 5’- CGAACGCAGAAGAATTAACATTCATTTTTGTCGTCAGGAACACCCAAAAGCTGTGGTTGAC-3’, zhf1 5’-GTTGCCAGCCATATGTGTATTTTGGTTCGTGAGATGTTGAATGTGCTAGACGAGTAGCCCA-3’, zrt1 5’- CCATATTCGTTGAATTCATTGGCATCACCTCCACAAGTCACAGTAGCAGAGCTATCATCGTC-3’, and act1 5’-GTCCCATACCTACCATAATACCATGGTGACGGGGTCTACCGAC-3’. Act1 probes were diluted with unlabeled probe where indicated. Fluorescent microscopy of live cells was performed with an Olympus FV 1000 Filter Confocal system, using filter sets for GFP.
Data are presented as the mean ± standard deviation (SD). Statistical analyses were performed using GraphPad Prism 5 software (GraphPad Software, La Jolla, CA, USA). Where appropriate, data were analyzed by a Student unpaired t-test. A p value of <0.05 was considered statistically significant.
|
10.1371/journal.pntd.0000690 | Secondary Syphilis in Cali, Colombia: New Concepts in Disease Pathogenesis | Venereal syphilis is a multi-stage, sexually transmitted disease caused by the spirochetal bacterium Treponema pallidum (Tp). Herein we describe a cohort of 57 patients (age 18–68 years) with secondary syphilis (SS) identified through a network of public sector primary health care providers in Cali, Colombia. To be eligible for participation, study subjects were required to have cutaneous lesions consistent with SS, a reactive Rapid Plasma Reagin test (RPR-titer ≥1∶4), and a confirmatory treponemal test (Fluorescent Treponemal Antibody Absorption test- FTA-ABS). Most subjects enrolled were women (64.9%), predominantly Afro-Colombian (38.6%) or mestizo (56.1%), and all were of low socio-economic status. Three (5.3%) subjects were newly diagnosed with HIV infection at study entry. The duration of signs and symptoms in most patients (53.6%) was less than 30 days; however, some patients reported being symptomatic for several months (range 5–240 days). The typical palmar and plantar exanthem of SS was the most common dermal manifestation (63%), followed by diffuse hypo- or hyperpigmented macules and papules on the trunk, abdomen and extremities. Three patients had patchy alopecia. Whole blood (WB) samples and punch biopsy material from a subset of SS patients were assayed for the presence of Tp DNA polymerase I gene (polA) target by real-time qualitative and quantitative PCR methods. Twelve (46%) of the 26 WB samples studied had quantifiable Tp DNA (ranging between 194.9 and 1954.2 Tp polA copies/ml blood) and seven (64%) were positive when WB DNA was extracted within 24 hours of collection. Tp DNA was also present in 8/12 (66%) skin biopsies available for testing. Strain typing analysis was attempted in all skin and WB samples with detectable Tp DNA. Using arp repeat size analysis and tpr RFLP patterns four different strain types were identified (14d, 16d, 13d and 22a). None of the WB samples had sufficient DNA for typing. The clinical and microbiologic observations presented herein, together with recent Cali syphilis seroprevalence data, provide additional evidence that venereal syphilis is highly endemic in this region of Colombia, thus underscoring the need for health care providers in the region to be acutely aware of the clinical manifestations of SS. This study also provides, for the first time, quantitative evidence that a significant proportion of untreated SS patients have substantial numbers of circulating spirochetes. How Tp is able to persist in the blood and skin of SS patients, despite the known presence of circulating treponemal opsonizing antibodies and the robust pro-inflammatory cellular immune responses characteristic of this stage of the disease, is not fully understood and requires further study.
| Venereal syphilis is a sexually transmitted disease caused by the bacterium Treponema pallidum (Tp). We describe 57 patients (age 18–68 years) from Cali, Colombia diagnosed with secondary syphilis (SS). Most were women (64.9%); predominantly Afro-Colombian (38.6%) or mestizo (56.1%), and all of low socio-economic status. Three (5.3%) were newly diagnosed with HIV infection at study entry. The typical palmar and plantar rash of SS was the common clinical finding (63%). Whole blood (WB) samples and skin biopsies were assayed for Tp DNA by using molecular methods. 46% of the WB samples had circulating Tp DNA and 64% were positive when the DNA was extracted on the same day of collection. Tp DNA was also present in the skin of 66% (12/26) of biopsies tested by PCR. We conclude that primary care providers in countries like Colombia need to remain highly vigilant for the clinical presentation of SS. The study also provides, for the first time, qualitative and quantitative evidence that untreated SS patients have significant numbers of spirochetes in blood and skin, and that this occurs despite the known presence of circulating anti-treponemal antibodies and strong cellular immune responses associated with this stage of the disease.
| Syphilis is a sexually transmitted disease (STD) caused by the spirochetal bacterium Treponema pallidum (Tp) subspecies pallidum [1], [2]. Despite the existence of inexpensive and effective antibiotic treatment regimens, syphilis continues to be a major public health problem. According to the most recent World Health Organization (WHO) estimates, approximately 10.6 million new syphilis cases occur yearly throughout the globe [3]. Although venereal syphilis has re-emerged in developed countries [4], most individuals (>90%) who acquire the disease reside in less affluent regions of the world [3]. In Cali, Colombia, our Latin American study site, the yearly incidence of venereal syphilis over the last decade was estimated to be around 32 cases per 100,000 (GE Aristizabal, City of Cali Health Department's STD Division - personal communication). This approximation is significantly higher than in Western Europe [5] or in the United States [4], and based on the very high rates of well documented gestational and congenital syphilis rates in the city [6], it very likely underestimates the true prevalence of venereal syphilis in Cali. Indeed, in a recent study conducted by our collaborators [7], 1.2–6.8% of 23,190 sexually active 15–24 year old men and women, from different socioeconomically deprived Cali districts “comunas”, had serum RPR values ≥1∶8. In the same study, 13.8% of men who have sex with men (MSM), 28.8% of female commercial sex workers and 39.2% of transsexuals in Cali were also found to be sero-positive for syphilis [7]. Strategies to control sexual transmission of Tp are, thus, urgently needed in Colombia, not only because of the harmful consequences of syphilis to infected pregnant women or their unborn children [5], [8], [9], but also because of the strong association of venereal syphilis with an increased risk for acquiring and transmitting human immunodeficiency virus (HIV) [4], [10]–[12]. To begin to curb the spread of venereal syphilis it is very important that health care providers become more adept at distinguishing the typical and atypical signs and symptoms associated with early syphilitic infection.
Unlike most invasive bacterial infectious diseases, venereal syphilis is a multistage illness with clinical manifestations that reflect the propensity of Tp to disseminate systemically and to induce persistent chronic inflammation in diverse tissues and organ systems [1], [2], [13], [14]. Infection begins when the bacterium comes in contact with skin or mucosal membranes, multiplying locally over several days, while simultaneously disseminating through blood vessels and lymphatics [2], [15]. The appearance of a painless ulcer, more commonly known as a “chancre”, typically only appears 2–4 weeks after the initial contact with the spirochete [2], [15], [16]. By this time, organisms have disseminated from the primary site of infection to various organ systems and throughout the dermis [2], [15], setting the stage for what is classically known as secondary syphilis. This stage of the disease, which is the principal focus of the current study, is characterized by the most overt systemic clinical features, including a variety of dermal manifestations as well as systemic signs and symptoms typically appearing within 4–10 weeks of the initial infection [13], [14]. Despite the evolving nature of the adaptive immune response [17], including the presence of opsonic antibodies [18], it may take weeks and in some cases months for the host to gain the upper hand against the invading pathogen, ultimately giving rise to an asymptomatic stage known as latent syphilis. During early latency (the first 4 years post-infection) patients may experience recurrences of spirochetemia as well as clinical relapses [13], [15], [19], both indicative of the host's inability to fully eradicate and control the bacterium. In due course, patients enter late latency and several years later 15–40% of them develops recrudescent forms of the disease; collectively referred to as tertiary syphilis [14], [20], [21]. How the bacterium disseminates from its primary or secondary sites of infection, and why it persists in its human host for extended periods of time despite the vigorous cellular and humoral adaptive immune responses it evokes [2], [17], remains unresolved.
Greater progress towards deciphering the pathogenesis of the paradoxical nature of venereal syphilis has been hampered due to the inability to readily propagate Tp in vitro, as well as the lack of a suitable inbred animal model to study the disease. Nevertheless, much can be learned about the pathogenesis of venereal syphilis through a combined analysis of the epidemiologic, clinical and microbiologic features of the various stages of the disease. In the current study we review the clinical, histopathologic and laboratory features of 57 patients diagnosed with SS through a network of public sector primary health care providers in Cali, Colombia. Concurrently we determined spirochetal DNA burdens in WB and skin samples from a subset of these patients by using a highly sensitive real-time PCR assay. In conjunction with available Cali syphilis seroprevalence data [7], this study provides clinical and microbiologic evidence that venereal syphilis is highly endemic in this region of Colombia. Our findings also make evident that in this population early syphilis patients may go undiagnosed and untreated for several weeks, in some cases for several months. We also provide, for the first time, quantitative and/or qualitative molecular evidence that spirochetes are present in significant numbers in skin and blood of untreated SS patients. Paradoxically, spirochetes persisted despite the known presence of circulating antitreponemal opsonizing antibodies in the serum of SS patients [18], as well as the robust pro-inflammatory dermal cellular immune response characteristic of this stage of the disease [17].
SS patients from Cali, Colombia were recruited from 2003 to 2009 as part of an ongoing syphilis immunology study [10]. Cali is the 3rd largest city in Colombia with a population of 2,139,535; many of whom belong to middle or low socio-economic strata (32.1% and 31.6% respectively). Out of 22 distinct comunas in the city, 11 are considered very poor, have less than adequate access to health care services and as already alluded to above, very high RPR seropositivity rates [7]. Five public health institutions called Empresas Sociales del Estado (ESEs) are strategically located throughout the city, and are responsible for providing regional health care to the local population residing in the various comunas (Figure 1). Prior to study initiation, nurses and physicians working in decentralized hospitals and health care centers affiliated to individual ESEs were trained by study personnel to properly recognize and treat early syphilis patients. Patients who met criteria for a diagnosis of SS and who agreed to participate in the study were referred by participating providers to the “Centro Internacional de Entrenamiento e Investigaciones Médicas” (CIDEIM) for further examination by study physicians and for confirmatory blood tests. At CIDEIM all participants were required to sign informed consent. Study procedures were reviewed and approved by the human subjects boards at the Connecticut Children's Medical Center, the University of Connecticut Health Center (UCHC), Walter Reed Army Institute of Research (WRAIR) and CIDEIM. The study protocol was also reviewed by ethics committee from each of the participating ESEs. The Institutional review board at the Centers for Disease Control and Prevention (CDC) approved the molecular analysis of Tp DNA in blood samples and skin biopsies obtained from secondary syphilis patients and controls.
The diagnosis of secondary syphilis was based on a compatible medical history, the appearance of characteristic skin or mucosal lesions (see below), a reactive RPR of ≥1∶4 dilutions, and a positive confirmatory treponemal test (Fluorescent Treponemal Antibody Absorption, FTA-ABS). All subjects had a complete physical examination performed by one of two trained dermatologists (AC or RT) who are well versed in the identification of venereal syphilis. Individual case histories, serologic data, and photographs of dermal and cutaneous lesions were then reviewed by at least one of two infectious disease experts (JR or JS). Serologic tests for syphilis and HIV were conducted for all participants at a centralized reference laboratory in Cali. Study subjects were not eligible for participation if they were <18 years old, if they were known to be HIV+ or if they had used antibiotics within 4 weeks prior to study entry. Pregnant women were also excluded from participation. Socio-demographic characteristics and relevant clinical and epidemiological information were obtained by means of a standardized questionnaire. Study dermatologists obtained skin biopsies from 11 patients who had representative syphilis dermal lesions. Whole blood and skin biopsies were also collected from a subset of patients for real time polymerase chain reaction (PCR) Tp DNA quantitation (see below). All patients were treated with 2.4 million units of intramuscular benzathine penicillin in accordance with Colombian public health guidelines, which are in accord with available CDC treatment guidelines [22]. Patients were asked to return within 45–60 days for a follow-up clinical examination and repeat RPR titers. When possible, partner notification and treatment were done by health care providers from the city of Cali's health department. Diagnosis and treatment of additional STDs was done through the patient's primary health care provider.
All patients had non treponemal (RPR), and treponemal (FTA-ABS) tests performed at a reference laboratory in Colombia. An ECLIA (Electro-chemiluminescence Immunoassay) HIV serum antibody test, a complete blood count (CBC) with manual differential, and quantitation of the erythrocyte sedimentation rate (ESR) were also done. All women underwent a serum pregnancy test (none were positive). The three positive HIV ECLIAs identified in study patients were subsequently confirmed by Western-blot analysis. HIV-positive patients were referred to a public health care sector HIV clinic for additional testing and treatment. Dark field microscopy was not required in this study. Punch biopsies from affected dermal sites were obtained by the study dermatologist from 11 patients deemed to have distinctive clinical manifestations classically associated with SS. Available tissues from these patients were stained with both hematoxilin-eosin (H&E) and Warthin-Starry silver stain (one (9%) revealed spirochetes by silver stain). Individual biopsy samples were then systematically analyzed by trained pathologists in Colombia and subsequently corroborated by a pathologist in the United States.
The clinical sensitivity of the polA real-time PCR was determined by testing blood spiked with either purified Tp DNA or Tp organisms. Fifty milliliters of peripheral blood was collected in EDTA tubes from a single donor. The blood sample was maintained at room temperature prior to performing spiking experiments. Tp was grown in the testis of New Zealand white rabbits and harvested as previously described [23]. The cultivation of T. pallidum in rabbits was approved by the CDC animal care committee. Treponemes were diluted to a concentration of 1.5×107/ml using prewarmed TpCM (Tp cultivation medium). A ten-fold serial dilution of the suspension was performed using prewarmed TpCM and a 200µl aliquot of each ten-fold dilution was added in duplicate to 1.8 ml of blood. One set of tubes was kept at room temperature for 1 hr while the second set was placed in a refrigerator for 26 hrs. DNA was extracted from a 200µl aliquot of the 10−1 dilution of Tp organisms in TpCM using the QIAamp DNA Mini Kit (Qiagen, Valencia, CA). The purified DNA was serially diluted ten-fold in sterile PBS and a 200µl aliquot of each ten-fold dilution was added in duplicate to 1.8 ml of blood. Tubes were left at room temperature for 1 hr or stored at 4°C as for blood spiked with Tp organisms. DNA extraction from all spiked blood samples was achieved using the QIAamp DNA Midi kit (Qiagen) and the purified DNA was eluted in 500µl Buffer AE. A 10µl sample was tested in duplicate using a real-time PCR that targets the DNA polymerase I gene of Tp (polA).
Five to ten milliliters of whole blood were collected in tubes containing EDTA (1.8mg EDTA per milliliter of blood) from 26 SS patients enrolled during the last three years of the study. The first 15 WB samples collected were stored frozen for several days and subsequently shipped on dry ice overnight to UCHC. DNA was subsequently extracted from 0.4 ml of whole blood using the QIAamp DNA Blood Mini Kit (Qiagen) following procedures recommended by the manufacturer. For the last eleven patients enrolled into the study DNA was extracted from whole blood samples on site by CIDEIM personnel, and subsequently shipped on dry ice to UCHC. A 4 mm-punch skin biopsy of SS lesions was also obtained from the same group of patients and controls, snap-frozen and stored in liquid nitrogen in preparation for overnight transportation on dry ice to UCHC. Skin from three healthy controls was also obtained at CIDEIM and handled in the same fashion. Upon arrival at UCHC, DNA was extracted from all skin samples using the Qiagen DNAeasy Blood and Tissue kit (Qiagen). DNA samples from both the skin and WB were eventually shipped on dry ice to the Laboratory Reference and Research Branch, Division of STD Prevention, at the CDC in Atlanta, Georgia for diagnostic PCR testing. Samples were tested using a real-time PCR targeting the Tp polA gene (Gene Bank Accession No. U57757) as described below.
PCR amplification was performed using forward primer TP-1 (5′CAGGATCCGGCATATGTCC3′), reverse primer TP-2 (5′AAGTGTGAGCGTCTCATCATTCC3′), and probe TP-3 (5′CTGTCATGCACCA GCTTCGACGTCTT3′) as previously published [24] with some exceptions. The probe was labeled with Cyanine (Cy5) at the 5′ end and black-hole quencher 3 (BHQ3) at the 3′ end. PCR was performed in 50µl reaction volumes containing the following: 4µl of deoxynucleoside triphosphate mix (2.5mM of dATP, dCTP, dGTP, and 5.0mM of dUTP), 6µl of MgCl2 (25 mM), 0.2µM of each primer, 0.6U uracil N-glycosylase, 5U of AmpliTaq Gold polymerase, 5µl of 10× PCR buffer (All Applied Biosystems, Foster City, CA), 0.2µM of probe, and 12µl of template DNA. Thermocycling was performed in a Rotor-Gene 6000 instrument (Qiagen) as follows: 50°C for 2 min and 95°C for 10 min and 45 cycles of 95°C for 20 sec and 60°C for 1 min. Each PCR run included positive and negative (no template) control reactions. The Tp copy numbers for each WB specimen were extrapolated from the standard curve generated using ten-fold serial dilutions of purified Tp DNA. A human ribonuclease (RNase) P gene PCR assay was used, as previously described [25], to test for PCR inhibition in blood samples that were negative by the polA real-time PCR assay.
Strain typing was attempted for all WB and skin samples obtained from SS patients, which were Tp-positive by diagnostic PCR. PCR amplification and sizing of the 60-bp tandem repeats within the arp (acidic repeat protein) gene and PCR-restriction length polymorphism (RFLP) analysis of tpr (T. pallidum repeat) E, G, and J genes was done as previously described by Pillay et al. [26], [27], with two modifications. First, the tandem repeat region within the arp gene was amplified with a new primer pair, N1 (5′ATCTTTGCCGTCCCGTGTGC3′) and N2 (5′CCGAGTGGGATGGCTGCTTC3′) using the existing PCR conditions for the arp assay. Second, all PCR amplicons were analyzed on an Agilent 2100 Bioanalyzer (Agilent Technologies, San Diego, California).
A total of 57 patients (age 18–64 years, median 31 years) met the case definition for SS and were recruited by a network of primary health care providers located throughout Cali (Figure 1). All enrolled SS patients resided in comunas of the city of low or very–low socio-economic conditions. Most participants were either unschooled (10.5%) or had only partially completed elementary school education (59.6%). Because we did not directly target more affluent communities in Cali, we are unable to decisively conclude that SS principally affects underprivileged groups. Consistent with prior SS case series [14], the majority of SS patients enrolled were women (64.9%). Most participants were either Afro-Colombian (38.6%) or of mixed race (mestizo) (56.1%). Given that 25.3% of the general population in this region of the country is of Afro-Colombian background [28], it is evident that in this study blacks were proportionately over-represented. The majority of subjects self-reported a high-risk sexual behavior history (62.5%), including having multiple sexual partners, rarely using condoms, and/or having contact with commercial sex workers. Almost a quarter (24.6%) gave a history of illicit drug use and 59.6% stated that they consumed alcohol routinely. One male patient admitted to having had unprotected sex with other men as the principal risk factor for acquiring syphilis. 5.3% of the subjects enrolled in our study were newly diagnosed with HIV. A diagnosis of HIV co-infection is consistent with available epidemiologic and biologic evidence demonstrating that infection with Tp increases the likelihood of both transmitting and acquiring HIV [12], [29]–[31].
The mean duration of signs and symptoms at the time of presentation was 78.6 days (range 5–240), with a median duration of 30 days. A history of undocumented fever was not uncommon (15.8%); however, none of the patients enrolled was febrile at the time of the initial physical examination. Many patients reported mild to moderate flu-like symptoms (42.1%), and all had dermal and mucosal findings which were in accord with previously described dermatologic manifestations of SS [32], [33] (Table 1). The typical palmar and plantar exanthem of SS [13], [15], [19], [32]–[34] was the most common dermal manifestation (59.6%). As seen in figure 2, palmar and plantar lesions were often surrounded by hyperkeratosis and thin white rings or collar of scales; the latter has been classically known as Biett's collarette [34]. In agreement with Chapel's [32] and Mindel's classic descriptions of SS [32], [33], most subjects also had faint macular and papular eruptions, which were diffusely disseminated over the trunk and upper and lower extremities (Figure 3A–B). In many cases, the dermal lesions were hyperpigmented (Figure 3G), a finding which, not surprisingly, was more evident in darker skinned individuals. Condyloma lata, which are known to be highly infectious [1], [13], [14], were present in 15% of patients (Figure 3C and D). In one subject, several very large frambesiform, pustular lesions, a rare manifestation of the disease [34], were plainly evident over the naso-labial folds and lower jaw (not shown). Three patients had the characteristic patchy “moth-eaten-like” alopecia (Figure 3E), which resolved after penicillin treatment. Mucosal patches in genital areas and/or the oral mucosa, also commonly seen in this stage of the disease, were observed in 14% of patients (Figure 3F). Although early dissemination to the central nervous system (CNS) is known to occur in up to 40% of SS patients [1], [35], [36], none of the participants had overt clinical manifestations associated with meningitis or encephalitis (i.e. meningismus, cranial nerve disorders, visual changes or intense headache). The majority of patients (94.7%) had RPR titers ≥than 1∶16 and over a third (36.8%) had titers ≥1∶128. RPR titers at follow-up decreased at least four-fold in all patients followed at 40–60 days. Hematologic anomalies, indicative of the systemic inflammatory nature of this stage of the disease, were the norm in most patients studied. Indeed, more than half of all patients had an elevated erythrocyte sedimentation rate (ESR) (63.2%), 30% were anemic and several were lymphopenic (43.9%). It is possible, although unlikely, that the hematologic changes described herein are a reflection of other conditions which might be present in underserved individuals from Cali.
Table 2 summarizes the various histopathologic anomalies that were seen in lesional punch biopsies obtained from 11 of the 57 SS patients enrolled. In concert with prior histologic descriptions of SS [14], [34], [35], [37]–[40], skin biopsies revealed superficial and deep dermal cellular infiltrates of varying intensity, often in a perivascular distribution and primarily comprised of lymphocytes and plasma cells. Several other histologic patterns involving the epidermis were also seen. These included areas of focal spongiosis, basal vacuolar changes, parakeratosis, and epidermal acanthosis. Figure 4 depicts representative histopathologic abnormalities from four SS patients. In concert with the known low sensitivity of the Warthin-Starry stains to detect Tp in tissues [39], [40], spirochetes were only seen in one (9%) of the 11 biopsies studied.
We previously determined that WB was the best sample to detect spirochetes in blood obtained from infected rabbits [41]. In unpublished observations we confirmed that using spiked human blood, WB was better than serum, plasma or peripheral blood mononuclear cells (PBMCs) for detecting spirochetes. In this study, we therefore elected to use WB samples to quantitate the level of spirochetemia in untreated SS patients. Using a real-time qPCR assay targeting the polA gene, we first established the sensitivity of the assay to be between 15 and 150 spirochetes/ml, irrespective of whether freshly extracted Tp DNA or live whole spirochetes were used to spike the blood sample (Table S1). The polA PCR detection limit in blood samples that were immediately refrigerated upon spiking and kept at 4°C for a total of 26 hrs prior to DNA extraction, were one log higher than samples kept at room temperature for 1 hr and then processed. We then used this highly sensitive PCR method to amplify Tp DNA in 46% (11/26) of the WB samples obtained from SS patients (Table 3). The polA copy numbers in these patients ranged from 194.92 to 1954.2 copies/ml of WB, which is well above the cutoff for the assay. PCR inhibition was not observed in any of the DNA samples that tested negative for the Tp polA target. It is important to note that the ability to detect Tp DNA in SS patients' samples greatly improved when WB DNA was extracted within a few hours of procurement of the sample. Indeed, 63% (7/11) of the samples handled in this fashion had detectable Tp DNA, whereas only 5/15 (30%) WB samples were positive when WB was collected, frozen and shipped to UCHC for subsequent DNA extraction at the CDC. These findings highlight the importance of timely specimen processing and handling, and show how subtle differences in technique can greatly alter the ability to amplify Tp DNA.
Tp DNA was also detected by qualitative real-time PCR in 8/12 (66%) skin biopsies studied (Table 4). Three of the four skin biopsies that did not have detectable Tp DNA by RT-PCR were obtained from hyperkeratotic plantar plaques. Five of the 8 patients had Tp DNA present in both the blood and the skin, one patient with spirochetemia had a negative PCR in the skin; and two patients with detectable DNA in punch biopsy material had negative PCR results in WB.
None of the WB samples analyzed had sufficient quantity of Tp DNA to satisfactorily perform molecular strain typing. On the other hand, six of the eight positive skin samples were amenable for typing. By combining the 60-bp arp repeat sizes and the tpr E, G, and J RFLP patterns, we identified four different strain types; two each were 14d and 16d, and one each was 13d and 22a, respectively. The remaining strain was partially typeable by tpr RFLP analysis (pattern a). Although it is not possible to generalize our results due to the small sample size, our data suggests high strain diversity, which reflects the pattern seen in South Africa [27], where syphilis is known to be highly endemic.
The WHO estimates that up to a quarter of all yearly cases of infectious syphilis occur in Latin America and the Caribbean [3]. Because syphilis is not rigorously notified, and often not recognized by health care providers in the region, country-specific disease prevalence and incidence rates most likely underestimate the true magnitude of the problem. Existing published studies do provide evidence that venereal syphilis is not only highly endemic but also a very important sexually transmitted disease in tropical regions of the Americas [42]–[57]. For instance, syphilis has been shown to be a leading cause of genital ulcerative disease in both Peru and the Dominican Republic [56], second only to genital herpes. Likewise, in another study from Peru, the prevalence of venereal syphilis was estimated to be 10.5% amongst MSM and 2.0% in socially marginalized men and women [42]. In a Brazilian study, high syphilis prevalence rates were the norm in prisoners, commercial sex workers, and MSM [47]. High syphilis prevalence rates have also been documented for MSM (5% and 13%) and female sex workers (6.8% and 15.3%) in Honduras and Guatemala respectively (Source: PAHO web site http://new.paho.org accessed July, 2009). In one of the few available studies describing the epidemiology of venereal syphilis in Colombia, 10% of female sex workers in Bogota had serologic and clinical evidence of the disease [51]. Although the current study was not designed to be a comprehensive epidemiologic or microbiologic investigation of syphilis in Cali, our combined observations do provide further evidence that venereal syphilis is highly endemic in this region of Colombia. This assertion was further substantiated by the very high syphilis seroprevalence rates documented in 15–24 year old sexually active men and women from poor Cali districts [7].
Our findings also make evident that early syphilis patients in Cali may go undiagnosed and untreated for several weeks, in some cases for several months. This is not at all surprising given that the clinical manifestations of SS are often subtle and can be easily overlooked and/or dismissed by primary care providers and patients alike. A principal objective of the current case series is thus, to review for health care providers, particularly those who practice medicine in tropical regions like Colombia, the typical clinical manifestations associated with SS. Although a large proportion of patients presented with the classic palmo-plantar exanthem of SS [19], [32], [34], some had faint hypo or hyper-pigmented macules and papules that could have easily been dismissed for other dermatologic conditions. Indeed, the rash of SS may be frequently confused for other skin disorders including pityriasis rosea, psoriasis, seborrheic dermatitis and dermato-mycosis, amongst others [13]. Given the relative high prevalence of venereal syphilis in this population, it is not unreasonable to suggest that a diagnosis of SS should be considered in all sexually active individuals who present with any form of unexplained skin and mucous membrane pathology, and perhaps even in individuals with otherwise unexplained constitutional flu-like symptoms.
Several other investigators have previously used molecular methods to detect Tp DNA in WB and tissues from untreated early syphilis patients [58]–[65]. Our study is the first to measure spirochetal loads by real time qPCR in the blood of untreated SS patients. The ability to detect spirochetal DNA in these samples has been quite variable and highly dependent on various factors including; the type of sample collected, the stage of the disease (primary vs. secondary vs. latent), and the gene target used. Nevertheless, it is readily apparent that a significant proportion of early syphilis patients, regardless of the stage of the disease, have circulating spirochetal DNA. In the current study, the detection limit in blood samples that were spiked with either DNA or whole Tp organisms and kept at 4°C for 26 hrs was 15 polA copies/ml blood compared to 150 polA copies/ml in the same dilutions stored at room temperature for 1 hr. While frozen blood samples, stored over several days, may not be conducive to diagnostic PCR testing, our spiking experiments do indicate that blood samples can be stored up to 26 hours at 4°C without significantly affecting the capacity to detect Tp DNA. This may prove to be particularly useful in settings where blood samples cannot be processed on the day of collection. No PCR inhibition was observed with the use of 2ml blood for spiking experiments, despite the inherent high concentration of human DNA in these samples. It is possible that several other SS patients enrolled, if not all, were spirochetemic but not detected by qPCR. This may have been due to some patients having spirochetemia which was below the threshold of the assay or alternatively, treponemal DNA might have been degraded as a result of not being extracted within 24 hours of blood collection. For future studies the ability to detect low copy numbers may be enhanced by extracting DNA on site and using a larger volume of blood (2ml vs. 400µl).
In this study we also performed Tp strain type analysis in DNA material obtained from the skin of several SS patients. Using the method described by Pillay [27], [66], [67] we provide evidence that in the city of Cali there is considerable heterogeneity in circulating Tp strains. Subtype 14d, which is known to have a worldwide distribution [27], [62], [68], was present in the skin of two SS patients. Subtypes 13d and 16d, which were previously identified in syphilis patients in several cities in South Africa [27], were also present in skin samples in Cali. Strain type diversity in Cali, provides additional evidence that venereal syphilis is highly endemic in this population. In a much larger molecular epidemiology study conducted in several South African cities [27], greater geographic Tp strain diversity was found to correlate statistically with higher syphilis prevalence rates. It is our contention that an improved understanding of the heterogeneity of syphilis subtypes, not only helps to identify the introduction of new strains into an endemic population but also helps to evaluate if public health strategies have been successful at eradicated indigenous strains.
As already alluded to above, the molecular methods used herein provide evidence that significant numbers of spirochetes are not only present in the skin, but are also capable of spreading in significant numbers through the blood stream of untreated SS patients. Paradoxically, SS patients exhibit robust cellular and humoral adaptive immune responses [10], [17], [69], [70]. A careful analysis of Tp's unique ultrastructural features provides several explanations for this paradox. Unlike the outer membrane of gram-negative bacteria, that of Tp lacks the potent proinflammatory glycolipid lipopolysaccharide (LPS) [71], [72]. In addition, freeze-fracture microscopy studies have shown that the spirochete is largely devoid of integral outer-membrane proteins [73], [74]. Although Tp does contain an abundance of highly antigenic hydrophilic polypeptides, these molecules are tethered by covalently bound N-terminal lipids to the periplasmic leaflet of the cytoplasmic membrane [74], [75]. This unusual topology is thought to contribute to the ability of intact spirochetes to avoid recognition by innate immune cell (i.e. macrophages) pattern recognition receptors (PRR); thus delaying or impeding their activation at the sites of initial inoculation or in skin or organs to where spirochetes have disseminated. Inefficient antibody binding to the small number of potential antigenic targets present on the spirochete's outer membrane could also allow the spirochete to shun rapid and efficient binding by opsonizing anti-treponemal antibodies and thus avoid phagocytosis. One can envision a model were the paucity of outer membrane antigenic targets, perhaps in combination with the very slow rate of bacterial replication, facilitates intermittent low level spread of the spirochete from affected skin and mucous lesions into the blood stream of untreated SS patients. Lastly, sequence variation of the Tp repeat (Tpr) family of polymorphic multi-copy repeat proteins has been postulated as an additional mechanism of immune evasion and persistent infection by the spirochete [17], [76]–[78]. Of the several proteins with predicted outer membrane location, TprK has received the most attention. Although controversy remains as to the actual location of TprK [79], sequence diversity of tprK in samples obtained from several syphilis patients [76] has bolstered the idea that this molecule could play an important role in immune evasion. Several other candidate outer membrane proteins, including Tp92 [80], are currently being studied to determine if they meet the structural features of other known bacterial outer membrane proteins, if they bind syphilitic opsonic antibodies, and if they too can undergo antigenic variation.
We conclude that high syphilis prevalence rates in the region should prompt health care workers in countries like Colombia to maintain a high index of suspicion for the common and uncommon manifestations of early syphilis. In concert with the clinical findings highlighted herein, a diagnosis of SS must be considered as part of the differential diagnosis in any subject who presents with chronic skin and/or mucosal lesions. Public health care authorities must redouble their efforts to enhance early detection of venereal syphilis, to institute timely treatment of the disease and to improve follow-up of patients diagnosed with the disease. Lastly, research efforts designed to better understand the immunopathogenesis of the disease, in particular how the bacterium is able to elude host immunologic defenses and spread from sites of bacterial replication, as demonstrated herein, will greatly contribute to more effective and novel prevention strategies, including the development of an effective vaccine.
|
10.1371/journal.pgen.1000845 | Uncoupling of Satellite DNA and Centromeric Function in the Genus Equus | In a previous study, we showed that centromere repositioning, that is the shift along the chromosome of the centromeric function without DNA sequence rearrangement, has occurred frequently during the evolution of the genus Equus. In this work, the analysis of the chromosomal distribution of satellite tandem repeats in Equus caballus, E. asinus, E. grevyi, and E. burchelli highlighted two atypical features: 1) several centromeres, including the previously described evolutionary new centromeres (ENCs), seem to be devoid of satellite DNA, and 2) satellite repeats are often present at non-centromeric termini, probably corresponding to relics of ancestral now inactive centromeres. Immuno-FISH experiments using satellite DNA and antibodies against the kinetochore protein CENP-A demonstrated that satellite-less primary constrictions are actually endowed with centromeric function. The phylogenetic reconstruction of centromere repositioning events demonstrates that the acquisition of satellite DNA occurs after the formation of the centromere during evolution and that centromeres can function over millions of years and many generations without detectable satellite DNA. The rapidly evolving Equus species gave us the opportunity to identify different intermediate steps along the full maturation of ENCs.
| Centromeres are the functional elements controlling chromosome segregation during cell division. Vertebrate centromeres, which typically contain large amounts of tandem repeats (satellite DNA), are highly conserved for function but not for DNA sequence, suggesting that centromeric function is mainly determined by epigenetic factors. Evolutionary centromere repositioning is the shift of a centromere to a new position in the absence of structural chromosome rearrangements. In previous work, we demonstrated that centromere repositioning was exceptionally frequent during the evolution of the genus Equus (horses, asses, and zebras). In the present paper, we show that several Equus centromeres, including all the previously described evolutionary new centromeres, are apparently satellite-free, supporting the idea that large blocks of repeats are not necessarily required for the stability of centromeres. Our results suggest that centromere repositioning might be a two-step event: first, a neocentromere arises in a satellite-less region; satellite repeats may then colonize this repositioned centromere at a later stage, giving rise to a “mature” centromere. The rapidly evolving Equus species gave us the opportunity to catch snapshots of several evolutionary novel centromeres in different stages during their maturation.
| Centromeres, cytologically appearing as visible primary constrictions in metaphase chromosomes, are essential for the proper segregation of sister chromatids during cell division. They are the sites of kinetochore assembly and spindle fiber attachment and consist of protein-DNA complexes, in which the DNA component is typically characterized by the presence of extended arrays of tandem repeats (called satellite DNA). Satellite DNA, initially purified by density gradient centrifugation experiments [1],[2], is organized as long arrays of head-to-tail repeats, located in the constitutive heterochromatin.
Two observations have suggested that, although satellite DNA sequences and centromeres are often associated with one another, satellite DNA itself is not required for centromere function. Firstly it became clear that, in spite of the proposed involvement of these sequences in a highly conserved cell division-related function, they are remarkably different among different species. This observation, known as the “centromere paradox”, pointed to epigenetic factors as being responsible for centromere function through binding of the DNA with kinetochore proteins [3]. Secondly, and perhaps more influentially, the group of Choo [4] and subsequently several other groups [5] were able to identify and analyse neocentromeres in rare human clinical material. The analysis of neocentromeres demonstrated that full centromere function can occur in the absence of the sequence organization characteristic of most natural centromeres and that a DNA fragment may acquire centromere function without any sequence alteration, a phenomenon defined “centromerization” [6]. The existence of neocentromeres and the rapid evolution of centromeric DNA suggested that an epigenetic mark rather than DNA sequence determines centromere function. The identity of this mark remains a matter of investigation. Some have argued that the mark is the ability to be bound by CENP-A, a centromere specific variant of the histone H3 [3], while others have argued that the mark is a feedback loop in which centromere stretching at metaphase plays a critical role [7].
Another phenomenon supporting the epigenetic nature of centromeres is evolutionary repositioning, that is the shift along the chromosome of the primary constriction together with the centromeric function. Comparative studies of chromosomes in primates, other placental mammals, marsupials and birds have demonstrated that the positioning of centromeres can change over the course of evolution, in the absence of any other significant and detectable change in marker order along the chromosome, generating evolutionary new centromeres (ENCs) [8]–[13]. It has been proposed that the initial event of evolutionary repositioning may be the loss of function of the original centromere followed by the gain of epigenetic signals in a non-centromeric position. Such a sequence of events would lead to the formation of a centromere in a new chromosome region devoid of satellite DNA [10],[11],[14]. This “young” neocentromere may then gradually accumulate, during several successive generations, repetitive DNA through various recombination-based mechanisms. These events would lead to the formation of a centromere in a new permissive chromosome region devoid of satellite DNA, without involvement of DNA sequence alterations. Since all natural centromeres described so far, including ENCs, contain satellite DNA sequences, Marshall and co-workers [5] proposed that satellite sequences are incorporated at repositioned centromere sites, because they probably confer an adaptive advantage possibly by increasing the accuracy of chromosome segregation. Alternatively, the accumulation of satellite sequences may simply be a neutral process driven by the presence of heterochromatin in the centromeric DNA.
In this scenario, we might expect to find evolutionarily immature centromeres, lacking satellite DNA, in rapidly evolving species. Equids are a representative example of quickly radiating organisms; the eight living species of the Equidae family belong to the genus Equus and comprise: two horses (E. caballus and E. przewalskii), two Asiatic asses (E. kiang and E. hemionus), one African ass (E. asinus) and three zebras (E. grevyi, E. burchelli and E. zebra). The Equus species shared a common ancestor about 2–3 million years ago and the extant species emerged about 1 million years ago, that is in a very short evolutionary time [15]. These animals are valuable for comparative cytogenetics because, in spite of their recent divergence, morphological similarity and capacity to interbreed, their karyotypes differ extensively [16]–[18]. The variation involves both the structure and the number of chromosomes, which ranges from 32 in E. zebra to 66 in E. przewalskii. Cross-species chromosome painting has confirmed the great karyotypic variability of this genus [19]. In addition, we have shown that at least nine centromere repositioning events took place during the evolution of this genus, six of which occurred in E. asinus (donkey) [12],[20] and one of which occurred in horse chromosome 11 (ECA 11). These results demonstrate that the phenomenon of centromere repositioning played a key role in the rapid karyotypic evolution of the equids and point to these species as an ideal model system for the analysis of neocentromere formation and centromere evolution. The observation that a number of evolutionary novel centromeres are present in the rapidly evolving Equus species, prompted us to investigate their sequence organization in order to ascertain whether any of them lack satellite DNA, in agreement with the above described model of centromere shift during evolution.
The first part of this analysis was the determination of the DNA sequence of the evolutionary new centromere on horse chromosome 11, which demonstrated that this centromere lacks any satellite DNA sequences [21]. This observation strongly supports the hypothesis that this centromere was formed recently during the evolution of the horse lineage and, in spite of being functional and stable in all horses, did not acquire all the marks typical of mammalian centromeres, probably representing the first example of an evolutionary “immature” centromere. Here, with the goal of identifying other possible cases of satellite-less ENCs, we performed an extensive cytogenetic analysis of the organization of centromeric sequences in four Equus species: the domestic horse (E. caballus), the domestic donkey (E. asinus), and two zebras (E. grevyi and E. burchelli). The results suggest that several such “immature” ENCs may indeed be present in these species. The presence of so many apparently-satellite-free evolutionary new centromeres suggests that, at least in this genus, there is no adaptive requirement for the acquisition of centromeric satellite DNA once neocentromeres are formed.
Two satellite DNA sequences were previously isolated from a horse genomic library in lambda phage [22] using two procedures. A satellite (37cen) was identified in a phage clone containing a large restriction fragment following double digestions with frequently cutting restriction enzymes. The second satellite (2PI) was isolated as a by-product of a screen of the same library for minisatellites. The phage clones were sub-cloned in plasmid vector and sequenced. The 37cen sequence, consisting of a 221 bp repeat (Accession number: AY029358), is 93% identical to the horse major satellite family independently identified by Wijers and colleagues [23] and by Sakagami and co-workers [24]. The 2PI sequence, consisting of a 23 bp repeat (Accession numbers: AY029359S1 and AY029359S2), belongs to the e4/1 family described by Broad and colleagues [25],[26] and shares 83% identity with it. Zoo-blot analysis showed that the two horse satellites are undetectable in cow, goat, sheep, man, dog, mouse, Syrian hamster, mediterranean fruit fly and yeast, while they are present in several species of the genus Equus, including E. caballus, E. asinus, E. grevyi and E. burchelli (data not shown).
To localize these satellites, two color FISH experiments were performed using the 37cen and 2PI sequences as probes on metaphase chromosomes from E. caballus (ECA, horse) (Figure 1A, column 1), E. asinus (EAS, domestic donkey) (Figure 1B, column 1), E. grevyi (EGR, Grevy's zebra) (Figure 1C, column 1) and E. burchelli (EBU, Burchelli's zebra) (Figure 1D, column 1). The chromosomal distribution of the two satellites was analyzed from single and double color FISH experiments and the results are schematically reported in the top rows of each panel of Figure 2A–2D.
In E. caballus (Figure 2A, top row), the majority of centromeres contained both satellites (yellow), five chromosomes (1, 4, 5, 12 and X) showed only 37cen signals (green) and chromosome 2 showed only the 2PI signal (red). The centromere of chromosome 11 was the only one lacking any signal. Thus, 37cen was localized at the centromeric region of all chromosomes except 2 and 11; these results are essentially in agreement with those from Sakagami and colleagues [24] who localized a satellite DNA sequence, belonging to the same family, on all horse centromeres except three; this discrepancy is not surprising, considering that we used fluorescence-based approaches whereas they used radioactive probes, which are known to be less sensitive. The 2PI sequence was present at the centromere of all the acrocentric horse chromosomes, as well as at the centromere of eight meta- or submeta-centric chromosomes (2, 3, 6, 7, 8, 9, 10 and 13). Thus, all centromeres have either one or both satellites while ECA11 is the only E. caballus chromosome lacking signals from both satellites.
In E .asinus (Figure 2B, top row), the distribution of the two satellites was different when compared to E. caballus; in fact, several chromosomes, while lacking satellite signals at their centromeres, contained such signals at one non-centromeric terminus. In particular, the 37cen sequence was localized on one telomeric end of six meta- or submeta-centric chromosome pairs (1p, 7p, 9p, 12p, 13p, and 14q) and in the centromeric region of three chromosomes only (1, 2 and 30), chromosome 1 showing a very large subcentromeric signal; thus, in chromosome 1, this probe recognized both the p arm terminus and the extended subcentromeric heterochromatic region. The 2PI satellite was located at one terminus of thirteen meta- or submeta-centric donkey chromosomes (1p, 4p, 6p, 7p, 8p, 9p, 11p, 12p, 13p, 17p, 14q, 15q and 30q) and on the centromeric region of eleven chromosomes (1, 2, 3, 20, 21, 23, 24, 25, 28, 29 and 30), the extended chromosome 1 subcentromeric region showing two clearly distinguishable separate signals.
In E. grevyi (Figure 2C, top row), 37cen was much less represented, being detectable only on the centromeric region of the submetacentric chromosome 7. Conversely, 2PI was abundant, since it was found in one non-centromeric end of thirteen chromosomes (1p, 2p, 5p, 6p, 7p, 8p, 10p, 12p, 13p, 14p, 15p, 19q and 21q) and on the centromeric region of chromosomes 7, 9, 12 and 20; thus, chromosomes 7 and 12 contain 2PI sequences both at the centromere and at the p arm terminus.
Finally in E. burchelli (Figure 2D, top row), the 37cen sequence was undetectable whereas the 2PI sequence was abundant, hybridization signals being present on the centromere of ten chromosomes (1, 3, 4, 5, 10, 11, 12, 14, 15 and 19); on chromosomes 1 and 5, an additional signal was clearly detectable in the subcentromeric region and on chromosome 4 in the proximal region of the short arm. On EBU 1p, 12p, 14p, 17q and 20q, terminal 2PI signals were also present; therefore, chromosomes 1, 12 and 14 contain this satellite both at the centromere and at one end.
A further point to be remarked is that fourteen EGR and EBU meta- or submeta-centric autosomes are syntenic, as shown by chromosome painting [27], share the same banding pattern, being presumably derived from fusion of ancestral acrocentrics. However, we observed that the majority of these chromosomes showed a different distribution of the 2PI satellite; these EGR/EBU chromosomes were: 1/1, 2/2, 3/3, 5/5, 6/6, 7/7, 8/8, 10/10, 12/13, 14/15, 15/16, 19/19 and 20/20. Only the syntenic chromosomes EGR 16 and EBU 18 have the same satellite distribution. The discrepancy in the distribution of satellite DNA sequence that we observed may be mainly ascribed to a differential retention of repetitive sequences at sites corresponding to centromeres of ancestral acrocentric chromosomes.
The data reported in Figure 1, column 1, were obtained by high stringency hybridization (see Materials and Methods). Hybridizations at low stringency were also performed and the results were super-imposable to those obtained at high stringency except for a higher background (data not shown).
The absence of detectable 37cen and 2PI FISH signals from the centromeres of E. caballus (horse) chromosome 11 and of several E. asinus (donkey), E. grevyi (Grevyi's zebra) and E. burchelli (Burchelli's zebra) chromosomes, raises the question whether satellite DNA, belonging to other families, might be present at such centromeres. To investigate this possibility, we performed FISH analysis on the chromosomes of the four species, using their total genomic DNA as probe, at both high and low stringency (Figure 1, columns 2 and 3 and bottom rows of Figure 2A–2D). Also in this case, the data obtained with high and low stringency were essentially super-imposable, except for a higher background in the latter (compare columns 2 and 3 in Figure 1). This procedure can allow the identification of regions containing very abundant tandem repeats due to the different hybridization kinetics of highly reiterated sequences versus single copy DNA. This approach is especially effective for the identification of satellite DNA in the Equus species, providing a resolution comparable to that of FISH performed with cloned satellite probes, as clearly shown by the high specificity of the pattern of hybridization signals and by the overall similarity of signal distribution in the top and bottom rows of each panel in Figure 2. The particular adequacy of this approach to localize satellite sequences on Equus chromosomes may be due to a high degree of homogeneity in the organization of tandem repeat arrays in these genomes.
In the horse, when the chromosomes were hybridized with total horse genomic DNA (Figure 1A, column 2 and column 3), all the centromeres, except the one of chromosome 11 (white arrows), were labelled with specific signals; the distribution of these signals (Figure 2A, bottom row) corresponded to that observed with a 1∶1 mix of the single satellite probes (Figure 2A, top row), with one exception consisting in a faint interstitial signal on the long arm of the X chromosome detectable only by hybridization with genomic DNA. This observation indicated that satellite sequences other than 37cen or 2PI are present on chromosome X. Strikingly, we obtained a similar pattern of hybridization when we used donkey, Grevy's zebra or Burchelli's zebra genomic DNAs as probes on horse chromosomes; however, a certain degree of variation in signal intensities was observed on specific sites (data not shown). Very similar hybridization patterns were also observed on donkey and zebra chromosomes probed with their own genomic DNA or with genomic DNA from the other species (data not shown). These results indicated that 37cen and 2PI are the most abundant satellite sequences in these four species.
Also in the donkey (Figure 2B), Grevy's zebra (Figure 2C) and Burchelli's zebra (Figure 2D) the distribution of the FISH signals using the two approaches was not exactly comparable. In fact, following hybridization with genomic DNA, a few sites of hybridization were observed that were not detected with the 37cen and 2PI probes; these (Figure 2B–2D) involved chromosomes X in all the three species, Y in the donkey (no information on the Y chromosome of EBU and EGR is available), EAS 11cen, EGR 5qtel, EGR 19cen, EGR 20qtel, EGR 21cen, EBU 2cen, EBU 2ptel, EBU 7ptel, EBU 13cen, EBU 18qtel, EBU 20cen, EBU 21cen, EBU 21qtel. It must be mentioned here that the telomeric signal on EGR 5q represents a polymorphic marker since it was repetitively observed on one only of the two homologues. In addition, EGR 9, EBU 12 and EBU 14 showed hybridization signals with the cloned satellite probes and not with genomic DNA; this might have been due to a relatively low abundance of the repeats located at these sites. This observation indicates that we cannot rule out the presence of low abundance tandem repeats at some of the centromeres where FISH signals were not detected. Altogether these results suggested that, although 37cen and 2PI are the major satellite DNA families in the four Equus species, other repetitive DNA families exist.
It must be emphasized here that, among horse chromosomes, the only one lacking any signal (both with specific satellites and with the genomic DNA) was ECA 11 and we actually demonstrated, by sequence analysis, that this centromere is totally devoid of satellite tandem repeats [21]. Some of the centromeres lacking any signal in the three other species may actually be completely devoid of satellite repeats, like ECA 11; however, since a molecular characterization of Equus centromeres other than ECA 11 is not available, we cannot exclude that short arrays of satellite-type tandem repeats may be present and undetectable by FISH on non-horse Equus centromeres. In any case, either the absence or low abundance of tandem repeats at numerous Equus centromeres demonstrate that they are characterized by an atypical sequence organization, possibly related to their evolutionary history (see Discussion).
An important observation to be underlined is that, in the present analysis (Figure 2), no 37cen, 2PI or genomic DNA signal was observed on the nine evolutionarily new centromeres that we previously identified in the genus Equus, namely the centromeres of ECA 11, EAS 8, EAS 9, EAS 11, EAS 13, EAS 15, EAS 18/EBU 20, EAS 19 [12] and EAS 16/EBU 17 [20].
In all the horse chromosomes, with the exception of ECA 11, satellite DNA was detected at centromeres (identified as primary constrictions) as in the majority of mammalian species described so far; on the contrary, in the three other Equus species, no consistent correlation between the presence of satellite DNA and the primary constriction was observed. In order to confirm that these centromeres are actually sites of centromeric function, we performed immuno-FISH experiments on horse and donkey chromosomes using: 1) an antibody directed against the human protein CENP-A (the H3 histone variant that was previously shown to bind all horse centromeres [21]) for the immuno-identification of centromere function, and 2) horse total genomic DNA, for the localization of satellite DNA (Figure 3). In the horse (Figure 3A) both CENP-A and satellite DNA co-localized on the primary constriction of all chromosome pairs, except ECA 11, which is devoid of satellite DNA and therefore shows only the CENP-A green fluorescent signal. Conversely, in the donkey (Figure 3B), the anti-CENP-A antibody labelled the primary constriction of all the chromosomes, but several centromeres were devoid of satellite DNA, which was instead located at one end of several meta- and submeta-centric chromosomes; on these chromosomes, uncoupling of CENP-A binding and satellite DNA localization was clearly evident.
In conclusion, in the horse, satellite DNA consistently colocalizes with the centromeric protein CENP-A, with the exception of chromosome 11 in which CENP-A but not satellite DNA is present at the centromere; in the donkey, as expected, CENP-A is present at all centromeres (primary constrictions) but satellite signals are often absent at these sites while present at several non centromeric ends.
The analysis of the chromosomal distribution of satellite tandem repeats in the four Equus species showed that in this genus the organization of such sequences is atypical in two ways: 1) several centromeres seem to be devoid of satellite DNA and 2) satellite repeats are often present at non-centromeric termini (Figure 1 and Figure 2).
The 37cen and 2PI satellites, cloned from horse, represent the two major satellite families in the four Equus species. In the horse, either one or both these satellites are present on all chromosomes, except chromosome 11, and only at centromeres. In the other three species, although these satellites are abundant, they are undetectable at several centromeres and tend to be localized at terminal positions. The possibility that other families of satellite DNA may be present at these centromeres was explored by hybridizing the chromosomes with total genomic DNA; this analysis confirmed the presence of highly repetitive tandem arrays in the positions corresponding to those of the 37cen and 2PI probes and demonstrated also the existence of other still non characterized satellites on a few positions (see lower rows in the four panels of Figure 2), in agreement with the early indication obtained by Wichman et al. [28]. Nonetheless, several centromeres still failed to show any satellite hybridization signal. This absence could be due either to the lack of satellite DNA at these sites or to the presence of a number of tandem repeats too low to be detected by FISH.
The total absence of satellite repeats on a centromere has been already proven in one case: ECA 11, at the FISH resolution level, is completely devoid of any satellite DNA signal and the availability of the horse genome sequence assembly allowed us to rule out the presence of any satellite tandem repeat on this primary constriction also at the sequence level [21]. We wondered whether the centromeric function actually resides within the cytogenetically defined primary constriction of ECA 11. An array of this genomic region was hybridized with horse chromatin, cross-linked and immuno-precipitated with an antibody against the kinetochore proteins CENP-A or CENP-C, definitely demonstrating that the centromeric function resides within a DNA sequence totally devoid of satellite DNA [21]. In the same work [21], we also found that ECA 11 showed no accumulation of L1 transposons or KERV-1 elements, which were previously hypothesized to influence ENC formation [29],[30]. Although sequence data are not yet available on other Equus centromeres in which satellite DNA is not detectable by FISH, an immuno-FISH analysis with an anti-CENP-A antibody and with satellite DNA showed that, while in the horse satellite DNA and the kinetochore protein co-localize on all chromosomes (with the exception of ECA 11), in the donkey the centromeric function is often uncoupled from satellite DNA (Figure 3). In light of all these observations, it is conceivable that, besides the ECA 11 centromere, other FISH-negative centromeres of donkey and zebras may also be totally devoid of satellite repeats. Although we cannot rule out the presence of short arrays of tandem repeats on FISH negative non-horse Equus centromeres, these are nonetheless atypical and are likely to represent evolutionarily “immature” centromeres, that have recently undergone satellite DNA incorporation.
The absence of satellite repeats at some centromeres and their presence at terminal positions are in agreement with our previous observation that several centromere repositioning events occurred during the evolution of the Equidae [12],[20]; in this scenario, these evolutionarily recent events would have generated new centromeres that, at present, are still “immature” and did not yet acquire the sequence complexity typical of the vertebrate centromeres described until now. Conversely, the presence of satellite DNA at terminal positions in meta- and submeta-centric chromosomes, may be interpreted as the trace, left over by centromere repositioning events, of ancient, now inactive, terminal centromeres. In fact, comparative analyses performed using painting probes suggested that the ancestral Perissodactyla karyotype was probably composed of acrocentric chromosomes [19].
In Figure 4 a schematic representation of the possible steps leading to the formation of meta- or submeta-centric evolutionarily novel centromeres from an acrocentric ancestral chromosome (Figure 4A) is depicted. According to this scheme, and as proposed also by other authors [10],[11],[14], the first step would consist in the shift of the centromeric function to a new position lacking satellite DNA, while the satellite DNA from the old centromere remains in the terminal position (Figure 4B). A subsequent step would be the loss of the terminally located leftover satellite sequences (Figure 4C). The organization of satellite-free immature centromeres may be similar to that of the neocentromeres described in human clinical cases [14]. Finally, the new centromere could reach its maturity by acquiring satellite DNA (Figure 4D) as, for example, in the numerous ENCs described in primates and other species [8]–[13]. Thus, in the case of ECA 11 we may surmise that, while the new centromere did not acquire satellite DNA, the old inactivated centromere lost its satellite repeats, giving rise to a chromosome completely devoid of satellites (as in Figure 4C). The complete or nearly complete loss of satellite sequence from the sites where ancestral centromeres were inactivated could be due to deletion, translocation or recombination events, possibly favoured by the repetitive nature of these sequences. Conversely, it is conceivable that other repositioning events were not followed by the loss of all satellite repeats at the old inactivated centromere, giving rise to chromosomes with satellite repeats at terminal positions only (as in Figure 4B).
In a previous study [12], we demonstrated that the centromeres of several Equus chromosomes derived from repositioning events. This analysis was based on marker order comparisons in E. caballus, E. asinus and E. burchelli. We then used the same markers to extend the analysis to E. grevyi (data not shown). In Figure 5 we combined the data on centromere repositioning with the new data on the localization of satellite DNA presented in Figure 1 and Figure 2 of the present work. In these figures the four most informative groups of orthologous chromosomes are represented together with a sketch of the hypothetical ancestral chromosomes and the phylogenetic reconstruction of the events possibly leading to the centromere organization of the chromosomes in the four species.
Figure 5A shows the comparison of ECA 11 with its counterparts in E. asinus (EAS 13), E. grevyi (EGR 10q) and E. burchelli (EBU 10q). As mentioned above, the analysis of marker order on horse chromosome 11 and on the corresponding orthologous chromosomes in E. asinus and E. burchelli [12], demonstrated that ECA 11 and EAS 13 carry evolutionarily new centromeres. In the present work (see Figure 2A and 2B and Figure 5A) we observed that the two new centromeres lack satellite DNA that is instead localized at the p terminus of EAS 13, at the centromere of EBU 10 and at the p terminus of EGR 10. We hypothesize that the ancestral chromosome from which ECA 11, EAS 13, EGR 10q and EBU 10q derived, was the acrocentric outlined on the left of Figure 5A, containing satellite sequences at its centromere. The centromeric location of this hypothetical ancestral chromosome now corresponds to ECA 11qtel, EAS 13ptel, EGR 10cen and EBU 10cen. In E. caballus, the centromere was shifted in its present position, where no satellite DNA is present. The centromere of EAS 13 is also evolutionarily new and lacks any satellite DNA, at the FISH resolution level; the satellite sequences of the now inactive old centromere, have been lost in ECA 11, as in Figure 4C, while they are still present on EAS 13qtel as a relic, as in Figure 4B. Musilova et al. [27], using painting probes, demonstrated that EGR 10p and EBU 10p are orthologous to ECA 10q. It can be supposed that, after the fusion that gave rise to EGR 10 and EBU 10, centromeric satellite DNA was maintained in EBU 10 and lost in EGR 10; alternatively, short arrays of tandem repeats may still be present on the EGR 10 centromere at a level not detectable by FISH. The satellite DNA found on EGR 10ptel might represent the relic of the centromere of an ancestral acrocentric chromosome. Therefore, the absence of satellite DNA is the consequence of an evolutionarily recent repositioning event at the ECA 11 and EAS 13 centromeres, while, at EGR 10 centromere, it is a consequence of the fusion event.
The chromosomes shown in Figure 5B are the orthologs of ECA 14. The ancestral chromosome that presumably gave rise to these chromosomes is represented. Marker order analysis demonstrated that the centromere of EAS 9 was repositioned during evolution [12]. As in the preceding example, the new centromere does not show any satellite FISH signal, while the presence of satellite DNA at EAS 9 short arm terminus might be the fossil evidence of the ancestral centromere position (as in the scheme of Figure 4B). EGR 5 and EBU 5 are probably derived by the fusion of two ancestral acrocentric chromosomes (Figure 5B, right); this hypothesis is confirmed by chromosome painting data [19] which demonstrate that present day ECA 13, EGR 5p and EBU 5p are the orthologs of an acrocentric chromosome in tapirs and rhinoceroses. Presumably, after the fusion, EGR 5 centromere lost satellite DNA while EBU 5 conserved it. The 2PI satellite signal found at the p arm terminus of EGR 5 may be the remnant of an ancestral centromere. The 2PI positive region found in the subcentromeric region of EBU 5 may be the outcome of recombination events involving centromeric repeats.
In Figure 5C, ECA 17 with its donkey and zebra counterparts are shown. The hypothetical ancestral form is reported on the left of Figure 5C. Marker order analysis demonstrated that EAS 11 carries an evolutionarily new centromere [12]. EAS 11 shows no satellite sequences at the centromere while a 2PI positive region is present in the same physical position of the centromere of the ancestral chromosome corresponding to nowadays ECA 14 (as proposed in the scheme of Figure 4B). As in the previous cases, the zebra chromosomes were presumably derived from the fusion of acrocentric chromosomes. The satellite sequences were lost from EGR 6 and EBU 6 centromere. Grevy's zebra chromosome 6 shows satellite DNA signal at the p arm end. Again, this satellite sequence may represent the relic of the centromere of an ancestral acrocentric.
ECA 22 together with its donkey and zebra orthologs are shown in Figure 4D. The arrangement of ECA 22 represents an ancestral organization in mammals [31]. Marker order analysis demonstrated that EAS 15 carries an inversion, encompassed by a red line on the left of the chromosome, and that its centromere is evolutionarily new. The position of the centromere in EBU 12 can be ascribed to an additional zebra-specific centromere repositioning event or to a small inversion (red line on the left of the chromosome) [12]. EAS 15 centromere is devoid of satellite DNA, while the FISH signal present at EAS 15q terminus would represent the relic of the ancestral centromere, as in the scheme of Figure 4B. Both EGR 9 and EBU 12 centromeres were FISH positive. The ancestor of ECA 22, EAS 12, EGR 9q and EBU 12q is sketched on the left in Figure 5D. As hypothesized in the previous examples, the satellite DNA found at EBU 12p end could represent the fossil remains of an ancestral centromere that was inactivated during evolution.
Literature data suggest that the ancestral Perissodactyla karyotype might be very similar to the Rhinocerotidae one, which is characterized by high chromosome numbers (2n = 82−84), most chromosomes being acrocentric [19]. Horse chromosomes 11, 14, 17, and 22 are syntenic to black rhinoceros chromosomes 12, 5, 10 and 25, respectively [19]. These rhinoceros chromosomes are acrocentric; this evidence supports the hypothesis that the satellite DNA found at the non centromeric end of EAS chromosomes carrying ENCs is actually the reminder of the ancestral centromere.
The results presented in this work rise a number of questions concerning the underlying molecular mechanisms. The molecular marks responsible for centromeric function and stability remain elusive, considering that satellite-less centromeres appear to be functional and stable in Equus species. While neocentromere formation in human clinical cases is often accompanied by chromosomal rearrangements affecting the normal centromere, it is not clear whether centromere shift during evolution is a consequence of rearrangements of the ancestral centromere leading to loss of function. On the one end, the persistence of satellite DNA at some inactivated centromere sites could simply be a fossil relic or may be maintained by selective pressure. On the other hand, the loss of satellite sequences at some inactivated centromeres, such as the one of ECA 11, could be the consequence of recombination events eliminating functionally irrelevant sequences. Several studies on centromere repositioning in other mammalian orders and in birds [8]–[13] showed that ENCs are apparently less frequent than in the genus Equus and that, although evolutionarily novel, they are endowed with satellite sequences. According to the model presented in Figure 4, the Equus ENCs are in a still “immature” stage (Figure 4B or 4C), while the previously described ENCs of other orders have acquired satellite DNA reaching “maturity” (stage D in Figure 4). In this scenario, it remains to be established why mature centromeres possess satellite sequences considering that in the genus Equus some centromeres can stably function in their absence. Does the mechanics of centromeric function provide a molecular “sink” attracting and conserving repetitive sequences or do such sequences provide some selective advantage to centromere function? All these questions remain open for future investigation that may draw advantage from the study of the rapidly evolving Equus centromeres.
The complex evolution of satellite sequence distribution in the genus Equus, observed in the present paper, is in agreement with the instability and exceptional plasticity of the karyotype of these species [16]–[19]. In fact, the centromeric function and the position of satellite DNA turned out to be often uncoupled. Satellite-less centromeres arose from two different evolutionary events: fusions between ancestral acrocentric chromosomes and centromere repositioning. The latter event is unexpectedly frequent in this genus and occurs independently of the acquisition of satellite DNA. This observation supports the hypothesis that large blocks of satellite repeats are not necessarily required for the stability of centromeres. According to this view, satellite repeats may colonize new centromeres at a later stage giving rise to “mature” centromeres according to the pathway schematized in Figure 4. Thus, the rapidly evolving Equus species gave us the opportunity to catch snapshots of several ENCs in different stages of “immaturity”.
Fibroblasts were isolated and established from skin biopsies of a male and a female horse and from a male donkey. Grevy's zebra and Burchelli's zebra fibroblasts from female individuals were purchased from Coriell Repositories. Horse, donkey and zebras fibroblasts were cultured in Dulbecco's modified Eagle's medium (CELBIO), supplemented with 20% foetal calf serum (CELBIO), 2 mM glutamine, 2% non essential amino acids, 1x penicillin/streptomycin. Cells were maintained at 37°C in a humidified atmosphere of 5% CO2.
For metaphase spread preparation, cell cultures were treated with Colcemid (30 ng/ml, Roche) for 3 h, or mitoses were mechanically collected by direct blowing the medium on the dish surface. Chromosome preparations were performed with the standard air-drying procedure.
Whole genomic DNA from horse, donkey, Grevy's and Burchelli's fibroblasts was extracted according to standard procedures [32]. Lambda phage 37cen and 2PI DNA clones, were extracted from 10 ml of bacteria cultures with the Quantum Prep Plasmid miniprep kit (BioRad), according to supplier instructions.
Whole genomic DNA, and 37cen and 2PI satellites, were labelled by nick translation with Cy3-dUTP or Cy5-dUTP (Perkin Elmer) and hybridized to metaphase spreads of primary fibroblasts from the four equid species as described in Nergadze et al. [33]. Briefly, for each slide 250 ng of each satellite, and 25 ng of labelled whole genomic DNA was used. High stringency hybridizations were carried out overnight at 37°C in 50% formamide and post-hybridization washes were performed at 42°C in 2xSSC, 50% formamide; low stringency hybridizations were carried out at 37°C in 25% formamide and post-hybridization washes were performed at 37°C in 2xSSC, 25% formamide. Chromosomes were counterstained with Hoechst 33258. Digital grey-scale images for Cy3, Cy5 and Hoechst fluorescence signals were acquired with a fluorescence microscope (Zeiss Axioplan) equipped with a cooled CCD camera (Photometrics). Pseudocoloring and merging of images were performed using the IpLab software. Chromosomes were identified by computer-generated reverse Hoechst banding according to the published karyotypes.
Combined immunofluorescence/FISH was performed using a slight modification of the procedure previously describe by Saffery et al. [34]. Fibroblasts were incubated for 2h with 30 ng/ml Colcemid (Roche). The cells were harvested, washed once with phosphate-buffered saline and re-suspended at a concentration of 4×104 cells/ml in 0.075M KCl for 15 minutes at room temperature. 200 µl of cell suspension were cyto-spun (BHG Hermle Z380) onto slides at 1200 rpm for 10 minutes. Slides were incubated in KCM (120 mM KCl, 20 mM NaCl, 10 mM Tris-HCl, 0.5 mM NaEDTA, 0.1% (v/v) Triton X-100) for 15 minute at 37°C and blots air dried. The primary antibody (CENP-A, Upstate) was added and the slides incubated at 37°C for 1 hour followed by three 5 minute washes in KB- (10 mM Tris-HCl, 150 mM NaCl, 1% bovine serum albumin). A FITC conjugated secondary antibody was then added and the slides were incubated for a further hour at 37°C. Two KB- washes were then carried out before fixation in 4% formalin for 15 minutes. Two washes in H20 were carried out and the slides were air dried before further fixation in methanol:acetic acid (3∶1) for 15 minutes. Finally the slides were dried overnight in dark sealed boxes on hygroscopic salts. FISH and immuno-FISH image analysis were performed as described above.
|
10.1371/journal.pmed.1002766 | Independent and combined effects of improved water, sanitation, and hygiene (WASH) and improved complementary feeding on early neurodevelopment among children born to HIV-negative mothers in rural Zimbabwe: Substudy of a cluster-randomized trial | Globally, nearly 250 million children (43% of all children under 5 years of age) are at risk of compromised neurodevelopment due to poverty, stunting, and lack of stimulation. We tested the independent and combined effects of improved water, sanitation, and hygiene (WASH) and improved infant and young child feeding (IYCF) on early child development (ECD) among children enrolled in the Sanitation Hygiene Infant Nutrition Efficacy (SHINE) trial in rural Zimbabwe.
SHINE was a cluster-randomized community-based 2×2 factorial trial. A total of 5,280 pregnant women were enrolled from 211 clusters (defined as the catchment area of 1–4 village health workers [VHWs] employed by the Zimbabwean Ministry of Health and Child Care). Clusters were randomly allocated to standard of care, IYCF (20 g of small-quantity lipid-based nutrient supplement per day from age 6 to 18 months plus complementary feeding counseling), WASH (ventilated improved pit latrine, handwashing stations, chlorine, liquid soap, and play yard), and WASH + IYCF. Primary outcomes were child length-for-age Z-score and hemoglobin concentration at 18 months of age. Children who completed the 18-month visit and turned 2 years (102–112 weeks) between March 1, 2016, and April 30, 2017, were eligible for the ECD substudy. We prespecified that primary inferences would be drawn from findings of children born to HIV-negative mothers; these results are presented in this paper. A total of 1,655 HIV-unexposed children (64% of those eligible) were recruited into the ECD substudy from 206 clusters and evaluated for ECD at 2 years of age using the Malawi Developmental Assessment Tool (MDAT) to assess gross motor, fine motor, language, and social skills; the MacArthur–Bates Communicative Development Inventories (CDI) to assess vocabulary and grammar; the A-not-B test to assess object permanence; and a self-control task. Outcomes were analyzed in the intention-to-treat population. For all ECD outcomes, there was not a statistical interaction between the IYCF and WASH interventions, so we estimated the effects of the interventions by comparing the 2 IYCF groups with the 2 non-IYCF groups and the 2 WASH groups with the 2 non-WASH groups. The mean (95% CI) total MDAT score was modestly higher in the IYCF groups compared to the non-IYCF groups in unadjusted analysis: 1.35 (0.24, 2.46; p = 0.017); this difference did not persist in adjusted analysis: 0.79 (−0.22, 1.68; p = 0.057). There was no evidence of impact of the IYCF intervention on the CDI, A-not-B, or self-control tests. Among children in the WASH groups compared to those in the non-WASH groups, mean scores were not different for the MDAT, A-not-B, or self-control tests; mean CDI score was not different in unadjusted analysis (0.99 [95% CI −1.18, 3.17]) but was higher in children in the WASH groups in adjusted analysis (1.81 [0.01, 3.61]). The main limitation of the study was the specific time window for substudy recruitment, meaning not all children from the main trial were enrolled.
We found little evidence that the IYCF and WASH interventions implemented in SHINE caused clinically important improvements in child development at 2 years of age. Interventions that directly target neurodevelopment (e.g., early stimulation) or that more comprehensively address the multifactorial nature of neurodevelopment may be required to support healthy development of vulnerable children.
ClinicalTrials.gov NCT01824940
| Some 43% of children globally fail to reach their full developmental potential due to stunting and poverty.
Current evidence shows that improved nutrition has a modest effect on early child development.
Improving water, sanitation, and hygiene (WASH) may plausibly benefit neurodevelopment through reduced illness and improved gut health (through improving nutrient absorption and optimizing gut–brain communication).
The Sanitation Hygiene Infant Nutrition Efficacy (SHINE) trial tested the individual and combined effects of improved complementary feeding (provision of a small quantity of lipid-based nutrient supplement from 6 to 18 months of age, with complementary feeding counseling) and improved household WASH (provision of a pit latrine, handwashing stations, soap, chlorine, and hygiene counseling) on early child development at 24 months.
In all, 1,655 children born to HIV-negative women were assessed for gross motor, fine motor, language, cognitive, and social development using tools that were designed and adapted for rural Zimbabwe.
We found little evidence that the complementary feeding or WASH interventions tested improved child neurodevelopment at 2 years of age.
Complementary feeding and WASH interventions (as described above) may not have a clinically significant impact on child neurodevelopment.
More holistic approaches and interventions that explicitly target early child development may be needed to substantially impact child neurodevelopment.
| Globally, nearly 250 million children (43% of all children under 5 years of age) are at risk of compromised neurodevelopment due to poverty, stunting, and lack of stimulation [1]. Stunting has now been inextricably linked to poor early child development (ECD) [2] and affects 150 million children globally [3,4]. Although studies have demonstrated some improvements in ECD related to improved feeding practices, these studies have not demonstrated as much effect as hoped [5]. To address this “silent emergency” of compromised developmental potential in children [6], the Global Strategy for Women’s, Children’s and Adolescents’ Health (2016–2030) [7] and the recent Nurturing Care Framework [8]—promoted by international organizations including the World Health Organization, the World Bank, and UNICEF—are calling for an urgent scale-up of explicit ECD interventions, such as age-appropriate stimulation, responsive care, and increased access to high-quality pre-primary education. In parallel, the World Health Assembly has called for a 40% reduction in stunting by 2025 [9]. It is clear that action toward reducing the network of underlying factors that indirectly cause poor developmental outcomes is necessary. Predominant among these factors are nutritionally inadequate infant diets, and low and inequitable coverage of clean water, sanitation, and hygiene.
Among nutrition interventions, improved breastfeeding practices including early initiation [10], exclusive breastfeeding to age 6 months, increased duration of breastfeeding, and continued breastfeeding to age 24 months [11] have been shown to reduce diarrhea and child mortality, and improve educational attainment and adult income [12]. This is likely due to direct effects of nutrient provision on brain development as well as indirect effects of nutrition on physical growth, motor development, and physical activity [13]. Despite this, currently only 50% of children are breastfeed in the first hour after birth, and only 37% are exclusively breastfed [11]. Improved complementary feeding can reduce stunting [14], which is one of the strongest risk factors for poor ECD [2]; furthermore, long-term follow-up of randomized trials demonstrates that improving the nutritional adequacy of children’s diets between age 6 months and 3 years improves adult cognition and economic productivity [15].
Water, sanitation, and hygiene (WASH) interventions may also plausibly improve ECD. In a randomized trial in Pakistan, children whose households had received a 9-month intensive handwashing promotion during the first 30 months of life, which reduced diarrhea during that period but not subsequently, had higher global developmental quotients at age 5–7 years in comparison to control children, despite similar anthropometric measurements in children across the groups [16]. WASH interventions may impact ECD through several interlinked pathways. First, sanitation and handwashing with soap can reduce childhood diarrheal disease, which in some studies has been linked to poor childhood cognition and school performance [17,18], although this effect does not remain once stunting is taken into account [19,20]. Second, WASH may plausibly improve cognition by preventing environmental enteric dysfunction (EED), which may be an underlying cause of stunting [21]. EED is a disorder of the small intestine that is virtually ubiquitous among people living in conditions of poor sanitation and hygiene and is characterized by villous atrophy, increased permeability, malabsorption, and inflammation [22]. Third, WASH may modulate the composition and function of the gut microbiota, thereby influencing brain development through the microbiota–gut–brain axis [23]. Finally, EED is accompanied by systemic inflammation, which may directly impair neurodevelopment [24], and indirectly drive anemia through reduced erythropoiesis and hepcidin-mediated iron deficiency. Iron deficiency directly compromises brain development through its role in myelination, neurotransmission, and protein expression [25], and anemia causes listlessness. Similarly, being sick with diarrhea is likely to also impact children’s ability or willingness to engage in learning or active play. This resulting listlessness and lack of interest can then lead to reduced caregiver–child interaction and the capacity for children to engage in stimulating interactions and positive exploratory play [26].
It is therefore plausible that combining improved WASH and improved infant and young child feeding (IYCF) could impact ECD. The objective of this substudy within the Sanitation Hygiene Infant Nutrition Efficacy (SHINE) trial [27] was to test this hypothesis by evaluating the independent and combined effects of improved WASH and improved IYCF on ECD.
The design and methods of SHINE have been previously described [27]; the full protocol and statistical analysis plan are at https://osf.io/w93hy. Briefly, SHINE was a cluster-randomized community-based 2×2 factorial trial testing the independent and combined effects of a WASH intervention and an IYCF intervention on linear growth and hemoglobin at 18 months of age [27]. The study area comprised 2 rural districts of central Zimbabwe, which were divided into 212 clusters, defined as the catchment area of 1–4 village health workers (VHWs) employed by the Zimbabwean Ministry of Health and Child Care. Clusters were randomly allocated to 1 of 4 treatment arms (standard of care [SOC] alone, WASH, IYCF, or WASH + IYCF) at a public event using highly constrained randomization, which achieved balance across arms on 14 measures of geography, demography, water access, and sanitation coverage [28] (described more fully in S1 Text). Due to the nature of the interventions, masking was not possible. Between November 2012 and March 2015, VHWs identified pregnancies through prospective surveillance; women were eligible if they permanently resided in 1 of the rural study clusters, were confirmed pregnant (<14 gestational weeks), and provided written informed consent. Over the recruitment period, the cutoff of gestational age for recruitment eligibility was increased to 18 weeks (August 22, 2013), 24 weeks (January 3, 2014), and any time prior to parturition (October 20, 2014), through trial protocol amendments, to maximize recruitment. Recruitment took place in 211 of the 212 clusters.
All women were scheduled to receive 15 VHW visits between enrollment and 12 months postpartum (approximately 1 visit/month). Interventions were informed by extensive formative research and piloting [27,29,30]. Participatory behavior change interventions delivered during these visits were arm-specific and grounded in behavior change theory [27]. The IYCF intervention was based on formative research to identify and target cultural barriers [31]. The WASH intervention was based on the model of planned, motivated, and habitual hygiene behavior and was designed to invoke motivating emotions for hygiene and nurture [32]. At each visit, previous information was reviewed before introducing new information to create a sequenced integrated longitudinal intervention. Between 13 and 17 months postpartum, VHWs undertook monthly visits to provide routine care, deliver intervention supplies, and provide informal reminders to practice relevant behaviors, but formal modules were not delivered. At 18 months postpartum, an intervention review module was delivered to mothers in all trial arms.
VHW supervisors assessed timing and fidelity of implementation during scheduled visits and spot-checks (conducted every 3 months, or more often if VHW performance was not optimal). The content of VHW visits and the commodities provided were arm-specific.
Research nurses, separate from the intervention teams, made home visits at baseline (2 weeks after consent), at 32 weeks’ gestation, and at 1, 3, 6, 12, and 18 months postpartum to assess maternal and household characteristics and trial outcomes. At baseline, mothers had weight, mid-upper arm circumference (MUAC), and hemoglobin (Hemocue, Ängelholm, Sweden) measured, and were tested for HIV using a rapid test algorithm. HIV-positive women were urged to seek immediate antenatal care for PMTCT interventions. The following baseline indices were assessed: household minimum dietary diversity, food insecurity (Coping Strategies Index), household wealth (asset ownership index) [33], and maternal capabilities (perceived physical health, mental health, stress, social support, decision-making autonomy, gender norms attitudes, time use, and mothering self-efficacy), as detailed in the trial design paper [27].
Infant birth date, weight, and delivery details were transcribed from health records. The trial provided Tanita BD-590 infant scales to all health institutions in the study area and conducted training. Gestational age at delivery (prematurity) was calculated from the date of the last menstrual period. Infant weight, length, head circumference, and MUAC were measured at every postnatal visit. Children with acute malnutrition or illness were referred to local clinics. Mothers testing HIV-negative at baseline were retested at 32 weeks’ gestation; those testing HIV-negative during pregnancy were retested at 18 months postpartum.
Intervention uptake was assessed at all visits and is reported here, as pre-specified, for the 12-month postnatal visit. Nurses assessed WASH-related behaviors through maternal report (open defecation among household members, treatment of drinking water, disposal of nappy water, and child geophagia) and observation of the latrine (type of latrine, whether the path to latrine was trodden, whether the latrine was used for storage; and whether the latrine was shared with other households), handwashing station (presence of Tippy Taps and whether they were filled with soap and water), and play yard (visible cleanliness). Nurses assessed IYCF behaviors through maternal report of whether the child was still breastfeeding; the mother’s understanding of how to feed a child during illness; 24-hour recall of infant minimum dietary diversity and consumption of iron-rich, animal-source, and vitamin-A-rich foods; and 24-hour recall of infant SQ-LNS consumption.
All analyses were intention-to-treat at the child level. For tests with continuous outcomes (MDAT, MacArthur–Bates CDI, and A-not-B test), the absolute difference in mean score between treatment groups was estimated. For tests with dichotomous outcomes (self-control and grammar), the relative risk (RR) of passing the test was estimated in comparing treatment groups. Although the study was not powered to detect a statistical interaction between the IYCF and WASH treatments, it was estimated for each outcome. We accounted for the interaction in the model if it was significant (p < 0.05, according to the Wald test) or had a sizeable point estimate (i.e., difference in mean score > 0.25 SD for continuous outcomes; RR > 2 or <0.5 for dichotomous outcomes). Otherwise, we used a regression model with 2 terms to represent the treatment arms; we estimated the effect of IYCF by comparing the 2 IYCF arms with the 2 non-IYCF arms and estimated the effect of WASH by comparing the 2 WASH arms with the 2 non-WASH arms. If interaction was significant, we used a regression model with 3 terms to represent the 4 treatment arms. We used generalized estimating equations that accounted for within-cluster correlation to estimate effect size, unadjusted for other covariates, with an exchangeable working correlation structure [39]. A log-binomial specification was used to facilitate estimation of RRs. We compared baseline characteristics between arms while handling within-cluster correlation using multinomial and ordinal regression models with robust variance estimation, and Somers’ D for medians. We used Stata (version 14.1) for all analyses.
Adjusted analyses controlled for prespecified baseline covariates (as in our statistical analysis plan), which were initially assessed in bivariate analyses to identify those with an important association with the outcome (for dichotomous outcomes: p < 0.2 or RR > 2.0 or < 0.5; for continuous outcomes: p < 0.2 or difference > 0.25 SD). Selected covariates were entered in a multivariable regression model; a forward stepwise selection procedure was implemented, with p < 0.2 required for a variable to enter the model.
A per-protocol analysis was conducted to examine intervention effects when delivered at high fidelity (prespecified for WASH + IYCF arm as receiving all 10 core modules; for other arms predefined as receiving all modules scheduled at the same time-points when WASH + IYCF core modules were delivered). A prespecified subgroup analysis by child sex was planned if the interaction terms were p < 0.05. In a prespecified sensitivity analysis, children of women who seroconverted to HIV after pregnancy were excluded.
Other ECD intervention trials and comparisons of preterm versus term children have reported effect sizes of 0.3–0.4 standard deviations for similar ECD outcomes [42,43]. Accordingly, we calculated our sample size requirements to detect a 0.2 standard deviation shift with >80% power and a type I error rate of 5%, assuming an ICC of 0.07, 10 children per cluster, 33 clusters per arm, and a total of 132 clusters. We therefore aimed to recruit at least 1,320 children.
The Medical Research Council of Zimbabwe and the Institutional Review Board of the Johns Hopkins Bloomberg School of Public Health approved the study protocol (Zimbabwe: MRCZ/A/1675; Johns Hopkins University: IRB#4205). The ECD substudy protocol was included as an amendment to the main SHINE trial protocol, submitted to and approved by the 2 institutional review boards. The SHINE statistical analysis plan included the prespecified ECD outcomes. These documents can be found in S1 Text and at https://osf.io/w93hy. An independent data and safety monitoring board comprising 2 physicians from Zimbabwe and a statistician from the UK (listed in Acknowledgments) reviewed interim adverse event data in the main trial between enrollment and 18 months of age, but not in the ECD substudy since no interventions were provided between 18 and 24 months of age. The trial was registered at ClinicalTrials.gov (NCT01824940).
Between November 22, 2012, and March 27, 2015, 5,280 pregnant women were enrolled from 211 clusters at median 12 (IQR 9, 16) gestational weeks (Fig 1). Of 3,989 HIV-unexposed live births, 198 (4.8%) children died, 5 (0.1%) voluntarily left the study, and 100 (2.5%) were lost to follow-up or moved outside Zimbabwe; 3,686 children were therefore assessed at the 18-month visit. As previously reported, mean length-for-age Z-score (LAZ) was 0.16 (95% CI 0.08, 0.23) higher and hemoglobin concentration 2.03 (95% CI 1.28, 2.79) g/l higher among children in the IYCF compared to non-IYCF arms, but there was no evidence that the WASH intervention affected either LAZ or hemoglobin [44]. There was a modest impact on weight by the IYCF intervention, which we have previously reported (increase in weight-for-age Z-score: 0.13 [95% CI 0.07, 0.20], p < 0.001) [44].
Of the 3,686 children who provided trial primary outcomes at 18 months, 2,601 (70.6%) became 102 weeks to 112 weeks of age during the enrollment period. Of these 2,601 eligible children, 1,655 (63.6%; from 206 clusters) were enrolled in the ECD substudy and assessed at 24 months. The remaining 946 children were not enrolled: 12 (1.3%) declined; 2 (0.2%) died between 18 and 24 months of age; 464 (49.0%) had relocated temporarily or permanently from their study home; 194 (20.5%) could not be scheduled at a mutually agreeable time within the required age window; and 274 (29.0%) were not reachable by telephone or home visit to determine availability and interest in joining the ECD substudy. The mean (SD) age of children at the time of ECD assessment was very similar across trial arms (SOC: 105.3 [2.0] weeks; IYCF: 104.9 [1.9] weeks; WASH: 105.3 [2.0] weeks; WASH + IYCF: 105.2 [2.0] weeks). Fifteen children (0.9%) (5, 3, 4, and 3 from the SOC, IYCF, WASH, and WASH + IYCF arms, respectively) were assessed for ECD but excluded from analysis because they scored “moderate to severe” on the Washington screening tool, and 25 (1.5%) (8, 2, 9, and 6 from the SOC, IYCF, WASH, and WASH + IYCF arms, respectively) were excluded because they were found to be outside the allowable age window of 102–112 weeks, leaving 1,615 HIV-unexposed children in the present analysis. Baseline characteristics of mothers/children who joined and did not join the ECD substudy are shown in Table 1, split into 3 groups: those who were not eligible, those who were eligible but not enrolled, and those who were enrolled.
Baseline characteristics of enrolled mothers and infants were broadly similar between randomized groups, although there were minor imbalances in wealth, electricity supply, improved water source, water treatment, availability of a handwashing station, observed feces in the yard, and dietary diversity score (Table 2). Almost half of households practiced open defecation, and only one-third had an improved latrine at baseline. Few households had a handwashing station or treated their drinking water. The median walk time to an improved water source was 10 minutes; per capita volume of water collected per day was around 10 liters. Mothers were generally married and well-educated, but very few were employed. Average infant birth weight was 3.10 kg; the majority of infants were born in institutions by normal vaginal delivery.
Fidelity of intervention implementation was high (Table 3). Among households in the WASH arms, ≥98% received ventilated improved pit latrines, handwashing stations, baby mats, and play yards, and around 90% received ≥80% of planned soap and chlorine solution deliveries. Among IYCF households, almost 90% received ≥80% of planned SQ-LNS deliveries. Across all arms, VHWs completed 90%–95% of planned intervention visits.
Intervention implementation, assessed by observed and reported behaviors at the 12-month postnatal visit, achieved substantial contrast between arms (Table 3). In the WASH arms, open defecation among household members was virtually eliminated (0.6% compared to 40.4% in the non-WASH arms). Almost all households in the WASH arms (>99%) had an improved latrine; in 87% of households, the latrine had a well-trodden path and was not being used for storage (compared to 25% in non-WASH arms). In all, 85.6% of WASH households had a handwashing station with observed soap or rubbing agent and water (compared to 2.6% of non-WASH households). Among WASH households compared to non-WASH households, 26.4% versus 73.5% of mothers reported ever seeing their child ingest soil, and 2.9% versus 21.2% reported ever seeing their child ingest chicken feces. Compared to children in the non-IYCF arms, a higher proportion of children in the IYCF arms met minimum dietary diversity, and children in the IYCF arms had consumed more animal-source, more iron-rich, and more vitamin-A-rich foods in the previous day; >90% of children in the IYCF arms consumed SQ-LNS in the previous day. More than 95% of infants in all groups were still being breastfed at 12 months.
The effects of the randomized interventions on primary ECD outcomes at 24 months are shown in Table 4. For all ECD outcomes presented in this paper, there was no interaction between the IYCF and WASH treatments; accordingly, we estimated the effects of the interventions by comparing the 2 IYCF groups with the 2 non-IYCF groups and the 2 WASH groups with the 2 non-WASH groups. The IYCF intervention had a small but significant effect on the total MDAT score (unadjusted difference 1.35, 95% CI 0.24, 2.46; p = 0.02), which was non-significant on adjustment (adjusted difference 0.79, 95% CI −0.22, 1.60; p = 0.06). This effect size corresponds to a 0.15-SD increase in total MDAT score among children randomized to the IYCF intervention. The total MDAT score difference was driven by slightly higher scores in the language component (unadjusted difference 0.66, 95% CI 0.12, 1.19; p = 0.02) and social component (unadjusted difference 0.26, 95% CI 0.01, 0.51; p = 0.04) for children in the IYCF groups; both differences were attenuated in adjusted analyses (Table 4). We found no evidence that the WASH intervention affected the total MDAT score or any of its components.
There was no impact of IYCF on the MacArthur–Bates CDI grammar checklist in unadjusted or adjusted analyses. The WASH intervention had no impact in unadjusted analyses, but on adjustment it had a small but significant impact on the total number of words reported to be used by the child (adjusted difference 1.81, 95% CI 0.01, 3.61; p = 0.049). This effect size corresponds to a 0.09 SD increase in MacArthur–Bates CDI score among children randomized to the WASH intervention.
Neither intervention had any evidence of impact on the A-not-B test or self-control task in unadjusted or adjusted analyses (Table 4).
The IYCF intervention had a small but significant impact on the proportion of children reported to use plurals (adjusted RR 1.23, 95% CI 1.04, 1.45; p = 0.013) but no evidence of impact on the proportion of children reported to combine 2 words or the proportion of children using imperatives or the progressive tense. The WASH intervention had a significant impact on the use of plurals (adjusted RR 1.30, 95% CI 1.09, 1.55; p = 0.003) but had no evidence of impact on either the proportion of children reported to combine 2 words or the proportion of children reported to use imperatives or the progressive tense.
In the per-protocol analysis, effects of IYCF and WASH among the 1,310 children of mothers who had high-fidelity intervention delivery showed slightly reduced point estimates compared to the intention-to-treat findings, and differences between arms were no longer significant (S1 Table). In a preplanned subgroup analysis, there was no interaction between treatment group and child sex.
We investigated the independent and combined effects of improved WASH and improved IYCF on ECD in a setting of high stunting and poverty in rural Zimbabwe. Overall, we found little evidence that either package of interventions improved child development scores at 2 years of age. There was a small but significant impact of the IYCF intervention on unadjusted MDAT total, language, and social developmental scores; however, the differences between IYCF and non-IYCF groups were extremely modest (<1 item on the MDAT) and not significant in adjusted analyses. There was a small but significant impact of the WASH intervention in adjusted analyses for the CDI language test (which was not present in unadjusted analyses), but this was not reflected in the MDAT language score. There was no impact of WASH on any other ECD test.
Previous studies have reported larger effects on psychomotor development (i.e., changes of 2 to 8 points on the Bayley or Griffiths scales) following an intervention to improve complementary feeding [42,48,49]. In our study the IYCF intervention increased the overall MDAT score by only 1–2 points (0.15 standard deviations) in unadjusted analyses, equivalent to a child completing 1 or 2 extra tasks at the age of 2 years (e.g., running, saying 2 words together, or being able to thread beads or stack objects). These findings are consistent with the main trial results, in which IYCF modestly increased LAZ, head circumference-for-age Z-score, and hemoglobin concentration; however, these improvements in growth and anemia appear to translate into very small measurable differences in child development scores. Although the MDAT is a direct assessment tool with good cultural validity, it includes fewer items per age band than the Griffiths assessment (approximately 8 items per age band rather than 12), making it easier to use in a large trial in a rural Zimbabwean setting but potentially less sensitive to change. We did not separate out items in the MDAT to see if children achieved individual items earlier [50] as we felt it was important to concentrate on the prespecified overall global developmental effect; however, this may be an interesting future analysis. Although we found a higher reported number of words spoken by children in the WASH arms, this finding was apparent only in adjusted analyses and the effect size was very small (an additional 1.8 words in the WASH group, equivalent to <0.1 standard deviations of improvement in language).
We have previously highlighted that the interventions typically delivered by WASH programs in rural areas of low- and middle-income countries are insufficiently effective to reduce highly contaminated environments enough to reduce diarrhea or promote linear growth [44]. We have argued that a paradigm shift is needed in the way WASH is delivered, to develop interventions that are more effective and less reliant on behavior change. Whether a more comprehensive and effective WASH intervention can confer benefits for ECD requires evaluation in future studies. Two recent WASH Benefits trials implemented similar interventions to SHINE and evaluated ECD outcomes. In the Kenyan WASH Benefits trial [50], there were no differences in ECD measures at 2 years in either IYCF or WASH intervention groups. By contrast, the Bangladesh WASH Benefits trial [51] found an impact of every WASH intervention delivered singly or in combination, and of the nutrition intervention alone or combined with WASH, on multiple ECD outcomes. However, this trial compared each intervention to a control arm in which families received no promoter visits, making it difficult to disentangle the effects of the interventions from the impact of regular home visits. It is well established that home visiting with promotion of sensitive caregiving can impact ECD [42], and it is plausible that home visiting alone has similar benefits [52].
Our study has strengths and limitations. We undertook one of the only cluster-randomized trials of WASH interventions in early life with and without an IYCF intervention. The current analysis was a substudy of the larger trial, which was primarily designed to evaluate effects of the interventions on linear growth and hemoglobin. We designed the substudy to include a broad assessment of ECD, including executive function, memory, and self-control (often not evaluated at young ages); undertook extensive piloting and validation; and conducted regular quality control checks. Despite this, cognitive tests available to assess 2-year-old children are less sensitive than more complex cognitive assessments used at older ages. Furthermore, although the tests we used have shown sensitivity to change in other field studies [43,53], more sensitive tests or biological techniques such as EEG [54,55] may have given us information that these assessments did not; however, these other approaches are expensive and very difficult to do in field studies. Testing at older ages, such as school entry, would be helpful because tests may be more sensitive to small changes in cognitive function at this age.
We used a constrained randomization technique to mitigate any imbalances during enrollment, and conducted unadjusted and adjusted analyses for all ECD outcomes, using a large number of prespecified covariates to increase the precision of our estimates. In general, the point estimates of the effects of the interventions on ECD outcomes were attenuated after adjustment, although the effect of WASH on the MacArthur–Bates CDI score increased after adjustment. In evaluating the public health impact of the IYCF and WASH interventions, it is important to interpret both models. The adjusted analyses included several trial-related factors (child age, fieldworker, and calendar time) that may have an important influence on ECD measurement; the differences between unadjusted and adjusted estimates may therefore partly reflect the challenge of conducting child development assessments at 2 years of age. Although we adjusted for multiple factors known to influence child development (including socioeconomic status, maternal education, low birth weight, and prematurity), there are likely to be other unmeasured factors influencing child development. The current analysis did not investigate whether or not intermediate factors may have been affected by the WASH intervention, which could interplay with ECD (e.g., mother–child interaction or maternal capabilities); future analyses will study these interactions in more detail.
In summary, we found little effect of improved complementary feeding or an elementary household-level WASH intervention on measures of child development at 2 years of age. There was a very small increase in the total child development (MDAT) score among children receiving the IYCF intervention (in unadjusted analysis only) and a very small increase in language score among children receiving the WASH intervention (in adjusted analysis only); these small effects suggest that neither intervention at scale would meaningfully impact the compromised neurodevelopment that affects 43% of children under 5 years old globally. Neurodevelopment is a complex process impacted by multiple factors, not all of which could be addressed in this trial, such as low birth weight, prematurity, mother–child interaction, poverty, and child stimulation. Collectively, our data suggest that more holistic approaches and interventions that explicitly target ECD, as recommended in the Nurturing Care Framework, may be required to substantially improve child development [52,56].
|
10.1371/journal.pbio.2003404 | Saccades are phase-locked to alpha oscillations in the occipital and medial temporal lobe during successful memory encoding | Efficient sampling of visual information requires a coordination of eye movements and ongoing brain oscillations. Using intracranial and magnetoencephalography (MEG) recordings, we show that saccades are locked to the phase of visual alpha oscillations and that this coordination is related to successful mnemonic encoding of visual scenes. Furthermore, parahippocampal and retrosplenial cortex involvement in this coordination reflects effective vision-to-memory mapping, highlighting the importance of neural oscillations for the interaction between visual and memory domains.
| In everyday life, we constantly move our eyes to sample visual information. In order to make the sampling efficient, these eye movements need to be coordinated with the intrinsic brain dynamics that constrain visual computations. The present study provides novel evidence for how this coordination is achieved at the neuronal level, from 2 independent data sets: direct brain recordings in epileptic patients and noninvasive magnetoencephalography recordings in healthy participants. Both studies showed that eye movements are locked to the phase of alpha oscillations—synchronous and coherent neuronal electrical activity at 7–14 Hz—just prior to a saccade, i.e., a rapid eye movement that abruptly changes the point of fixation. Importantly, this coordination is predictive of successful memory encoding.
| Sampling of visual information has been shown to be rhythmic rather than continuous [1–3]. In particular, brain rhythms clocked by oscillations in the alpha (7–14 Hz) range [4] constrain visual sampling: electroencephalography (EEG)/magnetoencephalography (MEG) studies in humans have shown that the trial-by-trial fluctuations in near-threshold visual perception performance depend on the phase of alpha oscillations prior to stimulus presentation [5,6]. Saccadic eye movements overtly sample visual scenes. Here we ask how brain oscillations and saccades are coordinated in order to allow visual information to be encoded in memory areas.
We addressed this question by tracking eye movements in separate memory experiments involving MEG in healthy adults and intracranial recordings in epileptic patients (Fig 1). Participants were asked to remember images of visual scenes, and we later probed their memory. The phase locking [7] between presaccadic brain oscillations in relation to saccade onset was contrasted between later-remembered and later-forgotten images. Building on prior evidence on the cortical origins of alpha activity underlying visual information sampling [8,9], we hypothesized that higher phase locking in occipital lobe would be related to successful memory performance. MEG and intracranial data both showed that eye movements are locked to the phase of alpha oscillations prior to a saccade. Importantly, this coordination was related to successful memory encoding.
In order to investigate the temporal coordination of saccades and brain oscillations, the time-frequency representations of phase and power of the MEG and intracranial data were aligned to saccade onsets. Accordingly, high presaccadic phase locking would demonstrate an effective coordination of saccades in relation to brain oscillations. The intracranial data recorded from 3 patients with occipital depth electrodes (Fig 2A) revealed a significantly higher phase locking for later-remembered as compared to later-forgotten trials in the alpha band (12–14 Hz, cluster randomization: p < 0.005, controlling for multiple comparisons over frequencies, 2-sided test, fixed-effects statistics). Fig 2B depicts a time-frequency representation of the difference in phase locking, indicating that the effect is centered around 250 ms prior to saccade onset at 12–14 Hz. When aligning the data to saccade offset (i.e., fixation onset), no significant differences in phase locking were found (presaccade: p > 0.21, S1A Fig; postsaccade: p > 0.25, S1B Fig), in line with the idea that activity timed to saccade onset is important for visual processing [10]. The intracranial results then guided the analyses in the group-level study by confining the frequency of interest to 12–14 Hz, where MEG data here is presented from 22 healthy participants performing the memory task (see “Materials and methods” for exclusion criterions). A cluster-based permutation test revealed a significant difference in presaccade phase locking between later-remembered and later-forgotten images in the alpha band (12–14 Hz; cluster randomization: p < 0.01, controlling for multiple comparisons over sensors, 2-sided test, Fig 3A).
In the posterior sensors forming a cluster, the difference was most pronounced approximately 250 ms prior to saccade onset (Fig 3A). Unfiltered data from exemplar depth electrodes and MEG sensors depicting saccade-onset locked potentials are shown in S2 Fig. Analyzing the phase locking for later-remembered and later-forgotten pictures separately suggested the existence of a preferred alpha phase for later-remembered, but not later-forgotten, trials (S3 Fig).
Additional analyses of presaccadic spectral power indicated that the phase-locking results were not biased by spectral power (S4 Fig). A control analysis, in which we related phase locking to stimulus onset (irrespective of saccadic eye movements), revealed no significant differences between later-remembered and later-forgotten scenes (S8 Fig). However, when analyzing power after stimulus onset (irrespective of saccadic eye movements), significantly less alpha power was found for remembered as compared to forgotten scenes (12–14 Hz; cluster-randomization: p < 0.019, controlling for multiple comparisons over sensors and time, 2-sided test, S8 Fig), highlighting the difference between stimulus-onset-related subsequent memory studies (for an overview, see [11]) and the present saccade-related phase-locking analyses during free viewing. No difference was found when analyzing phase locking or power after saccade onset (S9 Fig).
Since the average fixation duration between typical eye movements is less than 500 ms, we also analyzed saccades with a minimum fixation duration of 200 ms prior to saccade onset. In this analysis, significantly higher phase locking for later-remembered trials than for later-forgotten trials at 10 Hz was found (cluster randomization: p < 0.05, controlling for multiple comparisons over sensors, 2-sided test, S5 Fig). Note that the shorter time window of 200 ms is at the expense of frequency resolution now being approximately 5 Hz). Because this analysis produced higher trial numbers, data from all 36 participants could be analyzed. Again, significantly higher phase locking for later-remembered trials than for later-forgotten trials was found (cluster randomization: p < 0.005, controlling for multiple comparisons over sensors, 2-sided test, S5 Fig). We conclude that the memory encoding related to saccades phase locked to alpha oscillations is robust with respect to presaccadic epochs of different lengths.
In order to identify the sources of the effects, we computed phase locking in the alpha band for virtual sensors, applying a dynamic imaging of coherent sources (DICS) beamformer [12]. Cluster-based permutation statistics at the source level yielded a significantly higher phase-locking index for later-remembered trials than for later-forgotten trials (p < 0.01, 2-sided test; Fig 3B). The cluster spanned from visual to parietal and temporal areas, extending into the cerebellum (Fig 3B). The largest differences were found in the parahippocampal gyrus and the retrosplenial cortex, which have been shown to support the encoding of visual scenes [13–15], and extended into the posterior hippocampus. The MEG source localization is supported by intracranial data from parahippocampal depth electrodes in the 3 patients, showing significantly higher phase locking for later-remembered trials versus later-forgotten trials in the alpha range (8–10 Hz, p cluster < 0.05; 2-sided test, fixed-effects statistics; S6 Fig)
The memory performance of the 22 participants included in the main MEG analyses (d-prime = 2.13, SE = 0.11) was considerably higher than the memory performance in patients (d-prime = 0.79). In total, 38,177 saccades were detected in the eye tracking data (22 participants, mean = 1,735.3, SE = 83.6), resulting in an average saccade rate of 2.30 Hz (SE = 0.11). The saccade rate is at the lower end of the typically reported range, which can be partly explained by the use of conservative saccade detection criterions to exclude ambiguous eye tracking data. The saccade rate was significantly higher (t21 = 7.34, p = 3.17 * 10−7) for later-remembered (mean = 2.37 Hz, SE = 0.10) versus later-forgotten (mean = 2.01 Hz, SE = 0.11) scenes, which has been reported previously [16,17], but see [18] for conflicting evidence. The average saccade duration was 28.8 ms (SE = 0.6), and the average fixation duration was 342.5 ms (SE = 13.8). Saccade directions displayed a horizontal bias but were not different for later-remembered versus later-forgotten trials (Kuiper 2-sample test for each participant, all p-values > 0.1; S7 Fig).
When only events with a minimum fixation period of 500 ms prior to saccade onset were included (as in the main analyses), 3,837 saccades remained (mean = 174.4, SE = 13.26) in the eye tracking data, resulting in an average saccade rate of 0.36 Hz (SE = 0.01). There was no significant difference between saccade rates for later-remembered (0.22, SE = 0.02) and later-forgotten (0.25, SE = 0.02) scenes (t21 = −1.77, p = 0.091), indicating that the subsequent memory effect found in all saccades (above) cannot be generalized across all types of saccades. The average saccade duration was 28.2 ms (SE = 0.8), and the mean fixation duration was 802.9 ms (SE = 20.4). Saccade directions displayed a horizontal bias but were not different for later-remembered versus later-forgotten trials (Kuiper 2-sample test for each participant, all p-values > 0.1; S7 Fig).
In the intracranial data, a total of 1,415 saccades were detected (3 participants, mean = 471.7), resulting in an average saccade rate of 1.25 Hz and a mean fixation duration of 474 ms. Note that this low saccade rate can partly be explained by the fact that electrooculography (EOG) signals were used to detect saccades in patients, which is less sensitive than eye tracking, and by conservative saccade detection criterions to exclude ambiguous EOG data. The mean saccade duration in patients was 33.7 ms. The saccade rate for later-remembered scenes (mean = 1.1551) was lower than for later-forgotten scenes (mean = 1.2906) for the patient data.
When only events that were free of saccades and blinks in a 0.5-s interval prior to saccade onset were included (as in the main analyses), 434 saccades remained (mean = 144.7) in the EOG data, resulting in an average saccade rate of 0.42 Hz. The saccade rate for later-remembered scenes (mean = 0.44) was similar to that of later-forgotten scenes (mean = 0.40) for the patient data. The average saccade duration was 26.2 ms, and the mean fixation duration was 878.6 ms.
In 2 independent data sets, we provide novel evidence for a functionally relevant coordination of saccadic eye movements and brain activity. Both the intracranial and the MEG data show that retinal inputs are temporally aligned to a preferential alpha phase. Importantly, this coordination was related to successful memory encoding, suggesting a mechanistic role for alpha oscillations in coordinating the encoding of visual information. Furthermore, our results point to an active involvement of task-relevant brain areas in this coordination: MEG and intracranial data yielded the occipital cortex, the parahippocampal gyrus, and the retrosplenial cortex as sources of the coordination of saccades and alpha phase, which have been shown to support the encoding of visual scenes [13–15]. The engagement of scene-selective areas may reflect effective vision-to-memory mapping along visual, parietal, and posterior temporal cortices [19].
Our findings are in line with work from the 1960s [20] suggesting a relationship between alpha oscillations and saccades; however, this effect was not related to perception and memory. They also support the notion of a preferred alpha phase for the execution of eye movements [21], by suggesting that during optimal information encoding, the execution of saccades is on hold until the end of an alpha duty cycle. We propose that effective coordination of saccades and brain oscillations allows for optimizing the speed of processing in the visual system [22]. The intracranial data in occipital and parahippocampal electrodes showed enhanced phase locking in the alpha band for later-remembered trials as compared to later-forgotten trials, albeit with the frequencies being slightly lower in the parahippocampal (8−10 Hz) than in the occipital depth electrodes (12−14 Hz). This could be interpreted as a shift in the dominant frequency of brain areas along the hierarchy, from visual to memory areas.
The main results presented here rely on events with a minimum fixation duration of 500 ms prior to saccade onset. The upside of this selection is the exclusion of other saccades or blinks that would contaminate the time window of interest while keeping a reasonable frequency resolution. On the downside, these events may not reflect stereotypical eye movement behavior, which display an average fixation duration of approximately 250 to 300 ms. However, analyzing events with a minimum fixation duration of 200 ms (at the expense of frequency resolution) showed very similar phase-locking effects, thus underscoring the robustness of our core findings.
Although memory studies often treat eye movements as artifacts, their interaction with memory processes has gained recent interest in the field [23,24]. Importantly, investigating naturalistic behavior in free-viewing paradigms, as used in the present study, has been shown to provide crucial insight into the interaction of eye movement behavior and memory processes, as, for example, relationships between visual sampling and recognition memory performance [16,17] or hippocampal blood oxygen level-dependent (BOLD) activity [25]. Going beyond these prior findings, the present results indicate that eye movements already have an effect on memory performance at the stage of their initiation, depending on their coordination with brain rhythms implicated in the sampling of visual information.
The increase in memory encoding with saccades locked to alpha phase might be supported by anticipatory attentional deployment [26]. The fact that the phase-locking difference was found prior to saccade onset might suggest planning of the upcoming to-be-attended location [27], resulting in a stronger locking between saccades and the phase of the alpha oscillation and ultimately improved memory encoding. The present results highlight the necessity for a coordination of alpha oscillations and eye movements for optimal memory encoding. Efficiently sampled visual information could then be integrated by the hippocampal memory system. A recent nonhuman primate study demonstrated that saccades were aligned to hippocampal oscillations of approximately 10 Hz [28]. Future studies should explore interregional synchronization in relation to oculomotor behavior, visual information sampling, and memory.
All participants gave written informed consent before the start of experiment in accordance with the Declaration of Helsinki. The study was approved by the local ethics committee (commission for human related research CMO-2014/288 region Arnhem/Nijmegen NL). The patients, who volunteered to participate in the study, had depth electrodes implanted for diagnostic reasons. The patients gave written informed consent. The study was approved by the ethics committee of the University of Munich.
For the MEG part, 36 young healthy adults were included in the study. Initially, 48 participants were recruited; however, 12 were removed because of not completing the study (7 participants), excessive movement artifacts (2 participants), or technical problems during the recordings (3 participants). The 36 participants included in this study (24 females; mean age 23.1 y, range 18−30 y; 35 right handed) reported no history of neurological and/or psychiatric disorders and had normal or corrected-to-normal vision.
Additionally, 3 male patients (age range 30−60 y) with a history of drug-resistant epilepsy were recruited from the Epilepsy Center, Department of Neurology, University of Munich, Germany.
The study design comprised an MEG and an fMRI (not reported here) session. Session order was counterbalanced across participants. For each session, 3 stimulus sets of 100 photographs each were constructed. Half of the pictures depicted indoor scenes, the other half outdoor scenes (exemplary scenes are shown in Fig 1). Pictures were presented in the MEG chamber on a 39 × 46 cm back-projection screen subtending a visual angle of approximately 27° × 32°. Out of the 3 sets, 2 sets (200 scenes) were presented during encoding. During test, these 2 sets were presented again, plus the third set (100 scenes as foils). Assignment of a set to encoding or test was counterbalanced across participants. Nine additional scenes were presented during a short practice session before encoding and test in order to explain the task. Participants were made aware about the memory test before the start of the experiment.
Fig 1 illustrates the experimental procedure. At study, the pictures were presented for 4 s in random order with the constraint that no more than 4 scenes of the same type (indoor/outdoor) were shown consecutively. The participants were instructed to judge whether the depicted scene was indoors or outdoors by button press during the fixation cross. This encoding task was chosen to ensure attention to each scene and promote encoding of the images. Participants freely viewed the scenes; i.e., they were not expected to fixate. A fixation cross with variable duration (1–2 s) followed each scene.
The study phase was followed by a distracter phase during which the participants solved simple mathematical problems for approximately 1 min, underwent approximately 5 min of fixation to different locations on the screen used to evaluate eye tracker accuracy, and spent approximately 1 min with eyes open and approximately 1 min with eyes closed. The distracter phase prevented participants from covert rehearsing. The distracter period was followed by the memory test. At test, the 200 pictures from the study phase and 100 new pictures (foils) were presented for 4 s each. The presentation order was randomized, with the constraint that no more than 4 scenes of the same type (old/new) were shown consecutively. After each scene, participants were prompted to indicate their confidence on whether the scene was old or new using a 6-point response scale, ranging from “very sure old” (1) to “very sure new” (6). This picture of the rating scale remained until the participants responded. Before the next scene, a fixation cross with variable duration (750–1,250 ms) was presented. The procedure for the patients with intracranial electrodes deviated slightly (see below).
MEG was recorded using a 275 whole-brain axial gradiometer system (VSM MedTech/CTF MEG, Coquitlam, Canada) installed in a magnetically shielded room. The data were sampled at 1,200 Hz following a low-pass antialiasing filter with a cutoff at 300 Hz. Additionally, horizontal and vertical electro-oculograms were recorded from bipolar Ag/AgCl electrodes (<10kΩ impedance) placed below and above the left eye and at the bilateral outer canthi. To track the position of the head during MEG recording, we used 3 head coils placed at anatomical landmarks (nasion and both ear canals). Using a real-time head localizer [29], the position of the head relative to the MEG helmet was tracked. Each participant’s nasion, left and right ear canal, and head shape were digitized with a Polhemus 3Space Fasttrack.
Preprocessing of the data was done using the Fieldtrip toolbox [30]. Data were divided into single epochs ranging from 0 to 4 s after picture onset. Epochs were corrected for cardiac artifacts using independent component analysis (ICA) and sorted according to the behavioral performance of each participant’s confidence judgments during the recognition test phase. Pictures that were confidently judged as old (responses 1, 2, and 3) constituted later-remembered scenes, and the remaining pictures were classified as later-forgotten scenes.
An Eyelink 1000 (SR Research) eyetracker was used to monitor the horizontal and vertical movements of the participant’s left eye. Before recording, the eye tracker was calibrated by collecting gaze fixation samples from known target points to map raw eye data on screen coordinates. Participants fixated on 9 dots sequentially on a 3-by-3 grid. After the calibration run, a validation run was performed during which the difference between current gaze fixations and fixations during the calibration was obtained. The calibration was only accepted if this difference was smaller than 1° of visual angle.
Eye tracking and MEG data were simultaneously recorded and analyzed using the Fieldtrip toolbox. Vertical and horizontal eye movements were transformed into velocities. Velocities exceeding a certain threshold (velocity > 6× the standard deviation of the velocity distribution, duration > 12 ms, see Engbert and Kliegl [31]) were defined as saccades. Saccade onsets during stimulus presentations in the study phase defined the events of interest (trials). To avoid potential artifacts from other eye movements and provide a reasonable frequency resolution of 2 Hz, only events that were free of saccades and blinks in a 0.5-s interval prior to saccade onset (i.e., a minimum fixation period of 500 ms) were included. Saccades that occurred during the presentation of scenes that were subsequently judged as old (responses 1, 2, and 3) constituted later-remembered trials. Saccades that occurred during scenes that were subsequently judged as new (responses 4, 5, and 6) constituted later-remembered trials.
After excluding all participants that had less than 30 remaining trials per condition (later remembered or later forgotten), 22 participants were included in the further analyses (3,837 trials in total, mean = 174.4, SE = 13.26; mean number of remembered trials = 109.7, SE = 8.8; mean number of forgotten trials = 64.7, SE = 9.7). In order to display the temporal dynamics of the phase locking, the trials were zero-padded to a length of 1.5 s (i.e., adding 500 ms of zeros before and after the 500 ms of data). Since typical eye movements occur approximately every 250–300 ms, the events with a minimum fixation period of 500 ms may not be representative. Therefore, we conducted additional phase-locking analysis on events with a minimum fixation period of 200 ms prior to saccade onset, including approximately 66% of all detected saccades. In the 22 participants, a total of 25,077 saccades (mean = 1,139.9, SE = 49.6; mean number of remembered trials = 802.7, SE = 57.5, mean number of forgotten trials = 337.1, SE = 26.5) were included. Since this analysis produced higher trial numbers, data from all 36 participants could be analyzed (total number of saccades = 43,226; mean = 1,201.8, SE = 41.3; mean number of remembered trials = 918.7, SE = 49.6, mean number of forgotten trials = 282.9, SE = 23.1). These trials were zero-padded to a length of 0.6 s (i.e., adding 200 ms of zeros before and after the 200 ms of data).
The frequency spectra of the phase and the power of the data were computed by applying a Fourier transformation to the 500 ms of data prior to saccade onset in each event, after multiplication with a hanning taper. Phase and power were calculated for frequencies between 2 and 30 Hz in steps of 2 Hz. The frequency spectra of the phase and the power were used to statistically test differences between conditions (see “Statistics”).
Synthetic planar gradient representations were approximated by relating the field at each sensor with its neighbors’ [32]. On each of the resulting 2 orthogonal gradients, Fourier coefficients were normalized by their amplitude, and the phase-locking index (PLI) [7] was calculated, by extracting the length of the resulting vector after averaging the phase angles:
PLItf=|n-1∑r=1neiktfr|,
where n = number of trials, and eik equals the complex polar representation of phase angle k in trial r, for time-frequency point tf.
This was done for later-forgotten and later-remembered trials. The PLI quantifies the consistency of phases across trials at each given time-frequency point. To control for a bias in PLI due to different trial numbers in conditions, a sample of trials from the condition with the larger number of trials was randomly drawn, with the number of trials in this sample being equal to the number of trials in the condition with less trials. The PLI for this sample was computed. After repeating this procedure 1,000 times, PLI values were averaged. This average reflects an unbiased estimate of the PLI for all trials in the respective condition. After this step, the 2 planar gradients were combined.
In order to depict the temporal dynamics of phase and power in the data, time-frequency representations were computed by a sliding time window approach with a window length of 0.5 s in steps of 50 ms across the zero-padded data. After multiplying a hanning taper to each window, the Fourier transformation was calculated for frequencies between 2 and 30 Hz in steps of 2 Hz.
To identify potential confounds due to differences in spectral power, power was calculated on synthetic planar gradients, using the same approach as outlined above. Instead of computing the PLI, power values were calculated from the Fourier coefficients (amplitude squared).
The PLI analysis on events with a minimum fixation period of 200 ms prior to saccade onset was performed as defined above, with the exception that the window length was 200 ms in the sliding time window approach. Due to the resulting frequency resolution of 5 Hz, phase information was extracted from 5 to 30 Hz in steps of 5 Hz.
To identify PLI differences in source space, a virtual sensor approach applying frequency-domain adaptive spatial filtering (DICS beamformer [12]) was implemented. This algorithm constructs a spatial filter for each specified location (each grid point; 10-mm3 grid). The cross spectral density for the construction of the spatial filter was calculated for the frequency of interest (12–14 Hz, using orthogonal Slepian tapers around a center frequency of 13 Hz with spectral smoothing of +/− 2 Hz), for all trials (common filter approach).
Individual structural MR images, acquired on a 3T Siemens Magnetom Prisma MRI system (Siemens, Erlangen, Germany), were aligned to the MEG coordinate system, utilizing the fiducials (nasion, left and right preauricular points) and individual head shapes recorded after the experiment. A realistic single-shell brain model [33] was constructed for each participant, based on the structural MRIs. The forward model for each participant was created using a common dipole grid (10-mm3 grid) of the grey matter (derived from the anatomical automatic labeling atlas [34]) volume in MNI space warped onto each participant’s anatomy. The Fourier data were projected into source space by multiplying them with the spatial accordant filters, allowing for the phase to be estimated. The PLI was computed on the 2 orientations of the source model, and later averaged, for later-remembered and later-forgotten trials, respectively.
Statistics followed a 2-step approach: first, differences in the intracranial data’s phase locking (later-remembered versus later-forgotten trials) were evaluated in a fixed-effect manner, by concatenating all electrodes from all patients. Cluster-based nonparametric permutation statistics [35] identified continuous frequency clusters with significant differences between later-remembered and later-forgotten PLI while controlling for multiple comparisons over frequencies. Only the cluster with the largest summed value was considered and tested against the permutation distribution. The null hypothesis that later-forgotten and later-remembered trials showed no difference in PLI was rejected at an alpha level of 0.05 (2-tailed).
Second, statistical quantification of the MEG sensor-level data was performed by a cluster-based nonparametric permutation approach [35], identifying clusters of activity on the basis of rejecting the null hypothesis while controlling for multiple comparisons over sensors. The frequency range (12–14 Hz) for the sensor-level statistics was restricted to the outcome of the intracranial data analyses. For each sensor, a test statistic was calculated, based on a paired samples t test comparing the PLI for later-remembered versus later-forgotten trials. Sensors showing a significant effect (p < 0.05, 2-sided t test) were clustered based on spatial adjacency, with a minimum of 2 adjacent sensors required for forming a cluster. T-statistics were summed in each cluster. Again, only the cluster with the largest summed value was considered and tested against the permutation distribution. The null hypothesis that later-forgotten and later-remembered trials showed no difference in PLI was rejected at an alpha level of 0.05 (2-tailed).
Statistical quantification of the source-level data was also performed by a cluster-based nonparametric permutation approach, now considering the clustering in voxel space. The frequency range for the source-level statistics was defined by the outcome of the sensor-level statistics, and the alpha level was set to 0.05 (2-tailed). Cluster-based nonparametric permutation statistics [35] identified continuous spatial clusters with significant differences between later-remembered and later-forgotten PLI while controlling for multiple comparisons over voxels. Only the cluster with the largest summed value was considered and tested against the permutation distribution. The null hypothesis that later-forgotten and later-remembered trials showed no difference in PLI was rejected at an alpha level of 0.05 (2-tailed).
Condition-specific PLIs for later-remembered and later-forgotten trials, separately (see S3 Fig), at the time and frequency of interest (12–14 Hz, −0.25 ms) were statistically quantified by comparing them to a distribution of surrogate PLI values. The surrogate PLI distribution was constructed for each participant and condition by shifting the data points in each condition’s trial circularly along the time axis with a random lag, for each sensor. PLI values were computed as explained above, for 1,000 random shifts. Subsequently, 10,000 surrogate grand averages were constructed by randomly drawing 1 PLI value from each participant’s surrogate distribution for each surrogate grand average. Condition-specific PLI grand averages were compared to these 10,000 surrogate grand averages on each sensor and considered to be significant if they were larger (or smaller) than 97.5% (or 2.5%) of the values in the surrogate grand average values (2-sided test).
Three male patients (age range 30–60 y) with occipital depth electrodes were included in the study. The patients had a history of drug-resistant focal epilepsy and were implanted for diagnostic reasons. Recordings were performed at the Epilepsy Center, Department of Neurology, University of Munich, Germany. The patients gave written informed consent. The procedure and design of the study was identical to the MEG procedure and design (see above), with the exception that only 100 pictures were presented during study and 200 scenes (100 old and 100 new) were presented during the memory test. This was done to compensate for inferior memory performance in a clinical setting.
Patient 1 had 10 depth electrodes implanted, covering bilateral temporal, parietal, and frontal regions and left occipital regions. Patient 2 had 10 depth electrodes implanted, covering right temporal, parietal, and occipital regions. Patient 3 had 11 depth electrodes implanted, covering left frontal, temporal, parietal, and occipital regions. The locations of the electrodes were determined using coregistered preoperative MRIs and postoperative CTs. Electrode locations were converted to MNI coordinates. Intracranial EEG was recorded from Spencer depth electrodes (Ad-Tech Medical Instrument, Racine, Wisconsin, United States) with 4–12 contacts each, 5 mm apart. Data were recorded using XLTEK Neuroworks software (Natus Medical, San Carlos, California, US) and an XLTEK EMU128FS amplifier, with voltages referenced to a parietal electrode site (1,000 Hz sampling rate). All electrodes that either were identified as located in the seizure onset zone or showed interictal spiking activity were excluded from analyses. Data were rereferenced offline to each contact’s neighboring contact (bipolar montage). All bipolar electrodes with both contacts in the occipital cortex were included in the analyses. Additionally, horizontal and vertical eye movements were recorded from bipolar Ag/AgCl electrodes (<10kΩ impedance) placed below and above the left eye and at the bilateral outer canthi.
Study phase data were cut into single epochs, ranging from 0 to 4 s after picture onset. Saccade onsets were extracted from EOG recordings using the method described above (see “Eye tracking acquisition, analyses, and trial definition”). Saccade onsets during stimulus presentations defined the events of interest (trials). All trials were visually inspected for artifacts (e.g., epileptiform spikes). Contaminated trials were excluded from the analyses. The encoding trials were sorted according to each participant’s confidence judgments during the test phase. Pictures that were confidently judged as old (responses 1, 2, and 3) constituted hits, and the remaining pictures were classified as misses. Time-frequency analyses, PLI, and statistics were computed as described above.
|
10.1371/journal.pgen.0030048 | Genome-Wide Linkage Analysis of Malaria Infection Intensity and Mild Disease | Although balancing selection with the sickle-cell trait and other red blood cell disorders has emphasized the interaction between malaria and human genetics, no systematic approach has so far been undertaken towards a comprehensive search for human genome variants influencing malaria. By screening 2,551 families in rural Ghana, West Africa, 108 nuclear families were identified who were exposed to hyperendemic malaria transmission and were homozygous wild-type for the established malaria resistance factors of hemoglobin (Hb)S, HbC, alpha+ thalassemia, and glucose-6-phosphate-dehydrogenase deficiency. Of these families, 392 siblings aged 0.5–11 y were characterized for malaria susceptibility by closely monitoring parasite counts, malaria fever episodes, and anemia over 8 mo. An autosome-wide linkage analysis based on 10,000 single-nucleotide polymorphisms was conducted in 68 selected families including 241 siblings forming 330 sib pairs. Several regions were identified which showed evidence for linkage to the parasitological and clinical phenotypes studied, among them a prominent signal on Chromosome 10p15 obtained with malaria fever episodes (asymptotic z score = 4.37, empirical p-value = 4.0 × 10−5, locus-specific heritability of 37.7%; 95% confidence interval, 15.7%–59.7%). The identification of genetic variants underlying the linkage signals may reveal as yet unrecognized pathways influencing human resistance to malaria.
| In tropical Africa, virtually all children become infected with malaria parasites. Most of them experience several malaria attacks per year, and over a million die from disease complications. Sickle-cell anemia, thalassemias, and other inherited red blood cell disorders indicate that malaria has selected for human genetic variants, but no attempts have so far been reported to systematically screen the human genome for malaria-resistance factors. We describe a genome-wide linkage analysis performed in children living in rural Ghana, West Africa, including approaches to select an informative study cohort and to assess, over a period of 8 mo, individual disposition to malaria parasitemia, fever episodes, and anemia. Families carrying the known malaria-protective red blood cell disorders were excluded, infection intensities were adjusted to the use of mosquito-protection devices, and parasitological and clinical findings were corrected according to the state of partial malaria immunity, which, under constant exposure, gradually develops over the first 10 y of life. The study revealed several genomic regions showing evidence for linkage to the various malaria phenotypes recorded, among them a prominent signal on Chromosome 10 correlated to the frequency of fever episodes. Future identification of genes involved is expected to reveal previously unrecognized pathways that may protect children against malaria.
| Malaria caused by Plasmodium falciparum is one of the leading causes of human morbidity and mortality worldwide, predominantly affecting populations of resource-poor countries in the south [1]. Drawbacks in developing effective control measures have stressed the demand for research aiming at a better understanding of basic elements of parasite biology and disease pathology.
The blood stages of the parasite comprise asexual forms, which maintain the infection and cause disease, and sexual forms, which transmit the infection [2]. Asexual blood parasite counts are the established measure of infection intensity [3], whereby reports on substantial variations over a short period of time indicated that many measurements may be required for appropriate estimates [4].
Clinically, malaria presents as a mild form of acute febrile episodes and anemia, or as a severe form, which comprises a complex syndrome of life-threatening complications [5]. While the severe form causes an enormous humanitarian burden, it does not affect more than 1%–2% of the residents of endemic areas [6], whereas the mild form predominates in terms of quantitative morbidity and economic reasoning [1,7,8]. While the non-specific symptoms of fever, headache, and nausea make the diagnosis of malaria fever episodes difficult to ascertain, a simple case definition proposed by the World Health Organization (WHO) based on fever and parasitemia is generally accepted due to its high sensitivity and specificity in endemic areas, where the vast majority of such episodes are in fact caused by malaria [9].
A second clinical feature of mild malaria is anemia. It affects an enormous number of children in endemic areas [10] and may present as a chronic, subacute, or acute, sometimes life-threatening form [5]. Its pathogenesis is considered multifactorial and may include the destruction of infected and uninfected erythrocytes and bone-marrow dysfunction, whereby the relative contributions of these factors and their roles in the various forms of malarial anemia have not yet been resolved [11].
The effect of human genetics on malaria has long been recognized when the theory of balancing selection was substantiated for thalassemias, sickle-cell anemia, and other red blood cell disorders [12]. Twin studies and heritability estimates have subsequently confirmed the influence of host genetics, which was shown to be most pronounced in children [13–15]. Candidate gene approaches have indicated a number of additional variants to be involved including those of the major histocompatibility complex and a cytokine-gene cluster on Chromosome 5q31-q33 [16]. However, no systematic analysis has been reported to address human genetics in malaria more comprehensively.
Here we report on an autosome-wide linkage analysis for P. falciparum infection intensity and mild clinical malaria among African children selected not to carry any of the classic malaria resistance genes. As markers, 10,000 single-nucleotide polymorphisms (SNPs) were used.
392 siblings of 108 families resident in West Africa were followed over a period of 31 wk, which covered an entire rainy season. Prevalences of P. falciparum blood trophozoites, parasite densities, and interim or present malaria fever episodes were monitored weekly and anemia as indicated by the packed blood-cell volume (PCV) was determined biweekly. Compliance was as follows; 98.8% of 12,152 parasitemia assessments, 95.4% of 6,272 PCV assessments, and 98.5% of 12,152 assessments for malaria fever episodes were recorded with a maximum of data missing per planned visits of single participants of 13/31, 6/16, and 18/31 records, respectively. Results from regression models for analyzing the effect of age, bednet use, and intake of antimalarials on the various phenotypes are summarized in Figure 1. Gender had no significant effect on any of the phenotypes, and therefore was not included in the final regression models used for phenotype corrections.
Based on a ranking that favored high levels of parasite densities in conjunction with high intrafamilial variability (see Materials and Methods), 377 individuals of 68 families were selected for genotyping, including 136 parental individuals and 241 siblings, who formed 330 sib-pairs. Applying the Affymetrix Human Mapping 10K array yielded an overall autosomal calling fraction of 94.5% for the raw genotypes. These were defined as SNPs for which definitive genotypes were obtained. After application of the quality control procedure, 1,524 autosomal markers (15.2%) were excluded from further analysis. The remaining markers yielded a mean information content of 0.976 (SD ± 0.029, range 0.510–1.000).
The nonparametric linkage analysis (NPL) and Haseman-Elston multipoint linkage analysis (HE) were applied (Figure 2A–2D). Parasite prevalence, parasite density, fever episodes, and anemia were analyzed as quantitative phenotypes. The most prominent result was a linkage signal for malaria fever episodes on Chromosome 10p15.3–10p14, which reached statistical significance in both the NPL and HE analyses. NPL showed an asymptotic z score of 4.37 (empirical p-value = 4.0 × 10−5) between SNP markers rs952153 and rs1964428 marking the interval of 5.9–12.0 cM and 2.5–3.5 Mb of the genetic and physical chromosomal maps, respectively (Figure 2D). HE showed a maximum asymptotic logarithm of odds (LOD) score of 3.03 (empirical p-value = 2.1 × 10−4) at marker rs1964428 corresponding to 12.0 cM/3.5 Mb (Figure 2D). The locus-specific heritability was estimated to be 37.7% (95% confidence interval, 15.7%– 59.7%) at 11.2 cM. The linkage region was termed PFFE-1 for P. falciparum-fever episode 1. The signal was robust to variations in data analysis, including the use of a raw phenotype without adjustments for covariates (z score of 4.52), or the use of a wider definition of malaria fever episodes that included afebrile malaria episodes diagnosed by the study physicians (z score of 4.04). The 2.2 z score support interval (corresponding to a 1-LOD support) encompassed a 27.4 cM/11.0 Mb distance containing 71 annotated or hypothetical genes. Functional candidates include genes encoding a platelet-type phosphofructokinase (PFKP) also expressed in red blood cells, an inducible 6-phosphofructo-2-kinase/fructose-2,6-bisphosphatase (iPFK-2/PFKFB3), the alpha chain of the interleukin-15 receptor (IL15RA), the alpha chain of the interleukin 2 receptor (IL2RA), protein kinase C theta (PRKCQ), the GATA-binding protein 3 (GATA3), and a gene similar to that of the interleukin 9 receptor precursor (LOC439945).
A further region with evidence for linkage was found using parasite density as the phenotype. The NPL analysis yielded a signal on Chromosome 13q with a maximum asymptotic z score of 3.73 (empirical p-value = 2.3 × 10−4) between rs2147363 at chromosomal position 55.0 cM/51.4 Mb and rs726540 at 55.5 cM/52.3 Mb (Figure 2B). HE resulted in a LOD score of 1.19 at this position (Figure 2B). The locus-specific heritability was estimated to be 33.7% (95% confidence interval, 9.8%–57.6%). The region was termed PFPD-2 for P. falciparum-parasite density 2, whereby another linkage region with parasite density had previously been reported [17–19]. The 2.2-z support interval of PFPD-2 encompassed 24.2 cM/32.4 Mb containing 158 annotated or hypothetical genes. Possible functional candidates include genes encoding the lymphocyte cytosolic protein 1 (LCP1), S-formylglutathione hydrolase (esterase D, ESD), the cysteinyl leukotriene receptor 2 (CYSLTR2), and the endothelin receptor, nonselective type, (EDNRB).
Furthermore, a signal on Chromosome 1p36 at 18 cM/9 Mb provided evidence for linkage with both parasite prevalence (LOD score of 2.31; empirical p-value = 5.3 × 10−4), and PCV (LOD score of 2.45; empirical p-value = 3.9 × 10−4) at adjacent marker positions (rs205474 and rs966134, respectively; Figure 2A and 2C). The NPL z scores were low in both instances (2.75 and 2.36, respectively).
Finally, no evidence was obtained for linkage of parasite density or malaria fever episodes to 5q31-q33 and to the MHC region on 6q23, respectively, which had previously been reported. In the present study, weak evidence was obtained that malarial anemia might be linked to 5q31-q33 (z score = 2.7, LOD score = 1.8) (Figure 2C).
To our knowledge, this is the first genome-wide approach to identify human genetic variants influencing susceptibility and resistance to malaria. Since the seminal observations on balancing selection with inborn red blood cell disorders, malaria is a prominent element in human genetics. The importance of the classic malaria-protective red blood cell traits is in the present study highlighted by the large proportion of 86% of families found to be affected in the initial survey of our study population. These were excluded from the study in order to concentrate the search on as yet unrecognized human genetic variants [16,20]. As genetic influences were reported to be of particular relevance in childhood malaria [14], we limited our study to children aged 0.5–11 y.
Assessing the phenotype of malaria infection intensity remains a challenge because it is uncertain to which extent any limited number of parasite counts truly reflect the infection intensity [4]. In addition, infection intensities may strongly depend on exposure, which is a variable difficult to assess in field studies. In the present study, the use of bednets and window screens to reduce exposure by preventing mosquito bites was addressed by data adjustments and exclusions of families, respectively (see Materials and Methods). It may be considered an advantage that the NPL and HE methods applied are based on intra-familial evaluations because malaria exposure is likely to be homogeneous within families living in the same households.
As expected, antimalarial treatments had an effect on all phenotypes studied. The influence on parasite prevalences and parasite densities was found to be limited to the two subsequent assessments, therefore it was addressed by correcting the respective values of prevalences and by excluding the corresponding densities (see Materials and Methods). In contrast, the influence on anemia was corrected for by adjusting the overall phenotype because epidemiological observations on the effect of drug resistance on anemia suggest possible long-term effects [22,23]. Concerning the number of fever episodes, no adjustments were made because they might have neutralized the essential phenotypic information due to the direct causal relationship between disease episodes and treatments.
Of all covariates tested, age had the strongest effect and was included in all phenotype adjustments. In children older than 6 mo as in the present cohort, the age effect on malaria in endemic areas is dominated by the gradual development of a certain degree of adaptive immunity, termed semi-immunity. This is reflected by a successive decrease over age of the number of fever episodes, the degree of anemia, parasite densities, and, at relatively high age, parasite prevalences [24–27]. Therefore, the phenotypes addressed may be influenced by both innate resistance and adaptive immunity, whereby innate resistance may have a predominant influence in younger children and adaptive immunity in older ones. This may focus the linkage signals obtained in this study on variants that are relevant under both conditions.
The phenotypes studied showed significant correlations between each other. This is in agreement with the general understanding that all signs and symptoms of malaria result from parasitemia. The explained variances in most instances were low, however, leaving room for separate genetic influences. As expected, the correlation between parasite prevalences and parasite densities was exceptionally high. Despite this, both were included as separate phenotypes because there is evidence to suggest that they are under distinct genetic influences. First, epidemiological findings including those of the present study (unpublished data) indicate that semi-immunity suppresses high parasite densities significantly more efficiently than low parasite densities [24], which suggest distinct elements of adaptive immunity. More importantly, HbS has been shown to protect from high parasite density but not from parasitemia itself [28], indicating that mechanisms of genetic resistance may affect high parasite density specifically.
Evaluation of the data using established linkage methods revealed several prominent linkage signals. Interestingly, locus-specific heritability calculations performed for two of these linkage regions indicated that, in both cases, approximately 35% of the total phenotype variability was attributable to these loci in families who did not carry any of the established malaria resistance factors. These estimates allow us to postulate the effect of a major locus in both instances, which would support a recent conclusion that susceptibility and resistance to infectious diseases may be governed by single major genes rather than by a large number of genes each exerting a small influence [29].
The region showing strongest and significant linkage concerned the phenotype of malaria fever episodes (PFFE-1). Notably, the signal was found in both model-free approaches. Furthermore, it was robust to variations in phenotype definitions, which may be of particular importance because the non-specific symptoms of malaria fever episodes make the clinical diagnosis uncertain. That we found the strongest linkage signal with this particular phenotype may relate to the fact that fever regulation might be similar regardless of whether it is influenced by innate resistance or adaptive immunity, with respect to the age-dependent bias introduced into our study cohort by these two factors, as described above. The underlying genetic variant may be of more general interest because it may relate to the regulation of the systemic inflammatory response.
A number of additional regions with evidence for linkage were identified which did not reach statistical significance. Therefore, they are not discussed in any detail, although experiences in other complex diseases have shown that weaker linkage signals may as well lead to the identification of relevant genetic variants [30]. The linkage regions described comprise a number of genes which may be classified as functional candidates because their products are operative in immune regulation or red blood cell metabolism. However, regarding their established functions, we consider none of them a prime candidate.
No support for our data can be derived from previous linkage studies in mouse malaria. Studying parasite density in murine Plasmodium chabaudi infection, evidence has been obtained for linkage regions on Chromosomes 3, 5, 9, 11, and 17 [31] but not on Chromosome 14, which covers the synteny of the linkage region on human 13q we obtained for P. falciparum- parasite density (NCBI, http://www.ncbi.nlm.nih.gov/Homology/). This is not unexpected because P. falciparum substantially differs from P. chabaudi in that P. falciparum-infected red blood cells adhere to the vascular endothelium [2], which may have a strong influence on parasite biology. Further linkage studies on murine malaria are limited to the phenotype of cerebral manifestations in Plasmodium berghei infections [31], which cannot be compared to our clinical phenotypes of uncomplicated malaria, and identified regions on Chromosomes 1, 11, and 17 but not on 13 and 2, which cover the regions syntenic to PFFE-1 on 10p.
To our knowledge, this is the first time that the Affymetrix HMA10k chip was used for genotyping individuals of African descent. The raw genotypes yielded a call rate of 94.5%, which nearly reached 95% considered sufficient for optimal assay performance [32] and was comparable to 96.9% reported for Caucasians [33]. This provides a basis for using the chip in African populations.
The study was conducted in the Asante Akim North District, Ashanti Region of Ghana, West Africa, a region classified as hyperendemic for malaria by a cross-sectional prevalence of 0.54 for P. falciparum.
Ethical approval was obtained from the Committee for Research, Publications and Ethics of the School of Medical Sciences, Kwame Nkrumah University of Science and Technology, Kumasi, Ghana. All procedures were explained in the local language, and consent was obtained from both parents. Parents of 2,551 families were recruited who had three or more children below the age of 12 y and agreed to participate. Venous blood samples of 2 ml were obtained from both parents and preserved by addition of an equal volume of 8 M urea. The genetic variants of hemoglobin (Hb)S, HbC, alpha+ thalassemia deletion 3.7, and glucose-6-phosphate-dehydrogenase (G6PD) deficiency A-, which were considered to possibly influence susceptibility to P. falciparum parasitemia and mild malaria, were determined, and 346 (13.6%) families were identified not segregating any of the traits.
Of the 346 families, a study group of 392 siblings of 108 families was selected based on (i) the logistic criterion that their homes clustered in 16 of the 30 villages included in the initial survey and (ii) that they did not live in homes equipped with window screens, which a posteriori were found to significantly reduce parasite prevalences from 0.54 to 0.35 (p < 0.001) and other parameters of malaria infection intensity. All families belonged to the ethnic group of Akan.
A subset of 377 members of 68 families were selected from the study group by excluding siblings who were absent at more than two assessments and by a ranking that favored high levels of parasite densities of P. falciparum in conjunction with high intrafamilial variability. The ranking score was based on the product of the mean of log parasite densities of P. falciparum within sib-ships multiplied by the standard deviation of log parasite densities within sibships. Families with highest scores were selected until 377 individuals were identified for genotyping. The genetic study group comprised 136 parents with 241 children, 52.5% boys and 47.5% girls, who, with a mean of 3.54 siblings per family, formed 330 sib pairs. Their median age was 5 y (range 0.5–11 y; IQR 3–8 y).
The children were phenotyped from May 20 to December 20, 2002. Weekly assessments by the visit of a trained physician included a medical history, measurement of body temperature by an infrared ear thermometer, a blood sample by finger prick or heel prick (approximately 100 μl), and in case of disease symptoms, a physical examination. The installation of window screens in homes and the use of bed-nets were recorded.
Weekly malaria smears were prepared at the study site, and in the laboratory they were stained with Giemsa and examined [34]. Parasite species were assessed, and parasite counts were recorded per 200 leukocytes (if >10 parasites/200 leukocytes) or 500 leukocytes (if ≤10 parasites/200 leukocytes) by two independent examiners. Parasite densities were calculated assuming a leukocyte count of 8,000/μl [34]. If the densities as determined in the two counts differed by a factor of three or more, a third independent count was obtained. The median parasite density of two or three counts was included in the analysis.
Weekly point prevalences of malaria parasitemias showed a median of 54% (range 40%–61%) yielding a cumulative period prevalence of 99%. Contributing 98% of all parasitemias, P. falciparum was the predominant species; Plasmodium ovale or Plasmodium malariae were found in 19% and in 90% of these in combination with P. falciparum. For the assessments of parasite prevalences and parasite densities, only P. falciparum and only the disease-causing asexual forms were included, which showed a median point prevalence of 53.1% and a period prevalence of 99%. Parasite densities of P. falciparum ranged from 0–317,360 parasites per μl, with an overall median of 32 and a 75th percentile of 680 in 12,011 assessments.
Anemia was on the spot assessed as PCV by capillary hematocrit centrifugation using 70 μl EDTA anticoagulated capillary tubes (Becton Dickinson, Germany) and mobile centrifuges. To reduce iron-deficiency as a possible confounder, prior to phenotyping, all children were treated against hookworm infection with 400 mg albendazole followed by oral iron supplementation of 2 mg/kg Fe2+ over 6 wk. The median PCV value of 16 biweekly assessments was used as the phenotype in all cases because none of the children showed evidence for any other disease causing substantial anemia. As determined by 16 biweekly PCV measurements, mild anemia defined by a PCV of <33% in the age group of 0.5–6 y and of <36% in the age group of 6–14 y [11] was found at 1,625 of 5,986 assessments (27%) affecting 326 (66%) of the children.
Mild malaria attacks were defined following WHO recommendations [9], first, by either assessing fever by a tympanic temperature of >37.7 °C, corresponding to a rectal temperature of >37.6 °C [35], or reported fever within the previous 4 d and second, a blood smear positive for asexual forms of P. falciparum at any density. The number of fever episodes during the observation period of 31 wk was counted for each individual, whereby multiple episodes within 3 wk were considered recrudescences and counted as one episode [36].
Applying this definition resulted in the identification of 504 malaria fever episodes affecting 257 children, which corresponds to a period prevalence of 66%. In addition, 34 febrile states of other etiologies including upper respiratory tract infections, lower respiratory tract infections, pneumonia, measles, and urinary tract infections were diagnosed by the study physicians. Suspected malaria attacks were treated by a standard dose of chloroquine or amodiaquine following national guidelines, other illnesses as deemed appropriate.
In greater detail, a tympanic temperature of >37.7 °C was measured at 365 visits. Subsequent malaria smears showed that 248 (68%) of the fever attacks occurred in conjunction with P. falciparum parasitemias (and four with other malaria parasites). Fever attacks reported by parents or guardians to have occurred between two examinations were recorded at 870 visits, whereby blood smears at the time of examination were positive for P. falciparum at 522 of these instances, and 133 of the attacks had reportedly been treated with antimalarials. The study physicians at 587 instances suspected malaria attacks at the time of examination, 389 (66%) of which were retrospectively supported by the presence of P. falciparum parasitemia. Of 631 mild malaria attacks as defined by WHO recommendations, 57% were in agreement with malaria diagnoses made by the study physicians, whereas 3% were classified by the study physicians as a febrile illness other than malaria.
Regression models were used for adjusting the raw phenotypes for influences of covariates (Figure 1). Continuous covariates were modeled using multivariable fractional polynomials [37,38] at the nominal 5% test level. The choice of the 5% test level has been discussed in [38]. Specifically, we looked for non-linearity by fitting a second-order fractional polynomial to the data. The best power transformation xp of covariate x is found, with the power p chosen from −2, −1, −0.5, 0, +0.5, +1, +2, +3, where x0 denotes ln x. For example, for p = 0, 0.5 the model is b0 + b1 ln(x) + b2 √x. The set includes the straight line, i.e., no transformation p = 1, and the reciprocal, logarithmic, square root, square, and cubic transformations. Even though the set is small, the powers offer considerable flexibility. The test is performed by comparing the difference in model deviances with a χ2 distribution on 1 degree of freedom. The resulting p-value is approximate and is justified by statistical arguments [37]. Pearson residuals from logistic regression and ordinary residuals from linear and Tobit regressions [39] were used for further analyses.
The difference between the sum of observed weekly P. falciparum parasitemias (1 or 0) minus the sum of predicted probabilities for parasitemias of an individual were calculated and used as phenotype. The predictions were derived from the study population by logistic regression models for parasitemia for each week's set of data with respect to the influences of age, time period since the last use of antimalarials, and the use of bed-nets (Figure 1).
Parasite densities recorded within 2 wk following the administration of antimalarials were found to be significantly lower than parasite densities determined at other time points (Wilcoxon rank sums tests, p < 0.05) and were therefore excluded. Since the 31 weekly parasite-density values (or their log-transformations) of an individual deviated from a normal distribution, quantiles were used. The 75th percentiles were chosen because they were substantially more informative than the median values, which would have been zero in nearly half the assessments. The 75th percentiles were log-transformed, whereby half the detection threshold of 16 parasites per μl was added before taking the logarithm. The phenotype was adjusted to age by calculating residuals from a Tobit regression model, whereby 74 of 392 observations were left censored (Figure 1).
The median values of 16 biweekly PCV assessments were adjusted using a multiple regression model which included the variables of age, number of antimalarial treatments, and use of bednets. Age was modeled as a second-degree fractional polynomial with exponents of −2, 2, the number of antimalarial treatments were considered untransformed, and bednet use was coded as a dummy variable (Figure 1).
The numbers of malaria fever episodes of each individual were adjusted for age and bednet use by calculating residuals from a Poisson regression model. Age was modeled as a second-degree fractional polynomial with exponents of −0.5, −0.5, bednet use was coded as a dummy variable. To avoid outliers for HE, phenotypic values above the 95th percentile were winsorized [40] (Figure 1).
Data analysis showed that siblings of families using mosquito protection by window screens had significantly reduced parasite prevalences, parasite densities, and numbers of fever episodes. As this indicated a marked reduction in exposure, the siblings were excluded a posteriori. The use of bednets was associated with slightly increased parasite prevalences, anemia, and numbers of fever episodes. As bednet use was heterogeneous within one family, all phenotypes were corrected accordingly. Intuitively, the associations appear paradoxical but may be explained by hypothesizing that households living under particularly high exposure, which implies particular molestation by mosquito bites, might use bednets preferentially, albeit with insufficient effectiveness [21].
Venous blood samples of 2 ml were obtained at the end of the study period and preserved by the addition of urea. DNA was extracted by a standard procedure (NucleoMag 96 Blood, Macherey-Nagel, Germany). HbS, HbC [41], and alpha+ thalassemia [42] were determined as described. G6PD variants A (A376G), and A- (A376G, G202A) [43] were typed following a PCR-flourescence-resonance-energy-transfer hybridization method [41], whereby forward and reverse primers (F, R) as well as fluorescein (FL) or LC640red (LC640) (Roche, Germany) labeled anchor (A) and sensor (S) oligonucleotides were used: Variant A376G was typed with F: 5′-tgtgtgtctgtctgtccgtgt-3′, R: 5′-aacggcaagccttacatctg-3′, A: 5′-FL-cgatgatgcagcctcctaccagcgc-3′, S: 5′-LC640-tcaacagccacatggatgccctcc-3′ and variant G202A with F: 5′-cagctgccctgccctcag-3′, R: 5′-cttgaagaagggctcactctgtttg-3′, A: 5′-LC640-cagaaggccatcccggaacagcc-phosphate-3′, S: 5′-gcatagcccacgatgaaggtgttttc-FL-3′. Allele frequencies of HbS, HbC, alpha+ thalassemia, and G6PD A- were estimated to be 0.071, 0.058, 0.175, and 0.063, respectively. G6PD variant A, which was previously reported to show only a slightly decreased enzyme activity of 80% compared to the normal variant B (A376) [44], was retrospectively found not to show a significant influence on any of the phenotypes (Wilcoxon rank sums tests of genotypes AA or A, AB, and BB or B on the phenotypes parasite prevalence (p = 0.13), parasite density (p = 0.55), anemia (p = 0.36), and fever episodes (p = 0.80); therefore children being homozygous, heterozygous, or hemizygous for G6PD B and/or A, were included in the study group. Paternities and maternities were assessed by typing the short tandem repeat markers D1S2782, D6S273 (Genethon, www.genethon.fr), D5S816, D7S2212, D11S1984, D17S1299, and D17S1294 (The Cooperative Human Linkage Center, http://lpgws.nci.nih.gov/CHLC/), in(TG)n of CD36 [45], and IL10G [46], all showing heterozygosity of ≥0.6 in the study population.
A high density SNP genome scan was performed using a whole-genome sampling analysis (WGSA) approach [32] with the Affymetrix GeneChip Human Mapping 10K v2 Array (early access) comprising 10,660 SNP markers with an average heterozygosity in Caucasians of 38% and a mean spacing of 258 kb/0.36 cM (Affymetrix , NetAffx Annotation files, http://www.affymetrix.com). Mapping order and genetic distances of markers were obtained from Affymetrix, the genetic position of 86 markers was unavailable, 295 were X-linked, and 10,279 were from autosomes. Allele frequencies were estimated from 134 founders. The physical positions of the markers were aligned to human DNA sequence information available from NCBI/NIH (http://www.ncbi.nlm.nih.gov/mapview/maps.cgi).
In a first step, gender of participants was verified by investigating heterozygosity and hemizygosity, respectively, at X-linked markers. The relationships between individuals were confirmed by using Graphical Relationship Representation [47]. Genotypes incompatible with Mendelian inheritance were identified with PedCheck [48] and removed in members of the respective families. Unlikely genotypes, e.g., double recombinants, were investigated with Merlin [49], and apparent errors were resolved by deleting the respective genotypes of all family members. The genotypes of two participants showing <80% called genotypes were completely removed from the data. In a second step, SNP markers were excluded if they showed either (i) a deviation from Hardy-Weinberg equilibrium in founders at the nominal p < 0.001 test level, (ii) a genotype calling fraction of <80%, or (iii) a heterozygosity of <5%.
A quantitative trait locus autosomal linkage analysis was performed using the model-free nonparametric linkage method (NPL) [50] and the Haseman-Elston method (HE) [51], assuming an additive genetic model. In order to allow the simultaneous analysis of all SNP loci on a given chromosome in a multipoint approach, we adapted GENEHUNTER [52] to a 64 bit version [53]. Asymptotic z scores from NPL and LOD scores from HE are reported. Additionally, empirical p-values were determined using 100,000 replicates on the basis of the mean information content in the multipoint SNP analysis, which was 0.976 (SD ± 0.029, range 0.510–1.000). Specifically, a single marker was simulated with 20 equally frequent alleles, thus yielding a heterozygosity of 0.95. For the simulation, family structures and phenotypes were left unchanged. Details on this Monte-Carlo approach can be found, e.g., in [54]. A linkage signal of p < 10−4 in either NPL or HE was considered significant [55]. In addition, we describe two linkage signals which were below the threshold of significance, one signal of p < 5·10−4 and another signal of p < 10−3 obtained at the same genomic region with two largely independent phenotypes. We calculated locus-specific heritabilities of additive effects using the regression approach of Sham and colleagues [56]. Applying this method, we estimated the mean and variance from the data and fixed the heritability to 0.5. Decreasing the overall heritability led to moderate increases of the locus specific heritability. Candidate genes were defined as genes relevant to red blood cell structure, red blood cell metabolism or the inflammatory response. |
10.1371/journal.pbio.1002332 | A Latex Metabolite Benefits Plant Fitness under Root Herbivore Attack | Plants produce large amounts of secondary metabolites in their shoots and roots and store them in specialized secretory structures. Although secondary metabolites and their secretory structures are commonly assumed to have a defensive function, evidence that they benefit plant fitness under herbivore attack is scarce, especially below ground. Here, we tested whether latex secondary metabolites produced by the common dandelion (Taraxacum officinale agg.) decrease the performance of its major native insect root herbivore, the larvae of the common cockchafer (Melolontha melolontha), and benefit plant vegetative and reproductive fitness under M. melolontha attack. Across 17 T. officinale genotypes screened by gas and liquid chromatography, latex concentrations of the sesquiterpene lactone taraxinic acid β-D-glucopyranosyl ester (TA-G) were negatively associated with M. melolontha larval growth. Adding purified TA-G to artificial diet at ecologically relevant concentrations reduced larval feeding. Silencing the germacrene A synthase ToGAS1, an enzyme that was identified to catalyze the first committed step of TA-G biosynthesis, resulted in a 90% reduction of TA-G levels and a pronounced increase in M. melolontha feeding. Transgenic, TA-G-deficient lines were preferred by M. melolontha and suffered three times more root biomass reduction than control lines. In a common garden experiment involving over 2,000 T. officinale individuals belonging to 17 different genotypes, high TA-G concentrations were associated with the maintenance of high vegetative and reproductive fitness under M. melolontha attack. Taken together, our study demonstrates that a latex secondary metabolite benefits plants under herbivore attack, a result that provides a mechanistic framework for root herbivore driven natural selection and evolution of plant defenses below ground.
| Plant roots produce diverse and abundant blends of bioactive metabolites. One potential function of these compounds is to protect roots against the devastating effects of below ground herbivore attack. However, examples demonstrating such a protective function in native plant-herbivore systems are lacking. Here, we investigated the interaction between the dandelion (T. officinale) and its native root feeding enemy, larvae of the common cockchafer beetle (M. melolontha). Dandelion is known to release secondary metabolite-rich latex from wounded roots, thus we specifically focused on the potential defensive role of these metabolites. By combining natural variation, genetic manipulation and in vitro assays, we demonstrate that taraxinic acid glucoside, a highly concentrated chemical, deters cockchafer larvae and thereby protects the roots. Dandelion plants with high levels of taraxinic acid benefited from this protection in terms of both vegetative and reproductive fitness. Our study demonstrates that a latex metabolite benefits plant fitness under root herbivore attack, a result which provides a mechanistic framework for root herbivore driven natural selection and the evolution of root defenses.
| Plants produce over 200,000 different metabolites that are not directly needed for their growth and development [1]. Many of these so-called secondary metabolites have a negative impact on insect herbivores [2–6], leading to the hypothesis that they evolved as defenses against the latter [7]. Indeed, recent studies demonstrated that leaf secondary metabolites reduce herbivore damage and thereby counteract the negative impact of herbivores on plant growth, that herbivore abundance covaries with secondary metabolites across different environments, that the exclusion of herbivores leads to rapid changes in genotype frequencies and associated metabolites, and that genes encoding for defensive metabolites can be under differential selection [8–12]. Together, these studies provide strong evidence for the hypothesis that above ground herbivores drive the evolution of leaf secondary metabolites.
In contrast to the leaves, less is known about the role of secondary metabolites in root–herbivore interactions. Roots are often attacked by below ground herbivores, and root herbivore infestation can strongly reduce plant growth and reproduction [13–15]. Furthermore, roots produce diverse and abundant blends of secondary metabolites [16,17], many of which can affect root herbivore behavior and reduce their performance [18]. Furthermore, root secondary metabolites can determine host species ranges in below ground feeding insects [19]. However, if root secondary metabolites enable plants to maintain growth (i.e., vegetative fitness) and reproduction (i.e., reproductive fitness) under root herbivore attack remains unclear. Common milkweed (Asclepias syriaca) families with high and low root cardenolides, for instance, did not differ in their above-ground biomass accumulation when attacked by Tetraopes tetraophthalmus below-ground [20]. Maize lines with high root benzoxazinoid concentrations on the other hand suffered less root damage by Diabrotica virgifera virgifera and had higher yields than lines with low benzoxazinoid concentrations [21]. However, follow-up experiments conducted under more controlled conditions failed to confirm this pattern [5,22]. The lack of knowledge regarding fitness benefits of root secondary metabolites makes it difficult to understand their role in the evolution of plant–herbivore interactions.
In both leaves and roots, secondary metabolites often accumulate in specialized structures including laticifers [23,24], which are among the most common secretory structures of flowering plants [25–27]. Laticifers are elongated individual or interconnected cells whose cytoplasm is called latex [28,29]. Laticifers are often under pressure and release large quantities of latex upon wounding, which can deter or even kill insect herbivores [28,30]. Surprisingly, however, direct evidence that laticifers are defensive, i.e., that they are positively associated plant vegetative or reproductive fitness in the presence but not in the absence of herbivory, is virtually absent [28,31,32]. A study by Agrawal [31] showed that latex exudation is under positive selection in common milkweed under ambient insect pressure. However, whether this pattern is herbivore dependent remains to be elucidated.
One of Europe’s most prevalent native latex-producing plants is the common dandelion (T. officinale agg.) (Flora Helvetica, 5th edition). T. officinale is a species complex consisting of sexual, outcrossing diploids that are native to central and southern Europe and a multitude of apomictic, clonal triploids that are spreading across the globe [33–35]. Similar to many other perennials in temperate ecosystems, the plant relies on its roots for resprouting and flowering in spring. As a perennial plant, both vegetative and reproductive performance contribute to the fitness of the plant. T. officinale produces latex in all major organs, with the highest amounts exuding from wounded tap roots [36]. The latex is dominated by three classes of secondary metabolites: phenolic inositol esters (PIEs), triterpene acetates (TritAcs) and the sesquiterpene lactone taraxinic acid β-D-glucopyranosyl ester (TA-G) [36]. Each compound class accounts for 5%–7% of latex fresh mass [36]. Sesquiterpene lactones and TritAcs can have deterrent and toxic effects against a wide range of organisms [37–40]. In its native range, T. officinale is frequently attacked by the larva of the common cockchafer (also called May bug), M. melolontha (Coleoptera: Scarabaeidae). M. melolontha is among Europe’s largest and most prevalent native root-feeding insects and periodically causes devastating damage to crops and pastures [41–43]. Although the larvae are highly polyphagous, they preferentially feed on T. officinale [44,45].
In this study, we explored the putative defensive function of T. officinale latex secondary metabolites against M. melolontha larvae. First, we investigated which latex secondary metabolites are likely to be involved in root herbivore defense using a correlative approach. Second, we decreased the production of the major candidate compound TA-G by identifying the gene encoding the first committed biosynthetic step and silencing it by RNA interference (RNAi), which allowed testing the effect of TA-G deficiency on plant and insect performance. Third, we purified TA-G to investigate its impact on M. melolontha in vitro. Fourth, we performed a common garden experiment with different T. officinale genotypes to determine whether TA-G reduces the negative impact of M. melolontha on plant vegetative and reproductive performance in the field. Through the above approaches, we demonstrate that TA-G protects the roots and thereby benefits plant fitness in the presence of root herbivores.
Three classes of secondary metabolites dominate the latex of T. officinale: PIEs (Fig 1A, left panel), the sesquiterpene lactone TA-G (Fig 1A, left panel), and TritAcs (Fig 1A, right panel) [36]. We measured the concentrations of the major latex secondary metabolites in 40 triploid T. officinale genotypes collected across central and northern Europe and selected 17 genotypes that displayed maximal variation in latex traits, but minimal variation in growth (S1 Text, S1 Table) to correlate latex secondary metabolites with herbivore performance. M. melolontha larval mass gain was negatively correlated with the concentration of TA-G (Fig 1B, left panel, p = 0.007, r2 = 0.40, linear model), with TA-G accounting for 26% of the observed variance. By contrast, larval mass gain was not correlated to the total concentrations of PIEs or TritAcs (Fig 1B, middle and right panel, p = 0.58 for PIEs; p = 0.53 for TritAcs, n = 17, linear models). Also, the concentration of TA-G was not correlated to the amount of latex that was released from wounded roots (S1 Fig). Surprisingly, latex mass was positively correlated to larval mass gain when analyzed together with TA-G concentration using multiple linear regression (S2 Table, p(TA-G) = 0.003, p(latex mass) = 0.03, linear model). The total amount of TA-G (latex mass * TA-G concentration), on the other hand, was not correlated to larval mass gain (S3 Table). Across the different genotypes, TA-G was constitutively produced and not induced by M. melolontha attack. On the contrary, we observed a trend for a reduction of TA-G concentration in the latex of M. melolontha-attacked roots (S2 Fig, p = 0.08 t-test,). The magnitude of this effect was similar across genotypes (S3 Fig, p = 0.0004, linear model).
To test if TA-G predominantly accumulates in laticifers and to what extent this accumulation is responsible for the overall TA-G concentration in the roots, we measured TA-G concentrations in latex-drained and latex-containing main roots, as well as latex-free root cortex cells. Draining latex from the roots decreased TA-G concentration by a factor of four (S4 Fig, p = 4x10-6, one-way ANOVA). TA-G concentration in the root cortex was as low as in drained roots (S4 Fig, p = 4x10-6, one-way ANOVA). Across the 17 different genotypes, TA-G concentrations in the latex and in the entire main roots were strongly positively correlated, with TA-G concentrations being about 100-fold higher in latex than in main roots (S5 Fig, p = 0.004, linear model). Together, these experiments show that TA-G is predominantly stored in the laticifers, and that latex TA-G is responsible for the overall concentration of TA-G in T. officinale roots.
To investigate the effect of TA-G on M. melolontha preference and T. officinale performance, we identified and silenced a gene that encodes for a germacrene A synthase, the enzyme that mediates the first committed step of TA-G biosynthesis, by RNAi (Fig 2A). To identify germacrene A candidate genes in T. officinale, we sequenced a transcriptome of the main root and the latex and constructed a reference transcriptome with the pooled reads. Putative germacrene A synthases were identified based on amino acid sequence similarity with two known germacrene A synthases from chicory [46]. Through this approach, we obtained full-length sequences of two putative germacrene A synthase genes, ToGAS1 and ToGAS2, which share 71% identity at the amino acid level. Phylogenetic comparison with other Asteraceae terpene synthases revealed that ToGAS1 belongs to the larger of two germacrene A synthase clusters, while ToGAS2 belonged to the smaller cluster (Fig 2B). Heterologous expression in Escherichia coli showed that both recombinant proteins produced (+)-germacrene A when incubated with the substrate farnesyl diphosphate (FDP) (Fig 2C, S6 Fig). To further characterize the two genes, we analyzed their expression in the outer root cortex, latex, and the entire main root. As ToGAS1 was more strongly expressed than ToGAS2 in both latex and entire main roots (Fig 2D), we targeted ToGAS1 through RNAi by expressing a 191 base pair fragment of this gene under the control of the constitutive 35S promoter. A reduction of TA-G by over 90% compared to wild type was observed in three independently transformed lines: −1, −12b, and −16 (“TA-G-deficient lines”). No reduction in TA-G concentration was found in two other lines, −9 and −15, compared to wild type (all designated as “control lines”) (Fig 2E). The amount of exuded latex did not differ between TA-G-deficient and control lines (S7 Fig). ToGAS1 was suppressed by more than 90% in the TA-G deficient lines compared to control lines, whereas ToGAS2 expression was not affected (S8 Fig). These results show that ToGAS1 is involved in TA-G biosynthesis in T. officinale latex.
To test the function of TA-G in planta using the transgenic lines, we first measured the effect of M. melolontha attack on 8 wk-old TA-G-deficient and control T. officinale lines. As noninfested TA-G-deficient and control lines differed in their growth (S9 Fig), we expressed the biomass of herbivore-infested plants relative to the mean biomass of control plants of each genotype. After herbivory, TA-G-deficient lines had lower main and side root mass (Fig 3A, main roots: p = 0.04; side roots: p = 0.01, Kruskal-Wallis rank sum test), but not leaf mass (S10 Fig, p = 0.8, Kruskal-Wallis rank sum test), expressed relative to noninfested plants of each genotype, showing that TA-G-deficient lines suffered a higher percentage of root biomass reduction than control lines. To exclude the possibility that the observed effects are due to differences in root growth, we performed a choice experiment with the TA-G-deficient and control lines using 5 wk-old plants, which did not show any differences in growth or biomass accumulation (S11 Fig). M. melolontha larvae preferred to feed on TA-G-deficient rather than on control lines (Fig 3B, top panel, p = 0.03, binomial test), resulting in three times higher root mass loss in the TA-G deficient than in the control lines under M. melolontha attack (Fig 3C, p = 0.04, paired Student’s t test). Additional metabolic profiling revealed that TA-G-deficient and control lines differed in total root protein levels (S12–S14 Figs). However, no correlation of this trait with M. melolontha behavior was found (S15 Fig). To specifically test the effect of TA-G silencing on latex bioactivity, we painted 6 wk-old carrot seedlings with latex from TA-G-deficient and control plants. M. melolontha preferred to feed on carrots painted with latex from the TA-G-deficient lines compared to that from the control lines as measured three hours after the start of the experiment (Fig 3B, lower panel, p = 0.01, binomial test).
Latex profiling revealed that TA-G-deficient lines also had lower PIE levels, suggesting an interaction between the two pathways (S16 and S17 Figs). To test whether TA-G alone is sufficient to reduce larval consumption, we isolated and purified TA-G by preparative chromatography and performed a feeding experiment with M. melolontha larvae feeding on artificial diet containing TA-G. To determine physiologically relevant TA-G concentrations, we first quantified TA-G in different T. officinale tissues. Latex contained 75 μg TA-G per mg per fresh mass, and the main roots, side roots and leaves contained 0.2–0.7 μg TA-G per mg fresh mass (Fig 4A). For the artificial diet experiment, we used a concentration of 3 μg TA-G per mg diet to represent a natural situation in which M. melolontha feeds on a root that accumulates latex at the site of wounding. Over 24 h, M. melolontha larvae consumed 40% less TA-G containing diet than control diet (Fig 4B, p = 0.045, Student’s t test).
To investigate whether TA-G benefits vegetative and reproductive fitness under M. melolontha attack in the field, we grew 2,040 T. officinale individuals of the experimental population (consisting of the 17 genotypes as described above) in a common garden. We established 20 circular plots, each of them containing 6 individuals of each genotype, and infested half of the plots with 72 M. melolontha larvae (23 larvae per m2) each (S18 Fig), a density similar to the damage threshold in pastures [47]. In the first year during which most plants did not flower, we measured the length of the longest leaf (“maximal leaf length”)—a reliable predictor for leaf and root mass under greenhouse conditions (S19 Fig)—and correlated this parameter with latex secondary metabolite concentrations. To standardize growth rates, we expressed the size increase of the longest leaf of the herbivore-infested plants relative to the size increase of the longest leaf of control plants of the same genotype (“relative leaf growth”). Shortly after infestation of the plants in June, no reduction in leaf growth was observed in the infested plants, and relative leaf growth was not correlated with the concentration of the three latex secondary metabolite classes (Fig 5A, p(June) = 0.38, Pearson’s product–moment correlation). In the course of the growing season, M. melolontha infestation reduced overall plant growth, and a positive correlation between relative leaf growth and TA-G concentration emerged, suggesting that TA-G reduced the negative impact of M. melolontha on plant performance (Fig 5A, p(September) = 0.01, Pearson’s product–moment correlation). In absolute terms, TA-G concentration and leaf growth tended to be positively correlated under M. melolontha attack and negatively correlated in the absence of M. melolontha (S20 Fig). No correlation between relative leaf growth and the total concentrations of PIEs, TritAcs, latex mass, or the total amount of TA-G (latex mass * TA-G concentration) was observed throughout the entire growing season (S21 Fig, S5 Table). Similarly, latex mass did not significantly account for relative leaf length when analyzed in a multiple regression together with TA-G concentration (S6 Table). Leaf length of the herbivore-infested plants was proportional to leaf length of noninfested plants, indicating that plant size did not affect the degree of damage (S22 Fig). To assess whether TA-G also benefits plant reproductive fitness, we correlated the number of flowers to latex secondary metabolite concentration in the following year. At the beginning of the flowering season, TA-G was positively correlated with the relative number of flowers (number of flowers of the herbivore-infested plants expressed relative to noninfested plants of each genotype) in the genotypes that flowered at this time point (Fig 5B, left panel). Genotypes that flowered did not differ in their TA-G concentration from genotypes that did not flower at this time point (p = 1, Wilcoxon rank sum test). No correlation between the relative number of flowers and the total concentrations of PIEs and TritAcs were observed (Fig 5B, middle and right panel). The positive correlation between the relative number of flowers, and TA-G disappeared at the end of the flowering period (p = 0.33, Pearson’s product–moment correlation), likely because almost all M. melolontha larvae had stopped feeding by this time (S7 Table). Together, these data strongly suggest that TA-G reduces the negative effect of root herbivore attack on plant vegetative and reproductive fitness.
In this study, we demonstrate that the sesquiterpene lactone TA-G, a major secondary metabolite of T. officinale, protects the plant against its major native root herbivore M. melolontha. TA-G deters M. melolontha larvae from feeding and thereby directly protects the roots, resulting in a reduction of the negative impact of the root feeder on vegetative and reproductive fitness. The observed pattern indicates that root herbivores may exert positive selection pressure on latex secondary metabolites and may thereby drive their evolution.
Our experiments involving natural variation, chemical manipulation, and genetic modification provide parallel lines of evidence for a negative effect of TA-G on M. melolontha larvae. First, across different T. officinale genotypes, TA-G concentration was negatively correlated with M. melolontha growth. Surprisingly, latex mass was positively associated with larval mass gain. Other unidentified plant traits that benefit the larvae and covary with latex exudation may account for this pattern. Second, purified TA-G reduced food consumption in vitro. Third, TA-G suppression through ToGAS1-silencing increased the attractiveness and consumption of T. officinale roots and decreased the deterrent effect of T. officinale latex towards M. melolontha. Interestingly, silencing ToGAS1 not only affected sesquiterpene lactone biosynthesis, but also plant growth and the accumulation of PIEs. Germacrene A synthases convert FDP into germacrene A. The substrate FDP is a common precursor for sesquiterpenes, triterpenes, and phytosterols [48,49], and the farnesyl residue can bind to growth-regulating proteins of the ras family [50]. It is therefore possible that ToGAS1 silencing affects other branches of the metabolism of T. officinale by changing FDP pool sizes. These observations illustrate the limitations of transgenic approaches as a stand-alone method and highlight the power of combining genetic manipulation, natural variation, and chemical complementation to elucidate the role of plant secondary metabolites in plant–herbivore interactions.
Many studies demonstrate that plant secondary metabolites are toxic to root and leaf herbivores [2–6]. Surprisingly, however, the benefits for the plant often remain unclear, especially for root herbivores [28,51,52]. Plant secondary metabolites may reduce food quality for herbivores and may thereby trigger compensatory feeding, leading to higher plant damage [53]. In addition, herbivore-imposed loss of biomass can lead to an overcompensation of plant growth and sexual reproduction, which may mask the fitness benefits of resistance factors [54–56]. Secondary metabolites can also reduce plant performance in the field by attracting specialized herbivores that use the chemicals as oviposition [57] and foraging cues [22,58,59]. All these factors may constrain the fitness benefits of bioactive secondary metabolites. Finally, the heterogeneity of natural environments, including varying herbivore communities and abiotic factors, can render the detection of fitness benefits difficult [10–12]. We manipulated the abundance of a major root herbivore within artificial populations consisting of plant genotypes that differ substantially in their capacity to produce plant secondary metabolites. This approach allowed us to demonstrate herbivore-dependent vegetative and reproductive fitness benefits under field conditions. Similar experimental designs could be used in combination with transgenic or genetic mapping populations to quantify the contribution of individual herbivore species and herbivore communities to secondary metabolite-dependent fitness benefits in heterogeneous environments [10–12].
So far, clear evidence for the fitness advantage of particular metabolites under insect attack has remained particularly scarce for below-ground plant–herbivore interactions. Vaughan et al. [60] showed that silencing the production of a semivolatile diterpene increased root damage of Arabidopsis thaliana by the opportunistic fungus gnat Bradysia spp. However, it remains unknown to what extent this effect translates into improved plant performance in nature. The lack of knowledge regarding the benefits of secondary metabolites under root herbivory limits our understanding for the evolution of root metabolites. Eschscholzia californica (Fabaceae) mainland populations that are exposed to pocket gopher herbivory had 2.5 times higher root alkaloid concentrations than island populations that are free from this herbivore pressure [61], suggesting that pocket gophers may exert positive selection on this metabolite. We show here that high TA-G concentration benefits plant vegetative and reproductive performance in the presence of T. officinale’s major native root herbivore, M. melolontha, thus providing an evolutionary framework for root herbivore-driven natural selection. TA-G-deficient T. officinale lost more root mass than control lines upon feeding by M. melolontha. In a common garden experiment, TA-G concentration was positively correlated with leaf growth and flower production across natural T. officinale genotypes.
While our data provides evidence that TA-G benefits the plants under M. melolontha attack, we did not obtain strong evidence for costs of TA-G production. Although TA-G concentration tended to be negatively correlated to plant growth across the 17 genotypes in the common garden experiment in the absence of M. melolontha, the correlation was not significant, possibly due to the relatively low number of genotypes that were used for the experiment. Experiments that evaluate putative fitness costs of TA-G production in different environments may provide further insights into the varying fitness effects of TA-G.
Laticifers are commonly assumed to be defensive [28,30]. Toxic metabolites or proteins in the latex can reduce herbivore performance [28], while the stickiness of latex can trap entire insects or glue their mouthparts together [30,62]. Despite the overwhelming evidence that latex reduces herbivore performance, experimental validation that latex benefits plant fitness under herbivore attack remains scarce [31,32]. We show that a toxic metabolite in the latex benefits the plant in the presence but not in the absence of an herbivore and thereby provide an experimental validation of the assumption that microevolutionary processes govern intraspecific variation in plant defense traits. These microevolutionary processes are consistent with the observed macroevolutionary patterns in which latex represents a key innovation that has spurred the evolution of the angiosperms [25]. Taken together, our results furnish an ecological and evolutionary explanation for the high concentrations of root and latex secondary metabolites and highlight the potential of soil-dwelling insects to shape the chemical defenses of their host plants.
All indoor experiments were performed in a climate chamber operating under the following conditions: 16 h light 8 h dark; light supplied by a sodium lamp NH 360 FLX SUNLUX ACE Japan; light intensity at plant height: 58 μmol m2 s−1; temperature: day 22°C; night 20°C; humidity: day 55%, night 65% (unless specified otherwise). Plants were potted in 0.7–1.2 mm sand and watered with 0.01%–0.05% fertilizer with N-P-K of 15-10-15 (Ferty 3, Raselina, Czech Republic).
M. melolontha larvae (S23 Fig) were collected from meadows in Switzerland and Germany. Experiments were performed with larvae in the third larval stage (L3) unless indicated otherwise. Insects were reared individually in 200 ml plastic beakers filled with a mix of potting soil and grated carrots in a phytotron operating under the following conditions: 12 h day 12 h night; temperature: day 13°C, night 11°C; humidity: 70%; lighting: none.
All statistical analyses were performed in R version 3.1.1 [63]. Pairwise comparisons were performed with the agricolae [64] and lsmeans [65] package. Results were displayed using ggplot2 [66] and gridExtra [67]. More details on the individual statistical procedures are given in the experimental sections below.
To investigate the effects of latex secondary metabolites on M. melolontha performance, we measured growth of M. melolontha larvae on 17 T. officinale genotypes. To establish an experimental T. officinale population, we screened 40 triploid genotypes from central and northern Europe [68] for secondary metabolite concentrations and growth rates. Twenty genotypes were selected based on maximal difference of latex chemistry with minimal variation in plant growth rate using cluster analysis (S1 Table, S1 Text). Among these 20 genotypes, three genotypes completely lacked TA-G but were later found to contain other unidentified sesquiterpene lactone glycosides (S24 Fig). These genotypes were subsequently excluded from analysis. The remaining 17 genotypes were used to correlate larval growth with latex secondary metabolite concentrations. For each genotype, 12 plants were infested at an age of 7 wk with one preweighed M. melolontha larva, while 12 plants were left herbivore-free. Eleven days after infestation, M. melolontha larvae were recovered and larval mass difference was determined. To measure the concentration of latex secondary metabolites, main roots were cut 1 cm below the tiller and exuding latex collected into Eppendorf tubes and glass vials, immediately flash-frozen in liquid nitrogen and stored at −80°C until extraction.
For extraction, 1 ml methanol was added to the plastic tubes, and 1 ml hexane containing 0.1 mg*ml−1 cholesteryl acetate as internal standard was added to the glass vials. Both types of vessels were vortexed for 5 min, centrifuged, and the supernatant was stored at −80°C until analysis. Methanol samples were measured on a high pressure liquid chromatograph (HPLC 1100 series equipment, Agilent Technologies), coupled to a photodiode array detector (G1315A DAD, Agilent Technologies) and an Esquire 6000 ESI-Ion Trap mass spectrometer (Bruker Daltonics, Bremen, Germany). For quantification, peak areas were integrated at 245 nm for TA-G and at 275 nm for PIEs, and quantified using external standard curves. Hexane samples were analyzed on an Agilent series 6890 gas chromatograph coupled to a flame ionization detector (GC-FID). Individual TritAcs were quantified based on the internal standard. Methodological details for the analytical procedure have previously been described [36].
Correlations between TA-G, total PIE, total TritAc concentrations and total amount of TA-G (TA-G concentration * latex mass) and M. melolontha mass gain, as well as between TA-G concentration and latex fresh mass, were analyzed using linear models on the mean values of each of the 17 genotypes using the metabolite concentration and latex mass of the noninfested plants, as these measurements were not confounded by differential larval feeding activities. The combined effect of TA-G concentration and latex fresh mass on M. melolontha growth was analyzed with a multiple regression based on the mean values of the 17 genotypes of the control treatment. Differences in TA-G concentration between M. melolontha-infested and control plants were analyzed with Student’s t tests. The correlation across the 17 genotypes between TA-G concentrations of the control, and M. melolontha-infested plants were analyzed with a linear model. The assays were performed in two blocks within two months, and latex for GC analysis was collected from a third batch of plants grown in the same growth chamber under identical conditions.
To investigate to which extent latex contributes to TA-G measured in the main roots, we measured the correlation between TA-G concentration in the latex and main roots, as well as the difference in TA-G concentration between main roots, latex-drained main roots, and largely latex-free outer main root cortex. To analyze the correlation of TA-G concentration in the latex and main roots, we grew 12 plants of each of the above-mentioned 17 genotypes. Main roots of 9 wk-old plants were cut 1 cm below the tiller and the exuding latex was collected. Main roots were separated from side roots and flash-frozen in liquid nitrogen. Latex was extracted using 1 ml methanol containing 10 μg*ml−1 loganin as an internal standard and analyzed on HPLC-DAD as described above. Peak area was integrated at 245 nm for TA-G and normalized to loganin as an internal standard. Main roots were ground to a fine powder, and 100 mg tissue was extracted with 1 ml methanol containing 10 μg*ml−1 loganin, vortexed, centrifuged, and the supernatant transferred to HPLC vials. Main root samples were analyzed the same way as the latex samples, except that the mobile phases consisted of 0.1% acetic acid (A) and acetonitrile (B) using following gradient: 0 min: 5% B, 18 min: 43% B, followed by column reconditioning. Peak area was integrated at 245 nm for TA-G and normalized to loganin as internal standard. The correlation between TA-G concentration in the latex and main roots was analyzed with a linear model.
To investigate the TA-G concentration of main roots, latex-drained main roots and laticifer-free main root cortex, we grew 16 plants from genotype A34 for 12 wk. To harvest drained and nondrained roots, the main roots of 8 plants were cut 2 cm below the tiller into two 1 mm slices. From one slice, the latex was collected using filter paper (Whatman 40) before freezing it in liquid nitrogen (“drained”). The other slice was flash-frozen without collecting latex (“nondrained”). To harvest the laticifer-free cortex tissue, the outer cortex zones of the main roots of 8 plants were dissected with a razor blade and frozen in liquid nitrogen. All samples were ground to a fine powder, weighed, extracted, and analyzed for TA-G concentrations as described above. Differences in the TA-G concentration between root samples were analyzed with a one-way ANOVA. Pairwise comparisons were performed with a Tukey posthoc test.
To identify putative germacrene A synthases, we sequenced the transcriptome of T. officinale main root and latex using Illumina HiSeq 2500. The main roots of six 10 wk-old plants from genotype A34 were cut, the exuding latex was collected into 100 μl homogenization buffer [69] (4 M guanidine isothiocyanate, 100 mM Tris-HCl, pH 7.0, and 5 mM dithiothreitol) and the latex as well as main root samples were frozen in liquid nitrogen. Main root samples were ground to a fine powder and RNA was extracted from 100 mg tissue with the RNAeasy plant mini kit (Qiagen) following the manufacturer’s instructions. For latex RNA extraction, 900 μl QIAzol lysis reagent was added to the latex samples, vortexed, and RNA isolated using RNAeasy Plant Lipid Tissue Mini Kit (Qiagen) following the standard procedure. On-column DNA digestion for main root and latex samples was performed using DNase free RNase (Qiagen). The six samples for main root and latex were pooled equimolarly. TruSeq RNA-compatible libraries were prepared and PolyA enrichment was performed before sequencing the two transcriptomes on an Illumina HiSeq 2500 with 20 Mio reads per library, 100 base pair, paired end. De novo transcriptome assembly on pooled reads from main root and latex sample was performed using Trinity (version Trinityrnaseq_r20131110) [70,71] running at default settings. Raw reads were deposited in the NCBI Sequence Read Archive (SRA) (BioProject Accession PRJNA301484).
To identify putative germacrene A synthase genes in the T. officinale transcriptome, we performed a BLAST analysis using the amino acid sequences of two known germacrene A synthases from chicory as templates [46]. Two putative germacrene A synthase genes were identified and designated as ToGAS1 and ToGAS2. Sequences were deposited in GenBank with the accession numbers KT898039 (ToGAS1) and KT898040 (ToGAS2). For the estimation of a phylogenetic tree of ToGAS1, ToGAS2, and characterized terpene synthases from other Asteraceae (S4 Table), we used the MUSCLE algorithm (gap open, −2.9; gap extend, 0; hydrophobicity multiplier, 1.5; clustering method, upgmb) implemented in MEGA5 [72] to compute an amino acid alignment using a neighbor-joining algorithm (Poisson model). All positions with less than 80% site coverage were eliminated. A bootstrap resampling analysis with 1,000 replicates was performed to evaluate tree topology.
The two putative germacrene A synthases were heterologously expressed in E. coli to verify their biochemical function. The complete open reading frames (S2 Text) encoding putative proteins with 559 amino acids for ToGAS1 and 583 amino acids for ToGAS2 could be amplified from root cDNA using the primers GAS1fwd (ATGGCAGCAGTTGAAGCCAATGGG) and GAS1rev (TTACATGGGCGAAGAACCTACA) for ToGAS1 and the primers GAS2fwd (ATGGCTCTAGTTAGAAACAACAGTAG) and GAS2rev (TCAGTTTTCGAGACTCGGTGGAGGAC) for ToGAS2. The genes were cloned into the vector pET100/D-TOPO (Invitrogen, Carlsbad, CA, USA) and an E. coli strain BL21 Codon Plus (Invitrogen) was used for heterologous expression. Expression was induced by addition of isopropyl-1-thio-D-galactopyranoside to a final concentration of 1 mM. The cells were collected by centrifugation at 4,000 g for 6 min and disrupted by a 4 × 30 sec treatment with a sonicator in chilled extraction buffer (50 mM Mopso, pH 7.0, with 5 mM MgCl2, 5 mM sodium ascorbate, 0.5 mM phenylmethanesulfonylfluoride, 5 mM dithiothreitol and 10% v/v glycerol). The cell fragments were removed by centrifugation at 14,000 g, and the supernatant was desalted into assay buffer (10 mM Mopso, pH 7.0, 1 mM dithiothreitol, 10% v/v glycerol) by passage through an Econopac 10DG column (BioRad, Hercules, CA, USA). Enzyme assays were performed in a Teflon-sealed, screw-capped 1 ml GC glass vial containing 50 μl of the bacterial extract and 50 μl assay buffer with 10 μM (E,- E)-FPP, 10 mM MgCl2, 0.2 mM NaWO4 and 0.1 mM NaF. An SPME (solid phase microextraction) fiber consisting of 100 μm polydimethylsiloxane (SUPELCO, Belafonte, PA, USA) was placed into the headspace of the vial for 60 min incubation at 30°C and then inserted into the injector of the gas chromatograph for analysis of the adsorbed reaction products. GC-MS analysis was conducted using an Agilent 6890 Series gas chromatograph coupled to an Agilent 5973 quadrupole mass selective detector (interface temp, 250°C; quadrupole temp, 150°C; source temp, 230°C; electron energy, 70 eV). The GC was operated with a DB-5MS column (Agilent, Santa Clara, USA, 30 m x 0.25 mm x 0.25 μm). The sample (SPME) was injected without split at an initial oven temperature of 50°C. The temperature was held for 2 min, then increased to 240°C with a gradient of 7°C*min−1, and further increased to 300°C with a gradient of 60°C*min−1 and a hold of 2 min. For the GC-MS analysis with a cooler injector, the injector temperature was reduced from 220°C to 150°C. Chiral GC-MS analysis was performed using a R-βDEXsm-column (Restek, Bad Homburg, Germany) and a temperature program from 50°C (2 min hold) at 2°C*min−1 to 220°C (1 min hold). A (+)-germacrene A synthase (MrTPS3) from chamomile (Matricaria recutita) [73] was used to prepare an authentic (+)-germacrene A standard.
To measure the expression of ToGAS1 and ToGAS2, we harvested latex, main roots and outer cortex cells of 8 wk-old A34 plants. Plants we cultivated in a growth chamber at 18°C and 75% humidity with a 16-h photoperiod (250 μmol m−2 s−1) in 50% Ricoterlanderde (RICOTER Erdaufbereitung AG, Aarberg, Switzerland), 40% sphagnum peat and 10% sand. Plants were fertilized every week with 0.1% Plantaktiv 16 + 6 + 26 Typ K (Hauert HBG Dünger, Grossaffoltern, Switzerland) according to the manufacturer`s instructions. Total RNA was isolated from roots using the GeneJET Plant RNA Purification Mini Kit (Thermo Scientific) according to the manufacturer’s instructions. Total RNA was isolated from latex by dissecting the main root with a razor blade and harvesting 10 μl of expelling latex in 100 μl homogenization buffer (see above). After the addition of 900 μl QIAzol Lysis Reagent, RNA was isolated using the RNeasy Lipid Tissue Mini Kit (Qiagen) according to the manufacturer’s instructions. All RNA samples were treated on column with RNase-free DNase I (Qiagen), and the RNA quality and quantity was determined on agarose gels as well as by spectrophotometric analysis using a ND-1000 spectrophotometer (NanoDrop Technologies).
From each sample, 1 μg total RNA was used for reverse transcription using oligo(dT) primers and SuperScript II Reverse Transcriptase (Invitrogen) according to the manufacturer’s instructions. The cDNA quality was determined by PCR using the primer combination ToActin-fwd (5`-CGTGACATCAAGGAGAAGC-3`) und ToActin-rev (5`-GCTTGGAGATCCACATCTG-3`). Quantitative real-time PCR (qRT-PCR) was performed according to the Minimum Information for Publication of Quantitative Real-Time PCR Experiments (MIQE) guidelines [74]. All primer sequences were validated in silico (Oligo Proberty Scan, Eurofins MWG, http://www.mwg-biotech.com) and accepted when they yielded single amplicons as it was proven by melt curve analysis, agarose gel electrophoresis, and sequencing.
qRT-PCR primers for ToGAS1 (ToGAS1-fwd, 5`-AAATTTCCCTCCTTCAGTATGGGG-3`; ToGAS1-rev, 5`-CTTATTGGAATCCATGGTTGGATCTAC-3`) and ToGAS2 (ToGAS2-fwd, 5`-CTGATACTACCATTGATGCAACCAC-3`; ToGAS2-rev, 5`-CAGCATCAATCTCTTCTGGATAAAG-3`) were designed to anneal at positions of significant sequence divergence between these two GAS genes to yield specific products. The T. officinale transcription elongation factor encoding EF-1α gene was used as a reference and amplified with the primer combination ToEF1α-fwd (5`-ACTGGTACTTCCCAGGCCGATTGC-3`) and ToEF1α-rev (5`-TTGTTTCACACCAAGGGTGAAGGCG-3`). qRT-PCR experiments were carried out with the LightCycler 96 Real-Time PCR System (Roche Diagnostics International Ltd) using the KAPA SYBR FAST qPCR Kit (Kapa Biosystems) according to the manufacturer’s instructions. For each experiment, three biological replicates were performed with two technical replicates for each biological triplicate. Relative gene expression levels were calculated with the LightCycler 96 Application Software (Version 1.1.0.1320, Roche Diagnostics International Ltd). Expression between different tissues and genes was analyzed with a two-way ANOVA, and pairwise comparison of the expression levels of the two genes performed with Tukey posthoc test.
Based on the transcriptome data, we targeted ToGAS1 by RNAi. For silencing, we used the triploid genotype A34 from the above-mentioned 17 T. officinale genotype based on transformation compatibility and intermediate levels of TA-G concentration. A34 is a triploid, synthetic apomict created by crossing a diploid mother from France with diploid pollen from a triploid apomict from the Netherlands [75]. For the construction of the germacrene A synthase RNAi vector, a 191-bp germacrene A synthase PCR fragment was amplified from T. officinale leaf cDNA using the RNAi-dicer optimized primers ToGermA-RNAi-BamHI_fw (5’-aaaGGATCCGGGATAGAGTACCAGAGATT-3’) and ToGermA-RNAi-XhoI_rev (5’-aaaCTCGAGGGCACTAATGTCCCACCTA-3’). This fragment was digested with BamHI and XhoI and inserted into the respective sites of the Gateway vector pENTR4 (Invitrogen). The resulting vector was used for LR recombination (mediated by LR clonase, Invitrogen) with the GW-compatible destination vector pFGC5941 (http://www.chromDB.org), which contains the CaMV 35S promoter and the chalcone synthase intron from Petunia hybrida. The integrity of the constructs was verified by sequencing and subsequently used for Agrobacterium tumefaciens-mediated stable transformation of the T. officinale A34 genotype using the same method as described previously [76].
The T1 generation of 13 transformed lines was screened for latex secondary metabolite concentrations using three individuals of each line. Main root latex of 8 wk-old T. officinale was collected into Eppendorf tubes and frozen in liquid nitrogen. Latex was extracted as described above using 1 ml methanol containing 10 μg*ml−1 loganin as internal standard. Samples were analyzed on HPLC-DAD as described above. Five lines were selected for further molecular and phenotypic characterization. First, the transgenic lines were confirmed to be triploid by flow cytometry. Second, the insertion of the transgene was verified by PCR and sanger sequencing on genomic DNA using the primer combination P2 + ToGermA-RNAi-XhoI_rev and P3 + ToGermA-RNAi-XhoI_rev, with 5‘-TACCTTCCCACAATTCGTCG-3‘f for P2, 5‘-CAGGTATTGGATCCTAGGTG-3‘ for P3 and 5‘-AAACTCGAGGGCACTAATGTCCCACCTA-3’ for ToGermA-RNAi-XhoI_rev. Third, transcript levels of ToGAS1 and ToGAS2 were determined in the T2 generation by qPCR using the primers described above. Four individuals of the TA-G-deficient (RNAi-1, -12b, -16) and control (RNAi-9, -15, WT) lines grown in soil were harvested at an age of 8 wk. Main root tissue was frozen in liquid nitrogen and ground under liquid nitrogen to a fine powder. RNA was extracted using the GeneJET Plant RNA Purification Mini Kit (Thermo Scientific) according to the manufacturer’s instructions. RT-qPCR was performed for ToGAS1, ToGAS2 and ToEF1α as described above (n = 4). Relative gene expression levels were calculated with the LightCycler 96 Application Software. Gene expression was analyzed with generalized linear models using a gamma error distribution for ToGAS1 and a Gaussian error distribution for ToGAS2. Forth, we determined latex fresh mass, latex secondary metabolites, total protein, amino acid and sugar concentrations in the roots. To analyze concentration of TA-G and total PIEs in the transgenic plants, we harvested six individuals of three TA-G-deficient (RNAi-1, -12b, -16) and three control (wild type, RNAi-9, -15) lines in the T2 generation at an age of 8 wk. Main root latex was collected into Eppendorf tubes, which were flash-frozen in liquid nitrogen. 1 ml methanol containing 10 μg*ml−1 loganin and 100 μg*ml−1 salicin as internal standards for TA-G and PIEs, respectively, were added to the Eppendorf tubes. Samples were extracted and analyzed as described above. Differences in the latex fresh mass, as well as in the concentration of TA-G and total PIEs between TA-G-deficient and control lines, were analyzed with one-way ANOVAs. To determine whether total TritAc concentration was affected by silencing, we collected main root latex from 6 individuals of 9 wk-old TA-G-deficient (RNAi-1, -12b, -16) and control lines (wild type, RNAi-9, 15) into glass vials, which were immediately frozen in liquid nitrogen. Samples were extracted with 1 ml hexane containing 100 μg*ml−1 cholesteryl acetate as internal standard. Samples were processed and analyzed on GC-FID as described above. Differences in the concentration of total TritAcs were analyzed with a one-way ANOVA.
To determine soluble protein, free amino acid and soluble sugar concentrations in the roots of the TA-G-deficient (RNAi-1, -12b, -16) and control (wild type, RNAi-9, -15) lines, we harvested 5 individuals of each line at an age of 12 wk. Root systems were exposed, washed, and main and side roots frozen in liquid nitrogen. Root tissue was ground under liquid nitrogen to a fine powder. For extraction, 1 ml 0.1 M TRIS-HCl, pH = 7.0 was added to 100 mg ground tissue, vortexed and centrifuged at room temperature at 17,000 g for 10 min. The supernatant was stored at −20°C until analysis.
Soluble protein concentration was determined using the Bradford assay and quantified using a standard curve of albumin [77]. Differences in soluble protein concentrations between TA-G-deficient and control lines, as well as between root tissues, were analyzed with two-way ANOVAs. To determine free amino acid concentrations, 10 μl of the diluted samples were mixed with 90 μl 13C, 15N labelled amino acid mix (20 μg amino acids*ml−1) (Isotec, Miamisburg, USA) and 100 μl borate buffer (0.9 M, pH = 10.0). To derivatize amino acids, 22 μl 30 mM fluorenylmethoxy-carbonyl chloride was added and samples were vortexed. After 5 min, 800 μl hexane was added to stop the reaction; the samples were vortexed and placed at room temperature until phases had separated. The lower, aqueous phase was analyzed on an Agilent 1200 HPLC system coupled to an API 5000 tandem mass spectrometer according to [78]. To determine soluble sugar concentrations, main root samples were diluted 1:10 and side root samples 1:5 in 0.1 M TRIS-HCl, pH = 7.0. Samples were analyzed on an Agilent 1200 HPLC system (Agilent Technologies, Germany) coupled to an API 3200 tandem mass spectrometer (Applied Biosystems, Germany) equipped with a turbospray ion source operating in negative ionization mode. Injection volume was 1 μl. Metabolite separation was accomplished by an apHera NH2, 15 cm x 4.6 mm x 3 μm. The mobile phase consisted of water (A) and acetonitrile (B) utilizing a flow of 1 ml*min−1 with the following gradient: 0 min: 20% A, 0.5 min: 20% A, 13 min: 45% B, followed by column reconditioning. The column temperature was maintained at 20°C. The ion spray voltage was maintained at −4.5 keV. The turbo gas temperature was set at 600°C. Nebulizing gas was set at 50 psi, curtain gas at 20 psi, heating gas at 60 psi and collision gas at 5 psi. Multiple reaction monitoring (MRM) was used to monitor analyte parent ion → product ion: m/z 178.9 →89 (collision energy (CE) −10 V; declustering potential (DP) −25 V), for glucose; m/z 178.9 →89 (CE −12V; DP −25V) for fructose; m/z 341.03 →58.96 (CE -52V; DP -45V) for sucrose; Both Q1 and Q3 quadrupoles were maintained at unit resolution. Analyst 1.5 software (Applied Biosystems, Darmstadt, Germany) was used for data acquisition and processing. All compounds were identified based on comparison of retention times and mass spectra to those of commercial standards. Glucose, fructose, and sucrose concentrations were quantified using external standard curves obtained from commercial standards. Differences in the sugar concentrations between TA-G-deficient and control lines, as well as between root tissues, were analyzed with two-way ANOVAs.
To investigate whether silencing of ToGAS1 affects plant performance, we measured root and leaf mass of three TA-G-deficient (RNAi-1, -12b, -16) and three control (wild type, RNAi-9, -15) lines. For each line, 24 plants of the T2 generation were cultivated for 8 wk. Half of the plants were infested with one preweighed M. melolontha larva. One week after infestation, plants were separated into side roots, main roots and leaves, and plant material was dried for three days at 60°C before weighing. As TA-G-deficient lines had 50% lower root mass than control lines, resistance was expressed relative to the control plants of each genotype (100*(1 − (mass herbivore plant / mean mass control plants of its genotypes))) and analyzed with Kruskal-Wallis rank sum tests.
In order to test M. melolontha preference for and plant resistance of TAG-deficient and wild type T. officinale plants, three TA-G-deficient (RNAi-1, -12b, -16) and three control (wild type, RNAi-9, -15) lines were tested in a choice experiment with M. melolontha larvae. Larvae were starved for three days prior to the experiment. Each larva was placed into a 180 ml plastic beaker, which was filled with 2–3 mm vermiculite. The roots of 5 wk-old T. officinale seedlings of the T2 generation (grown in soil in seedling trays) were washed, briefly dried with a tissue and the mass of the plants determined. One TA-G-deficient and one control plant was embedded into the vermiculite-filled beaker at opposite edges, with 9 replicates of each possible pairwise combination. Larval feeding site was scored visually after 3 h by inspecting the beakers from the outside. To determine root mass consumption, plants were recovered after 4 h. The plants were separated into shoots and roots and dried for three days at 60°C. Fresh mass was calculated from dry mass using a common conversion coefficient based on the fresh/dry mass ratio of five non-manipulated seedlings of each genotype. Root mass consumption was analyzed using paired Student’s t tests. To obtain sufficiently large sample sizes for a binomial test, larval preference was analyzed by pooling the data for the three TA-G-deficient and control lines. In order to test whether differences in primary metabolites affected M. melolontha choice, we correlated M. melolontha preference and root mass consumption to total main root protein concentration as determined from 12 wk-old plants as described above. Data were analyzed with Pearson’s product–moment correlation.
To test whether M. melolontha preference for TA-G-deficient T. officinale is mediated by latex metabolites, we recorded larval choice among carrot seedlings painted with latex of three TA-G-deficient (RNAi-1, -12b, -16) and three control (RNAi-9, -14, -15) lines. M. melolontha larvae were starved for two days. Each larva was placed into the center of a 180 ml plastic beaker, which was filled with 2–3 mm vermiculite. The roots of the 6 wk-old carrot seedlings were completely covered with latex of 5 mo-old TA-G-deficient and control T. officinale of the T1 generation, cultivated in 21 pots in soil (identical growth conditions as described above, except light source from NH 360 FLX SUNLUX ACE Japan). Seedlings painted with TA-G-deficient and control latex were pairwise arranged on opposite edges of the beaker, resulting in 6–11 replicates of each possible pairwise combination. Larval feeding site was visually scored after three hours. Larval preference was analyzed based on pooled data for the three TA-G-deficient and control lines using a binomial test.
To determine physiologically relevant TA-G concentrations for bioassays, we analyzed TA-G concentration from latex, main, and side roots and leaves from three wild type A34 plants. Main root latex of 11 wk-old T. officinale was collected into Eppendorf tubes, frozen in liquid nitrogen and extracted with 1 ml methanol containing 10 μg*ml−1 loganin as an internal standard as described above. Main roots, side roots, and leaf tissues were flash-frozen in liquid nitrogen and ground to fine powder. 100 mg tissue was extracted with 1 ml methanol containing 10 μg*ml−1 loganin, vortexed, centrifuged, and the supernatant transferred to HPLC vials. All samples were analyzed as described above for the analysis of TA-G in the main roots. Peak area was integrated at 245 nm for TA-G and normalized to loganin as internal standard.
To test whether TA-G deters M. melolontha, we isolated TA-G from latex and added it to artificial diet at a concentration of 3 μg TA-G*mg−1 diet. TA-G was isolated using 300 ml latex methanol extracts obtained from 300 A34 plants grown in the greenhouse. 10 ml water was added to the methanol extract before methanol was completely evaporated using rotary-evaporation. The aqueous solution was loaded on a Sephadex LH20 (GE-Healthcare, Germany) column with 2.5 cm x 30 cm dimensions. The compounds were eluted from the column using water at a flow speed of 1 ml*min−1. 15 ml fractions were collected and analyzed for TA-G on an Agilent 1200 HPLC system (Agilent Technologies, Germany) coupled to an API 3200 tandem mass spectrometer (Applied Biosystems, Germany) equipped with a turbospray ion source operating in negative ionization mode. Injection volume was 5 μl using flow injection analysis. The mobile phase consisted of 0.05% formic acid (A) and acetonitrile (B) utilizing a flow of 1 ml*min−1. 50% A was maintained for 0.5 min. The column temperature was kept at 20°C. The ion spray voltage was maintained at −4.5 keV. The turbo gas temperature was set at 600°C. Nebulizing gas was set at 50 psi, curtain gas at 30 psi, heating gas at 60 psi and collision gas at 5 psi. Multiple reaction monitoring (MRM) was used to monitor analyte parent ion → product ion: m/z 423 →261 (collision energy (CE) −14 V; declustering potential (DP) -40 V), for TA-G; m/z 447 →151 (CE -26V; DP -100V) for di-PIEs; m/z 581 →151 (CE -38V; DP -140V) for tri-PIEs. Both Q1 and Q3 quadrupoles were maintained at unit resolution. Analyst 1.5 software (Applied Biosystems, Darmstadt, Germany) was used for data acquisition and processing.
Pure fractions were pooled and lyophilized using an Alpha 1–4 LD plus freeze dryer (Martin Christ GmbH, Germany). 30 M. melolontha larvae were starved for two days before providing them 300 mg artificial diet supplemented with either 30 μl 30 mg*ml−1 TA-G or with 30 μl water as solvent control (artificial diet: 25 g bean flower, 2.4 g Agar from Roth, Agar-Agar bacteriologist, 105 ml tap water, 33.3 g cooked and mashed carrots). Larvae were allowed to feed for 24 h inside a 180 ml plastic beaker covered with a moist tissue before the remaining food was weighed. Food consumption was analyzed using Student’s t test. Larvae that consumed less than 30 mg diet were considered inactive and were excluded from the analysis.
In order to examine the effects of latex secondary metabolites on plant resistance in the field, we cultivated 2,400 T. officinale from the above-mentioned 20 genotypes in a common garden with and without M. melolontha infestation over one year at a field site in Jena, Germany (50°54'34.8"N; 11°34'00.1"E). Seeds were surface sterilized, germinated on moist filter paper in petri dishes in spring 2013, and the emerging seedlings were transferred onto peat balls after 10 d. One month after germination, seedlings were conditioned outside for one week before planting them into the field site. At the field site, the top 50 cm soil layer was removed and a metal mesh installed on the ground to confine vertical M. melolontha movement. Experimental units (“plots”) were set up using 20 circular plastic tubes (50 cm depth, 2 m diameter) that were placed on the top of the mesh and filled up with the original soil. One wheelbarrow of peat was mixed with the top 20 cm of soil to facilitate plant growth. In each plot, 6 replicates of all 20 genotypes were placed randomly in a quadratic grid with 10 cm distance between plants. To buffer edge effects, these experimental plants were surrounded with an additional row of T. officinale plants, which were excluded from data analysis. Plants were watered as necessary during the first two months after planting and plots weeded monthly. Three weeks after planting, the length of the longest leaf (“maximal leaf length”) was measured for each plant. Subsequently, 72 late L2 or early L3 M. melolontha larvae were homogenously distributed in half of the plots (“herbivory”), while the remaining plots were not manipulated (“control”). As a nondestructive measurement of plant performance, we measured maximal leaf length—a reliable predictor for above and below ground biomass under greenhouse conditions (S19 Fig)—of each plant every month until the end of the growing season.
For statistical analysis, the length of the longest leaf at the beginning of the experiment was subtracted from the maximal leaf length measured each month to reduce the impact of initial differences in plant size (“leaf growth”). To normalize between genotypes, leaf growth of herbivore-treated plants was expressed relative to control plants of the same genotype (“relative leaf growth”).
Relativeleafgrowth(j)=Mean(MaxleaflengthH(ij)−InitialmaxleaflengthH(ij))Mean(MaxleaflengthC(ij)−InitialmaxleaflengthC(ij))
with H = herbivore-infested plants
C = control plant
i = individual plant
j = genotype
Initial max leaf length = maximal leaf length in June before infestation
Correlations between relative leaf growth and TA-G, total PIEs, total TritAcs, latex fresh mass and total TA-G (latex mass * TA-G concentration) were performed based on mean values for each genotype for each month separately with Pearson’s product–moment correlations in R. The combined effect of latex mass and TA-G concentration on relative leaf growth was analyzed with a multiple linear regression. Secondary metabolite concentrations and latex fresh mass were obtained from the experiment with the 20 genotypes in the greenhouse as described above. Three genotypes lacking TA-G were excluded from the analysis due to the presence of unknown and thus unquantifiable sesquiterpene lactone glycosides.
In order to test whether damage caused by M. melolontha in the field is proportional to plant size, we assessed the correlation between leaf length of herbivore-infested individuals and leaf length of non-infested individuals of the 17 genotypes with Pearson’s product-moment correlations.
To correlate secondary metabolite concentrations to reproductive plant fitness, the number of flowers was counted every month in the following year. Correlations between relative number of flowers (number of flowers of the herbivore-infested plants expressed relative to noninfested plants of each genotype) and TA-G, total PIEs, and total TritAcs were analyzed with linear models and Pearson product–moment correlations based on the mean value of each of the 17 genotypes. Difference in TA-G concentration between genotypes that flowered and genotypes that did not flower at the beginning of the flowering season was analyzed with a Wilcoxon rank sum test based on the mean value of each of the 17 genotypes.
|
10.1371/journal.pntd.0005861 | Host regulation of liver fibroproliferative pathology during experimental schistosomiasis via interleukin-4 receptor alpha | Interleukin-4 receptor (IL-4Rα) is critical for the initiation of type-2 immune responses and implicated in the pathogenesis of experimental schistosomiasis. IL-4Rα mediated type-2 responses are critical for the control of pathology during acute schistosomiasis. However, type-2 responses tightly associate with fibrogranulomatous inflammation that drives host pathology during chronic schistosomiasis. To address such controversy on the role of IL-4Rα, we generated a novel inducible IL-4Rα-deficient mouse model that allows for temporal knockdown of il-4rα gene after oral administration of Tamoxifen. Interrupting IL-4Rα mediated signaling during the acute phase impaired the development of protective type-2 immune responses, leading to rapid weight loss and premature death, confirming a protective role of IL-4Rα during acute schistosomiasis. Conversely, IL-4Rα removal at the chronic phase of schistosomiasis ameliorated the pathological fibro-granulomatous pathology and reversed liver scarification without affecting the host fitness. This amelioration of the morbidity was accompanied by a reduced Th2 response and increased frequencies of FoxP3+ Tregs and CD1dhiCD5+ Bregs. Collectively, these data demonstrate that IL-4Rα mediated signaling has two opposing functions during experimental schistosomiasis depending on the stage of advancement of the disease and indicate that interrupting IL-4Rα mediated signaling is a viable therapeutic strategy to ameliorate liver fibroproliferative pathology in diseases like chronic schistosomiasis.
| Liver fibroproliferative diseases drive a considerable fraction of the overall human mortality. This is closely linked to the absence of efficient control measures against such diseases. Schistosomiasis, a chronic disease that affects humans, preferentially causes liver fibrosis and is responsible for devastating economic losses in developing nations where the disease is still endemic. Using reverse genetics, loss-of-function mouse models have helped uncover a protective role for Interleukin-4 receptor (IL-4Rα) in the host survival to experimental schistosomiasis. However, given the contributing role for this receptor in the etiology of some models of tissue fibrosis, its role during chronic schistosomiasis where the highly fibrotic liver of the infected individuals mediate the morbidity had not been properly addressed hitherto. Taking advantage of a third generation mouse model of inducible loss of a gene, we found a debilitating role for IL-4 receptor during chronic schistosomiasis as signaling via this receptor supported both liver inflammation and fibrosis. These findings demonstrate that although the host requires IL-4Rα to survive the acute phase of schistosomiasis, the more clinically relevant morbid phase of the disease is driven by the excessive utilization of this receptor. A therapeutic potential of blocking IL-4Rα to ameliorate liver fibroproliferative disease is therefore suggested.
| Schistosomiasis is a parasitic disease caused by blood-dwelling parasitic flatworms of the genus Schistosoma, mainly, Schistosoma mansoni (S. mansoni), S. japonicum and S. haematobium that are infective to humans and the most clinically relevant [1]. Schistosomiasis is estimated to affect more than 200 million people worldwide and causes up to 200,000 deaths per annum in developing countries [1]. The disease is caused by parasite eggs trapped in the microvasculature of the host organs (liver, intestine and bladder) that induce a vigorous inflammatory response [1]. The kinetics of the ensuing immune responses induced by S. mansoni infection are well defined and characterized [2,3]. Briefly, the outcomes of disease persistence and progression are organ enlargement, fibrosis, scarring, portal hypertension or hematuria (S. haematobium specifically) that drive host morbidity and eventually death in severe cases [1].
The immune response to schistosomiasis, similarly to that against other tissue-dwelling helminth infections [4–6], is highly polarized as it progresses, going from i) an early Th1 response to ii) a powerful Th2 response that culminates as the adult parasite-released eggs are trapped in the host tissues [2,3] and finally iii) a chronic regulatory phase with a minimized but still dominant Th2 response [3,7,8] with a more clinically relevant tissue fibroproliferative pathology. Our current understanding of schistosomiasis pathology heavily relies on the use of experimental murine models [2]. Studies aimed at uncovering factors that drive host protection or susceptibility to schistosomiasis have been conducted using gene-deficient mice. The disease associates with the formation of granulomas and excessive collagen deposition (fibrosis) around tissue-trapped eggs [3,7,8]. An important role was defined for the host immune effector responses in these pathognomonic processes as nude mice [9], T cell-depleted [10–13] or mice with severe combined immunodeficiency [14] failed to form proper fibrogranulomatous responses. Even though Th1, Th17 and Treg responses have been shown to play major roles in regulating schistosomiasis pathogenesis, type 2 immune responses, which are typically induced by the disease-mediating eggs of the parasite [15–17], have been ascribed a more dominant role [3,7,8,18].
Initiation and polarization of type 2 immune responses is orchestrated by interleukin-4 (IL-4) and IL-13 signaling via a common IL-4Rα chain [2,19]. Signaling via this receptor drives the activation of the transcription factor STAT6 in hematopoietic cells, the proliferation of T and B cells, the production of immunoglobulins by B cells, the priming and chemotaxis of mast cells and basophils [2,19]. In non-hematopoietic cells, this receptor plays a central role in inducing airway hyper-responsiveness by enhancing contractions and mucus secretion by gut epithelial cells [20] and has been shown to play a role in STAT6-dependent fibroblast activation leading to collagen deposition that define fibro-proliferative diseases [21,22]. Understandably, mice deficient in this receptor show impaired granuloma formation, enhanced liver damage and augmented gut inflammation that leads to endotoxemia and septic shock during acute schistosomiasis [23–26]. Moreover, studies conducted in our laboratory have refined the requirement of IL-4Rα to a cell-specific level showing that IL-4Rα-responsive macrophages [25], pan-T cells [27] and smooth muscles cells [28] are individually essential for driving host survival and limiting tissue pathology during acute schistosomiasis.
In all these studies employing mice constitutively deficient in IL-4Rα, a critical role for IL-4Rα mediated signaling during acute [25,26] and chronic schistosomiasis [25,29,30] is suggested. However, the constitutive lack of IL-4Rα led such transgenic mice to succumb prematurely to experimental schistosomiasis with high (acute model) as well as low (chronic model) infection doses i.e. chronic model of infections succumb during the acute phase in the absence of IL-4Rα [30] casting an equivoque on the reliability of using models of constitutive deletion of IL-4Rα to assess the role of this receptor during chronic schistosomiasis. Moreover, congenital IL-4Rα deletion has now been shown to affect the development of animals [31], challenging our current knowledge on the role of IL-4Rα throughout experimental schistosomiasis (acute and chronic) using mouse models of constitutive IL-4Rα deficiency.
In this study, the role of IL-4Rα during acute and chronic schistosomiasis was investigated using a novel murine model that allows for inducible deletion of il-4rα gene at any time point during S. mansoni infection. Our findings further confirmed a protective role played by IL-4Rα mediated signaling during acute schistosomiasis. Contrastingly, we showed for the first time that partial deletion of the il-4rα gene, specifically, at the chronic stage of schistosomiasis ameliorates the tissue pathology by reducing type-2 immune responses, improving immune balance between T helper cytokines and skewing the diminished immune response towards a more regulatory profile without affecting animal viability.
Inducible IL-4Rα deficient C57BL/6 mice (RosaCreERT2IL-4Rα-/lox mice, termed iCre-/+IL-4Rα-/lox mice) were established using a modified cyclization recombinase (Cre) under the control of the ubiquitously expressed Rosa promoter. This modified Cre incorporated a mutated fragment of the ligand-binding domain of the estrogen receptor (ERT2), that makes the activity of Cre conditional to the specific presence of Tamoxifen, an estrogen ligand homologue [32]. RosaCreERT2 C57BL/6 mice were intercrossed with IL-4Rα-/- C57BL/6 mice [33] to generate RosaCreERT2IL-4Rα-/- mice (Fig 1A) and subsequently intercrossed with floxed IL-4Rα (IL-4Rαlox/lox) C57BL/6 mice (exon 6 to 8 flanked by loxP) (Fig 1B, [25]) to generate RosaCreERT2-/+IL-4Rα-/lox C57BL/6 mice (Fig 1A). Tamoxifen feeding (Fig 1C) did not impair the fitness of naïve RosaCreERT2-/+IL-4Rα-/lox C57BL/6 mice, as judged by body weight change (Fig 1D). In Tamoxifen-fed RosaCreERT2-/+IL-4Rα-/lox C57BL/6 mice, CreERT2-mediated deletion of the exon 6 to 8 of the il-4rα gene (Fig 1B) was identified by specific Cre-, loxp- and il-4rα- PCR genotyping from tail DNA (Fig 1E), and real-time qPCR from liver (Fig 1G) and spleen DNA (Fig 1H).
Analysis of IL-4Rα surface expression on total cells from different organs by flow cytometry (Fig 2A) demonstrated that IL-4Rα was considerably depleted following administration of Tamoxifen to RosaCreERT2-/+IL-4Rα-/lox mice (Fig 2A and 2B and S1A Fig). To rule out a non-specific toxic effect or bystander immune alteration in Tamoxifen-fed RosaCreERT2-/+IL-4Rα-/lox mice, spleen weights (S1B Fig), organ cellularity (Fig 2C and S1C Fig), seric liver enzymes (S1D Fig), baseline IgE levels (S1E Fig), IL-2-driven proliferative responses of total splenocytes (S1F Fig), frequencies of major myeloid and lymphoid cells (S2A Fig and S2B Fig) and total CD4+ (S2C Fig) and CD8+ (S2D Fig) T cell numbers in spleens and mesenteric lymph nodes (MLN) were determined. This revealed that, amid a minimal cellular deficiency in Spleen CD4+ and CD8+ T cells at baseline in our murine model, organ cellularity, weight and baseline cellular and humoral immune responses were not generally affected in Tamoxifen-fed RosaCreERT2-/+IL-4Rα-/lox mice (Fig 2C, S1 Fig and S2 Fig). Tamoxifen treatment of RosaCreERT2-/+IL-4Rα-/lox mice significantly reduced or even abrogated surface IL-4Rα expression on spleen CD4+ T cells, MLN CD19+ B cells, peritoneal macrophages as well as bone marrow-derived dendritic cells (Fig 2D). Robustness of IL-4Rα knockdown in Tamoxifen-fed RosaCreERT2-/+IL-4Rα-/lox mice was monitored in white blood cells over a period of 16 weeks following Tamoxifen administration (Fig 2E). A relative expression level of 0% was attributed at all times to blood B cells from global IL-4Rα-/- mice, whereas a relative expression level of 100% was attributed to IL-4Rα-/lox mice. Blood B cells from oil-fed RosaCreERT2-/+IL-4Rα-/lox mice oscillated around a level of IL-4Rα expression of 100%, Tamoxifen-fed RosaCreERT2-/+IL-4Rα-/lox mice expressed only 20% of IL-4Rα (Fig 2E), which increased to a maximum of 30% in 16 weeks with no significant body weight changes throughout the monitoring period (Fig 2F).
To assess the cellular knockdown of IL-4Rα functionally, splenocytes from IL-4Rα-/lox littermate controls, oil-fed RosaCreERT2-/+IL-4Rα-/lox, Tamoxifen-fed RosaCreERT2-/+IL-4Rα-/lox and IL-4Rα-/- mice were cultured with or without recombinant IL-4 for 48h then harvested, stained for IL-4Rα expression and analyzed by flow cytometry (Fig 2G). As expected, spleen T (Fig 2H) and B cells (Fig 2I) derived from IL-4Rα-/lox mice and oil-fed RosaCreERT2-/+IL-4Rα-/lox controls up-regulated IL-4Rα expression after the addition of IL-4 (Fig 2G). In contrast, rIL-4 stimulated spleen T and B cells derived from Tamoxifen-fed RosaCreERT2-/+IL-4Rα-/lox or from IL-4Rα-/- mice showed no upregulation of IL-4Rα expression (Fig 2H and 2I). This showed functional impairment of IL-4Rα mediated signaling on cells from Tamoxifen-fed RosaCreERT2-/+IL-4Rα-/lox mice, complementing the above-demonstrated genotypic and phenotypic impairments. Taken together, these results indicated that Tamoxifen administration to the RosaCreERT2-/+IL-4Rα-/lox mouse model leads to a timely, efficient, safe and stably induced IL-4Rα knockdown mouse model.
A protective role for IL-4Rα mediated signaling has been established during acute schistosomiasis, where IL-4Rα deficient mice but not wild-type (wt) mice died around 6 to 8 weeks after natural infection with S. mansoni [25]. To determine whether IL-4Rα mediated signaling is required throughout the course of experimental schistosomiasis, IL-4Rα was knocked down in S. mansoni-infected RosaCreERT2-/+IL-4Rα-/lox mice at the early acute (Tamoxifen administration at 2 weeks post-infection termed Tam2), late acute (Tamoxifen administration at 6 weeks post-infection termed Tam6) and chronic phase (Tamoxifen administration at 16 weeks post-infection termed Tam16) (Fig 3A), as previously defined [3]. As expected, most of IL-4Rα deficient mice (70%) succumbed prematurely to infection with 35 S. mansoni cercariae as early as from 7 weeks post-infection (Fig 3B). Similarly, the viability of Tam2- and Tam6-fed RosaCreERT2-/+IL-4Rα-/lox mice declined rapidly (60 and 50% respectively at week 8 post-infection). From Tam-2-fed, Tam-6 fed or IL-4Rα deficient mice, no death was further reported as from 12 weeks post-infection, at the chronic phase of the disease. This indicated that IL-4Rα is necessary for host survival during acute schistosomiasis, but not required for host survival at the chronic phase of the disease. Indeed, removal of IL-4Rα in Tam16-fed S. mansoni-infected RosaCreERT2-/+IL-4Rα-/lox mice failed to affect the morbidity (as indicated by serum levels of alanine transaminase as a marker of liver disease (S3 Fig) and the mortality (Fig 3B) up to 24 weeks post-infection, further supporting a dispensable role of IL-4Rα mediated signaling during chronic schistosomiasis. Taken together our results suggest that IL-4Rα mediated signaling differentially regulates schistosomiasis disease depending on the stage of the infection.
The findings above demonstrated an impaired viability of S. mansoni-infected mice following IL-4Rα knockdown at 2 weeks post-infection (Tam2). Hence, the immune and histopathological response of Tam2-fed RosaCreERT2-/+IL-4Rα-/lox mice (termed iCre-/+IL-4Rα-/lox Tam2, Fig 4A), which might associate with the host premature death during experimental schistosomiasis was dissected. A consistent reduction of MLN CD4+ (S4A Fig) and CD8+ (S4B Fig) T cell counts was observed in S. mansoni-infected iCre-/+IL-4Rα-/lox Tam2 animals at week 7 when compared to littermate controls, consistent with the S. mansoni-infected global IL-4Rα-/- animals (S4A Fig and S4B Fig). Ex vivo stimulation with a cocktail of PMA/Ionomycin/Monensin for 4h at 37°C and subsequent intracellular FACS analysis of MLN cells from S. mansoni-infected iCre-/+IL-4Rα-/lox Tam2 animals at week 7 resulted in impaired IL-4 production, but similar IFNγ production when compared to control mice (Fig 4B–4D), suggesting a type2 impairment. This was paralleled by a significantly higher rate of reduction in the number of IL-4-producing CD4+ T cells in S. mansoni-infected iCre-/+IL-4Rα-/lox Tam2 animals (~50%, S4C Fig) when compared to the minimal reduction in IFNγ-producing CD4+ T cells reported (~20%, S4D Fig). This impaired IL-4 production was confirmed within the supernatant of anti-CD3-stimulated MLN cells from S. mansoni-infected iCre-/+IL-4Rα-/lox Tam2 animals by ELISA where a greatly diminished production of other type-2 cytokines as well, i.e. IL-13, IL-5 and IL-10, amid rather minimally altered IFNγ responses (Fig 4E) was observed. Subsequently, Type 2 antibody responses (IgG1 and total IgE) appeared markedly reduced, whereas Type 1 antibodies (IgG2a) were similar to control mice (Fig 4F–4H). However, liver egg burden was similar between the different groups (Fig 4I), ruling out a differential level of infection as the cause of the observed diminished type-2 responses in iCre-/+IL-4Rα-/lox Tam2 and IL-4Rα-/- mice. Together, these results demonstrate that knocking down IL-4Rα at the early acute phase of experimental schistosomiasis considerably diminishes host ability to subsequently mount a type-2 immune response. Liver granuloma size (Fig 4J and 4K) and fibrosis (Fig 4L and 4M) were reduced in S. mansoni-infected iCre-/+IL-4Rα-/lox Tam2 similar to global IL-4Rα-/- mice, translating into a significantly reduced level of hepato- (Fig 4N) and splenomegaly (Fig 4O) compared with IL-4Rα-responsive control mice. As expected, from our previous mortality studies [25,27], body weights of S. mansoni-infected IL-4Rα-/- and iCre-/+IL-4Rα-/lox Tam2 mice rapidly declined starting 6 weeks post-infection (Fig 4P) that preceded the death of these animals (Fig 4Q) when compared to IL-4Rα-responsive control mice. Bleeding was visible in the gut of the animals that rapidly succumbed to infection following removal of IL-4Rα. Taken together, these results suggest that IL-4Rα knockdown at the early acute phase of experimental schistosomiasis considerably diminishes the host ability to mount a protective fibro-granulomatous response around the S. mansoni eggs and this was associated with gut bleeding, rapid weight loss and premature death.
As impaired viability of S. mansoni-infected mice following IL-4Rα knockdown at 6 weeks post-infection (Tam6) was observed (Fig 3B), the associated immune and histopathological responses of Tam6-fed RosaCreERT2-/+IL-4Rα-/lox mice (termed iCre-/+IL-4Rα-/lox Tam6, Fig 5A) was investigated. As expected, surface IL-4Rα protein on lymphocytes from S. mansoni-infected iCre-/+IL-4Rα-/lox Tam6 mice was abrogated as demonstrated by flow cytometry (S5A Fig and S5B Fig). A significant reduction of T lymphocytes in the MLN of S. mansoni-infected iCre-/+IL-4Rα-/lox Tam6 animals at week 7 post infection (S5C Fig and S5D Fig) was observed. Ex vivo stimulation and subsequent intracellular FACS analysis of MLN cells from S. mansoni-infected iCre-/+IL-4Rα-/lox Tam6 animals at week 7 post infection revealed impaired IL-4 production, similar to global IL-4Rα-/- mice (Fig 5B and 5C), whereas IFN-γ responses were similar compared to IL-4Rα-/lox and IL-4Rα+/+ control mice (Fig 5B and 5D). This suggests an impairment of type 2 immune responses in S. mansoni-infected iCre-/+IL-4Rα-/lox Tam6 animals as confirmed by the drastic reduction of IL-4-producing CD4+ T cell numbers in the MLN (~83%, S5E Fig), that paralleled a significant but less important reduction of IFNγ-producing CD4+ T cell numbers (~60%, S5F Fig). This reduction of IL-4 production was confirmed by anti-CD3-stimulated MLN cells from S. mansoni-infected iCre-/+IL-4Rα-/lox Tam6 mice and analysis of the released cytokines by ELISA (Fig 5E). Consistently, we observed a significant decrease in the production of type 2 cytokines, i.e. IL-4 and IL-10, but minimally altered IFN-γ responses (Fig 5E). As a result of reduced IL-4, type 2 antibody responses (IgG1 and total IgE) were markedly reduced (Fig 5F and 5G), whereas type 1 antibodies (IgG2a) were similar to control mice (Fig 5H). Liver egg burden was similar between the different groups (Fig 5I), ruling out a differential level of infection as the cause of the observed diminished type 2 responses in iCre-/+IL-4Rα-/lox Tam6 and IL-4Rα-/- mice. Together, these results demonstrate that knock down of IL-4Rα after egg deposition does diminish host ability to maintain the type 2 immune responses. Reduced type 2 responses decreases pathological features, including liver granuloma size (Fig 5J and 5K) and fibrosis (Fig 5L and 5M), hepato- (Fig 5N) and splenomegaly (Fig 5O) compared with IL-4Rα-responsive control mice. However, the body weights of S. mansoni-infected iCre-/+IL-4Rα-/lox Tam6 mice rapidly declined following Tamoxifen-driven removal of IL-4Rα at 6 weeks post-infection similar to IL-4Rα-/- mice (Fig 5P) and culminated into the early death of these animals (Fig 5Q), when compared to IL-4Rα-responsive control mice. Bleeding was visible in the gut of the animals that rapidly succumbed to infection following removal of IL-4Rα. No premature mortality was reported with S. mansoni-infected Tam6-fed IL-4Rα+/+ (control for Tamoxifen side effects), RosaCreERT2-/+IL-4Rα+/+ (control for activated CreERT2) and RosaCreERT2-/+IL-4Rα-/lox (control for CreERT2 Tamoxifen-independent activity) mice when compared to S. mansoni-infected IL-4Rα+/+ (positive control) mice (S6 Fig) ruling out any non-specific effect(s) of Tamoxifen or CreERT2 as mediator(s) of the impaired survival of S. mansoni-infected iCre-/+IL-4Rα-/lox Tam6 mice. Taken together, these results show that IL-4Rα knockdown after egg deposition during the acute phase of experimental schistosomiasis considerably diminishes the host ability to maintain a type 2 immune response around the S. mansoni eggs which associates with gut bleeding, rapid weight loss and premature death.
S. mansoni-infected mice following IL-4Rα knockdown at 16 weeks post-infection, i.e. iCre-/+IL-4Rα-/lox Tam16 mice (Fig 6A) did not result in any weight loss (Fig 6B) or mortality (Fig 6C), for up to 24 weeks post infection. Liver egg burden was similar between the control IL-4Rα-/lox (Fig 6D), ruling out a differential level of infection between both groups of mice. IL-4Rα knockdown considerably reduced liver (Fig 6E) and spleen (Fig 6F) enlargement in chronically infected mice. Apparent scarification was visible on the liver lobes of control mice whereas IL-4Rα knockdown resulted in the removal/reversal/inhibition of liver scarification (Fig 6G). Moreover, IL-4Rα knockdown considerably reduced granuloma size (Fig 6H and 6I) and collagen levels (Fig 6J and 6K) in the livers of chronically infected mice. These data indicated that IL-4Rα knockdown ameliorate granulomatous inflammation, hepato- and splenomegaly and liver fibrosis during chronic schistosomiasis further consolidating the idea of a deleterious role for IL-4Rα signaling in mediating fibroproliferative pathology during chronic schistosomiasis.
To analyse the immune polarization and responses that are triggered by IL-4Rα knockdown during chronic schistosomiasis and associate with the amelioration of tissue disease, IL-4Rα was knockdown in mice chronically infected with S. mansoni at week 16 post infection and the immune response analyzed at week 18 post infection (Fig 7A). A significant reduction of CD4+ (S7A Fig) and CD8+ (S7B Fig) T lymphocytes in the MLN of S. mansoni-infected iCre-/+IL-4Rα-/lox Tam16 animals at week 18 post infection was observed. Our analyses revealed a reduced Th2-mediated production of IL-4 and IL-13, but present IFN-γ and IL-10 production by MLN T cells of IL-4Rα knockdown animals, as judged by frequencies (Fig 7B, gated as per S7C Fig), total numbers (Fig 7C) and ratios (Fig 7D) of cytokine-producing MLN CD4+ T cells (S6 Fig and Fig 7B). Canonical transcription factor analysis (Fig 7E) confirmed this conclusion with reduction of GATA3 but normal Tbet production in effector T cells (Fig 7E and 7F). Interestingly, the frequencies of Foxp3+ regulatory T cell responses were increased in the MLNs of S. mansoni-infected iCre-/+IL-4Rα-/lox Tam16 mice (Fig 7G and 7H), when compared with S. mansoni-infected IL-4Rα-/lox control mice. However, most likely as a result of total CD4+ T cell drop (S7A Fig), Treg cell numbers were reduced following Tam16 treatment in S. mansoni-infected iCre-/+IL-4Rα-/lox Tam16 animals when compared to their littermate controls (Fig 7I). Serum titers of type 2 antibodies (IgG1 and total IgE) were reduced (Fig 7J and 7K) but not type 1 antibodies (IgG2a, Fig 7L), supporting reduced type 2 responses in S. mansoni-infected iCre-/+IL-4Rα-/lox Tam16 mice. Of interest, regulatory B cell frequencies increased during infection and particularly in infected iCre-/+IL-4Rα-/lox Tam16 mice (Fig 7M and 7N) amid a rather stable total count in S. mansoni-infected iCre-/+IL-4Rα-/lox Tam16 animals when compared to littermate controls (Fig 7O).
Innate type 2 immune effectors i.e. eosinophils [34,35], ILC2 [36] and macrophages [37] have been positively linked to liver fibrosis, the pathophysiological process that drives the host morbidity during chronic schistosomiasis. Conversely, arginase expression by macrophages has been shown to counter tissue inflammation and fibrosis [38]. The analysis of the MLN cells of S. mansoni-infected iCre-/+IL-4Rα-/lox Tam16 mice for these cell types by flow cytometry (S7D Fig–S7G Fig) revealed that the pro-fibrotic innate effectors i.e. eosinophils (Fig 7P, S8A Fig and S8B Fig), ILC2 (Fig 7Q and S8C Fig) and macrophages (S8D Fig) were significantly diminished. Conversely, the mean arginase expression by macrophages (S8E Fig) was not affected in S. mansoni-infected iCre-/+IL-4Rα-/lox Tam16 mice, when compared to S. mansoni-infected IL-4Rα-/lox control mice. This suggests that IL-4Rα knockdown in chronically infected mice does skew the MLN response away from a pro-fibrotic response. Taken together, these results suggest that knocking down IL-4Rα at the chronic phase of experimental schistosomiasis considerably skews the host immune response away from the type 2 arm of the immune response, fosters a qualitatively more regulatory, anti-inflammatory and anti-fibrotic profile with no deleterious effect on host survival.
Taking advantage of a newly established temporal inducible IL-4Rα deficient mouse model, we demonstrated that interrupting IL-4Rα mediated signaling prevents the onset and maintenance of egg-driven type 2 immune responses and its associated fibro-granulomatous inflammation during schistosomiasis. Whereas early knockdown of the receptor during the acute phase of the disease led to aggravated morbidity and mortality, late targeting at the chronic phase considerably ameliorated fibrogranulomatous inflammation and reduced hepato- and splenomegaly without impairing the animal viability. Amelioration of chronic schistosomiasis pathology was further associated with reduction of type 2 immune effector responses but expansion of regulatory T and B cells, suggesting that IL-4Rα mediated immune responses are detrimental in chronic schistosomiasis. Hence, therapeutic intervening of IL-4Rα mediated signaling to reduce type 2 responses might provide a strategy to ameliorate fibroproliferative pathology in diseases like chronic schistosomiasis.
The group of Cheever et al. (1994) were the first to show that abrogation of type 2 immune responses in S. mansoni-infected mice resulted in impaired granulomatous inflammation around the trapped eggs in tissue. A subsequent study by Chiaramonte et al. [21] showed the importance of IL-13-mediated signaling in fibrogenesis through the blockade of IL-13. Moreover, this group further reported a critical role for IL-13 in granuloma formation induced by S. mansoni eggs [39]. An independent group further reported on the achievement of significantly reduced tissue fibrosis by blocking type 2 responses in S. mansoni-infected mice with anti-IL-4 antibody treatment [40]. This indicated that IL-4-orchestrated type 2 responses, as well as IL-13-driven responses, are all causally linked to fibrogranulomatous pathology. More recently, using Schistosomiasis-infected IL-4Rα deficient mice, we and others demonstrated reduced fibrogranulomatous inflammation. However, these mice died during acute schistosomiasis due to cachexia [25]. Using inducible IL-4Rα deficient mice in the present study, we now dissected the role of IL-4/IL-13-mediated type 2 responses during acute and chronic murine schistosomiasis. IL-4Rα removal (knockdown) early during schistosomiasis infection led to impaired type 2 responses with reduced fibrogranulomatous inflammation around the trapped eggs of the parasites, as demonstrated before. This resulted in exacerbated morbidity and premature death of the animals, as demonstrated previously in IL-4Rα deficient mice [25]. Thus, IL-4Rα-elicited type 2 immune effector responses like granuloma formation and fibrosis are important for the host survival during acute schistosomiasis. This concept has been previously established where a tissue protective role of these responses against the toxic secretions of the parasite eggs was suggested [41]. Interestingly, liver integrity was not affected after acute knockdown of IL-4Rα in S. mansoni-infected animals. This finding argues against liver toxicity being the pathological event that drives death in these animals. A more likely explanation for their premature death would be the intensive gut bleeding reported in our previous study on IL-4Rα deficient mice [25], and similarly observed in this study. As a result of the compromised gut integrity, bacteria would translocate to the blood stream and death by septic shock would ensue, as previously demonstrated [25].
Tamoxifen-induced knockdown of the IL-4Rα after egg deposition during the chronic phase (16 weeks post-infection) uncovered a hitherto unappreciated facet of the IL-4Rα mediated type 2 responses. Indeed, IL-4Rα knockdown during chronic schistosomiasis did not lead to gut bleeding and did not affect animal viability but ameliorated liver pathology with reduced granuloma size and fibrosis in the liver and no visible scarification and reduced level of liver and spleen enlargement. This clearly suggests that IL-4Rα mediated type 2 responses are detrimental during chronic schistosomiasis and the cause for fibroproliferative liver pathology. Of interest, regulatory T and B cell compartments were significantly increased following IL-4Rα removal during chronic schistosomiasis. It is tempting to associate the beneficial effect of IL-4Rα blockade on tissue pathology during chronic schistosomiasis to the enhanced regulatory response observed. In fact, previous studies have reported an amelioration of the fibrogranulomatous inflammation during chronic schistosomiasis by Foxp3+ regulatory T cells [42,43]. Whether IL-4Rα mediated signaling causally dictates the anti-inflammatory and anti-fibrotic activities of these regulatory cells during chronic schistosomiasis is not known.
As of now, a role for IL-4Rα signaling in the development of immune hyporesponsiveness after chronic exposure of host immune cells to schistosomal antigens has been demonstrated [44]. This further re-emphasizes the potential of IL-4Rα in modulating the dynamics of the host regulatory responses during chronic diseases such as schistosomiasis. Such a potential has already been widely reported with a negative regulation of Foxp3+ Tregs and a loss of their suppressive capacity suggested to occur when the IL-4Rα signaling was solicited [45–47]. Alternatively, however, the remnant IL-4Rα mediated signaling in Tam16 mice argues against the absence of Th2 responses as the sole driver of the ameliorated pathology observed. A rather noticeable finding is the upregulation of other cytokines i.e. IL-10 and IFNγ resulting in a better balanced cytokine profile between T cells producing IL-4, IFN-γ and/or IL-10. Consequently, the impairment of IL-4Rα mediated signaling during chronic schistosomiasis by inducing a more equilibrated and mixed Th profile might prevent untoward immune polarization and tissue immunopathology. This hypothesis is strongly supported by the recently demonstrated role for immune balance rather than strong immune polarization in controlling fibrogranulomatous pathology during experimental schistosomiasis [48]. Further experiments are now required to empirically disentangle these hypotheses. What remains clear and worth focus at present is the fact that targeting IL-4Rα mediated signaling for the management of non-communicable type 2-mediated diseases in humans is in advanced clinical trials [49–51]. Understandably, building on the present study, the translatability of targeting IL-4Rα mediated signaling during fibroproliferative diseases like chronic schistosomiasis is further supported.
What do we add to the current knowledge on the control of fibroproliferative disease? It should be recalled that our present report builds on the previous observations made during IL-13 blockade experiments where a key role for this cytokine, and the indication of the potential of the IL-4Rα signaling axis in driving fibroproliferative responses during experimental schistosomiasis was defined [21]. In as much as an efficient anti-fibrotic strategy already transpired from the sole blockade of IL-13 [21], the noticeable and independent pro-fibrotic effect of IL-4 [21,40] altogether argues for the higher anti-fibrotic potential of dually targeting IL-4 and IL-13 by blocking IL-4Rα rather than IL-13 alone. The picture might not be that straightforward, however, as caution should also be exerted in dually targeting IL-4 and IL-13 via IL-4Rα given that IL-4 unlike IL-13 is critical for type 2 immune responsiveness. A state of immune deficiency might therefore arise from IL-4Rα targeting as opposed to IL-13 targeting where Th2 responses are optimally elicited [21]. Also, consistent with the observation that IL-13 targeting was not toxic for the host [21], our present report shows that IL-4R targeting does not impair animal fitness. This strongly argues for the safety of our approach. Conclusively, as of yet, one could therefore speculate on an added value of targeting IL-4Rα rather than just IL-13 given the different profibrotic potentials of IL-13 [21,39,52,53] and IL-4 [40,54–58] as both cytokines signal through IL-4Rα. Clearly, such a conclusion would still need to be experimentally validated.
In summary, we provide evidence on the role of IL-4Rα during experimental schistosomiasis whereby early signaling helps the host survive the acute phase of the disease whereas signaling at the late chronic phase mediate the morbidity. Targeting IL-4Rα might therefore represent a novel therapeutic strategy against the fibroproliferative pathology that drives the morbidity of fibrotic diseases like chronic schistosomiasis.
IL-4Rα-/-, IL-4Rα-/lox and CreERT2 mice on a C57/BL6 background were previously described [2,25,33]. We generated a novel inducible IL-4Rα deleting mouse strain (RosaCreERT2-/+IL-4Rα-/lox) by intercrossing transgenic RosaCreERT2-/+ mice with IL-4Rα-/- and IL-4RαLox/Lox mice. CreERT2 transgenic negative littermates (IL-4Rα-/lox) expressing functional IL-4Rα were used as controls in all experiments. Mice were maintained in the University of Cape Town specific pathogen-free animal facility in accordance with the guidelines established by the Animal Research Ethics committee of the Faculty of Health Science of the University of Cape Town and the South African Veterinary Council (SAVC).
All animal experiments were conducted under strict recommendation of the South African national guidelines and of the University of Cape Town practice for laboratory animal procedures as outlined in protocols 010/048 and 014/003 reviewed and approved by the Animals Research Ethics Committee of the Faculty of Health Science of the University of Cape Town. Both male and female mice aged 6–12 weeks were used for all experiments. Care was taken under these protocols to minimize animal suffering in accordance with the guidelines of the Animal Research Ethics committee of the Faculty of Health Science of the University of Cape Town and the South African Veterinary Council (SAVC).
Mice were infected percutaneously via the abdomen with 35, 80 or 100 cercariae, as indicated, with a Puerto Rican strain of Schistosoma mansoni obtained from infected Biomphalaria glabrata (a generous gift from Adrian Mountford, York, UK). Eggs were purified from digested sections of liver or ileum from infected animals and counted at 40× magnification as previously described [40]. To activate il-4rα gene excision by CreERT2, Tamoxifen (Sigma, Deisenhofen, Germany) solubilized in vegetable oil was administered by oral gavage to mice for four consecutive days (2.5mg/day).
Polymerase chain reaction was used to confirm the genotype of RosaCreERT2-/+IL-4Rα-/lox mice. PCR conditions were 94°C for 2 minute, 94°C for 20 seconds, 45°C for 30 seconds, and 72°C for 20 seconds for 40 cycles. To quantify the efficiency of deletion, real-time PCR was performed on genomic DNA from liver and spleen cells using primers specific for IL-4Rα exon 5 (control) and exon 8 (deleted by CreERT2 activation) as described previously [25].
Il-4Rα surface expression was detected on splenocytes, lymph node cells, lung cells, hepatocytes, bone marrow cells and peritoneal exudate cells by phycoerythrin (PE) anti-CD124 (IL-4Rα, M-1). Cell subpopulations were identified with Alexa Fluor 700, BD Horizon V500, BD Horizon V450, PerCP-Cy5.5, APC, APC-Cy7, Fluoroscein isothiocyanate, PE, PE-Cy7 or biotinylated monoclonal antibodies against CD3, CD4, CD8, CD19, Lineage, CD1d, CD5, Foxp3, Gata-3, T-bet, IL-4, IL-13, IFN-γ, IL-10, F4/80, Ly6G, CD11c, MHCII, SiglecF, T1/ST2, ICOS, Arginase, CD11b. Biotin-labeled antibodies were detected by Allophycocyanin or PercP-Cy5.5. For staining, cells (1x 106) were labeled and washed in PBS, 3% FCS and 0.1% NaN3. Between each step of staining, cells were washed extensively. For intracellular cytokine staining, cells were restimulated with a cocktail of PMA/Ionomycin/Monensin for 4h at 37°C then fixed in 2% PFA, permeabilized and cytokine production was analyzed as previously described [25]. For intranuclear staining, a commercially available transcription buffer set (BD Bioscience) was used as per the manufacturer’s instructions. All antibodies were from BD Pharmingen (San Diego, CA) except where noted otherwise. Stained cells were then acquired on a LSR Fortessa machine (BD Immunocytometry system, San Jose, CA, USA) and data were analyzed using Flowjo software (Treestar, Ashland, OR, Usa).
Tissue samples were fixed in neutral buffered formalin, processed, and 5–7 μm sections stained with hematoxylin and eosin (H & E). Granuloma diameter of 20–50 granulomas per animal was determined using an ocular micrometer (Nikon NIS-Elements, Nikon Corporation, Tokyo, Japan). For fibrosis assessment, tissue sections were stained with chromotrope 2R and analine blue solution (CAB) and counterstained with Wegert's hematoxylin for collagen staining. Complementarily, a modified protocol of tissue hydroxyproline quantification was used [59]. In brief, weighed liver samples were hydrolyzed and the supernatant was neutralized with 1% phenolphthalein and titrated against 10 M NaOH. An aliquot was mixed with isopropanol and added to a chloramine-T/citrate buffer solution (pH 6.0) (Sigma). Ehrlich's reagent solution was added and measured at 570 nm. Hydroxyproline levels were calculated by using 4-hydroxy-L-proline (Calbiochem) as standard, and results were expressed as μg hydroxyproline per weight of liver tissue that contained 104 eggs.
Statistical analysis was conducted using GraphPad Prism 4 software (http://www.prism-software.com). Data were calculated as mean ± SD. Statistical significance was determined using the unpaired Student's t test, One-Way or Two-Way ANOVA with Bonferroni's post test, defining differences to C57BL/6, IL-4Rα-/lox or oil-treated RosaCreERT2-/+IL-4Rα-/lox as significant (*, p≤0.05; **, p≤0.01; ***, p≤0.001).
|
10.1371/journal.pbio.1000609 | Retinoic Acid Functions as a Key GABAergic Differentiation Signal in the Basal Ganglia | Although retinoic acid (RA) has been implicated as an extrinsic signal regulating forebrain neurogenesis, the processes regulated by RA signaling remain unclear. Here, analysis of retinaldehyde dehydrogenase mutant mouse embryos lacking RA synthesis demonstrates that RA generated by Raldh3 in the subventricular zone of the basal ganglia is required for GABAergic differentiation, whereas RA generated by Raldh2 in the meninges is unnecessary for development of the adjacent cortex. Neurospheres generated from the lateral ganglionic eminence (LGE), where Raldh3 is highly expressed, produce endogenous RA, which is required for differentiation to GABAergic neurons. In Raldh3−/− embryos, LGE progenitors fail to differentiate into either GABAergic striatal projection neurons or GABAergic interneurons migrating to the olfactory bulb and cortex. We describe conditions for RA treatment of human embryonic stem cells that result in efficient differentiation to a heterogeneous population of GABAergic interneurons without the appearance of GABAergic striatal projection neurons, thus providing an in vitro method for generation of GABAergic interneurons for further study. Our observation that endogenous RA is required for generation of LGE-derived GABAergic neurons in the basal ganglia establishes a key role for RA signaling in development of the forebrain.
| The vitamin A metabolite retinoic acid is an important signaling molecule needed for development of the central nervous system. Previous studies have shown a role for retinoic acid in regulating genes involved in the generation of motor neurons both in the hindbrain and spinal cord, but the role of retinoic acid in the forebrain has remained elusive. Here, we investigated mice that lack the ability to metabolize vitamin A into retinoic acid in the forebrain. Although no defects were observed in the generation of forebrain cortical neurons, we did observe a serious deficiency in GABAergic neurons, which provide inhibitory input to cortical neurons. Specifically, our results reveal that retinoic acid is required for forebrain neurons to activate an enzyme that converts glutamate to the inhibitory neurotransmitter GABA. We also find that retinoic acid treatment of human embryonic stem cells could stimulate production of GABAergic neurons. Deficiencies in GABAergic neurons have been associated with several neurological disorders, including Huntington's disease, autism, schizophrenia, and epilepsy. Knowledge of how GABAergic neurons are generated may aid efforts to treat these diseases.
| The embryonic forebrain, deriving from the most anterior part of the neural tube, comprises a complex set of structures in the developing brain. This complexity arises mainly due to the heterogeneity of the neurons comprising it in terms of morphology, structure, function, and genetic specification. During forebrain development, the dorsal domain (pallium) gives rise to the cortex while the ventral region (subpallium) generates the basal ganglia, i.e. the pallidum and the striatum, which, respectively, originate from the medial and lateral ganglionic eminences (MGE, LGE) [1]. The progenitor zones of the subpallial ganglionic eminences are the origin of chemically diverse populations of gamma-aminobutyric acid (GABA)ergic interneurons and projection neurons. GABAergic interneurons are inhibitory local circuit neurons modulating neuronal activity and synaptic plasticity. GABAergic neurons comprise ∼20% of all neurons within the cortex and hippocampus and ∼95% of the neurons within the striatum [2]–[4]. Whereas GABAergic projection neurons generated in the germinal zones of the LGE migrate radially to the adjacent striatum, GABAergic interneurons arise from both the MGE and LGE and migrate using multiple tangential routes to the olfactory bulb, cortex, and hippocampus [5]–[8]. Disturbed GABAergic neuron function has been associated with several neurological disorders including Huntington's disease, autism, schizophrenia, bipolar depression, and epilepsy [9]–[12]. Thus, a source of GABAergic neurons for cell replacement therapy may be useful for treatment of these neurological diseases.
GABAergic neuronal diversity emerges during embryogenesis and depends on both the timing and the creation of specific anteroposterior and dorsoventral progenitor domains by the coordinated action of several transcription factors expressed by distinct progenitor populations [13],[14]. In contrast, little is known about the extrinsic signaling pathways coordinating GABAergic specification in the basal ganglia. Retinoic acid (RA) functions as an extrinsic signal that regulates patterning of rhombomeres in the hindbrain and neuronal differentiation in the spinal cord [15]–[17], but the role of RA in forebrain development remains unclear. RA is derived from vitamin A through a two-step enzymatic process, employing retinol dehydrogenase (Rdh10) for oxidation of retinol to retinaldehyde, and retinaldehyde dehydrogenases Raldh1 (Aldh1a1), Raldh2 (Aldh1a2), and Raldh3 (Aldh1a3) for oxidation of retinaldehyde to RA, which then functions as a ligand for nuclear RA receptors [18]. A role for RA signaling during mouse striatal development is evident after E12.5 when Raldh3, expressed in the subventricular zone of the LGE [19], plays a required role in the up-regulation of dopamine receptor D2 expression [20]. Consistent with this finding, loss of RA receptor-beta (RARβ) in null mutant mice is associated with defects in striatal dopaminergic neurogenesis after E13.5 resulting in motor behavioral defects [21]. A recent study using Rdh10 mutant embryos with reduced RA synthesis in the meninges suggested that RA is required for normal radial expansion of the dorsal cortex [22]. However, other studies have suggested that RA may not act in the embryonic cortex, as RA activity was detected in the LGE but not the cortex [23].
Here, we employ null mutants for Raldh3 (Aldh1a3) and Raldh2 (Aldh1a2) to ascertain the anatomical sites, cellular targets, and consequences of RA signaling in the embryonic forebrain. Our results provide evidence that Raldh3 expression in the LGE is a major source of RA production in the embryonic forebrain and is required for GABAergic differentiation from LGE-derived progenitors in the basal ganglia. Furthermore, our findings suggest that RA generated in the meninges by Raldh2 is not required to stimulate radial expansion of the cortex as previously suggested. We also report that RA induces GABAergic differentiation in neurons generated from LGE-derived neurospheres and human embryonic stem cells, thus implicating a role for RA as a GABAergic differentiation factor both in vivo and in vitro.
Although Raldh3 expression in the LGE from E12.5 to early postnatal stages suggests the LGE is a major site of RA action in the embryonic forebrain [19],[24], Raldh2 and Rdh10 are expressed in the meninges beginning at E12.5–E13.5, suggesting that RA synthesized there may regulate corticogenesis [22],[25]. In order to better define the timing and location of RA signaling in the developing forebrain from E12.5–E14.5, we employed a tissue explant RA reporter cell assay [26]. Cortex and LGE tissues were dissected from E12.5 to E14.5 embryos and grown as explants in co-culture with the RA-reporter cells. As positive controls, eye (E12.5 to E13.5), which expresses Raldh1 and Raldh3, and meninges (E14.5), which expresses Raldh2, were dissected from the same embryos. Reporter cells co-cultured with cortex or LGE from E12.5 wild-type embryos displayed no RA activity, whereas eye explants did (Figure 1A–C); lack of RA activity in E12.5 LGE explants may be due to low initial Raldh3 expression. In accordance with the increase in Raldh3 expression in the LGE after E12.5 [19], LGE explants from both E13.5 and E14.5 induced strong RA activity in the surrounding reporter cells (Figure 1F,I). In contrast, E13.5 and E14.5 cortical explants remained unable to induce RA activity, whereas meninges and eye explants at these stages exhibited RA activity (Figure 1D,E,G,H). To verify that RA activity detected in the LGE is due to Raldh3 expression, we found that loss of RA synthesis by Raldh3 resulted in lack of RA activity in Raldh3−/− LGE explants but had no effect on RA activity in meninges (Figure 1M–O). Using an Raldh2−/− mutant model we found that all RA activity detectable in wild-type meninges at E14.5 was eliminated in Raldh2−/− meninges (Figure 1G,J). Raldh2−/− cortical explants contained no RA activity as observed in wild-type (Figure 1H,K), while RA activity was still observed in Raldh2−/− LGE (Figure 1I,L). Together, the above findings demonstrate that RA produced by Raldh3 in the LGE can activate transcription in the basal ganglia, whereas RA produced by Raldh2 in the meninges does not activate transcription in the adjacent cortex.
Exposure of the reporter cells to a range of RA concentrations between 1 nM and 1 µM provided a dose-response for RA activity (Figure S1); a concentration of 1 nM was sufficient to activate the reporter line as previously reported [26]. Thus, the fact that RA activity was not detected in cortical explants from E12.5 to E14.5 indicates that RA is present at very low levels in the cortex. However, recent studies proposed a role for RA in corticogenesis and additionally reported a value for the concentration of RA in the mouse E14.5 cortex (0.28 µmole/mg) [22], which is seven orders of magnitude higher than that previously reported for mouse E13.5 forebrain (12 pmol/g) [27]. The former value (presented as µmole/mg rather than pmol/g) is most likely in error as other studies reported RA concentrations in adult mouse cortex as 16 pmol/g and adult striatum as 78 pmol/g [28], but this leaves in doubt how much RA was actually detected in E14.5 cortex.
Our observation that RA activity during forebrain development is due primarily to Raldh3 expression in the LGE prompted us to investigate if neural precursors isolated from E14.5 LGE maintain their Raldh3 expression and RA activity when expanded in vitro under mitogen stimulation. Hence, we employed immunocytochemistry with a Raldh3 antibody together with the tissue explant RA bioassay using neurospheres generated from E14.5 LGE and cortex of wild-type and Raldh3−/− embryos. RA activity and Raldh3 immunostaining were detected in neurospheres derived from wild-type E14.5 LGE (Figure 2C–D). In contrast, both Raldh3 immunostaining and RA activity were eliminated in LGE-derived neurospheres from Raldh3−/− embryos (Figure 2A–B). Neither Raldh3 immunostaining nor RA activity were found in neurospheres derived from the cortex of either wild-type or Raldh3−/− embryos (Figure 2E–H). These results further confirm that Raldh3 expression in the LGE is responsible for RA synthesis and additionally showed that neurospheres expanded from LGE cells maintain their RA activity.
We investigated whether RA has an effect on the differentiation potential of regionally derived neurosphere cultures. Neurospheres from LGE and cortex of E14.5 wild-type and Raldh3−/− embryos were differentiated for 7 d and subsequently analyzed immunocytochemically with antibodies against the pan-neuronal marker β-tubulin-III (Tuj1), the GABA-synthesizing enzyme glutamic acid decarboxylase (Gad67), the neural progenitor marker nestin, and the astrocytic marker glial fibrillary acidic protein (GFAP). Many wild-type LGE neurospheres untreated with RA were found to co-express Tuj1 and Gad67, indicating they have differentiated and matured into a GABAergic phenotype (44.3%±13.0%), however very few Tuj1/Gad67-positive cells were detected in differentiated cultures of Raldh3−/− LGE neurospheres (13.2%±4.1%) (Figure 3A–B,K). Tuj1-expressing neurons that differentiated from cortical Raldh3−/− neurospheres appeared to have a similar morphology to those generated from cortical wild-type neurospheres, and Gad67 was never detected (Figure 3C–D). Nestin-positive progenitors and cells with an astrocytic morphology expressing GFAP appeared similar in LGE and cortical differentiated cultures derived from wild-type and Raldh3−/− neurospheres (Figure 3E–H).
LGE neurospheres from both wild-type and Raldh3−/− embryos were differentiated in the presence of RA in order to further test the effect of RA on GABAergic neuronal differentiation. After 1 wk of differentiation in the presence of 100 nM RA, the majority of the generated neurons in Raldh3−/− and wild-type cultures were GABAergic, as observed by double staining for Tuj1 and Gad67 (Figure 3I–J). However, when cortical neurospheres were differentiated in the presence of RA, Gad67 was never detected in either wild-type or Raldh3−/− differentiating cultures (Figure 3K–L). Quantification of GABAergic neuron differentiation from wild-type and Raldh3−/− LGE neurospheres with or without added RA showed a significant increase of Gad67-positive neurons in the presence of RA. The proportion of GABAergic cells derived from Raldh3−/− LGE neurospheres increased significantly from 13.2%±4.1% under control conditions to 83.37%±11.75% when the neurospheres had been differentiated in 100 nM RA (Figure 3M). The proportion of GABAergic neurons in cultures of wild-type LGE neurospheres increased from 44.3%±13.0% under control conditions to 76.45%±11.75% after treatment with 100 nM RA (Figure 3M). Thus, E14.5 LGE derived cells from Raldh3−/− embryos can be expanded as neurospheres and are able to differentiate into neurons and glia, but they are unable to differentiate into GABAergic neurons unless RA is added.
The observation that RA generated by Raldh3 induces GABAergic differentiation of neural precursors in vitro prompted us to investigate if RA signaling is required for GABAergic differentiation in the developing forebrain. We examined a panel of markers for both neural progenitors and differentiated neurons in forebrains from E14.5 wild-type and Raldh3−/− embryos. Raldh3 protein was observed at high levels in the SVZ of the LGE; Raldh3 was not detected in the Raldh3−/− forebrain (Figure 4A–B). Nestin and RC2 immunoreactivity was not significantly changed in Raldh3−/− versus wild-type basal ganglia at E14.5, suggesting that generation of neural progenitors is not affected when RA signaling is disrupted (Figure 4C–F). In order to determine the proliferative capacity of these neural progenitors, double immunohistochemistry was performed with the proliferation marker Ki67 and the radial glia/progenitor marker nestin (Figure S2A–B). We observed no reduction in the number of LGE proliferating progenitors in Raldh3−/− embryos compared to control embryos, showing that both generation and proliferation of neural progenitors is not affected when RA signaling is disrupted (Figure S2C).
MAP2 immunostaining marking postmitotic neurons [29] was unaffected in Raldh3−/− basal ganglia (Figure 4G–H). The numbers of MAP2-expressing neurons were quantified in the striatum, cortex, and septum from wild-type and Raldh3−/− embryos. MAP2-expressing cells did not change in number in these regions of the forebrain in Raldh3−/− embryos, confirming that neurogenesis was not affected (Figure S2D).
A defect in GABAergic differentiation was observed when RA signaling is lost in the basal ganglia. Gad67-positive cells are normally present along the SVZ from the LGE to the septum (Figure 4K), and the pattern of GABA immunoreactivity is normally similar to Gad67 although GABA-positive cells extend into the ventricular zone (Figure 4I); this is probably due to the fact that while Gad67 immunoreactivity marks only cells with GABA production (i.e. GABAergic cells), GABA immunoreactivity could mark cells synthesizing GABA plus cells that uptake GABA released by Gad67-positive cells. In Raldh3−/− embryos, detection of both Gad67 and GABA was nearly eliminated in the LGE and septum (Figure 4J,L). At E12.5, when Raldh3 expression in the LGE has just initiated, we observed that GABA was detected in the MGE and LGE in a pattern that was not significantly different between wild-type and Raldh3−/− forebrain; GABA detection in the LGE at E12.5 was at a lower level than that seen at E14.5 (Figure S3A–B). As RA activity is not yet detected in the LGE at E12.5 but is seen by E13.5 (Figure 1C,F), these findings demonstrate that RA signaling initiating after E12.5 is required to stimulate the high level of GABAergic differentiation normally observed in the LGE by E14.5.
The cellular source of RA in the LGE has previously been associated with newly born neurons in the subventricular zone expressing Raldh3 [24]. We analyzed Raldh3 immunoreactivity in two distinct types of cells, radial glia and postmitotic neurons of the LGE at E14.5. Double-labeling studies demonstrated that none of the RC2-positive radial glia exhibit Raldh3 detection, although the radial processes of these cells were observed next to Raldh3-positive cells localized in the SVZ that did not possess radial processes (Figure S4A–C). In contrast, most Raldh3-expressing cells were also labeled with neuronal marker MAP2 (Figure S4D–F). The above results provide further evidence that Raldh3-expressing cells in the LGE are newly born neurons defining a discrete region of the SVZ.
In E18.5 wild-type embryos, Raldh3 detection remains strongest along the SVZ of the LGE (particularly high in the dorsal LGE) with weaker detection further ventrally along the septum; Raldh3 immunoreactivity was eliminated in Raldh3−/− forebrain (Figure 5A–B). As observed at E14.5, MAP2 immunostaining was unaffected in Raldh3−/− versus wild-type forebrain at 18.5 (Figure 5G–H). At E18.5, GABAergic differentiation in the striatum was nearly eliminated in Raldh3−/− forebrain as monitored by Gad67 immunoreactivity (Figure 5C–D); GABA detection was reduced in striatum but less so than Gad67 possibly due to diffusion of GABA still generated ventral of the striatum (Figure 5E–F). Thus, not all regions of the basal ganglia were affected by the disruption of RA signaling, as both Gad67 and GABA immunoreactivity appear relatively normal in the pallidum and septum at E18.5 (Figure 5C–F).
The subcortical telencephalon is known to be the source of GABAergic projection neurons that migrate radially from the ventricular progenitor zone to reach their final destination. The LGE gives rise to GABAergic striatal projection neurons [1],[30]–[35], while the MGE gives rise to GABAergic projection neurons of the pallidum, septum, and nucleus basalis [1],[30]. In order to determine if differentiation of striatal projection neurons is affected in E18.5 Raldh3−/− embryos, we examined Foxp1, a marker for these neurons [36], and found that Raldh3−/− embryos display normal Foxp1 immunoreactivity in the striatum (Figure 5I–J). This observation supports our previous findings demonstrating that Raldh3 is not required to generate DARPP32-positive neurons, another marker of striatal medium-sized spiny projection neurons [20]. Instead, our findings demonstrate that RA is required for striatal projection neurons to acquire a GABAergic fate.
Our results indicate that RA is required to stimulate GABA synthesis in LGE-derived progenitors. To characterize other aspects of the GABAergic phenotype in Raldh3−/− mutants, we analyzed expression of the vesicular GABA transporter (VGAT, Viaat), a transporter that mediates accumulation of GABA into the synaptic vesicles before exocytotic release to the synaptic cleft [37]. Expression of VGAT appeared normal in Raldh3−/− forebrain (Figure S5A–B), indicating that RA is not required for this aspect of GABAergic differentiation.
Next, we wanted to investigate whether disruption of RA signaling could induce defects in the specification of other neuronal populations. Previous studies have shown that loss of Raldh3 or RARβ in the striatum results in down-regulation of dopamine receptor D2 in the nucleus accumbens [20],[21]. However, in Raldh3−/− forebrain no difference was found in the expression of tyrosine hydroxylase (TH), a marker of dopaminergic neurons (Figure S5C–D). Also, we observed no difference in vesicular glutamate transporter (VGLUT), a marker of glutamatergic neurons (Figure S5E–F).
In addition to neuronal markers, we examined whether loss of RA affects glia. We showed above that radial glia differentiation is not affected by loss of Raldh3 (Figure 4E–F). We also analyzed astrocyte differentiation by analyzing expression of the astrocytic marker GFAP. We did not detect any difference in GFAP immunoreactivity in E18.5 Raldh3−/− forebrain compared to wild-type controls (Figure 5K–L). Our findings thus suggest that RA is not required for gliogenesis or generation of radial projection neurons.
Taken together, our observations at E12.5–E18.5 demonstrate that RA is required to stimulate a high level of GABAergic differentiation first in the LGE and then later in the striatum but that a Raldh3-independent mechanism for GABAergic differentiation occurs in the MGE/pallidum and septum.
In addition to GABAergic projection neurons, progenitor cells in the LGE produce GABAergic interneurons that migrate tangentially mostly within the cortical intermediate zone, whereas GABAergic interneurons migrating from the MGE disperse into the cortical plate [5],[38]–[40]. Also, cells derived from the dorsal SVZ of the LGE generate many olfactory bulb interneurons via a rostral migratory pathway [40]–[42]. At E18.5, Gad67 immunoreactivity normally extends from the striatum into the intermediate zone of the cortex marking a population of LGE-derived interneurons, but this zone of Gad67 detection was markedly reduced in Raldh3−/− cortex (Figure 5C–D). Dlx2 is an early marker of GABAergic progenitors in the basal ganglia that is required for GABAergic interneuron migration to the cortex [14]. Dlx2 immunoreactivity was not changed in Raldh3−/− forebrain from E12.5–E18.5, demonstrating that interneurons are generated in the basal ganglia and migrate to the cortex in the absence of RA signaling (Figure S3C–H). Additionally, detection of Gad67 in the Raldh3−/− olfactory bulb was also clearly reduced compared to wild-type (Figure 5M–N). Apart from the dorsal LGE, recent studies have shown that additional telencephalic areas may also contribute to olfactory bulb interneurons [43]–[45], which may explain our observed partial elimination of Gad67 immunoreactivity in the Raldh3−/− olfactory bulb. The above findings provide evidence that RA synthesis controlled by Raldh3 is required for GABAergic differentiation of interneurons that originate from progenitors in the LGE then migrate tangentially to the cortex and olfactory bulb.
We investigated whether our findings may be useful to generate GABAergic neurons from human embryonic stem (ES) cells for potential cell replacement therapies. Following RA treatment of embryoid bodies and propagation of neural rosettes, cultures were processed immunocytochemically with antibodies against Pax6 (a marker for neural progenitors), Doublecortin (DCX; a marker for immature migrating neurons), the pan-neuronal marker Tuj1, plus GABA and Gad67. With no RA added in the differentiation medium a large proportion of the cells were Pax6-positive (79.9%±5.01%), indicating they were neural progenitors (Figure 6J,M), and many cells exhibiting neuronal processes were Tuj1-positive (32.8%±10.3%) and colocalized with DCX (31.3%±8.6%), suggesting they were immature neurons (Figure 6G,M). However, very few Gad67 (2.7%±0.6%) positive cells were detected in cultures with no RA added, suggesting that GABAergic differentiation is not favored under these conditions (Figure 6A,D). Treatment of embryoid bodies with 1 µM RA resulted in a significant increase in both the number of Tuji1/DCX+ neurons that were forming extensive neuronal networks and GABAergic neurons detected with Gad67 (16.2%±2.1%) (Figure 6B,E,M). Moreover, the proportion of Pax6-positive cells was reduced to almost half with 1 µM RA (49.9%±11.6%), suggesting that more progenitor cells had differentiated to immature migrating neurons co-expressing Tuj1 and DCX (Figure 6H,K,M). Addition of 10 µM RA further increased the percentage of GABAergic neurons positive for Gad67 (41.9%±10.7%) (Figure 6C,F,M) and further decreased the number of Pax6-positive progenitors (17.9%±4.9%) (Figure 6L,M). Under these differentiation conditions, RA was able to drive GABAergic differentiation in almost half of the cells generated (Figure 6M), suggesting that RA treatment is quite useful for induction of GABAergic differentiation in vitro.
To gain further insight on the subtype identity of the RA-induced GABAergic neurons generated in our cultures, we examined expression of region-specific transcription factors previously associated with the specification of both GABAergic interneurons and projection neurons. Many Tuj1-positive cells were also immunopositive for Dlx2, which is expressed in GABAergic precursors in the basal ganglia of the telencephalon, differentiating into both interneurons and striatal projection neurons (Figure S6A). However, no Tuj1-positive cells were found to be positive for Foxp1, a striatal projection neuron marker (Figure S6B). Lim1/2 is a LIM homeodomain protein marking interneurons of the diencephalon and spinal cord [17],[46]–[48]. Many GABA-positive neurons also expressed Lim1/2, suggesting they acquire a GABAergic interneuron phenotype (Figure S6C). Islet1 (Isl1), another LIM homeodomain protein, is expressed in the ventral forebrain where it marks telencephalic GABAergic projection neurons; Isl1 also marks diencephalic interneurons when co-expressed with Lim1/2 and GABA [46],[48]. No cells in our culture co-expressed GABA and Isl1, further suggesting that GABA-positive neurons in our cultures do not acquire striatal projection neuron or diencephalic interneuron identities (Figure S6D). We found that 43.7%±11.94% of GABA-positive cells expressed Lim1/2 and 37.5%±11.5% expressed Dlx2 (Figure S6E). Together, these data show that our human ES cell differentiation protocol induces a heterogeneous population of GABA-positive interneurons acquiring either telencephalic or spinal cord identities, but that it does not favor generation of GABA-positive striatal projection neurons.
Previous studies using an Rdh10 ethylnitrosourea (ENU) mutant, which reduces retinaldehyde needed for Raldh2 to catalyze RA synthesis in the meninges, suggested that this source of RA is required for radial expansion of the cortex; a reduction in radial expansion of the cortical postmitotic neuronal layer was proposed to result in a concomitant lateral increase in the proliferative progenitor population in the ventricular zone [22]. Raldh2−/− embryos, which completely lack meninges RA activity (Figure 1G,J), present an excellent model to examine whether RA is required for corticogenesis since Raldh2 catalyzes the final step of RA synthesis in the meninges. At E14.5, the head region of Raldh2−/− embryos appeared to have developed relatively normally while they invariably displayed stunted forelimbs (Figure 7A–B), which we have previously shown is due to a lack of RA synthesis by Raldh2 in trunk mesoderm [49],[50]. Thus, Raldh2−/− mutants do not exhibit massive head deformities like those reported for Rdh10 mutants [51]. We analyzed expression of Tuj1 and the proliferative marker Ki67 in coronal brain sections of both wild-type and Raldh2−/− embryos at E14.5. The medial-lateral length of the ventricular zone in the dorsal forebrain of the Raldh2−/− mutant appeared similar to that of the wild-type embryo (Figure 7C–D). Moreover, double immunostaining for Tuj1 and Ki67 revealed no changes in radial expansion of the postmitotic Tuj1-expressing cortical layer nor the Ki67 proliferative zone in the Raldh2−/− cortex when compared to wild-type (Figure 7E–J). Examination of MAP2, another marker for postmitotic neurons, also demonstrated no difference in medial-lateral length for the mutant dorsal ventricular zone (Figure S7A–B) and no difference in radial width for the mutant cortex (Figure S7C–D). Finally, examination of RC2, a marker of radial glia whose somata reside in the ventricular zone of the cortex and whose radial processes span the entire distance to the pial surface, showed no difference between the Raldh2−/− and wild-type cortex (Figure S7E–F). The fact that Raldh2−/− embryos retain a normal ratio of cortical progenitors to postmitotic neurons with no apparent morphological defects, in conjunction with a complete lack of RA activity in mutant meninges and cortical explants, suggests that RA is not required for embryonic corticogenesis. The contradiction between our results with Raldh2−/− embryos and the results of others with Rdh10 mutants [22] may be explained by the observation that Rdh10 mutants, unlike Raldh2−/− embryos, exhibit severe craniofacial defects that distort the cranium and forebrain possibly resulting in a thinner cortex [51].
Previous studies have shown that RARα and RARβ are expressed during mouse forebrain development, whereas RARγ is undetectable [52],[53]; RARα was reported to be widespread in the embryonic forebrain, while RARβ was detected primarily in the striatum and is induced by RA. We examined expression of RARα and RARβ in E18.5 wild-type forebrains by in situ hybridization. RARα mRNA was widespread in the E18.5 forebrain including both the striatum and cortex, but expression was low or undetectable in the ventricular zone (Figure S8A). RARβ mRNA was detected in the striatum but not in the cortex or ventricular zone (Figure S8B). Taken together with our observation that the LGE/striatum is a major localized site of RA synthesis during forebrain development due to Raldh3 expression, overlapping expression of both RARα and RARβ in the striatum further suggests that this is a major site of local RA-mediated induction and signaling.
In this study we demonstrate a novel requirement of RA generated by Raldh3 for GABAergic differentiation in the basal ganglia. In contrast, RA activity is not detected in the cortex at any stages examined despite detection of RA activity in the adjacent meninges, which requires Raldh2. Even if the cortex does receive a small amount of RA from the meninges that we cannot detect, our findings with Raldh2−/− embryos lacking RA activity in the meninges demonstrate that this source of RA is unnecessary for cortical expansion as suggested by a recent study [22]. Thus, unlike the cortex, the LGE represents an unambiguous site of RA action during forebrain development, and loss of RA in the LGE results in a loss of GABAergic differentiation.
Our studies revealed that at E12.5, when Raldh3 expression is barely detectable in the LGE and RA activity is not yet detected, Raldh3−/− embryos maintain expression of the regulatory gene Dlx2 and early aspects of GABAergic differentiation in the progenitor domains of the basal ganglia. However, by E14.5, when Raldh3 expression has intensified in the LGE and RA activity is easily detectable, RA generated by Raldh3 is required to stimulate GABAergic differentiation in the LGE by inducing Gad67 needed for GABA synthesis. At E18.5, Raldh3 is required to maintain GABAergic differentiation in the LGE, whereas a Raldh3-independent mechanism controls GABA synthesis in the MGE and septum. This observation suggests that the LGE is the main site of RA action along the SVZ. We observed that Raldh3 expression in newly generated neurons at the border of the proliferative and postmitotic zones in the LGE coincides with a region that generates both GABAergic striatal projection neurons and GABAergic interneurons [5],[31]–[34],[40]. Our studies in Raldh3−/− embryos revealed that RA signaling stimulates a GABAergic phenotype in LGE-derived interneurons migrating to the olfactory bulb and cortex and that RA is required for Foxp1-positive striatal projection neurons to further differentiate to a GABAergic fate. As Dlx2 and VGAT were still expressed normally in the absence of RA, the role of RA in GABAergic differentiation may be limited to stimulation of Gad67 activity in the LGE to promote GABA synthesis.
As the appearance of GABAergic interneurons in the olfactory bulb and cortex is reduced rather than eliminated, our findings suggest that interneurons can still migrate to these locations in the absence of RA but that less interneurons have matured to a GABAergic phenotype. Other studies have shown that Gsx2 (Gsh2), a homeobox gene specifying ventral character in the forebrain, is required for Raldh3 expression in the LGE [54] and that Gsx2 is required for specification of GABAergic interneurons that migrate from the LGE to the olfactory bulb [13]. In addition, differentiation of DARPP-32 striatal projection neurons is greatly reduced in Gsx2 null embryos but not in conditional Gsx2 mutants when Gsx2 is progressively inactivated from E10.5–E18.5 [13] and also not in Raldh3−/− mutants [20]. Thus, early expression of Gsx2 is required for correct DARPP-32 striatal projection neuron development, a time when there is no Raldh3 expression in the forebrain. Taking into consideration the above, one can conclude that RA signaling exerts a specific role in specifying the GABAergic phenotype both for production of GABAergic interneurons and for further differentiation of striatal projection neurons to a GABAergic fate. Examination of the Gad67 promoter proximal region revealed no evidence of a canonical RA response element (unpublished data), suggesting that Gad67 may be an indirect target or may be controlled post-transcriptionally by RA signaling in the basal ganglia during GABAergic differentiation. As it is clear that RARα and RARβ are both expressed in the basal ganglia, null mutants or antagonists for these RA receptors may be useful to further examine the mechanism through which RA functions during stimulation of GABAergic differentiation. Further, as we show that endogenous RA signaling is preserved in primary LGE neurosphere cultures and is required to generate GABAergic neurons in vitro, such cells may prove useful in studying the mechanism of RA action during GABAergic differentiation.
A previous study suggested that Foxc1 mutants fail to form a complete forebrain meninges and exhibit increased lateral expansion of the cortical ventricular zone and reduced neurogenic radial expansion due to the loss of RA produced by Rdh10 and Raldh2 in the meninges [22]. The major conclusions of that study were drawn by comparison of the cortical phenotype of the Foxc1 mutants with that of an Rdh10 ENU mutant [22],[55]. However, our studies on Raldh2−/− embryos lacking RA activity in the meninges demonstrate that RA is not required for radial expansion of the embryonic cortex. Additionally, RA receptors were not detected in the ventricular zone of the developing cortex, where RA was proposed to be required to induce neurogenic division of cortical progenitors. Together, these findings suggest that the dorsal forebrain phenotype in Foxc1 mutants is RA-independent. The Rdh10 ENU mutant employed for those forebrain studies [22] exhibits a very similar phenotype to another published Rdh10 ENU mutant, which has severe neural crest-derived craniofacial defects that are responsible for distortion of the cranium as well as forebrain [51]. Thus, reduced radial expansion of the cortex and increased lateral expansion of the ventricular progenitor zone reported for Rdh10 mutants [22] may not be due to a specific effect of RA on corticogenesis but rather a defect in cranial neural crest migration and differentiation that leads to the altered cortical morphology. Indeed, Rdh10 mutants lack all RA activity in the head during the time when cranial neural crest is migrating due to loss of all retinaldehyde synthesis [51], whereas Raldh2−/− embryos still retain most cranial RA synthesis during this time due to expression of Raldh1 and Raldh3 in ocular and olfactory tissues [56]. Thus, head development in Raldh2−/− embryos is not grossly altered, allowing us to conclude that a lack of cranial RA activity specifically in the meninges does not lead to a defect in radial expansion of the cortex. Furthermore, Raldh2 is not expressed in the dorsal meninges until E12.5 [22] or E13.5 [25], while the lengthening of the dorsal forebrain in Foxc1 mutants is already evident at E12.5 [22]. Based on the above, it seems unlikely that RA produced and secreted in the dorsal meninges could be the neurogenic factor inducing the switch from symmetric to asymmetric division in the ventricular zone to affect embryonic cortical expansion. Alternatively, RA generated in the meninges by Rdh10 and Raldh2 might have another function. RA could diffuse in the opposite direction and control development of the skull, which is populated by cranial neural crest cells. Interestingly, a recent study showed that ablation of all three RA receptors (RAR alpha, beta, and gamma) in cranial neural crest cells results in agenesis or malformations of most of the craniofacial skeletal elements including the frontal and parietal bones, which are adjacent to the dorsal meninges [57]. Additionally, Foxc1 hypomorphic mutants also exhibit malformation of the frontal bone [58], providing further evidence that RA generated in the meninges downstream of Foxc1 may function in cranial neural crest differentiation.
RA treatment is known to facilitate terminal differentiation of neural progenitors derived from ES cells [48],[59]–[63]. Here we demonstrated that exposure of human ES-derived embryoid bodies to high concentrations of RA promotes differentiation of neuronal precursors to a high percentage of immature GABAergic neurons. Interestingly, although a low endogenous concentration of RA is sufficient to stimulate GABAergic differentiation of cells in the LGE at E14.5, a high concentration of RA is needed for GABAergic differentiation in embryoid bodies derived from ES cells. This may be due to the much more primitive nature of cells in an embryoid body (similar to cells in an early gastrula) compared to neuroepithelial cells of the late embryonic forebrain. As RA binds directly to DNA-bound nuclear receptors that interact with co-repressors and co-activators, we suggest that high concentrations of RA may exert tremendous epigenetic effects on embryoid body cells, driving them to both a neuronal and GABAergic fate. In addition, our RA treatment protocol generated GABAergic neurons exhibiting expression of interneuron transcription factors of either anterior (forebrain) or posterior (spinal cord) identity, but not striatal projection neuron identity. A previous study proposed that mouse ES cells differentiating in medium without RA acquired a GABAergic identity of ventral forebrain co-expressing Gad67 and Isl1 (most likely striatal projection neurons), while exposure to RA resulted in acquisition of a spinal cord interneuron identity [48]; those studies differed from ours in that RA treatment occurred at a later window during embryoid body formation and a lower concentration of RA was used. Thus, differences in the effects of RA on GABAergic interneuron identity in various culture systems may be dependent upon the timing and concentration of RA used.
Production of the inhibitory neurotransmitter GABA in the central nervous system depends on local neurons, and disturbed GABAergic neuron function has been associated with numerous neurological disorders including Huntington's disease, autism, schizophrenia, bipolar depression, and epilepsy [9]–[12]. GABAergic interneurons are a particularly attractive cell population for cell-based therapies of these disorders due to their ability to migrate, differentiate, and function following transplantation [64]–[66]. GABAergic interneuron precursors derived from mouse ES cells were shown to migrate, survive for several months, and exhibit neurochemical and electrophysiological characteristics of mature interneurons when transplanted into postnatal cortex [67]. Additionally, transplantation of GABAergic interneuron precursors reduced the number of seizures in epileptic mice [68]. Thus, generation of GABAergic interneurons from RA-treated human ES cells as we report here coupled with isolation of cells with forebrain character may provide useful candidate cells in cell replacement therapies for one or more of these neurological conditions.
Raldh3−/− embryos exhibiting postnatal lethality just after birth were previously described [20]. Raldh2−/− embryos exhibiting midgestation lethality have been described previously [69]. To prevent Raldh2−/− early lethality, the maternal diet was supplemented for a short time with a very low dose of RA as described previously [50]. Briefly, 0.1 mg of all-trans-RA (Sigma Chemical Co.) was added per gram of standard mouse chow and provided fresh to pregnant females at E6.75–E8.75. At E9.25 mice were returned to standard chow until embryos were collected at E14.5; Raldh2−/− mutants obtained with this method invariably exhibit stunted forelimbs and interdigital defects, demonstrating a loss of RA function in regions where Raldh2 is responsible for RA synthesis [50]. Dietary supplementation with this dose of RA is indeed low as HPLC measurements have shown that it provides less RA to embryos than Raldh2 normally generates [70]. Administered RA is cleared within 12–24 h after treatment ends [69], thus allowing one to examine embryos at E10.5–E14.5 that now lack RA activity normally generated by Raldh2. Embryos were genotyped by PCR analysis of yolk sac DNA and were staged by designating noon on the day of the vaginal plug as E0.5. All mouse studies conformed to the regulatory standards adopted by the Animal Research Committee at the Sanford-Burnham Medical Research Institute.
Lateral ganglionic eminence (LGE) and cortex were dissected from E14.5 wild-type and Raldh3−/− mutant embryos and incubated in DMEM containing 0.1% trypsin and 0.05% DNase for 15 min at 37°C followed by mechanical dissociation. The cells were spun down and resuspended at a concentration of 100,000 cells/ml in basic medium DMEM-F12 supplemented with B27, 10 ng/ml basic fibroblast growth (bFGF) factor, and 20 ng/ml epidermal growth factor (EGF). No apparent differences in growth rate or appearance were observed for wild-type compared to Raldh3−/− neurosphere cultures. Neurospheres were differentiated by culturing on plates pre-coated with poly-L-ornithine. EGF and bFGF were removed from the expansion medium and 1% serum was added (differentiation medium). The spheres were maintained under differentiation conditions for 7 d in the presence or absence of 100 nM RA before fixation.
Tissue explants or neurospheres from wild-type and mutant embryos were cultured overnight on Sil-15 F9-RARE-lacZ RA reporter cells followed by detection of β-galactosidase activity as previously described [26].
E12.5–E14.5 heads and E18.5 brains were fixed overnight at 4°C in 4% paraformaldehyde and paraffin sections (7 µm) were processed immunohistochemically as described [71]. The primary antibodies included rabbit anti-Raldh3 1∶50 [72], mouse anti-nestin 1∶100 (Millipore), mouse anti-RC2 1∶100 (Developmental Studies Hybridoma Bank at University of Iowa; DSHB), mouse anti-MAP2 1∶200 (Sigma; M4403), rabbit anti-GABA 1∶500 (Millipore), mouse anti-Gad67 1∶50 (Millipore; MAB5406), rabbit anti-GFAP 1∶1000 (Dako), rabbit anti-Foxp1 1∶100 (Abcam), rabbit anti-Ki67 1∶200 (Abcam), rabbit anti-Dlx2 1∶100 (Abcam), mouse anti-VGAT 1∶200 (Synaptic Systems), rabbit anti-VGLUT (Synaptic Systems), and mouse anti-TH (Sigma). In situ hybridization of E18.5 brain sections was performed as described [73] using RARα and RARβ riboprobes.
Human ES cells (line H9) were cultured and passaged weekly on a feeder of irradiated embryonic mouse fibroblasts as described previously [74]. The protocol for RA-induced GABAergic differentiation was based on previous methods [75]. Briefly, human ES cell-derived embryoid bodies (EBs) were cultured in EB growth medium in non-adherent plates for 3 d, followed by RA treatment for 3 d (RAd3). RA was removed and the RAd3 EBs were plated on culture dishes pre-coated with poly-L-ornithine and fibronectin and cultured for an addition 4 d (RAd7) in serum-free neuronal induction medium, comprised of neurobasal medium supplemented with B27, bFGF, and EGF. At RAd7 neuroepithelial rosettes were isolated mechanically from the differentiation cultures with a 2 ml serological pipette. After isolation, rosettes were replated in the same neuronal induction medium for an additional 14 d (RAd21) before fixation.
Immunocytochemistry on neurospheres and ES cells was carried out as described [75]. Primary antibodies used included rabbit anti-GABA 1∶1,000 (Millipore), mouse anti-Gad67 1∶200 (Millipore; MAB5406), rabbit anti-GFAP 1∶1,000 (Dako), mouse anti-Tuj1 1∶1,000 (Covance), and guinea pig anti-DCX 1∶500 (Millipore), rabbit anti-Dlx2 1∶200 (Abcam, AB18188), rabbit anti-Foxp1 1∶100 (Abcam), mouse anti-Islet-1 1∶100 (40.2D6) (DSHB), and mouse anti-Lim1/2(4F2) 1∶200 (DSHB).
Immunopositive cells and total DAPI-stained nuclei were counted to calculate the percentage of immunopositive cells. Five randomly picked areas from three independent experiments were counted for each marker. Data were presented as mean ± SEM; for pair-wise analysis of treatment conditions and/or genotypes, an ANOVA test was used.
|
10.1371/journal.pcbi.1004680 | Topological Phenotypes Constitute a New Dimension in the Phenotypic Space of Leaf Venation Networks | The leaves of angiosperms contain highly complex venation networks consisting of recursively nested, hierarchically organized loops. We describe a new phenotypic trait of reticulate vascular networks based on the topology of the nested loops. This phenotypic trait encodes information orthogonal to widely used geometric phenotypic traits, and thus constitutes a new dimension in the leaf venation phenotypic space. We apply our metric to a database of 186 leaves and leaflets representing 137 species, predominantly from the Burseraceae family, revealing diverse topological network traits even within this single family. We show that topological information significantly improves identification of leaves from fragments by calculating a “leaf venation fingerprint” from topology and geometry. Further, we present a phenomenological model suggesting that the topological traits can be explained by noise effects unique to specimen during development of each leaf which leave their imprint on the final network. This work opens the path to new quantitative identification techniques for leaves which go beyond simple geometric traits such as vein density and is directly applicable to other planar or sub-planar networks such as blood vessels in the brain.
| Planar reticular networks are ubiquitous in nature and engineering, formed for instance by the arterial vasculature in the mammalian neocortex, urban street grids or the vascular network of plant leaves. We use a topological metric to characterize the way loops are nested in such networks and analyze a large database of 186 leaves and leaflets, revealing for the first time that the nesting of the networks’ cycles constitutes a distinct phenotypic trait orthogonal to previously used geometric features. Furthermore, we demonstrate that the information contained in the leaf topology can significantly improve specimen identification from fragments, and provide an empirical growth model that can explain much of the observed data. Our work can improve understanding of the functional significance of the various leaf vein architectures and their correlation with the environment. It can pave the way for similar analyses in diverse areas of research involving reticulate networks.
| The angiosperm leaf vein network fulfills the combined requirements of efficient liquid transport within the leaf and high robustness against load fluctuations and damage, while at the same time providing structural reinforcement [1–4].
Modern leaf vein networks evolved gradually from simple dendritic branching patterns by introduction of anastomoses [5, 6], leading to leaf vascular networks that are highly reticulate, exhibiting nested, hierarchically organized vein loops. The reticulate leaf vascular system is an example of evolutionary adaptation under various constraints [1, 7–10].
Despite some common trends, the diversity of vein morphology in dicotyledonous plants is striking (see for instance Fig 1a–1f). Current models of vascular development in the model species Arabidopsis thaliana predict several overlapping phases in which the leaf primordium at first mainly grows by cell division, then later by cell expansion [3, 11]. Lower order (major) veins are thought to be formed during the first phases, whereas minor veins are formed primarily during the later, leaving an imprint in the higher order vascular system of the leaf.
The morphology, anatomy, and correlations with climate of the lower order vascular architecture have been extensively studied [12, 13], and primary and secondary vein traits can be easily quantified [3]. Certain leaf traits such as vein density are closely linked to photosynthetic efficiency [14–16]. Links to climatic conditions and vegetation type have been proposed as well [5, 13, 17, 18].
The hydraulic resistance of the whole plant is strongly affected by the leaf hydraulic resistance. The smallest veins, by virtue of their combined length and small hydraulic diameter are responsible for the bulk of this resistance. At the same time, the smallest veins, and in particular the small free-ending veinlets, are perhaps the most crucial for water delivery [19]. However, the architecture of higher order vein reticulation has been largely ignored in the literature. Other than an extensive descriptive nomenclature [20] and mainly qualitative measures [21], to this day there is no quantitative work that goes beyond obvious geometric characteristics, like minor vein density, areole size, angle distribution, vascular segment length and width distribution [3, 22, 23]. These characteristics by themselves are not sufficient to describe the full architecture, in particular the organization of the loops. Loops typically show a large degree of hierarchical nesting, i.e. larger loops composed of larger-diameter veins contain many smaller loops with smaller vein diameter (see Fig 1e).
Although topological studies of spatial network architectures such as street networks are quite common [24], a detailed quantitative characterization of the topological properties related to reticulation has been elusive in the past, and only recently have researchers started to seriously attack the question [23, 25, 26].
We use ideas inspired by computational topology [27] to define a metric suitable to quantify the architecture of higher order venation of leaves. We apply our topological metric to a dataset of 186 leaves and leaflets, demonstrating that our characterization constitutes a new phenotypic trait in plant leaves and carries information complementary to previously used quantities. We then show that this information can be useful in the task of identifying leaves from fragments, significantly improving identification accuracy. We finally present a growth model that reproduces most of the observed variation in the topological traits.
Our results suggest that topological and geometric venation traits are approximately independent, and that the higher order venation topology is mainly controlled by a small set of parameters regulating noise during vein morphogenesis.
The topological venation traits we use can be employed in much broader contexts than leaves, being applicable to any (sub-)planar, anastomosing network such as blood vessels in the brain, liver or retina, foraging networks build by slime molds, lowland river networks, urban street networks or force chains in granular media, thereby possibly opening up an entire new line of research.
Our topological metric quantifies the hierarchical nesting of loops within the network as well as the topological lengths of tapered veins. The analysis follows an existing hierarchical decomposition algorithm [25, 26, 28], constructing from a weighted network a binary tree graph termed the nesting tree which contains information about nesting of loops. The algorithm is schematically shown in Fig 1g and discussed in the supplement.
We stress that the method depends not on exact measurements of vein diameters but only on relative order. Similarly, transformations which slightly alter node positions do not affect the outcome (see Fig 1h).
Once the binary nesting tree (see Fig 1g) has been obtained, its structure can be quantified. Here, for each node j in the nesting tree, we calculate the nesting ratio q j = s j r j[29], where rj ≥ sj are the numbers of leaf nodes in the right and left subtrees of node j. We then define the nesting number as a weighted average i = ∑j wj qj, where ∑j wj = 1. We employ an unweighted nesting number iu, with wj = 1, and a degree-weighted nesting number iw, with wj ∝ dj − 1 = rj + sj − 1, where dj is called subtree degree. A high value of iu,w qualitatively represents graphs that are highly nested such as those in the top row of Fig 1i.
The presence and extent of tapered veins is quantified as follows. Starting from some edge e, we find the next edge by taking the maximum width edge amongst all with smaller width than e. We count how many steps can be taken until no more edges with smaller width are adjacent, resulting in a topological length Le assigned to each edge in the network. The mean topological length L top = 1 N E ∑ e L e, where NE is the number of edges, characterizes tapered veins in a network. Fig 1i shows a qualitative representation of various example network topologies using mean topological length and nesting number.
Instead of using just the nesting number, we additionally calculate pairwise topological distances between networks as the two-sample Kolmogorov-Smirnov statistic DKS between the cumulative distributions of nesting ratios in order to quantify the statistical similarity between nested loop topologies. Other methods to quantify the degree of topological dissimilarity between binary trees representing biological systems have been proposed on the basis of a “tree edit distance” [30]. Despite promise, this distance suffers from being dominated by differences in the size of the compared trees. In its local form [31], it suffers from the opposite problem, quantifying only the similarity between the n most similar subtrees. In contrast, our method is designed to capture statistical similarities between nesting trees, making it more suitable for dissimilarly sized, noisy networks.
We show that the topological characteristics described above provide a new dimension in the phenotypic space of leaf venation morphology.
For this, we analyze a dataset consisting of 186 leaflets from various species primarily belonging to the Burseraceae family (see S1 Text and S1 Table). Although most of species are therefore closely related, their venation patterns show considerable diversity (see Fig 1a–1f), rendering them a good test set for our metrics The leaves were chemically cleared and stained to make their higher order venation network apparent [32], then scanned at high resolution (6400 dpi) and vectorized in-house (see S1 Text). Scanning whole leaves and digitizing at high resolution is computationally expensive but necessary for this work to accurately represent the statistics of the high order veins [33]. Publicly available databases of scanned specimens [34] contain mostly low resolution images.
From the vectorized data, we obtained for each leaf five local geometric quantities: vein density σ (total length of all veins/leaf area), mean distance between veins a, mean areole area A, areole density ρA, and average vein diameter weighted by length of venation between junctions d. The (un)weighted nesting number i(u)w was calculated from all subtrees of the nesting tree with degree d ≤ 256 in order to remove leaf size effects for the full networks; the mean topological length was calculated from the whole network. Together, these metrics form a “leaf venation fingerprint” encompassing local features of the network, that can be estimated from leaf segments alone if necessary. Fig 1a shows the complete dataset plotted in the space of unweighted nesting number and mean topological length. We plot the most abundant genera Protium (98 specimen in the dataset), Bursera (21 specimen), and Parkia (8 speciment) as different symbols. Although the dataset does not allow for firm conclusions at this taxonomic level, both Protium and Parkia appear to show a modest trend towards clustering around characteristic nesting numbers. We then employed Principal Component Analysis (see Fig 2b) and found that together, the first two principal components explain 81% (= 52% + 29%) of the total variance in the dataset. Component 1 can be interpreted as containing mostly metrics derived from geometry, whereas Component 2 contains mostly metrics from topology. Topological lengths contribute roughly equally to either. Even though small correlations between them exist, this reveals local geometrical and topological leaf traits as approximately orthogonal traits for the description of the phenotype of leaf venation (see S1 Text, also for further analysis of the data in terms of latent factors).
Pairs of leaves (see Fig 2a and Fig 2e and 2f) which are close according to the topological distance defined by the DKS metric applied to the nesting ratio statistics can possess similar “by eye” venation traits. In the samples in Fig 2e and 2f, cycle nestedness and vein thickness are traits that appear correlated. However, the topology of leaf venation constitutes a new phenotypic trait that provides information orthogonal to geometric traits.
Topological information significantly helps in identifying leaf samples to species, especially when only a segment of the leaf is available. We fragmented all leaf samples in silico into equally sized segments of ca. 1.2 × 1.2cm and calculated all venation traits for the individual pieces (see S2 Table). Here, we thresholded the nesting ratios at subtree degree d ≤ 128. We employed Linear Discriminant Analysis (LDA) [35] to classify the fragments based on specimen membership (see also S1 Text). We then calculated the the probability of correctly identifying a segment as belonging to one of the 186 leaves and leaflets (the accuracy, see Fig 2c). Using only geometrical degrees of freedom, we found a 10-fold cross-validated accuracy of 0.35 (95% CI: [0.31, 0.39]). Adding topology improves the accuracy to 0.54 (95% CI: [0.48, 0.60]). Additionally, for each pair of individual leaves in the dataset, the same procedure was applied to obtain a mean pairwise accuracy score (the probability of correctly identifying a fragment as belonging to one of two leaves.) Again, using topological traits significantly improved the summary result (see Fig 2d and S1 Text). The same classification was applied towards identification of segments to species, as opposed to samples, with quantitatively similar results (see S1 Text).
It must be noted that there can be considerable variance among leaf traits, even when comparing among specimen from a single plant—in particular between sun- and shade leaves [6, 36]— that should be taken into account if the information is available.
In order to explain the nesting ratio and topological length distributions measured in our dataset, we examine a developmental model for the formation of higher-order venation in which the interplay between strictly hierarchical loop genesis and random noise is the major factor affecting nestedness.
Empirically, during the expansion growth phase of the leaf lamina, high order vein loops grow and are subdivided by the appearance of new veins, subsequent vein orders appearing discretely one after the other [11, 37]. Our model intends to capture this phenomenological fact (see Fig 3a for an illustration). The model is compatible with models of vein morphogenesis that invoke either auxin canalization [38] or mechanical instabilities [39], or a combination. It is similar in spirit to that described in the supporting information of [39] or [40] but adds fine-grained control of stochasticity.
We stipulate that each leaf is subject to a species dependent characteristic amount of noise during development, resulting in unique characteristic statistics of minor venation patterns.
The model as a whole is controlled by four dimensionless parameters (see Methods section). In Fig 3b and 3c we show the distributions of normalized areole size, mean topological lengths and nesting ratios for the same two leaves as in Fig 2e and 2f. The real distributions can be explained well by tuning two of the parameters. Thus, noise during growth of cycles can explain the observed local hierarchical nesting characteristics.
It should be noted that different mechanisms may underlie the organization of low order veins. Indeed, both models [41] and empirical observations [42] have found strong links between low order vein structure and leaf shape that may be connected to the overall growth pattern and developmental constraints of the lamina [43].
The leaf vasculature is a complex reticulate network, and properly chosen and defined topological metrics can quantify and highlight aspects of the architecture that have been ignored until now. The topological metrics presented in this work provide a new, independent dimension in the phenotypic space of leaf venation, allowing for more precise characterization of leaf features and improved identification accuracy, including identification of fragments. The extensive nomenclature for characterization of the vascular morphology [20] offers a discrete set of attributes that is mathematically insufficient to properly quantify a continuum of leaf venation phenotypes. However, this descriptive terminology can be incorporated as additional topological dimensions in the phenotypic space and alongside the metrics presented in this work can provide a tool to quantify inter- and intra- species diversity. In addition, we show that the local hierarchy of nested loops in the leaf venation network can be explained by very simple stochastic processes during development, pointing toward a universal mechanism governing (minor) vein morphogenesis.
The topological measures we employ have possible applications that range far beyond the leaf data set explored here, being usable on any loopy complex weighted network which possesses an embedding on a surface. Examples of systems that could benefit from an analysis along the lines of this work include the blood vessels in the retina, liver or brain, anastomosing foraging networks built by slime molds and fungi, lowland river networks, human-made street networks, force chain networks in granular materials, and many more, thereby possibly opening up an entire new line of research.
The extraction the networks from the original high-resolution scans (6400 dpi) can be divided into two main steps: segmentation of the image to create a suitable binary representation and skeletonization of the shapes. To segment the image we use a combination of Gaussian blurring to reduce noise, local histogram equalization and recombination with the original image to increase contrast, and Otsu thresholding [44] to find the optimal threshold for the creation of the binary image. For the skeletonization we use a vectorization technique known from optical sign recognition [45, 46]. The approach relies on the extraction and approximation of the foreground feature’s contours using the Teh-Chin dominant point detection algorithm [47] and subsequent triangulation of the contours via constrained Delaunay triangulation [48]. Therefore the foreground is partitioned into triangles which can be used to create a skeleton of the shape. Each triangle contributes a “center” point to the skeleton which is determined by looking for local maxima in the euclidean distance map [49] of the binary and together these center points approximate the skeleton. By looking at edges shared between two triangles, neighborhood relations can be established and an adjacency matrix can be created. This adjacency matrix defines a graph composed of nodes (the former triangle centers) and edges (the connections between two adjacent triangles). In addition to the topology of the graph the original geometry of the network including coordinates of the nodes and lengths and radii of edges are preserved and stored in the graph. The processing is done using algorithms implemented in python. The framework uniting all the aforementioned functionality is freely available at [50].
A complete and detailed description of the hierarchical decomposition algorithm to extract the nesting tree from leaf network graphs can be found in the supplement S1 Text. The software package used to calculate nesting numbers, topological lengths, and geometric metrics is freely available at [51].
The model starts from a single rectangular loop of veins (Fig 3a). The loops grow and subdivide when they reach a threshold size A0 by introduction of a new vein. Not all loops subdivide at exactly the same size: the probability of subdivision as a function of areole area is a sigmoidal of width σA (Fig 3). All veins start with a fixed small width and grow linearly with time. The relative growth rate of vein lengths and widths is controlled by the nondimensional parameter α. The areole subdivision is only approximately symmetric: the new vein is randomly positioned close to the midline of the areole and the extent of the asymmetry is controlled by a parameter ρ ∈ [0, 1] (see S1 Text). After the growing leaf has a certain size, the simulation is terminated and random Gaussian noise with zero mean and standard deviation proportional to the parameter fn is added to the vein diameters.
The model is controlled by the four dimensionless parameters ρ, β = σA/A0, α and fn.
|
10.1371/journal.pbio.1000505 | Polarized Secretion of Drosophila EGFR Ligand from Photoreceptor Neurons Is Controlled by ER Localization of the Ligand-Processing Machinery | The release of signaling molecules from neurons must be regulated, to accommodate their highly polarized structure. In the developing Drosophila visual system, photoreceptor neurons secrete the epidermal growth factor receptor ligand Spitz (Spi) from their cell bodies, as well as from their axonal termini. Here we show that subcellular localization of Rhomboid proteases, which process Spi, determines the site of Spi release from neurons. Endoplasmic reticulum (ER) localization of Rhomboid 3 is essential for its ability to promote Spi secretion from axons, but not from cell bodies. We demonstrate that the ER extends throughout photoreceptor axons, and show that this feature facilitates the trafficking of the Spi precursor, the ligand chaperone Star, and Rhomboid 3 to axonal termini. Following this trafficking step, secretion from the axons is regulated in a manner similar to secretion from cell bodies. These findings uncover a role for the ER in trafficking proteins from the neuronal cell body to axon terminus.
| Cells secrete signaling molecules that trigger a variety of responses in neighboring cells by activating their respective cell-surface receptors. Because many cells in an organism are polarized, regulating the precise location of ligand secretion is important for controlling the position and nature of the response. During the development of the compound eye of the fruit fly Drosophila, for example, a ligand of the epidermal growth factor family called Spitz (Spi) is secreted from both the apical and basal (axonal) poles of photoreceptor cells but with different outcomes. Photoreceptor cells are recruited to the developing eye following apical secretion of Spi. Conversely, basal secretion of this same ligand, at a significant distance from the cell body, triggers differentiation of cells in the outer layer of the brain. Although secretion of Spi is known to occur at both poles of the cell, one important question is how Spi and its processing machinery are trafficked throughout the length of the photoreceptor axon to achieve basal secretion. In this study we show that the key to axonal trafficking is the regulated localization of Spi and its processing machinery, including the intramembrane protease Rhomboid, to sites within the endoplasmic reticulum (ER), which extends along the length of the axon. Two different Rhomboid proteins are expressed in photoreceptor cells, but only one of them is localized to the ER. We show that this ER-localized Rhomboid is indeed necessary and sufficient for Spi processing at axon termini. Our work therefore demonstrates how variations in intracellular localization of conserved signaling components can alter signaling outcomes dramatically. It also highlights the importance of the ER in trafficking proteins along the axon.
| Communication between cells and their environment entails the release and reception of signaling molecules. In polarized cells, such as epithelia or neurons, the unique cellular architecture imposes constraints on the precise sites where signal release and reception occur. For example, the distribution of axonal guidance receptors is restricted to specific proximal or distal axon segments [1]. Similarly, secretion of molecules from neurons must be highly polarized for the ligand to propagate in the appropriate receptive field. In some cases, ligand is secreted along the axon, where it interacts with ensheathing glia [2],[3], whereas in other cases ligand is secreted locally from cell bodies or growth cones [4]–[6]. Thus, polarized secretion is an essential aspect of ligand processing in neurons.
An example of ligand secretion from both cell bodies and axonal termini is that of the Drosophila epidermal growth factor receptor (EGFR) ligand Spitz (Spi). In the Drosophila eye imaginal disc, photoreceptors differentiate in the wake of a progressive morphogenetic furrow, which sweeps from the posterior of the disc to its anterior [7],[8]. Secretion of Hedgehog (Hh) from nascent photoreceptor cell bodies promotes the continued movement of the furrow [9],[10]. Photoreceptor neurons subsequently secrete the EGFR ligand Spi from their cell bodies, triggering neurogenesis in closely neighboring cells [11],[12].
Once specified as neurons, R1–R6 photoreceptor axons grow across the basal surface of the eye disc, funnel through the optic stalk, and reach the lamina, where they locally induce the differentiation of lamina cartridge neurons [13],[14]. Secretion of Hh from photoreceptor axon termini triggers an initial phase of neurogenesis in the lamina precursor cells, marked by the expression of Dachshund (Dac) and the EGFR itself [5]. The subsequent phase of lamina neurogenesis requires Spi, which is also locally delivered by the incoming retinal axons. EGFR activation by Spi in the lamina leads to the differentiation of five neurons in each cartridge, which express the pan neuronal marker ElaV [6]. Thus, local secretion of Spi at the two distinct poles of photoreceptor neurons controls neurogenesis in both the eye disc and the lamina. While the mechanisms that regulate Hh delivery to axons have been explored [4], how Spi is secreted from both cell bodies and axonal termini remains unknown.
Spi is the cardinal EGFR ligand throughout Drosophila development. It is broadly expressed as an inactive precursor [15]. Spi secretion is dependent on processing by the intramembrane protease Rhomboid-1 (Rho-1) [16]. The inactive Spi precursor is retained in the endoplasmic reticulum (ER) by a COPI-dependent mechanism [17]. Trafficking of Spi from the ER to the Rho-1 compartment requires the type II transmembrane protein Star (S) [18],[19]. Upon arrival at this late secretory compartment, Spi is cleaved by the Rho-1 protease and subsequently released to the extracellular milieu.
Rho-1 also cleaves the chaperone S, thereby rendering it incompetent to traffic additional Spi molecules [20]. We have previously shown that two additional Rhomboid family members, Rho-2 (also called Stet) and Rho-3 (also called Roughoid [Ru]), which are dedicated to oogenesis and eye development, respectively [21],[22], localize to the ER, as well as to the late secretory compartment [23]. Although Rho-2 and Rho-3, like Rho-1, promote Spi release from the late compartment, their ER presence attenuates EGFR signaling, primarily because of premature cleavage of S [23]. Thus, in photoreceptor neurons, Spi secretion from cell bodies is promoted by both Rho-1 and Rho-3 acting in the late compartment, with the ER activity of the latter also attenuating the overall levels of secreted ligand.
The presence of ER markers has been observed in axons and dendrites from various neurons [24],[25], and the ER has been suggested to be continuous in Purkinje cell axons [26]. However, the traditional role assigned to axonal ER is in localized translation of transported mRNA, rather than translocation of secreted proteins. Recently, a role for the ER in promoting trafficking of NMDA glutamate receptor to dendrites in cultured rat hippocampal neurons has been described [27].
Here we examined the mechanisms that regulate Spi release from axonal termini. We find that, unlike secretion from cell bodies, axonal secretion of Spi relies exclusively on Rho-3. Furthermore, the ability of Rho-3 to promote axonal secretion of Spi stems from its combined ER and late compartment localization. Supplementing an ER presence to Rho-1 or eliminating the ER localization of Rho-3 alternates their potencies vis-à-vis axonal Spi secretion. Our data indicate that the importance of the ER stems from its ability to promote axonal trafficking of Rhomboids, a feature that we suggest is linked to the extension of the ER throughout the axon. Finally, we characterize the apical compartment in which Spi is processed in cell bodies, and suggest that it is also present at axonal termini, where Spi is processed following trafficking along the axon. Our results show that subcellular localization of the EGFR-ligand-processing machinery in photoreceptors dictates the polarity of ligand secretion, and highlight the role of the ER in facilitating protein trafficking from the neuronal cell body to the axon terminus.
To investigate the requirement for Rho-mediated cleavage in promoting Spi release from photoreceptor axons, we assessed the effect of rho-1 or rho-3 mutations on lamina neurogenesis. In late third-instar larvae, EGFR activation by Spi delivered from photoreceptor axons leads to the expression of the pan-neuronal marker ElaV at the posterior part of the lamina (Figure 1A and 1B). Visual systems rendered homozygous for a null rho-1 allele, using the Eyeless Gal4 UAS Flip (EGUF) system [28], occasionally show some morphological defects, but ElaV expression in the lamina is not perturbed (Figure 1C). Thus, rho-1 is dispensable for Spi release from photoreceptor axons. We next examined ElaV expression in rho-3 EGUF clones (Figure 1D) or in homozygous mutant animals (Figure 1H). While ElaV is properly expressed in the eye disc and brain lobula, we could not detect any ElaV expression in the lamina, indicating that rho-3 is essential for EGFR activation in this tissue. Thus, whereas Rho-1 and Rho-3 can redundantly promote Spi release from cell bodies in the eye disc, only Rho-3 mediates EGFR activation in the lamina.
Since Rho-3 is also involved in photoreceptor neurogenesis, the lack of EGFR activation in the lamina of rho-3 mutants may be a secondary effect of defective neuronal development or axonal mistargeting. However, rho-3 mutant photoreceptors properly express the pan-neuronal marker ElaV, as well as markers of specific photoreceptor subtypes (Figure 1D′ and unpublished data; [23]), demonstrating that the general program of photoreceptor differentiation is not perturbed. The only defect we observed at the larval stage is an extra number of neurons, at the expense of non-neuronal cells [23]. Importantly, no overt axonal targeting defects were detected in the mutant, as seen with anti–horseradish peroxidase (HRP) staining (Figure 1D). Furthermore, the normal expression of the Hh target genes dac (Figure 1D′′) and EGFR (Figure 1E) in the brain reveals that there is no general secretion defect in rho-3 mutants. It thus appears that the rho-3 mutant phenotype reflects a specific defect in processing and secretion of Spi from axon termini.
To critically test the functionality of rho-3 mutant photoreceptors, we performed electroretinogram (ERG) recordings on adult flies (Figure S1). Photoreceptor neurons from wild-type or rho-3 eyes properly depolarize in response to light. However, “on/off transients,” which represent the activity of the post-synaptic lamina neurons [29], are absent in rho-3 ERG recordings, thus reflecting the defects in lamina neurogenesis. Conversely, “on/off transients” are detected in rho-1 EGUF clones. Hence, in the absence of Rho-3, Rho-1 facilitates all aspects of photoreceptor development, but not the induction of EGFR activation in the lamina.
Rhomboids promote EGFR signaling by processing the ligand Spi in the signal-sending cell prior to its secretion [30],[31]. This suggests that the lack of EGFR activation in rho-3 mutant laminae stems from a failure in cleavage and secretion of Spi from photoreceptors. To follow Spi processing and secretion, we monitored the localization of Spi–green fluorescent protein (GFP), a biologically active variant of the ligand, tagged by GFP at the extracellular domain [19]. The construct was expressed under the control of GMR–Gal4 [32], to restrict expression exclusively to the eye disc.
Inspection of EGFR distribution in the laminae of wild-type flies reveals many endocytic puncta, which are associated with the ElaV-expressing cartridge neurons (Figures 1E and S1D). We found that Spi–GFP secreted from the eye co-localized in the lamina with EGFR in these puncta, reflecting the release of the ligand from photoreceptor axons and endocytosis of ligand–receptor complexes by lamina cells (Figure 1F and 1J). This co-localization is dependent on cleavage by Rhomboid proteases, since a similarly expressed Spi–GFP construct in which the Rhomboid cleavage site was mutated [33] failed to co-localize with the receptor (Figure 1G and 1K).
We next examined the distribution of EGFR in rho-3 mutant laminae, and found that it is uniform compared to wild-type, and lacks the bright endocytic puncta (Figures 1H and S1E). In rho-1 mutant visual systems, the distribution and intensity of laminar EGFR staining were comparable to wild-type (Figure S1F). Furthermore, following expression of Spi–GFP in rho-3 mutant eye discs, GFP-positive puncta could not be detected in the laminae (Figure 1I and 1K). These results indicate that Rho-3 cleaves Spi within the transmembrane domain in photoreceptor neurons, to promote ligand release from their axons to the lamina.
In summary, our results show that, whereas both Rho-1 and Rho-3 are capable of mediating Spi secretion from cell bodies in the eye disc, only Rho-3 promotes the secretion of Spi from photoreceptor axons to the lamina.
Each of the approximately 750 ommatidia in the Drosophila eye contains eight photoreceptor neurons of distinct identities. R1–R6 neurons project their axons to the lamina, whereas R7 and R8 project their axons to the medulla. To ask which of these neurons provides Spi for patterning the lamina, we used a repertoire of Gal4 lines to drive Rho-3 expression in different subsets of photoreceptors, and monitored their ability to rescue the rho-3 mutant phenotype. All Gal4 drivers used are normally expressed in rho-3 mutant eye discs (unpublished data). As a complementary assay, we expressed Spi–GFP with the same lines, and monitored its co-localization with the internalized EGFR in the signal-receiving lamina neurons. Our findings are summarized in Table 1, showing that Rho-3 acts to promote Spi secretion from the axons of R2 and R5. We note that these axons also play a pivotal role in axonal pathfinding, as their mistargeting can lead to defective guidance of the entire ommatidial fascicle [34]. The concordance between the assays of ElaV induction and Spi internalization in the lamina suggests that the difference between the photoreceptors that do or do not provide the signal lies in their ability to process or secrete Spi, rather than in the capacity of the lamina cells to respond only to Spi that is secreted from distinct photoreceptors.
A mechanism that may account for the importance of Rho-3 in promoting Spi secretion from axons is RNA transport and localized translation. However, we have found no rho-3 RNA in axons, even after Rho-3 overexpression, which rescues the rho-3 phenotype (Figure S2). We have previously shown that Rho-1 and Rho-3 differ in their subcellular localization within photoreceptor cell bodies. When ectopically expressed with the Gal4–UAS system, Rho-1 localized to apical punctate structures, whereas Rho-3 was localized to the ER, as well as to the apical puncta [23]. We set out to test the hypothesis that the distinct intracellular localizations of Rho-1 and Rho-3 account for the difference in their capacity to trigger Spi processing and secretion in photoreceptor axons.
First, we examined the endogenous localization of the two proteases, without resorting to overexpression. Since antibodies that recognized the endogenous proteins could not be raised, we used recombineering [35] to generate ∼45-kb genomic fragments encompassing the rho-1 or rho-3 locus that express C-terminally tagged Rho-1–yellow fluorescent protein (YFP) and Rho-3–GFP in patterns and levels identical to the endogenous proteins. Transgenic lines were generated, in which the recombineered genes were inserted at the same chromosomal location. In the eye disc, genomic Rho-1 (gRho-1)–YFP localized exclusively to the apical compartment, whereas gRho-3–GFP was enriched in the ER, with staining also at the apical compartment (Figure 2A and 2B). These distributions demonstrate that despite the caveats associated with overexpression, the localizations obtained previously by the UAS–Gal4 system faithfully reflected the endogenous localization of these proteins.
To identify the sequences mediating the subcellular localization of Rhomboids, we swapped different fragments between Rho-1 and Rho-3. The resulting chimeras were GFP tagged, and transgenic animals were generated. In all cases the constructs were inserted at the same genomic location, to avoid a difference in expression levels. We find that the subcellular localization of Rhomboids depends on their cytoplasmic N terminus and the first intraluminal loop. Replacing these fragments of Rho-1 with the corresponding fragments from Rho-3, to yield GFP–R3L1-R1, relocalized Rho-1 to a Rho-3-like distribution, encompassing the ER and apical compartment (Figure 2C and 2F). Conversely, Rho-3 in which the N terminus and first loop were replaced by those of Rho-1 (GFP–R1L1-R3) retained localization to the apical compartment, but was absent from the ER (Figure 2D and 2E). Importantly, since the active site of the proteases is formed by residues embedded within the fourth and sixth transmembrane helices [36]–[38], the chimeras uncouple the subcellular localization signal from the catalytic activity. Therefore, the GFP–R1L1-R3 and GFP–R3L1-R1 constructs allow us to specifically define the role of subcellular localization in promoting Spi secretion from axonal termini.
Although both Rho-1 and Rho-3 promote Spi secretion from photoreceptor cell bodies, only Rho-3 facilitates Spi secretion from axons. To investigate whether this is due to its ER localization, we assayed the ability of GFP–Rho-1 or GFP–Rho-3 to rescue the rho-3 lamina phenotype. In addition, we tested a modified Rho-1 targeted to the ER and late compartment (GFP–R3L1-R1) and an ER-excluded Rho-3 (GFP–R1L1-R3) using the same assay. All constructs were shown to be efficient in cleaving Spi in cell culture assays and in vivo (unpublished data). Furthermore, since Rho-1 and Rho-3 are normally expressed at low levels in the eye disc, we inserted all the transgenes into attP18, a genomic landing site that was reported to yield low expression levels [39], and expression was driven in R2, R5, and R8 by MT14–Gal4.
As expected from their in vivo activities, GFP–Rho-3 rescued the rho-3 mutant lamina phenotype, whereas GFP–Rho-1 did not (Figure 3). Importantly, while GFP–Rho-1 failed to promote Spi secretion from the axons, supplementing it with an ER localization yielded a construct (GFP–R3L1-R1) capable of rescuing the rho-3 phenotype (Figure 1E and 1F). Conversely, whereas GFP–Rho-3 rescued the rho-3 phenotype, a Rho-3 version which is not ER localized (GFP–R1L1-R3) failed to do so (Figure 1D and 1F). These experiments show that ER localization is a critical feature that enables Rhomboid proteases to promote Spi secretion from the axons.
We next asked whether intact endogenous Rho-1, which cannot substitute for Rho-3 in Spi processing for axonal release, can facilitate Spi secretion when enriched in the ER. Passage through the ER is an essential step in Rho-1 maturation, as a protein bearing transmembrane domains. We thus attempted to compromise Rho-1 exit from the ER, by removing one copy of the syntaxin sed5, which is required for the fusion of ER-derived vesicles with the Golgi [40],[41]. When HA-tagged Rho-1 was expressed in sed5 homozygous mutant clones, its subcellular distribution shifted almost completely to the peri-nuclear ER (Figure 3G and 3H). In rho-3 mutants in which sed5 gene dosage was halved, we found that some ElaV expression was restored to the lamina (Figure 3I and 3J). Therefore, when endogenous Rho-1 trafficking out of the ER is compromised, it can substitute for Rho-3 and promote Spi release from axons.
We note here that under strong overexpression conditions, Rho-1 also rescues the rho-3 phenotype. This may reflect the perdurance of some Rho-1 in the ER when its export machinery is heavily burdened. Indeed, a low endogenous level of ER activity by Rho-1 en route to the apical compartment has been suggested previously [17]. Accordingly, the ER levels of Rho-1–HA in sed5 heterozygotes were too low to be detected by anti-HA staining, yet restored some laminar ElaV expression to rho-3 mutants. In summary, our results indicate that the difference in subcellular localization is the cause of the distinct ability of Rho-3, but not Rho-1, to promote Spi processing and secretion from photoreceptor axons.
The combined ER and secretory compartment localization of Rho-3 is critical for its ability to promote Spi secretion from axons. We next asked whether the ER component of this localization is sufficient for Rho-3 function in lamina induction. We uncoupled the two localizations by tagging Rho-3 with a KDEL sequence at its luminal C-terminus, thereby retaining it in the ER. This construct, as well as a KDEL-tagged Rho-1, were fused at their N-termini to GFP, and inserted into the same genomic landing site as the constructs previously described. Although GFP–Rho-3–KDEL and GFP–Rho-1–KDEL localize to the ER, and efficiently cleave Spi in cell culture assays and in vivo (unpublished data), they could not rescue the rho-3 lamina phenotype upon expression in the eye by MT14–Gal4 (Figure 4A–4D). This indicates that the ER localization of Rho-3 is not sufficient to promote EGFR signaling in the lamina, and suggests that the active Spi molecules secreted from the axons are not processed in the ER.
Since Spi that is secreted by photoreceptor axons is not cleaved in the ER, we monitored the capacity to traffic the Spi precursor to axonal termini. GMR–Gal4-driven expression in a wild-type eye disc of the Spi precursor marked with GFP at the N terminus, gave rise to translocation of the GFP tag across the entire length of the axon bundle (Figure 4E and 4H). However, it is not possible to determine by this assay whether the ligand that reaches the axon termini represents the precursor form or the cleaved ligand. Two lines of evidence suggest that the ligand precursor can be trafficked from the cell body to the axon terminus. First, a non-cleavable form of Spi also reached the axonal growth cones, when expressed in the eye disc (Figure 4F and 4H). Second, expression of mSpi–GFP in a rho-3 mutant background, in which the precursor does not undergo cleavage in the ER, gave rise to a ligand distribution in axons that was similar to wild-type (Figure 4G and 4H). Taken together, these experiments demonstrate that the Spi precursor can be trafficked along the axon, and suggest that it is cleaved outside of the ER prior to its secretion.
To support this conclusion, we assayed the ability of a cleaved form of the ligand (cSpi), which is localized to the ER [17], to rescue the rho-3 phenotype upon expression by MT14–Gal4 in R2, R5, and R8. Biologically active cSpi, tagged with HA or HRP, failed to induce ElaV expression in rho-3 laminae (Figure S3). This is consistent with the notion that cleavage of Spi in the ER is not the mode by which Rho-3 promotes secretion, and suggests that the importance of the ER to Rho-3 function stems from a different mechanism.
The above experiments demonstrate that while ER localization is crucial for the ability of Rho-3 to promote axonal secretion of Spi, the functional ligand is not cleaved in the ER. We therefore examined whether the ER could promote Rho-3-dependent signaling by facilitating the trafficking of the ligand-processing machinery to axons.
Examination of the endogenous ER markers protein disulfide isomerase (PDI) and BiP reveals that the ER extends throughout the axons of developing photoreceptor neurons (Figure 5A and unpublished data), as does the detection of KDEL-tagged ER luminal proteins (Figure 5B). ER markers were also observed in axons of adult flies (unpublished data), consistent with previous reports indicating that the ER is continuous in the axons of various neurons [26],[42]. We also detected the presence of endogenous ER exit sites (marked by dSec16 [43]) along the axons and at their termini in the lamina (Figure 5C), suggesting that proteins are released from the ER in these locations. Consistently, Golgi outposts (marked by Mannosidase II (ManII)–GFP [44]) were also evident along the entire axon length (Figure 5F). These observations suggest that in photoreceptor axons, the ER can be used by secreted proteins to reach a given exit site, prior to progressing along the secretory pathway.
To further test this idea, we expressed an ER-localized GFP (GFP–KDEL) [45] in the eye disc. GFP immunofluorescence was observed throughout the axons, while GFP mRNA was confined to the cell bodies (Figure 5D and 5E). Thus, proteins localized to the ER in the cell body can also reach the axon, by utilizing the extension of the ER to the axon.
Since Rho-3 is ER localized in the cell body, it could use this compartment in a manner similar to GFP–KDEL to move distally. Indeed, whereas rho-3 mRNA is not detected in the axons (Figure 5G), gRho-3–GFP is found in a continuous distribution in axons (Figure 5H). Conversely, gRho-1–YFP, which is not localized to the peri-nuclear ER, fails to reach the optic stalk (Figure 5I). To examine the possibility that ER localization would promote the axonal delivery of a Rhomboid protease, we generated another gRho-1–YFP construct, with a C-terminal KDEL tag. In contrast to gRho-1–YFP, gRho-1–YFP–KDEL was robustly distributed along the entire length of the axons (Figure 5J).
Taken together, these results imply that the importance of the ER for Spi signaling in this physiological context stems from its ability to promote trafficking to the axons, where Spi processing subsequently occurs.
Besides Rho-3, Spi and S are also localized to the ER in the eye disc. Therefore, the three components could associate in this compartment for joint trafficking to the axons. To test this hypothesis, we examined the co-localization of biologically active, HA-tagged versions of Spi or S with Rho-3–GFP. S–HA co-localizes with Spi–GFP in the axons at the optic stalk (Figure 6B). In photoreceptor cell bodies S was shown to stabilize Spi [46].We observed that S stabilizes Spi in axons, and promotes its trafficking through the axons, as more Spi–GFP molecules arrive at the lamina when co-expressed with S–HA (Figure 6D–6F). S–HA also co-localizes with Rho-3–GFP in the axons. Both the ligand and chaperone thus appear to co-localize with Rho-3–GFP in axons traveling through the optic stalk (Figure 6A and 6C).
We have previously shown that S is a substrate for ER-localized Rhomboid proteases [23], and that cleaved S cannot traffic Spi [20]. ER-based cleavage of S has a functional significance, as it limits the trafficking of the Spi precursor by the S chaperone out of the ER. This results in an increased sensitivity of EGFR signaling to S levels. Indeed, S heterozygous flies exhibit reduced EGFR signaling during oogenesis and eye development, where the ER-active Rho-2 and Rho-3 mediate Spi processing, respectively [23]. Thus, a sensitivity to S levels is indicative of exposure to Rhomboid-based cleavage in the ER. We find that S heterozygous flies show a severe reduction in ElaV expression in the lamina (Figure 6G and 6H). Importantly, the defect in EGFR signaling in the laminae of these flies is significantly more severe than the compromised induction of photoreceptors in the eye disc. This may reflect a longer exposure of S to ER cleavage by Rho-3 during trafficking to the axon termini. Thus, the hypersensitivity of the lamina to S gene dosage supports the notion that S and Rho-3 are jointly trafficked through the ER in photoreceptor axons.
Following its trafficking to the axonal termini, Spi seems to be secreted locally at a precise location [6]. In the eye disc, Spi is also secreted locally, from a late secretory compartment where Rho-1 and Rho-3 reside [23]. To gain insight into the mechanism of Spi release, we set out to identify the “late compartment” in the eye disc. A variety of compartment markers were tested for co-localization with Rho-1–HA expressed in the eye disc (see also [23]), including a collection of YFP-tagged Rab proteins [47]. The only significant co-localization was observed with YFP–Rab4 and YFP–Rab14 (Figure 7A and 7B). This co-localization was also verified in cell culture, where a significant proportion of Rho-1-, Rab4-, and Rab14-positive puncta overlap (Figure S4A). YFP–Rab4 and YFP–Rab14 also co-localize with apical, but not peri-nuclear, Rho-3–HA staining in the eye disc (Figure S5).
Interruption of Rab4 and Rab14 function in photoreceptors by RNA interference (RNAi) or dominant negative (DN) approaches did not result in any discernible phenotypes. However, both Rab proteins interact with effectors of Rab11 [48],[49], suggesting a role for this major conserved regulator of endosomal trafficking in Spi exocytosis. Indeed, expression of a DN form of Rab11 in Drosophila cell culture disrupted the morphology of Rab4/14 endosomes, marked by Rho-1–red fluorescent protein (RFP) or Spi–HA, when the latter was co-expressed with S (Figure S4). Furthermore, in the eye imaginal disc, Rho-1–GFP, which is normally localized to discrete puncta, is misocalized upon co-expression of Rab11DN by GMR–Gal4 (Figure 7C and 7D). Thus, although Rab11 does not co-localize to the Rho-1-containing endosomes, its function is essential for their correct formation.
We then asked whether EGFR signaling is affected by impairment of the Rab4/14 compartment. Indeed, expression of Rab11DN by GMR–Gal4 led to a reduction in the number of ElaV-expressing cells in the eye disc (unpublished data), as did expression of a Rab11 RNAi construct (Figure 7E). Importantly, there was no alteration of photoreceptor R8 differentiation, which is not dependent upon EGFR signaling. Since this phenotype may reflect a requirement for Rab11 in the signal-receiving cells, downstream to EGFR, we expressed the Rab11DN construct specifically in R8, which is the only photoreceptor that acts exclusively as a signal-emitting cell. Again, EGFR phenotypes such as missing photoreceptors and mis-rotated ommatidia were readily apparent (Figures 7F, S6A, and S6B). This indicates that Rab11 acts non-autonomously in R8, where it is required for EGFR ligand secretion.
When larvae expressing UAS–Rab11DN by GMR–Gal4 in the eye disc were allowed to develop, the resulting adults had very small and rough eyes, as previously reported (Figure 7G; see also [47]). Although Rab11 has pleiotropic functions, this phenotype is at least partly due to a specific failure in EGFR ligand secretion, since co-expression of Rho-1 with Rab11DN considerably ameliorated the phenotype (Figure 7H). We conclude that in the eye disc, Spi is cleaved and secreted from Rab4/14 endosomes, and that the normal function of these endosomes is required for EGFR ligand trafficking and processing.
The requirement for Spi cleavage to take place after ligand is trafficked out of the ER in both the cell bodies and axons, raised the possibility that subsequent trafficking steps also share common features. We therefore sought to determine whether Spi secretion from the axons similarly involves Rab4/14 endosomes, and is dependent upon Rab11 function. Indeed, we found that Rab4 or Rab14, expressed in the eye disc by GMR–Gal4, reached axonal growth cones, as did Rab11. Note that GMR–Gal4 does not drive expression in the lamina ([6] and Figure 4E–4G). As in the eye disc, co-localization between Rho-3–HA and YFP–Rab4 or YFP–Rab14 was observed in axonal termini (Figure S5), but not along the length of the axons in the optic stalk (unpublished data).
Expression of Rab11DN in the eye disc by GMR–Gal4 led to a significant reduction in the number of ElaV-positive cells in the lamina, while Dac expression was normal (Figure 7I). Importantly, expression of Rab11DN in R8, which does not secrete Spi to the lamina, severely impairs EGFR signaling in the eye disc but not in the lamina (Figure S6). To further separate the axonal function of Rab11 from its requirement in photoreceptor differentiation, we expressed Rab11DN by GMR–Gal4 together with RasV12, which induces massive photoreceptor recruitment ([50] and Figure S7A). In the eye disc RasV12 was epistatic to Rab11DN, where all cells were converted to ElaV-expressing neurons, supporting the notion that Rab11 acts upstream to Ras (Figure S7). Expression of RasV12 in the eye induces an enlarged lamina with extra lamina neurons. Co-expression of Rab11DN attenuated the effects of RasV12 on lamina development in seven of 12 specimens, leading to wild-type or even reduced ElaV expression (Figure S7). In other words, we have uncoupled the requirement for Rab11 for secretion of the ligand in the eye disc and in the lamina by using RasV12 to bypass the requirement for the ligand in the eye disc. Therefore, this effect specifically represents the requirement for Rab11 to allow secretion of the ligand at the axon termini. This is consistent with the notion that after trafficking of mSpi, S, and Rho-3 to the axonal termini, secretion occurs in a similar manner to the eye disc, utilizing a Rab11-dependent mechanism.
Polarized secretion of ligands from a signal-emitting cell to the appropriate receptive field is crucial for correct intercellular communication. Control over EGFR ligand secretion, and consequently EGFR activation, in Drosophila is achieved through trafficking and compartmentalization of the ligand-processing machinery. This work identifies a link between the subcellular localization of the Spi-processing machinery and the polarized release of Spi from axons.
Subcellular localization of Rhomboid proteases, which process the inactive Spi precursor, impinges on ligand secretion [23]. Both Rho-1 and Rho-3 are localized to apical Rab4/14 endosomes, where they are redundant in promoting Spi release from cell bodies. In contrast, only the Rho-3 protease mediates axonal secretion of Spi. This is evident from the rho-3 mutant phenotype, which shows a complete loss of EGFR activation in the lamina. Since the two proteases are expressed in the neurons which secrete Spi, and share the same substrate specificity, these features cannot account for the specific requirement for rho-3. RNA transport and localized translation of Rho-3 are also inconsistent with the following observations: (a) no rho-3 RNA was detected in axons, (b) gRho-3–GFP, reflecting endogenous expression, is localized throughout the axon, rather than concentrated at a point of localized translation, and (c) Rho-3 cDNA, devoid of 3′ or 5′ UTRs, rescued the mutant phenotype. The RNA of the rescuing construct was also not localized to axons.
Our results indicate that the exclusive requirement for Rho-3 is due to its ER localization. Re-localization of some of the Rho-1 pool to the ER, or removal of Rho-3 from the ER, achieved by swapping specific sequences, alternated their potencies to promote axonal secretion of Spi. Furthermore, when the ER export of endogenous Rho-1 was compromised, EGFR activation was partially restored to the lamina of rho-3 mutants. Thus, the ER localization of Rho-3 in photoreceptor neurons serves a dual function: it negatively regulates Spi secretion from cell bodies, via premature cleavage of S [23], and positively promotes Spi secretion from the axons to the lamina, by facilitating trafficking of the ligand-processing machinery to axon temini (schematized in Figure 7J).
How does the ER localization of Rho-3 contribute to Spi secretion from axons? The inability of GFP–Rho-3–KDEL or cSpi–HA to rescue the rho-3 phenotype demonstrates that the axonally secreted Spi is not cleaved in the ER, and prompted investigation into the role of the ER in promoting axonal trafficking.
We have shown that in Drosophila photoreceptor neurons, the ER extends throughout the axons. ER exit sites and Golgi outpost markers were also detected in axons. The continuity of the ER was previously demonstrated in Purkinje neurons [26] and in other cell types, including Drosophila oocytes [42],[51]. This implies that ER-localized proteins could use this compartment to move distally in the axon. Indeed, GFP–KDEL expressed in the eye disc reaches the axonal termini. Furthermore, the ER-localized Rho-3 is enriched in axons, as opposed to Rho-1, which is restricted to endosomes. Importantly, restricting the gRho-1 construct to the ER with a KDEL sequence gave rise to a robust translocation of the protease throughout axons, reaching their growth cones in the lamina.
ER-facilitated trafficking of Rho-3 could occur through diffusion in the ER membrane, with exit and retrieval of ER-derived vesicles being biased distally. Alternatively, and perhaps more likely, the ER presence of Rho-3 could lead it to an exit site localized at the axon base, from which trafficking would be directed towards the growth cones. This would explain the ability of Rho-1 to rescue the rho-3 phenotype under strong overexpression conditions. Distinction between these possibilities would require co-localization of Rho-3 or Spi immunoreactivity with known compartment markers in axons. So far, and despite a large number of markers examined, we could not detect such co-localization (unpublished data). Since the extension of the ER is correlated with the growth of the axons [52],[53], ER-facilitated trafficking also provides a means of ensuring that ligand is released only once the axons have reached their target layer, and ER exit sites and Golgi membranes are set in place.
Spi, S, and Rho-3 are all localized to the peri-nuclear ER in the eye disc. Since all three proteins can interact with one another [19],[46], this implies that the processing machinery could assemble in the ER for joint trafficking. Indeed, we found that Spi, S, and Rho-3 also co-localize in photoreceptor axons. Further evidence for the joint trafficking of S and Rho-3 is the marked sensitivity of EGFR signaling in the lamina to S levels. We have previously shown that S cleavage in the ER leads to compromised EGFR activation phenotypes upon halving S gene dosage [23]. The observation that EGFR signaling in the lamina is even more sensitive to S gene dosage than in the eye suggests that Rho-3 and S spend a significant time in the ER, where the chaperone is exposed to inactivation by cleavage.
How targeting of Spi–S–Rho-3 complexes to the basally located axons or the apical Rab4/14 endosomes is achieved is unclear. In the case of Hh, the presence or absence of the C-terminal cleavage fragment in the Hh-containing vesicle determines its destination [4]. The Spi C-terminus is not required for axonal targeting, since a Spi–GFP construct lacking most of the C-terminus showed the same distribution as intact Spi–GFP upon expression in the eye (unpublished data). Alternatively, another factor, which would be ER localized, could promote the trafficking of the processing machinery to axons. This factor is also expected to be expressed mainly in R2, R5, and R8, accounting for their importance in Spi secretion to the lamina. In the Drosophila oocyte, the polarized ER exit of another EGFR ligand, Gurken, is regulated by Cornichon. Somatic functions for Cornichon and its homolog Cornichon related have also been identified but not thoroughly explored yet [54].
While the presence of ER markers in axons or dendrites has been previously reported [27], the biological significance of such observations, commonly derived from protein localization data in cultured neurons, could only be speculated upon, since no functional readout was examined. The unique properties of photoreceptor axons in Drosophila, which not only conduct electrical signals but are also involved in transmitting developmental cues at an earlier phase, have allowed us to functionally demonstrate the essential role of the ER in trafficking the complete EGFR ligand-processing apparatus to axon termini. This mechanism is clearly distinct from the established roles of the axonal ER in allowing local translation of secreted or transmembrane proteins whose mRNAs are enriched at axon termini.
Spi is released to the extracellular milieu following cleavage by Rho-1. Different experimental systems have yielded conflicting reports as to the compartment in which the protease resides [18]–[20],[55]. We now find that in both photoreceptor neurons and Schneider cells, Rho-1 is localized to an endosomal population marked by Rab4 and Rab14. Rab4 localizes to fast recycling endosomes, which mediate the retrieval of endocytosed cargo to the plasma membrane [56],[57]. Rab14 mediates trafficking between the Golgi and endosomes [58],[59]. Both Rab4 and Rab14 share binding proteins with Rab11 [48],[49], a major regulator of vesicle transport.
The role of endosomal dynamics in Spi secretion is manifested by the EGFR phenotypes obtained following expression of Rab11 RNAi or DN constructs. While Rab11 has pleiotropic functions and is not dedicated to EGFR signaling, perturbing Rab11 directly impinges on Spi secretion. This was evident from the mislocalization of Rho-1–GFP in Rab11DN-expressing photoreceptors, and from similar effects in cell culture. This mislocalization is likely the cause of the phenotype, since co-expression of Rho-1 or Rho-3 with Rab11DN abrogated the small eye phenotype associated with Rab11DN expression. Although interfering with endosomal dynamics may also perturb signaling downstream of the receptor, we did not observe a mislocalization of EGFR itself (unpublished data). Furthermore, the expression of Rab11DN in R8 impaired the differentiation of nearby cells into photoreceptor neurons, demonstrating that Rab11 acts non-autonomously upstream of the receptor, consistent with a role in ligand secretion.
Rho-1 and some of the Rho-3 pool are localized to Rab4/14 endosomes. The intracellular route by which they reach these compartments remains to be explored. From the ER accumulation of Rho-1–HA in sed5 mutant clones, we infer that the proteases do not undertake a Golgi-independent route to the Rab4/14 endosomes [41]. Furthermore, Rab14 mediates trafficking between the Golgi and endosomes [59], and Rab11 endosomes can be reached without passing through the plasma membrane (see for example [60]–[62]). Therefore, there is no indication that Rhomboids must pass through the plasma membrane to reach the endosomal compartment. Nevertheless, if Spi is secreted by fusion of Rhomboid-containing endosomes with the membrane, then retrieval by endocytosis should play a role in shaping the steady-state distribution of Rhomboids. Accordingly, we have found that upon expression of a DN form of the Dynamin Shibire, Rho-1–HA immunofluorescence is detected on the plasma membrane (unpublished data).
Trafficking of Spi to endosomes also provides an efficient means of disposing of the ligand in cells that do not express a Rhomboid protease, to prevent nonspecific cleavage on the plasma membrane. In this case, the membrane-bound precursor could be sorted to a membrane domain that segregates to multi-vesicular bodies, and then degraded in the lysosome. Accordingly, distinct membrane domains have been described for Rab4 and Rab11 endosomes [63].
Finally, we detected a co-localization between Rab4/14 and Rho-3 at axonal termini, but not in the optic stalk, and found that disrupting Rab11 function in the eye disc compromised EGFR signaling in the lamina. This effect was not due to defects in eye development, as Rab11DN expressed in R8 also impaired eye development but had no effect on the lamina. This finding raises the possibility that the final steps of secretion from axonal termini and cell bodies are regulated in a similar manner, although Rab11 seems to play a more prominent role in secretion from cell bodies. A precedent supporting such a hypothesis is the requirement for Sec15, which interacts with Rab11, for the localization of several molecules at both photoreceptor cell bodies and axonal termini [64].
In summary, our results describe a mechanism of ER-facilitated trafficking of secreted molecules in axons, prior to processing and secretion at the axon tip. This mechanism could also be utilized for other proteins that are secreted in a polarized manner in neurons.
For the generation of gRho-1–YFP and gRho-3–GFP, 40–45 kb from the rho-1 or rho-3 loci, encompassing the ORFs and flanking region, were cloned into P[acman–attB, AmpR] by recombineering-mediated gap repair [35]. The domains extend between 3L:1437674 and 1475379 and 3L:1355719 and 1397235 (release 5.23) for rho-1 and rho-3, respectively. A YFP tag or a YFP–KDEL was inserted at the rho-1 C-terminus by GalK positive/negative selection [65]. rho-3 was GFP tagged at the C-terminus using the PL452 C-EGFP tag template vector [66]. Both constructs were injected into VK00005 landing site.
For GFP–Rho-1, GFP–Rho-3, GFP–R1L1-R3, and GFP–R3L1-R1, eGFP was cloned into pUAST–attB at the BglII–EcoRI sites. cDNAs were then cloned using EcoRI and XhoI. All constructs were sequenced, and injected into attP18 lines [39].
cSpiHA contains a triple HA tag from pTWH, inserted after the Spi cleavage site.
mSpi–HA was generated by a site-directed mutagenesis insertion of an XhoI site after T58 of Spi, into which a triple HA tag was subsequently inserted.
mSpi–GFPmut was obtained from S. Urban [33], and cloned into pTWM. Cleavage assays in S2 cells verified that this construct cannot undergo Rhomboid-dependent cleavage (unpublished data).
S–HA is the S cDNA cloned into pTHW. mSpi–GFP and cSpi–HRP were previously described [17],[19]. The cleavage activity of all Rhomboid constructs has been tested in cell culture, and the biological activity of all UAS-based constructs was assayed by expression in wing or eye imaginal discs.
Climbing late third-instar larvae were dissected and fixed in PBS containing 4% PFA. All subsequent washes and antibody incubations were done in PBS with 0.1% Triton X-100.
Primary antibodies used were anti-FasIII (mouse, 1∶50), anti-EGFR (rat, 1∶1,000), anti-Senseless (guinea pig, 1∶2,000; from H. Bellen), anti-dSec16 (rabbit, 1∶1,000; from C. Rabouille), anti-Myc (mouse, 1∶100; Santa Cruz Biotechnology), anti-GFP (chick, 1∶2,000; Abcam), anti-HA (mouse, 1∶1,000; Roche), and anti–Troponin H to detect BiP (rat, 1∶100; Babraham Bioscience Technologies). Anti-ElaV (rat, 1∶2,000, or mouse, 1∶500) and anti-Dac (mouse, 1∶500) were obtained from the Developmental Studies Hybridoma Bank, University of Iowa. Cy-5-conjugated goat anti-HRP, as well as Cy-2-, Cy-3-, and Cy-5-conjugated secondary antibodies (1∶200) were obtained from Jackson ImmunoResearch.
In situ hybridizations using rho-3 or GFP probes were done using standard techniques.
The following lines were used: GMR–Gal4, Sca–Gal4, m™–Gal4 (from M. Mlodzik), Lz–Gal4, K25–Gal4, MT14–Gal4 ([34], from I. Salecker), UAS–GFP–KDEL [45], MS1096–Gal4, PDI–GFP [67], sed5AR113 (From C. Rabouille), SIIN23, a collection of YFP-tagged, native or DN UAS–Rab transgenes [47], UAS–Rab11DN (from M. Gonzalez-Gaitan), UAS–ManII–GFP (from Y. Jan), and UAS–Rab11–RNAi (VDRC22198). Null alleles of rho-1 (rho-1Δp38) and rho-3 (ruPLLb) were recombined with FRT2A, and crossed to ey–Gal4,UAS–FLP/Cyo;FRT2a,GMR–hid,l(3)CL–L1/TM6B to generate entirely mutant eyes [28]. To generate sed5 AR113 MARCM clones expressing Rho-1–HA, C155–Gal4,UAS–CD8GFP,hsFLP;Gal80,FRT40A females were crossed to sed5 AR113,FRT40A/+;UAS–Rho-1HA/+ males. Wild-type clones were generated with a chromosome bearing only FRT40A. Clones expressing ManII–GFP were induced in animals of the following genotype: C155–Gal4,hsFLP/+;UAS–ManII–GFP/+;FRT82B.
UAS–mSpi–GFPmut, UAS–cSpi–HA, UAS–mSpi–HA, UAS–GFP–Rho-1, UAS–GFP–Rho-3, UAS–GFP–R1L1-R3, and UAS–GFP–R3L1-R1 were generated by standard P-element or phi31 germline transformation procedures.
ERG recordings were performed as described in [29].
|
10.1371/journal.pntd.0001642 | Characterisation of the Native Lipid Moiety of Echinococcus granulosus Antigen B | Antigen B (EgAgB) is the most abundant and immunogenic antigen produced by the larval stage (metacestode) of Echinococcus granulosus. It is a lipoprotein, the structure and function of which have not been completely elucidated. EgAgB apolipoprotein components have been well characterised; they share homology with a group of hydrophobic ligand binding proteins (HLBPs) present exclusively in cestode organisms, and consist of different isoforms of 8-kDa proteins encoded by a polymorphic multigene family comprising five subfamilies (EgAgB1 to EgAgB5). In vitro studies have shown that EgAgB apolipoproteins are capable of binding fatty acids. However, the identity of the native lipid components of EgAgB remains unknown. The present work was aimed at characterising the lipid ligands bound to EgAgB in vivo. EgAgB was purified to homogeneity from hydatid cyst fluid and its lipid fraction was extracted using chloroform∶methanol mixtures. This fraction constituted approximately 40–50% of EgAgB total mass. High-performance thin layer chromatography revealed that the native lipid moiety of EgAgB consists of a variety of neutral (mainly triacylglycerides, sterols and sterol esters) and polar (mainly phosphatidylcholine) lipids. Gas-liquid chromatography analysis showed that 16∶0, 18∶0 and 18∶1(n-9) are the most abundant fatty acids in EgAgB. Furthermore, size exclusion chromatography coupled to light scattering demonstrated that EgAgB comprises a population of particles heterogeneous in size, with an average molecular mass of 229 kDa. Our results provide the first direct evidence of the nature of the hydrophobic ligands bound to EgAgB in vivo and indicate that the structure and composition of EgAgB lipoprotein particles are more complex than previously thought, resembling high density plasma lipoproteins. Results are discussed considering what is known on lipid metabolism in cestodes, and taken into account the Echinococcus spp. genomic information regarding both lipid metabolism and the EgAgB gene family.
| The larva of the cestode parasite Echinococcus granulosus affects a wide range of livestock mammals and humans, causing cystic echinococcosis (hydatid disease), a zoonosis with significant economic and public health impact. The disease is characterised by the growth of a fluid-filled cyst in the host's viscera (mainly liver and lung). The most relevant antigen for hydatid disease diagnosis is antigen B (EgAgB), a highly abundant lipoprotein present in the cyst fluid. There is overwhelming literature regarding EgAgB antigenicity and molecular characterisation at the protein and gene levels, but the knowledge of the lipids physiologically bound to EgAgB protein subunits is very scarce. Indeed, there is only one report showing that delipidated EgAgB binds fatty acids in vitro. This work describes the physiological lipids of EgAgB, an important piece of information to complete our knowledge on EgAgB molecular composition. In contrast to what was thought, EgAgB consists of a variety of neutral and polar lipid classes, associated to protein subunits, forming a plasma lipoprotein-like particle of 229 kDa in size. Taken into account that E. granulosus cannot synthesise fatty acids and sterols, these data suggest that EgAgB plays a role in the uptake and transportation of these essential lipids across parasite structures.
| The larval stage of the cestode parasite Echinococcus granulosus is the causative agent of cystic echinococcosis (hydatid disease) in a range of mammalian species (mainly domestic ungulates) as well as in humans. It is a unilocular fluid-filled cyst, which steadily grows inside host visceras (mostly liver and lung). One of the major molecules produced in large amounts by the cyst is a highly immunogenic lipoprotein named antigen B (EgAgB) [1], [2], which represents a major diagnostic antigen for human infection [3]–[5]. This antigen is present in various larval locations including the parasite cellular layer of the cyst wall (germinal layer), the larval worms or protoscolex (asexually produced towards inside the cyst) and the hydatid cyst fluid (HCF). HCF is a complex mixture of parasite excretory-secretory products and host-derived molecules that constitutes the liquid content of the cyst [6]–[8]. Evidence for EgAgB presence in host circulation is very limited [9]. The strong antibody response mounted by infected patients against this antigen indicates that it is likely released into the host-parasite interface. However, it is unknown whether it is released throughout the infection or just at a certain time point [10], [11].
A lot of efforts have been made to understand the molecular composition/organization of EgAgB (reviewed by [12]). The native antigen is a lipoprotein which exhibits an estimated molecular weight of 120 to 160 kDa according to sedimentation equilibrium and gel filtration studies, respectively [1], [2]. The apolipoprotein components of EgAgB are encoded by a polymorphic multigene family that comprises five clades named EgAgB1 to EgAgB5 [13]–[17]. There is a long and yet unsettled controversy regarding EgAgB gene copy number. Based on the characterisation of E. granulosus isolates from different geographic origins, a recent study has proposed that there are at least 10 EgAgB distinct genes, including four and three different genes corresponding to the EgAgB3 and EgAgB4 clades [18]. However, a recent analysis of EgAgB loci in the current assembly of E. granulosus genome revealed the presence of seven EgAgB loci clustered on a discrete region of the genome, with one copy each of EgAgB1, EgAgB2, EgAgB4 and EgAgB5, as well as three slightly differing copies of EgAgB3 [19]. Outside this cluster only an EgAgB pseudogene was detected. However, the authors did not rule out the possibility of additional EgAgB genes in extra-chromosomal DNA arrays that might have slipped the genome assembly process [19]. There is evidence that EgAgB genes are differentially expressed in single life-cycle parasite stages, and also within distinct tissues of a same parasite stage (i.e. protoscolex and germinal layer) [18], suggesting that structural and/or functional differences between individual EgAgB lipoproteins may exist. The comparison of the amino acid sequences between members of EgAgB family showed that members of the EgAgB1, EgAgB3 and EgAgB5 clades are more similar among each other than to members of the EgAgB2 and EgAgB4 clades and vice versa [18]. The polypeptides encoded by these genes are between 65 and 71 amino acids long, and have approximately 8 kDa in mass; reason by which these apolipoproteins have traditionally been called EgAgB8/1 to EgAgB8/5. Some of them were found to be capable of self-associating into homo- or hetero-oligomers of 16 and 24 kDa [20] or even into higher order homo-oligomers [21].
Although EgAgB has been studied in some detail at the protein level, very little is known concerning its lipid moiety. EgAgB was originally described as a lipoprotein on the basis that lipids were non-covalently bound to the protein component since they could be mostly removed by extraction with alcohol/ether mixtures [2]. However, the characterisation of the lipid component has not been attempted; neither has the protein/lipid stoichiometry been determined nor the class lipid composition. More recently, it has been shown that EgAgB apolipoproteins belongs to a family of hydrophobic ligand binding proteins, referred to as HLBPs, found exclusively in cestode organisms. To date, members of this family include intracellular HLBP identified in Monienza expansa [22], [23] and Hymenolepis diminuta [24], [25] as well as extracellular HLBP identified as secreted components of Taenia solium and Echinococcus granulosus [26]. All these proteins have been found to be highly abundant and immunogenic, and exists as high-molecular-mass oligomers composed by α helix-rich subunits of about 7–11 kDa. Related immunogenic proteins were also described in Taenia crassiceps and Taenia hydatigena although their lipid-binding properties have not been analysed [26]–[29]. The ligand specificity of intracellular HLBPs has been characterised in vitro [22]–[25]; they bind saturated and unsaturated fatty acids (but not their CoA-ester derivatives), retinoids, and some antihelminthic drugs, and the M. expansa protein can also bind cholesterol. The in vitro lipid binding properties of extracellular HLBPs has been partially examined by binding assays using fluorescent lipid analogues and shown to bind fatty acids only [26], [30], [31]. In the case of EgAgB, the delipidated native molecule and the recombinant EgAgB8/1 and EgAgB8/2 apolipoproteins showed ability to bind a palmitic acid fluorescent analogue with high affinity, but the possibility that these proteins could bind lipids other than fatty acids was not evaluated [30].
Cestodes have a very restricted lipid metabolism. On the one hand, lipids are not suitable substrates for energy metabolism because they cannot be oxidised due to the limited aerobic capacity of tissue-dwelling parasites (reviewed by [32] and [33]). On the other hand, cestodes are unable to synthesise fatty acids, phospholipids and cholesterol de novo. Yet, lipids are required for biosynthetic purposes, and thus, parasite lipid-binding proteins play a key role in cestode metabolism, as they are likely involved in the uptake of lipids or their precursors from the host. In this scenario, it is generally thought that EgAgB and its secreted homolog could have an important role in the biology of cestodes, controlling the acquisition and distribution of lipids to specific tissues. Alternatively, it has been proposed that HLBPs could act as messenger molecules by carrying signalling lipids which would play a role in cell activation and/or differentiation processes involved in parasite adaptation to the host immune system. In the case of EgAgB, in vitro evidence suggests that this lipoprotein may modulate host defenses by down-regulating neutrophils and dendritic cell-mediated innate responses as well as T-cell dependent mechanisms, which globally influence the intensity and quality of the adaptive immune responses [34]–[37].
The present work was aimed at identifying the native lipid moiety of EgAgB in order to complete our knowledge on the EgAgB molecular composition; this information could simultaneously shed light into EgAgB structure and function. Of particular relevance was to determine whether EgAgB binds in vivo lipid classes other than fatty acids. For that purpose, we purified EgAgB to homogeneity, using a protocol based on ion exchange chromatography coupled to immunoaffinity with a monoclonal antibody (Mo EB7), and then purified the EgAgB lipid moiety by extraction with organic solvents. Characterisation of immunopurified EgAgB of bovine origin showed that lipoprotein particles are constituted mostly by EgAgB8/1 apolipoprotein and that the native lipid moiety of this antigen comprises neutral and polar lipids that have not been previously described as ligands of this HLBP family.
Inorganic salts, 3,5-di-tert-butyl-4-hydroxytoluene (BHT), ethylenediaminetetraacetic acid (EDTA) and authentic lipid standards including cholesterol (CH), fatty acids (FA), triacylglycerols (TAGs), phosphatidylethanolamine (PE), cardiolipin (CLP), phosphatidylinositol (PI), phosphatidylserine (PS) and phosphatidylcholine (PC) were acquired from Sigma Chemicals (USA). Solvents (HPLC grade or better) and α-naphthol were purchased from Merck (Germany) or Fisher Scientific (USA).
E. granulosus HCFs from cysts containing protoscoleces of bovine origin were obtained by aspiration of the content of cysts present in lungs and livers of naturally infected cattle. Cysts were collected during the routine work of local abattoirs in Montevideo (Uruguay). E. granulosus HCFs of human origin, collected from surgically-removed hepatic hydatid cysts, were generously donated by Dr A. Leites and Dr E. Torterolo (Hospital Militar, Montevideo, Uruguay). All HCF samples were preserved by addition of 5 mM EDTA and 20 µM BHT, and maintained at −20°C. For EgAgB purification, three batches of bovine HCF (each one containing a pool of HCF from individual cysts) and two samples of individual human HCF were used.
Native EgAgB was purified from HCF following a previously described protocol [20] with slight modifications. HCF was centrifuged at 10000 g for 30 min at 4°C and the resulting supernatant filtered consecutively through 5, 2, 0.8 and 0.45 µm filter membranes (Millipore). The clarified HCF was firstly fractioned by anion exchange chromatography on a Q-Sepharose column (Pharmacia Biotech, Uppsala, Sweden) previously equilibrated in 20 mM phosphate buffer, pH 7.2 containing 200 mM NaCl, 5 mM EDTA and 20 µM BHT. After washing in equilibration buffer, the retained material was eluted by changing ionic strength to 400 mM NaCl in a single step. The eluted fraction (enriched in EgAgB and almost free of host albumin and immunoglobulins) was used to purify the antigen to homogeneity by immunoaffinity chromatography based on the utilization of a monoclonal antibody (MoAb) -named EB7- that specifically recognises the native lipoprotein [20]. For this purpose, the Q-Sepharose eluted fraction was diluted in 20 mM phosphate buffer, pH 7.2 containing 5 mM EDTA and 20 µM BHT, to reach a final concentration of 200 mM NaCl and then applied to the EB7-Sepharose column. After washing, EgAgB was eluted with 100 mM glycine-HCl, pH 3, immediately neutralised with 2 M Tris pH 9.6 and then equilibrated in 20 mM phosphate buffer, pH 7.2 containing 5 mM EDTA and 20 µM BHT (PBS-EDTA-BHT) using a PD-10 desalting column (Amershan, Biosciences). The homogeneity of EB7-affinity purified EgAgB (immunopurified EgAgB) was monitored by SDS-PAGE on 15% polyacrylamide gels followed by silver stain. In addition, samples were analysed by two-dimensional gel electrophoresis as described below.
First dimension was performed with commercially available IPG-strips (7 cm, linear 3–10, GE Healthcare). Immunopurified EgAgB was prepared and concentrated by using the 2-D Clean-Up kit (GE Healthcare) and dissolved in rehydration solution (7 M urea, 2 M thiourea, 2% CHAPS, 0.5% IPG buffer 3–10 (GE Healthcare), 0.002% bromophenol blue, DTT 17 mM). Samples in rehydration solution were loaded onto IPG-strips by passive rehydration during 12 h at room temperature. The isoelectric focusing was done in an IPGphor Unit (Pharmacia Biotech) employing the following voltage profile: constant phase of 300 V for 30 min; linear increase to 1000 V in 30 min; linear increase to 5000 V in 80 min and a final constant phase of 5000 V to reach total of 6.5 kVh. Prior running the second dimension, IPG-strips were reduced for 15 min in equilibration buffer (6 M urea, 75 mM Tris–HCl pH 8.8, 29.3% glycerol, 2% SDS, 0.002% bromophenol blue) supplemented with DTT (10 mg/ml) and subsequently alkylated for 15 min in same equilibration buffer but supplemented with iodoacetamide (25 mg/ml). The second-dimensional separation (SDS-PAGE) was performed in 15% polyacrilamyde gels using a SE 260 mini-vertical gel electrophoresis unit (GE Healthcare). The molecular size markers used were Amersham Low Molecular Weight Calibration Kit for SDS Electrophoresis (GE Healthcare). The gels were silver stained according to [38]. Images were digitalised using a UMAX Power-Look 1120 scanner and LabScan 5.0 software (GE Healthcare).
Identification of protein spots was performed by mass spectrometry (MS) using a 4800 MALDI TOF/TOF™ (AB Sciex). Briefly, peptide mass fingerprinting plus MS/MS ion search of selected spots were carried out by in-gel trypsin treatment (sequencing-grade, Promega) at 37°C, overnight. Peptides were extracted from gels using 60% acetonitrile in 0.1% TFA, concentrated by vacuum drying. Peptides were further concentrated and desalted using C18 reverse phase micro-columns (OMIX Pippete tips, Varian). Peptide elution from micro-column was performed directly into the mass spectrometer sample plate with 2 µl of matrix solution (α-cyano-4-hydroxycinnamic acid in 60% aqueous acetonitrile containing 0.1% TFA). Mass spectra of digestion mixtures were acquired in the MALDI-TOF/TOF mass spectrometer using the reflector mode. Spectra were externally calibrated using a mixture of peptide standards (Mix 1, AB Sciex). For increased confidence of identification, selected peptides were further fragmented by post-source decay (PSD) and collisional-induced dissociation (CID). Proteins were identified by NCBI nr database searching using the MASCOT program (Matrix Science http://www.matrixscience.com/search_form_select.html) and using the following search parameters: monoisotopic mass tolerance, 0.05 Da; fragment mass tolerance, 0.2 Da; carbamidomethyl cysteine and methionine oxidation as variable modifications and up to one missed tryptic cleavage allowed.
Total lipids were extracted according to the methodology described by [39], with slight modifications. Briefly, HCF (previously concentrated 10-times using a Savant SpeedVac System) or immunopurified EgAgB (1 mg protein in approximately 2 mL of PBS-EDTA-BHT) were mixed with 30 ml of a CHCl3∶CH3OH mixture (2∶1) and vigorously shaken for 2 minutes. Next, the homogenate was filtrated and the extract washed with NaCl solution to reach a final concentration of 0.73% and a CHCl3∶CH3OH∶H2O ratio of 2∶1∶0.2 in volume. After vigorous agitation for 1 minute, the separation of phases was achieved by centrifugation at 2400 rpm for 20 min and the upper phase was removed by aspiration and discarded. The lower phase containing lipids was recovered and taken to dryness by rotary evaporation at 40°C under vacuum; last traces of solvent were removed by a stream of N2 (g). Finally, the lipid fraction was dissolved in CHCl3 to a concentration of 10 mg/mL. As a control of lipid contaminants in buffers and/or solvents, an equal volume of PBS-EDTA-BHT was used in parallel for extraction. Purified lipid extracts were stored at −20°C under N2(g) until analysis.
The protein content of immunopurified EgAgB preparations was determined using bicinchoninic acid in a microtitre plate assay (BCA Protein Assay kit) with BSA as standard (Pierce, Rockford, Ill.). Total lipids in EgAgB were determined gravimetrically by weighting purified lipids immediately after extraction and solvent removal under a stream of N2(g).
Separation of lipid class components was performed by high-performance thin layer chromatography (HPTLC) on Kieselgel 60 plates (Merck). HPTLC plates (10×10 cm) were pre-washed by migration in CHCl3/CH3OH (2∶1 v/v) and then activated at 100°C for 30 min. Lipid samples (fractions obtained from HCF and EgAgB and standards) were spotted manually using a micro-syringe (Hamilton). Double development was initially carried out as follows: plates were first half-developed using a mobile phase for resolving polar lipids (PL), and after drying under a N2 (g) stream, were developed to completion in a mobile phase for neutral lipids (NL). Single development for resolving PL or NL was also performed. The solvent systems used as mobile phase were: for PL resolution- methyl acetate/isopropanol/chloroform/methanol/0.25%KCl (25∶25∶25∶10∶9, v/v/v/v/v) and for NL resolution- hexane/diethyl-ether/acetic acid (80∶20∶1, v/v/v). Lipid bands were visualised by spraying the plates with 8% (m/v) CuSO4 in a 10% (v/v) H3PO4 aqueous solution, and heating at 140°C; the identification of lipid classes was performed by comparison with primary and secondary standards run on the same HPTLC plate. The relative abundance of each lipid class respect to the total lipid content could not be estimated because not all lipids were resolved adequately in a single HPTLC using double development. We estimated the percentage of individual lipid classes in the total of NL or PL instead. 1-nonadecanol was used as internal standard for normalization. HPTLC plates were scanned and the intensity of the bands was determined using the ImageJ software (http://rsb.info.nih.gov/ij/). In addition, specific staining for sterol esters/sterols and glycolipids were carried out using FeCl3.6H2O and α-naphthol, respectively [40]. For the latter analysis a lipid fraction from murine macrophage-like J774.A1 cells was prepared to use as a control. Briefly, J774.A1 cells (generously donated by Dr. M. Noel Alvarez, Departamento de Bioquímica, Facultad de Medicina, UdelaR) were washed with PBS, lysed using a hypotonic solution (0.25 mM phosphate 3.75 mM NaCl) and centrifuged (15000×g, 4°C for 30 min) to obtain a cell membrane enriched fraction. Lipids were then extracted following the Folch method as described above.
The fatty acid composition of EgAgB and HCF lipid fractions was analysed by gas-liquid chromatography of their methyl esters derivatives (FAMEs); for these studies material of both human and bovine origin were used. Lipid fractions were subjected to acid methanolysis with 1% H2SO4 in methanol at 50°C for 16 h. The purified FAMEs were dissolved in hexane and then subjected to GLC analysis on a Hewlett-Packard 5890 equipped with a Carbowax 20 capillary column and flame ionisation detector (FID). The oven temperature was initially set at 180°C for 10 min, and then increased at 2.5°C/min to 212°C, level at which was held for 10 min. The individual FAMEs peaks were identified by comparison of their retention times with those of authentic FAMEs standards.
Light scattering analysis was carried out using a Superset 200 HR 10/30 SEC column (Amersham Biosciences, Piscataway, NJ), connected to an HPLC system (Schimatzu) at room temperature. Immunopurified EgAgB (200 µL of a 0.7 mg/mL solution in PBS) was applied onto the column previously equilibrated in PBS, and elution was monitored with on-line detection using the following detectors: multiangle laser light scattering (miniDAWN system, Wyatt Technology Corporation, Santa Barbara, CA), ultraviolet UV (SPD-20A, Shimadzu) and differential refractive index (RID-10A, Shimadzu). In addition, plasma-derived high density lipoprotein (HDL) was analysed in the same conditions for comparison. For HDL preparation, human plasma was obtained from a healthy donor and HDL purification was carried out following conventional ultracentrifugation methods [41]. A written consent was obtained from the donor according to the Ethic Committee of the Faculty of Chemistry (UdelaR) and the Executive Decree N° 379/008. Light scattering data were collected and processed with ASTRA software (v4.73.04, Wyatt Technology) using the Debye fit method with a dn/dc ratio set to 0.186 mL/g [42].
Most studies were performed with EgAgB of bovine origin since the availability of parasite material of human origin was very limited. EgAgB was purified to homogeneity from bovine HCF using a previously described procedure based on selective adsorption of this antigen on Q-Sepharose followed by immunoaffinity chromatography using immobilised MoAb EB7 [20]. This purification methodology was found to be suitable for the main objectives of this work, as it has been shown that affinity immunoadsorption permits isolation of lipoproteins under minimally perturbing conditions [43]. We characterised the apolipoprotein component of bovine EgAgB using 2-D gel electrophoresis and found that it contained two major spots electrofocused at pH 8.8 and 9.6 (Figure 1A, arrows). These two spots were identified as EgAgB8/1 by mass spectrometry analysis. Identification was mainly due to the analysis of the data arising from two main peptide signals with m/z of 1275.68 and 1399.69 that matched with the ELEEVFQLLR and YFFERDPLGQK sequences, respectively. These peptides lie in conserved regions of the amino acid sequence of EgAgB8/1. Therefore, we cannot discriminate which ones of the already described EgAgB8/1 isoforms is present in the samples. These isoforms include molecules having different theoretical isoelectric point (Figure 1B). In the case of the most basic spot, focused at pH 9.6 (Figure 1A, arrow), an additional peptide matching the MFGEVK sequence (Figure 1B, dashed line box) was observed. This sequence is present in at least four EgAgB8/1 isoforms. Interestingly, only one of them has a theoretical pI of 9.52, matching the observed value (Figure 1B, solid line box). A minor spot electrofocused at pH 7.9 was also detected (Figure 1A, head arrow), and peptide mass spectrometry of this spot indicated that it also corresponded to EgAgB8/1. When higher amounts of antigen (3-fold) were analysed, EgAgB8/4 was detected; identification was mainly based on the presence of a signal with m/z = 1152 that matched with the LGEIRDFFR sequence, which is only present in EgAgB8/4 isoform (data not shown).
Previous work suggested that EgAgB8 apolipoproteins are capable of binding fatty acids [30]. However, the fact that HCF, the physiological milieu to which EgAgB is secreted, contains a wide range of neutral and polar lipids [44], [45], opens up the possibility that the putative physiological ligands of EgAgB8 apolipoproteins include a more diverse set of lipids. For identification of these ligands, we firstly purified the lipid fractions of both, EgAgB and the HCF from which this antigen was immunopurified, using parasite material of bovine origin via the Folch extraction method. This procedure is broadly applied to the analysis of lipoproteins and has the advantage of solubilising all major lipid classes using a single solvent mixture, even though it does not allow the protein to be recovered because this fraction irreversibly precipitates during the procedure. From the dry mass of total lipids extracted and the protein concentration of the starting sample we estimated the lipid∶protein ratio (w∶w), finding that it was plainly higher for immunopurified EgAgB (between 0.6∶1 and 1.1∶1, n = 3 independent batches) than HCF (between 0.17∶1 and 0.19∶1, n = 3 independent batches). This implies that the lipid fraction of EgAgB represented approximately 40–50% of its total mass.
As an initial approach for examining the complexity of lipid classes, HCF and EgAgB lipids were analysed by HPTLC using double development. Under these conditions the majority of neutral and polar lipids are clearly separated, although the resolution of some lipid classes may not be optimal. By analysing in parallel a mixture of authentic lipid standards, we observed that the natively-bound lipid component of immunopurified EgAgB is highly heterogeneous, comprising several lipid classes, all of which are also found in the HCF. Indeed, the lipid fractions of both EgAgB and HCF showed to contain a wide variety of lipid classes from very hydrophobic ones (compatible with sterol esters and TAGs) to charged ones (phospholipids) (Figure 2A); this pattern was obtained for EgAgB and HCF samples derived from three independent batches (data not shown). Thus, these results indicated that the lipid fraction of EgAgB particles carrying EgAgB8/1 apolipoproteins consists not solely of fatty acids, but also of a variety of polar and neutral lipids present in HCF. For comparative purposes, we analysed the lipid fraction of EgAgB immunopurified from HCF of human origin (two independent batches), finding similar results in terms of lipid∶protein ratio as well as lipid class composition (data not shown).
In order to improve the identification of the lipid classes of EgAgB, the EgAgB lipid fraction was analysed by HPTLC using conditions for resolving separately neutral and polar lipids; in this analysis, 1-nonadecanol was added to all samples as an internal standard (IS) for normalization. This analysis was carried out only for HCF and EgAgB of bovine origin. As shown in Figures 2B and 2C, the lipid composition of EgAgB and HCF was very similar in terms of the variety of lipid classes. Among neutral lipids, three classes were assigned considering their mobility in comparison with standards: TAGs, free sterols and free fatty acids. In addition, three components having higher mobility than TAGs were observed in both EgAgB and HCF lipid fractions. Among these, there was a component that migrated slightly faster than FAMEs but slower than cholesteryl laureate, being compatible with dialkyl-monoacylglycerols and/or alkenyl-diacylaglycerols [40]. The other two components were not completely resolved and migrated as a wide smear in the assay conditions. The least mobile component of this smear could correspond to sterol esters according to the mobility of cholesteryl laurate. This was confirmed by using a sterol ester-specific staining method based on the formation of a purple complex with FeCl3 at acid pH (Figure 3A). The most mobile component was also observed in the control, indicating that it corresponded to a very hydrophobic contaminant derived from the extraction procedure. A second contaminant having a very low mobility in the assayed conditions was also detected in the control.
With respect to polar lipids, the analysis of EgAgB and HCF in parallel with phospholipid standards showed that phosphatidylcholine was the main phospholipid present in both samples. Phosphatidylethanolamine, phosphatidylinositol and phosphatidylserine were also detected in smaller amounts as well as traces of cardiolipin. Furthermore, minor components with higher mobility than phospholipids but lower than neutral lipids compatible with glycolipids were observed. The presence of glycolipids was confirmed by staining with α-naphthol, a dye that specifically reacts with sugar groups (Figure 3B).
The relative abundance of the major lipid classes found in immunopurified EgAgB and HCF within total neutral or polar lipids was estimated as shown in Figure 4. This estimation was carried out by analysing the intensity of HPTLC bands by densitometry, for which errors due to unequal sample application or irregular staining across the plate were normalised using the internal standard. TAGs and phosphatidylcholine corresponded to the most abundant neutral and polar lipids of EgAgB, reaching around 30% and 60%, respectively; similar percentages of these lipids were found in HCF. Moreover, the relative abundance of any lipid class within neutral and polar lipids was almost identical for EgAgB and HCF, suggesting a non-selective binding of lipids by EgAgB8/1 apolipoproteins. It is worth to mention that the relative abundance values are likely affected by two factors inherent to the method: i) the fact that the intensity of staining does not follow the same proportionality with the lipid mass for the different lipid classes, and ii) the spreading of the band affects densitometry, which means that how each lipid is resolved/focused during the chromatography could influence its quantitation.
We next analysed the fatty acid composition of total lipids extracted from human and bovine EgAgB and HCF (Figure 5). Globally, the results revealed that the fatty acid profiles of EgAgB obtained from both sources were very similar. The predominant fatty acids were 16∶0, 18∶0 and 18∶1(n-9), while 18∶2(n-6) and 20∶4(n-6) were present in a much lower proportion. Other fatty acids were present in minor quantities, representing less than 5% each. Furthermore, the content of fatty acids in EgAgB closely resembles that of the HCF used for its purification, suggesting that EgAgB8/1 apolipoproteins are capable of binding the most abundant fatty acids of HCF in a non-selective manner. Nevertheless, EgAgB8/1 apolipoproteins showed some degree of selectivity in fatty acid binding properties since the relative abundance of 13∶0, 16∶0, 18∶0 and 18∶1(n-9) were similar in bovine HCF, but not in bovine EgAgB, which contained lower percentage of 13∶0 and higher percentage of 18∶1(n-9) than bovine HCF. The comparison of the relative abundance of fatty acids in bovine vs. human HCF strongly suggests that the fatty acids are taken from the host, since 13∶0, a more abundant fatty acid in ruminants, was found in higher levels in bovine than human HCF (14 vs. 3% of total fatty acids).
Bovine EgAgB was analysed by SEC to study its physical state in aqueous solution. EgAgB eluted as a main peak centered at 12.5 mL, from which an apparent MW of 160 kDa was estimated by comparison with protein standards. This value is in agreement with previous results [1], [2]. In order to determine the absolute MW independently of hydrodynamic assumptions, we also analysed the MW of bovine EgAgB by SEC-MALLS. The molecular mass curve obtained by SEC-MALLS displayed two phases (Figure 6): an initial sharp decrease of the size at the beginning of elution (started at molecular-mass species of ∼400 kDa), rapidly followed by a plateau at a calculated mass of ∼229 kDa, which included the maximum and extended to the end of the peak. This behaviour suggests that EgAgB exists in solution as a heterogeneous population of lipoprotein particles, with most species showing an average molecular mass of 229±7 kDa. This heterogeneity could be, at least partially, associated to the capacity of EgAgB apolipoproteins to form particles accommodating variable amounts of lipids. Another possibility is that the lipoprotein could have some tendency to self-associate, leading to the formation of (less abundant) higher-order oligomers. On the other hand, the lipid/protein mass ratio, lipid composition and size of EgAgB suggested that, globally, it exhibits compositional and dimensional characteristics resembling those of plasma HDL. For comparison, human HDL was then analysed by SEC-MALLS under the same conditions. HDL showed a similar SEC-chromatographic profile to EgAgB, exhibiting molecular mass species from 150 to 400 kDa, with a mean MW of 209±4 kDa (Figure 6, inset).
This work aimed at characterising the native lipid moiety of EgAgB. As we have already mentioned, studies were mainly performed with EgAgB of bovine origin, yielding data that improve our knowledge on EgAgB lipoprotein composition. In addition, the characterisation of EgAgB protein∶lipid ratio and EgAgB lipid composition (major lipid classes and total fatty acids) was carried out with limited amounts of human EgAgB, obtaining similar results that confirm our findings.
EgAgB was purified from HCF using a previously described method based on anion exchange followed by immunoaffinity on MoAb EB7-Sepharose [20]. For a complete description of the lipoprotein composition we firstly characterised the apoliprotein component of immunopurified EgAgB of bovine origin. In addition, the identification of the apolipoprotein subunits forming EgAgB is relevant because the lipid-binding properties of EgAgB8 isoforms may differ. Native as well as recombinant EgAgB8/1 and EgAgB8/2 did not show differences in their capacity to bind fatty acid analogues in vitro [30]. Nevertheless, the physiologic lipid ligands of EgAgB8 isoforms could depend not only on the lipid-binding properties of individual isoforms, but also on the cell/tissue compartment where they are synthesised, assembled and/or transported (e.g. germinal layer vs. protoscolex). We found that EgAgB8/1 was the major apolipoprotein component of immunopurified EgAgB (Figure 1), while EgAgB8/4 was detected in much smaller amounts. In the original article describing this purification method, peptides matching sequences of EgAgB8/1, EgAgB8/2 and EgAgB8/4 were found, although those corresponding to EgAgB8/4 (SDPLGQK and LGEIR) were not initially assigned to this molecule, because its sequence was unknown. The purification of lipoproteins carrying EgAgB8/1 agrees with the fact that EgAgB8/1 is expected to be present in HCF of bovine origin [46] and MoAb EB7 strongly reacts with this isoform [20]. In contrast, MoAb EB7 does not bind to EgAgB8/2 [20] and it is unlikely that it binds to EgAgB8/4 since this isoform has much higher similarity to EgAgB8/2 than to EgAgB8/1 [18]. Therefore, purification by this method of particles carrying EgAgB8/2 and/or EgAgb8/4 apolipoproteins may result from lipoprotein particles carrying EgAgB8/1 along with EgAgB8/4 and/or EgAgB8/2, and/or associative interactions between lipoprotein particles carrying different EgAgB8 isoforms. In this regard, it is worthy to note that association of EgAgB particles could occur according to the EgAgB analysis by SEC-MALLS (Figure 6) and previous observations made by studying EgAgB sedimentation equilibrium [2]. The fact that we did not detect EgAgB8/2 in immunopurified EgAgB of bovine origin may be due to the sensitivity of the technique or to variations in the composition of distinct HCF pools. Indeed, the presence of EgAgB8/2 was detected in bovine HCF by Chemale et al. [30] but not by Aziz et al. [46] and it seems to be more variable than that of EgAgB8/1 and EgAgB8/4 among cysts of different origin/fertility [46]. Furthermore, EgAgB8/2 would not be one of the most abundant EgAgB isoforms in the HCF according to the expression levels of EgAgB family in E. granulosus metacestode [18]. Therefore, the characterisation of the apolipoprotein composition of EgAgB highlights that HCF-derived EgAgB preparations may differ depending on parasite material and these differences may be relevant when analysing EgAgB structural and/or functional properties. In any case, since EgAgB8/1 is highly expressed in the E. granulosus metacestode [18], [19], the EgAgB8/1-enriched lipoprotein that we used for lipid characterisation likely represents one of the most abundant lipoproteins of the EgAgB family present in HCF.
Analysis of the lipid composition of immunopurified EgAgB of both bovine and human origin, revealed a high diversity in terms of lipid classes ranging from highly hydrophobic lipids (mainly TAG, but also sterol esters) to a variety of phospholipids (mainly phosphatidylcholine). These findings indicate that EgAgB is a more complex lipoprotein than previously suggested from lipid-binding studies in vitro, in which fatty acids were described as the main lipid ligands of EgAgB8/1 and EgAgB8/2 [30]. Also, these results set EgAgB apart from the fatty acid binding protein family, whose members bind mainly fatty acids [47]. On the other hand, the fact that the lipid moiety of immunopurified EgAgB represented between 40 and 50% of the total mass reveals that EgAgB require to adopt a very well organised structure to accommodate a high proportion of lipid molecules in a single particle, suggesting similarities with animal lipoproteins found in both invertebrate hemolymph and vertebrate plasma [48]. The structural organization of these animal lipoproteins is well established; the most hydrophobic lipids (triacylglycerols, cholesteryl esters and other lipid-soluble components) are sequestrated in a central core, surrounded by an external hydrophilic shell that contains the apolipoproteins and amphipathic lipids (mostly phospholipids and unesterified cholesterol) (review by [49]). A structure like this could explain the heterogeneity of molecular mass species observed during analysis of immunopurified EgAgB by SEC-MALLS. Among vertebrate plasma lipoproteins, EgAgB would be more similar to the smallest HDL particles, referred to as HDL3 [48], [50], which exhibits a lipid∶protein mass ratio (w∶w) between 0.67∶1 and 1.2∶1 (0,6∶1–1.1∶1 for EgAgB) and an average molecular mass of 200 kDa (229 kDa for EgAgB); the comparative analysis of HDL and EgAgB by SEC-MALLS supports this hypothesis (Figure 6, inset). However, the content of TAG in EgAgB is much higher than in HDL3, which likely reflects differences in the lipid transport function of these lipoproteins. In addition, in the context of this structural organisation share by all plasma lipoproteins, and taken into account its size and chemical composition, each EgAgB particle would contain between 11 and 15 EgAgB8 apolipoprotein molecules inserted into the outer phospholipid monolayer. This new scenario is relevant when considering the biological effects of EgAgB through parasite's as well as host's receptors. The exposure of more than a dozen of apolipoproteins on the surface of the lipoprotein particle would facilitate the establishment of multiple interactions with receptors, increasing the avidity of the interaction and the signals derived from it. The formation of multiple EgAgB-B cell interactions likely contributes to the immunogenicity of this lipoprotein. On the other hand, it cannot be ruled out that lipids could participate, at least partially, in some of the EgAgB biological activities (for example those anti-inflammatory described in vitro). Surface charge and/or hydrophobic distribution resulting from a lipoprotein ensemble (as opposed to lipid or protein fractions alone) may alter the type of receptors involved, affecting its physiological effects.
The identification of the native lipid ligands of EgAgB provides relevant information for studying the function of this HLBP family member. Evidence for HLBP-mediated fatty acid binding and transportation across parasite membrane has been obtained in Taenia solium [26]. Our results suggest that EgAgB apolipoproteins are likely involved in solubilising and stabilising a variety of insoluble lipids in a lipoprotein particle, and that this may have an important role to deliver lipids from the tissues/sites where are synthesised/sequestered, to those that utilise or storage them. The fact that the lipids present in EgAgB are not only fatty acids, but also other essential building blocks such as sterols, highlights that EgAgB could serve a role for the E. granulosus metabolic demand of lipids. In this context is important to highlight that no enzymes for either fatty acid anabolism or squalene synthesis (the precursor of the whole family of animal sterols) have been found in the E. granulosus transcriptome (data base http://www.compsysbio.org/partigene/) or the Echinococcus multilocularis genome (unpublished observations, http://www.genedb.org/Homepage/Emultilocularis). In addition, metabolic studies have demonstrated that sterol synthesis in E. granulosus seems to stop at the level of farnesyl or nerolidol pyrophosphate and that the content of cholesterol in HCF derives from the cholesterol pool of the host [51], [52]. Thus, E. granulosus needs to take these lipids from its host. Whether EgAgB apolipoproteins are directly involved in the uptake of fatty acids and sterols from host tissues remains to be elucidated, but most likely they contribute to transport these lipids within metacestode tissues. Delivery of sterols to metacestode target cells may be crucial for biosynthetic purposes (i.e. cholesterol for biological membranes), but also for triggering signaling pathways associated with parasite development and growth. In fact, signaling pathways involving sterol-responsive nuclear receptors are conserved from simple invertebrates to mammals and regulate metabolism and development [53]. For signalling actions host sterols may be modified by the parasite; indeed, a couple of putative steroid modifying enzymes have been found in the E. granulosus trasncriptome (our unpublished observations). Interestingly, about twenty nuclear receptors have been recently identified in E. multilocularis and E. granulosus and one of them displayed structural similarities to the DAF-12 subfamily, which binds cholesterol modified compounds and regulates cholesterol homeostasis and longevity in metazoans [54]. In addition, TAG are also a major component of EgAgB. TAG are not synthesised de novo by E. granulosus, but may be synthesised from building blocks obtained from the host as it occurs in other cestodes [33]. Metabolic studies have not been performed in Echinococcus, but evidence of TAG synthesis via α-glycerophosphate–phosphatidic acid-diglyceride pathway exists for the cestode Hymenolepis diminuta [55]. TAG are the major reserve of energy in animals, but no evidence on the operation of fatty acid oxidation pathways has been obtained in flatworms [33] including Echinococcus (our unpublished observations from the E. granulosus transcriptome). This scenario suggests that TAG mainly provide a source of fatty acids and glycerol for synthetic purposes via an enzymatic hydrolysis. Consistent with this view is the identification of two ESTs clusters (EGC01304 and EGC 03444) that encode putative lipases in the E. granulosus transcriptome data base. With respect to the high content of phospholipids in EgAgB (mainly phosphatidylcholine), these molecules likely play a structural role in the lipoprotein by exposing a polar outer surface required for lipoprotein solubilization in the aqueous milieu. Phospholipids are not synthesised de novo by flatworms, but they can be synthesised from building blocks obtained from the host (fatty acids and the head group) [33].
The uptake and delivery of lipids by lipoproteins require the existence of lipoprotein receptors in target cells. The existence of parasite EgAgB receptors as well as the use of host receptors by EgAgB would be needed for EgAgB-mediated lipid traffic. In this sense it is worth to mention that in E. multilocularis and E. granulosus genomes, antigen B gene cluster is flanked by EmLDLR or EgLDLR genes, which encode proteins that display significant sequence similarities to low density lipoprotein (LDL) receptors from other species, and contain one single class A LDL receptor domain [19]. The N-terminal end of the LDL receptor contains seven successive class A domains (a cysteine-rich repeat of about 40 amino acids), which are involved in the binding of LDL as well as very low density lipoprotein (VLDL) [56]. Furthermore, domains with homology to class A LDL receptor occur in related lipoprotein receptors such as VLDL receptor as well as LDL receptor-related protein/alpha 2-macroglobulin receptor [57], [58].
Overall, a new picture for EgAgB structure and function has emerged from this work. EgAgB is a complex 229 kDa lipoprotein capable of transporting a variety of lipid classes including essential lipids that are not synthesised by the parasite, such as fatty acids and sterols. EgAgB uptake and delivery of these lipids may not only contribute to biosynthetic purposes, but also to signalling events associated with parasite metabolism and development. Whether the hydrophobic ligand binding properties of EgAgB, reflected by its lipid class composition, are an intrinsic feature of cestode HLBP family remains to be determined.
|
10.1371/journal.pgen.1007059 | A subset of sweet-sensing neurons identified by IR56d are necessary and sufficient for fatty acid taste | Fat represents a calorically potent food source that yields approximately twice the amount of energy as carbohydrates or proteins per unit of mass. The highly palatable taste of free fatty acids (FAs), one of the building blocks of fat, promotes food consumption, activates reward circuitry, and is thought to contribute to hedonic feeding underlying many metabolism-related disorders. Despite a role in the etiology of metabolic diseases, little is known about how dietary fats are detected by the gustatory system to promote feeding. Previously, we showed that a broad population of sugar-sensing taste neurons expressing Gustatory Receptor 64f (Gr64f) is required for reflexive feeding responses to both FAs and sugars. Here, we report a genetic silencing screen to identify specific populations of taste neurons that mediate fatty acid (FA) taste. We find neurons identified by expression of Ionotropic Receptor 56d (IR56d) are necessary and sufficient for reflexive feeding response to FAs. Functional imaging reveals that IR56d-expressing neurons are responsive to short- and medium-chain FAs. Silencing IR56d neurons selectively abolishes FA taste, and their activation is sufficient to drive feeding responses. Analysis of co-expression with Gr64f identifies two subpopulations of IR56d-expressing neurons. While physiological imaging reveals that both populations are responsive to FAs, IR56d/Gr64f neurons are activated by medium-chain FAs and are sufficient for reflexive feeding response to FAs. Moreover, flies can discriminate between sugar and FAs in an aversive taste memory assay, indicating that FA taste is a unique modality in Drosophila. Taken together, these findings localize FA taste within the Drosophila gustatory center and provide an opportunity to investigate discrimination between different categories of appetitive tastants.
| Fat represents a calorically potent food source that yields approximately twice the amount of energy as carbohydrates or proteins per unit of mass. Dietary lipids are comprised of both triacylglycerides and FAs, and growing evidence suggests that it is the free FAs that are detected by the gustatory system. The highly palatable taste of FAs promotes food consumption, activates reward centers in mammals, and is thought to contribute to hedonic feeding that underlies many metabolism-related disorders. Despite a role in the etiology of metabolic diseases, little is known about how dietary fats are detected by the gustatory system to promote feeding. We have identified a subset of sugar-sensing neurons in the fly that also responds to medium-chain FAs and are necessary and sufficient for behavioral response to FAs. Further, we find that despite being sensed by shared neuronal populations, flies can differentiate between the taste of sugar and FAs, fortifying the notion that FAs and sugar represent distinct taste modalities in flies.
| Fat represents a calorically potent food source that yields approximately twice the amount of energy as carbohydrates or proteins per unit of mass. In mammals, dietary lipids are detected by taste cells, mechanosensory and olfactory neurons, as well as by post-ingestive feedback [1–4]. Dietary lipids are comprised of triacylglycerides and FAs, and growing evidence suggests that it is the free FAs that are detected by the gustatory system [5–7]. Sensing of FAs promotes food consumption, activates reward circuitry, and is thought to contribute to hedonic feeding that underlies many metabolism-related disorders [8,9]. Despite a role in the etiology of metabolic diseases, little is known about how dietary fats are detected by the gustatory system to promote feeding.
In flies and mammals, tastants are sensed by dedicated gustatory receptors that localize to the taste cells or taste receptor neurons [10–12]. These cells are sensitive to different taste modalities such as sweet, bitter, salty, sour, or umami, and project to higher order brain structures for processing [10,13,14]. While these taste modalities have been extensively studied, much less is known about how FAs are detected and how this sensory stimulus is processed. Taste neurons in Drosophila are housed in gustatory sensilla located on the tarsi (feet), proboscis (mouth), and wings. Each sensillum contains dendrites of up to four gustatory receptor neurons (GRNs), which are activated by different categories of tastants [15]. Two main classes of neurons include one group that senses sweet tastants and promotes feeding, and a second, non-overlapping group that senses bitter tastants and promotes avoidance [16,17]. We previously showed that Drosophila is attracted to medium-chain FAs [18]. Consumption of FAs relies on taste, rather than smell, as it is not impaired by surgical ablation of olfactory organs [18]. Additionally, FA consumption is abolished in pox-neuro mutants in which all external taste hairs are converted to mechanosensory bristles, indicating that the chemical signature rather than oily texture of FAs is associated with perception [18]. Silencing of Gr64f-expressing neurons, which are required for sugar sensing [19,20], also abolishes behavioral responses to FAs, suggesting that shared populations of gustatory neurons detect FAs and sugars [18]. Unlike sugars, FA sensing is dependent on functional Phospholipase C (PLC), suggesting that independent intracellular molecular signaling regulates FA and sugar taste [18]. However, any further characteristics of the physiological response or the specific neuronal identity of the neurons mediating FA response are unknown.
Taste neurons from the legs and proboscis predominantly project to the subesophageal zone (SEZ), the primary taste center of the Drosophila central nervous system, but the downstream central brain circuitry contributing to taste processing is less well understood [21–24]. Determining how diverse tastants activate GRNs that convey information to the SEZ, and how this information is represented in higher order brain centers, is central to understanding the neural basis for taste processing and feeding behavior. Identifying the neural principles underlying FA taste processing requires localizing FA-responsive taste neurons and characterizing their innervation of the primary taste center. Recent studies in Drosophila have identified taste neurons that are responsive to diverse modalities including salt, sugar, amino acids, water, carbonation, bitter, polyamines, and electrophilic tastants [16,25–31], yet little is known about the populations underlying FA taste. Here, we show that GRNs identified by expression of the IR56d, which partially overlap with Gr64f-expressing neurons, are necessary and sufficient for the feeding response induced by medium-chain FAs. Our results reveal a defined population of neurons that sense FAs to promote food consumption, providing a mechanism for differentiation between attractive tastants of different modalities.
We previously reported that silencing Gr64f-expressing taste neurons abolishes behavioral responses to both sugars and FAs [10]. To directly investigate the responsiveness of these neurons to FAs, we expressed the Ca2+ sensor GCaMP5 under control of Gr64f-GAL4 [32,33] (Fig 1A). The Ca2+ responses to proboscis application of water, sucrose, or the medium-chain FA, hexanoic acid (HxA), were monitored in vivo in the axonal projections of Gr64f neurons within the SEZ (Fig 1B). Flies were provided either with 10mM sucrose or 1% HxA because these concentrations induce comparable levels of Proboscis Extension Reflex (PER) behavior. Both 10mM sucrose and 1% HxA induced activity in the SEZ, while little response was observed to water alone (Fig 1C–1F). The temporal dynamics of Ca2+ activity differed between the two tastants, with HxA eliciting a longer lasting response (Fig 1E and 1F), yet both elicited comparable peak changes in fluorescence (Fig 1C). Therefore, both sugars and FAs activate Gr64f-expressing, sweet-sensing GRNs, fortifying the notion that this neuronal class is generally responsive to attractive tastants.
To localize FA-sensitive neurons within the broad Gr64f population, we selectively silenced neuronal populations predicted to overlap with Gr64f and examined PER in response to sucrose and HxA. The synaptobrevin cleavage peptide Tetanus Toxin-Light Chain (TNT) was expressed under the control of Gustatory Receptor or Ionotropic Receptor promoters known to overlap with Gr64f (S1 Table) [19,29,34,35]. Of the 10 GAL4 lines tested, silencing with GAL4 drivers for Gr61a, IR56b, and Gr64f resulted in PER defects to sucrose and HxA. In contrast, silencing Gr64e neurons significantly reduced response to sucrose without affecting response to HxA, and silencing Gr5a, Gr43a, or IR56d neurons significantly reduced PER to HxA without affecting sucrose response.
We chose to further investigate the role of IR56d neurons in FA sensing, since IRs have been found to be involved in detection of non-sugar appetitive tastants, including salt and amino acids [27,36,37]. IR56d neurons have previously been reported to project to an SEZ region that overlaps with sweet-sensing neurons, and a second region that originates in the taste pegs of the proboscis [35]. Outside of sensing carbonation, little is known about ligands that activate the taste pegs or their role in gustation. The sweet-sensing projections of IR56d resembled the region of Gr64f projections that was activated by HxA (Fig 1).
Consistent with previous reports, expression of GFP in IR56d neurons (IR56d-GAL4>cd8::GFP) revealed two populations of projections: one innervating the posterior SEZ and previously defined as emanating from labellar bristles, and a second population emanating from the taste pegs innervating the anterior SEZ (Fig 2A–2C) [35]. To determine whether IR56d neurons are required for FA taste we silenced them with TNT and measured PER in response to FA presentation. To control for genetic background and any potential non-specific effects of TNT, we compared PER in flies with silenced IR56d neurons (IR56d-GAL4>UAS-TNT) to flies expressing an inactive variant of TNT (IR56d-GAL4>UAS-impTNT) [34]. Expression of impTNT in IR56d neurons did not affect PER in response to sucrose or HxA, while expression of TNT selectively inhibited HxA response (Fig 2D). Therefore, IR56d neurons are necessary for behavioral responses to HxA, but dispensable for responses to sucrose.
Broad activation of sweet-sensing neurons expressing the trehalose receptor Gr5a induces feeding response in the absence of tastants [16,38,39]. To determine whether activation of IR56d neurons is sufficient to induce PER, we selectively expressed the thermo-sensitive cation channel transient receptor potential A1 (TRPA1) in IR56d neurons, or Gr64f neurons as a positive control, and measured heat-induced PER [40,41]. TRPA1 induces neuronal activity at temperatures above 28°C, but has little effect on neuronal activity in flies at 22°C, allowing for thermogenetic modulation of neuronal activity [40,41]. In agreement with previous findings, broad activation of sweet-sensing neurons with Gr64f-GAL4 induced PER (Fig 2E) [26,42,43]. Similarly, PER was significantly greater upon selective activation of IR56d neurons than in control flies harboring UAS-TRPA1 or IR56d-GAL4 transgenes alone (Fig 2E). Therefore, activation of IR56d neurons alone is sufficient to induce PER.
We previously showed that PER response to HxA requires the PLC homolog no receptor potential A (norpA) in Gr64f neurons [18]. To determine whether norpA is required in IR56d-expressing neurons, we selectively expressed norpARNAi under control of IR56d-GAL4 and measured PER response to HxA and sucrose. The response to HxA was reduced in experimental flies (IR56d-GAL4>UAS-norpARNAi) using two different norpARNAi transgenes compared to controls harboring IR56d-GAL4 alone (Fig 2F). In agreement with previous findings examining norpA mutants or knock-down of norpA in all neurons expressing Gr64f, sucrose sensing was unaffected in IR56d-GAL4>UAS-norpARNAi flies, indicating that signaling through norpA in IR56d neurons is required for response to HxA, but dispensable for sucrose response.
IR56d-expressing neurons project to both taste peg and sweet-sensing regions of the SEZ, and each region can be distinguished anatomically (Fig 2A–2C; [35]). To determine whether sugars and FAs can differentially activate each region, we expressed GCaMP5 in IR56d neurons (IR56d-GAL4>GCaMP5) and measured tastant-evoked activity within anterior and posterior projections. The posterior region, which overlaps with Gr64f neurons, responded to both HxA and sucrose with similar magnitude (Fig 3A and 3B). The anterior projections from the taste pegs, however, responded to HxA, but not sucrose (Fig 3C and 3D). Therefore, the anterior and posterior projecting IR56d neurons are functionally distinguishable by their responsiveness to sucrose.
Flies exhibit feeding responses to the presentation of several FA classes [18]. It is possible that the IR56d neurons are broadly responsive to different classes of FAs. Alternatively, different classes of FAs may be sensed by independent, or partially overlapping, populations of sensory neurons. To measure the responsiveness of IR56d neurons to different classes of FAs, we first compared behavioral responses of IR56d-silenced flies to short-chain pentanoic acid (5 carbons, 5C), medium-chain octanoic acid (8C), and long-chain oleic acid (18C). As compared to control IR56d>impTNT flies, IR56d>TNT flies exhibited reduced PER to octanoic acid, but retained PER to pentanoic acid (Fig 4A), suggesting that PER to short-chain FAs is not dependent on IR56d neurons. Oleic acid did not elicit strong PER in control flies, suggesting flies do not respond to at least some long-chain FAs. We directly examined IR56d neuron responsiveness to different FAs with in vivo Ca2+ imaging in IR56d-GAL4>GCaMP5 flies. Octanoic acid activated both the posterior and anterior IR56d projections, while pentanoic acid selectively activated anterior IR56d projections (Fig 4B and 4C). Oleic acid, which did not induce PER, did not activate IR56d projections in either regions (Fig 4B and 4C). Together, these findings reveal that IR56d neurons respond to short and medium-chain FAs, and further, that sub-populations localized by SEZ projections have distinct FA response profiles.
The findings that silencing of IR56d or Gr64f neurons abolishes PER to hexanoic and octanoic acids raises the possibility that neurons co-expressing both receptors are required for FA response (Fig 5A). To validate co-expression of IR56d and Gr64f, we used the LexA system to label Gr64f neurons (Gr64f-LexA>LexAOp-CD8:GFP) and the GAL4 system to label IR56d neurons (IR56d-GAL4>UAS-RFP) (Fig 5A). Examining SEZ projections revealed co-localization within the posterior SEZ, with no co-localization detected in the anterior SEZ, suggesting the posterior IR56d neurons co-express Gr64f and IR56d. To determine whether the IR56d/Gr64f co-expressing neurons are required for FA taste, we used intersectional strategies to selectively silence anterior IR56d neurons of the taste pegs. Specifically, we repressed TNT expression in IR56d/Gr64f co-expressing neurons using Gr64f-LexA to drive expression of the GAL80 repressor (IR56d-GAL4>UAS-TNT; Gr64f-LexA>LexAop-GAL80) (Fig 5B). Selectively silencing IR56d neurons that do not overlap with Gr64f did not affect PER to HxA or sucrose compared to impTNT controls, suggesting the taste peg neurons are dispensable for the reflexive feeding response to FAs (Fig 5C). Flies lacking Gr64f-LexA, but still harboring a copy of LexAop-GAL80 (IR56D-GAL4>UAS-TNT; LexAop-GAL80/+), showed reduced PER as expected (S1 Fig).
To test if the IR56d taste peg neurons are sufficient to induce PER, we measured heat-induced PER in flies containing TRPA1 in the restricted expression pattern (IR56d-GAL4>UAS-TRPA1; Gr64f-LexA>LexAop-GAL80). Heat-induced PER was significantly reduced in flies expressing TRPA1 in only the IR56d taste peg neurons compared to control flies that lacked GAL80, and thus expressed TRPA1 in all IR56d neurons (Fig 5D). Together these results suggest IR56d neurons that do not overlap with Gr64f neurons are dispensable for PER in response to FAs.
Although both sugars and FAs induce feeding behavior, it is unclear whether flies can qualitatively differentiate between these two classes of appetitive tastants. We have developed an assay in which an appetitive tastant is applied to the tarsi, paired with a bitter quinine application to the proboscis, and the suppression of PER in subsequent responses to the appetitive tastant is then measured [39,44]. To determine whether flies can differentiate between sugars and FAs, we applied sucrose (conditioned stimulus) to the tarsi followed immediately by quinine application (unconditioned stimulus) to the proboscis. Following three training trials, memory was tested by application of either sucrose or HxA to the tarsi, in the absence of quinine, and measuring PER (Fig 6A). We then performed reciprocal experiments in which flies were trained with HxA and tested for PER to HxA or sucrose.
As previously reported, pairing sugar with quinine significantly reduced PER over the course of three training trials compared to flies offered only sugar without quinine (Fig 6B) [45,46]. This suppression of PER persisted in the testing phase where quinine is not presented (Fig 6B). On the contrary, there were no differences in PER to HxA between flies repeatedly provided sucrose in the absence of quinine and flies trained with sucrose-quinine pairing, indicating that the aversive taste memory formed to sucrose is not generalized to HxA. Conversely, the pairing of HxA and quinine resulted in PER suppression to HxA that was not generalized to sucrose (Fig 6C). We did observe a reduction in total PER response to HxA when flies had previously received sucrose tastant (Fig 6B and 6C). This occurred when prior stimulation was paired with quinine or unpaired, suggesting it is independent of memory formation and likely due to the comparative difference in salience between the two tastants. Quantification of the percentage reduction in PER revealed that the ‘matched’ groups, where quinine is paired with the tastant that is later tested, suppressed PER by 79–94%, while there was no significant PER suppression in the ‘opposite’ group, where the quinine-paired tastant and the tested tastant were different (Fig 6D). This reciprocal discrimination between sucrose and HxA is different from the unilateral discrimination between two sugars reported previously [44] and indicates that flies can discriminate between sucrose and HxA based on their identity. Thus, sugars and FAs act as independent taste modalities in flies.
Sweet-sensing neurons in Drosophila have been broadly classified as those responding to sugars and other attractive tastants such as glycerol and amino acids [16,27,47]. The findings presented here further localize the reflexive feeding response to hexanoic and octanoic acids, both medium-chain FAs, to a small population of FA-responsive taste neurons that partially overlap with sweet-sensing neurons. We have previously shown that genetic silencing of most sweet-sensing neurons using Gr64f-GAL4 abolished FA response, suggesting that these neurons detect sugars and FAs [18]. In flies, some subpopulations of Gr64f neurons are selectively sensitive to certain tastants including a Gr64e population that is sensitive to glycerol [47] and a Gr5a subset that is sensitive to trehalose [16]. To localize the Gr64f neurons responsible for FA taste, we conducted a targeted screen and silenced neurons that are believed to overlap with Gr64f neurons, which led us to study the IR56d population of neurons. Silencing IR56d neurons appears to selectively disrupt HxA response without affecting response to sucrose, supporting the notion that independent mechanisms within the Gr64f population mediate responses to sugars and FAs.
It is possible that FA-sensitive neurons are broadly tuned to FAs or selectively respond to distinct classes of FAs. Our Ca2+ imaging experiments indicate that IR56d neurons are responsive to medium-chain HxA (C6, saturated) and octanoic acid (C8, saturated) in both anterior and posterior regions if the SEZ, and to short-chain pentanoic acid (C5, saturated), but only in the anterior projections. We do not find IR56d neurons responsive to long-chain oleic acid (C18, mono-unsaturated). These findings are supported by behavioral data revealing that flies exhibit PER in response to pentanoic acid, HxA, and octanoic acid, but not oleic acid. Therefore, it seems likely that flies are strongly responsive to short/medium-chain FAs, but are less responsive to long-chain and/or unsaturated FAs. The finding that PER elicited by pentanoic acid occurs even when Ir56d neurons are genetically silenced suggests independent populations of taste neurons drive PER in response to short-chain and medium-chain FAs. Further, IR56d neurons may be activated by long-chain FAs that were not tested, and these could modulate feeding response and induce PER. Nevertheless, the findings presented here reveal specificity for medium-chain FAs within a defined population of taste neurons.
Many of the neurons identified by IR56d expression express multiple taste receptors including IR56d, Gr64f and Gr5a. These neurons likely express many additional candidate taste receptors, and future studies are needed to identify the receptor(s) that are responsive to FAs. IRs are related to ionotropic Glutamate receptors and respond to diverse tastants and odorants, making them excellent candidates for detecting FAs. [48,49]. While IR56d remains an excellent candidate, it will be necessary to examine potential IR co-receptors that may be critical for IR-dependent sensation. For example, IR25a is relatively broadly expressed and likely functions as a co-receptor for numerous IR-dependent sensory processes including temperature sensing and hygrosensation [37,50–52]. It is possible that multiple IRs are required for FA taste, with some acting as co-receptors and others detecting specific FAs. While future work is required to identify the molecular bases for FA taste, the identification of FA sensitivity in IR56d neurons provides a system to interrogate the cellular mechanisms of FA taste.
The PER response induced by two different medium-chain FAs, hexanoic and octanoic acids, suggests they may be part of Drosophila melanogaster diet. Typical dietary fats, including many plant based oils, such as coconut oil, are rich in longer-chain FAs including palmitic acid, oleic acid and linoleic acid [53]. However, medium-chain FAs are present in fermenting fruits such as guava and also in pollen [54,55]. Moreover, the medium-chain FAs (mostly C6-C10) are excreted by yeast during fermentation, possibly helping with finding yeast-rich feeding substrates, raising the possibility that flies have developed a response to FAs in order to locate suitable fermented food sources [56]. Further, we have previously shown that a diet of HxA alone is sufficient for survival [18]. Therefore, it is possible that FA attraction evolved to promote consumption of calorically rich fermenting fruits consumed by Drosophila.
The use of sucrose and HxA in an aversive taste memory paradigm reveals flies can discriminate between these attractive tastants. Sugars induce broad activation of Gr64f neurons, while the activation induced by HxA appears more restricted, and therefore it is possible that differences in activation allow for differentiation [57]. Alternatively, we find that HxA also activates anterior-projecting IR56d neurons that emanate from the taste pegs and do not co-express Gr64f, raising the possibility that differential response of these neurons to sucrose and FAs allows discrimination. Considering the different biochemical pathways involved in sugar and FA sensing [18], their identity may also be coded by unique temporal and spatial dynamics of sensory neuron activation [15,58,59]. Differences in activation are suggested to provide a mechanism for olfactory discrimination within defined neural populations, and it is possible that similar mechanisms are utilized for attractive tastants [60]. In Drosophila, attractive tastants have been found to induce a wide range of excitatory responses ranging from acute to prolonged firing [28,61], providing a potential mechanism for discrimination. While the sensillar response to FAs has not been reported, the differences in Ca2+ response to sugar or HxA presentation within the SEZ suggest differences in temporal activation.
Our findings reveal the population of IR56d neurons that innervate the anterior SEZ, which emanate from the taste pegs, are dispensable for PER in response to FAs. However, it is possible these neurons are still involved in discrimination between FAs and sugars. These neurons are not responsive to sucrose, therefore distinct anatomical activation may account for the gustatory discrimination between attractive substances. The taste pegs have previously been implicated in sensing non-sugar attractive tastants including polyamines and carbonation, raising the possibility that these neurons are responsive to multiple taste modalities [25,26]. Selectively silencing the IR56d taste peg neurons and measuring discrimination between FAs and sugars may determine whether distinct classes of IR56d neurons mediate taste feeding response and taste discrimination.
We find that flies can discriminate between sugars and FAs, but it is not known whether they can discriminate qualitatively between different classes of FAs. A previous study examining discrimination between different sugars found that flies are unable to discriminate based on quality, but could discriminate based on perceived palatability [44]. Here, we find that pentanoic acid elicits a PER response that is independent of IR56d neurons. The findings, coupled with evidence that distinct populations of neurons respond to FAs from different classes, raise the possibility that flies may discriminate between FAs based on the identity of neurons activated by each FA, or classes of FAs.
We previously reported that PLC signaling in sweet-sensing Gr64f neurons is required for FA taste [18]. Flies with mutation or knockdown of the PLC-ß ortholog norpA do not respond to HxA or octanoic acid but respond normally to sugars, revealing independent intercellular signaling mechanisms likely underlie response to FAs and sugars [18,62]. We find that knockdown of norpA in IR56d neurons abolishes FA taste without disrupting the taste of sucrose. These findings phenocopy norpA mutants and broad knockdown of norpA in Gr64f neurons, fortifying the notion that PLC signaling is selectively required for FA taste [18]. Testing the response of norpA deficient flies to FAs that are sensed independently of IR56d will inform whether PLC is more generally required for FA taste, or is only specific to medium-chain FAs detected by IR56d neurons.
While taste coding within the SEZ has been extensively investigated, less is known about the higher order circuits governing taste. Sweet-sensing neurons connect to the antennal mechanosensory and motor center (AMMC) and downstream PAM dopamine neurons that are activated by sugar [38,63]. In addition, a separate population of dopamine neurons, the PPL1 cluster, is required for olfactory appetitive memory and taste aversive conditioning [64–66]. To date, higher order neurons responsive to FA taste have not been identified. It is possible that sugar and FA taste signal through shared higher order dopamine neurons or, alternatively, each taste modality may activate distinct populations of higher order neurons that convey valence to the mushroom bodies, the memory and sensory integration center in insects [67–69].
While both sugars and FAs activate shared neurons, the ability to discriminate between these tastants provides a model for investigating sensory discrimination. There is growing evidence of multimodal coding within Drosophila sensory neurons, and in downstream targets. Flies harboring only a single functional type of olfactory receptor neurons are able to discriminate between odorants, presumably due to differences in temporal activation between the odorants [70]. Further, in the larval taste system, a single gustatory receptor neuron is responsive to both attractive and aversive compounds, and mediates the integration of these competing stimuli [71]. In addition to integration of distinct cues by the sensory system, the Drosophila mushroom bodies, and courtship circuitry integrate complex sensory cues within the brain [72,73]. Future studies on how the central brain processes sugar and FA taste will help elucidate mechanisms of discrimination between sugars and FAs.
Despite the role of FAs in promoting feeding, surprisingly little is known about how FAs promote taste in any model system. Fats contain many sensory cues and separating the taste of fat per se, from other cues such as texture, viscosity and smell is a particular challenge in mammals [74]. A number of studies have implicated the lipid binding protein CD36 as contributing to FA taste. CD36 is expressed in gustatory oral tissue, and appears to be selectively involved in FA taste [75]. CD36 knockout animals show no preference for FAs but retain preference for sugars [76]. The Drosophila homolog of CD36, Sensory neuron membrane protein 1, is expressed in the olfactory system and required for sensation of the pheromone cis-vaccenyl acetate [77], and therefore is unlikely to mediate FA taste. Additionally, a number of FA-binding GPCRs are expressed in taste cells, but their role in FA taste has not been identified. The ability to selectively manipulate and ablate defined classes of sensory neurons in Drosophila allows for the disambiguation of taste from other sensory processes [78]. Identifying FA receptors and neural circuitry mediating FA taste and discrimination will provide a framework for investigating similar processes in mammalian systems.
Taken together, this study provides insight into the coding of FAs within the fly gustatory system. Our results reveal a population of sweet-sensing neurons that are tuned for medium-chain FAs, but not short- or long-chain FAs. Further, the finding that flies are capable of discriminating between FAs and sugars suggests coding differences, either spatial or temporal neuronal activation, and provides a mechanism to distinguish between tastants of the same valence. Understanding how FAs are coded within the fly brain provides a model for understanding taste in more complex systems and will offer insight into generalizable mechanisms for taste discrimination.
Flies were grown and maintained on standard food (New Horizon Jazz Mix, Fisher Scientific). Flies were maintained in incubators (Powers Scientific; Dros52) at 25°C on a 12:12 LD cycle, with humidity set at 55–65%. The background control line used in this study is the w1118 fly strain unless otherwise noted. The following fly strains were ordered from The Bloomington Stock Center, UAS-impTNT (28840), UAS-TNT (28838), UAS-TRPA1 (26263); UAS-GFP (32186); UAS-GCaMP5 (42037); UAS-RFP/LexAop-GFP (32229); Gr43a-GAL4 (57637), Gr5a-GAL4 (57591), Gr61a-GAL4 (57658), Gr64a-GAL4 (57662), Gr64c-GAL4 (57663), Gr64d-GAL4 (57665), Gr64e-GAL4 (57666), Gr64f-GAL4 (57668), IR56b-GAL4 (60706), IR56d-GAL4 (60708), LexAop-Gal80 (32213), norpA-RNAi1 (31113), and norpA-RNAi2 (31197). Gr64f-LexA was a kind gift from H. Tanimoto and previously described in [19]. Seven to nine day old mated female flies were used for all experiments in this study, except when noted.
For all experiments, one-week-old flies were fasted for 48 hours prior to testing. For the initial screen using TNT, and specific testing of tarsal response, PER was measured by applying tastant to the tarsi, as previously described [18]. For all other PER experiments, including validation of IR56d phenotypes, tastant was applied to the proboscis to match behavioral results with functional imaging. Flies were anesthetized on CO2, mounted in a pipette tip so that their head and proboscis, but not tarsi, were exposed, and allowed to acclimate for a minimum of 30 minutes prior to testing. Flies that did not stop responding to water within 5 minutes were discarded. A small KimWipe (Kimberley Clark) wick saturated with tastant was manually applied to the tip of the proboscis for 1–2 seconds and proboscis extension reflex was monitored. Only full extensions were counted as a positive response. Each tastant was presented three times, with 10 seconds between each trial. Between tastant trials, the proboscis was washed with water and flies were allowed to drink. PER response was calculated as a percentage of proboscis extensions to total number of tastant stimulations. For experiments examining the effects of TRPA1 activation on PER, flies were mounted on a microscope slide using nail polish as described previously [18]. Flies were then placed on a heat plate heated to 34°C and video of activity was recorded for 1 minute. The number of flies for each genotype showing PER within the trial period was counted and the percentage of flies showing PER calculated.
Flies were anaesthetized on ice and placed in a small chamber with the head and proboscis accessible. A small hole was cut in tin foil and fixed to the stage leaving a window of cuticle exposed, then sealed using dental glue (Tetric EvoFlow–Ivoclar Vivadent). The proboscis was extended and a small amount of dental glue was used to secure it in place, ensuring the same position throughout the experiment.
The cuticle and connective tissue was dissected to expose the SEZ, which was bathed in artificial hemolymph (140mM NaCl, 2mM KCl, 4.5mM MgCl2, 1.5mM CaCl2, and 5mM HEPES-NaOH with pH = 7.0). Mounted flies were placed under a confocal microscope (Nikon A1) and imaged using a 25x water-dipping objective lens. The pinhole was opened to allow a thicker optical section to be monitored. Recordings were taken at 4Hz with 256 resolution. Tastants were delivered to the proboscis for 1–2 seconds with a KimWipe wick operated by micromanipulator (Narishige, GJ-1). For analysis, regions of interest were drawn manually, taking care to capture the same area between control and experimental. Baseline fluorescence was recorded over 5 frames, 10 seconds prior to tastant application. %ΔF/F was calculated for each frame as (fluorescence—baseline fluorescence)/baseline fluorescence * 100. Average fluorescence traces were created by taking the average and standard error of %ΔF/F for all samples per frame.
Fly brains were dissected in ice-cold PBS and fixed in 4% formaldehyde, PBS, 0.5% Triton-X 100 for 30 minutes. Brains were rinsed 3X with PBS, Triton-X for 10 min and incubated overnight at 4°C in NC82 (Iowa Hybridoma Bank [79]). The brains were rinsed again in PBS-TritonX, 3X for 10 minutes and placed in secondary antibodies (Donkey anti-Mouse 555; Life Technologies) for 90 minutes at room temperature. The brains were mounted in Vectashield (VectorLabs) and imaged on confocal microscope. Brains were imaged in 2um sections and are presented as the Z-stack projection through either the entire brain, or anterior and posterior regions of IR56d projections in the SEZ.
PER induction was performed in one week old mated females as described previously [5, 16]. Flies were collected and placed on fresh food for 24 hours and then fasted for 48 hours in vials containing wet KimWipe paper. Flies were later anaesthetized on CO2 pad and glued using nail polish (Cat#72180, Electron Microscopy Science) by their thorax and wing base on a microscopy slide and left to recover in a humidified box for 3-6h prior to experiments. For experiments, the slide was mounted vertically under a dissecting microscope (Olympus SZX12) during which PER was observed. Flies were satiated with water before and during the experiment. Flies that did not initially satiate within 5 minutes were excluded from conditioning. A 1ml syringe (Tuberculine, Becton Dickinson & Comp) with 200uL pipette tip attached was used for tastant presentation. We used purified water, 10mM and 1000mM sucrose, 0.4% hexanoic acid and 10mM quinine hydrochloride solutions. The protocol was adapted from [39]. Briefly, for pretest, each fly was given 10mM sucrose or 0.4% HxA on their tarsi three times with 20 second inter-trial interval and the number of full proboscis extensions was recorded. During training, the same stimulation as before was followed by placing a droplet of 10mM quinine on the extended proboscis, during which flies were allowed to drink for up to 2 seconds or until they retracted their proboscis. After each session, the tarsi and proboscis were washed with water and flies were allowed to drink to satiation. After training, flies were tested with either that same substance without quinine or with the untrained substance (matched or opposite trained groups). An independent group of flies were measured as described above but quinine was never presented (naïve groups). At the end of each experiment, flies were given 1000mM sucrose to check for retained ability of PER and non-responders were excluded [11].
Sucrose was purchased from Fisher Scientific (FS S5-500). All other tastants were purchased from Sigma Aldrich. Sucrose, hexanoic acid (SA 153745), octanoic acid (SA C2875), pentanoic acid (SA 240370) and quinine hydrochloride (SA 145904) were diluted in H20. Oleic Acid (SA O1008) was diluted in 1% DMSO (Sigma).
All statistical tests were performed in R. For normally distributed data, Welch’s t-test or ANOVA with Tukey’s post-hoc comparison were performed. For data that did not fit a normal distribution, Wilcoxon Rank-Sum or Kruskal-Wallis with Dunn’s post-hoc tests were used. Fisher’s Exact Test was used for binary categorical data. For all tests with multiple comparisons, a Bonferroni p-value adjustment was performed.
|
10.1371/journal.pgen.0040026 | Genome-Wide Expression of Azoospermia Testes Demonstrates a Specific Profile and Implicates ART3 in Genetic Susceptibility | Infertility affects about one in six couples attempting pregnancy, with the man responsible in approximately half of the cases. Because the pathophysiology underlying azoospermia is not elucidated, most male infertility is diagnosed as idiopathic. Genome-wide gene expression analyses with microarray on testis specimens from 47 non-obstructive azoospermia (NOA) and 11 obstructive azoospermia (OA) patients were performed, and 2,611 transcripts that preferentially included genes relevant to gametogenesis and reproduction according to Gene Ontology classification were found to be differentially expressed. Using a set of 945 of the 2,611 transcripts without missing data, NOA was further categorized into three classes using the non-negative matrix factorization method. Two of the three subclasses were different from the OA group in Johnsen's score, FSH level, and/or LH level, while there were no significant differences between the other subclass and the OA group. In addition, the 52 genes showing high statistical difference between NOA subclasses (p < 0.01 with Tukey's post hoc test) were subjected to allelic association analyses to identify genetic susceptibilities. After two rounds of screening, SNPs of the ADP-ribosyltransferase 3 gene (ART3) were associated with NOA with highest significance with ART3-SNP25 (rs6836703; p = 0.0025) in 442 NOA patients and 475 fertile men. Haplotypes with five SNPs were constructed, and the most common haplotype was found to be under-represented in patients (NOA 26.6% versus control 35.3%, p = 0.000073). Individuals having the most common haplotype showed an elevated level of testosterone, suggesting a protective effect of the haplotype on spermatogenesis. Thus, genome-wide gene expression analyses were used to identify genes involved in the pathogenesis of NOA, and ART3 was subsequently identified as a susceptibility gene for NOA. These findings clarify the molecular pathophysiology of NOA and suggest a novel therapeutic target in the treatment of NOA.
| Worldwide, approximately 15% of couples attempting pregnancy meet with failure. Male factors are thought to be responsible in 20%–50% of all infertility cases. Azoospermia, the absence of sperm in the ejaculate due to defects in its production or delivery is common in male infertility. In this study, we focused on non-obstructive azoospermia (NOA) because the etiologies of obstructive azoospermia are well studied and distinct from those of NOA. Microdeletions of the Y chromosome are thus far the only genetic defects known to affect human spermatogenesis, but most cases of NOA are unsolved. Because NOA results from a variety of defects in the developmental stages of spermatogenesis, the stage-specific expressions of genes in the testes must be investigated. Thus, genome-wide gene expression analyses of testes of NOA can provide insight into the several etiologies and genetic susceptibilities of NOA. In the present study, we analyzed several differentially expressed genes in NOA subclasses and identified ART3 as a susceptibility gene for NOA.
| Spermatogenesis, a major function of mammalian testes, is complex and strictly regulated. While spermatogenesis is a maturation of germ cells, other cells including Sertoli, Leydig, and peritubular myoid cells also play important roles, and defects at any differentiation stage might result in infertility. Male infertility is estimated to affect about 5% of adult human males, but 75% of the cases are diagnosed as idiopathic because the molecular mechanisms underlying the defects have not been elucidated. In consequence, an estimated one in six couples experiences difficulty in conceiving a child despite advances in assisted reproductive technologies. Male-factor infertility constitutes about half of the cases, and a significant proportion of male infertility is accompanied by idiopathic azoospermia or severe oligozoospermia, which may well have potential genetic components. It is well-recognized that men with very low sperm counts (<1 million/ml), identified through an infertility clinic, have a higher incidence of Y-chromosome microdeletion (up to 17%) [1,2]. However, the genetic causalities of most cases of azoospermia are not known.
Global gene-expression profiling with microarray technologies has been applied with great promise to monitor biological phenomena and answer biological questions. Indeed, microarray technologies have been successfully used to identify biomarkers, disease subtypes, and mechanisms of toxicity. We applied microarray analysis to testis specimens from infertile individuals including patients with obstructive azoospermia (OA) and non-obstructive azoospermia (NOA [OMIM %606766]) to characterize NOA and to identify the specific pathophysiology and molecular pathways of the disease. In addition, we attempted to identify genetic susceptibility to NOA from genes differentially expressed in NOA testes.
Genes related to spermatogenesis and candidate genes for azoospermia have been surveyed in humans and mice, especially since gene targeting technology accelerated the identification of genes that play crucial roles in spermatogenesis [3]. Because spermatogenesis is a complex process including meiosis, a germ cell–specific event, gene expression profiles specific to the differentiation stage, clinically classified by the Johnsen's score, were examined to provide insight into the pathogenesis of azoospermia [4]. In the current study, we performed microarray analyses on biopsied testes obtained from 47 NOA patients at diverse clinical stages without prior selection and 11 OA patients. The 47 NOA samples showed a wide range of heterogeneity, including a series of impairments at the differentiation stage of spermatogenesis that so far have been evaluated mainly by pathological findings. Thus, classification of NOA at the transcriptome level is a necessary first step in elucidation of the molecular pathogenesis of NOA. To do this, we adopted the non-negative matrix factorization (NMF) method, an unsupervised classification algorithm developed for decomposing images that has been applied in various fields of science including bioinformatics because of its potential for providing insight into complex relationships in large data sets [5–7]. 47 of the NOA-samples were divided into three subclasses by the NMF method, and each class was associated with clinical features. 149 transcripts were identified as differentially expressed genes among the NOA subclasses according to a statistical criterion, and the features involved in spermatogenesis based on Gene Ontology classification were demonstrated.
The genetic causality of NOA most likely involves the expression level of a susceptibility gene, which might be detected by genome-wide gene expression analysis. While it is daunting to identify genetic susceptibility from 100-1000 differentially expressed genes, genetic susceptibility might more readily be identified from random genes differentially expressed with high significance rather than by investigating only genes in a specific biological pathway. Based on a well-defined statistical procedure, 52 candidate genes for NOA were catalogued by gene expression profile and screened for allelic association study in a total of 442 NOA patients and 475 fertile male controls. After gene-centric selection of SNPs, 191 SNPs of 42 candidate genes were initially evaluated for allelic association with NOA. After two rounds of screening, SNPs of the ADP-ribosyltransferase 3 (ART3) gene were found to be significantly associated with NOA, and five of these SNPs were selected for haplotype construction. The most common haplotype was significantly under-represented in the patients and may be protective. The functional impact of this haplotype was further investigated.
As shown in Figure 1A, the most notable difference in histological findings between NOA and OA testes was that the NOA patients exhibited, at varying degrees, incomplete sets of spermatogenic germ cells (spermatogonia, spermatocytes, spermatids, and spermatozoa) in the seminiferous tubules. In severe NOA patients, we could not even detect Sertoli cells, the major somatic cells of the seminiferous tubules, on histological examination (figure not shown), indicating clinical heterogeneity of NOA testes. In order to elucidate the molecular systems underlying NOA at the transcriptome level, it is important to extract genes reflecting the diversity of NOA phenotypes. For this purpose, we first compared global gene expression profiles in NOA testes to those of OA testes using the Agilent Human 1A(v2) Oligo Microarray system. We chose the ‘standard reference design' in two-color microarray experiments as an experimental design for the expression analysis [8], where a single microarray was used to compare either NOA or OA to the testicular reference RNA as described in Materials and Methods (Figure 1B).
Of the 18,716 transcripts screened with the microarray, we obtained transcripts that showed a 2-fold mean expression difference between NOA and the reference, the NOA group; the OA group comprised transcripts showing less than 2-fold mean expression difference between OA and the reference (Figure 1B). Of the transcripts overlapping the two groups, 2,611 transcripts were found to be differentially expressed between NOA and OA testes after statistical filtering (based on lowess-normalized natural log[Cy5/Cy3], Bonferroni's corrected p < 0.05). This gene list, termed NOA-related target genes, comprised 902 elevated and 1,709 decreased transcripts in NOA testes. To characterize the gene list from the biological aspect, the 2,611 transcripts were subjected to functional clustering according to Gene Ontology (GO) classification for biological processes with GeneSpring software. We identified a total of 190 GO categories that were significantly (p < 0.05 without multiple testing correction) over-represented among the 2,611 transcripts. Table 1 shows the ten top-ranked GO categories in descending order of significance based on p-values with Fisher's exact test. It is notable that the GO categories involved in gametogenesis (GO:48232; 7283; 7276), reproduction (GO:19953; 3), and the cell cycle (GO:279; 51301; 7049; 7067) are significantly associated with the gene list. We further analyzed two separate subsets comprising 1,709 decreased (Figure 2, upper) and 902 elevated (Figure 2, lower) transcripts, based on their GO annotations. The top-ranked GO categories for NOA-related target genes are more similar to those of the 1,709 decreased transcripts than to those of the 902 elevated ones (Figure 2; Table 1). Thus, the predominant features may reflect spermatogenic defects common to NOA testes. On the other hand, 902 transcripts elevated in NOA testes exhibited a distinct GO profile that included several GO categories involved in biosynthesis and metabolism in cytoplasm (Figure 2), implying an increase in cytoplasmic turnover rates such as steroid turnover in NOA testes.
To clarify heterogeneity of NOA testes at the transcriptome level, we further examined NOA-related target genes to identify NOA subclasses without prior classification with pathological features. We adopted the NMF algorithm coupled with a model selection method [6] to a complete dataset of 945 out of the 2,611 transcripts without missing values of signal intensities for a total of 47 NOA samples. Figure 3A shows reordered consensus matrices averaging 50 connective matrices generated for subclasses K = 2, 3, 4, and 5. Distinct patterns of block partitioning were observed at models with 2 and 3 subclasses (K = 2 and 3), whereas higher ranks (K = 4 and 5) made block partitioning indistinct. Thus, the NMF method predicts the existence of reproducible and robust subclasses of NOA samples for K = 2 and 3. This prediction was quantitatively supported by higher values of cophenetic correlation coefficients (coph) for NMF-clustered matrices. The NMF class assignment for K = 3 showed the highest coph value (coph = 0.993), indicating that three molecular subclasses, termed NOA1, NOA2, and NOA3, are the most reasonable subclassification among 47 NOA samples. For comparative analysis of class discovery, a hierarchical clustering (HC) approach was applied to log-transformed normalized ratios for NOA-related target genes. As shown in Figure 3B, the HC dendrogram exhibited a clustering pattern similar to that of the NMF-based subclassification, as the three NMF-subclasses of NOA samples tended to form distinct clusters in the HC analysis. Thus, the HC clustering for NOA-related target genes appears to justify the three NMF-based subclasses of NOA samples.
To investigate the clinical features of the three NOA subclasses, we compared several clinical measures among the subclasses. The results obtained from statistical analyses in a total of four groups including the OA group are summarized in Table 2. We found significant differences in the three NOA-related clinical characteristics, testicular histological score (Johnsen's score, p = 1.4 × 10−6), serum FSH level (p = 9.8 × 10−4), and LH level (p = 0.0051) among the four groups using Kruskal-Wallis test, but there were no differences in age and serum testosterone level. Post hoc pairwise comparisons revealed that both the NOA1 and NOA2 groups exhibited low Johnsen's scores and high levels of serum FSH compared with the OA group (Table 2). In the NOA1 group, a high LH level (p < 0.01) also was found compared with the OA group. On the other hand, there were no significant differences in any of the parameters between the NOA3 and OA groups, as well as among three NOA subclasses in post hoc analysis. Elevations of serum FSH and LH concentrations often are observed in infertile patients with abnormal testicular histologies and are correlated, to some extent, with the severity of spermatogenic defects [9,10]. Testicular histologies of NOA and OA patients have been evaluated by the Johnsen's scores, ranging from 10 to 1 according to the presence or absence of spermatogenesis-related cell types (spermatozoa, spermatids, spermatocytes, spermatogonia, and Sertoli cells) in seminiferous tubules [11]. The NMF-based subclasses of testicular gene expression showed that the low score classes had heterogeneity (NOA1 and NOA2), presumably indicating the possibility of distinct spermatogenic defects at the molecular level that could not be detected by morphological examination.
Based on the three NOA subclasses, we conducted further statistical analyses to extract transcripts representing expression differences between NOA subclasses from the NOA-related target genes (Figure S1). 149 out of 2,611 transcripts showed significant differences (p < 0.05, Tukey's post hoc test) in testicular expression between the NOA subclasses, as summarized in Table S1. To characterize this gene list based on GO classification for biological processes, we examined which GO terms were highly associated with the 149 differentially expressed transcripts, relative to those for the NOA-related gene list (as shown in Table 1 and Figure 2). Figure 4 shows the 10 top-ranked GO categories for the 149 transcripts, using the 2,611 NOA-related target transcripts as a background set of genes for this GO analysis. Nine GO categories excluding gametogenesis appeared to be novel, indicating that highly significant enrichments of transcripts involved in DNA metabolism (GO:6259; 6325; 6323; 6281), chromosome organization and biogenesis (GO:51276; 7001), sex differentiation (GO: 7548), and response to endogenous stimulus (GO:9719; 6974) occurred after the extraction of 149 transcripts from the NOA-related target gene list (Figure 4). Other features of the 149 transcripts from the gene list (Table S1) were as follows: (1) a high frequency (24.2%) of sex chromosome-linked genes; (2) a high frequency (13.4%) of genes encoding cancer/testis antigens [12,13]; and (3) a moderate frequency (6.7%) of male infertility-related genes. Defect of these genes results in male infertility/subfertility in mice [3,14–16].
Twenty-five of the 149 transcripts showing differences in between-subclass expression displayed elevated expression in NOA, while the others (124 transcripts) had decreased expression (Table S1). The 25 NOA-elevated transcripts accounted only for differences in testicular expression between NOA1 and the other two subclasses, NOA2 and NOA3 (Figure S2; Table S1), suggesting testicular hyperactivity in NOA1 patients. For example, 3β-hydroxysteroid dehydrogenase, encoded by HSD3B2 and HSD3B1, plays a crucial role in biosynthesis of testosterone in Leydig cells [17]. Expression levels of the two transcripts in the NOA1 subclass were higher than those in the NOA2 and NOA3 subclasses, and the expression difference between NOA1 and NOA3 was significant by Tukey's post hoc test (Figure S2; Table S1). As the NOA1 subclass showed significantly high LH and slightly low testosterone levels (Table 2), the elevated levels of the two transcripts may be explained by a compensation process for maintaining normal testosterone level. Thus, such enhanced steroidogenesis of the NOA1 subclass might favor, even if only slightly, testicular hyperactivity in NOA1 patients.
On the other hand, among the 124 NOA-decreased transcripts, most transcripts (118/124) showed expression differences between NOA3 and the other two subclasses (Figures S2–S4; Table S1). Expression levels of these transcripts in the NOA3 subclass were similar to those in testis reference RNA (Figures S2–S4), indicating that the NOA3 subclass has a mild defect in spermatogenesis. This notion is supported by the fact that the expression of INHBB encoding inhibin β subunit B in the NOA3 subclass is normal while NOA1 and NOA2 subclasses showed low levels, indicating that inhibin β may be a marker of testicular dysfunction, as previously reported [18].
To evaluate the appropriateness of microarray data on transcripts representing expression differences between NOA subclasses, we selected 53 with high significance (p < 0.01, Tukey's post hoc test, Figure S1 and Table S1) out of the 149 differentially expressed transcripts and subjected them to real-time RT-PCR analysis. Of the 53 transcripts, the highly homologous VCX family genes, VCX, VCX2, and VCX3A, were detected with non-specific assay as a mixture of transcripts. Thus, 50 genes and one gene mixture were subjected to real-time RT-PCR. As shown in Figure S5, real-time RT-PCR data of the 51 transcripts were highly correlated with the results of microarray analysis, the squares of correlation coefficients (R2) ranging from 0.40 (CT45–2) to 0.90 (GAJ). This validation analysis also provided statistically positive evidence on between-subclass differences for all of the 51 transcripts (p < 0.05 with Kruskal-Wallis test, data not shown).
One approach to prioritizing candidate genes for genetic susceptibility underlying NOA is to adopt gene expression data from NOA tissues. Genes that show differences in expression level between NOA subclasses regardless of biological impact were selected based on the concept that polymorphic variation in gene expression among unrelated individuals is largely due to polymorphisms in DNA sequence [19,20]. 52 genes having statistical differences in expression (p < 0.01, Table S1) were regarded as candidates for allelic association with NOA. Despite the fact that these genes were not selected based on pathological relevance to NOA, genes such as SYCP3, DAZL, and INHBB, which were reported to function in spermatogenesis were included [21–23]. 191 single nucleotide polymorphisms (SNPs) of 42 genes were subjected to allelic association study with 190 NOA patients and 190 fertile men in the first round of screening. Ten genes (CTAG1B, LOC158812, LOC255313, MAGEA2, PEPP-2, TSPY1, TSPY2, VCX3A, VCY, and XAGE1) were not analyzed because no gene-based SNPs with minor allele frequency (MAF) > 0.05 could be found. We identified seven genes (ART3, LOC92196, NYD-SP20, PAGE5, TEX14, TKTL1, and XAGE5) with at least one SNP showing a discrepancy in MAF of 5% or greater between cases and controls (Table S2). Forty-four SNPs in the seven genes were subjected to a second round of screening by increasing sample size (380 NOA patients and 380 fertile men). After the two rounds of screenings, only one SNP (rs6836703) of ART3 (ADP-ribosyltransferase 3) was positively associated with NOA after Bonferroni's correction for multiple testing (Table 3; χ2 = 11.7, corrected p = 0.027).
We focused on ART3 based on the result of the two rounds of screenings, and identified 38 SNPs with MAF > 0.1 by database search or direct sequencing of the gene. 442 NOA patients (cases) and 475 fertile men (controls) were genotyped. Because we intended in this study to find a common genetic cause for NOA, patients with microdeletions of the Y chromosome at the azoospermia factor (AZF) locus, one of the major causes of NOA [1,2], were not excluded from the cases. However, to characterize the cases in regard to the AZF deletions, we examined the incidence of the deletions in a subset of the cases. Of the 442 NOA patients, 99 were examined by PCR-based screening. Fourteen (14.1 %) of the 99 cases examined showed the AZF deletions, and NOA patients with AZFc deletions were most frequent among the 14 cases (data not shown). The overall deletion frequency was comparable to those of other studies [1,2], in which the higher incidence of AZFc deletions also was observed. The clinical characteristics of patients with the AZF deletions did not differ from those of the other NOA patients (data not shown).
Linkage disequilibrium map showed that all of the SNPs of ART3 were in near complete LD evaluated with D' statistic (|D'| > 0.7) in both cases and controls (only controls are displayed in Figure 5). None of the SNPs in the controls showed deviation from Hardy-Weinberg's equilibrium at a threshold of p < 0.01 (data not shown). As shown in Table 4, SNPs showing positive associations based on nominal p-values were widely distributed throughout ART3. The most significant association was observed with ART3-SNP25 (rs6836703) located in intron 11 of ART3 (χ2 = 9.16, nominal p = 0.0025, odds ratio [95% CI] = 1.34 [1.11–1.63]). We applied the permutation method for adjustment of multiple testing to avoid a false positive result [24]. A total of four SNPs including ART3-SNP25 met the empirical significance level of p < 0.05 (Table 4).
For the haplotype-based association study, we first selected five SNPs (ART3-SNP1, 5, 23, 25, and 28) as tag SNPs captured through LD in ART3 from 15 SNPs with nominal p < 0.05 at a threshold of r2 ≥ 0.8 with Tagger software [25]. Haplotype frequencies were inferred using an expectation-maximization (EM) algorithm. After excluding rare haplotypes (frequency < 0.01), association of ART3 haplotypes with NOA was examined in 442 cases and 475 controls. Haplotype H1, the most common haplotype in controls, was under-represented in cases with significance (Figure 6; 26.6% in cases and 35.3 % in controls; χ 2 = 15.7, df = 1, nominal p = 0.000073), indicating a protective impact of haplotype H1. After Bonferroni's correction for multiple testing, a protective effect of haplotype H1 was still significant (corrected p = 0.00080). Other haplotypes showed no significant difference in frequencies between cases and controls (Figure 6). We also applied a Bayesian algorithm for phasing haplotypes with PHASE version 2.1.1 [26,27]. Regardless of haplotype-phasing methods, haplotype H1 was the most frequent in controls (26.4% in cases and 35.0% in controls), and a significant difference in haplotype H1 frequency between cases and controls was observed (permutation p < 0.0001 in global comparison, generated after 10,000 iterations).
The functional relevance of haplotype H1 in comparison with the clinical data was then explored. Diplotype was inferred with EM algorithm, and three categories (code 0, 1, and 2) were defined by the number of haplotype H1 carried without counting the other haplotypes, and nonparametric analysis of variance test with clinical data was performed. Serum levels of hormones (LH, FSH, and testosterone), other biochemical and pathophysiological markers, and Johnsen's score were analyzed by Kruskal-Wallis test with a Bonferroni/Dunn post hoc test between the three diplo-groups. Serum testosterone levels were significantly different among the three groups (Figure 7; df = 2, p = 0.0093), but there were no significant differences in other clinical markers. Post hoc pairwise comparisons revealed that serum testosterone levels were significantly higher in the subgroup having two copies of haplotype H1 than in a subgroup with one or no haplotype H1 (p = 0.0064 or p = 0.0004, respectively, Figure 7). PHASE-inferred individual diplotypes also revealed a similar correlation between diplo-groups of haplotype H1 and serum testosterone levels (data not shown).
ART3 protein expression in azoospermic testes was examined by immunohistochemical analysis. As shown in Figure 8, specific staining of ART3 protein was predominantly observed in spermatocytes in OA testes (Figure 8C–8E) as well as in normal testes from individuals of accidental sudden-death (Figure 8A and 8B). Staining was not observed in other stages of undifferentiated germ cells or Sertoli cells in the seminiferous tubules, or the interstitial tissues such as Leydig cells. On the other hand, we did not detect any ART3 protein in NOA testes with Johnsen's scores ranging from 2 to 3, which showed no spermatocytes, spermatids, or spermatozoa in the seminiferous tubules (n = 12 samples; Figure 8F–8H). There was no marked difference in testicular ART3 protein expression among the three ART3 diplo-groups carrying none, one, or two copies of haplotype H1.
Our investigation was designed to clarify the pathogenesis of NOA using global gene expression analyses of testis samples from NOA patients and to identify genetic susceptibilities underlying NOA from the genes differentially expressed. Large families with multiple generations having NOA cannot be expected due to the nature of infertility, so linkage study is impractical for NOA and has not been reported. Alternatively, allelic association study is a practical approach to identification of genetic susceptibility underlying NOA. Thus far, more than 80 genes have been identified as essential for male infertility in humans and mice [3]. Genes on the Y chromosome were emphasized because of observed microdeletions in patients, and genes such as DAZ and HSFY were examined for possible susceptibility genes [28,29]. Recently, homozygous mutation of the aurora kinase C gene was identified in large-headed multiflagellar polyploid spermatozoa, a rare form of infertility, using homozygosity mapping [30]. In the current study, we applied a novel approach to identify common susceptibility genes for NOA by applying global gene expression analysis of NOA testes. Based on the hypothesis that a common variant of a susceptibility gene has resulted in altered expression in tissues relevant to disease etiology [31], we first elucidated the gene expression profile in testes of NOA patients and characterized the genetic pathways that were either under-expressed or over-expressed. Because spermatogenesis is a complex differentiation process, NOA could result from a defect at any stage of the process. Thus, gene expression profiling of NOA tissues might well be confounded by the difficulty of discerning the differential stage and the pathological status. Feig et al. [4] examined stage-specific gene expression profiles in human NOA patients after classification on the basis of Johnsen's score. The testis tissues were classified into four groups showing Sertoli-cell only syndrome, meiotic arrest, testicular hypospermia, and testicular normospermia, corresponding to Johnsen's score 2, 5, 8, and 10, respectively, and stage-specific differential gene expression was monitored. We sought to identify susceptibility genes underlying NOA that could affect any stage of spermatogenesis. Testis samples subgrouped according to Johnsen's score in advance might identify genes affecting multiple stages of spermatogenesis. Therefore, we globally subgrouped the samples at diverse stages of differentiation using an NMF method for reducing multidimensionality that is appropriate for application to high dimensional biological data. The NMF method subgrouped three classes, NOA1, NOA2, and NOA3, which also were unequivocally subgrouped by the HC approach (Figure 3). Notably, NOA1 and NOA2 represent a pathologically similar type showing low Johnsen's score, but were subclassified because of their distinct gene expression pattern. NOA1 and NOA2 showed differences in LH, FSH, and testosterone levels, thus establishing meaningful biological significance of the sub-classes (Table 2).
In the current study, we adopted a novel approach to select candidate susceptibility genes for NOA. Global gene expression analyses were performed on NOA testes, and 52 genes were selected according to differential gene expression between NOA subclasses with a strict statistical criterion (p < 0.01 with Tukey's post hoc test). Despite the fact that our selection criteria relied only on data regarding differences in gene expression and did not include any biological assumptions, many of the genes were related to spermatogenesis based on Gene Ontology analyses (Figure 4; Table 1). 191 SNPs of 42 genes were screened, and only one gene, ART3, showed a positive association after the two rounds of screening. Multiple SNPs of ART3 were significantly associated with NOA, the most significant association being observed with ART3-SNP25 (rs6836703, nominal p = 0.0025, permutation p = 0.034; Table 4). We also detected a protective haplotype, H1, which was the most common form and was strongly associated with NOA (nominal p = 0.000073, corrected p = 0.00080, Figure 6). In addition, diplotype analysis showed that individuals carrying at least one haplotype H1 showed an elevated plasma testosterone level (Figure 7).
ART3 is a member of the mono-ADP-ribosyltransferase family genes. The biological function of ART3 remains obscure, as ART3 does not display any detectable arginine-specific transferase activity due to lack of the active site motif (R-S-EXE) that is essential for catalytic activity. Since differentiation of stage-specific expression of ART3 in testis has been reported, protein expression being exclusively present in spermatocytes but absent in spermatozoa [32], a genetic variation of ART3 might well lead to a functional defect in the process of spermatogenesis. Haplotype H1 of ART3, comprising all of the disease-protective alleles at the respective SNP sites, was under-represented in the patients. However, functional disturbance associated with haplotype H1 is so far undetermined despite the fact that several experiments designed to demonstrate haplotype-specific differences in expression level have been performed. Thus, it is possible that this haplotype represents fine tuning that maintains normal maturation of spermatocytes and improves the efficiency of spermatogenesis.
In conclusion, genome-wide gene expression analyses identified differentially expressed genes of NOA subclasses, and ART3 was identified as a susceptibility gene underlying NOA. This genetic study constitutes only first-stage evidence of association because only Japanese individuals were included, so further replication in independent case-control samples is required to confirm the role of the ART3 haplotype in genetic risk for NOA. Although further functional evidence is also required, these results provide insight into the pathoetiology of NOA as well as reproductive fitness at the molecular level, and suggest a target for therapy.
Testicular biopsy specimens for microarray analysis were obtained from 47 Japanese patients (aged from 24 to 52 years) with NOA and 11 (aged from 22 to 57 years) with OA, each of whom also underwent testicular sperm extraction (TESE) for assisted reproduction and/or diagnostic biopsy for histological examination. The biopsies for microarray analysis and histological examination were mainly sampled from unilateral, multiple testicular sites in the respective patients. Each patient was first assigned to azoospermia by showing no ejaculated spermatozoa in a semen examination. Subsequently, OA was defined as follows: (1) motile spermatozoa were sampled from microsurgical epididymal sperm aspiration (MESA), or (2) a considerable number of mature spermatozoa was sampled from TESE. NOA was tentatively defined as having no epididymal and/or testicular spermatozoa. The degree of spermatogenic defect was histologically evaluated according to Johnsen's score [11]. At least three biopsies from the same individual were taken, and the average Johnsen's scores in the NOA and OA groups ranged from 1 to 6.5 and from 5.1 to 9, respectively. In most patients, preoperative levels of serum follicle-stimulating hormone (FSH), leutenizing hormone (LH), and total testosterone were measured. The infertile male patients who visited Niigata University, Tachikawa Hospital, and St. Mother's Hospital received a routine semen examination according to 1999 WHO criteria. Based on this analysis, sperm were counted and the patients who had no ejaculated sperms were enrolled for a case-control association study. In total, 442 patients were ascertained to have NOA. In the current study, azoospermia patients with varicocele, ejaculatory dysfunction, endocrinopathy, or histologically examined OA as defined above were excluded. 475 fertile men having no specific clinical record were recruited in Niigata University. The ethics committees of Niigata University, Tachikawa Hospital, St. Mother's Hospital, and Tokai University approved the study protocols, and each participant gave written informed consent. Genomic DNA was prepared from blood white cells by Dneasy (Qiagen, Tokyo, Japan) or salivas by phenol/chloroform extraction.
To examine microdeletion of the Y chromosome in a subset of NOA patients, PCR-based diagnostic technique was used as follows: PCR amplifications with fluorescence (FAM or HEX)-labeled primers were performed to obtain fragments encompassing each of 13 STS markers in and around azoospermia factor (AZF) regions of the Y chromosome (in AZFa: SY83, SY95 and SY105; in AZFb: SY118, G65320, SY126 and SY136; in AZFc: SY148, SY149, SY152, SY283 and SY1291; in the heterochromatin distal to AZFc: SY166). Primer sequences and PCR conditions are available from the authors on request. PCR-amplified fragments were run on the ABI PRISM 3100 Genetic Analyzer (Applied Biosystems, Tokyo, Japan), and Y-chromosome microdeletion was determined with GENESCAN software (Applied Biosystems).
Total RNA from testicular biopsy was extracted using TRIzol reagent (Invitrogen, Carlsbad, CA, USA) and quantity and quality of the extracted RNA were examined with 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA, USA) using RNA 6000 Nano LabChip (Agilent Technologies). Human Testis Total RNA (BD Biosciences, San Jose, CA, USA), a histologically normal testicular RNA pooled from 39 Caucasians, was used as a common reference in two-color microarray experiments.
For fluorescent cRNA synthesis, high-quality total RNA (150 ng) was labeled with the Low RNA Input Fluorescent Linear Amplification Kit (Agilent Technologies) according to the manufacturer's instructions. In this procedure, cyanine 5-CTP (Cy5) and cyanine 3-CTP (Cy3) (PerkinElmer, Boston, MA, USA) were used to generate labeled cRNA from the extracted patient RNA and the reference RNA, respectively. Labeled cRNAs (0.75 μg each) from one patient and the common reference were combined and fragmented in a hybridization mixture with the In Situ Hybridization Kit Plus (Agilent Technologies). The mixture was hybridized for 17 hours at 65°C to the Agilent Human 1A(v2) Oligo Microarray, which carries 60-mer probes to 18,716 human transcripts. After hybridization, the microarray was washed with SSC buffer, and then scanned in Cy3 and Cy5 channels with the Agilent DNA Microarray Scanner model G2565AA (Agilent Technologies). Signal intensity per spot was generated from the scanned image with Feature Extraction Software ver7.5 (Agilent Technologies) in default setting. Spots that did not pass quality control procedures were flagged and removed for further analysis.
The Lowess (locally weighted linear regression curve fit) method was applied to normalize the ratio (Cy5/Cy3) of the signal intensities generated in each microarray with GeneSpring GX 7.3 (Agilent Technologies). Compared with the expression level of reference RNA, the NOA group, with expression undergoing a 2-fold mean change or more was extracted; the OA group comprised transcripts showing less than 2-fold mean expression change (Figure 1A). Of the transcripts included in both groups, only those with a statistically significant difference in expression between NOA and OA testes (based on lowess-normalized natural log[Cy5/Cy3], Bonferroni's corrected p < 0.05) were counted as NOA-related target genes. To elucidate the molecular subtypes of NOA, we adopted the non-negative matrix factorization (NMF) algorithm, which has been recently introduced to analysis of gene expression data [5,6]. For this analysis, a complete dataset without missing values was generated from raw values of Cy5 intensities for the NOA-related target genes in the NOA samples, and used to clarify NOA heterogeneity using three M-files (available from the following URL; http://www.broad.mit.edu/cgi-bin/cancer/publications/pub_paper.cgi?mode=view&paper_id=89) for MATLAB (Mathworks, Natick, MA, USA). According to the subclassification of NOA samples, transcripts differentially expressed between NOA subclasses were determined by one-way ANOVA, followed by Tukey's post hoc test in GeneSpring GX. For multiple test corrections in this statistical analysis, we used the Benjamini-Hochberg procedure [33] of controlling the false discovery rate (FDR) at the level of 0.05 or 0.01. To analyze which categories of Gene Ontology were statistically overrepresented among the gene lists obtained, we used GO Browser, an optional tool in GeneSpring GX, where the statistical significance was determined by Fisher's exact test. The microarray data reported in this paper have been deposited in the Gene Expression Omnibus (GEO, http://www.ncbi.nlm.nih.gov/geo/) database, and are accessible through GEO Series accession number GSE9210.
Quantitative real-time RT-PCR analysis was used to verify the microarray data on 53 transcripts representing differential expressions between NOA subclasses with high significance (p < 0.01). Among 53 transcripts, VCX (NM_013452), VCX2 (NM_016378), and VCX3A (NM_016379) were examined as a single transcript because sequence homologies between the three transcripts prevented development of appropriate assays for discrimination. Testicular total RNA (1 μg) subjected to microarray analysis was used as a template in first-strand cDNA synthesis with SuperScript III First-Strand Synthesis System (Invitrogen). Each single-stranded cDNA was diluted one-tenth for a subsequent real-time RT-PCR using SYBR Premix Ex Taq (Perfect Real Time) (TAKARA BIO, Otsu, Japan) on the ABI PRISM 7900HT Sequence Detection System (Applied Biosystems) according to the manufacturer's instructions. The PCR primers for 43 transcripts showing between-subclass differences with high significance and GAPDH were designed and synthesized by TAKARA BIO Inc., or QIAGEN GmbH (as the QuantiTect Primer Assay). In the real-time RT-PCR analysis for the nine remaining transcripts, we used TaqMan Gene Expression Assays (Applied Biosystems) with TaqMan Universal PCR Master Mix (No AmpErase UNG version) according to the manufacturer's instructions (Applied Biosystems). The detailed information on the primer sequences used and/or the assay system selected are summarized in Table S3. A relative quantification method [34] was used to measure the amounts of the respective genes in NOA testes, normalized to GAPDH as an endogenous control, and relative to Human Testis Total RNA (BD Biosciences) as a reference RNA. Statistical significance between NOA subclasses was determined by Kruskal-Wallis test, followed by multiple comparisons; p < 0.05 was considered significant.
Based on gene expression data of NOA testes, we selected 52 genes (encoding 53 transcripts) as candidates for genetic susceptibilities underlying NOA. SNPs of the candidate genes with minor allele frequency (MAF) > 0.05 were obtained from the NCBI dbSNP database (http://www.ncbi.nlm.nih.gov/SNP/), and applied to an initial screening. Of the 52 candidate genes, 10 genes (CTAG1B, LOC158812, LOC255313, MAGEA2, PEPP-2, TSPY1, TSPY2, VCX3A, VCY, and XAGE1) were excluded from the initial screening because gene-based SNPs with MAF > 0.05 were not found in the public SNP database. A total of 191 SNPs of 42 genes were genotyped in the screening with TaqMan SNP Genotyping Assays on the ABI PRISM 7900HT Sequence Detection System (Applied Biosystems). 190 NOA patients (cases) and 190 fertile men (controls) were genotyped in the first round screening. For genes with at least one SNP showing a discrepancy in MAF of 5% or greater between cases and controls, the sample size was increased to 380 cases and 380 controls in the second round.
After two rounds of initial screening, additional SNPs of ART3 were selected from dbSNP or identified by direct sequencing of all 12 exons of the gene (Ensemble transcript ID ENST00000355810) and splice acceptor and donor sites in the intron using the genomic DNAs from 95 infertile patients as PCR templates. A total of 38 SNPs of ART3 were finally genotyped on 442 cases and 475 controls by TaqMan SNP Genotyping Assays or by direct sequencing with BigDye Terminators v3.1 Cycle Sequencing Kit (Applied Biosystems) on ABI PRISM 3700 DNA analyzer.
Pairwise linkage disequilibrium (LD), using the standard definition of D' and r2 [35,36], was measured with SNPAlyze v5.0 software (DYNACOM, Mobara, Japan). To construct ART3 haplotypes in phase-unknown samples, tag SNPs of ART3 were selected with Tagger software [25], incorporated in the Haploview. The expectation-maximization (EM) algorithm [37] and PHASE version 2.1.1 [26,27] was used to infer haplotype frequencies and individual diplotypes for ART3. Differences in allelic and haplotype frequencies were evaluated using a case-control design with the chi-square test. For an adjustment of multiple testing, we applied a permutation method with Haploview version 3.32 software, or Bonferroni's method to determine corrected p-values.
To investigate association of the ART3 diplotype with clinical phenotypes such as serum hormone levels, differences among the three categories (code 0, 1, and 2), defined by the number of the most significant haplotype, were statistically examined by Kruskal-Wallis test, followed by Bonferroni/Dunn post hoc test (StatView version 5.0, SAS Institute, Cary, NC, USA).
To examine cellular localization of ART3 protein in azoospermic testes, testicular biopsy specimens from 15 OA and 12 NOA patients were subjected to immunohistochemistry. Four postmortem testicular tissues of accidental sudden-deaths were used as normal controls. The testicular tissues were fixed in 10% buffered formalin and embedded in paraffin. Cryosections (3 μm thickness) were pre-incubated with the Histofine Antigen-Retrieval Solution (1:10 dilution; Nichirei Bioscience, Tokyo, Japan) for 10 minutes at 95 °C. The sections were then incubated with primary ART3 antibody (1:4,000; Abnova, Taipei, Taiwan), then with IgG2b isotype (1:4,000; MBL International, Woburn, USA) for 60 minutes at room temperature. After washing with PBS, the sections were incubated with the Histofine Simple Stain Max-PO (Multi) (1:5 dilution; Nichirei Bioscience) for 30 minutes at room temperature, and then reacted with DAB (Nichirei Bioscience) for 10 minutes at room temperature. Haematoxylin was used for counterstaining. |
10.1371/journal.pcbi.1004756 | Oligomers of Heat-Shock Proteins: Structures That Don’t Imply Function | Most proteins must remain soluble in the cytosol in order to perform their biological functions. To protect against undesired protein aggregation, living cells maintain a population of molecular chaperones that ensure the solubility of the proteome. Here we report simulations of a lattice model of interacting proteins to understand how low concentrations of passive molecular chaperones, such as small heat-shock proteins, suppress thermodynamic instabilities in protein solutions. Given fixed concentrations of chaperones and client proteins, the solubility of the proteome can be increased by tuning the chaperone–client binding strength. Surprisingly, we find that the binding strength that optimizes solubility while preventing irreversible chaperone binding also promotes the formation of weakly bound chaperone oligomers, although the presence of these oligomers does not significantly affect the thermodynamic stability of the solution. Such oligomers are commonly observed in experiments on small heat-shock proteins, but their connection to the biological function of these chaperones has remained unclear. Our simulations suggest that this clustering may not have any essential biological function, but rather emerges as a natural side-effect of optimizing the thermodynamic stability of the proteome.
| The vast majority of living cells express molecular chaperones that suppress protein aggregation by inhibiting illicit protein–protein interactions. We refer to this class of chaperones as ‘passive molecular chaperones,’ since they do not require an external energy source in order to function. We use simulations of a minimal model of passive chaperones and aggregation-prone client proteins to show how these chaperones increase the solubility of the proteome as a whole. This anti-aggregation mechanism is surprisingly effective, even when the chaperones are expressed in very low concentrations. Most importantly, we predict that passive chaperones that are optimized to stabilize the proteome while avoiding irreversible aggregation are likely to cluster in chaperone-only oligomers. This behavior is not functional per se—that is, it is not required for these chaperones to perform their anti-aggregation function—but nevertheless emerges as a side-effect of this optimization. Our analysis thus provides an explanation for an unusual behavior that is commonly observed in experiments on passive molecular chaperones.
| Passive molecular chaperones inhibit the aggregation of cytosolic proteins and are thus a nearly ubiquitous component of living cells [1–3]. This class of chaperones comprises clusterin, α-crystallins and many other small heat-shock proteins (sHSPs), which promote tolerance to a wide range of cellular stressors such as elevated temperatures and hazardous nonspecific interactions [4, 5]. These chaperones cannot by themselves fold or refold misassembled proteins and do not require ATP to function. Instead, passive chaperones associate reversibly with aggregation-prone proteins. Even when present in sub-stoichiometric ratios with their client proteins, sHSPs and similar chaperones are effective at suppressing aggregation and coping with environmental stress [6–8]. Yet the mechanism by which this class of chaperones stabilizes the cytosol is not well understood despite significant efforts at determining the structural properties of these molecules.
Here we propose that passive chaperones function by increasing the overall solubility of the proteome. Through this mechanism, passive chaperones reduce the fraction of toxic oligomers in solution and suppress the nucleation of protein aggregates. It has recently become apparent that some sHSPs can also interact with protein aggregates in order to curtail further protein deposition [9–11]. These aggregates are often detrimental to cellular survival, in part because they can sequester other crucial proteins [12]. We provide simulation evidence that this effect on the proteome solubility is a generic feature of passive chaperones that associate promiscuously and reversibly with their clients.
There is substantial experimental evidence that passive chaperones interact promiscuously with client proteins in chemical equilibrium. Both the rate of client aggregation and the fraction of chaperones associated with insoluble proteins are concentration-dependent [1, 3]. Furthermore, chaperone binding responds directly to increases in the available client binding surfaces, including hydrophobic regions of destabilized clients that are only transiently exposed [13]. The binding of passive chaperones often modifies the size and structure of amorphous aggregates, leading to smaller soluble clusters in which the putative chaperone binding sites are associated with the hydrophobic interfaces of the client proteins [14–16]. On the basis of these dynamic chaperone–client aggregates, previous studies have suggested that such aggregates might serve as a relatively inert depot of misfolded proteins during cellular stress [2, 17–20].
However, client proteins are not the only substrates to which passive chaperones bind: these chaperones are commonly found in chaperone-only oligomers both in vitro and in vivo[7, 14–16, 21–24]. Recent experiments indicate that these dynamic oligomers are also under thermodynamic control [15, 16, 25, 26] and vary with the experimental conditions, such as the temperature and the ionic strength of the solution [25, 27, 28]. Because this tendency to form oligomers is highly conserved across the family of sHSPs and similar molecular chaperones, it has long been recognized that dynamic fluctuations in the oligomeric state play an important role in the organization of many passive chaperones [7, 25, 29, 30]. At present, however, it is unclear whether the formation of chaperone oligomers is a key functional event. In fact, there is considerable evidence to the contrary: experiments have shown that mutations and post-translational modifications that alter the tendency of chaperones to form oligomers do not necessarily affect their function [27, 31–34]. These observations raise the question of how, if at all, the presence of chaperone oligomers contributes to their ability to solubilize aggregation-prone proteins in vivo. Here we show that both the function and oligomerization of passive molecular chaperones can be explained by identifying the optimal conditions for a thermodynamically stable solution of chaperones and aggregation-prone proteins. Our results suggest that low concentrations of promiscuous chaperones are a generic means of stabilizing a biological mixture with respect to a variety of nonfunctional interactions.
To understand how passive molecular chaperones affect the thermodynamic stability of a protein solution, we consider a minimal model of two species in solution: an aggregation-prone protein and a simple molecular chaperone. Aggregation of the client proteins is primarily driven by highly directional interactions. These interactions are mediated by ‘patches,’ which represent primarily hydrophobic regions that are commonly involved in both functional and aberrant protein–protein interactions. Chaperone–client recognition is also driven by these directional associations between chaperone monomers and the exposed patches of client monomers. Both the chaperone and client proteins may also associate via weak nonspecific interactions, which we assume to be averaged over the relative orientations of the monomers. These pairwise isotropic interactions account for transient associations between proteins in a crowded environment [35, 36]. We do not explicitly model the overwhelming majority of proteins that may also experience this weak nonspecific interaction but are not prone to aggregation via directional interactions, as this simplification does not qualitatively affect our analysis.
In protein solutions under physiological conditions, the interactions between proteins are short-ranged in comparison to the size of the monomers, since the high ionic strength characteristic of physiological media leads to an effective screening of electrostatic interactions [37]. We therefore choose to model protein interactions through nearest-neighbor contacts on a three-dimensional lattice, where unoccupied lattice sites represent an implicit solvent. Monomers interact if they reside on adjacent lattice sites, and they are free to rotate and to move among lattice sites in accordance with the equilibrium Boltzmann distribution. We assume that each protein exists in a single coarse-grained conformation and that the interactions between proteins are determined by effective binding free energies (Fig 1). This coarse-graining of the internal degrees of freedom allows us to capture the effects of the intermolecular forces in a reduced set of parameters and is particularly suitable for both globular proteins in near-native states and misassembled proteins with exposed hydrophobic regions. All monomers on adjacent lattice sites experience an orientationally averaged nonspecific interaction, which is assigned a dimensionless free energy of −βϵ. (Interaction energies are expressed in thermal units: β−1 ≡ kB T, where kB is the Boltzmann constant and T is the absolute temperature.)
Because aggregation-prone proteins are likely to participate in directional protein–protein interactions via multiple binding sites [1, 38–40], which also promote interactions with sHSPs [41, 42], we choose a client model with three patches that is susceptible to aggregation by means of directional interactions alone (Fig 1). The directional interactions between client monomers are assigned an attractive free energy of −βϵs-s. These interactions are chosen to be strong enough to form insoluble client aggregates in the absence of both chaperones and additional nonspecific interactions [35].
A minimal model of a passive chaperone must be capable of binding exposed patches on the client monomers. Here we assume that the chaperone monomers have a single binding site and that the interaction free energy between chaperone and client patches is −βϵc-s (Fig 1). While this assumption is clearly a simplification of the structure of passive chaperones, which may interact with diverse clients via different binding sites, this representation captures the passivation of interactive client binding sites through the burial of hydrophobic surfaces. Most importantly, this representation has the physical features that are necessary to capture the qualitative effects of passive chaperones on the thermodynamics of a complex fluid. Because passive chaperones are known to function at low concentrations, we assume that there are always fewer chaperones than client binding sites. In what follows, the relative amounts of the chaperone and client monomers in solution are indicated by xc and xs, respectively, such that xc + xs = 1.
We have used Monte Carlo simulations and finite-size-scaling techniques to calculate the miscibility limit of this model, i.e., the point at which the chaperone–client mixture becomes unstable with respect to aggregation and/or demixing (see Methods). This miscibility limit coincides with a thermodynamic instability, where small, spontaneous fluctuations are sufficient to establish long-ranged spatial heterogeneity in an initially well-mixed solution. In a protein solution, a thermodynamic instability may have contributions from directional interactions, which cause the polymerization and demixing of the strongly interacting species, as well as orientationally averaged interactions, which drive the formation of thermodynamic phases with differing protein densities [35]. Strong directional interactions between the client proteins can thus lead to the formation of disordered aggregates and an accompanying loss of solubility.
As expected, the presence of chaperones inhibits the formation of client oligomers by competing for binding to patches on the client monomers. However, this passivation of directional interactions is not the only effect of chaperone binding: the interactions between chaperones and client proteins simultaneously increase the strength of the orientationally averaged nonspecific interactions that the solution can tolerate while remaining thermodynamically stable. This effect can be seen in Fig 2, which shows the miscibility limit, βϵ*, at which insoluble aggregates first appear in the solution. When the strength of the orientationally averaged nonspecific interactions increases beyond this limit, i.e., βϵ > βϵ*, the solution becomes unstable with respect to small fluctuations in the protein concentrations. Increasing the strength of these nonspecific interactions can thus cause the solution to become unstable without altering the strength of the directional interactions that drive the polymerization of the client monomers. Our calculations show that passive chaperones dramatically affect the miscibility limit by inhibiting polymerization and solubilizing transient clusters of client proteins, despite the fact that there are far fewer chaperones than there are client binding sites.
In the absence of chaperones, i.e., xc → 0, the solution is unstable due to the strong directional interactions between the aggregating client monomers. In this case, βϵ* is negative, indicating that a solution of sufficiently concentrated client proteins in a well-screened solvent will form insoluble aggregates. It is important to note that even when βϵc-s = 0, the chaperones still interact nonspecifically with the client monomers through the orientationally averaged interaction βϵ. Here we find that the addition of such ‘inert’ chaperones has a negligible effect on the miscibility limit relative to a client-only solution. This observation also implies that the majority of cytosolic proteins that are not aggregation-prone do not significantly affect the miscibility limit when the dominant instability is driven by strong directional interactions.
Our calculations further indicate that the thermodynamic forces driving these instabilities are qualitatively different in solutions with weakly and strongly binding chaperones. In the case of weakly binding chaperones (βϵc-s ≪ βϵs-s), the solution demixes into client-enriched and client-depleted phases primarily as a result of directional interactions. Insoluble client aggregates recruit monomers via the formation of directional contacts and exchange small oligomers with the coexisting solution. With strongly binding chaperones (βϵc-s ≫ βϵs-s), the solution forms a high-density condensate consisting of both chaperones and client proteins bound by nonspecific interactions. Under these conditions, the proteins in both the soluble and insoluble phases exist as amorphous clusters that decrease in size as the stoichiometric ratio xc/xs is increased. The introduction of strongly binding chaperones, even in low concentrations, significantly increases the solution miscibility limit towards the theoretical maximum for this model, β ϵ max * ≃ 0 . 87.
Because a chaperone that interacts with a variety of clients is very likely to engage in promiscuous interactions, it is reasonable to assume that chaperones do not distinguish among the various hydrophobic surfaces in solution. The strength of interactions between chaperone binding sites is likely to be similar to the strength of interactions between chaperones and clients, and thus it is natural to assume that βϵc-c = βϵc-s. However, if we instead prevent chaperone–chaperone binding by setting βϵc-c = 0, we find that the effect on the solution miscibility limit is negligible (Fig 2).
Since the parameter βϵc-c directly controls the probability of chaperone dimerization, our calculations suggest that the formation of chaperone oligomers has a very minor effect on chaperone function. Experimentally, the relationship between oligomerization and chaperone function has been probed by modifying or truncating sHSPs [27, 32–34]. The available experimental evidence indicates that alterations to the putative client binding sites on sHSPs affect the oligomer equilibria and the functionality of the chaperones independently, in qualitative agreement with the predictions of this model.
Putting these results into context, we now ask, “Is there an optimal chaperone–client binding strength for a biological mixture?” Fig 2 shows that strongly binding chaperones are best suited for increasing the miscibility limit. In this case, producing more chaperones (or reducing the total concentration of aggregation-prone clients) increases βϵ* in an approximately linear relationship, allowing an organism to respond effectively to an increase in nonspecific interactions. Nevertheless, strong promiscuous interactions come at a cost: nearly irreversible binding between a chaperone and any available association site, including other proteins that are not explicitly modeled in our simulations, sequesters both interfaces, thereby preventing their participation in further functional interactions. The optimal chaperone binding strength must balance these competing requirements for solution stability and reversible binding.
Despite the complexity of naturally occurring protein solutions, we can predict the optimal chaperone binding strength by considering a generic fitness function, which quantifies the trade-offs in biological costs and benefits. This fitness should be maximized for optimal biological function. For the present model, the fitness F is a function of the miscibility limit, βϵ*, as well as two biological costs that depend on the variables βϵc-s and xc. In the absence of any deleterious effects of chaperone action, increasing the solubility of the proteome must be beneficial, and thus F should be an increasing function of βϵ*. However, one potential cost of chaperone action arises from the sequestration of functional proteins (which are not explicitly modeled in our simulations but must be present in a naturally occurring protein solution) due to the promiscuous binding of chaperones. Another potential cost is associated with the production of chaperone molecules. These costs imply that the fitness function F [ β ϵ * ( β ϵ c-s , x c ) , β ϵ c-s , x c ] should satisfy both ∂ ℱ / ∂ β ϵ c-s | β ϵ * < 0 and ∂ ℱ / ∂ x c | β ϵ * < 0, respectively, when the miscibility limit βϵ* is held constant. Taking the total derivative of F with respect to both βϵc-s and xc, we find that this fitness function is maximized where
∂ β ϵ * ∂ β ϵ c-s = - ∂ F ∂ β ϵ c-s β ϵ * ∂ F ∂ β ϵ * and ∂ β ϵ * ∂ x c = - ∂ F ∂ x c β ϵ * ∂ F ∂ β ϵ * .
All partial derivatives of F depend on the precise nature of the biological system and thus cannot be determined precisely. Nevertheless, it is reasonable to assume that the cost derivatives of F are approximately constant: at low chaperone concentrations, the law of mass action implies that the cost due to reversible, promiscuous binding is approximately linear in both xc and βϵc-s, while the total cost associated with the production of chaperone molecules is also proportional to their concentration. We can therefore interpret the ratios of the derivatives in each of the above equations as the importance of each cost relative to the benefit of stabilizing the protein solution. Assuming that promiscuous chaperone binding and chaperone production are indeed significant biological costs, then these equations imply that we should seek to optimize the fitness by maximizing the response functions ∂βϵ*/∂βϵc-s and ∂βϵ*/∂xc.
More intuitively, maximizing these response functions directs the optimal chaperone design towards the region of parameter space in which the solution miscibility limit is most sensitive to small increases in either the chaperone–client binding strength or the number of chaperone molecules in solution. The first condition, ∂βϵ*/∂βϵc-s, biases the optimal chaperone design away from values of βϵc-s for which the miscibility limit increases asymptotically, thus discriminating against excessively strong binding between chaperones and clients. The second condition, ∂βϵ*/∂xc, requires that the miscibility limit be sensitive to changes in the chaperone stoichiometric fraction.
Our calculations show that it is indeed possible to satisfy both conditions simultaneously. In Fig 3a and 3b, we plot the calculated response functions ∂βϵ*/∂βϵc-s and ∂βϵ*/∂xc, in dimensionless units, as functions of the chaperone stoichiometric fraction and the chaperone binding strength. We identify a ‘design window’ for optimal chaperone operation by finding the approximate range of chaperone binding strengths over which both response functions are maximized given a fixed chaperone stoichiometric fraction. The region of parameter space in which both response functions can be maximized is relatively narrow, suggesting that optimized passive chaperones should have tightly constrained binding strengths. We further find that the optimal range of chaperone binding strengths is only weakly dependent on the chaperone stoichiometric fraction. This observation implies that chaperones with fixed binding strengths can operate close to optimality over a wide range of sub-stoichiometric concentrations. We also note that these protein–protein interaction free energies are in the physical range of a few kB T. The optimal chaperone binding strength is generally weaker than the client–client interactions, indicating that the chaperones need not out-compete the aggregation-prone clients for association with exposed binding interfaces.
Remarkably, our simulations reveal that the probability of finding chaperone oligomers is also highest in the region of parameter space where the optimal design conditions for chaperone activity are satisfied. In Fig 3c, we plot the probability of chaperone–chaperone binding at the miscibility limit, assuming that βϵc-c = βϵc-s. We find that this probability is maximal in the window of optimal chaperone binding strength over the complete range of simulated chaperone stoichiometric fractions. Under these conditions, a significant fraction of the chaperone binding sites are not associated with the aggregation-prone interfaces on the client proteins, but are rather buried in chaperone-only oligomers. This fraction may be even higher in the miscible fluid or in the presence of client proteins that are less prone to aggregation due to weaker directional interactions.
These calculations provide further evidence that the assembly of chaperone oligomers does not play a functional role. Although the choice of βϵc-c affects the magnitude of the effect shown in Fig 3c, we emphasize that simply allowing chaperone–chaperone binding does not imply that chaperone-only oligomers will be observed at the miscibility limit: there is a large region of parameter space over which this probability is very small. Furthermore, the results presented in Fig 3 are qualitatively unchanged for all reasonable choices of βϵc-c, i.e., 0 < β ϵc-c ≲ βϵc-s. Our simulations thus indicate that the ability to assemble chaperone oligomers affects neither the anti-aggregation function of the chaperones nor their adherence to the proposed design constraints.
We have shown that a simple model of chaperone–client mixtures reveals two generic and unexpected features of passive molecular chaperones. First, chaperone–chaperone interactions only marginally affect the stability of a protein solution in which strong directional interactions drive the aggregation of client proteins. Second, promiscuous passive chaperones tend to assemble chaperone oligomers under conditions where the chaperone–client binding strength balances the requirement for proteome stability with the need to avoid irreversible binding. Taken together, these results suggest that the assembly of oligomers of passive molecular chaperones is not an essential functional event for stabilizing a protein solution. Instead, this behavior emerges as a side-effect of operating under thermodynamically optimal conditions.
To arrive at this conclusion, we have proposed that passive chaperones perform their anti-aggregation function by increasing the miscibility limit of a protein solution. Through this mechanism, passive chaperones inhibit the sequestration of functional proteins and increase the thermodynamic stability of a biological mixture with respect to random nonspecific interactions. Our simulations demonstrate that this mechanism is physically plausible even when the aggregation-prone client proteins greatly outnumber the chaperones. We emphasize that only the ratio of chaperone molecules to client binding interfaces, not the total concentration of chaperones in solution, is relevant for chaperone function. In all cases considered here, the stoichiometric fraction of chaperones is much lower than xc = 0.75, the fraction that would be required to passivate all binding sites on the three-patch client monomers in the solution. The fact that the chaperones that fulfill this anti-aggregation function are highly conserved in both lower and higher organisms suggests that there is a strong evolutionary pressure to perform this role in an optimized fashion. Our calculations indicate that the range of suitable chaperone binding strengths is indeed narrow and that the principles for an optimal design emerge from thermodynamic arguments.
The generality of the present model suggests that the assembly of chaperone-only oligomers would not be affected by introducing additional detail in an off-lattice model. Such an extension, however, would allow a much wider variety of chaperone oligomers to be observed. The significant coarse-graining involved in the development of the present model and the high symmetry imposed by the lattice do not permit the reproduction of many structural features or the precise oligomeric distributions of specific passive chaperones. For instance, all three domains of αB-crystallin are believed to be involved in the assembly of higher-order oligomers [26], while the chaperones in the present model may only form dimers through client-binding interfaces. Nevertheless, such detailed molecular interactions are unlikely to affect the physical mechanism by which passive chaperones suppress aggregation.
Most importantly, the simplicity of this model allows us to make generic predictions about the thermodynamics of passive molecular chaperones. Regardless of the molecular-level details, the critical behavior of a fluid with short-ranged interactions falls within the Ising universality class, which is also known to describe phase separation in globular protein solutions [43–47]. In the vicinity of the miscibility limit, fluctuations in both the protein density and the intermolecular contacts within aggregates are significant, and a broad distribution of cluster sizes is observed at equilibrium. Our proposed mechanism therefore supports the assertion that subunit exchange is essential for the function of sHSPs and related chaperones [4, 7, 30, 48, 49]. Even if the aggregates are not fully equilibrated due to slow kinetics, the large concentration fluctuations in the vicinity of a metastable critical point are likely to enhance the formation of gel-like aggregates [50] or the nucleation of aggregated phases [51, 52]. For example, recent simulations have shown that clustering through nonspecific interactions plays an important role in the kinetics of amyloid fibril nucleation [53].
We have presented a minimal model of a mixture of passive molecular chaperones and aggregation-prone proteins. By calculating the limit of thermodynamic stability in this model protein solution, we have shown how passive chaperones that are expressed in sub-stoichiometric ratios with their clients can substantially suppress aggregation. We have further argued that the biological costs associated with chaperone production and promiscuous, irreversible binding significantly constrain the optimal design of an effective passive chaperone. We find that if passive chaperones interact promiscuously with exposed hydrophobic surfaces, then the assembly of chaperone oligomers emerges as a nonfunctional side-effect of this thermodynamically optimal design. Because of the generality of the model, these conclusions are relevant to a broad class of molecular chaperones. Fully atomistic simulations could provide further information on the parameters governing the interaction strengths between chaperones and their aggregation-prone targets as well as between the passive chaperones themselves. Such simulations could therefore provide a means of transferring the general thermodynamic principles uncovered by the coarse-grained simulations presented here to detailed models of specific chaperone–client mixtures.
In the lattice model considered here, the limit of thermodynamic stability of a well-mixed solution is encountered at the critical surface for phase separation. In what follows, we describe the Monte Carlo simulations and finite-size-scaling theory used to calculate points on this critical surface. Our approach is a generalization of the computational strategy described in detail in Ref. 35.
In general, the critical surface of a multicomponent mixture has dimension d − 2, where d is the total number of independent thermodynamic fields [54]. The independent thermodynamic fields in the present model are the dimensionless chemical potentials of both the chaperones and the clients, βμc and βμs, respectively, as well as the dimensionless interaction energies: βϵ, βϵs-s, βϵc-s and βϵc-c. The relevant critical surface in this model is thus a 4-dimensional surface.
We perform biased grand-canonical Monte Carlo simulations, as described in Ref. 35, to collect statistically independent lattice configurations near the critical surface. We use a L × L × L cubic lattice with periodic boundary conditions and set L = 12 so that all simulations are carried out in the scaling regime. We then apply the finite-size-scaling theory of Wilding and Bruce [55, 56] to solve self-consistently for the critical order parameter, M ^, and the critical orientationally averaged nonspecific energy, βϵ*, at fixed values of βϵs-s, βϵc-s, βϵc-c and xc. In order to determine each critical point plotted in Fig 2, we approximate the marginal probability distribution p ( M ) from the grand-canonical samples and then tune this distribution in order to match the known distribution of the critical ordering operator in the three-dimensional Ising universality class, p M. This computational procedure is described below.
In a two-solute solution, with two independent dimensionless chemical potentials βμc and βμs, the critical order parameter must account for fluctuations in the number densities of both the client and chaperone monomers, ρs and ρc, respectively, as well as fluctuations in the internal energy density, u. The critical fluctuations in the number densities can be described by the vector ν ^, which indicates the difference in compositions of the two incipient phases [36]. We therefore define M ^ to be the linear combination
M ^ ≡ ν s ρ s ^ + ν c ρ c ^ - s u ^ , (1)
where both ν ^ and the field-mixing parameter s must be determined self-consistently. The grand-canonical distribution of M is constructed from the simulation data according to
p gc , k ( M ) ≡ Λ ∑ v w v 1 δ M k ≤ ( ρ s , ρ c , u ) v · M ^ < δ M k + 1 , (2)
where the index v runs over all independent samples and 1{⋅} is the indicator function. Each sample has a statistical weight wv in the grand-canonical ensemble that depends on the values of the thermodynamic fields [35]. The system-dependent scaling constant Λ must be determined self-consistently. The bin size is chosen such that ( δ M k + 1 - δ M k ) = L - 3, where δ M ≡ Λ ( M - M * ) and M * is the ensemble-averaged mean value of M.
We then construct a χ2-function that seeks to minimize the difference between the observed distribution of M and the universal distribution, p M, while obeying the imposed composition constraint:
χ 2 ≡ ∑ k p gc , k ( M ) ( β f → ) - p M ( δ M k / Λ ) 2 σ k 2 + ∑ i ∈ { s , c } ρ i ( β f → ) / ∑ j ∈ { s , c } ρ j ( β f → ) - x i 2 σ i 2 , (3)
where β f → ≡ ( β ϵ , β ϵ s-s , β ϵ c-s , β ϵ c-c , β μ s , β μ c ) and the index k runs over all bins. In the second term, 〈ρi〉 indicates the ensemble-averaged number density of component i. We estimate the error in the sampled distribution of M to be
σ k 2 = ∑ v w v 2 1 k , v - ∑ v w v 1 k , v 2 / n samples ∑ v w v , (4)
where 1k,v is the indicator function written out explicitly in Eq (2), and we estimate the error in the observed composition at the critical point to be
σ i 2 = 1 ϕ 2 ∑ j , k ∈ { s , c } δ i j - ρ i ϕ 〈 δ ρ j δ ρ k 〉 δ i k - ρ i ϕ , (5)
where 〈δρj δρk〉 ≡ 〈ρj ρk〉 − 〈ρj〉〈ρk〉, ϕ ≡ ∑j∈{s,c} ρj and δij is the Kronecker delta.
Finally, we calculate the probability of chaperone dimerization, 〈pc-c〉, directly from the simulation data according to the definition
〈 p c-c 〉 * ≡ 2 n cc N c * , (6)
where ncc is the number of chaperone–chaperone patch contacts and Nc is the total number of chaperone monomers on the lattice. In this definition, 〈⋅〉* indicates a grand-canonical average obtained at the critical point with the specified chemical potentials and directional interaction energies.
|
10.1371/journal.pcbi.1000142 | Evolutionarily Conserved Substrate Substructures for Automated Annotation of Enzyme Superfamilies | The evolution of enzymes affects how well a species can adapt to new environmental conditions. During enzyme evolution, certain aspects of molecular function are conserved while other aspects can vary. Aspects of function that are more difficult to change or that need to be reused in multiple contexts are often conserved, while those that vary may indicate functions that are more easily changed or that are no longer required. In analogy to the study of conservation patterns in enzyme sequences and structures, we have examined the patterns of conservation and variation in enzyme function by analyzing graph isomorphisms among enzyme substrates of a large number of enzyme superfamilies. This systematic analysis of substrate substructures establishes the conservation patterns that typify individual superfamilies. Specifically, we determined the chemical substructures that are conserved among all known substrates of a superfamily and the substructures that are reacting in these substrates and then examined the relationship between the two. Across the 42 superfamilies that were analyzed, substantial variation was found in how much of the conserved substructure is reacting, suggesting that superfamilies may not be easily grouped into discrete and separable categories. Instead, our results suggest that many superfamilies may need to be treated individually for analyses of evolution, function prediction, and guiding enzyme engineering strategies. Annotating superfamilies with these conserved and reacting substructure patterns provides information that is orthogonal to information provided by studies of conservation in superfamily sequences and structures, thereby improving the precision with which we can predict the functions of enzymes of unknown function and direct studies in enzyme engineering. Because the method is automated, it is suitable for large-scale characterization and comparison of fundamental functional capabilities of both characterized and uncharacterized enzyme superfamilies.
| Enzymes are biological molecules essential for catalyzing the chemical reactions in living systems, allowing organisms to convert nutrients into usable forms and convert harmful or unneeded molecules into forms that can be reused or excreted. During enzyme evolution, enzymes maintain the ability to perform some aspects of their function while other aspects change to accommodate changing environmental conditions. In analogy to studies of enzyme evolution focused on conservation of sequence and structural motifs, we have examined a large number of enzyme superfamilies using a new computational analysis of patterns of substrate conservation. The results provide a more nuanced picture of enzyme evolution than obtained either by detailed small-scale studies or by large-scale studies that have provided only general descriptions of function and substrate similarity. The superfamilies in our set fall along the entire spectrum from the conserved substructure being mostly reacting to mostly nonreacting, with most superfamilies falling in the intermediate range. This view of enzyme evolution suggests more complex patterns of functional divergence than those that have been proposed by previous theories of enzyme evolution. The method has been automated to facilitate large-scale annotation of enzymes discovered in sequencing and structural genomics projects.
| The molecular functions of enzymes result from a complex evolutionary interplay between environmental constraints, requirements for organismal fitness, and the functional malleability of a particular enzyme scaffold. Within these constraints, existing enzymes are recruited during evolution to perform new or modified functions while often maintaining some aspects of the ancestral function [1]–[3]. Consequently, among contemporary enzymes we observe groups of evolutionarily related enzymes that share some aspects of molecular function and differ in others. The most divergent groups of evolutionarily related enzymes that still share aspects of function are called superfamilies. Within a superfamily, we define a family as a set of proteins that perform the same overall catalytic reaction in the same way. Why are some aspects of function shared and others allowed to change? By examining which aspects of function are shared among contemporary enzymes, we can gain insight into the requirements and constraints that govern this evolutionary process.
The focus of most studies of enzyme evolution has been the examination of conservation in sequence and structure. The data available to conduct such studies is enormous and still increasing due to the multiplicity of ongoing genomic and metagenomic sequencing efforts [4]. In tandem with the growth of sequence and structural data, a large number of new and sophisticated tools have been developed to improve our ability to identify the divergent members of superfamilies, allowing us to analyze patterns of conservation in sequence and structure that shed light on how enzyme functions have evolved and diversified (for some examples, see [5]–[7]). But such studies only capture aspects of enzyme evolution that can be inferred from the machinery that enables enzymatic catalysis, the enzymes themselves. Far fewer studies have focused on the substrates and products of these reactions, with most of these focused on the requirements of metabolism [8],[9]. In this work, our goal is to understand the details of how enzymes function and evolve by studying the conservation and variation in their substrates and products. In doing so, we aim for a more extensive view of enzyme evolution in order to improve our abilities to annotate enzymes of unknown function and to infer common aspects of function for superfamilies that have not yet been characterized.
The value of any analysis of the evolution of enzyme function depends on how we describe enzyme function, with respect to both the detailed molecular functions of individual enzymes and the properties of function shared across diverse members of enzyme superfamilies. Previous approaches to study enzyme evolution range from detailed manual analyses of small numbers of related enzyme families and superfamilies to automated analyses of many superfamilies. The former have often included not only analyses of sequences and structures but also comparisons of the substrates and reaction mechanisms of the constituent enzymes. These studies have been useful for annotating new sequences and structures and for generating and testing hypotheses about patterns of enzyme evolution (see [10]–[14] for examples). However, because of the expert knowledge required and their time-intensive nature, these types of analyses are not feasible for large numbers of superfamilies. Other semi-automated efforts have contributed to our understanding of enzyme evolution and data from these analyses have been made available in a number of online resources that include the Structure-Function Linkage Database [15], MACiE [16], the Catalytic Site Atlas [17], and EzCatDB [18]. Automated analyses more directly comparable to the large-scale and automated study described here [19]–[21] have used enzyme classification systems, like the Enzyme Commission (EC) system [22], to represent functional properties and determine what properties are conserved. The EC system represents a large proportion of known enzyme reactions, classifying each enzyme with a hierarchical set of four numbers that uniquely identify a reaction, and is easy to use for large-scale analyses. However, this system, developed before analyses of enzyme evolution were common, does not provide a detailed description of enzyme function or substrates at the atomic level [23]. Moreover, the EC classification of function often does not correspond with either the aspects of function that are conserved or those that can change during evolution. These issues make this system unsuitable for evaluating how enzyme function evolves, especially when evolutionary relationships are distant [24]. For enzymes, the Gene Ontology (GO) system's [25] molecular function classifications, also often used to describe and analyze function, largely recapitulate the EC system. More similar to the work reported here, several groups have analyzed enzyme relationships and evolution using substrate and reaction similarities [26]–[28]. Although these similarity metrics are useful, especially for clustering enzymes by their substrate similarities, they are not informative about what specific aspects of function are conserved, a specific goal of this work.
Here, we use graph isomorphism analyses to compare substrates of enzymes from 42 superfamilies to identify specific aspects of function conserved within each superfamily. We also use comparisons of substrates and their corresponding products to determine whether and how much of the conserved substructure is involved in the reaction. This comparison of substrates and products is similar to an analysis performed for a previous study with a different purpose, to predict EC numbers [29]. To simplify the interpretation of results across the multiple superfamilies in this study, only enzymes comprised of single domains and that catalyze unimolecular reactions were investigated. Automation of the analysis allows us to describe overall trends in functional conservation and variation across a large number of superfamilies. A descriptive representation of conserved enzyme molecular functions using chemical structures and SMILES strings [30],[31] is also provided. This representation should be useful for annotating new members of superfamilies discovered in sequencing projects and for characterizing new superfamilies.
Results are presented for 42 superfamilies from the Structural Classification of Proteins (SCOP) database [32]. These superfamilies meet the following criteria: (1) they consist of only single-domain enzymes that (2) perform only unimolecular reactions (or reactions with two substrates, of which one is water), and (3) the superfamilies include at least two different reactions (representing at least two different E.C. numbers) for which substrate and product information are available in the enzyme database BRENDA [33]. Sufficient data were available in BRENDA (the third criterion) for 46.2% of the superfamilies meeting the first two criteria. These 42 superfamilies include representatives of six of the seven SCOP fold classes; the only fold class not represented is the membrane proteins class. The enzymes in these 42 superfamilies represent a substantial proportion of the diversity of enzyme function, covering 25.4% of EC classes defined by the first two digits (subclasses) and 18.7% of EC classes defined by the first three digits (sub-subclasses). Conservation patterns were examined using only substrates and products as the data available in BRENDA were not sufficient to consider other aspects of reaction conservation, such as transition states and intermediates.
Our goal was to determine the molecular features that the substrates of a superfamily share and whether the shared features are involved in the reactions catalyzed by that superfamily. Thus, for each superfamily, we identified the conserved substructure, defined as the set of bonds and their connected atoms that are present in all substrates of the superfamily (Figure 1A). These conserved substructures for the 42 superfamilies in our dataset are shown in Figure 2. Additional information about the diversity and conservation of functions in these superfamilies is provided in a hyperlinked table in the supplementary information online (Table S1). Moreover, for each enzyme's substrate(s), we found the reacting substructure by determining what atoms and bonds change between the substrate and the product (Figure 1B). For each enzyme, we then determined whether the conserved substructure overlaps with the reacting substructure and by how much. This overlap was quantified by calculating the fraction of the conserved substructure that is reacting (fc) (Figure 1C, Table S2) and the fraction of the reacting substructure that is conserved (fr) (Figure 1D, Table S2). Results for these measures of overlap are presented with respect to both the number of atoms and the number of bonds.
For a given superfamily, the average fc and fr calculated using atoms often differ from the values obtained using bonds (Table S2). This difference arises because the number of bonds is frequently not proportional to the number of atoms in molecular structures (e.g., one bond consists of two atoms while three atoms can be connected by three bonds; a cyclic structure will have a different number of bonds compared to non-cyclic structure with the same number of atoms). In addition, different types of reactions vary in the ratio of atoms and bonds that are involved in the reaction (e.g., a lyase may break one bond involving two atoms while an intramolecular transferase may involve one bond and three atoms). Because both are valid measures of substructure size, both are provided in this report.
The distribution of average fc for the set of superfamilies (Figure 3A) indicates that there is a continuum among the superfamilies in how much of the conserved substructure is reacting, with superfamilies ranging from having little to having most of the conserved substructure participating in the reaction. This trend is observed regardless of whether we use atoms or bonds in our calculations of average fc. The results also show that all superfamilies with a conserved substructure have an average fc above zero, indicating that at least part of the conserved substructure is involved in the reaction.
Only one superfamily in our study set, the superfamily defined by SCOP as the metallo-dependent hydrolase superfamily, also known as the amidohydrolase superfamily [34],[35], has substrates so diverse that they do not share a common substructure of even a single conserved bond. Detailed analysis of the superfamily, including analysis of differences in the overall functions, how active site motifs are used for catalysis, and other factors such as metal ion dependence, suggests that this group may be more properly considered as multiple superfamilies (Brown and Babbitt, in preparation).
Plotting fr against fc illustrates the distribution of superfamilies (Figure 3B) across different patterns of overlap (Figure 3C) in the reacting and conserved substructures. For simplicity, only the data calculated using atoms is provided in Figure 3B. The values for each superfamily, calculated using both atoms and bonds, are provided in Table S2. The different regions in Figure 3B are intended merely to orient the reader to the range of variation across multiple superfamilies rather than to infer distinct categories implying fundamental differences between the superfamilies in different regions.
To determine whether there are differences in how a conserved substructure is used within a single superfamily, the variation of fc within each superfamily was also evaluated (Table S2). Most superfamilies have little variation in how much of the conserved substructure is reacting (Figure 4A). However, there are a few superfamilies with substantial variation in fc. We also evaluated the level of variation in which part of a superfamily's conserved substructure is used among the different reactions by calculating the average overlap in reacting and conserved substructures (or ∩ c) of every pair of substrates in the superfamily. A flatter distribution and more variation was observed among the superfamilies for the average or ∩ c (Figure 4B) than for the standard deviation of fc. The superfamilies that rank highest both in variation in fc and or ∩ c include the carbon-nitrogen hydrolase, metalloproteases (“zincins”) (catalytic domain), and the thioesterase/thiol ester dehydrase-isomerase superfamilies. Superfamilies that have low variation in fc and or ∩ c include the HD-domain/PDEase-like, dUTPase-like, and carbohydrate phosphatase superfamilies.
From these examples of superfamilies with high and low variation in fc and or ∩ c, we observe that the superfamilies with high variation tend to have smaller conserved substructures while superfamilies with low variation tend to have larger conserved substructures, though the correlation is not perfect. The superfamilies in the low variation group have phosphate groups in the conserved substructure. These tendencies may arise because different superfamilies and different types of reactions have different propensities for variation and conservation through evolution. Alternatively, variation in how different superfamilies are defined in SCOP may lead to some of the variation observed among these superfamilies. We also note that the set of reactions surveyed in this work represents only a subset of enzyme superfamilies, making it difficult to definitively address these hypotheses and questions. More extensive analyses will be required to confirm and further explore these initial observations.
As new superfamily members are characterized, modifications of these substructure conservation patterns may be required. To provide updates of this information, work is underway to incorporate this information into a searchable resource within our Structure-Function Linkage Database (http://sfld.rbvi.ucsf.edu/) [15]. Additional data generated in this study, including reacting substructures and how they overlap with conserved substructures for individual superfamily members, are available from the authors upon request. As described below, our method can also be used to determine conserved functional characteristics for superfamilies that have not yet been characterized. Programs and scripts required to perform these analyses are also available upon request.
Our analysis of the conservation of substrate substructures in enzyme superfamilies precisely determines aspects of chemical transformations that are conserved during divergent evolution. As such, it provides a view of conservation and divergence different from the view afforded by more common types of studies focused on enzyme sequences and structures. While our dataset of superfamilies and their associated substrates, products, and reactions is large, it is still limited as only single domain and unimolecular enzymes and superfamilies with sufficient data available were considered. Nevertheless, the results suggest a continuum in how enzyme superfamilies have evolved, from the reacting substructure being mostly conserved to being only slightly conserved (Figure 3A). Moreover, these superfamilies span a wide range in patterns of overlap (Figure 3B and 3C).
Previously, both large-scale and focused studies of enzyme evolution have recognized two primary models of how function is conserved. In the retro- or substrate-conserved model of enzyme evolution, Horowitz's original hypothesis describes how an existing enzyme in a pathway is duplicated and then evolves to convert new molecules into the substrate for the original enzyme in a metabolic pathway [36],[37]. In the resulting pathway, the newly evolved enzyme will function to provide a reaction required upstream of the original enzyme (i.e., the product of the newly evolved enzyme would be the substrate for the parent). In the second model, chemistry-constrained evolution, the ancestral enzyme, which can be from any pathway, is already promiscuous for or performs a fundamental type of chemistry (often a partial reaction) in common with the function of the daughter enzyme. The aspect of catalysis shared by the ancestral and daughter enzymes is maintained through conservation of structural features such as active site residues [1],[17],[38]. The key difference between these two models is in the pattern of function conservation within each. Related proteins that have diverged via the retro- or substrate-conserved model will bind substrates in common while the chemical reactions with those substrates differ. In the chemistry-constrained model, divergence can give rise to large superfamilies performing many different reactions. Members of such superfamilies will have conserved some aspect of the chemical reaction, which is often a partial reaction, while the substrates they use and their overall chemical reactions differ.
For the most part, the previous studies that have classified superfamilies into one or the other of these categories have been limited either in their scope (see the review by Glasner et al. for examples [39]) or in the type of data used [8],[9],[20],[21]. Although our current work cannot be directly compared with these previous analyses because of differences in methodologies, our results suggest that the evolution of enzyme function is too complex to be described by a few distinct categories. Instead, we see large variations in the patterns of substrate conservation across the set of superfamilies investigated in this study. Also, in these superfamilies, conserved substructures are not entirely reacting nor are they entirely non-reacting. This observation also suggests that the reacting and non-reacting substructures, the latter often including the part of the substrate that has binding interactions with the enzyme, are simultaneously relevant to the evolutionary process and should be analyzed together. Consistent with our observations, a recent network-based analysis of the evolution of metabolism concludes that the two models previously used to describe enzyme evolution are not mutually exclusive or independent [40].
Variations observed within individual superfamilies suggest additional complexity in the evolution of function and how conserved substrate substructures are used in catalysis. Although within most of the superfamilies we studied there is little variation in the extent to which conserved substructures are involved in the reaction (Figure 4), the observation of some variation, and in a few cases, considerable variation, demonstrates that even members of the same superfamily may not proceed with the same pattern of evolution.
As discussed in the sections below, these results also suggest potentially important implications for the analysis of individual superfamilies, functional annotation, and value of evolutionary information in providing guidance for enzyme engineering.
By automating the analysis of enzyme substrates and reactions, the methodology introduced in this work facilitates the analysis of previously unstudied enzyme superfamilies. This effort contrasts with previous analyses of enzyme superfamilies to determine patterns of functional conservation that have been highly labor-intensive, involving extensive manual analysis of reactions and literature-based curation of functional properties (see the SFLD, http://sfld.rbvi.ucsf.edu/, for examples). The substructures conserved among the substrates of all members of a superfamily (Figure 2) provide annotation information that describes how function has been conserved in each of these superfamilies. The certainty of these superfamily annotations will depend, however, on how well the range of substrates in each superfamily has been sampled. Thorough substrate sampling may be especially critical for complex superfamilies that include many different catalytic functions. While we have used all available reaction information in our analyses, the sampling of superfamily reactions may still be incomplete. As new reactions are discovered through the sequencing of new genomes and metagenomes, these results can be updated and improved.
Despite these limitations, the characterization of superfamily-conserved substructures presented here facilitates the annotation of individual sequences on a large scale, helping to address the need for new strategies for automated function annotation. This issue has become more pressing as the number of sequenced genomes increases and the era of metagenomics moves into high gear [41]. Sequences that can be classified into a superfamily but not into a specific family can be annotated with the substructure common to all characterized members. In these cases, often found in complex superfamilies exhibiting broad diversity in enzyme function, this may be the only level at which accurate annotation can be achieved, as insufficient information may be available to support annotation of a specific reaction or substrate specificity.
While substructure-based annotation does not by itself suggest a specific enzyme function, this information can be used as a starting point for additional analyses to determine specific function. For example, many structures have been solved through structural genomics efforts, but their functions remain unknown [42]. We have compiled a list of structures that have been classified into the SCOP superfamilies analyzed in this study, but have unknown functions. These structures, many of them from structural genomics projects, can be at least minimally annotated with the substructure identified here as conserved across that superfamily, illustrated by the examples given in Figure 5 (see Table S3 for the complete list). Using this information, characteristics of ligands likely to be bound or turned over by these proteins can be inferred, providing guidance for biochemical studies to determine specificity. These data also provide information about classes of small molecules that may be useful for co-crystallization trials to aid in solving the structures of these proteins or to capture them in functionally relevant conformations.
The variation found within superfamilies presents a caveat to be considered when using these substructures for function annotation. While most of the superfamilies analyzed here have conserved substructures that are used consistently among the different superfamily members (Figure 4), there are a few superfamilies that have significant variation in the degree to which the conserved substructure is used in the reactions. These superfamilies can be expected to be more difficult cases for function prediction since their variability makes it more difficult to determine conserved aspects of function. In contrast, superfamilies with less variation in the degree to which the conserved substructure is used in the reaction are expected to be more straightforward cases for function prediction.
Understanding the patterns of functional conservation associated with the evolution of functionally diverse enzyme superfamilies can provide useful information for guiding enzyme engineering experiments in the laboratory [43]. Using as a starting template for design or engineering an enzyme that already “knows” how to perform a critical partial reaction or how to bind a required substrate substructure ensures that some of the machinery required to perform a desired function is already in place. Although still daunting, the task then simplifies to modifying the enzyme to bind and turn over a new substrate that contains the substructure consistent with the underlying capabilities of the superfamily. As a corollary, aspects of function that have been conserved in all members of a divergent superfamily may be difficult to modify by in vitro engineering [43],[44]. Using such a strategy in a proof-of-concept study, two members of the enolase superfamily were successfully engineered to perform the reaction of a third superfamily member [45]. As shown in Figure 6, the superfamily-conserved substructure and the partial reaction associated with that substructure were not changed in these experiments. Rather, engineering the template proteins to perform the target reaction involved changing each to accommodate binding the part of the substrate that is unique to the new reaction desired.
To allow for generalization of this approach, our analysis provides for all of the superfamilies that we investigated 1) the parts of an enzyme's substrate and reaction that are not conserved among related enzymes, which, provided they can be associated with regions of a target structure that interact with them, may point to structural features amenable to engineering, and 2) the parts of the substrates that are conserved across all members of a superfamily, which may point to regions of the structure that may not be easily changed without loss of function or stability [46].
In this study, requirements for a sufficiently large sample of enzyme reactions for a comprehensive analysis restricted us to using only substrates and products. However, enzyme substrates can undergo intermediate changes during catalysis that are not adequately captured by looking only at substrates and products. In some reactions, such as those in the enolase superfamily [47], some portions of the substrate change and revert back to their original configuration during the reaction; these types of transformations are undetectable in the study described here. The enolase superfamily represents a well-characterized example of chemistry-conserved evolution. However, because our analysis does not currently detect such substrate changes, the average fc(atoms) for the enolase superfamily is 0.31 and the average fc(bonds) for the enolase superfamily is 0.34, which places this superfamily in the middle of the distribution among our superfamilies for these measures of overlap. Being able to detect the full extent to which structures change during a reaction would provide a better picture of substructure conservation in superfamilies like the enolase superfamily. But this will require compilation of additional data to capture all of the partial reactions involved in a given overall reaction, including structures of reaction intermediates. Emerging data resources, such as MACiE [16] and the SFLD [15], currently seek to catalog information about reaction steps and mechanisms. However, because this process is labor-intensive and often hampered by disagreement or ambiguity in the literature regarding the specific mechanisms of some reactions, these data resources are not yet sufficiently populated to support such broader analyses. As these types of resources grow, we are optimistic that the information required to analyze reaction mechanisms more fully will become increasingly available. Although it is beyond the scope of this study, correlating the conservation patterns we see in enzyme substrates with the conservation patterns in the sequence and structures of the enzymes themselves would also be a valuable extension for these analyses.
Finally, recent progress has been made in using in silico docking of small molecules to enzyme structures to infer molecular function. In one such study, a library of high-energy reaction intermediates was generated and used to predict substrate specificity of enzymes in the amidohydrolase superfamily [48]. As these methodologies are further developed, incorporation of predicted reaction intermediates into substructure analysis could improve prediction of substructures that are reacting. In addition to benefiting from such recent advances in docking, the type of analysis presented here may in turn be used to improve applications of docking to predicting substrate specificity in enzymes. Several such studies have recently focused on predicting functional specificity in the enolase [49],[50] and amidohydrolase [51] superfamilies using knowledge about conserved substrate substructures from earlier analyses [15],[52] to construct focused ligand libraries for docking. We expect that the set of conserved substructures generated by our analysis can be used similarly to guide the construction of chemical libraries of ligands to improve prediction of substrate specificity in other superfamilies.
This study presents an automated method for analysis of superfamilies to determine the conserved aspects of their functions, represented by patterns of substrate conservation. Our results show that superfamilies do not fall into discrete and easily separable categories describing how their functions may have evolved. Rather, the conserved substructures determined in this analysis define superfamily-specific conservation patterns. These results enable precise prediction of functional characteristics at the superfamily level for complex superfamilies whose members perform many different but related reactions, even when the evidence is insufficient to support more specific annotations of overall reaction and substrate specificity. For applications in enzyme engineering, we expect that the identification of the aspects of function that have been most and least conserved during natural evolution will provide guidance for identifying the structural elements of a target scaffold that are most and least amenable to modification, thereby informing engineering strategies for improved success.
For our analyses, we used a subset of superfamilies from SCOP, a database of manually classified protein superfamilies, filtered based on criteria chosen to be most informative about enzyme evolution at high levels of functional divergence. We included only superfamilies of single-domain enzymes with significant functional information in SCOPEC, a subset of SCOP with verified EC numbers, and in BRENDA, the most comprehensive database of enzyme experimental results. Although many enzymes and proteins function as multi-domain units, the nature and organization of which can affect the specificity and regulation of enzymes [53], for this study, we chose to use only single-domain enzymes as this allowed us to clearly assign a single function to one domain. We included examples of enzymes known to have multiple structural domains only when the composite acts as a single functional unit (e.g., the enolase superfamily).
To ensure that the members of each superfamily were sufficiently divergent in function to analyze conservation of their substructures, only superfamilies annotated with at least two different EC numbers were investigated. Compared to unimolecular reactions, bimolecular reactions have considerably more complex chemical and kinetic mechanisms for how substrates interact with the enzyme's catalytic site (i.e., in what order different substrates bind). Because these variations would have greatly complicated the analysis, we excluded superfamilies with any reactions that were not unimolecular. Using the top level of the EC annotation, superfamilies were selected in which all the characterized members belong to any one of the following classes: hydrolases (EC numbers 3.x.x.x), lyases (EC numbers 4.x.x.x), and isomerases (EC numbers 5.x.x.x).
Experimentally verified substrate and product data were taken from the licensed version of the BRENDA database (release 6.2) [54]. Reactions were excluded in which (1) the product(s) had more than five (non-hydrogen) atoms more than the substrate or (2) substrates and products both had three or fewer (non-hydrogen) atoms. Reactions in the first category are likely to be erroneous because they are not properly balanced. Reactions in the second category are unlikely to be informative for the analysis because they contain so few atoms.
A “conserved substructure” (Figure 1A) contains the maximal sets of bonds in a substrate that are present in all the substrates of a superfamily, plus their adjacent atoms. In all our analyses, we considered only bonds consisting of two atoms, neither of which is a hydrogen. The “unconserved substructure” is the set of bonds in a substrate that are not in the conserved substructure, plus their adjacent atoms. An atom can be in both the conserved and unconserved substructure if it is adjacent to both a bond in the conserved substructure and a bond in the unconserved substructure.
A “reacting substructure” (Figure 1B) consists of the bonds in a substrate that are not present in the product, their adjacent atoms, and any atoms that become connected in new bonds in the product. In the case of a racemization reaction, in which the chirality of an atom center changes, the reacting substructure is defined as including the chiral atom that changes in the reaction, the four adjacent bonds and their adjacent atoms. The “nonreacting substructure” is the set of bonds in a substrate that are also present in the product and their adjacent atoms. An atom can be in both the reacting and nonreacting substructure if it is adjacent to both a bond in the reacting substructure and a bond in the nonreacting substructure.
The substrate substructure conserved among all characterized members of each superfamily was calculated using the maximal common substructure (MCS) algorithm implemented in the Chemistry Development Kit (CDK) [55], an open source Java toolkit for manipulating small molecules. The molecules are represented as graphs in which the nodes represent atoms and the edges represent bonds. Each node is labeled with an atom type and each edge is labeled with the two atom types of the connected atoms and the bond order. This algorithm finds, for a pair of molecules, the maximum common substructure (MCS) present in both molecules. We extended this to find the MCS for the set of all known substrates for a superfamily. In this initial analysis, we treated different atoms as dissimilar as long as the element type was different and bonds as different when the bond order and the two pairs of connected atoms were not identical. The only exception to this rule was made for phosphate and sulfate groups, which we treated as similar in the substrate conservation analyses. Our code allowed for the possibility of multiple unconnected MCSs by representing them as an unconnected graph with each connected portion corresponding to one MCS. Although some of the pairwise MCSs contain multiple unconnected subgraphs, none of the superfamily-conserved substructures contain such multiple unconnected MCSs. Finally, each substrate has a unique unconserved substructure defined as the set of edges not present in the conserved substructure and the atoms adjacent to these edges.
For each enzymatic reaction in which both the substrate and its corresponding product(s) are known, we calculated the non-reacting substructure by finding the MCS between the substrate and the product(s). The reacting substructure is the set of edges in the substrate that are not present in the product, plus the atoms adjacent to these edges. The reacting substructure also includes atoms that form new bonds in the product.
To quantify the overlap between the reacting and conserved substructures, for each reaction in our dataset, we calculate fc (Figure 1C), the fraction of the conserved substructure that is reacting and fr (Figure 1D), the fraction of the reacting substructure that is conserved. The values for fc and fr are calculated in two ways, using atoms or bonds, and the results for both are reported as they provide different but useful views of the data. fc for bonds is determined by dividing the number of bonds that are in both the conserved and the reacting substructures (r ∩ c) by the number of bonds in only the conserved substructure. fc for atoms is determined similarly, using the number of atoms instead of bonds. Likewise, fr for bonds is determined by dividing the number of bonds that are in both the conserved and the reacting substructures by the number of bonds in only the reacting substructure; this value was also calculated using atoms. For each enzyme in the BRENDA database, there may be multiple substrates with corresponding reactions that have been characterized. For these cases, the values of fc and fr were obtained by averaging all the substrates of each enzyme and then these values were averaged for all the enzymes in each superfamily. We also determined the standard deviation in fc and fr for the enzymes of each superfamily.
To determine whether the same part of the superfamily-conserved substructure was used in the different reactions of the superfamily, every pair of reactions was analyzed in each of the superfamilies in our dataset. Each reaction has a substrate substructure that is both conserved and reacting (r ∩ c). For each pair of reactions, we calculated how much overlap is observed among the two (r ∩ c) substructures and normalized each of these overlaps by the smallest (r ∩ c) of each pair. The resulting measure of overlap (or ∩ c) was then averaged over every pair of reactions in each superfamily.
|
10.1371/journal.pcbi.1004923 | Stochastic Simulation of Biomolecular Networks in Dynamic Environments | Simulation of biomolecular networks is now indispensable for studying biological systems, from small reaction networks to large ensembles of cells. Here we present a novel approach for stochastic simulation of networks embedded in the dynamic environment of the cell and its surroundings. We thus sample trajectories of the stochastic process described by the chemical master equation with time-varying propensities. A comparative analysis shows that existing approaches can either fail dramatically, or else can impose impractical computational burdens due to numerical integration of reaction propensities, especially when cell ensembles are studied. Here we introduce the Extrande method which, given a simulated time course of dynamic network inputs, provides a conditionally exact and several orders-of-magnitude faster simulation solution. The new approach makes it feasible to demonstrate—using decision-making by a large population of quorum sensing bacteria—that robustness to fluctuations from upstream signaling places strong constraints on the design of networks determining cell fate. Our approach has the potential to significantly advance both understanding of molecular systems biology and design of synthetic circuits.
| Simulation algorithms have become indispensable tools in modern quantitative biology, providing deep insight into many biochemical systems, including gene regulatory networks. However, current stochastic simulation approaches handle the effects of fluctuating extracellular signals and upstream processes poorly, either failing to give qualitatively reliable predictions or being very inefficient computationally. Here we introduce the Extrande method, a novel approach for simulation of biomolecular networks embedded in the dynamic environment of the cell and its surroundings. The method is accurate and computationally efficient, and hence fills an important gap in the field of stochastic simulation. In particular, we employ it to study a bacterial decision-making network and demonstrate that robustness to fluctuations from upstream signaling places strong constraints on the design of networks determining cell fate.
| Dynamic simulation is an essential and widespread approach for studying biomolecular networks in cell biology [1]. However, the computational resources required can quickly become limiting for several reasons. Cellular networks are complex, containing many biomolecular species and reactions. The effects of biochemical stochasticity can be pervasive at the single-cell level [2, 3], implying that stochastic simulation approaches are often needed. And cells do not live in isolation, which requires simulation on multiple scales, ranging from the single cell to large ensembles of communicating cells [4, 5]. In these circumstances, parsimonious models of intracellular networks offer dimension reduction [6–8] and significant advantages [9]. However, such models often only provide accurate descriptions when they include the effects of interactions with other fluctuating processes in the cell and of signals arising extracellularly [10–12]. While it is straightforward to write a Chemical Master Equation describing the stochastic dynamics of these models, it is usually impenetrable to analysis and one needs to make use of simulation methods. The stochastic simulation algorithm (SSA) [13, 14] allows only the random timing of reactions in the network model to be taken into account (often known as intrinsic noise), but cannot be used when other processes interacting with the network cause its propensities to fluctuate between reaction occurrences. The SSA assumes constant propensities between reactions (and hence exponentially distributed waiting times). Here we present a new approach relaxing this assumption, called Extrande, for stochastic simulation of a biomolecular network of interest embedded in the dynamic, fluctuating environment of the cell and its surroundings. An extensible implementation of Extrande for general reaction networks with multiple inputs is given in the S1 File.
Biological processes that interact with the network or model of interest are sometimes called extrinsic processes [15]. They often significantly change the stochastic behaviour and dynamics of the network [16, 17]. We briefly give two illustrations of the biological importance of extrinsic processes as motivation for the development of our approach, the first well-established, and the second considered here. First, although intrinsic noise is an important contributor, extrinsic processes are known to be a substantial and sometimes dominant source of variation in gene expression levels across cells and over time [18–21]. We are now beginning to understand the underlying biological sources [22], which include effects related to circadian oscillations, temperature, chromatin remodelling, the cell-cycle and pulsatile transcription factors [23, 24]. To understand gene expression, it is therefore essential to move beyond the SSA, which can only account for intrinsic noise, and to include other sources of variation. Second, fluctuations in the expression, degradation and recycling of proteins inevitably affect the way networks containing those proteins function and the extent of stochasticity in the input they provide to other networks. Fluctuations in the component proteins of signal transduction networks limit information transfer [25], affect transduction network ‘design’ [26] and, although often overlooked, are inevitably conveyed (as extrinsic inputs) to the networks regulated by signaling. Here, the computational advantages of Extrande will allow us to demonstrate how fluctuations in the protein componentry of signal transduction networks are conveyed to signaling outputs and place strong constraints on the design of networks determining cell fate, thus influencing the distribution of phenotypes at the population level. Without the ability to simulate biomolecular networks that are exposed to fluctuating inputs, the ability to address such questions is severely restricted.
There are two existing approaches to stochastic simulation of reaction networks subject to dynamic, fluctuating inputs. The first class of algorithms [5, 13, 27] implements the SSA, under the approximation that the input remains constant between the occurrences of any two reactions. However, this approximation can give spurious results even when dynamic inputs to the network are changing relatively slowly. We term these collectively the Slow Input Approximation method (SIA). The second class of algorithms [28–30] involves step-wise numerical integration of reaction propensities until a target value for the integral is reached. Algorithms in this class would be (conditionally) exact, if it were not for the presence of numerical error in integration, but can impose large and impractical computational burdens, especially when cell ensembles are studied. We term these collectively the integral method (distinguishing next and direct integral approaches below). We perform a comparative analysis of both methods with Extrande and demonstrate that our method offers an accurate and computationally efficient alternative approach. Extrande involves no analytical or numerical integration but instead relies on ‘thinning’ techniques [31, 32]. Other approaches using rejection methods have also recently been proposed as a means to tackle systems with time-dependent propensities [12, 33].
The stochastic simulation algorithm (SSA) [13, 14] allows simulation of biomolecular reaction networks taking into account the discreteness of these systems as well as the intrinsic randomness in the timing of reaction events. The SSA assumes that the propensity of each reaction channel to fire, hence the probability of the reaction to occur over a small time interval, remains constant between reaction events. This naturally restrains the use of SSA to simulate networks embedded in dynamic, fluctuating environments because the reaction propensities then become time-varying quantities under the influence of extrinsic processes.
Extrande (Box 1)—or Extra Reaction Algorithm for Networks in Dynamic Environments—allows exact stochastic simulation of any downstream reaction network, conditional upon a time course of the dynamic inputs that is simulated up-front. The method involves no analytical or numerical integration, though we give a connection to the direct integral method below, and instead makes use of point process ‘thinning’ techniques [31, 32], where some simulated events are discarded. The only error incurred is any error associated with the input pre-simulation, typically an approximate simulation of a stochastic differential equation (Box 1).
The Extrande approach can be understood as introducing an extra, ‘virtual’ reaction channel into the system (whose occurrence does not change molecule numbers). The propensity of the extra channel is designed to fluctuate over time so that (when added to the sum of all other reaction propensities) the total propensity in the augmented system becomes constant between events and equal to an upper bound on the sum of the propensities in the original system. To accomplish this, the method exploits the exogeneity of the dynamic inputs—the assumption of negligible retroactivity [35] from network to inputs. In particular, their exogeneity means that Extrande is able to make use of the ‘future’ trajectory of the inputs to find an upper bound, B, on the total propensity, which is valid over a certain time interval L (see Step 3, Box 1).
Simulation of the augmented system is feasible by means of an SSA-like algorithm. The method uses the bound on the total propensity to generate a putative reaction reaction time τ (Step 4). If the reaction time exceeds the time horizon L, it is rejected; the system time advances by L (Step 6), and the procedure restarts by determining a new bound. Otherwise, time advances by τ and a reaction is chosen based on the updated reaction propensities (at time t+τ) (Steps 8–15). The reaction events of the virtual channel are discarded, leaving those of the other channels—because the simulated timing and types of the biochemical reaction channels are unaffected by the behaviour of the extra channel, the result is a trajectory of the original system (see Methods).
We study the decision to enter competence (for uptake of extracellular DNA) by the model organism Bacillus subtilis. It is well established [39–41] that the source of differentiation of 10–20% of the cell population under stress conditions is fluctuations in transcription of the master competence regulator, ComK. The ComS-MecA-ComK competence module is regulated by the activated transcription factor pComA, the output of the transduction mechanism relaying extracellular, quorum sensing signals (CSF and ComX), see Fig 3A. We study the effect of this upstream signaling on differentiation into the competent phenotype.
A useful approach to understanding the structure-function relationship in systems biology is to rewire networks found in nature and compare function with the wild-type, which can then shed light on why apparently similar network structures were not adopted naturally [42]. In the wild-type, upstream signaling acts via activation of the ComS promoter by pComA binding (Fig 3A, thick black arrow). We compare the behaviour of wild-type cells to those with a Synthetic Decision-Making network (SynDM) which is regulated, in addition, via activation of the ComK promoter by pComA binding (red dashed arrow). We model ComK-driven progress and entry into functional competence, and write Progress ( t ) = k ∫ 0 t ComK ( s ) d s, where k is an effective rate of ComK-driven differentiation. A cell is taken to enter (functional) competence at the time when Progress(t) = 1. The value of the parameter k is set so that the wild-type and SynDM networks give equal fractions of competent cells with a constant level of pComA (1000 molecules). We tune rate parameters associated with the ComK promoter of the SynDM network so that the fraction of SynDM cells entering competence (0.18) is the same as for wild-type cells, in the absence of fluctuations in pComA levels (see S1 Text). A table listing all reactions and parameter values used in our models of the competence module of wild-type B. subtilis and the SynDM networks is given in the S1 Text.
We use the linear noise approximation (LNA) [43] to model the the upstream signaling (with CSF and ComX fixed at steady-state levels), giving a mean for pComA of 1000 molecules throughout. Importantly, we include in the model gene expression and degradation of the proteins comprising the signal transduction mechanism because it is now understood that the resultant variation has important effects on signaling and information transfer [26]. A single Ornstein-Uhlenbeck (OU) process is sufficient to closely match the mean, variance and autocorrelation function of pComA given by the LNA (see S1 Text). We therefore use a single OU process for the pComA input in what follows. A range of protein lifetimes is considered, consistent with the broad range of cell-cycle periods observed for bacteria under different growth conditions [44], where nutrient limitation can result in periods in excess of 10h. Our baseline LNA model of the upstream signaling module gives a lifetime and CV of pComA fluctuations equal to 5h and 0.35. We take the pComA input to be exogenous to the ComS-MecA-ComK competence module since it is in high abundance relative to the 2 promoters it binds (the only interaction between the two modules).
The importance in determining cell fate of the time taken for the cell to complete different differentiation programs (to the point of irreversible commitment) has recently been emphasised [45]. The SynDM network creates a differentiated sub-population by activating the differentiation program in most or all of the cell population (Fig 3C & 3D), with entry to competence the outcome of a ‘race’ to differentiate over the relevant time window. In the SynDM network, binding of pComA to the ComK promoter results more often in periods of non-zero ComK expression than in the wild-type population, but when such periods occur, they are less sustained (see Fig 3B–3D, and Fig. E in S1 Text). The typical rate of progress of a SynDM cell to competence is increased by a higher level of pComA (see Fig. E in S1 Text), and extrinsic fluctuations in the pComA level therefore affect the fraction of cells entering competence (Fig 3C & 3D). In contrast, the wild-type activates the differentiation program in a smaller sub-population, the size of which is under modest regulation by pComA (Fig 3F)—a high proportion of the active wild-type cells then enter competence because, once activated, ComK expression rarely deactivates in the wild-type (see Fig 3B, and Fig. E in S1 Text).
We find two important advantages of the wild-type design (in addition to the implied reduction in the metabolic cost of gene expression). First, the fraction of cells entering competence is considerably more robust to the fluctuations from upstream signaling in pComA (Fig 3E). For example, with the baseline model of upstream signaling, the SynDM network has a competent fraction (40%) which is more than 2.25 times the competent fraction when pComA is held constant at its mean level, whereas the competent fraction of wild-type cells (17% cf 18%) has changed very little. The difference in robustness is explained by the sensitivity of the probability of competence for a SynDM cell as a function of the time average of the signal, 〈pComA〉, which switches quite rapidly from zero to one (Fig 3F). Since the fraction of competent cells is equal to the average of Prob[Competence|〈pComA〉] over the distribution of 〈pComA〉 (which is approximately the distribution of pComA for longer lifetimes), the competent fraction increases in the presence of extrinsic fluctuations for SynDM (recall the mean of pComA is 1000 molecules). In contrast, Prob[Competence|〈pComA〉] is approximately linear for the wild-type network, which implies that the competent fraction depends largely on the mean of pComA alone. Such plots (Fig 3F) should prove a useful diagnostic tool for the design of synthetic decision-making networks.
The second advantage of the wild-type design is that the fraction of cells entering competence is also considerably more robust than SynDM to heterogeneity across the cell population in the rate at which ComK-driven differentation proceeds (Fig 3G). The reason is evident from the progress to competence trajectories in Fig 3B–3D. We note that fluctuations from upstream signaling in pComA can also cause decreases in the fraction of competent SynDM cells, as seen for higher rates of differentiation (Fig 3G). Heterogeneity in the rate at which differentiation programs proceed is inevitable where cellular decisions are executed by large gene expression networks and involve substantial physiological changes [46].
These in silico experiments (Fig 3), made computationally feasible by Extrande, cast light on the wild-type network design in which quorum signaling input to the competence decision-making network (ComS-MecA-ComK) by the transcription factor pComA exerts its effect only at the promoter of ComS and not at the promoter of ComK. The experiments reveal exquisite robustness of the wild-type design to fluctuations from upstream signaling and to heterogeneity in downstream processes, and demonstrate the computational potential of Extrande for in silico network design.
Stochastic simulation of biomolecular networks is now indispensable for studying biological systems, from small reaction networks to large ensembles of cells. The effects of stochasticity can be pervasive at the single-cell level, determining the distribution of phenotypes in a population and thus potentially affecting evolutionary outcomes. However, studying such phenomena requires stochastic simulation of a large ensemble of cells that can take into account both intrinsic and extrinsic sources of cellular variation. This can be hugely costly in terms of CPU time, placing important in silico experiments out of reach. Here we provide the new Extrande approach—for stochastic simulation of a biomolecular network embedded in the dynamic environment of the cell and its surroundings—which substantially increases the computational feasibility of such experiments without compromising accuracy.
We show that previous approaches to this problem either can fail dramatically, even when inputs vary relatively slowly, or impose impractical computational burdens due to costly numerical integration of reaction propensities. Given a simulated trajectory of fluctuating network inputs, the Extrande approach provides a conditionally exact solution that can speed up simulation by several orders of magnitude compared to integral methods. In practice, we find that integral methods suffer from the high cost of propensity evaluations during numerical integration. Extrande bypasses numerical integration by introducing an extra reaction channel—one designed to keep the total propensity of the ‘augmented’ system constant between events—hence making the problem of finding the time to the next event analytically tractable. Importantly, our numerical results demonstrate that the overhead costs induced by the Extrande method—for example, due to thinning and rejection events, and due to obtaining the ceiling of the input process when a global ceiling is not available–are significantly lower than the cost of accurate numerical integration. In practice, we observe speed-ups by a factor as great as 2.5×104 (Fig 2C).
Recent work [12] proposes to handle fluctuating environments in a different manner, by deriving a network model for the biochemistry that takes account of the dynamic input and follows the correct (marginal) probability law. Explicit simulation of the input is bypassed. The resultant ‘uncoupled’ network model has time-varying reaction propensities and can then be simulated using integral or thinning methods. However, analytical derivation of the uncoupled network model is not always possible, particularly when there are multiple inputs. The accuracy of the method then depends on finding suitable approximation schemes.
There are two main limitations of modelling using the Extrande method. The first is that Extrande, being a method of obtaining trajectories of the chemical master equation (with time-dependent propensities), has the same applicability limitations as the master equation; namely there is an implicit assumption that the system is dilute (point particles) and well-mixed, conditions which are not met when molecular crowding is significant [47, 48]. The second limitation is that Extrande assumes that the inputs influence the system of interest but the latter does not influence the inputs (which implies the inputs can be pre-simulated). Hence the method is useful, for example, to understand how certain external stimuli such as light and temperature can affect the stochastic dynamics of a system. For the case of a chemical stimulus, the method can provide an accurate description of the stochastic dynamics if the system and its output do not significantly feedback to adjust the original chemical stimulus, for example by a regulatory mechanism.
We exploit the benefits of the proposed Extrande simulation method here to study the decision-making behaviour of a quorum sensing population of bacterial cells. The in silico experiments presented (Fig 3) took approximately two computing months using Extrande (and an Intel Xeon, 3.3GHz quad-core processor with 32GB of RAM), but would have been prohibitive using the integral method due to the approximate 70-fold slow down needed to ensure even modest accuracy (see Fig. D in S1 Text). The results elucidate the costs and benefits of alternative network designs for the probabilistic differentiation of a sub-population of cells in response to upstream signaling. Our findings argue for the biological significance of fluctuations in signaling inputs that arise from synthesis and degradation of the protein componentry of signal transduction networks, and show that these fluctuations have important consequences for downstream networks such as those deciding cell fate. We expect the accuracy and reductions in CPU time made possible by Extrande to help open up the landscape of computationally feasible simulation of biomolecular networks and cell ensembles. Extrande thus has the potential to accelerate both understanding of molecular systems biology and the design of synthetic networks.
The Extrande approach relies on augmenting the reaction network with an extra, ‘virtual’ channel (giving the augmented system, Z), so as to make simulation of the augmented system feasible, while ensuring that the simulated timings and types of biochemical reactions are unaffected by the firings of the extra channel. In the Extrande method, the conditional propensity of the extra channel depends on the history of the extra channel (as well as on the history of the original system, H t X), and so does the upper bound. A related Proposition in [32] does not allow for this dependence (see S1 Text). We therefore provide the new proof below. To see the dependence on the extra channel, note that the bound is in general updated in Step 3 of the Extrande algorithm (Box 1) after each firing of the extra channel.
The reaction network to be simulated (Box 1) has the number of molecules of each species at time t given by
X ( t ) = X ( 0 ) + S R ( t ) ,
where R(t) = {R1(t), …, RM(t)} is the vector of processes counting the number of times each biochemical reaction channel fires during the time interval [0, t], and S = {v1, …, vM} is the stoichiometric matrix. The ‘Poisson’ or random time-change representation [49] expresses R(t) in terms of M independent, unit rate Poisson processes, Y(t) = {Y1(t), …, YM(t)}, and so can be written here as
X ( t ) = X ( 0 ) + S Y 1 ∫ 0 t a 1 [ X ( s ) , I ( s ) ] d s , . . . , Y M ∫ 0 t a M [ X ( s ) , I ( s ) ] d s T , (1)
where I is the possibly multivariate input, superscript T denotes transpose of a vector, and aj[X(s), I(s)] is the propensity of the jth reaction, for j = 1, …, M, conditional on { H s X , I }. We denote by I (the σ-field generated by) the entire trajectory of the input.
We introduce as a simulation device the extra, virtual reaction RM+1: ∅ → ∅, to form the augmented system Z ( t ) = X ( t ) R M + 1 ( t ) = X ( 0 ) 0 + S 0 0 1 R ( t ) R M + 1 ( t ) .
The random time-change representation of the augmented system is in terms of (M+1) independent, unit rate Poisson processes, Y(t) = {Y1(t), …, YM+1(t)}
Z ( t ) = Z ( 0 ) + ( S 0 0 1 ) × ( [ … , Y j ( ∫ 0 t a j [ X ( s ) , I ( s ) ] d s ) , … ] , Y M + 1 ( ∫ 0 t a M + 1 ( s ) d s ) ) T (2)
where aM+1(s) is the propensity of the extra reaction channel (conditional on { H s Z , I }), and where we set aj[X(s), I(s)], for j = 1, …, M, as the propensity of the jth reaction conditional on { H s Z , I }, which now includes the history of the extra channel, RM+1.
Notice that Eq 2 is identical to Eq 1 in its expression of the original system, X(t), or equivalently of R(t). Therefore, if the propensity aM+1 is chosen to somehow make simulation of [R(t), RM+1(t)] straightforward, we are able to simulate our target, R(t), by simulating the augmented system in Eq 2 and then ignoring RM+1(t). To do this, let B(t) be an ( H t Z , I )-measurable random variable satisfying (with probability 1) that
a 0 ( t ) = ∑ j = 1 M a j [ X ( t ) , I ( t ) ] ≤ B ( t ) , t ≥ 0 ,
so that B(t) is a stochastic upper bound for the total biochemical reaction propensity. Now define the propensity of the extra channel (conditional on { H t Z , I }) as:
a M + 1 ( t ) = B ( t ) - a 0 ( t ) .
The ground process (see S1 Text) of [R(t), RM+1(t)] has propensity (conditional on { H t Z , I }) given by ∑ j = 1 M + 1 a j ( t ) = B ( t ) , by construction. The Extrande method chooses the stochastic bound, B(t), so that it is constant between firings of the augmented system (see Box 1), which makes straightforward the simulation of the ground process of [R(t), RM+1(t)]. We write the ith occurrence time of the ground process of [R(t), RM+1(t)] as Ti, i = 1, 2, … It is now the case that
Prob { T i + 1 - T i ≤ t | T 1 , Z 1 , . . . , T i , Z i , I } = 1 - exp { - t B ( T i ) } ,
where Zi is the channel corresponding to the ith firing. The waiting time has an exponential distribution and the occurrence times {T1, T2, …} are therefore just those of a ( H t Z , I )-Poisson process with propensity B(t), and can be simulated analogously to the SSA as in Step 4 of Box 1.
What remains is to assign each firing time Ti to one of the (M+1) channels of the augmented system. We do the allocation sequentially, using the result from counting process theory [50] that, for j = 1, …, (M+1):
Prob { Z i + 1 = j | T 1 , Z 1 , . . . , T i , Z i , T i + 1 , I } = a j [ X ˜ ( T i + 1 ) , I ˜ ( T i + 1 ) ] B ( T i ) , (3)
where we have used the left-continuous versions ( X ˜ ( t ) , I ˜ ( t ) ) of (X(t), I(t)), and B ˜ ( T i + 1 ) = B ( T i ). Eq 3 is implemented by Steps 9–15 in Box 1. The intuition for Eq 3 uses Bayes’ theorem. Consider a small interval of time dt. The probability that the channel is the jth one given that some reaction fires at time Ti+1, since the probability of more than one reaction can be neglected, is given by [ d t · a j ( X ˜ T i + 1 , I ˜ T i + 1 ) ] / [ d t · k = 1 M + 1 ∑ a k ( X ˜ T i + 1 , I ˜ T i + 1 ) ] . The target of the Extrande simulation, R(t), is now obtained by ignoring all the firing times of the extra channel after simulation of the augmented system is complete. This completes the proof. ■
We note that the condition limt → ∞ Rj(t) = ∞ (j = 1, …, M) is needed for the representation in Eq 1, but is not needed for the validity of the Extrande method. The random time-change representation is used here to make the proof more accessible. The Extrande algorithm results in a probability law, P, under which the functions aj[X(t), I(t)] give the propensities of the biochemical reactions conditional upon ( H t Z , I ). Because the aj[X(t), I(t)] are ( H t X , I )-measurable, they also give the ( H t X , I )-conditional propensities of the biochemical reactions under P, as required of the probability measure P resulting from the Extrande algorithm.
Finally, we remark that a description equivalent to the random time-change representation, Eq 1, is the Chemical Master Equation [49]. Specifically, for the conditional probability P ( n , t ) = Prob ( X ( t ) = n | X ( 0 ) = n 0 ; I ) one can write
d P ( n , t ) d t = ∑ j = 1 M a j [ n - v j , I ( t ) ] P ( n - v j , t ) - a j [ n , I ( t ) ] P ( n , t ) , (4)
whose propensities are time-varying, stochastic functions due to the dependence on the input process.
|
10.1371/journal.pcbi.1000008 | Diminished Self-Chaperoning Activity of the ΔF508 Mutant of CFTR Results in Protein Misfolding | The absence of a functional ATP Binding Cassette (ABC) protein called the Cystic Fibrosis Transmembrane Conductance Regulator (CFTR) from apical membranes of epithelial cells is responsible for cystic fibrosis (CF). Over 90% of CF patients carry at least one mutant allele with deletion of phenylalanine at position 508 located in the N-terminal nucleotide binding domain (NBD1). Biochemical and cell biological studies show that the ΔF508 mutant exhibits inefficient biosynthetic maturation and susceptibility to degradation probably due to misfolding of NBD1 and the resultant misassembly of other domains. However, little is known about the direct effect of the Phe508 deletion on the NBD1 folding, which is essential for rational design strategies of cystic fibrosis treatment. Here we show that the deletion of Phe508 alters the folding dynamics and kinetics of NBD1, thus possibly affecting the assembly of the complete CFTR. Using molecular dynamics simulations, we find that meta-stable intermediate states appearing on wild type and mutant folding pathways are populated differently and that their kinetic accessibilities are distinct. The structural basis of the increased misfolding propensity of the ΔF508 NBD1 mutant is the perturbation of interactions in residue pairs Q493/P574 and F575/F578 found in loop S7-H6. As a proof-of-principle that the S7-H6 loop conformation can modulate the folding kinetics of NBD1, we virtually design rescue mutations in the identified critical interactions to force the S7-H6 loop into the wild type conformation. Two redesigned NBD1-ΔF508 variants exhibited significantly higher folding probabilities than the original NBD1-ΔF508, thereby partially rescuing folding ability of the NBD1-ΔF508 mutant. We propose that these observed defects in folding kinetics of mutant NBD1 may also be modulated by structures separate from the 508 site. The identified structural determinants of increased misfolding propensity of NBD1-ΔF508 are essential information in correcting this pathogenic mutant.
| Deletion of a single residue, phenylalanine at position 508, in the first nucleotide binding domain (NBD1) of the Cystic Fibrosis Transmembrane Conductance Regulator (CFTR) is present in approximately 90% of cystic fibrosis (CF) patients. Experiments show that this mutant protein exhibits inefficient biosynthetic maturation and susceptibility to degradation probably due to misfolding of NBD1 and the resultant incorrect interactions of other domains. However, little is known about the direct effect of the Phe508 deletion on NBD1 folding. Here, using molecular dynamics simulations of NBD1-WT, NBD1-F508A, and NBD1-ΔF508, we show that the deletion of Phe508 indeed alters the kinetics of NBD1 folding. We also find that the intermediate states appearing on wild type and mutant folding pathways are populated differently and that their kinetic accessibilities are distinct. Moreover, we identified critical interactions not necessarily localized near position 508, such as Q493/P574 and F575/F587, to be significant structural elements influencing the kinetic difference between wild type and mutant NBD1. We propose that these observed alterations in folding kinetics of mutant NBD1 result in misassembly of the whole multi-domain protein, thereby causing its premature degradation.
| CF is the most common autosomal inherited disease with high morbidity among Caucasians. CF patients have altered epithelial ion transport that leads to decreased hydration of epithelial surfaces in the gut, kidney, pancreas, and airways [1]. Decreased surface liquid volume impairs mucociliary clearance which in turn leads to respiratory bacterial infection [2],[3]. Chronic pulmonary damage caused by bacterial infection dramatically decreases patients' life expectancies. The absence of a functional ABC protein, CFTR, from apical membranes of epithelial cells is the basis of this pathophysiology in cystic fibrosis [4],[5]. CFTR is a multidomain, integral membrane protein containing two transmembrane domains, two nucleotide-binding domains (NBD1 and NBD2), and a regulatory region (R domain) (Figure 1).
Although more than 1,400 mutations are known in CFTR (http://www.genet.sickkids.on.ca/cftr), approximately 90% of CF patients carry the allele with deletion of the codon for phenylalanine at position 508 [6], which is located in the first nucleotide-binding domain (NBD1) (Figure 1). Experimental studies suggest that the CFTRΔF508 may be arrested at two stages during its biogenesis. First, the loss of the Phe508 backbone may shift a fraction of the NBD1s of nascent CFTRΔF508 off the wild type folding pathway, causing misfolding and eventual rapid degradation [7]–[9]. Interestingly, recent studies show no significant structural difference between the wild type and mutant NBD1 structures nor in their thermodynamic stabilities [10]. Second, the absence of the Phe508 side-chain prevents the correct post-translational assembly of all CFTR domains [11]. The detailed structural origin of the perturbed kinetics of NBD1 leading to its co-translational arrest is unknown.
Nucleotide-binding domains of ABC proteins are highly conserved in sequence and structure. NBDs contain a typical F1 ATPase core subdomain, which consists of an α-helix surrounded by antiparallel β-sheets [9],[12]. This region contains the conserved Walker A and B motifs that are involved in binding ATP. The α-helical subdomain contains the ABC-signature motif important for ATP hydrolysis (Figure 1). From X-ray structures of bacterial transporters, the α-helical subdomain is also known to mediate contact with the transmembrane domains [13],[14].
Folding of multidomain proteins is aided by molecular chaperones to prevent and correct improper (non-native) associations between solvent-exposed hydrophobic regions. Smaller single-domain proteins correct and prevent formation of improper contacts through a sequence of partial folding-unfolding events en route to the native state. This sequence of partial folding-unfolding events reflects the ability of single-domain proteins to self-chaperone their folding. In NBD1, the attenuated refolding of the recombinant ΔF508 mutant is consistent with the notion that Phe508 reduces the activation energy of NBD1 folding in vivo as well as in vitro [11]. Lowering of the activation energy increases the folding rate, which in turn reduces the folding time for NBD1. Reduction of the folding lessens the propensity of NBD1 to correct the malformed contacts in the intermediate states. Here we propose that Phe508 deletion decreases NBD1's self-chaperoning capability.
To investigate the effect of the Phe508 deletion on the stability, dynamics and kinetics of NBD1, we performed equilibrium dynamics simulations and folding simulations of NBD1-WT and NBD1-ΔF508. Our analysis shows that there is no significant difference in their stability and equilibrium dynamics, which agrees with experiments. However, even in the presence of correcting mutants (G550E, R553Q, and R555K) [10], [15]–[17] in our model of NBD1-ΔF508, we still observe a significant change in dynamics at the folding transition. We further explore the difference in the folding transition by performing 300 folding simulations each for NBD1-WT and NBD1-ΔF508. We also perform simulations of another mutant NBD1-F508A to serve as control. These simulations allow the comparison of the mutant and wild type folding probabilities, their intermediate states, the structures of these intermediate states, and their folding pathways. Finally, we identify contacts between residues in NBD1 critical to its folding dynamics that are perturbed by Phe508 deletion, thus increasing the propensity of NBD1-ΔF508 to misfold.
To determine the equilibrium dynamics and stabilities of the wild type and mutant NBD1, we perform equilibrium simulations (106 time units∼0.5 millisecond [18]) of wild type and mutant NBD1 using discrete molecular dynamics [19],[20] (see Methods). From the equilibrium simulations, we calculate the thermal denaturation curve of both NBD1-WT and NBD1-ΔF508 (Figure 2) and observe two stable thermodynamic states, folded and unfolded. In agreement with previous experimental studies by denaturation experiments [7]–[9], the stabilities of wild type and ΔF508 NBD1 are not significantly different.
The slope at the transition temperature of the wild type (Tm∼0.68 ε/kB) is 9838 kb and the slope at the transition temperature of the mutant (Tm∼0.70 ε/kB) is 16201 kb (ε∼1–2 kcal/mol and kB is the Boltzman factor; see Methods for further discussion on units). This shift in slope at the transition temperature indicates a difference in folding cooperativity of NBD1-WT and NBD1-ΔF508 and therefore a difference in folding kinetics.
Folding is a stochastic process, thus to investigate in detail the difference in folding kinetics and dynamics of NBD1-WT and NBD1-ΔF508, we perform 300 folding simulations on each of the structures. Starting from fully unfolded chains of NBD1-WTand NBD1-ΔF508, we progressively reduce the temperature of the system to simulate thermal folding (see Methods). We find the folding probability [21] (number of runs that lead to the native structure/number of total folding simulations) of wild type to be 33±3% while that of the mutant is 13±2% (see Methods). The ratio of NBD1-WT and NBD1-ΔF508 correlates with the ratio of their folding yields derived from folding experiments. Folding yields of NBD1-WT is approximately twice that of NBD1-ΔF508 in the temperature range 10°C to 22°C [9]. Folding simulations of our control structure NBD1-F508A yield a folding probability of 26±4% which is intermediate to that NBD1-WTand NBD1-ΔF508. This folding probability value is in agreement with experimental studies showing intermediate folding efficiencies and maturation levels of NBD1-F508A relative to NBD1-WT [9],[11].
To investigate the molecular origin of the difference in folding yields and probabilities, we map the folding pathways of NBD1-WT, NBD1-F508A, and NBD1-ΔF508 by identifying their metastable folding intermediate states. The folding intermediate states of a folding trajectory are exhibited as peaks in the energy probability distributions (Figure 3; Figure S1). Thus, dominant intermediate states in the folding pathways are peaks in the average energy probability distributions (see Methods; Figure 3). The average energy probability distributions of wild type and the mutant are significantly different (Kolmogorov-Smirnov test; P-value<1.4×10−292), which suggests a significant difference in the folding kinetics of wild type and mutant NBD1. The dominant intermediate states are listed in Table S1. The average fraction of native contacts of NBD1 structures in an intermediate state follows a distinct distribution (Figure S2), thus, an intermediate state identified using energy as the folding reaction coordinate, forms a distinct collection of NBD1 conformations.
We find that some intermediate states are accessible only by either NBD1-WT (S6 and S9) or NBD1-ΔF508 (S5 and S10), further suggesting that Phe508 deletion leads the mutant to off-folding pathways (see below). While states S2, S3, S4, S7, and S8 are both traversed by NBD1-ΔF508 and NBD1-WT, their time occupancies (length of time NBD1 spends in an intermediate state) are different (Figure 3B). Since time occupancies are proportional to the free energy barriers between intermediate states, these observations suggest that the Phe508 deletion significantly perturbs the NBD1 folding free energy landscape.
To determine the difference between the sequence of folding events of the wild type, ΔF508, and the F508A control, we estimate the probability of transitions between intermediate states (see Methods and Figure S3). The difference in transition probabilities of NBD1-WT, NBD1-ΔF508, and NBD1-F508A is shown in Figure 4. The transition probabilities show some states accessible only to either wild type or mutant NBD1. The difference in state accessibilities between the two indicates a difference in contact pattern formation (nucleation events), which could cause the observed difference in folding yields.
We calculate the most dominant folding pathways in wild type and mutant NBD1. The most dominant path in wild type follows a sequence of transition Unfolded→S10→S8→S7→S5→S4→S1, while the dominant path in the mutant follows the sequence of transitions Unfolded→S9→S8→S7→S6→S4→S1. Thus, NBD1-WTand NBD1-ΔF508 undergo different sequences of folding events.
Because of the reduction in dimensionality of the folding process when energy is used as a reaction coordinate, each intermediate state represents an ensemble of NBD1 structures. To identify the primary structural characteristics of each intermediate state, we clustered structures in the corresponding state and calculated the frequency of contacts formed between pairs of residues (Figure 5; see Methods). In all intermediate states, we find the most notable structural difference between NBD1-WT and NBD1-ΔF508 occurs in the S7-H6 loop. For example, P574 interacts with Q493 in wild type but not in the mutant. Also, F575 interacts with F587 in the mutant but not in wild type (Figure 6). This pattern of contact formation reflects the difference in NBD1-WT and NBD1-ΔF508 crystal structures that is embedded in the interactions defined according to structure. Additionally, residue pairs that have similar interactions (i.e., attractive or repulsive) in the wild type and mutant crystal structures still exhibit different contacts in the folding intermediate states. These results show that the pattern of transient contact formation in the wild type is also perturbed by Phe508 deletion. This class of residue pairs include Q525/E585 and C524/I586.
We observe a number of folding trajectories reaching native energies (∼630 ε) and within a 2.5 Å root-mean-square deviation (RMSD) with respect to the native structure, but the resulting topological wiring of the secondary structures is incorrect. The “miswiring” consistently occurs in the H5-S6 loop. Interestingly, this H5-S6 loop is in the immediate neighbourhood of the loop containing Phe508. This suggests “weak” regions in NBD1 that are intrinsically prone to misfolding.
To verify that the identified contact pairs (Q493/P574 and F575/F587) found in the S7-H6 loop are indeed critical in the kinetics of NBD1, we revert their interactions in NBD1-ΔF508 to their interactions in NBD1-WT and perform folding simulations. In the case of the Q493/P574 pair, the residues are in close proximity in NBD1-WT but not in NBD1ΔF508, thus we changed the interaction between Q493 and P574 in NBD1ΔF508 from repulsive to attractive to mimic a possible rescuing mutation. Folding simulations of “rescued” NBD1-ΔF508 yield a folding probability of 19±2%. On the other hand, residues F575 and F508 are in close contact in NBD1-ΔF508 but not in NBD1-WT, thus we reverted their interaction in NBD1-ΔF508 from attractive to repulsive. Folding simulations of the second “rescued” NBD1-ΔF508 yield a folding probability of 20±2%. These folding probabilities of the two “rescued” NBD1-ΔF508s are higher than the 13±2% folding probability of the original NBD1-ΔF508, which supports our findings that the contacts between Q493 and P574 and between F575 and F587 are indeed critical to NBD1 folding.
Deletion of Phe508 in CFTR NBD1 is the most frequent mutation in patients with cystic fibrosis. Proteins with ΔF508 mutation in the first nucleotide binding domain NBD1 can not mature resulting in absence of functional CFTR from the plasma membrane. The molecular mechanism leading to this pathological situation is unknown. No significant difference in thermodynamic stabilities was experimentally observed between wild type and mutant NBD1 [7]–[10]. Crystal structures of NBD1-WT, NBD1-F508A, and NBD1ΔF508 are also practically identical except for the S7-H6 loop (Figure 1B) [9],[10],[12]. However, the folding yields of the wild type and mutant were observed to be different [9],[10],[12] suggesting that Phe508 deletion alters the folding kinetics and dynamics of NBD1. Using computational tools, we find that indeed Phe508 deletion causes the mutant to follow a different folding pathway from that of the wild type. We also investigate the molecular origin of NBD1ΔF508 aberrant folding.
Our model agrees with the experimental studies where wild type and mutant NBD1 did not exhibit significant thermodynamic difference. Thermal denaturation curves of both wild type and mutant were calculated from long equilibrium dynamics simulations (∼0.5 millisecond) (Figure 2). We find that there is no significant difference in the stabilities of wild type and mutant NBD1, which agrees with experimental observations showing unaltered thermodynamic stability upon Phe508 deletion from NBD1 [9],[10],[12]. The thermal denaturation curves likewise show a slight shift in the melting temperature, which agrees with the minor change in melting temperature from 49°C to 46°C upon deletion of Phe508 [8]. The agreement between our computational results and experimental studies validates our model and methodology. The shift in Tm reflects the attenuated folding of NBD1, consistent with the notion that Phe508 reduces the activation energy of NBD1 folding in vitro and in vivo [11]. These observations do not contradict biochemical and functional measurements that show rescued complete ΔF508 CFTR has a temperature-sensitive stability defect in post-ER compartments [22]. While the stability of NBD1 may be minimally perturbed upon Phe508 deletion, the impaired interaction of NBD1 with the rest of CFTR could still destabilize the whole CFTR. Misfolding of NBD1 may prevent its proper interaction with other CFTR domains.
Interestingly, even a simplified protein model (4-bead representation for non-aromatic residues and 5-bead for aromatic residues) and a simplified potential (Go̅-type interaction) show a significant difference in NBD1-WT and NBD1ΔF08 kinetics. Indeed the multiple folding simulations of NBD1 models show significantly higher folding efficiency for the wild type than the ΔF508 mutant, which correlate with some experimental studies that found higher folding yield for wild type NBD1 than its ΔF508 mutant [8]. A relevant control of our simulation protocol and modeling assumptions is the folding simulation of the NBD1-F508A that yields a folding probability of 26±4%, which is higher than that of NBD1ΔF08 but lower than that of NBD1-WT. This observation again correlates with the measured folding efficiencies of isolated NBD1 and maturation levels of whole CFTR [9],[11].
We also find that the folding time of NBD1ΔF508 mutant is smaller than that of wild type, suggesting an increase in the effective folding rate upon Phe508 deletion. Consistent with the notion that the refolding of recombinant ΔF508 mutant reduces the activation energy of NBD1 folding in vivo as well as in vitro [11]. The reduction in folding time diminishes the propensity of NBD1 to correct and prevent malformed contacts in the intermediate states. Thus, mutant NBD1 has a diminished self-chaperoning activity.
What is the origin of this loss in self-chaperoning capacity by the Phe508 mutant? To answer this question, we identify and compare the folding intermediates accessed by wild type and mutant en route to the native state. Calculation of transition frequencies from one state to another reveal differences in the accessibilities of intermediate states (Figure 5). Some intermediate states are only accessible in either wild type or mutant. These observations suggest that the progression of contact formation is different in the two structures. We also determine the primary differences in the progression of contact formation by calculating the most frequent contacts formed within the intermediate states. Between wild type and mutant NBD1 intermediates, the most notable differences are in the interaction between amino acid pairs F575/F587 and Q493/P574. Contact between Q493 and P574 is consistently formed in the intermediate states of the wild type but not of the mutant (Figure 4). Interestingly, the P574S mutation has been observed in a CF family also possessing the ΔF508 mutation but without significant pulmonary or pancreatic disease. The solubilizing F494N mutation, which is adjacent to Q493, has also been shown to partially correct the folding defect of CFTR-ΔF508 [23]. The mutations P574S and F594N may promote contact formation between Q493 and P574 during NBD1ΔF508 folding, thus rescuing NBD1ΔF508 from the misfolding defect. Interestingly, Thibodeau et al. found that NBD1F508W, the only F508X mutant with a lower folding efficiency than NBD1ΔF508, can be rescued by introducing the compensating mutation W496F, which is exactly in the same loop that contains Q493 and F594 [9].
A drawback that may arise from using Go̅ is that the properties of a protein are determined solely by its geometry, an assumption that apparently deviates from the observation that sequence is also a key determinant of folding properties. However, this potential drawback is not limiting in our study of the folding kinetics of wild type NBD1 and its mutants. The nuanced effect of a mutation or deletion at position 508 is already reflected in the S7-H6 loop conformation of NBD1-WT, NBD1-F508A, and NBD1ΔF508 crystal structures.
Changes in NBD1 folding kinetics have been shown earlier experimentally. Qu et al. observed dramatic changes in the temperature sensitivity of the folding process in the absence of Phe508 [8]. In the temperature and protein concentration range used in the refolding experiments, mutant NBD1 reached the native state less efficiently compared to wild type. The mutant NBD1 aggregated faster and to a larger extent, as observed by light scattering of the samples. In contrast, Lewis et al. observed no difference in folding of wild type and ΔF508 NBD1 monitored by CD spectroscopy or intrinsic Trp fluorescence [10]. This apparent disparity may be due to a substantial difference in experimental conditions of these two refolding experiments. While the former laboratory performed these experiments at different temperatures incubating the samples overnight, the latter made measurements immediately after dilution of the denaturant over a timescale of minutes. In the experiments of Qu et al. [7],[8], there is a possibility that the two constructs have different solubilities. On the other hand, Lewis and coworkers' studies of folding kinetics were carried out only at room temperature, where the differences between the refolding of the two different domains are already attenuated compared to lower temperature (10°C–16°C) [10].
Our results reveal the intrinsic property of NBD1ΔF508 to fold improperly and raise the possibility of redesigning NBD1ΔF508 to rescue it from misfolding. In case of the contact that is found in wild type but not in the ΔF508 mutant (e.g., Q493/P574), one can find amino acid substitutions that promote interaction between this pair of residues (Q493/P574). On the other hand, for the contact found only in the ΔF508 mutant but not in wild type (e.g., F475/F587), candidate rescue mutants are those that destabilize the interaction between this residue pair (F475/F587). Knowing the molecular details of the altered folding in the case of the mutant domain also provides a basis for design of small molecules to correct the most prevalent and pathogenic mutation in CFTR.
To access time scales of NBD1 folding, we use a simplified protein model but still maintain important features of the protein such as side-chain packing. Amino acid residues were modelled as follows: (1) glycines are represented by three beads (-N, Cα, C′); (2) phenylalanine, tyrosine, tryptophan, and histidine by five beads (-N, Cα, C′, Cβ, Cγ), and (3) all other residues by four beads (-N, Cα, C′, Cβ) [24]. This protein model successfully described protein aggregation [24]. In the simulations, we use PDB ID: 2BBO, PDB ID: 1XMI and PDB ID: 1XMJ [10] as models of NBD1-WT, NBD1-F08A, and NBD1-ΔF508, respectively. The missing loop between E403 and L436 in both wild type and mutant NBD1 is reconstructed using a loop-search algorithm in SYBYL (Tripos Assoc. Inc, St. Louis, MO).
Using discrete molecular dynamics [19],[20], long equilibrium simulations at various temperatures were performed to investigate the equilibrium dynamics of the CFTR NBD1. Interactions between beads were defined using the Go̅-model [25]. In the Go̅-model, interactions between residues are determined from the native structure of known NBD1 crystal structures. Pairwise, square-well interactions were assigned between beads in the model according to contacts formed in the native state. Specifically, two residues are said to be in contact if their atoms (excluding hydrogen) are within a distance of 4.5 Å. The strength of the interaction between residues in contact (denoted hereon as ε) defines the energy units. Physically ε∼1–2 kcal/mol, which is approximately a contribution to protein stability from a hydrogen bond. The temperature is measured in units of ε/kb, where kb is the Boltzmann constant. The time unit (tu) is estimated to be the shortest time between particle collisions in the system (∼0.1 nanosecond). From long equilibrium simulations of 106 tu, we were able to access the long time-scale dynamics of the CFTR NBD1 in the order of 0.5 millisecond. Each equilibrium simulation consumed approximately 300 CPU hours.
We perform 300 folding simulations for each NBD1-WT, NBD1-F508A, and NBD1-ΔF508. Starting from fully unfolded chains, the temperature of the system is progressively reduced to allow NBD1 to fold to its native structure. Folding simulations proceeded until τmax∼60,000 tu (time units), which is chosen to be longer than the typical folding time of the studied sequences [21]. A similar criterion was employed in the studies calculating the folding probability of proteins [26]. The NBD1 structure in a folding run is considered folded when (1) its energy is less than or equal to −620ε (the energy of the native state), (2) its structure is within 2.5 Å RMSD from the native, and (3) the structure possesses correct topological wiring of the secondary structure elements. To estimate the error in folding probabilities, each folding trajectory is considered a Bernoulli trial with a binary outcome, folded or unfolded. The variance of a Bernoulli process is σ2 = p(1−p)/n, where p is probability and n is the total number of trials.
To identify the positions of intermediate states, a sum of multiple Gaussian curves is fitted to the average energy probability distribution of successfully folded runs. ai, bi, and ci are the center, standard deviation and height of the ith Gaussian curve, respectively.
We estimate probability of transitions between states by counting the trajectories that underwent such transition. The sum of probabilities of paths emanating from a given state is normalized to 1, which physically means that the system always exits from its current intermediate state. The transition probabilities represent independent conditional probabilities, thus the most likely path from the unfolded state to the native is estimated by multiplying the probabilities of the traced edges.
We calculated a contact matrix for each structure in the intermediate state. An element of the contact matrix is 1 when two residues were within 4.5 Å or 0 otherwise. Dominant contacts between pairs of residues in NBD1 are determined from the average contact matrix of all the structures. |
10.1371/journal.pcbi.1004732 | PSAMM: A Portable System for the Analysis of Metabolic Models | The genome-scale models of metabolic networks have been broadly applied in phenotype prediction, evolutionary reconstruction, community functional analysis, and metabolic engineering. Despite the development of tools that support individual steps along the modeling procedure, it is still difficult to associate mathematical simulation results with the annotation and biological interpretation of metabolic models. In order to solve this problem, here we developed a Portable System for the Analysis of Metabolic Models (PSAMM), a new open-source software package that supports the integration of heterogeneous metadata in model annotations and provides a user-friendly interface for the analysis of metabolic models. PSAMM is independent of paid software environments like MATLAB, and all its dependencies are freely available for academic users. Compared to existing tools, PSAMM significantly reduced the running time of constraint-based analysis and enabled flexible settings of simulation parameters using simple one-line commands. The integration of heterogeneous, model-specific annotation information in PSAMM is achieved with a novel format of YAML-based model representation, which has several advantages, such as providing a modular organization of model components and simulation settings, enabling model version tracking, and permitting the integration of multiple simulation problems. PSAMM also includes a number of quality checking procedures to examine stoichiometric balance and to identify blocked reactions. Applying PSAMM to 57 models collected from current literature, we demonstrated how the software can be used for managing and simulating metabolic models. We identified a number of common inconsistencies in existing models and constructed an updated model repository to document the resolution of these inconsistencies.
| The broad application of genome-scale metabolic modeling has made it a useful technique for tackling fundamental questions in biological research and engineering. Today over 100 models have been constructed for organisms that carry out a diverse array of metabolic activities spanning all three kingdoms of life. These models, however, have been curated independently following different conventions. The maintenance of model consistency has been challenging due to the lack of consensus in model representation and the absence of integrated modeling software for associating mathematical simulations with the annotation and biological interpretation of metabolic models. To solve this problem, we developed a new software package, PSAMM, and a new model format that incorporates heterogeneous, model-specific annotation information into modular representations of model definitions and simulation settings. PSAMM provides significant advances in standardizing the workflow of model annotation and consistency checking. Compared to existing tools, PSAMM supports more flexible configurations and is more efficient in running constraint-based simulations. All functions of PSAMM are freely available for academic users and can be downloaded from a public Git repository (https://zhanglab.github.io/psamm/) under the GNU General Public License.
| The GEnome-scale Models (GEMs) of metabolic networks have broad applications in biological research and engineering [1]. Models have been developed for organisms of all three kingdoms of life [2–5] and have been used to simulate a wide variety of metabolic processes, such as photo- and chemo-autotrophic carbon fixation [6,7], fermentation [8], and the production of specific organic compounds [9]. GEMs can be applied in theoretical research to predict gene essentiality [10,11], simulate the thermo-tolerance of bacterial strains [12], and study the structural and functional evolution of metabolic networks [13]. They can also be used in practical studies to identify drug targets [14,15], illustrate the mechanism of human diseases [16], and to optimize the production of compounds of industrial significance [17–19]. By connecting genome annotations with the mathematical simulation of reaction networks, GEMs are particularly applicable for integrating heterogeneous datasets from high-throughput studies [20], such as the profiles of transcriptional regulation [21,22] and measurement of carbon isotope labeling [23].
Specialized MATLAB toolboxes have been released over the past two decades to support the mathematical simulations of metabolic networks. The COBRA Toolbox is a collection of widely used, open source tools. It includes diverse implementations of constraint-based modeling algorithms and has attracted a large number of user contributions from the modeling community [24]. This toolbox, however, is restricted to the MATLAB environment and requires users to maintain paid licenses from MathWorks Inc. The COBRApy software is a more recent implementation of the COBRA Toolbox functions using the Python programming language and Jython, a Java implementation of Python [25]. Like the COBRA Toolbox, COBRApy is released as a toolbox rather than a software package. Therefore, knowledge about the Python programming language is required for users to efficiently set up operations in COBRApy.
Other tools have been developed to support the annotation and visualization of metabolic networks. ModelSEED is a web-based platform that supports automated reconstruction of metabolic models from genome annotations [26,27]. It links protein functions with an internal reaction database and is associated with the SEED functional annotation database and the RAST genome annotation pipeline [28]. In contrast to the COBRA Toolbox, this platform is focused on the reconstruction instead of the mathematical simulation of GEMs. The automated pipeline of ModelSEED permits direct construction of a draft model from genome annotation. However, the draft model still requires extensive manual curations, and manually editing the draft models is not currently supported [27]. The RAVEN Toolbox supports semi-automatic reconstruction and visualization of genome-scale models [29]. It uses information from the KEGG database [30] and, similar to the COBRA Toolbox, can only be applied under the MATLAB environment. Finally, Pathway Tools is another software package that supports pathway annotation and visualization [31]. It uses pathway information from the MetaCyc database [32] and has recently been extended to include functions for flux balance analysis. Unlike the open source software mentioned above, Pathway Tools is released under a restricted license agreement, but it is free to use for academic users.
Despite the technological advances in supporting individual steps of network reconstruction and mathematical simulation, challenges remain in maintaining the quality of metabolic reconstructions and in associating the representation of mathematical problems with the annotation and biological interpretation of metabolic models. Human intervention and iterative manual examinations are required in various steps, such as defining biomass functions, assigning reaction directionality, establishing boundary conditions, and filling novel reaction gaps [26,33]. Frequently, these processes are carried out either manually or using customized tools that were developed by individual model curators. However, due to the absence of standardized procedures for quality checking and version tracking of iterative model annotations, the curation and manual editing of metabolic models are prone to introducing inconsistencies [34]. These inconsistencies may lead to false predictions of model viability, and as demonstrated in a recent study, may hinder the evaluation of new modeling approaches [35–37].
Another problem that prevents the effective curation and consistency checking of GEMs is the disassociations between mathematical and biological representations of model metadata. Despite a broad adoption of the SBML format, it is not designed to incorporate detailed annotations of the genes, reactions, compounds, and pathways. Two strategies have been implemented to help address this limitation. The first is to use user-defined, model-specific or toolbox-specific tags in SBML files (e.g. the <notes> tag in COBRA-compliant SBML format). These tags are not part of the standard SBML specifications and are not recognized by the standardized SBML parsers (e.g. libSBML [38]). The second strategy employs tables of annotation data. These tables are customized to represent specific models, but they are disconnected from the simulation problems defined in SBML files, and an automated parsing of these tables is impossible due to the lack of convention in table organizations. In either case, the mathematical representation of simulation problems is separated from the biological annotation of model components. As a result, the consistency between genome annotation and the definition of simulation settings have to be maintained by individual model curators. This again is prone to introducing misrepresentations in GEMs.
An additional disadvantage of the SBML format is the lack of modularity in network representations. The SBML definition of GEMs is composed of two major lists: listOfSpecies, which defines all of the metabolites in a GEM; and listOfReactions, which defines all of the reactions and their simulation settings in a single modeling state, including flux bounds, flux values, and objective coefficients. Since the definition of simulation settings interleaves with the definition of static reaction features, SBML does not support reusing the static model definitions. Anytime a new simulation is released on the same base model, information about the static features of metabolites and reactions will need to be duplicated in a new SBML file. Hence, the SBML format is not compatible with a modular organization of modeling components because it mixes the static model properties with the dynamic simulation settings.
Here, we present a Portable System for the Analysis of Metabolic Models (PSAMM), an open source software package that was implemented to support the iterative curation of GEMs by connecting model annotations with mathematical simulations. A novel model format was developed in PSAMM using the YAML language to integrate detailed annotations of model components and to provide flexibilities in both the data format (i.e. support direct referencing of annotation tables or the YAML-based list format) and the data content (i.e. support parsing of both standardized and model-specific data fields). The new YAML format assembles model definition and simulation settings into independent and reusable modules (S1 Text). By reducing the amount of characters used for data structure definition, it streamlines the content of the real data and enables tracking and management of GEMs with conventional line-based version control systems like Git [39]. Applying PSAMM, we constructed an updated repository of 57 published models using the modular representation of the YAML format, and we further demonstrated the importance of model formatting and consistency checking through identifying and correcting potential inconsistencies in existing GEMs.
PSAMM was developed as an integrated software package that supports calling simulation functions using simple one-line commands. This distinguishes PSAMM from existing metabolic modeling toolboxes, which require the users to go through individual steps of model loading, linear programming solver selection, model optimization, and simulation results mapping. An example of PSAMM commands is given (S1 Fig), in which a thermodynamically constrained flux balance analysis (tFBA) was performed through a simple one-line command. The result of the analysis contains metadata about the model name, model version, objective function, reaction fluxes, and reaction equations. In contrast, it takes three steps to set up a tFBA problem in COBRA, and the output is presented in a complex data structure that is dissociated from the biological interpretations of the model. The PSAMM implementations were also much more efficient than the COBRA toolbox functions. While it took only 7.15 seconds to solve a tFBA problem in PSAMM, the same process on the same model took 194.73 seconds in COBRA.
A comparison of the functions in the PSAMM software package with the functions in the COBRA and RAVEN toolboxes is provided in the S1 Table. Besides the availability of diverse functions and implementations of model import/export, model checking, and constraint-based analysis, a unique feature of PSAMM is the availability of easily accessible help information that assists the users in selecting program functions and parameter values. This feature is enabled because PSAMM integrates the various functions under a common framework of two universal programs, psamm-model and psamm-import. The help information can be accessed in PSAMM by adding the -h or --help option following any program functions listed in the S1 Table. In contrast, RAVEN and COBRA do not provide an integrated interface to the available functions but instead require the users to know the names of individual functions in order to retrieve help information. Moreover, to systematically evaluate the efficiency of PSAMM implementations, we recorded the running time of common operations in PSAMM and COBRA on the Escherichia coli model iJO1366 [40] (Materials and Methods). The results suggested that PSAMM is overall much more efficient than COBRA (Fig 1). For the computationally intensive functions like flux variable analysis (FVA), robustness analysis (Robustness), and thermodynamically constrained flux balance analysis (tFBA), PSAMM ranges from 9 to 25 times faster than implementations in the COBRA toolbox. For the FBA function, PSAMM appears to have an overhead of about 3 seconds on top of the time required for the problem-solving step (Fig 1 inset). This overhead is for the reading of YAML model files, which is also included in calculating the running times of FVA, Robustness, and tFBA, but solving FBA problem in PSAMM (0.03 seconds) is slightly more efficient than the same step in COBRA (0.08 seconds). The running time for RAVEN was not plotted in Fig 1 because many of the functions are unavailable in the RAVEN toolbox (S1 Table). A detailed record of the running times for available functions in COBRA, RAVEN, and PSAMM is provided in the S2 Table. Overall, PSAMM is more efficient than the COBRA and RAVEN toolboxes in carrying out constraint-based analysis of GEMs.
The PSAMM software package includes five main components: user interface, model input/output, internal model representations, model checking/simulation, and linear programming utilities (Fig 2). These components are interconnected with one another to form the internal workflow of model importing and model optimizations in PSAMM. Below, we provided a detailed description of these functions.
A new model format was implemented in PSAMM based on YAML, a serialization language that is designed to support both computational parsing and human readability of complex data structures [50]. The YAML format has several new features that distinguish it from the SBML format of constraint-based metabolic models. First, while SBML strictly couples the model definition with a single simulation condition, YAML allows the users to freely combine model definitions with simulation setups in a modular form. Hence, a model component can be reused in multiple simulation setups simply by using the include function (Fig 3 and S1 Text). In contrast, to publish the simulation results of multiple modeling conditions in the SBML format, one has to build duplicated instances of the entire model definition to introduce variability in simulation setups. Second, while SBML is suboptimal in working with the widely-used version control systems like Git [51], the YAML format is fully compatible and supports tracking of model edits with these systems. This is because YAML, by applying line breaks and whitespaces as meaningful structural information, can be tracked in a way similar to programming scripts using the longest common sequence (LCS) algorithm that is commonly used by Git [52]. In contrast, SBML does not have a convention of breaking lines, and therefore the LCS line-based version tracking will have difficulties in pinpointing the exact location of changes in the SBML files. Despite the development of new algorithms to support the comparison of SBML model files, the efficient and reliable tracking of changes in the SBML format is still challenging [34,51]. Third, while SBML relies on the verbose representation of markup tags, the YAML format minimizes the amount of characters used for marking up the structure of the data to streamline the integration of complex data (Fig 3 and S1 Text). Finally, while the parsing of SBML files requires knowledge of predefined markup tags to capture specific data fields, the parsing of YAML files can be achieved by the recognition of line-based data structure rather than relying on the identification of specific markup tags. For example, some features like the compound formula and reaction EC numbers are not an integral part of standardized SBML specifications [41], so they will be invisible to standard SBML parsers (e.g. libSBML [38] or JSBML [53]) even when they are included in a SBML file with user-defined markup tags. However, the same information can be easily represented in YAML files. A standard YAML parser will have no problem recognizing user-defined, non-standard data domains and subsequently integrate it into the presentation of metabolic models. Overall, the YAML format is an optimized solution for connecting manual curation with the mathematical simulation of GEMs.
The PSAMM YAML-based representation of GEMs includes the model name, biomass function, compound/reaction database(s), model reaction lists, reaction flux limits, and growth media. A flexible infrastructure was designed that permits the integration of model definitions either within a centralized file (e.g. model.yaml) or in additional files using the include function (Fig 3a). In an example shown in Fig 3a, the reaction database was divided into multiple files based on pathway classification, and both the YAML format (e.g. glycolysis.yaml and biomass.yaml) and the tab-delimited table format (e.g. tca-cycle.tsv) were illustrated. Similarly, the compound annotations can also be represented as YAML files (e.g. compound.yaml) or tab-delimited tables. The reaction and compound files can be used to define a broader database of all possible reactions with additional entries beyond the scope of a given model. For example, when running the gapfill or fastgapfill functions, users may include a broader reaction database that contains enzymatic reactions that can be potentially applied to close network gaps. In such cases, the model file (e.g. model_rxn.tsv) is used to identify a subset of reactions that are included in the model definition. The other reactions in the broader database will be ignored in model simulations and will only be probed in gapfill or fastgapfill to predict potential gap filling reactions. Finally, to define the constraints for metabolic simulations, the media file (e.g. media.yaml) is used to identify the exchange reactions, and the limits file (e.g. limits.yaml) is used to set the flux limits for internal reactions. The limits file is defined only when the flux bounds of an internal reaction deviate from the defaults in PSAMM, which automatically assigns a conventional boundary of [−x,x] for reversible reactions and [0,x] for irreversible reactions, where the value x is assigned using an option named default_flux_limit in the model.yaml file. When default_flux_limit is not assigned, PSAMM will use a default value of x = 1000 to define reaction bounds (S1 Text). This feature permits further streamlining of the YAML format to reveal model specific simulation setups. Although PSAMM supports references to model files with user-defined file names that are saved in distributed locations in the file system, it is recommended that users save all files of an individual model within a dedicated directory to facilitate model organization and maintenance. The users may also choose to combine all the model information into a single YAML file by replacing the include functions with the actual content of the data files. However, the include functions are recommended to maintain modularity of model definition and simulation settings in the YAML representation.
Using the YAML format, annotations of compound and reaction features can be further divided into multiple files that represent logical divisions of cellular compartments or metabolic pathways. Moreover, The annotations can be represented in either a YAML format or a tab-delimited table format. These flexibilities can enable the manual curation of model files and make the model components portable among different simulation conditions. The line-based text representation of YAML files also makes it compatible with broadly-used version control systems like Git, and the YAML format is especially efficient at tracking changes in large reaction equations, such as the biomass function and the protein, RNA, and lipid synthesis functions (Fig 3b). In contrast, the Excel format is not compatible with version control due to the binary form, and the lack of a standard model definition in Excel files also detracts from its suitability as a GEM format. The SBML files, in spite of a well-defined data structure, do not use line breaks or white spaces as a part of the data structure and hence are suboptimal in working with Git [34]. For example, each list item in the SBML file is frequently represented as a single line that contains multiple data entries describing this item. This will render the version control in Git ineffective because changes on any one entry will cause the version tracking system to highlight an entire line of multiple data entries, and it is difficult for a user to pinpoint the exact location of the change within a long line of data. Moreover, since the markup tags are included as a part of the data presentation, changes in the markup tag definitions during the periodical updates of the SBML specifications will show up in the tracking even though there is no change in the model data.
To test the application of PSAMM in supporting model annotations, we applied the software to 57 published GEMs collected from current literature or open-source model releases (S3 Table). These included the RECON2.04, which was released in the format of a MATLAB data file. All models were first converted to the PSAMM-specific YAML format and then analyzed individually using the stoichiometric and flux checking functions. FBA simulations were also performed using designated biomass function to verify if the published descriptions of model viability can be replicated. When inconsistencies occurred, the causes were identified by manually inspecting the corresponding models. Overall, PSAMM revealed several types of inconsistencies among existing models. These included inconsistencies in the syntax of SBML files, the annotation contents of Excel tables, the stoichiometric balance of metabolic reactions, and the connectivity of metabolic reactions. Below, we provide an overview of the existing inconsistencies identified by PSAMM.
The reconstruction and analysis of GEMs have broad applications in understanding genotype-phenotype connections, studying evolutionary processes, and predicting the outcomes of metabolic engineering [13,81]. Despite the development of tools that support individual steps along the modeling procedure [24,26,29], it is still challenging to associate mathematical simulation with the biological interpretation of GEMs. Often times, model files need to be curated iteratively to set up exchange reactions, assign reaction directionality, define biomass functions, and fill reaction gaps, and the results of mathematical simulations need to be mapped to the annotated data of model definitions. However, due to the absence of integrated data representations and the lack of standardized model checking procedures, the curation of GEMs are prone to introducing inconsistencies, which in turn may lead to errors in modeling and in interpreting the biological meaning of modeling outcomes.
The PSAMM software package was developed to solve the above challenges by integrating the annotation of metabolic pathways with the consistency checking and the constraint-based analyses of mathematical models (Fig 2). It offers a novel YAML format of model representation that integrates the heterogeneous annotation of model components with the formulation of mathematical simulation problems. PSAMM takes advantage of several useful features commonly supported in the Python programming language: (1) it is highly portable and can be easily installed using the pip package management system; (2) it is open source and does not rely on paid software like MATLAB; (3) it has extensive library support that permits both matrix-based operations and dictionary-based feature representations; (4) it is an integrated package that supports calling simulation functions using simple one-line commands (S1 Fig) and permits adjustments to modeling parameters using command options (S1 Table). PSAMM supports the import and export of diverse model formats and is applicable for the curation of draft models generated from model reconstruction pipelines like ModelSEED [26], RAVEN [29], and Pathway Tools [31]. It provides extensive help information (e.g. through the command-line -h or --help options) and an online tutorial that demonstrates the main functions and workflow of PSAMM (https://psamm.readthedocs.org/en/latest/tutorial.html). Additionally, the PSAMM functionality is easily expandable by the user community through an Application Programming Interface (API).
PSAMM integrates heterogeneous metadata in model annotations with the simulation of metabolic networks, so that users can perform model curation and mathematical simulations under the integrated framework of a unified software package. While COBRA and RAVEN are released as toolboxes and are dependent on the MATLAB software, PSAMM is an independent software package that is freely available for all academic users. Calling the simulation functions is largely simplified in PSAMM because the individual steps of model reading, solver selection, model simulation, and simulation results mapping are integrated into a single one-line command. In contrast, setting up a simulation problem in COBRA would require calling multiple functions in the MATLAB interface (S1 Fig). Moreover, PSAMM allows users to document their modifications to the model definition as well as the simulation conditions during model curations. In contrast, changes made through the MATLAB interface will be lost when the computing session is closed. Even if users write out their changed models into an SBML file, it will be difficult for the users to track the exact location of the changes due to the incompatibility of SBML files with line-based version control systems [51,82]. Finally, PSAMM is configured to perform with high efficiency a number of computationally intensive simulations, such as tFBA, FVA, robustness analysis and random minimal networks (Fig 1 and S2 Table). As an example, we recorded the running time of robustness analysis on iJO1366, which is one of the larger models in the collection we analyzed. Using a single computing thread, the median runtime of robustness analysis with 1000 steps on the EX_o2(e) exchange reaction using Cplex solver was 4.07 seconds in PSAMM (including the reading of model files), whereas COBRA takes 100.7 seconds under the same settings for just solving the simulation problem. This is because PSAMM modifies the simulation problem directly through an optimized LP utility interface (Fig 2), while COBRA requires the simulation problem to be redefined for every step of the robustness analysis.
The PSAMM YAML-based model representation provides additional support to model curation by offering a human-readable interface for the organization of heterogeneous data (Fig 3 and S1 Text). The YAML format incorporates information about both model annotations (e.g. in the reactions, compounds, and model definitions) and model simulations (e.g. in the biomass, media, and limits definitions), and it supports a flexible interface for storing annotation information either within a centralized model file or as independent feature files. The YAML format streamlines model representation by minimizing the use of markup tags. The use of line breaks and whitespaces as markups of data structure not only enables the parsing and association of user-defined, model-specific model annotations, but also enhances the compatibility of YAML with line-based version control systems. This compatibility with version tracking is especially useful in collaborative projects that involve multiple curators working on the same model since modifications made by different curators can be documented in parallel and later reconciled and merged into a new model release that is more comprehensive than what could be achieved by a single curator. Additionally, YAML supports the modular representation of diverse model components and the free combination of model definitions with diverse simulation conditions (Fig 5). Hence, instead of being restricted to a single simulation condition, YAML provides additional flexibilities for the documentation of multiple modeling conditions in a single model release.
Applying PSAMM to existing GEMs in public literature, a number of inconsistencies were identified. These included inconsistencies in the formatting of model files (e.g. SBML or Excel files), the definition of model stoichiometry, and the presence of blocked reactions. In general, the representation of SBML and Excel files varies among different GEMs both in the formatting of model annotations and in the completeness of information provided. For example, only a small fraction of the collected SBML models included complete metadata regarding protein-coding genes, functional annotations, Enzyme Commission (EC) numbers, and subsystem classifications. This was largely due to the lack of standardized definition of such information in SBML specifications and to some extent has prevented the SBML format from integrating heterogeneous annotation data. The Excel files, although frequently used by model curators for the annotation of pathway information, also lacks standardization and data integration. These limitations have caused a number of common inconsistencies in the representation of both SBML and Excel models. Using the annotation framework of PSAMM and the integrated YAML representation, we have corrected these inconsistencies and documented the changes in an online Git repository at https://github.com/zhanglab/psamm-model-collection (S2–S5 Text).
To ensure that the GEMs correctly represented the function and connectivity of metabolic networks, the PSAMM stoichiometric and flux consistency checking were applied to search for unbalanced or disconnected reactions. Surprisingly, only 30% of the analyzed models were stoichiometrically balanced, and the unbalanced reactions can be attributed to the inconsistent representation of compounds in different reaction equations. The analysis of flux consistency revealed that while the central metabolic pathways contain only a small fraction of blocked reactions across all GEMs, the pathways of lipid metabolism, glycan metabolism, and cofactors and vitamins metabolism were largely variable among different GEMs and were frequently disconnected from biomass production. Interestingly, the pathway distribution of metabolic reactions varied among different GEMs (S2 Fig). In the successive models of certain organisms (e.g. in E. coli), it appeared that the initial reconstructions have focused on central metabolic pathways, while the later iterations have contributed to the inclusion and completion of peripheral pathways like the transport reactions and the glycan and lipid metabolism. While it is not uncommon to have blocked reactions in GEMs due to the limited knowledge in current literature about certain metabolic pathways, the analysis of flux inconsistency should be a standard step especially in interpreting the biological significance of gene or reaction deletion simulations. Since the blocked reactions do not contribute to the production of biomass, they will always be predicted as non-essential reactions. However, such prediction is not biologically meaningful because it reflects an artifact caused by the disconnection of metabolic pathways.
A list of functions (marked in S1 Table) was examined in the PSAMM package as well as the COBRA and RAVEN toolboxes to compare the efficiency of the different tools. The simulations were carried out on the model iJO1366 [40] with the Cplex linear programming solver for PSAMM (version 0.17) and COBRA (version 12.6, IBM academic release for PSAMM and TOMLAB release for COBRA), and with the MOSEK solver for RAVEN (version 7.1.0.36). The specific simulation parameters were set according to the conditions specified for each function in the S2 Table. Outputs from different tools were manually examined to ensure that the same results were generated form the same simulation settings. The running time was recorded in a CentOS operating system with a single processor core allocated to each program (the solver was not restricted in this way and was able to use up to 20 cores). Each simulation was carried out at least seven times, from which the median was plotted in Fig 1 with the 75th and the 25th percentile values as the upper and the lower limits of the error bar.
PSAMM has the following dependencies, some of which are required and some of which are optional but limit the functionality of PSAMM in their absence. The dependencies PyYAML, xlrd, xlsxwriter, and NumPy can be automatically installed through the Python package manager pip. The linear programming solvers need to be installed by the user following instructions from corresponding releases.
A list of 56 GEMs was collected from current literature and a public model collection [84] as SBML files. Additionally, the RECON2.04 model was downloaded in the format of a MATLAB data file from a public release at https://vmh.uni.lu/#downloadview. Compound and reaction features of nine models were downloaded in the Microsoft Excel format (S3 Table). These models were used to test the application of PSAMM for model management and model curation. For constraint-based simulations, a biomass reaction was identified for each model by examining the objective coefficient defined under the SBML “kineticLaw” section. The reaction with a non-zero objective coefficient was treated as the biomass reaction. If the objective function was not specified in SBML file, the reaction identifiers were searched and the ones containing the string “biomass” were marked as biomass reactions. Information was also obtained from original publications of the collected models to identify biomass reactions and to select the main biomass reaction when more than one was present. The biomass reactions of each model were listed in S3 Table.
The stoichiometric consistency check was implemented in the PSAMM masscheck function, which supports both compound-based and reaction-based checking of the stoichiometric consistency. By default, the masscheck function excludes the exchange and biomass reactions from consideration because by design, stoichiometric balance is not required in these reactions. The compound-based stoichiometric consistency checking was implemented based on Thiele et al. [49] and was formulated as:
Maximize Σi zi
Subject to
This problem looks for compounds that failed to obtain a positive mass under optimizing conditions. By maximizing the values in z up to 1 for as many compounds as possible, it forced the mass values (m) to be at least 1 for the same compounds. The sum of the values in z can be used to approximate the number of compounds that were properly balanced.
Although the compound-based checking was useful for identifying compounds that may cause stoichiometric inconsistency, it provides no insights into the stoichiometric balance of individual reactions. Therefore, PSAMM implements a new algorithm to directly search for unbalanced reactions. This approach is performed using an LP problem that was modified from the above stoichiometric consistency check. A mass residual variable (r) is introduced into every reaction and bounded by the residual bound variable (z) such that rj ∈[−zj;zj]. The residual bound variables are then minimized, and the reactions that carry a non-zero residual value rj (i.e. bounded by a positive zj) are reported as a candidate of inconsistency. The mathematical formulation of this approach is described in the following LP problem:
Maximize Σj zj
Subject to
This approach can be applied iteratively to assist with the correction of stoichiometric inconsistencies in metabolic models. The residual variables indicate a mass that is missing from identified reactions, and the sign of the residual can be used to determine whether the left or right side of an equation has a missing mass. Additionally, PSAMM provides a “--checked” option in the masscheck function, which allows fixing certain residual values at zero when the corresponding reactions have been confirmed to be balanced, making the program able to converge on the remaining set of unbalanced reactions. When the stoichiometric consistency check was performed on the collection of models, the biomass and exchange reactions were excluded from consideration and the requirement of a positive mass was removed from compounds that were used to model photons or electrons.
The flux consistency check was implemented in the PSAMM fluxcheck function. A flux inconsistent reaction is a reaction that cannot take a non-zero flux under any simulation conditions. In other words, given the stoichiometric matrix S, the vector of fluxes v, and the reaction j, if no solution to Sv = 0 exists, where vj ≠ 0 the reaction is considered to be flux inconsistent. Conversely, if a solution exists the reaction is considered to be flux consistent. A consistent model is a model that only contains consistent reactions [85]. The flux consistency check in PSAMM was implemented based on two independent approaches. The first approach was based on the definitions in [86]. This was achieved by applying a modified version of the flux variability analysis (FVA) on each reaction while using default constraints on other reactions. For reversible reactions, the flux is first maximized and then minimized, and for irreversible reactions, only the maximization is necessary. A significant speedup was possible in PSAMM (S2 Table) by avoiding regenerating new problem definitions throughout the procedure. In previous implementations (e.g. COBRA) the LP problem was regenerated for optimizing each reaction. However, in PSAMM the LP problem definition was instead reused. This was feasible since the LP problems in this approach are all equivalent except for using a different objective function for each reaction. Reactions that carried non-zero fluxes were considered to be flux consistent, and vice versa. Alternatively, PSAMM provides another way of checking reaction flux consistency based on the fast consistency check (FASTCC) algorithm, which was designed to provide a more efficient solution in evaluating reaction flux consistency [87]. When analyzing the models in the collection for flux inconsistencies, the constraints on the flux limits of the exchange reactions were removed prior to the analysis to eliminate the influences of external settings and to provide a lower estimate for the fraction of inconsistent reactions (Fig 4).
|
10.1371/journal.pbio.2003663 | A novel unsupervised analysis of electrophysiological signals reveals new sleep substages in mice | Sleep science is entering a new era, thanks to new data-driven analysis approaches that, combined with mouse gene–editing technologies, show a promise in functional genomics and translational research. However, the investigation of sleep is time consuming and not suitable for large-scale phenotypic datasets, mainly due to the need for subjective manual annotations of electrophysiological states. Moreover, the heterogeneous nature of sleep, with all its physiological aspects, is not fully accounted for by the current system of sleep stage classification. In this study, we present a new data-driven analysis approach offering a plethora of novel features for the characterization of sleep. This novel approach allowed for identifying several substages of sleep that were hidden to standard analysis. For each of these substages, we report an independent set of homeostatic responses following sleep deprivation. By using our new substages classification, we have identified novel differences among various genetic backgrounds. Moreover, in a specific experiment with the Zfhx3 mouse line, a recent circadian mutant expressing both shortening of the circadian period and abnormal sleep architecture, we identified specific sleep states that account for genotypic differences at specific times of the day. These results add a further level of interaction between circadian clock and sleep homeostasis and indicate that dissecting sleep in multiple states is physiologically relevant and can lead to the discovery of new links between sleep phenotypes and genetic determinants. Therefore, our approach has the potential to significantly enhance the understanding of sleep physiology through the study of single mutations. Moreover, this study paves the way to systematic high-throughput analyses of sleep.
| Sleep is a heterogeneous process determined by a number of genetic and epigenetic factors. To investigate the biology of sleep, animal models, such as mice, are extensively used in sleep studies, and large-scale phenotypic datasets are required to reach meaningful conclusions. Currently, manual annotations of electrophysiological states by a researcher is the gold-standard approach to classifying sleep stages. Only a few sleep states are identified through such manual annotations, namely non-rapid-eye-movement (NREM) and rapid-eye-movement (REM) sleep. In this work, we present a new computational approach that identified multiple new substages of sleep based on standard electroencephalography (EEG)/electromyography (EMG) features. Using this new approach, we studied each individual identified state and discovered that many of these states respond to the basic principles of sleep physiology: for example, some states homeostatically respond to sleep deprivation. We also applied our method to different mouse strains and to a circadian mutant line of mice. In all experimental groups, we were able to refine our understanding by associating specific substages with the genetic variations. We conclude that our new unbiased computational approach can help refine the study of sleep by further dissecting sleep biology.
| Sleep is a physiological, metabolic, and behavioral state of the organism that plays an important role in many biological functions. It is described by 2 different states—namely, non-rapid-eye-movement (NREM) sleep and rapid-eye-movement (REM) sleep. In laboratory settings, sleep is conventionally defined by properties of electrophysiological signals, and many features of these biological signals are shared across species. For this reason and thanks to the advances in mouse genetics, the study of sleep in mice has become popular in many laboratories.
A guideline for sleep scoring was developed for human studies as a tool for visual analysis of electroencephalography (EEG) and electromyography (EMG) traces according to a predefined set of rules [1]. Based on these rules, a simplified catalog of electrophysiological properties was determined for rodents as well. Differently from humans, in rodents NREM sleep is considered to be a single state. It is characterized by low EEG frequencies (below 5 Hz) with a progressive reduction of muscular tone, and it is usually referred to as deep sleep. On the other hand, REM sleep is characterized by EEG frequencies in the range 5–9 Hz. This makes it an electrophysiologically active state, similar to wakefulness in many aspects, even though the muscle tone decreases dramatically during this state. Due to this fact, REM sleep is also called paradoxical sleep. These basic criteria and some additional properties (i.e., periodic electrophysiological phenomena that occur in EEG) are used to manually classify long series of time epochs (each epoch usually lasts 4 seconds) into either one of the two aforementioned sleep states or into the wakefulness state.
Visual scoring of the different time epochs is the current gold-standard approach that provides fundamental insights into sleep regulation and physiology; however, it is a time-consuming process that can be affected by errors and misjudgments across different scorers. This is caused by the subjective decisions that can easily bias the labeling process. Indeed, inter- and intrascorer variability is always present, and studies evaluating the consistency between scorers have shown an average agreement of about 80%, which varies significantly across different sleep states [2,3,4]. In addition to the difficulties related to manual scoring, it is well recognized that sleep states are nonhomogeneous units representing an aggregation of different substages [5]. For example, NREM sleep in humans is typically characterized by 3 substages, and in some animals, it can be divided into 2 substages (slow-wave sleep I and II) [6]. However, whether sleep stages should be divided into multiple substages, each one having a physiological relevance, remains an open question in sleep biology.
In this scenario, we attempted to go beyond the current state of knowledge in the study of sleep in mice, trying to extrapolate unforeseen phenomena from the data. Our study is driven by the assumption that a richer structure of electrophysiologically derived properties can be discovered from standard datasets. Thus, we present here a different conceptual and technical framework for the analysis of sleep in mice based on the adoption of unsupervised machine learning. The proposed scheme has the potential to represent a paradigm shift in the study of sleep, allowing for an investigation without prior assumptions on the different states, thus facilitating an agnostic exploratory data analysis. Indeed, employing models that are capable of detecting regularities in large datasets, we can discover many sleep substages that a human scorer would never detect by a standard visual inspection of signals.
The class of latent (hidden) variable models have proven to be very effective in capturing hidden regularities in the data, due to their ability to encode rich structured priors and infer latent properties of the data without requiring human annotations [7]. Out of all potential models, we decided to adopt the mean-covariance restricted Boltzmann machine (mcRBM) [8] to model brain and muscle activities. This model is capable of learning the joint distribution between a set of continuous observed random variables and a set of binary unobserved (latent) ones, modeling complex (multimodal) distributions from Gaussian-like data (like in our case, see Fig 1A), and exploiting correlations between the input variables. Specifically, it allows inferring a set of latent representations (namely, latent states) describing different modes in the input data distribution, each one ideally corresponding to a cluster associated with a reoccurring sleep pattern. These latent states can be therefore interpreted as model descriptors identifying different sleep substages. For this reason, in the following text, we will interchangeably refer to them as latent states or substages.
This model has been demonstrated to be capable of learning such regularities and discovering meaningful phenomena [7]. Indeed, thanks to the unsupervised nature of the approach, we were able to study unforeseen complex phenomena appearing in electrophysiological data that cannot be studied with common rule-based and supervised learning approaches [9,10,11,12] just because such phenomena are not entirely modeled or described by the experts’ prior knowledge. This approach, for the first time, allowed the identification of multiple substages both in sleep and wakefulness phases. Thus, we explored the variability in the sleep behavior between different mouse genotypes. In particular, we performed a multisubject analysis applied on groups of mice with different genetic backgrounds, i.e., a classical pure inbred mouse strain, an outbred strain, and a strain with mixed background. From this multisubject data analysis, we identified a repertoire of novel substages that characterize the different physiological states of mouse sleep. Taking this as a starting point, we identified novel sleep-behavioral differences among mouse genotypes. Indeed, the results of our experiments robustly show that there are several substages that are characteristic of 1 strain, and this can be informative of the genetic differences across mice.
In an effort to further investigate the use of our approach, we applied our analysis to the recently identified Zfhx3Sci/+ mouse strain, which presents a shortening of the circadian clock and alterations in sleep homeostasis [13]. Our new approach provides a deeper understanding of the sleep abnormalities in this circadian mutant mouse line, revealing specific EEG anomalies and circadian modulations. We assessed the physiological relevance of each new stage by testing the homeostatic response to a perturbation (i.e., sleep deprivation). Thus, we identified a general heterogeneous homeostatic response, suggesting that these states carry a potentially unique physiological role. Yet we observed that specific substages show distorted responses in mutants’ sleep, which may indicate the presence of subtle microstructures in the sleep of mice that are governed by single genes.
It is important to highlight that, beyond being a novel approach to the analysis of complex sleep phenomena, the proposed method allows management of sleep longitudinal studies, transferring the inferred knowledge across experiments. In addition, it can also be applied to other investigations in any biological domain involving the analysis of similar data (i.e., time series), such as other behavioral studies.
We explored the homogeneity of the 3 standard physiological states (NREM sleep, REM sleep, and wakefulness) using a cohort of 46 animals from 3 different strains: (i) a common inbred mouse strain (C57BL/6J); (ii) a mixed background strain (BALB/c × C3H/HeH × C57BL/6J); and (iii) an outbred mouse line (CD1). We investigated the existence of substages that are common across animals through a multisubject data analysis pipeline. To deal with the subject-to-subject variability, each individual dataset was preprocessed separately, obtaining a uniform representation across subjects (see Fig 1A). After preprocessing, it was possible to observe the presence of nonhomogeneous state distributions. For example, Fig 1B shows in one animal that REM sleep episodes are distributed across at least 2 main clusters.
Since we were looking for co-occurring regularities in the data, aiming at discovering common sleep substages across subjects and across groups, all individual datasets of the aforementioned 3 mouse lines were modeled jointly with a single mcRBM (Fig 1C). This allowed us to search for regularities within the dataset, inferred through the states of the model’s latent variables (Fig 1D). As a result, approximately 190 different latent states (i.e., binary configurations of latent variables) were identified, most of which were strongly representative of the labeling performed by manual scoring. In Fig 2A, we provide a visual representation of this relationship, showing the probability for each latent state (rows) to correspond to each of the 3 manually scored sleep states (columns). From this probability matrix, we can observe that a significant number of latent states map with high probability to only 1 of the 3 known states. There are also some latent states that could be associated with approximately equal probability (whitish color in the graph) to at least 2 states. For example, there are latent states mapping to both wakefulness and REM and others mapping to both NREM and REM. While these latter latent states are likely to be misclassified in the classical scoring system—for example, by different scorers—in our unbiased classification, they are independent states with a specific frequency and timing.
We also measured how informative the different latent states are about the 3 states by estimating the normalized mutual information (NMI) [14], which measures how well the learnt latent states have encoded the sleep states. In our case, it was approximately 0.6, indicating a good informativeness of the latent states. Mutual information (MI) and state entropy were also computed for each state separately and, as shown in the graph in Fig 2B, the observed latent states are highly informative about the 3 states (i.e., NMI > 0.5).
Thanks to the fact that the mcRBM is a generative model, we were able to further investigate the variability of the input data distributions associated with each inferred latent state. Specifically, a multivariate normal distribution (represented by its mean vector and covariance matrix) over the input variables was inferred for each latent state (see Fig 1D). Different latent states correspond to different distributions in the input data, even when they can be statistically associated with the same sleep stage (see Fig 2C and S1 Fig). This result points to the existence of a more elaborate electrophysiological characterization of sleep than expected, showing that the 3 main states are nonhomogeneous entities. This supports our hypothesis that the known 3 sleep stages can be described by multiple latent states in the model, hence by different sleep substages that experts cannot detect by a simple visual inspection of the signals.
In light of the findings described in the previous section, we investigated whether there are substages that are characteristic of specific mouse genotypes. Indeed, we observed that a significant number of substages are mostly associated to 1 of the 3 strains. In order to determine those that best describe each strain, we used the 2-sample independent t test, comparing the distributions of the substages across the mouse genotypes. Three examples are shown in Fig 3, in which the graphs depict for each strain the statistics on the number of epochs in which each latent state is observed. The arrows below each graph show the p-values of the 2-sample independent t test, giving a picture of the difference between each combination of 2 distributions. This approach allows determining analytically the sets of all latent states that best characterize each single genotype.
Prompted by the variability of sleep among mouse groups, we attempted to test whether we can use the substages to discriminate between the mouse genetic backgrounds with a supervised machine-learning approach. Each group and each subject is described by a discrete probability distribution over the observed latent states. In a leave-one-subject-out classification framework and using a majority-voting criterion, we achieved a discrimination performance of 85.11%. The resulting confusion matrix is shown in S2 Fig, in which we can see that only 2 subjects for each strain have been misclassified. We can also observe that both misclassified subjects of the mixed background strain have been classified as belonging to the C57BL/6J group, which is interesting because the C57BL/6J background and mixed background groups are genetically related. This experiment highlights the power of our study in detecting genetic background effects. Since different genetic backgrounds are associated with specific traits, our capability to discriminate between mouse groups indicates that some latent states we identified depend on specific genetic components.
Transitions between different substages can be informative in the analysis of sleep over 24 hours. We computed the transition probabilities between the latent states for the entire dataset. As expected, latent states mapping to a certain sleep stage among NREM, REM, and wakefulness have higher probability to be followed by latent states mapping to the same stage. We visualized the transition probabilities in a graph where nodes (the latent states) are clustered according to their overall connectivity using a multidimensional scaling algorithm (Fig 4 top). The absolute position of each node in this graph is not relevant; the overall grouping of nodes is what can give us useful insight. Indeed, as a first result, we noticed that clustering the latent states according to their connectivity surprisingly also results in a meaningful grouping of nodes in terms of the 3 known sleep stages. The clustered graph confirms basic knowledge regarding the transitions across sleep states. For example, the cluster of latent states mapping to NREM sleep is near to the cluster of latent states mapping to REM sleep, and both are far away from the cluster of latent states associated with wakefulness. Interestingly, latent states not having a clear mapping to any of the 3 known sleep stages (whitish substages in Fig 2A) are usually interleaving substages positioned between the 3 overall clusters and associated with a transitional state between the 3 known sleep stages (see purple nodes in Fig 4).
In a deeper investigation of the differences among mouse genetic backgrounds, we also built the graphs for each strain separately, following the same principle (see graphs in S3 Fig and interactive graphs at http://pavis.iit.it/datasets/mouse-sleep-analysis). This allowed us to look for possible differences among the 3 groups. We searched for nodes assuming a different role in the 3 strains (e.g., nodes that are central to a cluster in a group and become transitional between 2 clusters in another graph). Indeed, even though the general clustering is common across graphs, there are some nodes whose role changes significantly. For instance, node 53, while being associated with REM sleep according to human labeling, seems to be a transitional state between wakefulness and NREM sleep in the graph generated from the CD1 strain. On the contrary, the same node in the graphs generated from both C57BL/6J and mixed background groups has transitions only with NREM and REM sleep nodes (see black arrows pointing out this node in each graph in S3 Fig). Similarly, node 174 in the mixed background group appears as a transitional state between wakefulness and sleep, while in the other groups, it has transitions only with nodes associated with wakefulness (see orange arrows pointing out this node in each graph in S3 Fig). These results imply that specific substages can have different roles according to a specific genetic background and cannot be considered as isolated stages. This offers an unprecedented dynamic aspect in studying sleep physiology within multiple sleep microstructures.
To investigate how substages develop over time, we analyzed the daily behavior over the 24 hours for each genotype. Specifically, we computed the histogram of the 4-second epochs that fall in each single latent state using bins of 1 hour, starting from 7:00 AM. Histograms are computed for each latent state and for each group separately. A polynomial curve is fitted to the histogram to obtain a representation of daily oscillations. We observed that some latent states show different daily profiles across mouse genotypes. For example, the profile of the C57BL/6J strain in latent state 104 (see Fig 4 bottom left) is shifted compared to the other 2 strains. Interestingly, the latter state (as well as latent state 110 in S4B Fig) is quite numerous within the dataset and particularly in NREM sleep, suggesting that the daily distribution of the latent states is an additional powerful discriminator of mouse genotypes. A similar shift in the profile of C57BL/6J was observed in many other latent states, including those that do not have a clear mapping to only 1 of the 3 known sleep stages, like latent state 19 (see Fig 4 bottom right). In S4 and S5 Figs, we report further examples of daily profiles of specific latent states and the manually annotated known sleep stages, respectively. This shift in the profile of C57BL/6J is present in almost all the latent states, as well in the 3 known stages, and it is due to its genetic background that, to the best of our knowledge, it has not been described in literature before.
We further analyzed the latent states having maximum peaks during the subjective night (light phase: 7:00–19:00). We observed that there is a group of latent states having maximum peak in the first half of the light phase, while some others have a peak in the second half of the light phase. Interestingly, we observed that 91% of the latent states mapping with high probability to REM have peaks in the second half of the light phase (see Fig 5A). This compartmentalization of REM-like sleep substages across the subjective night resembles the classical distribution of REM sleep in humans, in which the majority of REM sleep occurs in the second part of the night. Our result, suggesting that a microstructure of REM-like sleep occurs in the second part of the subjective night of a mouse, was never described before. Thus, we can speculate that the temporal profile of these latent states could be informative of the full REM-like sleep in mice. Indeed, REM sleep is often reported to be 5%–10% of sleep in mice, while in humans, it reaches up to 20%. In Fig 5B, we compared the percentage of REM sleep annotated with manual scoring versus the percentage of REM-like sleep, considering all latent states that present a profile that peaks in the second half of the subjective night. Remarkably, the account of latent states is close to the percentage (in range of 20%–30%, see Fig 5B) of REM sleep reported in humans. These results may suggest that this new method has the potential to describe a microstructure of REM-like sleep states in mice that resemble the human one.
In order to explore in more detail the idea of using specific latent states as biomarkers for the identification of single gene determinants (i.e., in circadian clock alterations), we also performed an additional sleep analysis of the Zfhx3Sci/+ mutant strain, which is characterized by a reduced circadian period and some other differences in their sleep architecture compared to their littermate control Zfhx3+/+ mice [13].
Following the same processing pipeline as in the previous experiment, all datasets of the 2 groups were modeled jointly with an mcRBM. Analyzing the daily profile of the inferred configurations, we observed the existence of latent states where Zfhx3Sci/+ mutants have a reduced profile (see Fig 6B and 6C left). This is also evident from the analysis of the profiles of the 3 known states using the manually scored epochs (see Fig 6A), as was already shown in [13]. Moreover, some latent states show flat profiles during the dark phase for both strains (see Fig 6B right and Fig 6C right), showing more specific information in the data that is not evident when only relying on standard scoring. Looking at the mean of the distributions over the input data (see bottom plots in Fig 6B and 6C), we also observed that such states have different data distributions compared to the previous cases. Finally, we also noted that for some latent states, mainly associated with REM or NREM stages, the profile of Zfhx3Sci/+ mutants has a forward shift compared to their littermate control group (see Fig 6D, first 2 top graphs).
A further validation of the physiological value of each latent state came from the study of how each substage responds to sleep deprivation. Indeed, sleep is a homeostatic process that is best described by its rebound response following sleep deprivation; hence, we tested it for each latent state. In Fig 7A, we show the distribution of rebound responses in Zfhx3 wild-type and mutant mice. In particular, we classified between immediate (i.e., at zeitgeber time [ZT] 6), intermediate (i.e., within ZT 6 and ZT 12), and late (i.e., beyond ZT 12) responses over the 6 hours following sleep deprivation. The dynamic of rebound responses in the 2 groups of mice is similar, with both groups having similar peaks at different times along the recovery phase (Fig 7A). Then, we looked at the rebound for each latent state by assessing the different response compared to its baseline value for the same time of the day (Fig 7B shows the amount of rebound of latent states for the 2 models). This analysis provides an initial screening of latent states that present similar responses in the 2 groups (i.e., linearly correlated in Fig 7B corresponding to quadrants 1 and 3) and latent states that respond differently according to the genotype (i.e., nonlinearly correlated in Fig 7B corresponding to quadrants 2 and 4). Thus, we performed a more detailed analysis of individual responses to sleep deprivation, and this provided a set of heterogeneous behaviors across different latent states, ranging from high-frequency states (e.g., >17,000 epochs) to low-frequency ones (e.g., <1,000 epochs). As we can appreciate from Fig 7C, some latent states respond equally in the 2 groups of mice. The analysis also revealed specific latent states in which the homeostatic rebound following sleep deprivation is attenuated or fully suppressed in mutants compared to wild-type mice (Fig 7D). We also identified states in which the response to deprivation is increased in mutants (Fig 7E) or not present in both groups (Fig 7F). Overall, the results of applying our approach to a perturbed condition such as sleep deprivation indicate that the latent states we have identified are independent states that can differentially respond to homeostatic rebound and that single gene change may influence each state separately.
Our study presents an advancement in the methodologies allowing the study of EEG-based sleep physiology in mice. Using our approach, we were able to identify sleep substages using standard electrophysiological data collected from mice. The observed sleep substages are well contained within the standard NREM sleep, REM sleep, and wakefulness categories, providing a preliminary validation of the proposed approach. Moreover, this new unbiased method also provides a rich repertoire of substages in wakefulness, which are, however, difficult to characterize because they correspond to a large range of behaviors. Therefore, we focused our study mainly on the latent states that emerge within sleep. Our method, by its own nature, can be used independently from classical NREM/REM annotation, offering an unbiased approach to the analysis of sleep states.
In order to validate the biological relevance of the learnt latent states, we applied our method to a multisubject analysis framework including different groups of mice. Specifically, we used 3 groups of mice representing the most commonly used genetic backgrounds in mouse genetics. Our study has shown that a number of latent states are characteristic of specific genotypes, and this greatly enriches the study of the genetic components of EEG sleep. Indeed, the results of our study advocate for a new set of endophenotypes that can be studied in mice by reanalyzing electrophysiological data. Interestingly, we also observed that the majority of the latent states follow a circadian-like profile and are homeostatically regulated, which adds further information in the physiological relevance of states determined by our approach. Furthermore, the definition of latent states according to the circadian phase in which they are expressed or according to their homeostatic response after sleep deprivation generates new scientific questions to be investigated. For example, our study on the Zfhx3 circadian mutants refined the understanding of the sleep patterns affected by the mutation. In particular, we identified specific states that can account for increased sleep episodes in the dark phase of mutants. Within the NREM alterations reported for this line, we were also able to identify specific latent states that are characteristic of the mutants. Moreover, the oscillation of specific mutant states over 24 hours is shortened compared to littermate control mice, suggesting a specific correlation between sleep physiology and circadian clock. Yet, reanalyzing the homeostatic defect of this mutant line by means of the proposed analysis framework, we noticed that some specific latent states can account for the genotype differences, while this is not possible considering the whole NREM stage. Moreover, the mutation may be characterized by various latent states presenting either reduced rebound or increased rebound; therefore, the overall effect we observe in the mutants is the result of a highly dynamic sleep structure. This latter example indicates that our approach may be applied to additional datasets of sleep and circadian mutants. The general outcome emerging from our study is that, as expected, sleep in mice is more complex than what was analyzed so far, and applying our multistage dissection of sleep may highlight different dynamic structures of sleep that are physiologically relevant and genetically determined.
In the future, our methodological pipeline can be applied to datasets coming from different mouse mutants, promising a deeper understanding of the influence of genetic variations on specific physiological processes in sleep.
Last but not least, our method can naturally cope with a large dataset, allowing it to be employed in the analysis of high-throughput experiments, which are very much needed in large-scale screening in mouse genetics and drug discovery. Indeed, despite its unsupervised nature, the presented analysis pipeline can easily be turned into an automatic tool for scoring sleep in mice. Moreover, although it was out of the scope of the current work, the proposed approach allows the study of the physiological role of specific latent states corresponding to interesting modes, i.e., multivariate normal distributions over the input variables. In principle, it is possible to label relevant modes, creating a set of ground-truth distributions that can be used to identify and label the substages in new experiments.
All animal procedures were approved by our institutional animal committee (“Organismo preposto al benessere degli animali”, OPBA, IIT, Genova) and by the ethical national committee in Italy for IIT Genova. All procedures were done under the Italian Policy (license issued on 19 June 2009, decreto No106/2009-B). The study followed ARRIVE guidelines (http://www.nc3rs.org.uk/arrive-guidelines, see S1 Checklist). All mice were anesthetized IP with Ketamine/Xilazine, 90-150K/7.5-16X and implanted with telemetry transmitters (Data Sciences, F20-EET, Gold system) for recording EEG.
We analyzed EEG/EMG recordings coming from 46 male mice of 3 different groups:
Isogenic C57BL/6J mice are the most widely used mice in neuroscience due to the fact that, as with any other inbred genetic background model, it reduces variability in phenotypic expression. Moreover, a number of mouse mutant models have been developed and are available on C57BL/6J background. However, in the study of mouse models for translational medicine, the testing of mutations on mixed background models (e.g., BALB/c × C3H/HeH × C57BL/6J) or outbred mice (CD1), which resemble human genetic variations better, brings a great value.
To test whether our approach could be a valuable tool for the identification of specific electrophysiological alterations and circadian modulations, in a second experiment, we analyzed EEG/EMG recordings coming from 14 male mice:
All Zfhx3Sci/+ and Zfhx3+/+ mice were genotyped as described in [17]. For more details regarding animals’ breeding and maintenance, please refer to [13].
Mice (10–14 weeks of age) were anesthetized with Ketamine/Xylazine (65/5 mg/kg, 2 ml/kg, IP) and implanted with a subcutaneous transmitter that records EEG and EMG with 2 biopotential channels (Data Sciences, F20-EET, Gold system). Electrodes (1 mm diameter) were placed on the parietal cortex of the mouse, only on 1 hemisphere. EEG electrodes were implanted on the dura, and EMG electrodes were attached to the muscle in the nape of the neck to acquire signals (Dataquest A.R.T. software, DSI) while the animals freely move in their home cage. EEG activity was sampled at 500 Hz with a 50 Hz cut-off filter. EEG signals were band-pass filtered at 0.3 Hz (low-pass filter) and 0.1 kHz (high-pass filter). After 10 days of postsurgery recovery, continuous electrophysiological signals were recorded for 24 hours (first experiment—light/dark cycle 12:12 with light switched off and on at 7 PM and 7 AM, respectively) and for 72 hours (second and third experiment—same light/dark condition with 6-hour sleep deprivation after the first 48 hours). Sleep deprivation was performed by introducing novel objects in the home cage of the animal and/or by gentle touch. All data was analyzed in 4-second epochs (the total number of epochs per subject was 21,600). The range 0.25–50 Hz was partitioned into 5 frequency bands: Delta (0.25–5 Hz), Theta (5–9 Hz), Alpha (9–12 Hz), Beta (12–20 Hz), and Gamma (20–50 Hz). Each epoch was subject to a fast Fourier transformation (FFT) with 0.48-Hz resolution using the Hanning window method. The EMG integral was exported as a measure of the muscular activity. Semiautomatic sleep scoring to annotate the 3 sleep stages (wakefulness, NREM sleep, REM sleep) was performed using the SleepSign software followed by visual inspection and correction by experts. This hand labeling was used as ground truth to better characterize our experiments. Epochs associated with artefacts (i.e., 6.74% of the dataset in the first experiment, 0.835% in the second experiment—first 24 hours of baseline; 0.655% in the third experiment—second 24 hours of baseline plus 18 hours of recovery) were excluded from the analysis.
The resulting visual scoring statistics in terms of the 3 sleep stages were as follows:
A common issue to address in multisubject analysis is the variability across subjects’ physiology, which is reflected in different ranges of EEG/EMG signal for different subjects. This issue can be solved with a proper preprocessing of data. In our case, each individual dataset was preprocessed separately by first computing all possible ratios between the 5 EEG bands, because we observed that using the ratios instead of bands’ power produces less variability across subjects. A full representation of the EEG spectra is plotted in S6 Fig, showing the classical band distribution across NREM/REM stages, as determined by the manual scoring. Similarly, S7 Fig shows examples of the power bands of some latent states compared to those of the known sleep stages they are associated with. From this comparison, it can be observed that the latent states show a more compact representation of specific structure in the data than the standard sleep stages; hence, they are more specific concerning the microstructure that can be found in the data. Since our data are log-Gaussian, to deal with the skewness in the features’ distributions, we used the natural logarithm, obtaining more bell-shaped distributions [18] that facilitate the modeling process performed by the mcRBM. The zero-mean normalization was further applied to each single feature to make all the subjects comparable.
We seek for modes in the input data distribution by modeling the joint distribution of the input variables (ratios between EEG bands and EMG) using the mcRBM [8]. The mcRBM is a probabilistic energy-based graphical model consisting of 2 fully connected layers of stochastic random variables (also called units): a layer of visible variables representing the observed data (visible units, v) and a layer of latent variables (hidden units, h) that capture dependencies between the visible ones (see Fig 8 for an illustration).
The advantage of using the mcRBM for this task is related to its effectiveness in modeling real-valued Gaussian distributed data, together with the ability to exploit the correlations between the input variables. Another important feature making the selected model more suitable than other variants of restricted Boltzmann machines (RBMs) is the fact that its latent variables are divided into 2 different sets (see section mcRBM for more details regarding the model). One of these sets ({h1, …, hm}) is used to model the mean value of the input variables, while the other one ({hm+1, …, hm+c}) is thought to be for explicitly modeling their covariance. This second set of latent variables allows for a better fitting of the data distribution presenting dependencies between input variables than what can be achieved with simpler models (e.g., Gaussian RBMs).
Once the model is trained, its bipartite graph structure allows for representing regularities in the observed input variables through the inferred values of the latent binary variables. In other words, the model can be used to generate a set of latent representations (the states of the inferred hidden variables) for the observed EEG/EMG data. Indeed, the model processes each input vector (corresponding to one sample of the dataset—in our case, one 4-second epoch), generating its related latent representation consisting of a series of binary values according to the probability of each single latent variable to be active.
Notably, in this Bayesian framework, the observed variables are approximately jointly Gaussian distributed [8], given the latent ones, with the mean and covariance jointly defined by the specific values of latent variables and model parameters (see Eqs 6 and 7). This means that each single latent configuration is associated with one mode in the joint distribution of the input variables, reflecting the regularities in the data. Hence, one latent representation may be considered a model representation of a sleep or a wakefulness substage.
In each of the 2 experiments performed in this study, a single mcRBM was used to model the joint distribution of the EEG ratios and the EMG. After training the model, the latent states of the 2 sets of hidden units were inferred according to Eqs 4 and 5.
The size of the mcRBM (the number of its latent variables) was set after a series of validation trials in which we “calibrated” the model to the problem at hand. We observed that having too few hidden units results in a very small number of fuzzy clusters, while, as expected, the size escalation leads to a higher computational cost. We also realized that there is a certain number of hidden units, beyond which there is no real advantage in increasing the network size, since they remain inactive. We reached a good tradeoff between computational cost and quality of the results (i.e., significantly high MI, easy interpretability of visible data distributions, good performance in groups’ discrimination, etc.) using 11 hidden covariance units (hc), 11 visible-to-hidden covariance factors, and 10 hidden mean units (hm). Similarly, all the other parameters were set after a series of trials, taking into account the average quality of the results. Finally, the models were trained on GPU over minibatches of 256 data points with learning rate set to 10−2.
Like all RBM models, the mcRBM [8] is a probabilistic energy-based graphical model with a bipartite undirected graph structure, which consists of 2 fully connected layers of stochastic random variables (also called units): a layer of visible variables representing the observed data (visible units, v), and a layer of latent variables (hidden units, h) that capture dependencies between the visible ones. Unlike the standard RBM models, the mcRBM has 2 groups of hidden units: mean units (hm) that model the mean of the input elements and precision units (hc) that represent pairwise dependencies between the visible variables, modeling their covariance structure (see Fig 9 for an illustration).
Due to the absence of connections between the variables within the layers, the variables of a layer are independent of each other, given the variables of the other layer. The model is defined in terms of an energy function that is given by the sum of the energy functions of the 2 groups of variables:
Emc(v,hc,hm)=Ec(v,hc)+Em(v,hm).
(1)
The energy function of the precision variables Ec defines a zero-mean Gaussian distribution over the visible variables, while adding the second energy term Em allows the mcRBM to produce conditional distributions over the visible units having nonzero means.
Ec is defined as follows:
Ec(v,hc)=-dThc-(vTC)2Phc,
(2)
in which C is the weight matrix from the visible (v) units to the factors (F), P is a weight matrix with nonpositive entries for connections from the factors (F) to the hidden covariance units (hc), and d is the hc bias vector. P is constrained to have nonpositive entries in order to avoid having a model that assigns larger and larger probabilities (more negative energies) to larger and larger inputs. To control great disparity in the input features and make the model more robust, we used the normalized version of Eq 2 [8].
The energy function of mean variables Em is given by:
Em(v,hm)=-12(v-b)T(v-b)cThm-vTWhm,
(3)
with W denoting the direct connections from the hidden mean units (hm) to the visible ones (v), b being the visible bias, and c being the hidden mean bias.
After the training process, the model allows us to infer the states of its latent variables given the observed data. Given a training sample, the binary states of the hc units can be inferred according to the following conditional distribution:
p(hc|v)=σ(d+((vTC)2P)T),
(4)
in which σ(x) = 1 / (1 + exp(-x)) is the logistic function, while the states of hm units can be inferred according to the following conditional distribution:
p(hm|v)=σ(c+WvT).
(5)
Conditioning on the latent variables, the resulting distribution over the visible ones is approximately Gaussian, which depends on both the hidden covariance and hidden mean latent states and is given by:
p(v|hc,hm)∝N(ΣWhm,Σ),
(6)
in which Σ is given by:
Σ=(C(diag(-PThc))CT)-1.
(7)
To evaluate how well the model fits the observed data in our experiments, we analyzed how informative the learnt latent states are about the stages of wakefulness, NREM sleep, and REM sleep, as defined by the manual scoring. Since each latent state corresponds to a discrete probability distribution over the 3 manually scored stages, the informativeness we want to measure can be expressed in terms of entropy and MI.
Entropy [14] is a measure of the uncertainty of a random variable and is defined by:
HL(Y)=-∑y∈YpL(y)log2pL(y),
(8)
in which Y corresponds to the set of the known 3 sleep stages (wakefulness, NREM sleep, and REM sleep), while pL(y) is the probability of each of these 3 stages to appear in each latent state (hopefully, for each latent state, only one of the sleep stages will have a high probability).
MI [14] is a measure of the amount of information that one random variable contains about another random variable. In our case, we would like to see how informative the latent states X are about the 3 sleep stages Y. The MI I(X;Y) is the relative entropy between the joint distribution and the product distribution p(x)p(y):
I(X;Y)=-∑x,yp(x,y)log2p(x,y)p(x)p(y),
(9)
in which p(x,y) is the joint probability of latent state x and sleep stage y, p(x) is the probability of latent state x, and p(y) is the probability of sleep stage y.
Finally, the NMI is computed as follows:
In=I(X;Y)H(X).
(10)
NMI (In) is expressed as a real number in the range [0, 1], allowing us to compare it to its theoretical upper bound (i.e., which is equal to 1 when there is a perfect correlation between the 2 variables).
Transitions between sleep stages can provide additional information regarding the structure of sleep in 24 hours. For this reason, we analyzed the transition probabilities between the observed latent states, which were computed and summarized in a squared transition probabilities matrix. The matrix was visualized as a graph whose nodes are the latent states and edges are the transitions between them. The weight of each edge is the probability of the corresponding transition. We used the ForceAtlas2 [15] algorithm to cluster the network. Briefly, the algorithm randomly positions the nodes in a 2D space, which are then moved according to the attraction and repulsion forces among the nodes until the network reaches a state of equilibrium. These forces cause nodes with a big number of connections (hubs) to strongly repel each other and nodes connected with a heavily weighted edge (meaning a very probable transition) to attract each other [15].
One question we addressed was whether we could use the discovered latent states to discriminate between different mouse genotypes. To do so, each subject was represented with a discrete probability distribution over the k = 1,2, …, n observed latent states, with Pi(xk) being the probability of latent state k for subject i.
We then performed a classification with a leave-one-subject-out schema (train the model on N − 1 subjects and test over the left-out subject, iterating N times over all N subjects). We used an ensemble of classifiers based on a linear support vector machine (SVM), a linear discriminant analysis (LDA), and a 1-nearest neighbor (1-NN) to classify with a majority vote the left-out subject as belonging to 1 of the 3 groups.
|
10.1371/journal.pntd.0001608 | Increased Birth Weight Associated with Regular Pre-Pregnancy Deworming and Weekly Iron-Folic Acid Supplementation for Vietnamese Women | Hookworm infections are significant public health issues in South-East Asia. In women of reproductive age, chronic hookworm infections cause iron deficiency anaemia, which, upon pregnancy, can lead to intrauterine growth restriction and low birth weight. Low birth weight is an important risk factor for neonatal and infant mortality and morbidity.
We investigated the association between neonatal birth weight and a 4-monthly deworming and weekly iron-folic acid supplementation program given to women of reproductive age in north-west Vietnam. The program was made available to all women of reproductive age (estimated 51,623) in two districts in Yen Bai Province for 20 months prior to commencement of birth weight data collection. Data were obtained for births at the district hospitals of the two intervention districts as well as from two control districts where women did not have access to the intervention, but had similar maternal and child health indicators and socio-economic backgrounds. The primary outcome was low birth weight.
The birth weights of 463 infants born in district hospitals in the intervention (168) and control districts (295) were recorded. Twenty-six months after the program was started, the prevalence of low birth weight was 3% in intervention districts compared to 7.4% in control districts (adjusted odds ratio 0.29, 95% confidence interval 0.10 to 0.81, p = 0.017). The mean birth weight was 124 g (CI 68 - 255 g, p<0.001) greater in the intervention districts compared to control districts.
The findings of this study suggest that providing women with regular deworming and weekly iron-folic acid supplements before pregnancy is associated with a reduced prevalence of low birth weight in rural Vietnam. The impact of this health system-integrated intervention on birth outcomes should be further evaluated through a more extensive randomised-controlled trial.
| Low birth weight is an important risk factor for neonatal and infant morbidity and mortality and may impact on growth and development. Maternal iron deficiency anaemia contributes to intrauterine growth restriction and low birth weight. Hookworm infections and an iron-depleted diet may lead to iron deficiency anaemia, and both are common in many developing countries. A pilot program of deworming and weekly iron-folic acid supplementation for non-pregnant women aiming to prevent iron deficiency was implemented in northern Vietnam. We compared the birth weight of babies born to women who had had access to the intervention to babies born in districts where the intervention had not been implemented. The mean birth weight of the intervention districts' babies was 124 g more than the control districts' babies; the prevalence of low birth weight was also reduced. These results suggest that providing women with deworming and weekly iron-folic acid supplements before pregnancy is associated with increased birth weight in rural Vietnam. This intervention was provided as a health system integrated program which could be replicated in other at-risk rural areas. If so it could increase the impact of prenatal and antenatal programs, improving the health of both women and newborns.
| Low birth weight is widely recognised as a risk factor for neonatal mortality and morbidity, as well as reduced cognitive function and the development of chronic diseases in later life [1]–[3]. Iron deficiency anaemia during pregnancy is an important cause of restricted foetal growth leading to low birth weight and preterm delivery, and also maternal illness and death [4], [5].
It is estimated that more than one third of women in the world are anaemic [6], and iron deficiency is the most common cause of anaemia in the majority of settings [7]. In addition, many of these women live in rural communities of developing countries where intestinal parasitic infections are endemic. Hookworm infections contribute to anaemia severity and persistence by causing chronic blood loss [8], [9]. It has been hypothesized that the anti-helminthic drugs mebendazole and albendazole may have a positive impact on birth outcomes if administered during pregnancy, but conclusive evidence is still lacking. Preventive chemotherapy through mass deworming is recommended when the prevalence of infection with any soil-transmitted helminth exceeds 20% [10]. However, very few countries have promoted routine anti-helminthic treatment in women of reproductive age [11].
Previous studies have reported benefits to maternal and infant health through antenatal supplementation with iron-folic acid supplements and multiple micronutrients [12]–[14], although poor compliance and variable supply have limited the impact of this approach [15]. Preventative weekly iron-folic acid supplementation for women of reproductive age given before pregnancy is effective in improving iron stores and women are less likely to develop iron deficiency anaemia during pregnancy if iron stores are replete at the time of conception [16]. The WHO has recently recommended that weekly iron-folic acid supplementation be made available for women of reproductive age in areas where the prevalence of anaemia in women of reproductive age is above 20% [17].
In Vietnam, the prevalence of anaemia has previously been reportedly as high as 65% in pregnant and 54% in non-pregnant women, and in 2003 it was estimated that 21.8 million people had hookworm infections [18]–[20]. In November 2005 we conducted a survey of anaemia, iron deficiency and hookworm infection in women of reproductive age in Yen Bai province, northern Vietnam. The results showed prevalences to be 38%, 23% and 78% respectively [21]. Anaemia was associated with iron deficiency and meat consumption, however there was no association between hookworm infection and either anaemia or iron deficiency [21]. In response, a pilot weekly iron-folic acid supplementation and deworming program for women of reproductive age was introduced in two districts in May 2006. By September 2007 impact and compliance surveys identified that: 90% of women were taking the weekly iron-folic acid supplements; mean haemoglobin had risen from 122 g/L to 131 g/L; mean serum ferritin levels had risen from 28.1 µg/L to 44.7 µg/L; anaemia, iron deficiency and hookworm infection prevalence had dropped to 20%, 6% and 26% respectively [22], [23].
We hypothesized that the consequent improvement in women's nutritional status would translate to improvements in birth weight of their babies compared to babies born to women in adjacent districts where the intervention was not available.
The project was approved by the Human Research Ethics Committee of the National Institute of Malariology, Parasitology and Entomology (Hanoi, Vietnam), the Walter and Eliza Hall Institute of Medical Research (Melbourne, Australia, Project No. 03/07) and Melbourne Health (Melbourne, Australia). The birth weight survey was locally approved by the Yen Bai Ministry of Health. Extensive consultation was undertaken between the project team and community leaders, as well as liaison with village, district and provincial health staff. Village health workers provided participants with information regarding the intervention and signed informed consent was documented. For the birth weight survey, the mother's informed consent was obtained verbally before data collection, as approved by the Yen Bai Ministry of Health Research Committee and in accordance with the Vietnamese Ministry of Health protocols for surveys. Oral consent was documented by the presence of a witness.
The pilot intervention was conducted in Tran Yen and Yen Binh districts, with Luc Yen and Nghia Lo as control districts, all in Yen Bai province, a remote, mountainous province in Vietnam with low population density (104 people/km2) [24] and poor road and transport infrastructure. Intervention districts were selected based on advice from provincial authorities that they were representative of most other districts in the province in terms of population density and socioeconomic factors. All non-pregnant women between 16 and 45 years in the two districts (estimated 51,623) were targeted. Pregnant women were identified by asking women whether they were pregnant and the timing of their last menstrual period, and were not given deworming treatment if they were or thought they may be pregnant [25]. This protocol was approved by the provincial health authorities. In May 2006 the distribution of iron (FeSO4)/folic acid (60/0.4 mg) and 4-monthly albendazole (400 mg) to women was introduced. A detailed account of the protocol has been previously reported [25]. Briefly, iron-folic acid supplements were distributed to village health workers monthly (4 tablets, i.e. weekly consumption for one month) through the administrative strata of the Department of Preventive Medicine. The village health workers then distributed the supplements to individual women either through organized community meetings or home delivery. Provincial authorities chose to distribute 4-monthly deworming treatment through the commune health centres as they felt that this level of the health system was more appropriate for a mass chemotherapy intervention. The intervention was preceded by training of local health workers and delivered with educational activities, distribution of promotional materials to women and community educational meetings [25]. The intervention was not introduced or publicised in other districts of Yen Bai prior to the completion of data collection for this study.
Four hundred sixty-three infants were born in the 4 district hospitals during the period of observation (January 2008–Jun 2008), 168 in intervention districts (47.5% females) and 295 (47.2% females) in control districts.
Demographic data for the participants by study arm including maternal age at delivery, parity and socio-economic and educational background are shown in table 1: socio-economic background was similar in the two groups, while women in the intervention districts had higher maternal age but lower level of education.
The prevalence of low birth weight was 3% and 7.4% in the intervention and control districts respectively (Table 2). This equates to a 4.4% absolute and 59% relative lower prevalence of low birth weight (number needed to treat = 23, 8.84–59.11, i.e. one less low birth weight infant for every 23 delivering mothers who had access to deworming and iron-folic acid supplementation prior to pregnancy). The odds ratio of low birth weight comparing infants born to mothers from intervention versus control areas, controlling for the effect of potential confounders and incorporating the possible clustering of inhabitants of a district, was 0.29 (95% CI 0.10 to 0.81, p = 0.017; clustering-adjusted and covariate-unadjusted p = 0.040; clustering-unadjusted and covariate-unadjusted p = 0.077).
Mean birth weight was 124 g higher in intervention districts (3135 vs 3011 g, CI for mean birth weight increase 68 to 255, p<0.001) (Table 2). Mean birth weight positively correlated with socio-economic background, level of education, maternal age and parity (data not shown).
We present here results that show that the prevalence of low birth weight in infants born to mothers in rural Vietnam who had access to a pre-pregnancy program of 4-monthly albendazole treatment and weekly iron-folic acid was 3.0% compared to 7.4% for those born in neighbouring districts where women did not have access to this intervention. The latter is comparable to the Vietnam national average for low birth weight of 7.0% [28]. In addition, we observed a significantly higher mean birth weight in the intervention group.
While there are numerous studies assessing the benefits of ante-natal iron supplementation for pregnant women, birth outcome and infant health [14], [29], [30], there are few reported long-term studies of the impact of pre-pregnancy weekly iron-folic acid and deworming programs for women of reproductive age on birth outcomes and infant health [31]. Berger et al (2005) reported a prevalence of 2.9% low birth weight infants in a group of women who were participants in a pre-pregnancy weekly iron-folic acid supplementation program who subsequently became pregnant. The prevalence of low birth weight in the daily supplementation arm was 9.3%, however there was no standard treatment control group in the study [32].
Our study has limitations. The study design is not that of a randomized controlled trial, which would allow a more conclusive interpretation of the results. We have tried to adjust for potential confounders but other variables may exist that would bias the results. However, our study was conducted during the implementation of a large scale anaemia and iron deficiency prevention program, and therefore provides information about the impact of the intervention when implemented under field conditions through routine health services.
We are not sure about the exact supplements women took during pregnancy and with what frequency. When questioned, women were unsure of the number and source of free iron-folic acid supplements provided by health workers during pregnancy. Those who bought antenatal nutrition supplements privately did so on the recommendation of a doctor or relation/friend and again were unable to clearly state what the supplements were or how much iron they contained. We therefore cannot exclude a possible bias in the results due to differences in antenatal iron supplementation patterns between the intervention and control districts.
The intervention in Tran Yen/Yen Binh did not have separate arms for iron-folic acid or deworming alone so the relative contribution of each cannot be ascertained. A previous pregnancy supplementation and deworming study suggested that haematinics and anthelminthics had an additive effect on stabilizing haemoglobin during pregnancy when given to 125 Sierra Leone women, with haematinics having the greater effect [33].We find it plausible in our case that iron-folic acid and deworming acted additively or even synergistically, targeting the problem of maternal anaemia at different levels [6]. Regular hookworm control is likely to have complemented iron-folic acid supplementation by reducing iron loss due to chronic hookworm infection. Moreover, the pre-pregnancy population-based approach has been previously shown to result in a gradual and stable improvement in iron status prior to conception [22]. Another significant advantage of regular deworming for non-pregnant women in Vietnam is that Ministry of Health regulations proscribe administering deworming medication during pregnancy.
Another limitation was that we were not able to sample the entire pool of deliveries of mothers from intervention and control districts, as the cost and logistics in such a remote area were beyond the resources available to this study. District hospitals collect about a third of routine deliveries; although this is not a fully comprehensive sample, we believe it is representative enough across the study districts. Furthermore, if selection bias did exist, we assume it would be similar in intervention vs control districts.
It is also important to acknowledge that this intervention would increase the workload of community health workers, which is currently a topic of debate in the international development community. This may challenge the long term sustainability of the implementation and needs to be taken into consideration by health authorities in planning for the intervention.
The data presented here is the result of one of few studies of the impact of pre-pregnancy iron-folic acid supplementation and deworming for non-pregnant women on infant birth weight. Whilst it has been recently suggested that there is a need for stronger, more robust data to support long-term intermittent iron-folic acid supplementation in women of reproductive age [34], there is growing evidence that these programs are not only important [35] but also feasible and implementable in resource constrained settings [23], [25], [31].
The results of our study suggest that the pre-pregnancy combination of deworming and weekly iron-folic acid supplementation for women of reproductive age in northern Vietnam is associated with a reduced incidence of low birth weight and higher mean birth weight. Such a program could potentially represent a high-impact and easily implementable intervention to apply in settings with a high prevalence of hookworm infection and anaemia, for the health of both women and newborns.
|
10.1371/journal.pntd.0001516 | Excretory/Secretory-Products of Echinococcus multilocularis Larvae Induce Apoptosis and Tolerogenic Properties in Dendritic Cells In Vitro | Alveolar echinococcosis, caused by Echinococcus multilocularis larvae, is a chronic disease associated with considerable modulation of the host immune response. Dendritic cells (DC) are key effectors in shaping the immune response and among the first cells encountered by the parasite during an infection. Although it is assumed that E.multilocularis, by excretory/secretory (E/S)-products, specifically affects DC to deviate immune responses, little information is available on the molecular nature of respective E/S-products and their mode of action.
We established cultivation systems for exposing DC to live material from early (oncosphere), chronic (metacestode) and late (protoscolex) infectious stages. When co-incubated with Echinococcus primary cells, representing the invading oncosphere, or metacestode vesicles, a significant proportion of DC underwent apoptosis and the surviving DC failed to mature. In contrast, DC exposed to protoscoleces upregulated maturation markers and did not undergo apoptosis. After pre-incubation with primary cells and metacestode vesicles, DC showed a strongly impaired ability to be activated by the TLR ligand LPS, which was not observed in DC pre-treated with protoscolex E/S-products. While none of the larvae induced the secretion of pro-inflammatory IL-12p70, the production of immunosuppressive IL-10 was elevated in response to primary cell E/S-products. Finally, upon incubation with DC and naïve T-cells, E/S-products from metacestode vesicles led to a significant expansion of Foxp3+ T cells in vitro.
This is the first report on the induction of apoptosis in DC by cestode E/S-products. Our data indicate that the early infective stage of E. multilocularis is a strong inducer of tolerance in DC, which is most probably important for generating an immunosuppressive environment at an infection phase in which the parasite is highly vulnerable to host attacks. The induction of CD4+CD25+Foxp3+ T cells through metacestode E/S-products suggests that these cells fulfill an important role for parasite persistence during chronic echinococcosis.
| Parasitic helminths are inducers of chronic diseases and have evolved mechanisms to suppress the host immune response. Mostly from studies on roundworms, a picture is currently emerging that helminths secrete factors (E/S-products) that directly act on sentinels of the immune system, dendritic cells, in order to achieve an expansion of immunosuppressive, regulatory T cells (T-reg). Parasitic helminths are currently also intensely studied as therapeutic agents against autoimmune diseases and allergies, which is directly linked to their immunosuppressive activities. The immunomodulatory products of parasitic helminths are therefore of high interest for understanding immunopathology during infections and for the treatment of allergies. The present work was conducted on larvae of the tapeworm E. multilocularis, which grow like a tumor into surrounding host tissue and thus cause the lethal disease alveolar echinococcosis. The authors found that E/S-products from early infective larvae are strong inducers of tolerogenic DC in vitro and show that E/S-products of larvae of the chronic stage lead to an in vitro expansion of Foxp3+ T cells, suggesting that both the expansion of these T cells and poorly responsive DC are important for the establishment and persistence of E. multilocularis larvae within the host.
| The metacestode larval stage of the fox-tapeworm E. multilocularis is the causative agent of alveolar echinococcosis, one of the most dangerous zoonoses world-wide [1]. Apart from the strobilar adult stage that resides within the intestine of the definitive host (e.g. foxes, dogs), the life cycle of this cestode comprises three larval stages that are involved in the infection of the intermediate host (small rodents and, occasionally, humans). An infection of the intermediate host is initiated by the oral uptake of ‘infectious eggs’ that contain the first larval stage, the oncosphere. Upon activation within stomach and intestine, the oncosphere hatches, penetrates the intestinal wall, and gains access to the host's viscera. Almost exclusively within the intermediate host's liver, the oncosphere then undergoes a metamorphosis towards the metacestodes which is driven by totipotent parasite stem cells (germinal cells; neoblasts) that were carried into the host through the oncosphere. As a result of the oncosphere - metacestode metamorphosis, fully mature, cyst-like metacestode vesicles are formed that grow infiltratively, like a malignant tumor, into the surrounding host tissue and that consist of an inner, cellular ‘germinal layer’(GL) and an outer, glycan-rich and acellular ‘laminated layer’ (LL) [2]. At least in experimentally infected mice, the formation of the LL cannot be observed earlier than 2–3 weeks upon initial infection [3], [4], [5], [6]. Evidence has been obtained that the LL is one of the parasite's key structures for protection against the host immune system in the later phase of the infection [7]. Approximately 2 months after the infection of mice, ‘brood-capsules’ are formed from stem cells of the GL that later give rise to the third larval stage, the protoscolex, which is passed on to the definitive host [2]. The E. multilocularis infection process can thus be separated into 3 phases. The first phase starts with the oncosphere and culminates, after 2–3 weeks, in the formation of mature metacestode vesicles which, in the second phase, grow infiltratively into the host tissue. During the third stage, protoscoleces are formed in natural intermediate hosts, but only rarely in human infections [2].
Cellular effector mechanisms are considered to be the key defense against metacestode growth and dissemination in mice and humans since mouse strains that cannot develop cellular immune responses are highly susceptible to AE, whereas strains defective in humoral immunity can control parasite growth to a certain level [8]. Furthermore, in humans co-infected with E. multilocularis and the human immunodeficiency virus (HIV), very fast and unlimited parasite proliferation occurs [9], [10], whereas promotion of cellular immunocompetence has a beneficial effect on the outcome of the disease [8]. A significant number of studies on both humans and mice indicated that T helper 1 (Th1)-dominated immune responses, characterized by the release of interferon-γ (IFN-γ), after priming by DC that secrete interleukin (IL)-12, are effective in eliminating the parasite at an early stage, whereas a Th2 cytokine profile (IL-4, IL-5) and the release of immunosuppressive IL-10 and TGF-β is generally associated with susceptibility to the parasite and progressive disease [8]. Although it became clear from these studies that the parasite, most probably by E/S-products, actively influences the host immune response (e.g. gradually driving it into the Th2 branch), little is currently known on the molecular and cellular basis of E. multilocularis induced immune suppression, particularly for the early stages of the infection.
DC are professional antigen presenting cells that represent the link between the innate and the adaptive immune system and are crucially involved in the induction of Th1-, Th2- or Th17-dominated immune responses [11], [12]. Upon pathogen recognition, DC take up antigens and undergo maturation, as can be assessed by the up-regulation of surface markers such as the major histocompatibility complex II (MHC II) and co-stimulatory molecules such as CD86 and CD80 [11]–[13]. After migration to the T cell area of lymph nodes, DC interact with naïve T cells to promote adaptive immune responses towards the Th1-, Th2-, Th17-branches, depending on the pathogen pattern they have adopted [13]. However, DC are also targets of parasites to establish immune evasion, e.g. by induction of regulatory T cells (T-reg), which counteract T helper cell activities [11], [12], [14]. A potentially important role of DC in immunosuppressive mechanisms during AE has indeed been established in a recent in vivo study on secondary (intraperitoneal) AE in mice [15]. In this work, Mejri et al. demonstrated that peritoneal DC from chronically infected mice, representing the late stage of AE, express higher levels of TGF-β mRNA, lower levels of IL-10 and IL-12 mRNA, and display down-regulation of maturation-associated surface markers, when compared to DC from non-infected mice [15]. Furthermore, DC from intraperitoneally infected mice specifically modulated CD4+ and CD8+ T cell responses suggesting a role for immunosuppressive T-reg during chronic AE [15]. DC are also among the first cells encountered by parasites during an infection [11], [13] and may have a critical role in the Th1 to Th2 shift reported for the intermediate host during the chronic phase of AE [8], [10]. In line with this hypothesis are recent data demonstrating that immature human DC fail to mature in the presence of crude, non-fractionated E. multilocularis antigen [16]. Moreover, during infection of the intermediate host, migration of parasitic larvae from the intestinal entry site to the liver and late metastasis to other organs (lung, brain) [17] strongly suggest that these larvae encounter DC in vivo. However and in spite of the general importance of DC in cellular host-helminth interaction mechanisms [11], [12], [18], only few investigations have so far been carried out towards an identification and characterization of immunomodulatory molecules that are released by Echinococcus larvae and their influence on DC function. Apart from the above mentioned study concerning the influence of crude E. multilocularis antigen on human DC [16], there are merely reports on the activity of crude hydatid (vesicle) fluid or selected hydatid fluid protein compounds of the related tapeworm E. granulosus on DC maturation [19], [20].
Due to the limited availability of respective parasite material (oncospheres), no in vitro studies have so far been carried out concerning the interaction of host immune cells with early infective parasite stages. Our current picture concerning the effects of Echincoccus E/S-products on host cells thus mostly derives from studies in which easier accessible protoscoleces had been employed [21]–[27], with the considerable drawback that this stage is formed very late during an infection of the intermediate host (if at all), and that in intact metacestode vesicles, protoscoleces do not have direct contact to host tissue and cells. Notably, we have recently introduced a primary germinal cell cultivation system by which the early developmental phase within the intermediate host can be re-constituted in vitro [28]. In this system, isolated E. multilocularis primary cells proliferate and form cellular aggregates that give rise to mature metacestode vesicles (including LL) in a manner that closely mimics the natural oncosphere - metacestode metamorphosis process [28]. Even concerning gene and protein expression patterns, this system closely reflects early parasite development within the intermediate host, and parasite antigens originally described to be expressed specifically in the oncosphere are readily detectable in the regenerating parasite cell aggregates [29], [30]. Although it became clear from previous studies that E. multilocularis through its larval E/S-products tightly down regulates accessory cell functions of macrophages [31] little is currently known about the effect on DC. In the present study, we used our primary germinal cell cultivation system [28] to investigate the influence of E/S-products from primary cells (characteristic of the early phase of the infection) on DC and compared it with the effects of E/S-products of mature metacestode vesicles, characteristic for the chronic phase, and protoscoleces which, in intact parasite material, do not have direct contact with host immune cells.
All experiments were carried out in accordance with European and German regulations on the protection of animals (Tierschutzgesetz). Ethical approval of the study was obtained from the local ethics committee of the government of Lower Franconia (Regierung von Unterfranken; 621-2531.01-2/05 and 55.2-2531.01-73/07).
All experiments were performed with the natural E. multilocularis isolate JAVA [32] which was propagated in Mongolian jirds (Meriones unguiculatus) as described [33]. Isolation of metacestode tissue and axenic cultivation of metacestode vesicles was performed essentially as described previously by Spiliotis and Brehm [33]. For the isolation of protoscoleces, parasite tissue was isolated from infected jirds and homogenized as described [34]. The homogenate was subsequently filtered through a nylon mesh of 150 µm pore size, thus separating protoscoleces from larger pieces of metacestode tissue. The flow through was subsequently filtered through a nylon mesh of 30 µm pore size, separating protoscoleces from single cells and small cell clumps. Protoscoleces were then washed off the nylon mesh with sterile PBS and separated from equally sized metacestode vesicles by microscope-aided, manual picking with a pipette tip prior to applying axenic cultivation conditions in order to eliminate eventual host remnants [33] For the isolation of primary cells, axenically cultivated metacestode vesicles were mechanically sheared and trypsin-digested essentially as previously described [28]. Primary cells were then directly cultivated in hepatocyte-conditioned medium supplemented with reducing agents under a nitrogen atmosphere.
After one week of cultivation under axenic conditions [33], [35], the different larval stages (primary cells, metacestode vesicles, protoscoleces) were analyzed for host cell contamination by organism-specific PCR. Chromosomal DNA was isolated from the larvae and from liver tissue of a non-infected jird using the DNeasy isolation kit (Qiagen). Part of the parasite specific gene elp (ezrin-radixin-moesin-like [36]) was amplified using the primers Em10-15 (5′-TCC TTA CCT TGC AGT TTT GT -3′) and Em10-16 (5′-TTG CTG GTA ATC AGT CGA TC-3′). As a control for host-DNA contamination, a previously described β-tubulin-encoding gene from Meriones unguiculatus was used [37], employing primers Tub12-UP and TUB12-ST as described [38].
In vitro cultivated material of all three larval stages was isolated and cell lysates were generated by first passing larval material repeatedly through a pipette tip, followed by one washing step in 1×PBS, and subsequent treatment with 50 µl of 2× STOP mix (2 ml 0.5 M Tris–HCl pH 6.8, 1.6 ml glycerol, 1.6 ml 20% SDS, 1.4 ml H2O, 0.4 ml 0.05% (w/v) bromphenol blue, 7 µl β-mercapto-ethanol per 100 µl) and boiling for 10 min at 100°C. Proteins were separated by SDS-PAGE and analyzed by Western blotting using an antibody directed against β-actin (Cell signalling technology®; No. 4967) of a wide variety of metazoan organisms. Images were subsequently analyzed for the relative expression of β-actin using the Image-J program (http://rsb.info.nih.gov/ij/) [39]. The relative expression transcribed as values of area under the curve (AUC) was used to normalize the β-actin content of each sample. In a first set of experiments, different amounts of in vitro cultivated primary cells, metacestode vesicles and protoscoleces were analyzed (Figure S1A) and the relative β-actin content of each sample was then used as a basis for normalization. Based on previous studies showing that, in vivo, oncosphere-derived stem cells develop into mature metacestode vesicles within 2–3 weeks upon infection [3]–[6], we first determined the amount of primary cells which, in our in vitro system, lead to the production of metacestode vesicles within the same time. This was the case when we used 1/6th of the amount of primary cells that can be isolated from 40 ml of metacestode vesicles (Figure S1B). This amount was defined as 1 Unit and contained ∼600 µg of total protein (Figure S1B). We then carried out calculations for the remaining larval stages and found that 2000 protoscoleces as well as 4 metacestode vesicles (2 months of age) of a diameter of 5 mm after 1 week of axenic cultivation represented 1 Unit and also contained ∼600 µg of total protein (Figure S1B). The reliability of this quantitative approach was further assessed by comparing one half unit of parasite material from each stage (i.e. 1000 protoscoleces, 2 metacestode vesicles of a diameter of 5 mm and 1/12th of the amount of primary cells that can be isolated from 40 ml metacestode vesicles) which, as shown in Figure S1C, also resulted in comparable β-actin content.
C57BL/6 mice were purchased from Charles River/Wiga (Sulzfeld, Germany) and TCR transgenic OT2 B6 mice were a kind gift of Prof. F. Carbone (Melbourne, Australia). All mice were bred within the animal facility of the Institute of Virology and Immunobiology, University of Würzburg, under specific pathogen-free conditions. Female mice were used at the age of 6–14 weeks.
Single cell suspensions were obtained from the spleen of C57BL/6 mice by mechanically squeezing the tissue with glass slides in cold PBS and filtered through a 70 µm nylon cell strainer. Red blood cells in the filtrate were lysed with 1,4% NH4Cl for 5 minutes at 37°C, and the splenocytes were washed in R10 medium, that is RPMI 1640 (GIBCO BRL) supplemented with penicillin (100 U/ml, Sigma, Deisenhofen, Germany), streptomycin (100 µg/ml, Sigma), L-glutamine (2 mM, Sigma), 2-mercaptoethanol (50 µM, Sigma and 10% heat-inactivated fetal calf serum (FCS, PAA Laboratories, Parsching, Austria). Cell counts were subsequently determined using the trypan blue (No. 26323, Biochrom, Berlin, Germany) exclusion test on a bright-lined Neubauer counting chamber.
DC were generated from the bone marrow (BM) precursors of C57BL/6 mice as previously described [40]. Briefly, BM precursor cells were cultured for 8 days with GM-CSF. At day 8, non-adherent DC (70–80% CD11c+ cells) were harvested and seeded at a density of 106 cells/ml R10 culture medium.
A comparable amount of axenically cultivated parasite material, normalized for β-actin content, from each of the three larval stages was used throughout the stimulation process. Tissue culture inserts (Greiner Bio-One) of 1 µm pore size with or without larvae were thoroughly washed in R10 medium to completely remove hepatocyte-conditioned medium, were then added to DC or splenocyte cultures, and kept at 37°C in the presence of 5% CO2 for different time points. For LPS stimulation experiments, inserts containing parasite material were removed after 24 h, DC were harvested and re-plated at an equal number of living cells (5×105 cells/ml) in R10 culture medium with or without 0.1 µg/ml lipopolysaccharide (LPS; E. coli 0127:B8; Sigma Aldrich) for additional 48 h. Upon completion, DC or splenocyte viability was determined by trypan blue exclusion (No. 26323, Biochrom, Berlin, Germany) on a bright-lined Neubauer counting chamber. Flow cytometric assessment of DC surface staining was then performed using fluorochrome-conjugated antibodies (anti-mouse) against the surface lineage marker CD11c (CD11c-PE-Cy5.5, Caltag Laboratories), MHC II (eBiosciences) and CD86 (B7-2-FITC, eBiosciences or B7-2-PE, BD Pharmingen). Splenocytes were stained for CD19 (CD19-pecy5, BD Pharmingen) as a specific marker for B cells and an exclusion marker for T cells within lymphocytes. To monitor the level of DC apoptosis, annexin-V binding buffer (BD Pharmingen) and FITC-conjugated annexin-V ready-to-use solution (BD Pharmingen) were used coupled to 7-AAD staining solution (BD Pharmingen). To assess DC maturation, marker-specific antibodies (CD11c, MHC II and CD86) were applied and after 30 min incubation at 4–8°C in the dark, cells were washed twice with FACS buffer (3% FCS, 0.1% NaN3 in PBS) and fixed with 1% (v/v) formaldehyde in PBS. The staining procedure was identically conducted for CD 19 on splenocytes. In DC apoptosis assays, stimulated DC along with UV-irradiated DC (positive control for apoptosis) at a peak intensity of 9000 mW/cm2 at the filter surface and a peak emission of 313 nm (trans-illuminator), were directly resuspended in 50 µl of 1× annexin-V binding buffer. Next, 5 µl of 7-AAD and 2 µl of annexin-V–FITC were added to the tubes and incubated for 15 minutes at RT. Cells were then resuspended in 200 µl of 1× annexin-V binding buffer then acquired on a FACSCalibur™ (Beckton Dickinson) cytometer, equipped with CellQuest software. Results were further analyzed with FlowJo software (Tree Star, USA).
After stimulation of DC, the production of IL-6, IL-10, and IL-12p70 was measured in supernatant using sandwich enzyme-linked immunosorbent assays (ELISA, OptEIA kits, BD Pharmingen) according to the manufacturer's instructions. The kits detection limits were of 39 pg/ml for IL-12p70 and 19 pg/ml for IL-10 and IL-6.
Spleens and lymph nodes from 6–14 weeks old OT2 mice were isolated, and the separated splenocytes and lymph nodes cells, obtained as described above (Splenocytes isolation), were resuspended in cold PBS. CD4+ cells were isolated using an EasyStep negative selection mouse CD4+ T cell enrichment kit (Stemcell Technologies). After separation, the purity of T-cell preparation was routinely higher than 90%, as determined by flow cytometry. CD4+ T cells were subsequently enriched for CD25− cells using Miltenyi Biotec's LD columns with a suitable MACS separator usually achieving 90–95% of purity.
DC were incubated with 3-fold higher numbers of OT2 CD4+CD25− T cells and 200 ng/ml of OVA protein (Sigma, grade V) supplemented or not with parasite larvae E/S-products (supernatant of equal amounts of larvae kept in medium for 7 days). After 5 days of co-culture, the cells were harvested and stained using the T-reg detection Kit (Miltenyi Biotec) prior to flow cytometric analysis.
All results were expressed as mean ± standard deviation (SD). Differences observed between groups were evaluated using the Wilcoxon/Mann-Whitney U test, a nonparametric test that does not assume normality of the measurements (it compares medians instead of means). Values of p<0.05 were considered statistically significant. All statistical analysis were performed with STATISTICA version 8.0.725.0 (StatSoft GmbH)
The morphology of the three different E. multilocularis larval stages investigated in this study is depicted in Figure 1A. Primary cells were isolated from the GL of metacestode vesicles and seeded into culture dishes where they formed aggregates with central cavities within one week of cultivation. As previously outlined, these primary cell aggregates closely resemble the early developing metacestode both morphologically and physiologically, and routinely result in the production of fully mature metacestode vesicles after 3–4 weeks of cultivation [28]. Furthermore, primary cell aggregates express factors such as members of the EG95/W45 protein family (or host protective oncospheral antigens) that are specifically present in taeniid oncospheres and are known to play an important role in early parasite establishment [30]. Primary cell aggregates thus closely mimic the E. multilocularis larval stage at the onset of the oncosphere-metacestode metamorphosis [29], [30]. In all experiments, primary cell cultures were carefully checked for the absence of mature, LL-containing metacestode vesicles (Figure 1Aa). Mature metacestode vesicles had a size of approximately 5 mm (diameter) and were completely equipped with a LL surrounding a cellular GL (Figure 1Ab). Protoscoleces (50–100 µm in size) were covered by a tegument and were used in a non-activated, dormant state (i.e. no pre-activation with low pH and trypsin), as they typically occur within metacestode vesicles in the intermediate host (Figure 1Ac). After one week cultivation under axenic conditions [33], [35], the absence of contaminating, cellular host material in all parasite samples was confirmed by organism-specific PCR (Figure 1B). Parasite material of each of the three larval stages was subsequently normalized on the basis of β-actin content (Figure 1C; Figure S1). In all subsequent DC co-cultivation procedures, comparable amounts of parasite material were used with 1 Unit defined as 2000 protoscoleces, 4 metacestode vesicles of 5 mm of diameter, and 1/6th of primary cells generated from 40 ml of metacestode vesicles after 1 week of in vitro culture (Figure 1C).
In a first set of experiments, the influence of E. multilocularis E/S-products on DC viability was tested. To this end, E. multilocularis larval material (1 Unit each for all three larval stages) was co-incubated for 48 h with DC, physically separated through a transwell system (1 µm pore size), and the number of viable DC was assessed by trypan blue exclusion. As shown in Figure 2A, co-incubation of DC with primary cells and metacestode vesicles led to greatly reduced viability (30–40% surviving cells compared to the control), whereas a still significant, but reduced killing effect was observed in the presence of protoscoleces (76,6+/−3,7% survival). To exclude the possibility that cell death in DC-parasite co-cultures merely resulted from starvation due to the presence of proliferating larvae, conditioned medium of a comparable amount of primary cells was tested on DC. To this end, supernatant from primary cell cultures (7 days old) was collected, sterile filtered, and added to fresh cultures of DC. As depicted in Figure 2B, even in the absence of proliferating larvae, conditioned medium similarly induced DC death.
In order to assess whether Echinococcus E/S-products do have general cytolytic effects that might account for the observed DC death, a hemolysis assay was carried out in which parasite material was co-incubated with human erythrocytes. However, as shown in Figure S2, no such effects were observed for any of the larval stages. We further tested whether E. multilocularis metacestode vesicles, which displayed the strongest killing effect on DC, could also affect other immune effector cells. Splenocytes from C57BL/6 mice were exposed to metacestode vesicles through a transwell system for 48 to 72 h and host cell viability was assessed by trypan blue exclusion. As shown in Figure 2C, in contrast to what we observed for BMDC, splenocyte viability was not affected by E/S-products of metacestode vesicles. We also specifically analyzed the CD19+ (B cells) and CD19− (primarily T cells) splenic lymphocytes for the effect of MCE/S on viability. As observed for the whole splenocyte population, the splenic B and T cell proportions were not altered upon 48–72 h of exposure to MCE/S (Figure 2D). Taken together, these data indicated that E/S-products of E. multilocularis larvae, particularly those that are released by primary cells and metacestode vesicles, induce murine DC death, but do not have general cytolytic properties and do not lead to killing in whole spleen cell preparations or alter the splenic B and T cell compartments in vitro.
To examine by which mechanism the E. multilocularis E/S-products induced DC death, Annexin-V/7-AAD dual staining was performed to differentiate necrotic (7AAD+) from apoptotic (Annexin-V+) cells. DC were separately exposed to comparable amounts of parasite material from each of the larval stages through transwells for 24 h, harvested and processed for staining. As shown in Figure 2E,F, following exposure of DC to E/S-products of primary cells or metacestode vesicles, 2-fold more DC underwent apoptosis than DC exposed to protoscolex E/S-products, which showed a similar rate of apoptosis as untreated DC. Taken together, these results indicated apoptosis as the primary mechanism by which E/S-products from primary cells and metacestode vesicles induce DC death.
Having shown that E/S-products of E. multilocularis primary cells and metacestode vesicles induce apoptosis in part of the co-incubated DC, we subsequently analyzed the fate of the surviving DC concerning maturation and cytokine release. To this end, DC were first incubated for 72 h with each of the three different E. multilocularis larval stages, separated by transwells. Subsequently, DC maturation was assessed by measuring the expression of surface markers MHCII and CD86 by flow cytometry (Figure 3A,B). Interestingly, E/S-products from primary cells and metacestode vesicles significantly inhibited the spontaneous DC maturation rate as compared to untreated DC, which was not observed for E/S-products of protoscoleces (Figure 3A, B).
To investigate whether parasite larvae also alter DC cytokine release, DC were first exposed to each of the parasite stages through transwells for 24 h after which the parasite larvae were removed. Upon this brief exposure to the various larvae, the DC were harvested, counted and an equal number of surviving DC were re-seeded in fresh medium for an additional 48 h. Supernatant was then harvested and the level of secreted IL-12, IL-6 and IL-10 was assessed by ELISA. As expected, bacterial LPS, used here as a positive control, led to a strong induction of all three cytokines in DC (Figure 3C). On the other hand, none of the parasite E/S-products was able to trigger IL-12 release. However, substantial amounts of IL-10 were produced upon exposure of DC to E/S-products of primary cells, whereas those of protoscoleces induced the production of IL-6. Following challenge by metacestode E/S-products, neither IL-6 nor IL-10 production were induced in DC (Figure 3C). Taken together, these results indicated that E/S-products of primary cells and metacestode vesicles inhibited the ability of DC to spontaneously mature in vitro, and failed to promote (or blocked) the release of detectable amounts of the pro-inflammatory and Th1-associated cytokine, IL-12. Interestingly E/S-products of primary cells additionally acted on DC to induce IL-10 secretion, a feature which was not seen with E/S-products of the metacestode. In contrast, E/S-products of protoscoleces fostered DC maturation and induced IL-6 but not IL-12 secretion by DC.
The consistent absence of detectable amounts of IL-12 in culture supernatant of DC after treatment with E. multilocularis larvae suggested that parasite E/S-products either failed to induce the secretion of this Th1-associated cytokine, or simply impaired its production by treated DC. Since there is increasing evidence that parasite survival within the intermediate host depends on the ability to deviate immune polarization away from a potential parasitocidic Th1 type [8], we examined whether exposure of DC to E. multilocularis E/S-products could have an influence on subsequent DC maturation by LPS, a stimulus usually associated with strong IL-12 release. To this end, DC were first incubated with each of the larval stages physically separated through transwells for 24 h. After this brief incubation time, the parasite larvae were removed and the DC were harvested, counted and an equal number of surviving DC were re-seeded in fresh medium containing 0.1 µg/ml of LPS for an additional 48 h. At the end of this second incubation period, DC were assessed for their level of maturation (MHCII and CD86 surface molecule expression) and DC supernatant was analyzed by ELISA for the presence of IL12p70, IL10 and IL6. As shown in Figure 4A/B, in contrast to the situation for protoscolex co-cultures, LPS did not induce maturation when DC had previously been cultivated in the presence of primary cells or metacestode vesicles. Furthermore, pre-exposure to E/S-products of all three larval stages significantly altered the cytokine profile of DC upon subsequent LPS stimulation, again leading to a strongly diminished release of the Th1 associated cytokine IL-12 (Figure 4C). Taken together, our results indicate that exposure of DC to E/S-products of primary cells inhibits maturation in response to LPS, diminishes the ability to produce IL-12 and IL-10, but induces IL-6 production. E/S-products of metacestode vesicles act similarly on DC, but have a reduced effect on LPS-induced IL-10 and IL-6 production. In contrast, E/S-products of protoscoleces do not inhibit the expression of MHCII and CD86 surface markers by LPS-stimulated DC, but strongly affect LPS-induced IL-12 production.
A recent report has suggested an implication of Foxp3+ regulatory T cells (T-reg) in E. multilocularis larval establishment and/or persistence within the intermediate host [15]. Furthermore, a key role of DC in inducing the de novo generation of Foxp3+ T-reg is well established [41]–[43]. Using an in vitro OVA peptide-based assay for the co-cultivation of DC and CD4+ T cells [44] we therefore tested whether additionally present E/S-products of E. multilocularis larvae could lead to an expansion of CD4+CD25+Foxp3+ T cells, commonly assumed to be T-reg [41]–[43]. To this end, co-cultures of freshly generated DC (day 8) and naïve CD4+CD25− T cells from OT2 αβ TCR-transgenic mice at a DC/T cell ratio of 1/3, supplemented with OVA peptide, were exposed to E/S-products of all three larval stages. Upon 5 days of incubation, the cells were harvested and stained for Foxp3+ T-reg-specific cell markers (CD4, CD25 or IL2Rα chain and Foxp3). Interestingly, in contrast to co-cultures of DC and T-cells with E/S-products of primary cells and protoscoleces, there was a significant expansion (2.5-fold) of the population of T-reg (CD4+CD25+Foxp3+) in co-cultures that included E/S-products of metacestode vesicles (Figure 5). These results indicated that at least the developmental stage that characterizes the chronic phase of AE, the metacestode, is able to induce de novo CD4+CD25+Foxp3+ T-reg conversion in vitro in the presence of DC.
As typical in the case of helminth infections, AE is a long-lasting and chronic disease that is most probably associated with parasite-induced, immunosuppressory mechanisms around the primary site of infection [8]. For a number of nematode and trematode systems, research during recent years has demonstrated a crucial role of T-reg in the respective immunosuppressive mechanisms and emphasized the importance of DC in the induction of helminth-associated Th2- and tolerogenic immune responses [11], [12]. Compared to nematode and trematode infections, immunomodulatory functions of DC in cestode infections have drawn significantly less attention, although this is clearly an emerging field since several studies concerning the influence of parasite products on DC maturation and cytokine secretion profiles have been conducted very recently. In two of these recent reports, Reyes et al. [45] and Terrazas et al. [46] investigated the effects of E/S-products of Taenia crassiceps cysticerci, representing the metacestode larval stage of this Taenia infection model, on the activation of murine DC. These authors observed impaired DC maturation in response to TLR dependent stimuli, particularly when DC of infection susceptible mouse strains were pre-incubated with parasite E/S-products [45]. In the case of E. granulosus, a species closely related to E. multilocularis, the effects of hydatid cyst fluid (HCF) and isolated antigen B (AgB), a major constituent of HCF, were tested and led to DC maturation as well as DC cytokine profiles that were indicative of Th2 immune responses [19], [20]. However, whether these interactions are of major relevance in vivo remains questionable since intact parasite tissue usually prevents direct contact between HCF and host immune effector cells, and the spectrum of metacestode E/S-products does not necessarily overlap with the spectrum of proteins present in HCF. Although it is generally assumed that AgB might leak out of intact metacestode vesicles or be released early during an infection from damaged metacestode material [47], we could not detect AgB in the E/S-products of in vitro cultivated E. multilocularis metacestode vesicles despite the fact that this component was well expressed in HCF [48]. Crude metacestode antigen preparations containing vesicle fluid, somatic parasite proteins and contaminating host components [16] as well as isolated vesicle fluid of E. multilocularis [20] were also already tested concerning their effects on DC and failed to induce maturation as did a purified mucin-type glycoprotein (Em2) that is usually expressed at the surface of LL-containing metacestode vesicles [49], [50]. Hence, although all these reports indicate that larval cestode parasite products can exert immunomodulatory effects on host DC, depending on the source and form of application, their precise nature and role during the course of an infection of the intermediate host remains elusive so far. What became clear through a recent in vivo investigation on experimentally infected mice, on the other hand, was that, at least in the chronic phase of experimental AE, peritoneal DC display a significant down-regulation of surface markers that are associated with DC maturation, and over-expressed TGF-β mRNA, which might lead to an induction of T-reg in this phase of the disease [15], [51]. A potential role of T-reg in human AE has also already been suggested [8] based on the fact that immune-suppressive TGF-β and IL-10, major cytokines that are released by T-reg [52], can be predominantly found in the immediate vicinity of actively proliferating parasite tissue.
In order to more closely mimic the situation at the site of infection, we utilized in this study a cultivation system by which actively secreted E/S-products of living parasite material can be tested on DC. Furthermore, in addition to parasite components that are produced during late stages of the infection (metacestode and eventually protoscoleces), we also included larval material that represents early stages of the infection, prior to the establishment of metacestode vesicles, in which E. multilocularis should be highly susceptible to host attacks due to the absence of a protective LL [7].
First, although co-incubation with protoscolex E/S-products clearly induced DC maturation, no activation could be observed upon co-incubation with E/S-products of primary cells and metacestode vesicles. The cells were not only affected in the expression of surface activation markers but also failed to secrete pro-inflammatory cytokines. This inhibitory effect was even apparent in the presence of strong stimuli of TLR signaling since DC pre-incubated with E/S-products of primary cells and metacestode vesicles did not mature in response to LPS, whereas pre-incubation with E/S-products of protoscoleces had no such effect. Furthermore, in a much more pronounced manner than protoscolex compounds, E/S-products of both primary cells and metacestode vesicles induced DC death mediated by apoptosis. To our knowledge, this is the first report on the induction of apoptosis in host DC in response to cestode larval material, which should have important implications concerning immuno-suppressive activities, particularly at the very early stage of the infection. On the one hand, the induction of apoptosis in DC should be beneficial to the parasite since it depletes immune effector cells around the early parasite lesions that are important to induce inflammatory immune responses. Moreover, it is well established that apoptosis, extrinsically triggered by infectious agents such as viruses, parasites, or bacteria, usually results in a bystander effect of induced immunosuppression [53]. In parasitic helminths, the induction of DC apoptosis has already been reported for microfilariae of the nematode Brugia malayi which, as also shown herein for E. multilocularis, strongly limited their capacity to produce pro-inflammatory IL-12, and prevented T cell activation and proliferation [54]. Previous in vitro studies further demonstrated that apoptotic DC are rapidly taken up by immature DC, which prevents subsequent maturation of immature DC in response to TLR stimuli [53]. It is, therefore, conceivable that the strongly diminished ability of DC that were pre-incubated with E/S-products of primary cells and the metacestode to LPS, as observed in our study, is indirectly mediated by the induction of apoptosis in a subset of immature DC, rather than by direct inhibition of DC maturation through parasite E/S-products. Since the uptake of apoptotic DC induces immature DC to secrete TGF-β, which induces differentiation of naïve T cells into Foxp3+ T-reg [53], E/S-products of the metacestode, and particularly of primary cells, could thus establish a strongly immunosuppressive environment around parasite lesions already at the beginning of an infection.
As in the case of E/S-products produced by B. malayi microfilariae [54], we can currently only speculate about the molecular nature of Echinococcus E/S factors that might induce DC apoptosis. Among the various host-derived compounds that can extrinsically trigger DC apoptosis are ligands of the tumor necrosis factor (TNF) superfamily as well as glucocorticoids [53] and B. malayi microfilariae have already been demonstrated to induce DC apoptosis by triggering TNFα-dependent signaling mechanisms [55]. Although no TNF-like ligand has been described so far in Echinococcus or any other flatworm, there has been a recent report on the presence of a TNFα-receptor like surface protein in Schistosoma mansoni which presumably interacts with host TNFα [56]. Our own preliminary analyses on the E. multilocularis genome, which is currently being sequenced [29], [30], revealed that a very similar receptor is also expressed by cestodes (data not shown). Bioinformatically, TNF-ligands are difficult to identify in raw sequence data, which might be the reason why so far no such molecule was identified in the Schistosoma or Echinococcus genomes. However, the presence of a respective receptor in these organisms implies that they might also express cognate ligands, which subsequently could bind to members of the TNF-receptor family on host cells (such as DC), thus triggering apoptosis. Apart from components of the TNF signaling machinery, it has recently also been demonstrated that cestode larvae (T. crassiceps cysticerci) are capable of producing steroid hormones [57]. Although in this system only the production of sex steroids has been tested, it is conceivable that they also produce glucocorticoids which might, either together with TNF-ligands or as an alternative, be involved in triggering host DC apoptosis. By utilization of the culture system established in this work, these alternatives can now be addressed.
A marked difference between DC that were incubated in the presence of E/S-products of metacestode vesicles and primary cells was that, in the latter case, the production of anti-inflammatory IL-10 was significantly induced. To our knowledge, an elevated expression of IL-10 coinciding with DC apoptosis has so far never been described, indicating that this effect was not provoked by the elevated induction of DC apoptosis through E/S-products of primary cells, when compared to those of the metacestode. Hence, we rather suggest that primary cells secrete a set of factors that differs from E/S-products of the metacestode and contains additional components that are able to induce the expression of IL-10 by non-activated DC. This hypothesis is supported by the differential influence of E/S-products from primary cells and metacestode vesicles that we have observed in T-reg conversion assays. Only E/S-products of metacestode vesicles, but not those of primary cells (or protoscoleces), were able to significantly increase the number of CD4+CD25+Foxp3+ regulatory T cells in vitro. Although we cannot presently tell whether the in vitro T-reg conversion was exclusively mediated by the modified DC, or whether there is also a direct influence of parasite E/S-products on CD4+ T cells, these data nevertheless clearly support an emerging picture that points at Foxp3+ T-reg cells as potential mediators of the fine tuning of the host immune system during metacestode establishment and growth within the intermediate host [15]. Furthermore, our data suggest that an expansion of Foxp3 expressing T cells during chronic (peritoneal) AE, as observed by Mejri et al. [15], might not simply be an intrinsic consequence of an ongoing immune response, but that the parasite actively induces Foxp3+ T-reg through its E/S-products.
The factor(s) and mechanism(s) involved in the modulation of DC maturation and function are currently subject to ongoing investigations. These include studies on the possibility that Echinococcus E/S-products might directly interact with TLR ligands rendering the latter less able to elicit a ‘normal’ response from DC, or may be acting as antagonists of TLR-ligand binding interactions [11]. Furthermore, parasite E/S-products may affect DC directly through interactions with non-TLR pattern recognitions receptors such as DC-SIGN, Dectin family members [11]. In previous studies on other helminth systems, secreted compounds such as filarial cystatins [58] were shown to induce the expression of IL-10 in antigen presenting cells, including non-activated DC. Interestingly, our own preliminary analyses of the E. multilocularis genome sequence indicate that related molecules are also encoded by the cestode, although further experimentation is clearly necessary concerning a possible secretion of these molecules by PC or whether they exert immunomodulatory activities comparable to those of filarial cystatins. Regarding T-reg conversion, the so far best characterized component that elicited similar in vitro effects as E/S-products from metacestode vesicles was a secreted compound of the nematode Heligmosomoides polygirus with TGF-β-like activities [59]. Due to the fact that TGF-β-signaling mechanisms have already evolved very early in animal evolution, TGF-β-like cytokines are expressed by a wide variety of free-living, but also parasitic invertebrates [60], [61]. Notably, at least one gene that encodes a structural homolog of mammalian TGF-β is also present on the genome of E. multilocularis [62], and the involvement of this component in the in vitro T-reg conversion process induced by metacestode E/S-products is currently investigated by us using the in vitro cultivation models established in this study.
In sharp contrast to co-incubation with primary cells and metacestode vesicles, DC exposed to E/S-products of protoscoleces were clearly activated, as assessed by up-regulation of surface activation markers (MHCII and CD86), secreted elevated levels of IL-6 (but no IL-10), and strongly impaired the ability of DC to produce IL-12 in response to TLR stimuli (LPS). This phenotype resembles that of DC that had been incubated in the presence of E. granulosus HCF and isolated AgB [19], [20]. However, in contrast to these investigations, DC incubated with protoscolex compounds in our study did not release elevated levels of IL-10, as reported by Rigano et al. [20], or IL-12, as reported by Kanan and Chain [19]. This is most probably due to the fact that the spectrum of E/S-products of protoscoleces does not fully overlap with the content of HCF since, for example, AgB is only weakly expressed by protoscoleces [30], [63]. In general, however, the phenotype of DC upon co-incubation with E/S-products of protoscoleces in our study is largely comparable to that of DC incubated with certain Trypanosoma antigens which have been closely associated with the induction of Th2-dominated immune responses [64]. Whether the Th2 immune response that is characteristic of the chronic stage of AE [8] is provoked (or supported) by direct contact between protoscoleces and DC within the intermediate host remains highly questionable, since this larval stage is only produced very late in the infection and direct contact between protoscoleces and host cells is usually prevented by the parasite's surface layers. Furthermore, Th2-dominated immune responses can also be observed in chronic AE under conditions in which no protoscoleces are produced [65]. However, since intestinal luminal infections by adult cestodes are associated with Th2 immune responses [66], the phenotype we observed in this study for DC exposed to E/S-products of protoscoleces could rather be associated with immunological processes that are relevant for an infection of the definitive host [66]. In any case, the marked differences between the responses of DC to E/S-products of early versus late developmental stages of E. multilocularis clearly demonstrates that an induction of tolerance in DC is not a general characteristic of Echinococcus material, but rather that the E/S repertoire of primary cells and metacestodes has specifically evolved to fulfill these purposes. Care should therefore be taken in the interpretation of results that have been obtained concerning the immune response during echinococcosis (intermediate host infection) by using co-incubation-systems of Echinococcus protoscoleces with host cells [21]–[27] or by employing the mouse model of peritoneal, protoscolex-induced secondary alveolar echinococcosis for short-term infections [67].
In conclusion, in this study we provide for the first time evidence for the induction of apoptosis in host DC through E/S-products of early infectious stages of E. multilocularis. We further show that primary cells, as representative of the oncosphere stage that undergoes metamorphosis towards the metacestode, are able to induce poorly responsive, IL-10 secreting DC in vitro. This effect is somewhat reduced at the chronic stage (metacestode), leading to poorly responsive, immature DC, but a Foxp3+-T-reg-inducing environment, and is no longer present in the protoscolex stage (table 1). Although our study concentrated on in vitro interactions between parasite larvae and DC, thus excluding the possible influence of other immune effectors or epithelial cells, the clear induction of poorly responsive, apoptotic and IL-10 secreting DC in response to primary cells suggests that a similar mechanism might also be operative in the tissue surrounding the early metamorphosing oncosphere. If so, this process might be important for an early establishment of the parasite during a phase of relatively high vulnerability to the host immune system, whereas in the chronic phase, after production of the LL, a slightly altered profile of E/S-products that mainly induces T-reg could support long-term persistence and infiltrative growth of the metacestode, as previously suggested [15]. The molecular nature of Echinococcus E/S-products that are responsible for these effects is currently being investigated by us using the available genome sequence information [29], [30], recently established methods for genetic manipulation of primary cells [68], and the cultivation settings established in this work.
|
10.1371/journal.pcbi.1002025 | Cell-Sorting at the A/P Boundary in the Drosophila
Wing Primordium: A Computational Model to Consolidate Observed Non-Local Effects
of Hh Signaling | Non-intermingling, adjacent populations of cells define compartment boundaries;
such boundaries are often essential for the positioning and the maintenance of
tissue-organizers during growth. In the developing wing primordium of
Drosophila melanogaster, signaling by the secreted protein
Hedgehog (Hh) is required for compartment boundary maintenance. However, the
precise mechanism of Hh input remains poorly understood. Here, we combine
experimental observations of perturbed Hh signaling with computer simulations of
cellular behavior, and connect physical properties of cells to their Hh
signaling status. We find that experimental disruption of Hh signaling has
observable effects on cell sorting surprisingly far from the compartment
boundary, which is in contrast to a previous model that confines Hh influence to
the compartment boundary itself. We have recapitulated our experimental
observations by simulations of Hh diffusion and transduction coupled to
mechanical tension along cell-to-cell contact surfaces. Intriguingly, the best
results were obtained under the assumption that Hh signaling cannot alter the
overall tension force of the cell, but will merely re-distribute it locally
inside the cell, relative to the signaling status of neighboring cells. Our
results suggest a scenario in which homotypic interactions of a putative Hh
target molecule at the cell surface are converted into a mechanical force. Such
a scenario could explain why the mechanical output of Hh signaling appears to be
confined to the compartment boundary, despite the longer range of the Hh
molecule itself. Our study is the first to couple a cellular vertex model
describing mechanical properties of cells in a growing tissue, to an explicit
model of an entire signaling pathway, including a freely diffusible component.
We discuss potential applications and challenges of such an approach.
| In developing animal tissues, cells can often re-arrange locally and mix
relatively freely. However, in some stereotypic and crucially important
instances during body development, cells will strictly not intermingle, and
instead form sharp boundaries along which they will sort out from each other.
This mechanism helps organisms to establish signaling centers and to maintain
distinct cellular identities. Often, cells at such boundaries will remain in
close physical contact and are morphologically alike. Thus, the boundary itself
can be difficult to observe unless the expression status of specific marker
genes is monitored experimentally. How are these ‘compartment
boundaries’ established? Here we devise a computational model that aims to
describe one such boundary in a well-studied animal tissue: the developing wing
primordium of Drosophila melanogaster. We model the production,
diffusion and local sensing of an essential signaling molecule, the
Hedgehog protein. We reveal one possible mechanism by which
Hedgehog sensing can influence the mechanical properties of cells, and compare
the simulated outcome to observations in experimentally perturbed, actual wing
discs. Our relatively simple model suffices to establish a straight and stable
compartment boundary.
| During embryonic development of complex multicellular organisms, spatial reference
points need to be established within tissues. These are often formed by specialized
groups of cells that are capable of signaling to neighboring cells. Such signaling
centers define coordinate systems along which newly arising cells can orient
themselves and make crucial decisions regarding proliferation, differentiation or
migration [1],
[2], [3], [4], [5], [6]. Because of
their pervasive importance, tissue-organizing centers need to be precisely
controlled – both spatially and temporally, as well as with respect to their
signaling amplitude.
One possible mechanism for spatial control of tissue organizers is to restrict the
movement of cells at fixed boundary positions. This phenomenon is indeed observed,
and it involves the separation of groups of cells that have already been spatially
instructed to assume distinct identities, for example at segment- or
parasegment-boundaries. Akin to water in oil, the two cell populations are seen to
establish and maintain a relatively straight interface to each other, effectively
minimizing their contact area. The minimizing force is assumed to help stabilize the
interface against random perturbations that may arise from cell divisions or from
arbitrary cell movements; thus, any organizing activity that is associated with the
interface is likewise stabilized. How is this separation, or ‘sorting’,
of cells of distinct identities achieved? One line of work attributes this to
differential cell adhesion [7], [8]: cell populations might develop distinct adhesive
properties; these affinity differences would then allow them to sort out from one
another. Another line of reasoning is based on Differential Interfacial Tension
(DIT) [9], [10]: this
hypothesis suggests that cells might actively constrict surfaces that are in contact
with neighboring cells, depending on the cellular identity of neighbors and/or
depending on signaling events. Both mechanisms would ultimately lead to physical
forces that would help keep the cell populations apart.
The developing wing primordium of Drosophila (‘wing
disc’) is particularly well suited to study boundary formation (Figure 1). It is not required for
larval viability, can be manipulated experimentally through an advanced genetic
toolkit, and has been well characterized. The disc contains a compartment boundary
that separates anterior from posterior cells; this boundary is inherited from
specification events occurring early in the embryo. The initial embryonic events
that give rise to the boundary involve mutual signaling between stripes of cells,
mediated by an extensively studied network of genes (the ‘segment polarity
network’ [11], [12], [13], [14]). Once established, the cellular identities on both sides
of these boundaries are stable throughout larval development and well into adult
life. The compartment boundary in the disc is strictly respected by all cells, even
when cells on one side of the boundary are artificially provided with a competitive
growth advantage over cells on the other side of the boundary [15]. The wing disc itself is
a simple, flat, epithelial sheet, and the orientations of cell divisions appear
largely random [16]. Genetic analysis and computational modeling of this
tissue is simplified by the fact that daughter cells arising from cell divisions
usually remain in physical contact and do not migrate away from each other. This has
been shown experimentally by tracing descendents of single cells; in most cases such
a ‘clone’ of offspring cells will form a coherent patch of connected
cells. This behavior suggests that the complicated processes of cell intercalation
and migration can be neglected, to a first approximation, when studying boundary
maintenance in this tissue.
Working with such wing discs, a recent, seminal study has begun to shed light on
possible boundary formation mechanisms [16] (see also ref [17]). The authors
have directly demonstrated an increased mechanical tension at cell-to-cell
interfaces located immediately at the boundary, using laser ablation experiments.
Subsequent computer simulations then revealed that collectively such local forces
are sufficient to maintain a stable compartment boundary. These results are
intriguing, but they raise a number of new questions: Boundary formation in the wing
disc is known to depend on the secreted and diffusible signaling protein Hedgehog
(Hh), which is produced by posterior cells and specifically sensed and transduced by
anterior cells [18], [19] (Figure
2). If diffusible Hh indeed somehow influences mechanical tension, what
conditions must then be met to ensure a well-defined boundary? So far, all known
transcriptional responses of Hh signaling are occurring several cell-diameters wide
into the responding tissue. How is the response in this case restricted to the
immediate boundary region? Furthermore, experimental suppression of Hh signaling has
been shown to lead to ectopic boundary formation distant from the actual boundary
[20]. Does
this mean that the influence of Hh signaling does extend beyond the actual boundary,
and if so, why does this not have a noticeable consequence in the wild type
situation?
Here, we propose a mechanistic model that can generate a localized outcome of Hh
signaling with respect to physical forces and mechanical properties, despite a
longer range of the molecular response in terms of target gene expression.
Furthermore, we estimate the distance from the boundary, up to which Hh signaling
may be able ‘prime’ cells for boundary formation; this distance is
inferred using both experimental results as well as modeling results, and we
estimate it to be at least 10 cell diameters. We approach the problem by first
formulating an explicit, two-dimensional model of Hh production, diffusion and
transduction, and by then coupling this setup to a physical model of the growing
tissue. In our modeling approach, cells and their contact surfaces are described as
a graph of connected vertexes. Our model essentially follows the Differential
Interface Tension hypothesis; it is a modified version of a model that has been
previously established for the very same tissue [21]. We observe good
compartment boundary formation over a range of simulation parameters, and the
modeling outcomes agree qualitatively with experimental perturbations specifically
performed for this study.
In principle, at least two distinct molecular scenarios could explain the local
generation of tensile forces at the boundary (Figure 1). In the first scenario (ref [16]), two
different cell-surface molecules would form a heterotypic interaction at the
boundary; their expression would essentially be under the control of the anterior or
posterior “identities” of cells on either side of the boundary. The
heterotypic interaction of these two molecules would be sensed locally at the
cell-interaction interfaces, which would then respond by generating increased
physical tension. This is a simple and attractive model, but it is not
straightforward to consolidate with the known requirement, on the anterior side, for
reception and transduction of the Hh signal. Loss of Hh transduction can generate
ectopic boundaries in the anterior compartment ([20], this study), but it is
generally not presumed that such loss of Hh signal will change the identity of
anterior cells into that of posterior cells: the expression status of the selector
genes engrailed and ci is not affected by Hh signaling. Thus, if the cell-identity
seems unchanged, then both of the putative cell-surface molecules required for this
model would have to be under Hh control: one as a direct molecular target, and one
as an “inverse” molecular target (i.e., de-repressed upon the loss of Hh
signal). Only in such a setup would loss of Hh transduction lead to ectopic boundary
formation within the anterior compartment. However, target gene de-repression upon
loss of Hh signal has to our knowledge never been reported for any known Hh target
gene, and it would likely require further, more complicated indirect signaling
mechanisms. Alternatively, one might imagine that one of the two molecules was
expressed ubiquitously throughout the tissue, and only the other molecule would be a
target of Hh signaling. However, in such a scenario heterotypic binding would occur
throughout the entire Hh target gene expression domain; increased tension would thus
not be restricted to the immediate compartment boundary only.
We therefore propose an alternative, somewhat more parsimonious scenario (Figure 1): The increased tensile
forces at the boundary would be the consequence of a single cell-surface molecule,
which would be a simple, direct molecular target gene of Hh signaling. This molecule
would be able to transmit a signal to the inside of the cell, but only upon its
activation by a homotypic interaction with molecules of the same type from a
neighboring cell. Crucially, as discussed in more detail below, this signal and its
conversion into mechanical tension would have to be rate-limited: relative tension
would be highest at the section of the cell where most of the molecule has been
activated, but the overall tension per cell would be constant (i.e. independent of
the absolute amount of activated cell-surface molecule).
To accurately take into account the role of Hh signaling in the disc (Figure 2), we first devised a
formal model for the production, diffusion and transduction of Hh in this
two-dimensional tissue. The model (Figure 3) includes the Hh receptor Patched (Ptc), as well as the
essential downstream signaling component Smoothened (Smo), together with an unknown,
putative co-factor of Smo; this co-factor is not further specified but has been
speculated to be a lipid [22], [23]. The Hh protein is allowed to freely diffuse throughout
the tissue, following its production in posterior cells. For each individual cell
within the tissue, we compute the concentrations of the modeled entities as they
develop in time by numerically solving a set of partial differential equations
(Figure 3B). Apart from
known or suspected players in Hh signaling, we implement an additional, putative
target gene of Hh signaling, which we term “TMx”. Unlike Ptc, this gene
is presumed to play no active role in the signaling pathway itself, instead it is a
downstream target of the pathway and does not feed back into the sending or
receiving of the Hh signal. We assume this gene to have the simplest possible
connection to Hh signaling, namely a production term proportional to the amount of
active Smo molecule in anterior cells. We further assume that the product of this
gene has a function in regulating cortical tension at the inner surface of cells. We
do not specify the molecular mechanism by which it regulates tension, but one could
for example envisage TMx being a transmembrane protein whose intracellular domain
recruits or otherwise influences cortical actin filaments [24]. Since TMx is modeled as a Hh
target gene, it provides a way to connect transcriptional Hh responses to physical
forces acting on cell shapes (Figure
3C). Our model is based on three central assumptions with regard to TMx:
first, that it would increase cortical tension only in response to homotypic
activation, i.e. upon binding another TMx molecule presented on the surface of a
neighboring cell. Second, that it cannot increase the overall
propensity of the cell for exacting cortical forces, but instead merely
re-distributes cortical tension factors among the various interfaces that a given
cell has with its neighbors. Again, we do not specify why this might be the case,
but one could envisage a dynamic equilibrium of cytoskeleton filament deposition,
and removal, at the cortex. In such a situation, each section of the cellular
surface competes with all other sections within the same cell for the build-up of
cytoskeleton material, and activated TMx might simply tip the balance towards
deposition, locally. Lastly, the TMx molecule itself (while initially expressed
isotropically) would enrich at cell surfaces at which it is activated by homotypic
binding, perhaps because it is stabilized or preferentially re-deposited there.
Thus, the overall effect of TMx would be that it changes the relative strength of
contractile forces at each individual cell/cell contact segment; we model this as
scaling factor within the line tension term of the physical energy function (Figure 3C).
For our implementation of the full model, a challenge was to accurately compute the
two-dimensional diffusion of the Hh protein on a geometry that is itself constantly
changing. We achieve this by alternating the mechanical relaxation/growth
computations with an explicit diffusion of Hh on finite volumes established by the
shapes of the cells (see Supplemental Material, Text S1). It
should be noted that our model does not address questions related to overall
regulation of tissue growth or to the determination of final organ size (nor does it
address issues of correct developmental timing). Detailed models for growth control
and mechanical forces affecting the tissue as a whole have been developed already
[25],
[26], [27], but
they do not need to be applied here because our readouts are local, and because we
stop the simulations well before the tissue would normally cease growing.
Having specified the model, we next set out to parameterize it. Experimentally
quantified data regarding the various kinetic parameters in Hh signaling are
difficult to obtain and are at present quite sparse. We therefore focused our
parameter exploration and validation on the modeled shapes of the
various concentration gradients in the tissue (rather than on the absolute molecular
concentrations); these shapes are already much better known, mainly from antibody
staining experiments. For simplification, we performed parameter exploration in one
dimension only, by projecting molecular concentration gradients along an
anterior-posterior transect of the tissue (Figure S1). The Ptc protein in particular served
as a guide for our manual parameter optimization – it is itself a target gene
of Hh, and its expression and activity gradients are understood comparatively well
[28]. As is
shown in Figure
S1, our model resulted in the characteristic up-regulation of Ptc in a
small stripe of cells anterior to the boundary. Remarkably, the Ptc protein
concentration gradient shows an approximately sigmoidal shape when projected along
the antero-posterior axis, with highest values close to the boundary; this is not
specified in the model as such, but instead follows naturally from the wiring of the
pathway, with Ptc being both the receptor and a direct target gene of Hh. Because
the parameter space of our model is fairly large, and each simulation run takes
several CPU hours, a fully systematic scan of the possible parameters is difficult.
Instead, we explored the parameter space manually. Thus, our parameter set should
and will be updated as experimental data on concentrations and kinetic constants
become available; any updates will again have to reproduce the known shapes of
concentration gradients in the tissue. Initial test runs of our model revealed that
several parameter sets resulted in the formation of a stable lineage boundary at the
anterior/posterior interface (see for example Figure 3D and Figure 4). The resulting overall tissue-shapes
often revealed a small constriction of the tissue margins at the position of the
boundary, suggesting that the boundary exerts long-range mechanical forces on the
tissue as a whole, as might be expected (Figure 3D).
Next, we validated the overall distribution of cell shapes in the simulated tissue,
i.e. the distribution of cells over the various possible polygon classes (i.e.,
number of edges per cell), and the dependency between polygon class and cell surface
area. We based this on published experimental data (cell shape measurements) from
refs [16] and
[21]. This
test further constrained our model parameters (Figure 5). As shown previously, the relative
settings of the main parameters of the energy function (i.e., perimeter elasticity
factor vs. line tension factor ) can be varied over a
certain range, without resulting in much deviation between modeled and measured cell
shapes. In our case, the added requirement of a stable boundary, which should mimic
the actual boundary in the disc, constrained the parameters even further. For
example, we noticed that relaxing the relative strength of the ‘perimeter
elasticity’ parameter (third row in Figure 5) resulted in the best overall appearance
of the boundary; however this was accompanied with a reduced fit to the
polygon-distribution, and with somewhat unrealistic (elongated) cell shapes
immediately adjacent to the boundary. As the best subjective compromise, we
identified the parameter setting and
(first row in Figure 5). At this point in parameter space, we observed the best fit to
known cell sizes and shapes, while at the same time obtaining a fairly straight
boundary (see also the comparison to a negative control in Figure 3D). Immediately at the boundary, our
model posits an approximately twelve-fold difference between TMx expression levels
(anterior cells in row A1 having maximal TMx concentration of roughly 2.4 a.u. vs.
posterior cells in row P1 with a basal TMx concentration of 0.2 a.u.; see Figures 3D and S1D). For
average six-sided cells, this translates to a roughly two-fold increase in line
tension at the boundary (see Figure S3F) – in good agreement with
laser-ablation experiments [16], in which a 2.5 fold increase had been measured.
Our model qualitatively recapitulated the configurations observed in actual wing
discs (including a localized boundary in spite of a longer-ranging response to Hh),
so we next tested whether it would also correctly recapitulate the effects of
genetic perturbations in the pathway. As described previously [20], [29], the transduction and response
to the Hh signal can be blocked, in cells anterior to the boundary, by the removal
of the essential Hh pathway protein Smo. This is achieved experimentally by inducing
mitotic recombination in a small subset of cells, early in development, in larvae
that are heterozygous for a mutant in the smo gene. The resulting
small patches of homozygous mutant cells (clones) have been demonstrated to display
two types of behavior [20], [29]: first, when situated close to the boundary, they tend to
round off, minimizing their contact with neighboring cells. Second, when situated
immediately adjacent to the boundary (specifically: at its anterior side), they tend
to sort out from anterior cells and migrate into posterior territory. Both effects
are interpreted as evidence for ectopic boundary formation – cells inside the
clone are not receiving the Hh signal, but are juxtaposed to cells that do (this
mimicks the situation at the boundary and leads to the rounding off, and/or to the
migration into the posterior compartment that also does not transduce Hh). For our
present study, we have repeated these experiments for a number of wing discs, and
used automated image processing to quantify the extent of the
“rounding-off” effect (Figure 4). We observed a highly significant distance-dependence of the
rounding-off behavior: clones farther away from the boundary are rounding off less
strongly than clones closer to the boundary
(p = 6·10−6). As expected, this
effect is not observed on the posterior side of the boundary, where Hh signaling has
no known effects. This suggests that, whatever the molecular response to Hh
signaling that is contributing to boundary formation, this response does extend
further into the anterior tissue than just the immediate first row of cells at the
boundary. In essence, cells seem to be “primed” for boundary formation,
by Hh, several cell-diameters wide into the tissue.
In our model, we can arbitrarily set the Smo production rate to zero for any cell
(and its descendents), thus mimicking the experimental situation. We find that we
can qualitatively recapitulate the behavior of
smo− clones in our simulations (Figure 4): clones situated close
to the anterior side of the boundary, but not on the posterior side, can be observed
to round off; in addition, we observe a tendency of clones that immediately straddle
the boundary to migrate from anterior towards posterior territory (but not in the
opposite direction). Importantly, similar to the experimental situation, we also
observed a highly significant distance-dependence for the extent of rounding-off
(with respect to the distance to the boundary, again only on the anterior side).
This confirms that our model can correctly recapitulate this important aspect of the
perturbation, and it supports our interpretation of the situation in the wing disc:
a hypothetical transcriptional target of Hh signaling could be sufficient to
generate a strictly local force that can establish a clearly delinated compartment
boundary, despite this target being expressed (like all known transcriptional
targets) over a certain distance away from the boundary. By assessing the shape of
experimental smo− clones, we can effectively chart
out the predicted expression level of this putative gene; it appears to be expressed
roughly similar to ptc or dpp (in a graded stripe
of expression along the boundary, at least 10 cell diameters wide).
When mutant cells are generated experimentally using mitotic recombination, a sister
cell is generated that is not homozygous mutant, but instead homozygous wild-type in
the smo gene. This so-called “twin-spot” provides
another relevant input for our modeling: it presumably contains a larger amount of
Smo protein (relative to the surrounding heterozygous tissue). We note that, both in
the experiment and in our simulation, this difference in Smo levels does not suffice
to generate a significant rounding-up of twin-spots (Figure 4). Indeed, the roundness of twin-spots is
identical in the anterior and posterior compartment and independent from the
distance towards the compartment boundary. Effectively, this observed behavior of
experimental twin spots served as another constraint for our model parameterization:
Differences in Hh pathway activity that are at most two-fold should not be
sufficient to generate an observable boundary; and, the actual change in pathway
activity at the endogenous boundary can thus be inferred to be much higher.
In an earlier version of the model, we had assumed that the amount of cortical
constriction would simply be directly proportional to the hypothetical Hh target TMx
(data not shown). However, under this assumption we were unable to find a parameter
set that would satisfy all constraints and that would result in realistic cell
shapes. Cells were either visibly too small or too large in the TMx expression
stripe, and/or were showing imbalances in the relative contributions of cortical
forces and area elasticity, leading to distorted cellular shapes (data not shown).
In our view, this indicates that the processes at the boundary are not simply based
on increasing or decreasing overall cortical constriction, but instead on a local
redistribution of a pre-existing, basal propensity for cortical constriction. As an
important consequence, it appears that it is not the absolute level of TMx that is
important, but the ratio of TMx expression between two neighboring
cells.
Our model is the first to couple tissue growth, driven by explicit cell divisions in
a force-balanced cellular vertex approach, to signal transduction processes
including diffusion, transcriptional responses and mechanical effects. This general
approach should be applicable to a number of crucial developmental mechanisms,
including growth control and body axis specification [24], [30], [31], [32]. In our case, we chose to model
the Hh pathway, despite lacking many of the kinetic parameters that are needed to
fully describe the pathway. This is probably the situation faced for most
developmental signal transduction pathways today. However, we do believe this
approach is justified, as long the as the outcome of the modeling is challenged
experimentally, and as long as the sought-after answers are not addressing merely
quantitative nuances in the pathway, but instead more fundamental mechanistic
choices. Here, we essentially aimed to clarify whether a homotypic boundary model
can work in principle (Figure
1), and whether a single, classical transcriptional target of Hh could be the
missing link between pathway activity and physical forces at the cellular level. We
find that this could indeed be the case, and that such a target gene might even be
expressed at a basal level outside the Hh signaling stripe (since only relative
differences at the boundary are needed). Our findings provide one possible
explanation why previous attempts to search for this gene were unsuccessful: often
it was assumed that the gene would be strongly expressed anteriorly, but not at all
posteriorly. Instead, in our model the gene can indeed be expressed posteriorly (in
fact, many configurations are possible, as long as they include a localized
difference in expression at the boundary). Overall, our study indicates that
mechanistic pathway modeling within whole tissues can help to choose among
hypothetical, conflicting scenarios, and that it can even constrain properties of
postulated missing players in a pathway.
To generate smo mutant clones, the
smo3 allele was flipped against a CD2-marked FRT
chromosome. After mitotic recombination took place, non-CD2 expressing cells
were homozygous mutant for smo3. Cells of the
posterior compartment were marked by expression of a hh-lacZ
transgene. Flies had the following genotype: y w hsflp; FRT39
smo3/FRT39 hsCD2;
hhlacZ/+.
Antibody stainings of imaginal discs were done as described previously [33]. The
following antibodies were used: rabbit α-E-Cadherin (Santa Cruz
Biotechnology, 1∶200), mouse α-CD2 (Serotec, 1∶500), chicken
α-βGal (Immunology Consultants Laboratory 1∶1000).
The shapes of smo− clones were determined by
the absence of CD2 staining; correspondingly, the shapes of twin spots were
defined by increased CD2 staining. The ‘roundness’ of
smo− clones or twin spots was quantified
by the measure
[34], where
A is the area of the clone (or of the twin spot) and
L its perimeter. Circular clones have
, all other clones have . Clonal position
was defined by the distance of the center of mass of the clone to the A/P
boundary as marked by hh-lacZ staining. All geometry
measurements in confocal microscopy images of wing discs, as well as in the
corresponding images from simulations, were fully automatized with the help of
the ImageProcessingToolbox™ of Matlab.
We explicitly describe the Hedgehog pathway by a coupled system of ordinary and
partial differential equations. The Hh protein, produced in the posterior
compartment of the wing disc, diffuses into the anterior compartment, where it
binds reversibly to its receptor Patched (Ptc). Binding of Hh to Ptc relieves
the repression of the transmembrane protein Smoothened (Smo) by Ptc, but neither
the mechanism for Ptc repression of Smo nor the mechanism by which the complex
[Hh Ptch] relieves this repression has been fully understood. We
assume that the active form of Smo corresponds to a complex of Smo protein and
an unknown ligand Lx, [Lx Smo]. We further assume
that, in a membrane compartment inaccessible to Smo, there exists a reservoir of
Lx, from where it can flow towards Smo via a passive transport mechanism. We
assume that Lx gets pumped away from Smo (active transport) by unbound Ptc. Ptc
in turn is produced with a constant, low basal rate in the A-compartment, and is
additionally a transcriptional target downstream of the active form of Smo in
the A-compartment (via the transcription factor Ci, not modeled explicitly).
Finally, we assume that the putative transmembrane protein “TMx” is
likewise a transcriptional target downstream of the active form of Smo, with an
additional, basal expression throughout the tissue. In Figure 3A, the above players and their
interactions are summarized. This network of interactions is translated into a
system of coupled ordinary and partial differential equations, listed in Figure 3B. Since cell-to-cell
diffusion is restricted to the Hh molecule, only the first equation includes
spatial derivatives, whereas all other equations are ordinary differential
equations. We assume that protein kinetics can be described by a constant set of
parameters for each protein [35]. smo− clones were
mimicked by setting the corresponding production rate
to zero; the corresponding twin spots were modeled by
doubling this production rate as compared to wild type cells. The coefficients
appearing in the system of equations are provided and described in the
Supplemental Material (Text S1).
The apical side of Drosophila's wing disc is modeled as a two-dimensional
vertex model, where the junctions between cells are defined by straight lines
(edges) connecting vertices. The resulting tissue topology is obtained by
minimizing an energy function describing visco-elastic properties of the cells.
Our model is an extension of previously published models describing cells as
polygons [21], [26], [36]. In keeping with the framework of these previous
models, mechanical forces are not stated explicitly; instead, by minimizing the
‘work function’ that aims to reflect the potential energy of the
system, quasi-instantaneous relaxation of the system into local energy-minima is
achieved [37]. This is assumed to correspond to the outcome of balanced
forces acting in the elastic, dampened system of the tissue. It should be
stressed that the modeling takes place on three, well-separated time scales: at
the longest time scale (hours to days), cells divide and the tissue grows. At
the medium time scale (minutes to hours), signaling proteins diffuse and are
transduced into molecular responses inside the cell. The actual mechanics
(forces and movements) occur at the shortest time scale – on the order of
seconds – as has been demonstrated experimentally by tracking the
relaxation movements following laser ablations in the tissue [16].
In our extension of the published models, we assume that a putative transmembrane
protein downstream of the Hedgehog pathway (“TMx”) leads to a change
in the line tension term of the energy function. We assume that TMx molecules
are preferentially recruited to those edges that offer more binding partners
(i.e., other TMx molecules expressed on neighboring cells). At the inner side of
the cell membrane, TMx is assumed to signal to “effectors” that in
turn influence cortical tension. The total number of effectors in each cell is
not influenced by Hh signaling and is rate limiting. Both requirements are
reflected in the definition of an additional scaling
factor in the line tension contribution of the energy function
displayed in Figure 3. Note
that the sum in the definition of the scaling factor runs only over the edges of
cell α.
The size of the scaling factor only depends on the ratio of the
concentrations of two neighboring cells and is thus independent on the absolute
values of concentrations (Figure S3). For all cells outside the stripe
of increased TMx expression, the scaling factor computes to 1 and the energy
function thus corresponds to the original energy function published in ref [21]. The
scaling factor is strongly increased above the basal value of 1 on those edges
of posterior cells immediately straddling the boundary (and thus touching
anterior cells); and it is strongly decreased on all other edges of those cells.
The changes in the scaling factor for anterior cells are more subtle, as shown
in Figure
S3. As each edge belongs to two cells, and their scaling factors for
a given edge may not be the same, the energy function effectively takes into
account the average of the two factors.
With the dimensionless parameters and
, we obtain the following normalized energy function from
the energy function displayed in panel C of Figure 3:(1)We
minimize the normalized energy function eq. (1) by a conjugate gradient method,
leading to a shortening of edges that have an increased scaling factor and a
lengthening of edges with a decreased scaling factor. This causes a
straightening of the boundary between anterior and posterior cells, and (as an
interesting side-effect) an increased average area of cells immediately
posterior to the boundary (i.e., “P1” cells in Figure S2A;
this has been experimentally observed as well [16]). Final simulations were
run with the following parameter set for the normalized energy function:
, (and
for edges of cells constituting the outer margin of the
tissue). Assuming an average cell edge length of , we applied the
same target area to all cells, based on a regular hexagon with edge length
l: .
The simulation of tissue growth was implemented as described in ref. [21]. In
contrast to this previous work, we chose not to apply periodic boundary
conditions, but modeled the tissue margins explicitly. Diffusion of Hh was
discretized by the Finite Volume Method, using the cells as local control
volumes. Following each growth step, the diffusion step was executed, and then
the remaining of the differential equations displayed in Figure 3 (kinetic reactions) were solved
numerically within each cell (for further details see Supplementary Information,
Text
S1). Simulations were started with 220 cells placed in a regular grid
(always using the same starting formation). In simulations that included mutant
clones, 20 smo− cells (simulated by a zero Smo
production rate) adjacent to the corresponding twin spot cells (simulated by a
doubled production rate of Smo as compared to wild type cells) were distributed
uniformly in the starting configuration. We set the initial concentrations for
all proteins within each cell to zero. Between cells and the extracellular
medium we applied zero flux boundary conditions.
All simulations were run until the number of cells had increased to 6000; this
roughly corresponds to the total number of cells in the pouch of a third-instar
wing disc.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.